diff options
Diffstat (limited to 'lib/compiler/src')
56 files changed, 8570 insertions, 5724 deletions
diff --git a/lib/compiler/src/Makefile b/lib/compiler/src/Makefile index 299b2892fc..2408c76b48 100644 --- a/lib/compiler/src/Makefile +++ b/lib/compiler/src/Makefile @@ -1,7 +1,7 @@ # # %CopyrightBegin% # -# Copyright Ericsson AB 1996-2013. All Rights Reserved. +# Copyright Ericsson AB 1996-2018. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -49,7 +49,7 @@ MODULES = \ beam_a \ beam_asm \ beam_block \ - beam_bool \ + beam_bs \ beam_bsm \ beam_clean \ beam_dead \ @@ -62,6 +62,8 @@ MODULES = \ beam_opcodes \ beam_peep \ beam_receive \ + beam_reorder \ + beam_record \ beam_split \ beam_trim \ beam_type \ @@ -81,25 +83,24 @@ MODULES = \ core_scan \ erl_bifs \ rec_env \ + sys_core_alias \ + sys_core_bsm \ sys_core_dsetel \ sys_core_fold \ sys_core_fold_lists \ sys_core_inline \ sys_pre_attributes \ - sys_pre_expand \ v3_codegen \ v3_core \ v3_kernel \ - v3_kernel_pp \ - v3_life + v3_kernel_pp BEAM_H = $(wildcard ../priv/beam_h/*.h) HRL_FILES= \ beam_disasm.hrl \ core_parse.hrl \ - v3_kernel.hrl \ - v3_life.hrl + v3_kernel.hrl YRL_FILE = core_parse.yrl @@ -126,7 +127,7 @@ ERL_COMPILE_FLAGS += +native endif ERL_COMPILE_FLAGS += +inline +warn_unused_import \ -Werror \ - -I../../stdlib/include -I$(EGEN) -W + -I../../stdlib/include -I$(EGEN) -W +warn_missing_spec # ---------------------------------------------------- # Targets @@ -184,21 +185,19 @@ release_docs_spec: # ---------------------------------------------------- $(EBIN)/beam_disasm.beam: $(EGEN)/beam_opcodes.hrl beam_disasm.hrl -$(EBIN)/beam_listing.beam: v3_life.hrl -$(EBIN)/beam_validator.beam: beam_disasm.hrl +$(EBIN)/beam_listing.beam: core_parse.hrl v3_kernel.hrl $(EBIN)/cerl.beam: core_parse.hrl $(EBIN)/compile.beam: core_parse.hrl ../../stdlib/include/erl_compile.hrl $(EBIN)/core_lib.beam: core_parse.hrl $(EBIN)/core_lint.beam: core_parse.hrl $(EBIN)/core_parse.beam: core_parse.hrl $(EGEN)/core_parse.erl $(EBIN)/core_pp.beam: core_parse.hrl +$(EBIN)/sys_core_alias.beam: core_parse.hrl $(EBIN)/sys_core_dsetel.beam: core_parse.hrl $(EBIN)/sys_core_fold.beam: core_parse.hrl $(EBIN)/sys_core_fold_lists.beam: core_parse.hrl $(EBIN)/sys_core_inline.beam: core_parse.hrl -$(EBIN)/sys_pre_expand.beam: ../../stdlib/include/erl_bits.hrl -$(EBIN)/v3_codegen.beam: v3_life.hrl +$(EBIN)/v3_codegen.beam: v3_kernel.hrl $(EBIN)/v3_core.beam: core_parse.hrl $(EBIN)/v3_kernel.beam: core_parse.hrl v3_kernel.hrl $(EBIN)/v3_kernel_pp.beam: v3_kernel.hrl -$(EBIN)/v3_life.beam: v3_kernel.hrl v3_life.hrl diff --git a/lib/compiler/src/beam_a.erl b/lib/compiler/src/beam_a.erl index f0f2ee08c2..6fd4ace540 100644 --- a/lib/compiler/src/beam_a.erl +++ b/lib/compiler/src/beam_a.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2012-2013. All Rights Reserved. +%% Copyright Ericsson AB 2012-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -25,6 +25,9 @@ -export([module/2]). +-spec module(beam_asm:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> Fs = [function(F) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. @@ -39,8 +42,7 @@ function({function,Name,Arity,CLabel,Is0}) -> Is = beam_jump:remove_unused_labels(Is1), {function,Name,Arity,CLabel,Is} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -56,8 +58,17 @@ rename_instrs([{call_only,A,F}|Is]) -> rename_instrs([{call_ext_only,A,F}|Is]) -> [{call_ext,A,F},return|rename_instrs(Is)]; rename_instrs([{'%live',_}|Is]) -> - %% When compiling from old .S files. + %% Ignore old type of live annotation. Only happens when compiling + %% from very old .S files. rename_instrs(Is); +rename_instrs([{get_list,S,D1,D2}|Is]) -> + %% Only happens when compiling from old .S files. + if + D1 =:= S -> + [{get_tl,S,D2},{get_hd,S,D1}|rename_instrs(Is)]; + true -> + [{get_hd,S,D1},{get_tl,S,D2}|rename_instrs(Is)] + end; rename_instrs([I|Is]) -> [rename_instr(I)|rename_instrs(Is)]; rename_instrs([]) -> []. diff --git a/lib/compiler/src/beam_asm.erl b/lib/compiler/src/beam_asm.erl index a3201b0f4a..df0321e85a 100644 --- a/lib/compiler/src/beam_asm.erl +++ b/lib/compiler/src/beam_asm.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1996-2013. All Rights Reserved. +%% Copyright Ericsson AB 1996-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -24,19 +24,51 @@ -export([module/4]). -export([encode/2]). +-export_type([fail/0,label/0,reg/0,reg_num/0,src/0,module_code/0,function_name/0]). + -import(lists, [map/2,member/2,keymember/3,duplicate/2,splitwith/2]). -include("beam_opcodes.hrl"). -module(Code, Abst, SourceFile, Opts) -> - {ok,assemble(Code, Abst, SourceFile, Opts)}. +%% Common types for describing operands for BEAM instructions. +-type reg_num() :: 0..1023. +-type reg() :: {'x',reg_num()} | {'y',reg_num()}. +-type src() :: reg() | + {'literal',term()} | + {'atom',atom()} | + {'integer',integer()} | + 'nil' | + {'float',float()}. +-type label() :: pos_integer(). +-type fail() :: {'f',label() | 0}. + +%% asm_instruction() describes only the instructions that +%% are used in BEAM files (as opposed to internal instructions +%% used only during optimization). + +-type asm_instruction() :: atom() | tuple(). + +-type function_name() :: atom(). + +-type asm_function() :: + {'function',function_name(),arity(),label(),[asm_instruction()]}. + +-type module_code() :: + {module(),[_],[_],[asm_function()],pos_integer()}. + +-spec module(module_code(), [{binary(), binary()}], [{atom(),term()}], [compile:option()]) -> + {'ok',binary()}. -assemble({Mod,Exp,Attr0,Asm0,NumLabels}, Abst, SourceFile, Opts) -> +module(Code, ExtraChunks, CompileInfo, CompilerOpts) -> + {ok,assemble(Code, ExtraChunks, CompileInfo, CompilerOpts)}. + +assemble({Mod,Exp0,Attr0,Asm0,NumLabels}, ExtraChunks, CompileInfo, CompilerOpts) -> {1,Dict0} = beam_dict:atom(Mod, beam_dict:new()), {0,Dict1} = beam_dict:fname(atom_to_list(Mod) ++ ".erl", Dict0), NumFuncs = length(Asm0), {Asm,Attr} = on_load(Asm0, Attr0), + Exp = cerl_sets:from_list(Exp0), {Code,Dict2} = assemble_1(Asm, Exp, Dict1, []), - build_file(Code, Attr, Dict2, NumLabels, NumFuncs, Abst, SourceFile, Opts). + build_file(Code, Attr, Dict2, NumLabels, NumFuncs, ExtraChunks, CompileInfo, CompilerOpts). on_load(Fs0, Attr0) -> case proplists:get_value(on_load, Attr0) of @@ -61,7 +93,7 @@ insert_on_load_instruction(Is0, Entry) -> Bef ++ [El,on_load|Is]. assemble_1([{function,Name,Arity,Entry,Asm}|T], Exp, Dict0, Acc) -> - Dict1 = case member({Name,Arity}, Exp) of + Dict1 = case cerl_sets:is_element({Name,Arity}, Exp) of true -> beam_dict:export(Name, Arity, Entry, Dict0); false -> @@ -79,7 +111,7 @@ assemble_function([H|T], Acc, Dict0) -> assemble_function([], Code, Dict) -> {Code, Dict}. -build_file(Code, Attr, Dict, NumLabels, NumFuncs, Abst, SourceFile, Opts) -> +build_file(Code, Attr, Dict, NumLabels, NumFuncs, ExtraChunks, CompileInfo, CompilerOpts) -> %% Create the code chunk. CodeChunk = chunk(<<"Code">>, @@ -91,9 +123,9 @@ build_file(Code, Attr, Dict, NumLabels, NumFuncs, Abst, SourceFile, Opts) -> Code), %% Create the atom table chunk. - - {NumAtoms, AtomTab} = beam_dict:atom_table(Dict), - AtomChunk = chunk(<<"Atom">>, <<NumAtoms:32>>, AtomTab), + AtomEncoding = atom_encoding(CompilerOpts), + {NumAtoms, AtomTab} = beam_dict:atom_table(Dict, AtomEncoding), + AtomChunk = chunk(atom_chunk_name(AtomEncoding), <<NumAtoms:32>>, AtomTab), %% Create the import table chunk. @@ -150,25 +182,34 @@ build_file(Code, Attr, Dict, NumLabels, NumFuncs, Abst, SourceFile, Opts) -> Essentials1 = [iolist_to_binary(C) || C <- Essentials0], MD5 = module_md5(Essentials1), Essentials = finalize_fun_table(Essentials1, MD5), - {Attributes,Compile} = build_attributes(Opts, SourceFile, Attr, MD5), + {Attributes,Compile} = build_attributes(Attr, CompileInfo, MD5), AttrChunk = chunk(<<"Attr">>, Attributes), CompileChunk = chunk(<<"CInf">>, Compile), - %% Create the abstract code chunk. + %% Compile all extra chunks. - AbstChunk = chunk(<<"Abst">>, Abst), + CheckedChunks = [chunk(Key, Value) || {Key, Value} <- ExtraChunks], %% Create IFF chunk. - Chunks = case member(slim, Opts) of + Chunks = case member(slim, CompilerOpts) of true -> - [Essentials,AttrChunk,AbstChunk]; + [Essentials,AttrChunk]; false -> [Essentials,LocChunk,AttrChunk, - CompileChunk,AbstChunk,LineChunk] + CompileChunk,CheckedChunks,LineChunk] end, build_form(<<"BEAM">>, Chunks). +atom_encoding(Opts) -> + case proplists:get_bool(no_utf8_atoms, Opts) of + false -> utf8; + true -> latin1 + end. + +atom_chunk_name(utf8) -> <<"AtU8">>; +atom_chunk_name(latin1) -> <<"Atom">>. + %% finalize_fun_table(Essentials, MD5) -> FinalizedEssentials %% Update the 'old_uniq' field in the entry for each fun in the %% 'FunT' chunk. We'll use part of the MD5 for the module as a @@ -199,9 +240,7 @@ build_form(Id, Chunks0) when byte_size(Id) =:= 4, is_list(Chunks0) -> chunk(Id, Contents) when byte_size(Id) =:= 4, is_binary(Contents) -> Size = byte_size(Contents), - [<<Id/binary,Size:32>>,Contents|pad(Size)]; -chunk(Id, Contents) when is_list(Contents) -> - chunk(Id, list_to_binary(Contents)). + [<<Id/binary,Size:32>>,Contents|pad(Size)]. %% Build a correctly padded chunk (with a sub-header). @@ -223,15 +262,10 @@ flatten_exports(Exps) -> flatten_imports(Imps) -> list_to_binary(map(fun({M,F,A}) -> <<M:32,F:32,A:32>> end, Imps)). -build_attributes(Opts, SourceFile, Attr, MD5) -> - Misc = case member(slim, Opts) of - false -> - {{Y,Mo,D},{H,Mi,S}} = erlang:universaltime(), - [{time,{Y,Mo,D,H,Mi,S}},{source,SourceFile}]; - true -> [] - end, - Compile = [{options,Opts},{version,?COMPILER_VSN}|Misc], - {term_to_binary(set_vsn_attribute(Attr, MD5)),term_to_binary(Compile)}. +build_attributes(Attr, Compile, MD5) -> + AttrBinary = term_to_binary(set_vsn_attribute(Attr, MD5)), + CompileBinary = term_to_binary([{version,?COMPILER_VSN}|Compile]), + {AttrBinary,CompileBinary}. build_line_table(Dict) -> {NumLineInstrs,NumFnames0,Fnames0,NumLines,Lines0} = @@ -373,7 +407,17 @@ encode_arg({atom, Atom}, Dict0) when is_atom(Atom) -> {Index, Dict} = beam_dict:atom(Atom, Dict0), {encode(?tag_a, Index), Dict}; encode_arg({integer, N}, Dict) -> - {encode(?tag_i, N), Dict}; + %% Conservatily assume that all integers whose absolute + %% value is greater than 1 bsl 128 will be bignums in + %% the runtime system. + if + N >= 1 bsl 128 -> + encode_arg({literal, N}, Dict); + N =< -(1 bsl 128) -> + encode_arg({literal, N}, Dict); + true -> + {encode(?tag_i, N), Dict} + end; encode_arg(nil, Dict) -> {encode(?tag_a, 0), Dict}; encode_arg({f, W}, Dict) -> @@ -431,6 +475,8 @@ encode_alloc_list_1([{floats,Floats}|T], Dict, Acc0) -> encode_alloc_list_1([], Dict, Acc) -> {iolist_to_binary(Acc),Dict}. +-spec encode(non_neg_integer(), integer()) -> iolist() | integer(). + encode(Tag, N) when N < 0 -> encode1(Tag, negative_to_bytes(N)); encode(Tag, N) when N < 16 -> diff --git a/lib/compiler/src/beam_block.erl b/lib/compiler/src/beam_block.erl index 0321b1c07b..fe43163455 100644 --- a/lib/compiler/src/beam_block.erl +++ b/lib/compiler/src/beam_block.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2013. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -23,31 +23,36 @@ -module(beam_block). -export([module/2]). --import(lists, [mapfoldl/3,reverse/1,reverse/2,foldl/3,member/2]). --define(MAXREG, 1024). +-import(lists, [reverse/1,reverse/2,member/2]). -module({Mod,Exp,Attr,Fs0,Lc0}, _Opt) -> - {Fs,Lc} = mapfoldl(fun function/2, Lc0, Fs0), +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + +module({Mod,Exp,Attr,Fs0,Lc}, Opts) -> + Blockify = not member(no_blockify, Opts), + Fs = [function(F, Blockify) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. -function({function,Name,Arity,CLabel,Is0}, Lc0) -> +function({function,Name,Arity,CLabel,Is0}, Blockify) -> try %% Collect basic blocks and optimize them. - Is1 = blockify(Is0), - Is2 = embed_lines(Is1), - Is3 = move_allocates(Is2), - Is4 = beam_utils:live_opt(Is3), - Is5 = opt_blocks(Is4), - Is6 = beam_utils:delete_live_annos(Is5), - - %% Optimize bit syntax. - {Is,Lc} = bsm_opt(Is6, Lc0), - - %% Done. - {{function,Name,Arity,CLabel,Is},Lc} + Is1 = case Blockify of + false -> Is0; + true -> blockify(Is0) + end, + Is2 = embed_lines(Is1), + Is3 = local_cse(Is2), + Is4 = beam_utils:anno_defs(Is3), + Is5 = move_allocates(Is4), + Is6 = beam_utils:live_opt(Is5), + Is7 = opt_blocks(Is6), + Is8 = beam_utils:delete_annos(Is7), + Is = opt_allocs(Is8), + + %% Done. + {function,Name,Arity,CLabel,Is} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -62,56 +67,15 @@ blockify(Is) -> blockify([{loop_rec,{f,Fail},{x,0}},{loop_rec_end,_Lbl},{label,Fail}|Is], Acc) -> %% Useless instruction sequence. blockify(Is, Acc); -blockify([{test,is_atom,{f,Fail},[Reg]}=I| - [{select,select_val,Reg,{f,Fail}, - [{atom,false},{f,_}=BrFalse, - {atom,true}=AtomTrue,{f,_}=BrTrue]}|Is]=Is0], - [{block,Bl}|_]=Acc) -> - case is_last_bool(Bl, Reg) of - false -> - blockify(Is0, [I|Acc]); - true -> - %% The last instruction is a boolean operator/guard BIF that can't fail. - %% We can convert the three-way branch to a two-way branch (eliminating - %% the reference to the failure label). - blockify(Is, [{jump,BrTrue}, - {test,is_eq_exact,BrFalse,[Reg,AtomTrue]}|Acc]) - end; -blockify([{test,is_atom,{f,Fail},[Reg]}=I| - [{select,select_val,Reg,{f,Fail}, - [{atom,true}=AtomTrue,{f,_}=BrTrue, - {atom,false},{f,_}=BrFalse]}|Is]=Is0], - [{block,Bl}|_]=Acc) -> - case is_last_bool(Bl, Reg) of - false -> - blockify(Is0, [I|Acc]); - true -> - blockify(Is, [{jump,BrTrue}, - {test,is_eq_exact,BrFalse,[Reg,AtomTrue]}|Acc]) - end; blockify([I|Is0]=IsAll, Acc) -> - case is_bs_put(I) of - true -> - {BsPuts0,Is} = collect_bs_puts(IsAll), - BsPuts = opt_bs_puts(BsPuts0), - blockify(Is, reverse(BsPuts, Acc)); - false -> - case collect(I) of - error -> blockify(Is0, [I|Acc]); - Instr when is_tuple(Instr) -> - {Block,Is} = collect_block(IsAll), - blockify(Is, [{block,Block}|Acc]) - end + case collect(I) of + error -> blockify(Is0, [I|Acc]); + Instr when is_tuple(Instr) -> + {Block,Is} = collect_block(IsAll), + blockify(Is, [{block,Block}|Acc]) end; blockify([], Acc) -> reverse(Acc). -is_last_bool([{set,[Reg],As,{bif,N,_}}], Reg) -> - Ar = length(As), - erl_internal:new_type_test(N, Ar) orelse erl_internal:comp_op(N, Ar) - orelse erl_internal:bool_op(N, Ar); -is_last_bool([_|Is], Reg) -> is_last_bool(Is, Reg); -is_last_bool([], _) -> false. - collect_block(Is) -> collect_block(Is, []). @@ -126,7 +90,9 @@ collect_block([I|Is]=Is0, Acc) -> case collect(I) of error -> {reverse(Acc),Is0}; Instr -> collect_block(Is, [Instr|Acc]) - end. + end; +collect_block([], Acc) -> + {reverse(Acc),[]}. collect({allocate,N,R}) -> {set,[],[],{alloc,R,{nozero,N,0,[]}}}; collect({allocate_zero,N,R}) -> {set,[],[],{alloc,R,{zero,N,0,[]}}}; @@ -142,14 +108,15 @@ collect({put_tuple,A,D}) -> {set,[D],[],{put_tuple,A}}; collect({put,S}) -> {set,[],[S],put}; collect({get_tuple_element,S,I,D}) -> {set,[D],[S],{get_tuple_element,I}}; collect({set_tuple_element,S,D,I}) -> {set,[],[S,D],{set_tuple_element,I}}; -collect({get_list,S,D1,D2}) -> {set,[D1,D2],[S],get_list}; +collect({get_hd,S,D}) -> {set,[D],[S],get_hd}; +collect({get_tl,S,D}) -> {set,[D],[S],get_tl}; collect(remove_message) -> {set,[],[],remove_message}; collect({put_map,F,Op,S,D,R,{list,Puts}}) -> {set,[D],[S|Puts],{alloc,R,{put_map,Op,F}}}; -collect({get_map_elements,F,S,{list,Gets}}) -> - {Ss,Ds} = beam_utils:split_even(Gets), - {set,Ds,[S|Ss],{get_map_elements,F}}; -collect({'catch',R,L}) -> {set,[R],[],{'catch',L}}; +collect({'catch'=Op,R,L}) -> + {set,[R],[],{try_catch,Op,L}}; +collect({'try'=Op,R,L}) -> + {set,[R],[],{try_catch,Op,L}}; collect(fclearerror) -> {set,[],[],fclearerror}; collect({fcheckerror,{f,0}}) -> {set,[],[],fcheckerror}; collect({fmove,S,D}) -> {set,[D],[S],fmove}; @@ -170,21 +137,27 @@ embed_lines([{block,B2},{line,_}=Line,{block,B1}|T], Acc) -> embed_lines([{block,B1},{line,_}=Line|T], Acc) -> B = {block,[{set,[],[],Line}|B1]}, embed_lines([B|T], Acc); +embed_lines([{block,B2},{block,B1}|T], Acc) -> + %% This can only happen when beam_block is run for + %% the second time. + B = {block,B1++B2}, + embed_lines([B|T], Acc); embed_lines([I|Is], Acc) -> embed_lines(Is, [I|Acc]); embed_lines([], Acc) -> Acc. opt_blocks([{block,Bl0}|Is]) -> %% The live annotation at the beginning is not useful. - [{'%live',_,_}|Bl] = Bl0, + [{'%anno',_}|Bl] = Bl0, [{block,opt_block(Bl)}|opt_blocks(Is)]; opt_blocks([I|Is]) -> [I|opt_blocks(Is)]; opt_blocks([]) -> []. opt_block(Is0) -> - Is = find_fixpoint(fun opt/1, Is0), - opt_alloc(Is). + find_fixpoint(fun(Is) -> + opt_tuple_element(opt(Is)) + end, Is0). find_fixpoint(OptFun, Is0) -> case OptFun(Is0) of @@ -193,14 +166,39 @@ find_fixpoint(OptFun, Is0) -> end. %% move_allocates(Is0) -> Is -%% Move allocate instructions upwards in the instruction stream, in the -%% hope of getting more possibilities for optimizing away moves later. +%% Move allocate instructions upwards in the instruction stream +%% (within the same block), in the hope of getting more possibilities +%% for optimizing away moves later. +%% +%% For example, we can transform the following instructions: +%% +%% get_tuple_element x(1) Element => x(2) +%% allocate_zero StackSize 3 %% x(0), x(1), x(2) are live +%% +%% to the following instructions: +%% +%% allocate_zero StackSize 2 %% x(0) and x(1) are live +%% get_tuple_element x(1) Element => x(2) %% -%% NOTE: Moving allocation instructions is only safe because it is done -%% immediately after code generation so that we KNOW that if {x,X} is -%% initialized, all x registers with lower numbers are also initialized. -%% That assumption may not be true after other optimizations, such as -%% the beam_utils:live_opt/1 optimization. +%% NOTE: Since the beam_reorder pass has been run, it is no longer +%% safe to assume that if x(N) is initialized, then all lower-numbered +%% x registers are also initialized. +%% +%% For example, we must be careful when transforming the following +%% instructions: +%% +%% get_tuple_element x(0) Element => x(1) +%% allocate_zero StackSize 3 %x(0), x(1), x(2) are live +%% +%% to the following instructions: +%% +%% allocate_zero StackSize 3 +%% get_tuple_element x(0) Element => x(1) +%% +%% The transformation is safe if and only if x(1) has been +%% initialized previously. We will use the annotations added by +%% beam_utils:anno_defs/1 to determine whether x(a) has been +%% initialized. move_allocates([{block,Bl0}|Is]) -> Bl = move_allocates_1(reverse(Bl0), []), @@ -209,54 +207,77 @@ move_allocates([I|Is]) -> [I|move_allocates(Is)]; move_allocates([]) -> []. -move_allocates_1([{set,[],[],{alloc,_,_}=Alloc}|Is0], Acc0) -> - {Is,Acc} = move_allocates_2(Alloc, Is0, Acc0), +move_allocates_1([{'%anno',_}|Is], Acc) -> move_allocates_1(Is, Acc); -move_allocates_1([I|Is], Acc) -> - move_allocates_1(Is, [I|Acc]); -move_allocates_1([], Is) -> Is. - -move_allocates_2({alloc,Live,Info}, [{set,[],[],{alloc,Live0,Info0}}|Is], Acc) -> - Live = Live0, % Assertion. - Alloc = {alloc,Live,combine_alloc(Info0, Info)}, - move_allocates_2(Alloc, Is, Acc); -move_allocates_2({alloc,Live,Info}=Alloc0, [I|Is]=Is0, Acc) -> +move_allocates_1([I|Is], [{set,[],[],{alloc,Live0,Info0}}|Acc]=Acc0) -> case alloc_may_pass(I) of - false -> - {Is0,[{set,[],[],Alloc0}|Acc]}; - true -> - Alloc = {alloc,alloc_live_regs(I, Live),Info}, - move_allocates_2(Alloc, Is, [I|Acc]) + false -> + move_allocates_1(Is, [I|Acc0]); + true -> + case alloc_live_regs(I, Is, Live0) of + not_possible -> + move_allocates_1(Is, [I|Acc0]); + Live when is_integer(Live) -> + Info = safe_info(Info0), + A = {set,[],[],{alloc,Live,Info}}, + move_allocates_1(Is, [A,I|Acc]) + end end; -move_allocates_2(Alloc, [], Acc) -> - {[],[{set,[],[],Alloc}|Acc]}. +move_allocates_1([I|Is], Acc) -> + move_allocates_1(Is, [I|Acc]); +move_allocates_1([], Acc) -> Acc. +alloc_may_pass({set,_,[{fr,_}],fmove}) -> false; alloc_may_pass({set,_,_,{alloc,_,_}}) -> false; alloc_may_pass({set,_,_,{set_tuple_element,_}}) -> false; -alloc_may_pass({set,_,_,{get_map_elements,_}}) -> false; alloc_may_pass({set,_,_,put_list}) -> false; alloc_may_pass({set,_,_,put}) -> false; alloc_may_pass({set,_,_,_}) -> true. - -combine_alloc({_,Ns,Nh1,Init}, {_,nostack,Nh2,[]}) -> - {zero,Ns,beam_utils:combine_heap_needs(Nh1, Nh2),Init}. + +safe_info({nozero,Stack,Heap,_}) -> + %% nozero is not safe if the allocation instruction is moved + %% upwards past an instruction that may throw an exception + %% (such as element/2). + {zero,Stack,Heap,[]}; +safe_info(Info) -> Info. %% opt([Instruction]) -> [Instruction] %% Optimize the instruction stream inside a basic block. opt([{set,[X],[X],move}|Is]) -> opt(Is); +opt([{set,[Dst],_,move},{set,[Dst],[Src],move}=I|Is]) when Dst =/= Src -> + opt([I|Is]); +opt([{set,[{x,0}],[S1],move}=I1,{set,[D2],[{x,0}],move}|Is]) -> + opt([I1,{set,[D2],[S1],move}|Is]); +opt([{set,[{x,0}],[S1],move}=I1,{set,[D2],[S2],move}|Is0]) when S1 =/= D2 -> + %% Place move S x0 at the end of move sequences so that + %% loader can merge with the following instruction + {Ds,Is} = opt_moves([D2], Is0), + [{set,Ds,[S2],move}|opt([I1|Is])]; opt([{set,_,_,{line,_}}=Line1, {set,[D1],[{integer,Idx1},Reg],{bif,element,{f,0}}}=I1, {set,_,_,{line,_}}=Line2, {set,[D2],[{integer,Idx2},Reg],{bif,element,{f,0}}}=I2|Is]) when Idx1 < Idx2, D1 =/= D2, D1 =/= Reg, D2 =/= Reg -> opt([Line2,I2,Line1,I1|Is]); -opt([{set,[_|_],_Ss,{get_map_elements,_F}}=I|Is]) -> - [I|opt(Is)]; +opt([{set,[D1],[{integer,Idx1},Reg],{bif,element,{f,L}}}=I1, + {set,[D2],[{integer,Idx2},Reg],{bif,element,{f,L}}}=I2|Is]) + when Idx1 < Idx2, D1 =/= D2, D1 =/= Reg, D2 =/= Reg -> + opt([I2,I1|Is]); +opt([{set,Hd0,Cons,get_hd}=GetHd, + {set,Tl0,Cons,get_tl}=GetTl|Is0]) -> + case {opt_moves(Hd0, [GetTl|Is0]),opt_moves(Tl0, [GetHd|Is0])} of + {{Hd0,Is},{Tl0,_}} -> + [GetHd|opt(Is)]; + {{Hd,Is},{Tl0,_}} -> + [{set,Hd,Cons,get_hd}|opt(Is)]; + {{_,_},{Tl,Is}} -> + [{set,Tl,Cons,get_tl}|opt(Is)] + end; opt([{set,Ds0,Ss,Op}|Is0]) -> {Ds,Is} = opt_moves(Ds0, Is0), [{set,Ds,Ss,Op}|opt(Is)]; -opt([{'%live',_,_}=I|Is]) -> +opt([{'%anno',_}=I|Is]) -> [I|opt(Is)]; opt([]) -> []. @@ -268,98 +289,223 @@ opt_moves([D0]=Ds, Is0) -> case opt_move(D0, Is0) of not_possible -> {Ds,Is0}; {D1,Is} -> {[D1],Is} - end; -opt_moves([X0,Y0], Is0) -> - {X,Is2} = case opt_move(X0, Is0) of - not_possible -> {X0,Is0}; - {Y0,_} -> {X0,Is0}; - {_X1,_Is1} = XIs1 -> XIs1 - end, - case opt_move(Y0, Is2) of - not_possible -> {[X,Y0],Is2}; - {X,_} -> {[X,Y0],Is2}; - {Y,Is} -> {[X,Y],Is} - end; -opt_moves(Ds, Is) -> - %% multiple destinations -> pass through - {Ds,Is}. - + end. %% opt_move(Dest, [Instruction]) -> {UpdatedDest,[Instruction]} | not_possible %% If there is a {move,Dest,FinalDest} instruction %% in the instruction stream, remove the move instruction %% and let FinalDest be the destination. -%% -%% For this optimization to be safe, we must be sure that -%% Dest will not be referenced in any other by other instructions -%% in the rest of the instruction stream. Not even the indirect -%% reference by an instruction that may allocate (such as -%% test_heap/2 or a GC Bif) is allowed. opt_move(Dest, Is) -> - opt_move_1(Dest, Is, ?MAXREG, []). - -opt_move_1(R, [{set,_,_,{alloc,Live,_}}|_]=Is, SafeRegs, Acc) when Live < SafeRegs -> - %% Downgrade number of safe regs and rescan the instruction, as it most probably - %% is a gc_bif instruction. - opt_move_1(R, Is, Live, Acc); -opt_move_1(R, [{set,[{x,X}=D],[R],move}|Is], SafeRegs, Acc) -> - case X < SafeRegs andalso beam_utils:is_killed_block(R, Is) of - true -> opt_move_2(D, Acc, Is); - false -> not_possible + opt_move_1(Dest, Is, []). + +opt_move_1(R, [{set,[D],[R],move}|Is0], Acc) -> + %% Provided that the source register is killed by instructions + %% that follow, the optimization is safe. + case eliminate_use_of_from_reg(Is0, R, D) of + {yes,Is} -> opt_move_rev(D, Acc, Is); + no -> not_possible end; -opt_move_1(R, [{set,[D],[R],move}|Is], _SafeRegs, Acc) -> - case beam_utils:is_killed_block(R, Is) of - true -> opt_move_2(D, Acc, Is); - false -> not_possible +opt_move_1(_R, [{set,_,_,{alloc,_,_}}|_], _) -> + %% The optimization is either not possible or not safe. + %% + %% If R is an X register killed by allocation, the optimization is + %% not safe. On the other hand, if the X register is killed, there + %% will not follow a 'move' instruction with this X register as + %% the source. + %% + %% If R is a Y register, the optimization is still not safe + %% because the new target register is an X register that cannot + %% safely pass the alloc instruction. + not_possible; +opt_move_1(R, [{set,_,_,_}=I|Is], Acc) -> + %% If the source register is either killed or used by this + %% instruction, the optimimization is not possible. + case is_killed_or_used(R, I) of + true -> not_possible; + false -> opt_move_1(R, Is, [I|Acc]) end; -opt_move_1(R, [I|Is], SafeRegs, Acc) -> - case is_transparent(R, I) of - false -> not_possible; - true -> opt_move_1(R, Is, SafeRegs, [I|Acc]) - end. +opt_move_1(_, _, _) -> + not_possible. + +%% opt_tuple_element([Instruction]) -> [Instruction] +%% If possible, move get_tuple_element instructions forward +%% in the instruction stream to a move instruction, eliminating +%% the move instruction. Example: +%% +%% get_tuple_element Tuple Pos Dst1 +%% ... +%% move Dst1 Dst2 +%% +%% This code may be possible to rewrite to: +%% +%% %%(Moved get_tuple_element instruction) +%% ... +%% get_tuple_element Tuple Pos Dst2 +%% + +opt_tuple_element([{set,[D],[S],{get_tuple_element,_}}=I|Is0]) -> + case opt_tuple_element_1(Is0, I, {S,D}, []) of + no -> + [I|opt_tuple_element(Is0)]; + {yes,Is} -> + opt_tuple_element(Is) + end; +opt_tuple_element([I|Is]) -> + [I|opt_tuple_element(Is)]; +opt_tuple_element([]) -> []. + +opt_tuple_element_1([{set,_,_,{alloc,_,_}}|_], _, _, _) -> + no; +opt_tuple_element_1([{set,_,_,{try_catch,_,_}}|_], _, _, _) -> + no; +opt_tuple_element_1([{set,[D],[S],move}|Is0], I0, {_,S}, Acc) -> + case eliminate_use_of_from_reg(Is0, S, D) of + no -> + no; + {yes,Is1} -> + {set,[S],Ss,Op} = I0, + I = {set,[D],Ss,Op}, + case opt_move_rev(S, Acc, [I|Is1]) of + not_possible -> + %% Not safe because the move of the + %% get_tuple_element instruction would cause the + %% result of a previous instruction to be ignored. + no; + {_,Is} -> + {yes,Is} + end + end; +opt_tuple_element_1([{set,Ds,Ss,_}=I|Is], MovedI, {S,D}=Regs, Acc) -> + case member(S, Ds) orelse member(D, Ss) of + true -> + no; + false -> + opt_tuple_element_1(Is, MovedI, Regs, [I|Acc]) + end; +opt_tuple_element_1(_, _, _, _) -> no. + +%% Reverse the instructions, while checking that there are no +%% instructions that would interfere with using the new destination +%% register (D). -%% Reverse the instructions, while checking that there are no instructions that -%% would interfere with using the new destination register chosen. +opt_move_rev(D, [I|Is], Acc) -> + case is_killed_or_used(D, I) of + true -> not_possible; + false -> opt_move_rev(D, Is, [I|Acc]) + end; +opt_move_rev(D, [], Acc) -> {D,Acc}. + +%% is_killed_or_used(Register, {set,_,_,_}) -> bool() +%% Test whether the register is used by the instruction. + +is_killed_or_used(R, {set,Ss,Ds,_}) -> + member(R, Ds) orelse member(R, Ss). + +%% eliminate_use_of_from_reg([Instruction], FromRegister, ToRegister, Acc) -> +%% {yes,Is} | no +%% Eliminate any use of FromRegister in the instruction sequence +%% by replacing uses of FromRegister with ToRegister. If FromRegister +%% is referenced by an allocation instruction, return 'no' to indicate +%% that FromRegister is still used and that the optimization is not +%% possible. + +eliminate_use_of_from_reg(Is, From, To) -> + try + eliminate_use_of_from_reg(Is, From, To, []) + catch + throw:not_possible -> + no + end. -opt_move_2(D, [I|Is], Acc) -> - case is_transparent(D, I) of - false -> not_possible; - true -> opt_move_2(D, Is, [I|Acc]) +eliminate_use_of_from_reg([{set,_,_,{alloc,Live,_}}|_]=Is0, {x,X}, _, Acc) -> + if + X < Live -> + no; + true -> + {yes,reverse(Acc, Is0)} end; -opt_move_2(D, [], Acc) -> {D,Acc}. - -%% is_transparent(Register, Instruction) -> true | false -%% Returns true if Instruction does not in any way references Register -%% (even indirectly by an allocation instruction). -%% Returns false if Instruction does reference Register, or we are -%% not sure. - -is_transparent({x,X}, {set,_,_,{alloc,Live,_}}) when X < Live -> - false; -is_transparent(R, {set,Ds,Ss,_Op}) -> - case member(R, Ds) of - true -> false; - false -> not member(R, Ss) +eliminate_use_of_from_reg([{set,Ds,Ss0,Op}=I0|Is], From, To, Acc) -> + ensure_safe_tuple(I0, To), + I = case member(From, Ss0) of + true -> + Ss = [case S of + From -> To; + _ -> S + end || S <- Ss0], + {set,Ds,Ss,Op}; + false -> + I0 + end, + case member(From, Ds) of + true -> + {yes,reverse(Acc, [I|Is])}; + false -> + case member(To, Ds) of + true -> + case beam_utils:is_killed_block(From, Is) of + true -> + {yes,reverse(Acc, [I|Is])}; + false -> + no + end; + false -> + eliminate_use_of_from_reg(Is, From, To, [I|Acc]) + end end; -is_transparent(_, _) -> false. +eliminate_use_of_from_reg([I]=Is, From, _To, Acc) -> + case beam_utils:is_killed_block(From, [I]) of + true -> + {yes,reverse(Acc, Is)}; + false -> + no + end. + +ensure_safe_tuple({set,[To],[],{put_tuple,_}}, To) -> + throw(not_possible); +ensure_safe_tuple(_, _) -> ok. + +%% opt_allocs(Instructions) -> Instructions. Optimize allocate +%% instructions inside blocks. If safe, replace an allocate_zero +%% instruction with the slightly cheaper allocate instruction. + +opt_allocs(Is) -> + D = beam_utils:index_labels(Is), + opt_allocs_1(Is, D). + +opt_allocs_1([{block,Bl0}|Is], D) -> + Bl = opt_alloc(Bl0, {D,Is}), + [{block,Bl}|opt_allocs_1(Is, D)]; +opt_allocs_1([I|Is], D) -> + [I|opt_allocs_1(Is, D)]; +opt_allocs_1([], _) -> []. %% opt_alloc(Instructions) -> Instructions' %% Optimises all allocate instructions. -opt_alloc([{set,[],[],{alloc,R,{_,Ns,Nh,[]}}}|Is]) -> - [{set,[],[],opt_alloc(Is, Ns, Nh, R)}|opt(Is)]; -opt_alloc([I|Is]) -> [I|opt_alloc(Is)]; -opt_alloc([]) -> []. - +opt_alloc([{set,[],[],{alloc,Live0,Info0}}, + {set,[],[],{alloc,Live,Info}}|Is], D) -> + Live = Live0, %Assertion. + Alloc = combine_alloc(Info0, Info), + I = {set,[],[],{alloc,Live,Alloc}}, + opt_alloc([I|Is], D); +opt_alloc([{set,[],[],{alloc,R,{_,Ns,Nh,[]}}}|Is], D) -> + [{set,[],[],opt_alloc(Is, D, Ns, Nh, R)}|Is]; +opt_alloc([I|Is], D) -> [I|opt_alloc(Is, D)]; +opt_alloc([], _) -> []. + +combine_alloc({_,Ns,Nh1,Init}, {_,nostack,Nh2,[]}) -> + {zero,Ns,beam_utils:combine_heap_needs(Nh1, Nh2),Init}. + %% opt_alloc(Instructions, FrameSize, HeapNeed, LivingRegs) -> [Instr] %% Generates the optimal sequence of instructions for %% allocating and initalizing the stack frame and needed heap. -opt_alloc(_Is, nostack, Nh, LivingRegs) -> +opt_alloc(_Is, _D, nostack, Nh, LivingRegs) -> {alloc,LivingRegs,{nozero,nostack,Nh,[]}}; -opt_alloc(Is, Ns, Nh, LivingRegs) -> - InitRegs = init_yreg(Is, 0), +opt_alloc(Bl, {D,OuterIs}, Ns, Nh, LivingRegs) -> + Is = [{block,Bl}|OuterIs], + InitRegs = init_yregs(Ns, Is, D), case count_ones(InitRegs) of N when N*2 > Ns -> {alloc,LivingRegs,{nozero,Ns,Nh,gen_init(Ns, InitRegs)}}; @@ -375,19 +521,14 @@ gen_init(Fs, Regs, Y, Acc) when Regs band 1 =:= 0 -> gen_init(Fs, Regs, Y, Acc) -> gen_init(Fs, Regs bsr 1, Y+1, Acc). -%% init_yreg(Instructions, RegSet) -> RegSetInitialized -%% Calculate the set of initialized y registers. - -init_yreg([{set,_,_,{bif,_,_}}|_], Reg) -> Reg; -init_yreg([{set,_,_,{alloc,_,{gc_bif,_,_}}}|_], Reg) -> Reg; -init_yreg([{set,_,_,{alloc,_,{put_map,_,_}}}|_], Reg) -> Reg; -init_yreg([{set,Ds,_,_}|Is], Reg) -> init_yreg(Is, add_yregs(Ds, Reg)); -init_yreg(_Is, Reg) -> Reg. - -add_yregs(Ys, Reg) -> foldl(fun(Y, R0) -> add_yreg(Y, R0) end, Reg, Ys). - -add_yreg({y,Y}, Reg) -> Reg bor (1 bsl Y); -add_yreg(_, Reg) -> Reg. +init_yregs(Y, Is, D) when Y >= 0 -> + case beam_utils:is_killed({y,Y}, Is, D) of + true -> + (1 bsl Y) bor init_yregs(Y-1, Is, D); + false -> + init_yregs(Y-1, Is, D) + end; +init_yregs(_, _, _) -> 0. count_ones(Bits) -> count_ones(Bits, 0). count_ones(0, Acc) -> Acc; @@ -397,15 +538,34 @@ count_ones(Bits, Acc) -> %% Calculate the new number of live registers when we move an allocate %% instruction upwards, passing a 'set' instruction. -alloc_live_regs({set,Ds,Ss,_}, Regs0) -> +alloc_live_regs({set,Ds,Ss,_}, Is, Regs0) -> Rset = x_live(Ss, x_dead(Ds, (1 bsl Regs0)-1)), - live_regs(Rset). + Live = live_regs(0, Rset), + case ensure_contiguous(Rset, Live) of + not_possible -> + %% Liveness information (looking forward in the + %% instruction stream) can't prove that moving this + %% allocation instruction is safe. Now use the annotation + %% of defined registers at the beginning of the current + %% block to see whether moving would be safe. + Def0 = defined_regs(Is, 0), + Def = Def0 band ((1 bsl Live) - 1), + ensure_contiguous(Rset bor Def, Live); + Live -> + %% Safe based on liveness information. + Live + end. -live_regs(Regs) -> - live_regs_1(0, Regs). +live_regs(N, 0) -> + N; +live_regs(N, Regs) -> + live_regs(N+1, Regs bsr 1). -live_regs_1(N, 0) -> N; -live_regs_1(N, Regs) -> live_regs_1(N+1, Regs bsr 1). +ensure_contiguous(Regs, Live) -> + case (1 bsl Live) - 1 of + Regs -> Live; + _ -> not_possible + end. x_dead([{x,N}|Rs], Regs) -> x_dead(Rs, Regs band (bnot (1 bsl N))); x_dead([_|Rs], Regs) -> x_dead(Rs, Regs); @@ -415,233 +575,119 @@ x_live([{x,N}|Rs], Regs) -> x_live(Rs, Regs bor (1 bsl N)); x_live([_|Rs], Regs) -> x_live(Rs, Regs); x_live([], Regs) -> Regs. -%%% -%%% Evaluation of constant bit fields. -%%% - -is_bs_put({bs_put,_,{bs_put_integer,_,_},_}) -> true; -is_bs_put({bs_put,_,{bs_put_float,_,_},_}) -> true; -is_bs_put(_) -> false. - -collect_bs_puts(Is) -> - collect_bs_puts_1(Is, []). - -collect_bs_puts_1([I|Is]=Is0, Acc) -> - case is_bs_put(I) of - false -> {reverse(Acc),Is0}; - true -> collect_bs_puts_1(Is, [I|Acc]) - end. - -opt_bs_puts(Is) -> - opt_bs_1(Is, []). - -opt_bs_1([{bs_put,Fail, - {bs_put_float,1,Flags0},[{integer,Sz},Src]}=I0|Is], Acc) -> - try eval_put_float(Src, Sz, Flags0) of - <<Int:Sz>> -> - Flags = force_big(Flags0), - I = {bs_put,Fail,{bs_put_integer,1,Flags}, - [{integer,Sz},{integer,Int}]}, - opt_bs_1([I|Is], Acc) - catch - error:_ -> - opt_bs_1(Is, [I0|Acc]) - end; -opt_bs_1([{bs_put,_,{bs_put_integer,1,_},[{integer,8},{integer,_}]}|_]=IsAll, - Acc0) -> - {Is,Acc} = bs_collect_string(IsAll, Acc0), - opt_bs_1(Is, Acc); -opt_bs_1([{bs_put,Fail,{bs_put_integer,1,F},[{integer,Sz},{integer,N}]}=I|Is0], - Acc) when Sz > 8 -> - case field_endian(F) of - big -> - %% We can do this optimization for any field size without risk - %% for code explosion. - case bs_split_int(N, Sz, Fail, Is0) of - no_split -> opt_bs_1(Is0, [I|Acc]); - Is -> opt_bs_1(Is, Acc) - end; - little when Sz < 128 -> - %% We only try to optimize relatively small fields, to avoid - %% an explosion in code size. - <<Int:Sz>> = <<N:Sz/little>>, - Flags = force_big(F), - Is = [{bs_put,Fail,{bs_put_integer,1,Flags}, - [{integer,Sz},{integer,Int}]}|Is0], - opt_bs_1(Is, Acc); - _ -> %native or too wide little field - opt_bs_1(Is0, [I|Acc]) - end; -opt_bs_1([{bs_put,Fail,{Op,U,F},[{integer,Sz},Src]}|Is], Acc) when U > 1 -> - opt_bs_1([{bs_put,Fail,{Op,1,F},[{integer,U*Sz},Src]}|Is], Acc); -opt_bs_1([I|Is], Acc) -> - opt_bs_1(Is, [I|Acc]); -opt_bs_1([], Acc) -> reverse(Acc). - -eval_put_float(Src, Sz, Flags) when Sz =< 256 -> %Only evaluate if Sz is reasonable. - Val = value(Src), - case field_endian(Flags) of - little -> <<Val:Sz/little-float-unit:1>>; - big -> <<Val:Sz/big-float-unit:1>> - %% native intentionally not handled here - we can't optimize it. - end. - -value({integer,I}) -> I; -value({float,F}) -> F. - -bs_collect_string(Is, [{bs_put,_,{bs_put_string,Len,{string,Str}},[]}|Acc]) -> - bs_coll_str_1(Is, Len, reverse(Str), Acc); -bs_collect_string(Is, Acc) -> - bs_coll_str_1(Is, 0, [], Acc). - -bs_coll_str_1([{bs_put,_,{bs_put_integer,U,_},[{integer,Sz},{integer,V}]}|Is], - Len, StrAcc, IsAcc) when U*Sz =:= 8 -> - Byte = V band 16#FF, - bs_coll_str_1(Is, Len+1, [Byte|StrAcc], IsAcc); -bs_coll_str_1(Is, Len, StrAcc, IsAcc) -> - {Is,[{bs_put,{f,0},{bs_put_string,Len,{string,reverse(StrAcc)}},[]}|IsAcc]}. - -field_endian({field_flags,F}) -> field_endian_1(F). - -field_endian_1([big=E|_]) -> E; -field_endian_1([little=E|_]) -> E; -field_endian_1([native=E|_]) -> E; -field_endian_1([_|Fs]) -> field_endian_1(Fs). - -force_big({field_flags,F}) -> - {field_flags,force_big_1(F)}. - -force_big_1([big|_]=Fs) -> Fs; -force_big_1([little|Fs]) -> [big|Fs]; -force_big_1([F|Fs]) -> [F|force_big_1(Fs)]. - -bs_split_int(0, Sz, _, _) when Sz > 64 -> - %% We don't want to split in this case because the - %% string will consist of only zeroes. - no_split; -bs_split_int(-1, Sz, _, _) when Sz > 64 -> - %% We don't want to split in this case because the - %% string will consist of only 255 bytes. - no_split; -bs_split_int(N, Sz, Fail, Acc) -> - FirstByteSz = case Sz rem 8 of - 0 -> 8; - Rem -> Rem - end, - bs_split_int_1(N, FirstByteSz, Sz, Fail, Acc). - -bs_split_int_1(-1, _, Sz, Fail, Acc) when Sz > 64 -> - I = {bs_put,Fail,{bs_put_integer,1,{field_flags,[big]}}, - [{integer,Sz},{integer,-1}]}, - [I|Acc]; -bs_split_int_1(0, _, Sz, Fail, Acc) when Sz > 64 -> - I = {bs_put,Fail,{bs_put_integer,1,{field_flags,[big]}}, - [{integer,Sz},{integer,0}]}, - [I|Acc]; -bs_split_int_1(N, ByteSz, Sz, Fail, Acc) when Sz > 0 -> - Mask = (1 bsl ByteSz) - 1, - I = {bs_put,Fail,{bs_put_integer,1,{field_flags,[big]}}, - [{integer,ByteSz},{integer,N band Mask}]}, - bs_split_int_1(N bsr ByteSz, 8, Sz-ByteSz, Fail, [I|Acc]); -bs_split_int_1(_, _, _, _, Acc) -> Acc. +%% defined_regs(ReversedInstructions) -> RegBitmap. +%% Given a reversed instruction stream, determine the +%% the registers that are defined. +defined_regs([{'%anno',{def,Def}}|_], Regs) -> + Def bor Regs; +defined_regs([{set,Ds,_,{alloc,Live,_}}|_], Regs) -> + x_live(Ds, Regs bor ((1 bsl Live) - 1)); +defined_regs([{set,Ds,_,_}|Is], Regs) -> + defined_regs(Is, x_live(Ds, Regs)). %%% -%%% Optimization of new bit syntax matching: get rid -%%% of redundant bs_restore2/2 instructions across select_val -%%% instructions, as well as a few other simple peep-hole optimizations. +%%% Do local common sub expression elimination (CSE) in each block. %%% -bsm_opt(Is0, Lc0) -> - {Is1,D0,Lc} = bsm_scan(Is0, [], Lc0, []), - Is2 = case D0 of - [] -> - Is1; - _ -> - D = gb_trees:from_orddict(orddict:from_list(D0)), - bsm_reroute(Is1, D, none, []) - end, - Is = beam_clean:bs_clean_saves(Is2), - {bsm_opt_2(Is, []),Lc}. - -bsm_scan([{label,L}=Lbl,{bs_restore2,_,Save}=R|Is], D0, Lc, Acc0) -> - D = [{{L,Save},Lc}|D0], - Acc = [{label,Lc},R,Lbl|Acc0], - bsm_scan(Is, D, Lc+1, Acc); -bsm_scan([I|Is], D, Lc, Acc) -> - bsm_scan(Is, D, Lc, [I|Acc]); -bsm_scan([], D, Lc, Acc) -> - {reverse(Acc),D,Lc}. - -bsm_reroute([{bs_save2,Reg,Save}=I|Is], D, _, Acc) -> - bsm_reroute(Is, D, {Reg,Save}, [I|Acc]); -bsm_reroute([{bs_restore2,Reg,Save}=I|Is], D, _, Acc) -> - bsm_reroute(Is, D, {Reg,Save}, [I|Acc]); -bsm_reroute([{label,_}=I|Is], D, S, Acc) -> - bsm_reroute(Is, D, S, [I|Acc]); -bsm_reroute([{select,select_val,Reg,F0,Lbls0}|Is], D, {_,Save}=S, Acc0) -> - [F|Lbls] = bsm_subst_labels([F0|Lbls0], Save, D), - Acc = [{select,select_val,Reg,F,Lbls}|Acc0], - bsm_reroute(Is, D, S, Acc); -bsm_reroute([{test,TestOp,F0,TestArgs}=I|Is], D, {_,Save}=S, Acc0) -> - F = bsm_subst_label(F0, Save, D), - Acc = [{test,TestOp,F,TestArgs}|Acc0], - case bsm_not_bs_test(I) of - true -> - %% The test instruction will not update the bit offset for the - %% binary being matched. Therefore the save position can be kept. - bsm_reroute(Is, D, S, Acc); - false -> - %% The test instruction might update the bit offset. Kill our - %% remembered Save position. - bsm_reroute(Is, D, none, Acc) +local_cse([{block,Bl0}|Is]) -> + Bl = cse_block(Bl0, orddict:new(), []), + [{block,Bl}|local_cse(Is)]; +local_cse([I|Is]) -> + [I|local_cse(Is)]; +local_cse([]) -> []. + +cse_block([I|Is], Es0, Acc0) -> + Es1 = cse_clear(I, Es0), + case cse_expr(I) of + none -> + %% Instruction is not suitable for CSE. + cse_block(Is, Es1, [I|Acc0]); + {ok,D,Expr} -> + %% Suitable instruction. First update the dictionary of + %% suitable expressions for the next iteration. + Es = cse_add(D, Expr, Es1), + + %% Search for a previous identical expression. + case cse_find(Expr, Es0) of + error -> + %% Nothing found + cse_block(Is, Es, [I|Acc0]); + Src -> + %% Use the previously calculated result. + %% Also eliminate any line instruction. + Move = {set,[D],[Src],move}, + case Acc0 of + [{set,_,_,{line,_}}|Acc] -> + cse_block(Is, Es, [Move|Acc]); + [_|_] -> + cse_block(Is, Es, [Move|Acc0]) + end + end end; -bsm_reroute([{test,TestOp,F0,Live,TestArgs,Dst}|Is], D, {_,Save}, Acc0) -> - F = bsm_subst_label(F0, Save, D), - Acc = [{test,TestOp,F,Live,TestArgs,Dst}|Acc0], - %% The test instruction will update the bit offset. Kill our - %% remembered Save position. - bsm_reroute(Is, D, none, Acc); -bsm_reroute([{block,[{set,[],[],{alloc,_,_}}]}=Bl, - {bs_context_to_binary,_}=I|Is], D, S, Acc) -> - %% To help further bit syntax optimizations. - bsm_reroute([I,Bl|Is], D, S, Acc); -bsm_reroute([I|Is], D, _, Acc) -> - bsm_reroute(Is, D, none, [I|Acc]); -bsm_reroute([], _, _, Acc) -> reverse(Acc). - -bsm_opt_2([{test,bs_test_tail2,F,[Ctx,Bits]}|Is], - [{test,bs_skip_bits2,F,[Ctx,{integer,I},Unit,_Flags]}|Acc]) -> - bsm_opt_2(Is, [{test,bs_test_tail2,F,[Ctx,Bits+I*Unit]}|Acc]); -bsm_opt_2([{test,bs_skip_bits2,F,[Ctx,{integer,I1},Unit1,_]}|Is], - [{test,bs_skip_bits2,F,[Ctx,{integer,I2},Unit2,Flags]}|Acc]) -> - bsm_opt_2(Is, [{test,bs_skip_bits2,F, - [Ctx,{integer,I1*Unit1+I2*Unit2},1,Flags]}|Acc]); -bsm_opt_2([I|Is], Acc) -> - bsm_opt_2(Is, [I|Acc]); -bsm_opt_2([], Acc) -> reverse(Acc). - -%% bsm_not_bs_test({test,Name,_,Operands}) -> true|false. -%% Test whether is the test is a "safe", i.e. does not move the -%% bit offset for a binary. -%% -%% 'true' means that the test is safe, 'false' that we don't know or -%% that the test moves the offset (e.g. bs_get_integer2). - -bsm_not_bs_test({test,bs_test_tail2,_,[_,_]}) -> true; -bsm_not_bs_test(Test) -> beam_utils:is_pure_test(Test). +cse_block([], _, Acc) -> + reverse(Acc). -bsm_subst_labels(Fs, Save, D) -> - bsm_subst_labels_1(Fs, Save, D, []). +%% cse_find(Expr, Expressions) -> error | Register. +%% Find a previously evaluated expression whose result can be reused, +%% or return 'error' if no such expression is found. -bsm_subst_labels_1([F|Fs], Save, D, Acc) -> - bsm_subst_labels_1(Fs, Save, D, [bsm_subst_label(F, Save, D)|Acc]); -bsm_subst_labels_1([], _, _, Acc) -> - reverse(Acc). +cse_find(Expr, Es) -> + case orddict:find(Expr, Es) of + {ok,{Src,_}} -> Src; + error -> error + end. -bsm_subst_label({f,Lbl0}=F, Save, D) -> - case gb_trees:lookup({Lbl0,Save}, D) of - {value,Lbl} -> {f,Lbl}; - none -> F +cse_expr({set,[D],Ss,{bif,N,_}}) -> + case D of + {fr,_} -> + %% There are too many things that can go wrong. + none; + _ -> + {ok,D,{{bif,N},Ss}} end; -bsm_subst_label(Other, _, _) -> Other. +cse_expr({set,[D],Ss,{alloc,_,{gc_bif,N,_}}}) -> + {ok,D,{{gc_bif,N},Ss}}; +cse_expr({set,[D],Ss,put_list}) -> + {ok,D,{put_list,Ss}}; +cse_expr(_) -> none. + +%% cse_clear(Instr, Expressions0) -> Expressions. +%% Remove all previous expressions that will become +%% invalid when this instruction is executed. Basically, +%% an expression is no longer safe to reuse when the +%% register it has been stored to has been modified, killed, +%% or if any of the source operands have changed. + +cse_clear({set,Ds,_,{alloc,Live,_}}, Es) -> + cse_clear_1(Es, Live, Ds); +cse_clear({set,Ds,_,_}, Es) -> + cse_clear_1(Es, all, Ds). + +cse_clear_1(Es, Live, Ds0) -> + Ds = ordsets:from_list(Ds0), + [E || E <- Es, cse_is_safe(E, Live, Ds)]. + +cse_is_safe({_,{Dst,Interfering}}, Live, Ds) -> + ordsets:is_disjoint(Interfering, Ds) andalso + case Dst of + {x,X} -> + X < Live; + _ -> + true + end. + +%% cse_add(Dest, Expr, Expressions0) -> Expressions. +%% Provided that it is safe, add a new expression to the dictionary +%% of already evaluated expressions. + +cse_add(D, {_,Ss}=Expr, Es) -> + case member(D, Ss) of + false -> + Interfering = ordsets:from_list([D|Ss]), + orddict:store(Expr, {D,Interfering}, Es); + true -> + %% Unsafe because the instruction overwrites one of + %% source operands. + Es + end. diff --git a/lib/compiler/src/beam_bool.erl b/lib/compiler/src/beam_bool.erl deleted file mode 100644 index d14be83496..0000000000 --- a/lib/compiler/src/beam_bool.erl +++ /dev/null @@ -1,765 +0,0 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 2004-2013. All Rights Reserved. -%% -%% Licensed under the Apache License, Version 2.0 (the "License"); -%% you may not use this file except in compliance with the License. -%% You may obtain a copy of the License at -%% -%% http://www.apache.org/licenses/LICENSE-2.0 -%% -%% Unless required by applicable law or agreed to in writing, software -%% distributed under the License is distributed on an "AS IS" BASIS, -%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -%% See the License for the specific language governing permissions and -%% limitations under the License. -%% -%% %CopyrightEnd% -%% -%% Purpose: Optimizes booleans in guards. - --module(beam_bool). - --export([module/2]). - --import(lists, [reverse/1,reverse/2,foldl/3,mapfoldl/3,map/2]). - --define(MAXREG, 1024). - --record(st, - {next, %Next label number. - ll %Live regs at labels. - }). - -module({Mod,Exp,Attr,Fs0,Lc}, _Opts) -> - %%io:format("~p:\n", [Mod]), - {Fs,_} = mapfoldl(fun(Fn, Lbl) -> function(Fn, Lbl) end, 100000000, Fs0), - {ok,{Mod,Exp,Attr,Fs,Lc}}. - -function({function,Name,Arity,CLabel,Is0}, Lbl0) -> - try - {Is,#st{next=Lbl}} = bool_opt(Is0, Lbl0), - {{function,Name,Arity,CLabel,Is},Lbl} - catch - Class:Error -> - Stack = erlang:get_stacktrace(), - io:fwrite("Function: ~w/~w\n", [Name,Arity]), - erlang:raise(Class, Error, Stack) - end. - -%% -%% Optimize boolean expressions that use guard bifs. Rewrite to -%% use test instructions if possible. -%% - -bool_opt(Asm, Lbl) -> - LiveInfo = beam_utils:index_labels(Asm), - bopt(Asm, [], #st{next=Lbl,ll=LiveInfo}). - -bopt([{block,Bl0}=Block| - [{jump,{f,Succ}}, - {label,Fail}, - {block,[{set,[Dst],[{atom,false}],move}]}, - {label,Succ}|Is]=Is0], Acc0, St) -> - case split_block(Bl0, Dst, Fail, Acc0, true) of - failed -> - bopt(Is0, [Block|Acc0], St); - {Bl,PreBlock} -> - Acc1 = case PreBlock of - [] -> Acc0; - _ -> [{block,PreBlock}|Acc0] - end, - Acc = [{protected,[Dst],Bl,{Fail,Succ}}|Acc1], - bopt(Is, Acc, St) - end; -bopt([{test,is_eq_exact,{f,Fail},[Reg,{atom,true}]}=I|Is], [{block,_}|_]=Acc0, St0) -> - case bopt_block(Reg, Fail, Is, Acc0, St0) of - failed -> bopt(Is, [I|Acc0], St0); - {Acc,St} -> bopt(Is, Acc, St) - end; -bopt([I|Is], Acc, St) -> - bopt(Is, [I|Acc], St); -bopt([], Acc, St) -> - {bopt_reverse(Acc, []),St}. - -bopt_reverse([{protected,[Dst],Block,{Fail,Succ}}|Is], Acc0) -> - Acc = [{block,Block},{jump,{f,Succ}}, - {label,Fail}, - {block,[{set,[Dst],[{atom,false}],move}]}, - {label,Succ}|Acc0], - bopt_reverse(Is, Acc); -bopt_reverse([I|Is], Acc) -> - bopt_reverse(Is, [I|Acc]); -bopt_reverse([], Acc) -> Acc. - -%% bopt_block(Reg, Fail, OldIs, Accumulator, St) -> failed | {NewAcc,St} -%% Attempt to optimized a block of guard BIFs followed by a test -%% instruction. -bopt_block(Reg, Fail, OldIs, [{block,Bl0}|Acc0], St0) -> - case split_block(Bl0, Reg, Fail, Acc0, false) of - failed -> - %% Reason for failure: The block either contained no - %% guard BIFs with the failure label Fail, or the final - %% instruction in the block did not assign the Reg register. - - %%io:format("split ~p: ~P\n", [Reg,Bl0,20]), - failed; - {Bl1,BlPre} -> - %% The block has been splitted. Bl1 is a non-empty list - %% of guard BIF instructions having the failure label Fail. - %% BlPre is a (possibly empty list) of instructions preceeding - %% Bl1. - Acc1 = make_block(BlPre, Acc0), - {Bl,Acc} = extend_block(Bl1, Fail, Acc1), - try - {NewCode,St} = bopt_tree_cg(Bl, Fail, St0), - ensure_opt_safe(Bl, NewCode, OldIs, Fail, Acc, St), - {NewCode++Acc,St} - catch - %% Not possible to rewrite because a boolean value is - %% passed to another guard bif, e.g. 'abs(A > B)' - %% (in this case, obviously nonsense code). Rare in - %% practice. - throw:mixed -> - failed; - - %% There was a reference to a boolean expression - %% from inside a protected block (try/catch), to - %% a boolean expression outside. - throw:protected_barrier -> - failed; - - %% The 'xor' operator was used. We currently don't - %% find it worthwile to translate 'xor' operators - %% (the code would be clumsy). - throw:'xor' -> - failed; - - %% The block does not contain a boolean expression, - %% but only a call to a guard BIF. - %% For instance: ... when element(1, T) -> - throw:not_boolean_expr -> - failed; - - %% The optimization is not safe. (A register - %% used by the instructions following the - %% optimized code is either not assigned a - %% value at all or assigned a different value.) - throw:all_registers_not_killed -> - failed; - throw:registers_used -> - failed; - - %% A protected block refered to the value - %% returned by another protected block, - %% probably because the Core Erlang code - %% used nested try/catches in the guard. - %% (v3_core never produces nested try/catches - %% in guards, so it must have been another - %% Core Erlang translator.) - throw:protected_violation -> - failed; - - %% Failed to work out the live registers for a GC - %% BIF. For example, if the number of live registers - %% needed to be 4 because {x,3} was a source register, - %% but {x,2} was not known to be initialized, this - %% exception would be thrown. - throw:gc_bif_alloc_failure -> - failed - - end - end. - -%% ensure_opt_safe(OriginalCode, OptCode, FollowingCode, Fail, -%% ReversedPrecedingCode, State) -> ok -%% Comparing the original code to the optimized code, determine -%% whether the optimized code is guaranteed to work in the same -%% way as the original code. -%% -%% Throw an exception if the optimization is not safe. -%% -ensure_opt_safe(Bl, NewCode, OldIs, Fail, PrecedingCode, St) -> - %% Here are the conditions that must be true for the - %% optimization to be safe. - %% - %% 1. If a register is INITIALIZED by PrecedingCode, - %% then if that register assigned a value in the original - %% code, but not in the optimized code, it must be UNUSED or KILLED - %% in the code that follows. - %% - %% 2. If a register is not known to be INITIALIZED by PreccedingCode, - %% then if that register assigned a value in the original - %% code, but not in the optimized code, it must be KILLED - %% by the code that follows. - %% - %% 3. Any register that is assigned a value in the optimized - %% code must be UNUSED or KILLED in the following code, - %% unless we can be sure that it is always assigned the same - %% value. - - InitInPreceding = initialized_regs(PrecedingCode), - - PrevDst = dst_regs(Bl), - NewDst = dst_regs(NewCode), - NotSet = ordsets:subtract(PrevDst, NewDst), - MustBeKilled = ordsets:subtract(NotSet, InitInPreceding), - - case all_killed(MustBeKilled, OldIs, Fail, St) of - false -> throw(all_registers_not_killed); - true -> ok - end, - MustBeUnused = ordsets:subtract(ordsets:union(NotSet, NewDst), - MustBeKilled), - case none_used(MustBeUnused, OldIs, Fail, St) of - false -> throw(registers_used); - true -> ok - end, - ok. - -update_fail_label([{set,Ds,As,{bif,N,{f,_}}}|Is], Fail, Acc) -> - update_fail_label(Is, Fail, [{set,Ds,As,{bif,N,{f,Fail}}}|Acc]); -update_fail_label([{set,Ds,As,{alloc,Regs,{gc_bif,N,{f,_}}}}|Is], Fail, Acc) -> - update_fail_label(Is, Fail, - [{set,Ds,As,{alloc,Regs,{gc_bif,N,{f,Fail}}}}|Acc]); -update_fail_label([], _, Acc) -> reverse(Acc). - -make_block(Bl) -> - make_block(Bl, []). - -make_block([], Acc) -> Acc; -make_block(Bl, Acc) -> [{block,Bl}|Acc]. - -extend_block(BlAcc, Fail, [{protected,_,_,_}=Prot|OldAcc]) -> - extend_block([Prot|BlAcc], Fail, OldAcc); -extend_block(BlAcc0, Fail, [{block,Is0}|OldAcc]) -> - case extend_block_1(reverse(Is0), Fail, BlAcc0) of - {BlAcc,[]} -> extend_block(BlAcc, Fail, OldAcc); - {BlAcc,Is} -> {BlAcc,[{block,Is}|OldAcc]} - end; -extend_block(BlAcc, _, OldAcc) -> {BlAcc,OldAcc}. - -extend_block_1([{set,[_],_,{bif,_,{f,Fail}}}=I|Is], Fail, Acc) -> - extend_block_1(Is, Fail, [I|Acc]); -extend_block_1([{set,[_],As,{bif,Bif,_}}=I|Is]=Is0, Fail, Acc) -> - case safe_bool_op(Bif, length(As)) of - false -> {Acc,reverse(Is0)}; - true -> extend_block_1(Is, Fail, [I|Acc]) - end; -extend_block_1([_|_]=Is, _, Acc) -> {Acc,reverse(Is)}; -extend_block_1([], _, Acc) -> {Acc,[]}. - -%% split_block([Instruction], Destination, FailLabel, [PreInstruction], -%% ProhibitFailLabelInPreBlock) -> failed | {Block,PreBlock} -%% Split a sequence of instructions into two blocks - one containing -%% all guard bif instructions and a pre-block all instructions before -%% the guard BIFs. - -split_block(Is0, Dst, Fail, PreIs, ProhibitFailLabel) -> - case ProhibitFailLabel andalso beam_jump:is_label_used_in(Fail, PreIs) of - true -> - %% The failure label was used in one of the instructions (most - %% probably bit syntax construction) preceeding the block, - %% the caller might eliminate the label. - failed; - false -> - case reverse(Is0) of - [{set,[Dst],_,_}|_]=Is -> - split_block_1(Is, Fail, ProhibitFailLabel); - _ -> failed - end - end. - -split_block_1(Is, Fail, ProhibitFailLabel) -> - case split_block_2(Is, Fail, []) of - {[],_} -> failed; - {_,PreBlock}=Res -> - case ProhibitFailLabel andalso - split_block_label_used(PreBlock, Fail) of - true -> - %% The failure label was used in the pre-block; - %% not allowed, because the label may be removed. - failed; - false -> - Res - end - end. - -split_block_2([{set,[_],_,{bif,_,{f,Fail}}}=I|Is], Fail, Acc) -> - split_block_2(Is, Fail, [I|Acc]); -split_block_2([{set,[_],_,{alloc,_,{gc_bif,_,{f,Fail}}}}=I|Is], Fail, Acc) -> - split_block_2(Is, Fail, [I|Acc]); -split_block_2(Is0, _, Acc) -> - Is = reverse(Is0), - {Acc,Is}. - -split_block_label_used([{set,[_],_,{bif,_,{f,Fail}}}|_], Fail) -> - true; -split_block_label_used([{set,[_],_,{alloc,_,{gc_bif,_,{f,Fail}}}}|_], Fail) -> - true; -split_block_label_used([{set,[_],_,{alloc,_,{put_map,_,{f,Fail}}}}|_], Fail) -> - true; -split_block_label_used([_|Is], Fail) -> - split_block_label_used(Is, Fail); -split_block_label_used([], _) -> false. - -dst_regs(Is) -> - dst_regs(Is, []). - -dst_regs([{block,Bl}|Is], Acc) -> - dst_regs(Bl, dst_regs(Is, Acc)); -dst_regs([{set,[D],_,{bif,_,{f,_}}}|Is], Acc) -> - dst_regs(Is, [D|Acc]); -dst_regs([{set,[D],_,{alloc,_,{gc_bif,_,{f,_}}}}|Is], Acc) -> - dst_regs(Is, [D|Acc]); -dst_regs([_|Is], Acc) -> - dst_regs(Is, Acc); -dst_regs([], Acc) -> ordsets:from_list(Acc). - -all_killed([R|Rs], OldIs, Fail, St) -> - case is_killed(R, OldIs, Fail, St) of - false -> false; - true -> all_killed(Rs, OldIs, Fail, St) - end; -all_killed([], _, _, _) -> true. - -none_used([R|Rs], OldIs, Fail, St) -> - case is_not_used(R, OldIs, Fail, St) of - false -> false; - true -> none_used(Rs, OldIs, Fail, St) - end; -none_used([], _, _, _) -> true. - -bopt_tree_cg(Block0, Fail, St) -> - Free = free_variables(Block0), - Block = ssa_block(Block0), -%% io:format("~p\n", [Block0]), -%% io:format("~p\n", [Block]), -%% io:format("~p\n", [gb_trees:to_list(Free)]), - case bopt_tree(Block, Free, []) of - {Pre0,[{_,Tree}]} -> - Pre1 = update_fail_label(Pre0, Fail, []), - Regs0 = init_regs(gb_trees:keys(Free)), -%% io:format("~p\n", [dst_regs(Block0)]), -%% io:format("~p\n", [Pre1]), -%% io:format("~p\n", [Tree]), -%% io:nl(), - {Pre,Regs} = rename_regs(Pre1, Regs0), -%% io:format("~p\n", [Regs0]), -%% io:format("~p\n", [Pre]), - bopt_cg(Tree, Fail, Regs, make_block(Pre), St); - _Res -> - throw(not_boolean_expr) - end. - -bopt_tree([{set,[Dst],As0,{bif,'not',_}}|Is], Forest0, Pre) -> - {[Arg],Forest1} = bopt_bool_args(As0, Forest0), - Forest = gb_trees:enter(Dst, {'not',Arg}, Forest1), - bopt_tree(Is, Forest, Pre); -bopt_tree([{set,[Dst],As0,{bif,'and',_}}|Is], Forest0, Pre) -> - {As,Forest1} = bopt_bool_args(As0, Forest0), - Node = make_and_node(As), - Forest = gb_trees:enter(Dst, Node, Forest1), - bopt_tree(Is, Forest, Pre); -bopt_tree([{set,[Dst],As0,{bif,'or',_}}|Is], Forest0, Pre) -> - {As,Forest1} = bopt_bool_args(As0, Forest0), - Node = make_or_node(As), - Forest = gb_trees:enter(Dst, Node, Forest1), - bopt_tree(Is, Forest, Pre); -bopt_tree([{set,_,_,{bif,'xor',_}}|_], _, _) -> - throw('xor'); -bopt_tree([{protected,[Dst],Code,_}|Is], Forest0, Pre) -> - ProtForest0 = gb_trees:from_orddict([P || {_,any}=P <- gb_trees:to_list(Forest0)]), - case bopt_tree(Code, ProtForest0, []) of - {ProtPre,[{_,ProtTree}]} -> - Prot = {prot,ProtPre,ProtTree}, - Forest = gb_trees:enter(Dst, Prot, Forest0), - bopt_tree(Is, Forest, Pre); - _Res -> - throw(not_boolean_expr) - end; -bopt_tree([{set,[Dst],As,{bif,N,_}}=Bif|Is], Forest0, Pre) -> - Ar = length(As), - case safe_bool_op(N, Ar) of - false -> - bopt_good_args(As, Forest0), - Forest = gb_trees:enter(Dst, any, Forest0), - bopt_tree(Is, Forest, [Bif|Pre]); - true -> - bopt_good_args(As, Forest0), - Test = bif_to_test(Dst, N, As), - Forest = gb_trees:enter(Dst, Test, Forest0), - bopt_tree(Is, Forest, Pre) - end; -bopt_tree([{set,[Dst],As,{alloc,_,{gc_bif,_,_}}}=Bif|Is], Forest0, Pre) -> - bopt_good_args(As, Forest0), - Forest = gb_trees:enter(Dst, any, Forest0), - bopt_tree(Is, Forest, [Bif|Pre]); -bopt_tree([], Forest, Pre) -> - {reverse(Pre),[R || {_,V}=R <- gb_trees:to_list(Forest), V =/= any]}. - -safe_bool_op(N, Ar) -> - erl_internal:new_type_test(N, Ar) orelse erl_internal:comp_op(N, Ar). - -bopt_bool_args([V0,V0], Forest0) -> - {V,Forest} = bopt_bool_arg(V0, Forest0), - {[V,V],Forest}; -bopt_bool_args(As, Forest) -> - mapfoldl(fun bopt_bool_arg/2, Forest, As). - -bopt_bool_arg({T,_}=R, Forest) when T =:= x; T =:= y; T =:= tmp -> - Val = case gb_trees:lookup(R, Forest) of - {value,any} -> {test,is_eq_exact,fail,[R,{atom,true}]}; - {value,Val0} -> Val0; - none -> throw(mixed) - end, - {Val,gb_trees:delete(R, Forest)}; -bopt_bool_arg(Term, Forest) -> - {Term,Forest}. - -bopt_good_args([A|As], Regs) -> - bopt_good_arg(A, Regs), - bopt_good_args(As, Regs); -bopt_good_args([], _) -> ok. - -bopt_good_arg({Tag,_}=X, Regs) when Tag =:= x; Tag =:= tmp -> - case gb_trees:lookup(X, Regs) of - {value,any} -> ok; - {value,_} -> throw(mixed); - none -> throw(protected_barrier) - end; -bopt_good_arg(_, _) -> ok. - -bif_to_test(_, N, As) -> - beam_utils:bif_to_test(N, As, fail). - -make_and_node(Is) -> - AndList0 = make_and_list(Is), - case simplify_and_list(AndList0) of - [] -> {atom,true}; - [Op] -> Op; - AndList -> {'and',AndList} - end. - -make_and_list([{'and',As}|Is]) -> - make_and_list(As++Is); -make_and_list([I|Is]) -> - [I|make_and_list(Is)]; -make_and_list([]) -> []. - -simplify_and_list([{atom,true}|T]) -> - simplify_and_list(T); -simplify_and_list([{atom,false}=False|_]) -> - [False]; -simplify_and_list([H|T]) -> - [H|simplify_and_list(T)]; -simplify_and_list([]) -> []. - -make_or_node(Is) -> - OrList0 = make_or_list(Is), - case simplify_or_list(OrList0) of - [] -> {atom,false}; - [Op] -> Op; - OrList -> {'or',OrList} - end. - -make_or_list([{'or',As}|Is]) -> - make_or_list(As++Is); -make_or_list([I|Is]) -> - [I|make_or_list(Is)]; -make_or_list([]) -> []. - -simplify_or_list([{atom,false}|T]) -> - simplify_or_list(T); -simplify_or_list([{atom,true}=True|_]) -> - [True]; -simplify_or_list([H|T]) -> - [H|simplify_or_list(T)]; -simplify_or_list([]) -> []. - -%% Code generation for a boolean tree. - -bopt_cg({'not',Arg}, Fail, Rs, Acc, St) -> - I = bopt_cg_not(Arg), - bopt_cg(I, Fail, Rs, Acc, St); -bopt_cg({'and',As}, Fail, Rs, Acc, St) -> - bopt_cg_and(As, Fail, Rs, Acc, St); -bopt_cg({'or',As}, Fail, Rs, Acc, St0) -> - {Succ,St} = new_label(St0), - bopt_cg_or(As, Succ, Fail, Rs, Acc, St); -bopt_cg({test,N,fail,As0}, Fail, Rs, Acc, St) -> - As = rename_sources(As0, Rs), - Test = {test,N,{f,Fail},As}, - {[Test|Acc],St}; -bopt_cg({inverted_test,N,fail,As0}, Fail, Rs, Acc, St0) -> - As = rename_sources(As0, Rs), - {Lbl,St} = new_label(St0), - {[{label,Lbl},{jump,{f,Fail}},{test,N,{f,Lbl},As}|Acc],St}; -bopt_cg({prot,Pre0,Tree}, Fail, Rs0, Acc, St0) -> - Pre1 = update_fail_label(Pre0, Fail, []), - {Pre,Rs} = rename_regs(Pre1, Rs0), - bopt_cg(Tree, Fail, Rs, make_block(Pre, Acc), St0); -bopt_cg({atom,true}, _Fail, _Rs, Acc, St) -> - {Acc,St}; -bopt_cg({atom,false}, Fail, _Rs, Acc, St) -> - {[{jump,{f,Fail}}|Acc],St}; -bopt_cg(_, _, _, _, _) -> - throw(not_boolean_expr). - -bopt_cg_not({'and',As0}) -> - As = [bopt_cg_not(A) || A <- As0], - {'or',As}; -bopt_cg_not({'or',As0}) -> - As = [bopt_cg_not(A) || A <- As0], - {'and',As}; -bopt_cg_not({'not',Arg}) -> - bopt_cg_not_not(Arg); -bopt_cg_not({test,Test,Fail,As}) -> - {inverted_test,Test,Fail,As}; -bopt_cg_not({atom,Bool}) when is_boolean(Bool) -> - {atom,not Bool}; -bopt_cg_not(_) -> - throw(not_boolean_expr). - -bopt_cg_not_not({'and',As}) -> - {'and',[bopt_cg_not_not(A) || A <- As]}; -bopt_cg_not_not({'or',As}) -> - {'or',[bopt_cg_not_not(A) || A <- As]}; -bopt_cg_not_not({'not',Arg}) -> - bopt_cg_not(Arg); -bopt_cg_not_not(Leaf) -> Leaf. - -bopt_cg_and([I|Is], Fail, Rs, Acc0, St0) -> - {Acc,St} = bopt_cg(I, Fail, Rs, Acc0, St0), - bopt_cg_and(Is, Fail, Rs, Acc, St); -bopt_cg_and([], _, _, Acc, St) -> {Acc,St}. - -bopt_cg_or([I], Succ, Fail, Rs, Acc0, St0) -> - {Acc,St} = bopt_cg(I, Fail, Rs, Acc0, St0), - {[{label,Succ}|Acc],St}; -bopt_cg_or([I|Is], Succ, Fail, Rs, Acc0, St0) -> - {Lbl,St1} = new_label(St0), - {Acc,St} = bopt_cg(I, Lbl, Rs, Acc0, St1), - bopt_cg_or(Is, Succ, Fail, Rs, [{label,Lbl},{jump,{f,Succ}}|Acc], St). - -new_label(#st{next=LabelNum}=St) when is_integer(LabelNum) -> - {LabelNum,St#st{next=LabelNum+1}}. - -free_variables(Is) -> - E = gb_sets:empty(), - free_vars_1(Is, E, E, E). - -free_vars_1([{set,Ds,As,{bif,_,_}}|Is], F0, N0, A) -> - F = gb_sets:union(F0, gb_sets:difference(var_list(As), N0)), - N = gb_sets:union(N0, var_list(Ds)), - free_vars_1(Is, F, N, A); -free_vars_1([{set,Ds,As,{alloc,Regs,{gc_bif,_,_}}}|Is], F0, N0, A0) -> - A = gb_sets:union(A0, gb_sets:from_list(free_vars_regs(Regs))), - F = gb_sets:union(F0, gb_sets:difference(var_list(As), N0)), - N = gb_sets:union(N0, var_list(Ds)), - free_vars_1(Is, F, N, A); -free_vars_1([{protected,_,Pa,_}|Is], F, N, A) -> - free_vars_1(Pa++Is, F, N, A); -free_vars_1([], F0, N, A) -> - F = case gb_sets:is_empty(A) of - true -> - %% No GC BIFs. - {x,X} = gb_sets:smallest(N), - P = ordsets:from_list(free_vars_regs(X)), - ordsets:union(gb_sets:to_list(F0), P); - false -> - %% At least one GC BIF. - gb_sets:to_list(gb_sets:union(F0, gb_sets:difference(A, N))) - end, - gb_trees:from_orddict([{K,any} || K <- F]). - -var_list(Is) -> - var_list_1(Is, gb_sets:empty()). - -var_list_1([{Tag,_}=X|Is], D) when Tag =:= x; Tag =:= y -> - var_list_1(Is, gb_sets:add(X, D)); -var_list_1([_|Is], D) -> - var_list_1(Is, D); -var_list_1([], D) -> D. - -free_vars_regs(0) -> []; -free_vars_regs(X) -> [{x,X-1}|free_vars_regs(X-1)]. - -rename_regs(Is, Regs) -> - rename_regs(Is, Regs, []). - -rename_regs([{set,[Dst0],Ss0,{alloc,_,Info}}|Is], Regs0, Acc) -> - Live = live_regs(Regs0), - Ss = rename_sources(Ss0, Regs0), - Regs = put_reg(Dst0, Regs0), - Dst = fetch_reg(Dst0, Regs), - rename_regs(Is, Regs, [{set,[Dst],Ss,{alloc,Live,Info}}|Acc]); -rename_regs([{set,[Dst0],Ss0,Info}|Is], Regs0, Acc) -> - Ss = rename_sources(Ss0, Regs0), - Regs = put_reg(Dst0, Regs0), - Dst = fetch_reg(Dst0, Regs), - rename_regs(Is, Regs, [{set,[Dst],Ss,Info}|Acc]); -rename_regs([], Regs, Acc) -> {reverse(Acc),Regs}. - -rename_sources(Ss, Regs) -> - map(fun({x,_}=R) -> fetch_reg(R, Regs); - ({tmp,_}=R) -> fetch_reg(R, Regs); - (E) -> E - end, Ss). - -%%% -%%% Keeping track of register assignments. -%%% - -init_regs(Free) -> - init_regs_1(Free, 0). - -init_regs_1([{x,I}=V|T], I) -> - [{I,V}|init_regs_1(T, I+1)]; -init_regs_1([{x,X}|_]=T, I) when I < X -> - [{I,reserved}|init_regs_1(T, I+1)]; -init_regs_1([{y,_}|_], _) -> []; -init_regs_1([], _) -> []. - -put_reg(V, Rs) -> put_reg_1(V, Rs, 0). - -put_reg_1(V, [R|Rs], I) -> [R|put_reg_1(V, Rs, I+1)]; -put_reg_1(V, [], I) -> [{I,V}]. - -fetch_reg(V, [{I,V}|_]) -> {x,I}; -fetch_reg(V, [_|SRs]) -> fetch_reg(V, SRs). - -live_regs([{_,reserved}|_]) -> - %% We are not sure that this register is initialized, so we must - %% abort the optimization. - throw(gc_bif_alloc_failure); -live_regs([{I,_}]) -> - I+1; -live_regs([{_,_}|Regs]) -> - live_regs(Regs); -live_regs([]) -> - 0. - - -%%% -%%% Convert a block to Static Single Assignment (SSA) form. -%%% - --record(ssa, - {live=0, %Variable counter. - sub=gb_trees:empty(), %Substitution table. - prot=gb_sets:empty(), %Targets assigned by protecteds. - in_prot=false %Inside a protected. - }). - -ssa_block(Is0) -> - {Is,_} = ssa_block_1(Is0, #ssa{}, []), - Is. - -ssa_block_1([{protected,[_],Pa0,Pb}|Is], Sub0, Acc) -> - {Pa,Sub1} = ssa_block_1(Pa0, Sub0#ssa{in_prot=true}, []), - Dst = ssa_last_target(Pa), - Sub = Sub1#ssa{prot=gb_sets:insert(Dst, Sub1#ssa.prot), - in_prot=Sub0#ssa.in_prot}, - ssa_block_1(Is, Sub, [{protected,[Dst],Pa,Pb}|Acc]); -ssa_block_1([{set,[Dst],As,Bif}|Is], Sub0, Acc0) -> - Sub1 = ssa_in_use_list(As, Sub0), - Sub = ssa_assign(Dst, Sub1), - Acc = [{set,[ssa_sub(Dst, Sub)],ssa_sub_list(As, Sub0),Bif}|Acc0], - ssa_block_1(Is, Sub, Acc); -ssa_block_1([], Sub, Acc) -> {reverse(Acc),Sub}. - -ssa_in_use_list(As, Sub) -> - foldl(fun ssa_in_use/2, Sub, As). - -ssa_in_use({x,_}=R, #ssa{sub=Sub0}=Ssa) -> - case gb_trees:is_defined(R, Sub0) of - true -> Ssa; - false -> - Sub = gb_trees:insert(R, R, Sub0), - Ssa#ssa{sub=Sub} - end; -ssa_in_use(_, Ssa) -> Ssa. - -ssa_assign({x,_}=R, #ssa{sub=Sub0}=Ssa0) -> - {NewReg,Ssa} = ssa_new_reg(Ssa0), - case gb_trees:is_defined(R, Sub0) of - false -> - Sub = gb_trees:insert(R, NewReg, Sub0), - Ssa#ssa{sub=Sub}; - true -> - Sub1 = gb_trees:update(R, NewReg, Sub0), - Sub = gb_trees:insert(NewReg, NewReg, Sub1), - Ssa#ssa{sub=Sub} - end. - -ssa_sub_list(List, Sub) -> - [ssa_sub(E, Sub) || E <- List]. - -ssa_sub(R0, #ssa{sub=Sub,prot=Prot,in_prot=InProt}) -> - case gb_trees:lookup(R0, Sub) of - none -> R0; - {value,R} -> - case InProt andalso gb_sets:is_element(R, Prot) of - true -> - throw(protected_violation); - false -> - R - end - end. - -ssa_new_reg(#ssa{live=Reg}=Ssa) -> - {{tmp,Reg},Ssa#ssa{live=Reg+1}}. - -ssa_last_target([{set,[Dst],_,_}]) -> Dst; -ssa_last_target([_|Is]) -> ssa_last_target(Is). - -%% is_killed(Register, [Instruction], FailLabel, State) -> true|false -%% Determine whether a register is killed in the instruction sequence. -%% The state is used to allow us to determine the kill state -%% across branches. - -is_killed(R, Is, Label, #st{ll=Ll}) -> - beam_utils:is_killed(R, Is, Ll) andalso - beam_utils:is_killed_at(R, Label, Ll). - -%% is_not_used(Register, [Instruction], FailLabel, State) -> true|false -%% Determine whether a register is never used in the instruction sequence -%% (it could still referenced by an allocate instruction, meaning that -%% it MUST be initialized). -%% The state is used to allow us to determine the usage state -%% across branches. - -is_not_used(R, Is, Label, #st{ll=Ll}) -> - beam_utils:is_not_used(R, Is, Ll) andalso - beam_utils:is_not_used_at(R, Label, Ll). - -%% initialized_regs([Instruction]) -> [Register]) -%% Given a REVERSED instruction sequence, return a list of the registers -%% that are guaranteed to be initialized (not contain garbage). - -initialized_regs(Is) -> - initialized_regs(Is, ordsets:new()). - -initialized_regs([{set,Dst,_Src,{alloc,Live,_}}|_], Regs0) -> - Regs = add_init_regs(free_vars_regs(Live), Regs0), - add_init_regs(Dst, Regs); -initialized_regs([{set,Dst,Src,_}|Is], Regs) -> - initialized_regs(Is, add_init_regs(Dst, add_init_regs(Src, Regs))); -initialized_regs([{test,_,_,Src}|Is], Regs) -> - initialized_regs(Is, add_init_regs(Src, Regs)); -initialized_regs([{block,Bl}|Is], Regs) -> - initialized_regs(reverse(Bl, Is), Regs); -initialized_regs([{bs_context_to_binary,Src}|Is], Regs) -> - initialized_regs(Is, add_init_regs([Src], Regs)); -initialized_regs([{label,_},{func_info,_,_,Arity}|_], Regs) -> - InitRegs = free_vars_regs(Arity), - add_init_regs(InitRegs, Regs); -initialized_regs([_|_], Regs) -> Regs. - -add_init_regs([{x,_}=X|T], Regs) -> - add_init_regs(T, ordsets:add_element(X, Regs)); -add_init_regs([_|T], Regs) -> - add_init_regs(T, Regs); -add_init_regs([], Regs) -> Regs. diff --git a/lib/compiler/src/beam_bs.erl b/lib/compiler/src/beam_bs.erl new file mode 100644 index 0000000000..5f1b9ed488 --- /dev/null +++ b/lib/compiler/src/beam_bs.erl @@ -0,0 +1,280 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% +%% Purpose : Partitions assembly instructions into basic blocks and +%% optimizes them. + +-module(beam_bs). + +-export([module/2]). +-import(lists, [mapfoldl/3,reverse/1]). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + +module({Mod,Exp,Attr,Fs0,Lc0}, _Opt) -> + {Fs,Lc} = mapfoldl(fun function/2, Lc0, Fs0), + {ok,{Mod,Exp,Attr,Fs,Lc}}. + +function({function,Name,Arity,CLabel,Is0}, Lc0) -> + try + Is1 = bs_put_opt(Is0), + {Is,Lc} = bsm_opt(Is1, Lc0), + {{function,Name,Arity,CLabel,Is},Lc} + catch + Class:Error:Stack -> + io:fwrite("Function: ~w/~w\n", [Name,Arity]), + erlang:raise(Class, Error, Stack) + end. + +%%% +%%% Evaluation of constant bit fields. +%%% + +bs_put_opt([{bs_put,_,_,_}=I|Is0]) -> + {BsPuts0,Is} = collect_bs_puts(Is0, [I]), + BsPuts = opt_bs_puts(BsPuts0), + BsPuts ++ bs_put_opt(Is); +bs_put_opt([I|Is]) -> + [I|bs_put_opt(Is)]; +bs_put_opt([]) -> []. + +collect_bs_puts([{bs_put,_,_,_}=I|Is], Acc) -> + collect_bs_puts(Is, [I|Acc]); +collect_bs_puts([_|_]=Is, Acc) -> + {reverse(Acc),Is}. + +opt_bs_puts(Is) -> + opt_bs_1(Is, []). + +opt_bs_1([{bs_put,Fail, + {bs_put_float,1,Flags0},[{integer,Sz},Src]}=I0|Is], Acc) -> + try eval_put_float(Src, Sz, Flags0) of + <<Int:Sz>> -> + Flags = force_big(Flags0), + I = {bs_put,Fail,{bs_put_integer,1,Flags}, + [{integer,Sz},{integer,Int}]}, + opt_bs_1([I|Is], Acc) + catch + error:_ -> + opt_bs_1(Is, [I0|Acc]) + end; +opt_bs_1([{bs_put,_,{bs_put_integer,1,_},[{integer,8},{integer,_}]}|_]=IsAll, + Acc0) -> + {Is,Acc} = bs_collect_string(IsAll, Acc0), + opt_bs_1(Is, Acc); +opt_bs_1([{bs_put,Fail,{bs_put_integer,1,F},[{integer,Sz},{integer,N}]}=I|Is0], + Acc) when Sz > 8 -> + case field_endian(F) of + big -> + %% We can do this optimization for any field size without + %% risk for code explosion. + case bs_split_int(N, Sz, Fail, Is0) of + no_split -> opt_bs_1(Is0, [I|Acc]); + Is -> opt_bs_1(Is, Acc) + end; + little when Sz < 128 -> + %% We only try to optimize relatively small fields, to + %% avoid an explosion in code size. + <<Int:Sz>> = <<N:Sz/little>>, + Flags = force_big(F), + Is = [{bs_put,Fail,{bs_put_integer,1,Flags}, + [{integer,Sz},{integer,Int}]}|Is0], + opt_bs_1(Is, Acc); + _ -> %native or too wide little field + opt_bs_1(Is0, [I|Acc]) + end; +opt_bs_1([{bs_put,Fail,{Op,U,F},[{integer,Sz},Src]}|Is], Acc) when U > 1 -> + opt_bs_1([{bs_put,Fail,{Op,1,F},[{integer,U*Sz},Src]}|Is], Acc); +opt_bs_1([I|Is], Acc) -> + opt_bs_1(Is, [I|Acc]); +opt_bs_1([], Acc) -> reverse(Acc). + +eval_put_float(Src, Sz, Flags) when Sz =< 256 -> + %%Only evaluate if Sz is reasonable. + Val = value(Src), + case field_endian(Flags) of + little -> <<Val:Sz/little-float-unit:1>>; + big -> <<Val:Sz/big-float-unit:1>> + %% native intentionally not handled here - we can't optimize + %% it. + end. + +value({integer,I}) -> I; +value({float,F}) -> F. + +bs_collect_string(Is, [{bs_put,_,{bs_put_string,Len,{string,Str}},[]}|Acc]) -> + bs_coll_str_1(Is, Len, reverse(Str), Acc); +bs_collect_string(Is, Acc) -> + bs_coll_str_1(Is, 0, [], Acc). + +bs_coll_str_1([{bs_put,_,{bs_put_integer,U,_},[{integer,Sz},{integer,V}]}|Is], + Len, StrAcc, IsAcc) when U*Sz =:= 8 -> + Byte = V band 16#FF, + bs_coll_str_1(Is, Len+1, [Byte|StrAcc], IsAcc); +bs_coll_str_1(Is, Len, StrAcc, IsAcc) -> + {Is,[{bs_put,{f,0},{bs_put_string,Len,{string,reverse(StrAcc)}},[]}|IsAcc]}. + +field_endian({field_flags,F}) -> field_endian_1(F). + +field_endian_1([big=E|_]) -> E; +field_endian_1([little=E|_]) -> E; +field_endian_1([native=E|_]) -> E; +field_endian_1([_|Fs]) -> field_endian_1(Fs). + +force_big({field_flags,F}) -> + {field_flags,force_big_1(F)}. + +force_big_1([big|_]=Fs) -> Fs; +force_big_1([little|Fs]) -> [big|Fs]; +force_big_1([F|Fs]) -> [F|force_big_1(Fs)]. + +bs_split_int(0, Sz, _, _) when Sz > 64 -> + %% We don't want to split in this case because the + %% string will consist of only zeroes. + no_split; +bs_split_int(-1, Sz, _, _) when Sz > 64 -> + %% We don't want to split in this case because the + %% string will consist of only 255 bytes. + no_split; +bs_split_int(N, Sz, Fail, Acc) -> + FirstByteSz = case Sz rem 8 of + 0 -> 8; + Rem -> Rem + end, + bs_split_int_1(N, FirstByteSz, Sz, Fail, Acc). + +bs_split_int_1(-1, _, Sz, Fail, Acc) when Sz > 64 -> + I = {bs_put,Fail,{bs_put_integer,1,{field_flags,[big]}}, + [{integer,Sz},{integer,-1}]}, + [I|Acc]; +bs_split_int_1(0, _, Sz, Fail, Acc) when Sz > 64 -> + I = {bs_put,Fail,{bs_put_integer,1,{field_flags,[big]}}, + [{integer,Sz},{integer,0}]}, + [I|Acc]; +bs_split_int_1(N, ByteSz, Sz, Fail, Acc) when Sz > 0 -> + Mask = (1 bsl ByteSz) - 1, + I = {bs_put,Fail,{bs_put_integer,1,{field_flags,[big]}}, + [{integer,ByteSz},{integer,N band Mask}]}, + bs_split_int_1(N bsr ByteSz, 8, Sz-ByteSz, Fail, [I|Acc]); +bs_split_int_1(_, _, _, _, Acc) -> Acc. + +%%% +%%% Optimization of bit syntax matching: get rid +%%% of redundant bs_restore2/2 instructions across select_val +%%% instructions, as well as a few other simple peep-hole +%%% optimizations. +%%% + +bsm_opt(Is0, Lc0) -> + {Is1,D0,Lc} = bsm_scan(Is0, [], Lc0, []), + Is2 = case D0 of + [] -> + %% No bit syntax matching in this function. + Is1; + [_|_] -> + %% Optimize the bit syntax matching. + D = gb_trees:from_orddict(orddict:from_list(D0)), + bsm_reroute(Is1, D, none, []) + end, + Is = beam_clean:bs_clean_saves(Is2), + {bsm_opt_2(Is, []),Lc}. + +bsm_scan([{label,L}=Lbl,{bs_restore2,_,Save}=R|Is], D0, Lc, Acc0) -> + D = [{{L,Save},Lc}|D0], + Acc = [{label,Lc},R,Lbl|Acc0], + bsm_scan(Is, D, Lc+1, Acc); +bsm_scan([I|Is], D, Lc, Acc) -> + bsm_scan(Is, D, Lc, [I|Acc]); +bsm_scan([], D, Lc, Acc) -> + {reverse(Acc),D,Lc}. + +bsm_reroute([{bs_save2,Reg,Save}=I|Is], D, _, Acc) -> + bsm_reroute(Is, D, {Reg,Save}, [I|Acc]); +bsm_reroute([{bs_restore2,Reg,Save}=I|Is], D, _, Acc) -> + bsm_reroute(Is, D, {Reg,Save}, [I|Acc]); +bsm_reroute([{label,_}=I|Is], D, S, Acc) -> + bsm_reroute(Is, D, S, [I|Acc]); +bsm_reroute([{select,select_val,Reg,F0,Lbls0}|Is], D, {_,Save}=S, Acc0) -> + [F|Lbls] = bsm_subst_labels([F0|Lbls0], Save, D), + Acc = [{select,select_val,Reg,F,Lbls}|Acc0], + bsm_reroute(Is, D, S, Acc); +bsm_reroute([{test,TestOp,F0,TestArgs}=I|Is], D, {_,Save}=S, Acc0) -> + F = bsm_subst_label(F0, Save, D), + Acc = [{test,TestOp,F,TestArgs}|Acc0], + case bsm_not_bs_test(I) of + true -> + %% The test instruction will not update the bit offset for + %% the binary being matched. Therefore the save position + %% can be kept. + bsm_reroute(Is, D, S, Acc); + false -> + %% The test instruction might update the bit offset. Kill + %% our remembered Save position. + bsm_reroute(Is, D, none, Acc) + end; +bsm_reroute([{test,TestOp,F0,Live,TestArgs,Dst}|Is], D, {_,Save}, Acc0) -> + F = bsm_subst_label(F0, Save, D), + Acc = [{test,TestOp,F,Live,TestArgs,Dst}|Acc0], + %% The test instruction will update the bit offset. Kill our + %% remembered Save position. + bsm_reroute(Is, D, none, Acc); +bsm_reroute([{block,[{set,[],[],{alloc,_,_}}]}=Bl, + {bs_context_to_binary,_}=I|Is], D, S, Acc) -> + %% To help further bit syntax optimizations. + bsm_reroute([I,Bl|Is], D, S, Acc); +bsm_reroute([I|Is], D, _, Acc) -> + bsm_reroute(Is, D, none, [I|Acc]); +bsm_reroute([], _, _, Acc) -> reverse(Acc). + +bsm_opt_2([{test,bs_test_tail2,F,[Ctx,Bits]}|Is], + [{test,bs_skip_bits2,F,[Ctx,{integer,I},Unit,_Flags]}|Acc]) -> + bsm_opt_2(Is, [{test,bs_test_tail2,F,[Ctx,Bits+I*Unit]}|Acc]); +bsm_opt_2([{test,bs_skip_bits2,F,[Ctx,{integer,I1},Unit1,_]}|Is], + [{test,bs_skip_bits2,F,[Ctx,{integer,I2},Unit2,Flags]}|Acc]) -> + bsm_opt_2(Is, [{test,bs_skip_bits2,F, + [Ctx,{integer,I1*Unit1+I2*Unit2},1,Flags]}|Acc]); +bsm_opt_2([I|Is], Acc) -> + bsm_opt_2(Is, [I|Acc]); +bsm_opt_2([], Acc) -> reverse(Acc). + +%% bsm_not_bs_test({test,Name,_,Operands}) -> true|false. +%% Test whether is the test is a "safe", i.e. does not move the +%% bit offset for a binary. +%% +%% 'true' means that the test is safe, 'false' that we don't know or +%% that the test moves the offset (e.g. bs_get_integer2). + +bsm_not_bs_test({test,bs_test_tail2,_,[_,_]}) -> true; +bsm_not_bs_test(Test) -> beam_utils:is_pure_test(Test). + +bsm_subst_labels(Fs, Save, D) -> + bsm_subst_labels_1(Fs, Save, D, []). + +bsm_subst_labels_1([F|Fs], Save, D, Acc) -> + bsm_subst_labels_1(Fs, Save, D, [bsm_subst_label(F, Save, D)|Acc]); +bsm_subst_labels_1([], _, _, Acc) -> + reverse(Acc). + +bsm_subst_label({f,Lbl0}=F, Save, D) -> + case gb_trees:lookup({Lbl0,Save}, D) of + {value,Lbl} -> {f,Lbl}; + none -> F + end; +bsm_subst_label(Other, _, _) -> Other. diff --git a/lib/compiler/src/beam_bsm.erl b/lib/compiler/src/beam_bsm.erl index 62356928ae..1c8e0e9854 100644 --- a/lib/compiler/src/beam_bsm.erl +++ b/lib/compiler/src/beam_bsm.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2007-2015. All Rights Reserved. +%% Copyright Ericsson AB 2007-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -60,19 +60,26 @@ %%% data structures or passed to BIFs. %%% +-type label() :: beam_asm:label(). +-type func_info() :: {beam_asm:reg(),boolean()}. + -record(btb, - {f, %Gbtrees for all functions. - index, %{Label,Code} index (for liveness). - ok_br, %Labels that are OK. - must_not_save, %Must not save position when - % optimizing (reaches - % bs_context_to_binary). - must_save %Must save position when optimizing. + {f :: gb_trees:tree(label(), func_info()), + index :: beam_utils:code_index(), %{Label,Code} index (for liveness). + ok_br=gb_sets:empty() :: gb_sets:set(label()), %Labels that are OK. + must_not_save=false :: boolean(), %Must not save position when + % optimizing (reaches + % bs_context_to_binary). + must_save=false :: boolean() %Must save position when optimizing. }). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,Lc}, Opts) -> - D = #btb{f=btb_index(Fs0)}, - Fs = [function(F, D) || F <- Fs0], + FIndex = btb_index(Fs0), + Fs = [function(F, FIndex) || F <- Fs0], Code = {Mod,Exp,Attr,Fs,Lc}, case proplists:get_bool(bin_opt_info, Opts) of true -> @@ -92,14 +99,13 @@ format_error({no_bin_opt,Reason}) -> %%% Local functions. %%% -function({function,Name,Arity,Entry,Is}, D0) -> +function({function,Name,Arity,Entry,Is}, FIndex) -> try Index = beam_utils:index_labels(Is), - D = D0#btb{index=Index}, + D = #btb{f=FIndex,index=Index}, {function,Name,Arity,Entry,btb_opt_1(Is, D, [])} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -118,20 +124,21 @@ btb_opt_1([{test,bs_get_binary2,F,_,[Reg,{atom,all},U,Fs],Reg}=I0|Is], D, Acc0) end, btb_opt_1(Is, D, Acc) end; -btb_opt_1([{test,bs_get_binary2,F,_,[Ctx,{atom,all},U,Fs],Dst}=I0|Is], D, Acc0) -> - case btb_reaches_match(Is, [Ctx,Dst], D) of +btb_opt_1([{test,bs_get_binary2,F,_,[Ctx,{atom,all},U,Fs],Dst}=I0|Is0], D, Acc0) -> + case btb_reaches_match(Is0, [Ctx,Dst], D) of {error,Reason} -> Comment = btb_comment_no_opt(Reason, Fs), - btb_opt_1(Is, D, [Comment,I0|Acc0]); + btb_opt_1(Is0, D, [Comment,I0|Acc0]); {ok,MustSave} when U =:= 1 -> Comment = btb_comment_opt(Fs), - Acc1 = btb_gen_save(MustSave, Ctx, [Comment|Acc0]), - Acc = [{move,Ctx,Dst}|Acc1], + Acc = btb_gen_save(MustSave, Ctx, [Comment|Acc0]), + Is = prepend_move(Ctx, Dst, Is0), btb_opt_1(Is, D, Acc); {ok,MustSave} -> Comment = btb_comment_opt(Fs), Acc1 = btb_gen_save(MustSave, Ctx, [Comment|Acc0]), - Acc = [{move,Ctx,Dst},{test,bs_test_unit,F,[Ctx,U]}|Acc1], + Acc = [{test,bs_test_unit,F,[Ctx,U]}|Acc1], + Is = prepend_move(Ctx, Dst, Is0), btb_opt_1(Is, D, Acc) end; btb_opt_1([I|Is], D, Acc) -> @@ -144,6 +151,12 @@ btb_gen_save(true, Reg, Acc) -> [{bs_save2,Reg,{atom,start}}|Acc]; btb_gen_save(false, _, Acc) -> Acc. +prepend_move(Ctx, Dst, [{block,Bl0}|Is]) -> + Bl = [{set,[Dst],[Ctx],move}|Bl0], + [{block,Bl}|Is]; +prepend_move(Ctx, Dst, Is) -> + [{move,Ctx,Dst}|Is]. + %% btb_reaches_match([Instruction], [Register], D) -> %% {ok,MustSave}|{error,Reason} %% @@ -179,15 +192,14 @@ btb_gen_save(false, _, Acc) -> Acc. %% a bs_context_to_binary instruction. %% -btb_reaches_match(Is, RegList, D0) -> +btb_reaches_match(Is, RegList, D) -> try Regs = btb_regs_from_list(RegList), - D = D0#btb{ok_br=gb_sets:empty(),must_not_save=false,must_save=false}, #btb{must_not_save=MustNotSave,must_save=MustSave} = - btb_reaches_match_1(Is, Regs, D), - case MustNotSave and MustSave of + btb_reaches_match_1(Is, Regs, D), + case MustNotSave andalso MustSave of true -> btb_error(must_and_must_not_save); - _ -> {ok,MustSave} + false -> {ok,MustSave} end catch throw:{error,_}=Error -> Error @@ -205,8 +217,15 @@ btb_reaches_match_1(Is, Regs, D) -> btb_reaches_match_2([{block,Bl}|Is], Regs0, D) -> Regs = btb_reaches_match_block(Bl, Regs0), btb_reaches_match_1(Is, Regs, D); -btb_reaches_match_2([{call,Arity,{f,Lbl}}|Is], Regs, D) -> - btb_call(Arity, Lbl, Regs, Is, D); +btb_reaches_match_2([{call,Arity,{f,Lbl}}|Is], Regs0, D) -> + case is_tail_call(Is) of + true -> + Regs1 = btb_kill_not_live(Arity, Regs0), + Regs = btb_kill_yregs(Regs1), + btb_tail_call(Lbl, Regs, D); + false -> + btb_call(Arity, Lbl, Regs0, Is, D) + end; btb_reaches_match_2([{apply,Arity}|Is], Regs, D) -> btb_call(Arity+2, apply, Regs, Is, D); btb_reaches_match_2([{call_fun,Live}=I|Is], Regs, D) -> @@ -360,6 +379,10 @@ btb_reaches_match_2([{line,_}|Is], Regs, D) -> btb_reaches_match_2([I|_], Regs, _) -> btb_error({btb_context_regs(Regs),I,not_handled}). +is_tail_call([{deallocate,_}|_]) -> true; +is_tail_call([return|_]) -> true; +is_tail_call(_) -> false. + btb_call(Arity, Lbl, Regs0, Is, D0) -> Regs = btb_kill_not_live(Arity, Regs0), case btb_are_x_registers_empty(Regs) of @@ -369,15 +392,15 @@ btb_call(Arity, Lbl, Regs0, Is, D0) -> D = btb_tail_call(Lbl, Regs, D0), %% No problem so far (the called function can handle a - %% match context). Now we must make sure that the rest - %% of this function following the call does not attempt - %% to use the match context in case there is a copy - %% tucked away in a y register. + %% match context). Now we must make sure that we don't + %% have any copies of the match context tucked away in an + %% y register. RegList = btb_context_regs(Regs), - YRegs = [R || {y,_}=R <- RegList], - case btb_are_all_unused(YRegs, Is, D) of - true -> D; - false -> btb_error({multiple_uses,RegList}) + case [R || {y,_}=R <- RegList] of + [] -> + D; + [_|_] -> + btb_error({multiple_uses,RegList}) end; true -> %% No match context in any x register. It could have been diff --git a/lib/compiler/src/beam_clean.erl b/lib/compiler/src/beam_clean.erl index 919ee3ee7d..207f1c4deb 100644 --- a/lib/compiler/src/beam_clean.erl +++ b/lib/compiler/src/beam_clean.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2000-2013. All Rights Reserved. +%% Copyright Ericsson AB 2000-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -24,7 +24,10 @@ -export([module/2]). -export([bs_clean_saves/1]). -export([clean_labels/1]). --import(lists, [map/2,foldl/3,reverse/1,filter/2]). +-import(lists, [foldl/3,reverse/1]). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. module({Mod,Exp,Attr,Fs0,_}, Opts) -> Order = [Lbl || {function,_,_,Lbl,_} <- Fs0], @@ -39,6 +42,10 @@ module({Mod,Exp,Attr,Fs0,_}, Opts) -> {ok,{Mod,Exp,Attr,Fs,Lc}}. %% Remove all bs_save2/2 instructions not referenced by a bs_restore2/2. + +-spec bs_clean_saves([beam_utils:instruction()]) -> + [beam_utils:instruction()]. + bs_clean_saves(Is) -> Needed = bs_restores(Is, []), bs_clean_saves_1(Is, gb_sets:from_list(Needed), []). @@ -98,15 +105,20 @@ add_to_work_list(F, {Fs,Used}=Sets) -> %%% want to see the expanded code in a .S file. %%% --record(st, {lmap, %Translation tables for labels. - entry, %Number of entry label. - lc %Label counter +-type label() :: beam_asm:label(). + +-record(st, {lmap :: [{label(),label()}], %Translation tables for labels. + entry :: beam_asm:label(), %Number of entry label. + lc :: non_neg_integer() %Label counter }). +-spec clean_labels([beam_utils:instruction()]) -> + {[beam_utils:instruction()],pos_integer()}. + clean_labels(Fs0) -> - St0 = #st{lmap=[],lc=1}, + St0 = #st{lmap=[],entry=1,lc=1}, {Fs1,#st{lmap=Lmap0,lc=Lc}} = function_renumber(Fs0, St0, []), - Lmap = gb_trees:from_orddict(ordsets:from_list(Lmap0)), + Lmap = maps:from_list(Lmap0), Fs = function_replace(Fs1, Lmap, []), {Fs,Lc}. @@ -141,7 +153,7 @@ renumber_labels([{bif,is_record,{f,_}, renumber_labels(Is, Acc, St); renumber_labels([{test,is_record,{f,_}=Fail, [Term,{atom,Tag}=TagAtom,{integer,Arity}]}|Is0], Acc, St) -> - Tmp = {x,1023}, + Tmp = {x,1022}, Is = case is_record_tuple(Term, Tag, Arity) of yes -> Is0; @@ -175,7 +187,8 @@ is_record_tuple(_, _, _) -> no. function_replace([{function,Name,Arity,Entry,Asm0}|Fs], Dict, Acc) -> Asm = try - replace(Asm0, [], Dict) + Fb = fun(Old) -> throw({error,{undefined_label,Old}}) end, + beam_utils:replace_labels(Asm0, [], Dict, Fb) catch throw:{error,{undefined_label,Lbl}=Reason} -> io:format("Function ~s/~w refers to undefined label ~w\n", @@ -185,69 +198,6 @@ function_replace([{function,Name,Arity,Entry,Asm0}|Fs], Dict, Acc) -> function_replace(Fs, Dict, [{function,Name,Arity,Entry,Asm}|Acc]); function_replace([], _, Acc) -> Acc. -replace([{test,Test,{f,Lbl},Ops}|Is], Acc, D) -> - replace(Is, [{test,Test,{f,label(Lbl, D)},Ops}|Acc], D); -replace([{test,Test,{f,Lbl},Live,Ops,Dst}|Is], Acc, D) -> - replace(Is, [{test,Test,{f,label(Lbl, D)},Live,Ops,Dst}|Acc], D); -replace([{select,I,R,{f,Fail0},Vls0}|Is], Acc, D) -> - Vls1 = map(fun ({f,L}) -> {f,label(L, D)}; - (Other) -> Other end, Vls0), - Fail = label(Fail0, D), - case redundant_values(Vls1, Fail, []) of - [] -> - %% Oops, no choices left. The loader will not accept that. - %% Convert to a plain jump. - replace(Is, [{jump,{f,Fail}}|Acc], D); - Vls -> - replace(Is, [{select,I,R,{f,Fail},Vls}|Acc], D) - end; -replace([{'try',R,{f,Lbl}}|Is], Acc, D) -> - replace(Is, [{'try',R,{f,label(Lbl, D)}}|Acc], D); -replace([{'catch',R,{f,Lbl}}|Is], Acc, D) -> - replace(Is, [{'catch',R,{f,label(Lbl, D)}}|Acc], D); -replace([{jump,{f,Lbl}}|Is], Acc, D) -> - replace(Is, [{jump,{f,label(Lbl, D)}}|Acc], D); -replace([{loop_rec,{f,Lbl},R}|Is], Acc, D) -> - replace(Is, [{loop_rec,{f,label(Lbl, D)},R}|Acc], D); -replace([{loop_rec_end,{f,Lbl}}|Is], Acc, D) -> - replace(Is, [{loop_rec_end,{f,label(Lbl, D)}}|Acc], D); -replace([{wait,{f,Lbl}}|Is], Acc, D) -> - replace(Is, [{wait,{f,label(Lbl, D)}}|Acc], D); -replace([{wait_timeout,{f,Lbl},To}|Is], Acc, D) -> - replace(Is, [{wait_timeout,{f,label(Lbl, D)},To}|Acc], D); -replace([{bif,Name,{f,Lbl},As,R}|Is], Acc, D) when Lbl =/= 0 -> - replace(Is, [{bif,Name,{f,label(Lbl, D)},As,R}|Acc], D); -replace([{gc_bif,Name,{f,Lbl},Live,As,R}|Is], Acc, D) when Lbl =/= 0 -> - replace(Is, [{gc_bif,Name,{f,label(Lbl, D)},Live,As,R}|Acc], D); -replace([{call,Ar,{f,Lbl}}|Is], Acc, D) -> - replace(Is, [{call,Ar,{f,label(Lbl,D)}}|Acc], D); -replace([{make_fun2,{f,Lbl},U1,U2,U3}|Is], Acc, D) -> - replace(Is, [{make_fun2,{f,label(Lbl, D)},U1,U2,U3}|Acc], D); -replace([{bs_init,{f,Lbl},Info,Live,Ss,Dst}|Is], Acc, D) when Lbl =/= 0 -> - replace(Is, [{bs_init,{f,label(Lbl, D)},Info,Live,Ss,Dst}|Acc], D); -replace([{bs_put,{f,Lbl},Info,Ss}|Is], Acc, D) when Lbl =/= 0 -> - replace(Is, [{bs_put,{f,label(Lbl, D)},Info,Ss}|Acc], D); -replace([{put_map=I,{f,Lbl},Op,Src,Dst,Live,List}|Is], Acc, D) - when Lbl =/= 0 -> - replace(Is, [{I,{f,label(Lbl, D)},Op,Src,Dst,Live,List}|Acc], D); -replace([{get_map_elements=I,{f,Lbl},Src,List}|Is], Acc, D) when Lbl =/= 0 -> - replace(Is, [{I,{f,label(Lbl, D)},Src,List}|Acc], D); -replace([I|Is], Acc, D) -> - replace(Is, [I|Acc], D); -replace([], Acc, _) -> Acc. - -label(Old, D) -> - case gb_trees:lookup(Old, D) of - {value,Val} -> Val; - none -> throw({error,{undefined_label,Old}}) - end. - -redundant_values([_,{f,Fail}|Vls], Fail, Acc) -> - redundant_values(Vls, Fail, Acc); -redundant_values([Val,Lbl|Vls], Fail, Acc) -> - redundant_values(Vls, Fail, [Lbl,Val|Acc]); -redundant_values([], _, Acc) -> reverse(Acc). - %%% %%% Final fixup of bs_start_match2/5,bs_save2/bs_restore2 instructions for %%% new bit syntax matching (introduced in R11B). @@ -304,7 +254,7 @@ bs_restores([_|Is], Dict) -> bs_restores([], Dict) -> Dict. %% Pass 2. -bs_replace([{test,bs_start_match2,F,Live,[Src,Ctx],CtxR}|T], Dict, Acc) when is_atom(Ctx) -> +bs_replace([{test,bs_start_match2,F,Live,[Src,{context,Ctx}],CtxR}|T], Dict, Acc) -> Slots = case gb_trees:lookup(Ctx, Dict) of {value,Slots0} -> Slots0; none -> 0 @@ -353,8 +303,21 @@ maybe_remove_lines(Fs, Opts) -> end. remove_lines([{function,N,A,Lbl,Is0}|T]) -> - Is = filter(fun({line,_}) -> false; - (_) -> true - end, Is0), + Is = remove_lines_fun(Is0), [{function,N,A,Lbl,Is}|remove_lines(T)]; remove_lines([]) -> []. + +remove_lines_fun([{line,_}|Is]) -> + remove_lines_fun(Is); +remove_lines_fun([{block,Bl0}|Is]) -> + Bl = remove_lines_block(Bl0), + [{block,Bl}|remove_lines_fun(Is)]; +remove_lines_fun([I|Is]) -> + [I|remove_lines_fun(Is)]; +remove_lines_fun([]) -> []. + +remove_lines_block([{set,_,_,{line,_}}|Is]) -> + remove_lines_block(Is); +remove_lines_block([I|Is]) -> + [I|remove_lines_block(Is)]; +remove_lines_block([]) -> []. diff --git a/lib/compiler/src/beam_dead.erl b/lib/compiler/src/beam_dead.erl index ead88b57e9..efad082152 100644 --- a/lib/compiler/src/beam_dead.erl +++ b/lib/compiler/src/beam_dead.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2002-2013. All Rights Reserved. +%% Copyright Ericsson AB 2002-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -29,6 +29,10 @@ -import(lists, [mapfoldl/3,reverse/1]). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,_}, _Opts) -> {Fs1,Lc1} = beam_clean:clean_labels(Fs0), {Fs,Lc} = mapfoldl(fun function/2, Lc1, Fs1), @@ -52,8 +56,7 @@ function({function,Name,Arity,CLabel,Is0}, Lc0) -> Is = move_move_into_block(Is3, []), {{function,Name,Arity,CLabel,Is},Lc} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -239,38 +242,77 @@ backward([{test,is_eq_exact,Fail,[Dst,{integer,Arity}]}=I| backward([{label,Lbl}=L|Is], D, Acc) -> backward(Is, beam_utils:index_label(Lbl, Acc, D), [L|Acc]); backward([{select,select_val,Reg,{f,Fail0},List0}|Is], D, Acc) -> - List = shortcut_select_list(List0, Reg, D, []), + List1 = shortcut_select_list(List0, Reg, D, []), Fail1 = shortcut_label(Fail0, D), Fail = shortcut_bs_test(Fail1, Is, D), - Sel = {select,select_val,Reg,{f,Fail},List}, - backward(Is, D, [Sel|Acc]); + List = prune_redundant(List1, Fail), + case List of + [] -> + Jump = {jump,{f,Fail}}, + backward([Jump|Is], D, Acc); + [V,F] -> + Test = {test,is_eq_exact,{f,Fail},[Reg,V]}, + Jump = {jump,F}, + backward([Jump,Test|Is], D, Acc); + [{atom,B1},F,{atom,B2},F] when B1 =:= not B2 -> + Test = {test,is_boolean,{f,Fail},[Reg]}, + Jump = {jump,F}, + backward([Jump,Test|Is], D, Acc); + [_|_] -> + Sel = {select,select_val,Reg,{f,Fail},List}, + backward(Is, D, [Sel|Acc]) + end; backward([{jump,{f,To0}},{move,Src,Reg}=Move|Is], D, Acc) -> To = shortcut_select_label(To0, Reg, Src, D), Jump = {jump,{f,To}}, - case beam_utils:is_killed_at(Reg, To, D) of + case is_killed_at(Reg, To, D) of false -> backward([Move|Is], D, [Jump|Acc]); true -> backward([Jump|Is], D, Acc) end; -backward([{jump,{f,To}}=J|[{bif,Op,_,Ops,Reg}|Is]=Is0], D, Acc) -> +backward([{jump,{f,To}}=J|[{bif,Op,{f,BifFail},Ops,Reg}|Is]=Is0], D, Acc) -> try replace_comp_op(To, Reg, Op, Ops, D) of - I -> backward(Is, D, I++Acc) + {Test,Jump} -> + backward([Jump,Test|Is], D, Acc) catch - throw:not_possible -> backward(Is0, D, [J|Acc]) + throw:not_possible -> + case To =:= BifFail of + true -> + %% The bif instruction is redundant. See the comment + %% in the next clause for why there is no need to + %% test for liveness of Reg at label To. + backward([J|Is], D, Acc); + false -> + backward(Is0, D, [J|Acc]) + end end; -backward([{test,bs_start_match2,F,_,[R,_],Ctxt}=I|Is], D, - [{test,bs_match_string,F,[Ctxt,Bs]}, - {test,bs_test_tail2,F,[Ctxt,0]}|Acc0]=Acc) -> - case beam_utils:is_killed(Ctxt, Acc0, D) of - true -> - Eq = {test,is_eq_exact,F,[R,{literal,Bs}]}, - backward(Is, D, [Eq|Acc0]); - false -> - backward(Is, D, [I|Acc]) +backward([{jump,{f,To}}=J|[{gc_bif,_,{f,To},_,_,_Dst}|Is]], D, Acc) -> + %% The gc_bif instruction is redundant, since either the gc_bif + %% instruction itself or the jump instruction will transfer control + %% to label To. Note that a gc_bif instruction does not assign its + %% destination register if the failure branch is taken; therefore, + %% the code at label To is not allowed to assume that the destination + %% register is initialized, and it is therefore no need to test + %% for liveness of the destination register at label To. + backward([J|Is], D, Acc); +backward([{test,bs_start_match2,F,Live,[Src,_]=Args,Ctxt}|Is], D, Acc0) -> + {f,To0} = F, + case test_bs_literal(F, Ctxt, D, Acc0) of + {none,Acc} -> + %% Ctxt killed immediately after bs_start_match2. + To = shortcut_bs_context_to_binary(To0, Src, D), + I = {test,is_bitstr,{f,To},[Src]}, + backward(Is, D, [I|Acc]); + {Literal,Acc} -> + %% Ctxt killed after matching a literal. + To = shortcut_bs_context_to_binary(To0, Src, D), + Eq = {test,is_eq_exact,{f,To},[Src,{literal,Literal}]}, + backward(Is, D, [Eq|Acc]); + not_killed -> + %% Ctxt not killed. Not much to do. + To = shortcut_bs_start_match(To0, Src, D), + I = {test,bs_start_match2,{f,To},Live,Args,Ctxt}, + backward(Is, D, [I|Acc0]) end; -backward([{test,bs_start_match2,{f,To0},Live,[Src|_]=Info,Dst}|Is], D, Acc) -> - To = shortcut_bs_start_match(To0, Src, D), - I = {test,bs_start_match2,{f,To},Live,Info,Dst}, - backward(Is, D, [I|Acc]); backward([{test,Op,{f,To0},Ops0}|Is], D, Acc) -> To1 = shortcut_bs_test(To0, Is, D), To2 = shortcut_label(To1, D), @@ -295,7 +337,28 @@ backward([{test,Op,{f,To0},Ops0}|Is], D, Acc) -> is_eq_exact -> combine_eqs(To, Ops0, D, Acc); _ -> {test,Op,{f,To},Ops0} end, - backward(Is, D, [I|Acc]); + case {I,Acc} of + {{test,is_atom,Fail,Ops0},[{test,is_boolean,Fail,Ops0}|_]} -> + %% An is_atom test before an is_boolean test (with the + %% same failure label) is redundant. + backward(Is, D, Acc); + {{test,is_atom,Fail,[R]}, + [{test,is_eq_exact,Fail,[R,{atom,_}]}|_]} -> + %% An is_atom test before a comparison with an atom (with + %% the same failure label) is redundant. + backward(Is, D, Acc); + {{test,is_integer,Fail,[R]}, + [{test,is_eq_exact,Fail,[R,{integer,_}]}|_]} -> + %% An is_integer test before a comparison with an integer + %% (with the same failure label) is redundant. + backward(Is, D, Acc); + {{test,_,_,_},_} -> + %% Still a test instruction. Done. + backward(Is, D, [I|Acc]); + {_,_} -> + %% Rewritten to a select_val. Rescan. + backward([I|Is], D, Acc) + end; backward([{test,Op,{f,To0},Live,Ops0,Dst}|Is], D, Acc) -> To1 = shortcut_bs_test(To0, Is, D), To2 = shortcut_label(To1, D), @@ -321,6 +384,34 @@ backward([{kill,_}=I|Is], D, [{line,_},Exit|_]=Acc) -> false -> backward(Is, D, [I|Acc]); true -> backward(Is, D, Acc) end; +backward([{bif,'or',{f,To0},[Dst,{atom,false}],Dst}=I|Is], D, + [{test,is_eq_exact,{f,To},[Dst,{atom,true}]}|_]=Acc) -> + case shortcut_label(To0, D) of + To -> + backward(Is, D, Acc); + _ -> + backward(Is, D, [I|Acc]) + end; +backward([{bif,map_get,{f,FF},[Key,Map],_}=I0, + {test,has_map_fields,{f,FT}=F,[Map|Keys0]}=I1|Is], D, Acc) when FF =/= 0 -> + case shortcut_label(FF, D) of + FT -> + case lists:delete(Key, Keys0) of + [] -> + backward([I0|Is], D, Acc); + Keys -> + Test = {test,has_map_fields,F,[Map|Keys]}, + backward([Test|Is], D, [I0|Acc]) + end; + _ -> + backward([I1|Is], D, [I0|Acc]) + end; +backward([{bif,map_get,{f,FF},[_,Map],_}=I0, + {test,is_map,{f,FT},[Map]}=I1|Is], D, Acc) when FF =/= 0 -> + case shortcut_label(FF, D) of + FT -> backward([I0|Is], D, Acc); + _ -> backward([I1|Is], D, [I0|Acc]) + end; backward([I|Is], D, Acc) -> backward(Is, D, [I|Acc]); backward([], _D, Acc) -> Acc. @@ -339,6 +430,8 @@ shortcut_select_list([Lit,{f,To0}|T], Reg, D, Acc) -> shortcut_select_list(T, Reg, D, [{f,To},Lit|Acc]); shortcut_select_list([], _, _, Acc) -> reverse(Acc). +shortcut_label(0, _) -> + 0; shortcut_label(To0, D) -> case beam_utils:code_at(To0, D) of [{jump,{f,To}}|_] -> shortcut_label(To, D); @@ -348,6 +441,12 @@ shortcut_label(To0, D) -> shortcut_select_label(To, Reg, Lit, D) -> shortcut_rel_op(To, is_ne_exact, [Reg,Lit], D). +prune_redundant([_,{f,Fail}|T], Fail) -> + prune_redundant(T, Fail); +prune_redundant([V,F|T], Fail) -> + [V,F|prune_redundant(T, Fail)]; +prune_redundant([], _) -> []. + %% Replace a comparison operator with a test instruction and a jump. %% For example, if we have this code: %% @@ -368,14 +467,14 @@ shortcut_select_label(To, Reg, Lit, D) -> replace_comp_op(To, Reg, Op, Ops, D) -> False = comp_op_find_shortcut(To, Reg, {atom,false}, D), True = comp_op_find_shortcut(To, Reg, {atom,true}, D), - [bif_to_test(Op, Ops, False),{jump,{f,True}}]. + {bif_to_test(Op, Ops, False),{jump,{f,True}}}. comp_op_find_shortcut(To0, Reg, Val, D) -> case shortcut_select_label(To0, Reg, Val, D) of To0 -> not_possible(); To -> - case beam_utils:is_killed_at(Reg, To, D) of + case is_killed_at(Reg, To, D) of false -> not_possible(); true -> To end @@ -405,15 +504,22 @@ not_possible() -> throw(not_possible). %% F1: is_eq_exact F2 Reg Lit2 F1: is_eq_exact F2 Reg Lit2 %% L2: .... L2: %% -combine_eqs(To, [Reg,{Type,_}=Lit1]=Ops, D, [{label,L1}|_]) - when Type =:= atom; Type =:= integer -> +combine_eqs(To, [Reg,{Type,_}=Lit1]=Ops, D, Acc) + when Type =:= atom; Type =:= integer -> + Next = case Acc of + [{label,Lbl}|_] -> Lbl; + [{jump,{f,Lbl}}|_] -> Lbl + end, case beam_utils:code_at(To, D) of [{test,is_eq_exact,{f,F2},[Reg,{Type,_}=Lit2]}, {label,L2}|_] when Lit1 =/= Lit2 -> - {select,select_val,Reg,{f,F2},[Lit1,{f,L1},Lit2,{f,L2}]}; + {select,select_val,Reg,{f,F2},[Lit1,{f,Next},Lit2,{f,L2}]}; + [{test,is_eq_exact,{f,F2},[Reg,{Type,_}=Lit2]}, + {jump,{f,L2}}|_] when Lit1 =/= Lit2 -> + {select,select_val,Reg,{f,F2},[Lit1,{f,Next},Lit2,{f,L2}]}; [{select,select_val,Reg,{f,F2},[{Type,_}|_]=List0}|_] -> List = remove_from_list(Lit1, List0), - {select,select_val,Reg,{f,F2},[Lit1,{f,L1}|List]}; + {select,select_val,Reg,{f,F2},[Lit1,{f,Next}|List]}; _Is -> {test,is_eq_exact,{f,To},Ops} end; @@ -426,6 +532,22 @@ remove_from_list(Lit, [Val,{f,_}=Fail|T]) -> [Val,Fail|remove_from_list(Lit, T)]; remove_from_list(_, []) -> []. + +test_bs_literal(F, Ctxt, D, + [{test,bs_match_string,F,[Ctxt,Bs]}, + {test,bs_test_tail2,F,[Ctxt,0]}|Acc]) -> + test_bs_literal_1(Ctxt, Acc, D, Bs); +test_bs_literal(F, Ctxt, D, [{test,bs_test_tail2,F,[Ctxt,0]}|Acc]) -> + test_bs_literal_1(Ctxt, Acc, D, <<>>); +test_bs_literal(_, Ctxt, D, Acc) -> + test_bs_literal_1(Ctxt, Acc, D, none). + +test_bs_literal_1(Ctxt, Is, D, Literal) -> + case beam_utils:is_killed(Ctxt, Is, D) of + true -> {Literal,Is}; + false -> not_killed + end. + %% shortcut_bs_test(TargetLabel, ReversedInstructions, D) -> TargetLabel' %% Try to shortcut the failure label for bit syntax matching. @@ -509,6 +631,21 @@ shortcut_bs_start_match_1([{test,bs_start_match2,{f,To},_,[Reg|_],_}|_], shortcut_bs_start_match_1(_, _, To, _) -> To. +%% shortcut_bs_context_to_binary(TargetLabel, Reg) -> TargetLabel +%% If a bs_start_match2 instruction has been eliminated, the +%% bs_context_to_binary instruction can be eliminated too. + +shortcut_bs_context_to_binary(To, Reg, D) -> + shortcut_bs_ctb_1(beam_utils:code_at(To, D), Reg, To, D). + +shortcut_bs_ctb_1([{bs_context_to_binary,Reg}|Is], Reg, To, D) -> + shortcut_bs_ctb_1(Is, Reg, To, D); +shortcut_bs_ctb_1([{jump,{f,To}}|_], Reg, _, D) -> + Code = beam_utils:code_at(To, D), + shortcut_bs_ctb_1(Code, Reg, To, D); +shortcut_bs_ctb_1(_, _, To, _) -> + To. + %% shortcut_rel_op(FailLabel, Operator, [Operand], D) -> FailLabel' %% Try to shortcut the given test instruction. Example: %% @@ -818,3 +955,17 @@ get_literal(nil) -> get_literal({literal,_}=Lit) -> Lit; get_literal({_,_}) -> error. + + +%%% +%%% Removing stores to Y registers is not always safe +%%% if there is an instruction that causes an exception +%%% within a catch. In practice, there are few or no +%%% opportunities for removing stores to Y registers anyway +%%% if sys_core_fold has been run. +%%% + +is_killed_at({x,_}=Reg, Lbl, D) -> + beam_utils:is_killed_at(Reg, Lbl, D); +is_killed_at({y,_}, _, _) -> + false. diff --git a/lib/compiler/src/beam_dict.erl b/lib/compiler/src/beam_dict.erl index 2b5f8c1b7f..990e86062a 100644 --- a/lib/compiler/src/beam_dict.erl +++ b/lib/compiler/src/beam_dict.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1998-2014. All Rights Reserved. +%% Copyright Ericsson AB 1998-2016. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -24,11 +24,11 @@ -export([new/0,opcode/2,highest_opcode/1, atom/2,local/4,export/4,import/4, string/2,lambda/3,literal/2,line/2,fname/2, - atom_table/1,local_table/1,export_table/1,import_table/1, + atom_table/2,local_table/1,export_table/1,import_table/1, string_table/1,lambda_table/1,literal_table/1, line_table/1]). --type label() :: non_neg_integer(). +-type label() :: beam_asm:label(). -type index() :: non_neg_integer(). @@ -38,13 +38,16 @@ -type line_tab() :: #{{Fname :: index(), Line :: term()} => index()}. -type literal_tab() :: dict:dict(Literal :: term(), index()). +-type lambda_info() :: {label(),{index(),label(),non_neg_integer()}}. +-type lambda_tab() :: {non_neg_integer(),[lambda_info()]}. + -record(asm, {atoms = #{} :: atom_tab(), exports = [] :: [{label(), arity(), label()}], locals = [] :: [{label(), arity(), label()}], imports = gb_trees:empty() :: import_tab(), strings = <<>> :: binary(), %String pool - lambdas = [], %[{...}] + lambdas = {0,[]} :: lambda_tab(), literals = dict:new() :: literal_tab(), fnames = #{} :: fname_tab(), lines = #{} :: line_tab(), @@ -145,15 +148,11 @@ string(Str, Dict) when is_list(Str) -> -spec lambda(label(), non_neg_integer(), bdict()) -> {non_neg_integer(), bdict()}. -lambda(Lbl, NumFree, #asm{lambdas=Lambdas0}=Dict) -> - OldIndex = length(Lambdas0), +lambda(Lbl, NumFree, #asm{lambdas={OldIndex,Lambdas0}}=Dict) -> %% Set Index the same as OldIndex. Index = OldIndex, - %% Initialize OldUniq to 0. It will be set to an unique value - %% based on the MD5 checksum of the BEAM code for the module. - OldUniq = 0, - Lambdas = [{Lbl,{OldIndex,Lbl,Index,NumFree,OldUniq}}|Lambdas0], - {OldIndex,Dict#asm{lambdas=Lambdas}}. + Lambdas = [{Lbl,{Index,Lbl,NumFree}}|Lambdas0], + {OldIndex,Dict#asm{lambdas={OldIndex+1,Lambdas}}}. %% Returns the index for a literal (adding it to the literal table if necessary). %% literal(Literal, Dict) -> {Index,Dict'} @@ -186,6 +185,9 @@ line([{location,Name,Line}], #asm{lines=Lines,num_lines=N}=Dict0) -> {Index, Dict1#asm{lines=Lines#{Key=>Index},num_lines=N+1}} end. +-spec fname(nonempty_string(), bdict()) -> + {non_neg_integer(), bdict()}. + fname(Name, #asm{fnames=Fnames}=Dict) -> case Fnames of #{Name := Index} -> {Index,Dict}; @@ -195,15 +197,15 @@ fname(Name, #asm{fnames=Fnames}=Dict) -> end. %% Returns the atom table. -%% atom_table(Dict) -> {LastIndex,[Length,AtomString...]} --spec atom_table(bdict()) -> {non_neg_integer(), [[non_neg_integer(),...]]}. +%% atom_table(Dict, Encoding) -> {LastIndex,[Length,AtomString...]} +-spec atom_table(bdict(), latin1 | utf8) -> {non_neg_integer(), [[non_neg_integer(),...]]}. -atom_table(#asm{atoms=Atoms}) -> +atom_table(#asm{atoms=Atoms}, Encoding) -> NumAtoms = maps:size(Atoms), Sorted = lists:keysort(2, maps:to_list(Atoms)), {NumAtoms,[begin - L = atom_to_list(A), - [length(L)|L] + L = atom_to_binary(A, Encoding), + [byte_size(L),L] end || {A,_} <- Sorted]}. %% Returns the table of local functions. @@ -236,13 +238,16 @@ string_table(#asm{strings=Strings,string_offset=Size}) -> -spec lambda_table(bdict()) -> {non_neg_integer(), [<<_:192>>]}. -lambda_table(#asm{locals=Loc0,lambdas=Lambdas0}) -> +lambda_table(#asm{locals=Loc0,lambdas={NumLambdas,Lambdas0}}) -> Lambdas1 = sofs:relation(Lambdas0), Loc = sofs:relation([{Lbl,{F,A}} || {F,A,Lbl} <- Loc0]), Lambdas2 = sofs:relative_product1(Lambdas1, Loc), + %% Initialize OldUniq to 0. It will be set to an unique value + %% based on the MD5 checksum of the BEAM code for the module. + OldUniq = 0, Lambdas = [<<F:32,A:32,Lbl:32,Index:32,NumFree:32,OldUniq:32>> || - {{_,Lbl,Index,NumFree,OldUniq},{F,A}} <- sofs:to_external(Lambdas2)], - {length(Lambdas),Lambdas}. + {{Index,Lbl,NumFree},{F,A}} <- sofs:to_external(Lambdas2)], + {NumLambdas,Lambdas}. %% Returns the literal table. %% literal_table(Dict) -> {NumLiterals, [<<TermSize>>,TermInExternalFormat]} diff --git a/lib/compiler/src/beam_disasm.erl b/lib/compiler/src/beam_disasm.erl index 84a94f09e3..6cee9acae4 100644 --- a/lib/compiler/src/beam_disasm.erl +++ b/lib/compiler/src/beam_disasm.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2000-2014. All Rights Reserved. +%% Copyright Ericsson AB 2000-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -163,8 +163,8 @@ pp_instr(I) -> file(File) -> try process_chunks(File) - catch error:Reason -> - {error,?MODULE,{internal,{Reason,erlang:get_stacktrace()}}} + catch error:Reason:Stack -> + {error,?MODULE,{internal,{Reason,Stack}}} end. %%----------------------------------------------------------------------- @@ -314,10 +314,7 @@ get_funs({LsR0,[{func_info,[{atom,M}=AtomM,{atom,F}=AtomF,ArityArg]}|Code0]}) when is_atom(M), is_atom(F) -> Arity = resolve_arg_unsigned(ArityArg), {LsR,Code,RestCode} = get_fun(Code0, []), - Entry = case Code of - [{label,[{u,E}]}|_] -> E; - _ -> undefined - end, + [{label,[{u,Entry}]}|_] = Code, [#function{name=F, arity=Arity, entry=Entry, @@ -722,42 +719,6 @@ resolve_inst({wait,[Lbl]},_,_,_) -> {wait,Lbl}; resolve_inst({wait_timeout,[Lbl,Int]},_,_,_) -> {wait_timeout,Lbl,resolve_arg(Int)}; -resolve_inst({m_plus,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'+',W,[SrcR1,SrcR2],DstR}; -resolve_inst({m_minus,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'-',W,[SrcR1,SrcR2],DstR}; -resolve_inst({m_times,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'*',W,[SrcR1,SrcR2],DstR}; -resolve_inst({m_div,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'/',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_div,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'div',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_rem,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'rem',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_band,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'band',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_bor,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'bor',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_bxor,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'bxor',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_bsl,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'bsl',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_bsr,Args},_,_,_) -> - [W,SrcR1,SrcR2,DstR] = resolve_args(Args), - {arithbif,'bsr',W,[SrcR1,SrcR2],DstR}; -resolve_inst({int_bnot,Args},_,_,_) -> - [W,SrcR,DstR] = resolve_args(Args), - {arithbif,'bnot',W,[SrcR],DstR}; resolve_inst({is_lt=I,Args0},_,_,_) -> [L|Args] = resolve_args(Args0), {test,I,L,Args}; @@ -803,9 +764,6 @@ resolve_inst({is_nil=I,Args0},_,_,_) -> resolve_inst({is_binary=I,Args0},_,_,_) -> [L|Args] = resolve_args(Args0), {test,I,L,Args}; -resolve_inst({is_constant=I,Args0},_,_,_) -> - [L|Args] = resolve_args(Args0), - {test,I,L,Args}; resolve_inst({is_list=I,Args0},_,_,_) -> [L|Args] = resolve_args(Args0), {test,I,L,Args}; @@ -818,6 +776,9 @@ resolve_inst({is_tuple=I,Args0},_,_,_) -> resolve_inst({test_arity=I,Args0},_,_,_) -> [L|Args] = resolve_args(Args0), {test,I,L,Args}; +resolve_inst({is_tagged_tuple=I,Args0},_,_,_) -> + [F|Args] = resolve_args(Args0), + {test,I,F,Args}; resolve_inst({select_val,Args},_,_,_) -> [Reg,FLbl,{{z,1},{u,_Len},List0}] = Args, List = resolve_args(List0), @@ -840,11 +801,6 @@ resolve_inst({get_tuple_element,[Src,{u,Off},Dst]},_,_,_) -> {get_tuple_element,resolve_arg(Src),Off,resolve_arg(Dst)}; resolve_inst({set_tuple_element,[Src,Dst,{u,Off}]},_,_,_) -> {set_tuple_element,resolve_arg(Src),resolve_arg(Dst),Off}; -resolve_inst({put_string,[{u,Len},{u,Off},Dst]},_,Strings,_) -> - String = if Len > 0 -> binary_to_list(Strings, Off+1, Off+Len); - true -> "" - end, - {put_string,Len,{string,String},Dst}; resolve_inst({put_list,[Src1,Src2,Dst]},_,_,_) -> {put_list,resolve_arg(Src1),resolve_arg(Src2),Dst}; resolve_inst({put_tuple,[{u,Arity},Dst]},_,_,_) -> @@ -859,9 +815,6 @@ resolve_inst({case_end,[X]},_,_,_) -> {case_end,resolve_arg(X)}; resolve_inst({call_fun,[{u,N}]},_,_,_) -> {call_fun,N}; -resolve_inst({make_fun,Args},_,_,Lbls) -> - [{f,L},Magic,FreeVars] = resolve_args(Args), - {make_fun,lookup(L,Lbls),Magic,FreeVars}; resolve_inst({is_function=I,Args0},_,_,_) -> [L|Args] = resolve_args(Args0), {test,I,L,Args}; @@ -870,30 +823,6 @@ resolve_inst({call_ext_only,[{u,N},{u,MFAix}]},Imports,_,_) -> %% %% Instructions for handling binaries added in R7A & R7B %% -resolve_inst({bs_start_match,[F,Reg]},_,_,_) -> - {bs_start_match,F,Reg}; -resolve_inst({bs_get_integer=I,[Lbl,Arg2,{u,N},{u,U},Arg5]},_,_,_) -> - [A2,A5] = resolve_args([Arg2,Arg5]), - {test,I,Lbl,[A2,N,decode_field_flags(U),A5]}; -resolve_inst({bs_get_float=I,[Lbl,Arg2,{u,N},{u,U},Arg5]},_,_,_) -> - [A2,A5] = resolve_args([Arg2,Arg5]), - {test,I,Lbl,[A2,N,decode_field_flags(U),A5]}; -resolve_inst({bs_get_binary=I,[Lbl,Arg2,{u,N},{u,U},Arg5]},_,_,_) -> - [A2,A5] = resolve_args([Arg2,Arg5]), - {test,I,Lbl,[A2,N,decode_field_flags(U),A5]}; -resolve_inst({bs_skip_bits,[Lbl,Arg2,{u,N},{u,U}]},_,_,_) -> - A2 = resolve_arg(Arg2), - {test,bs_skip_bits,Lbl,[A2,N,decode_field_flags(U)]}; -resolve_inst({bs_test_tail,[F,{u,N}]},_,_,_) -> - {test,bs_test_tail,F,[N]}; -resolve_inst({bs_save,[{u,N}]},_,_,_) -> - {bs_save,N}; -resolve_inst({bs_restore,[{u,N}]},_,_,_) -> - {bs_restore,N}; -resolve_inst({bs_init,[{u,N},{u,U}]},_,_,_) -> - {bs_init,N,decode_field_flags(U)}; -resolve_inst({bs_final,[F,X]},_,_,_) -> - {bs_final,F,X}; resolve_inst({bs_put_integer,[Lbl,Arg2,{u,N},{u,U},Arg5]},_,_,_) -> [A2,A5] = resolve_args([Arg2,Arg5]), {bs_put_integer,Lbl,A2,N,decode_field_flags(U),A5}; @@ -908,8 +837,6 @@ resolve_inst({bs_put_string,[{u,Len},{u,Off}]},_,Strings,_) -> true -> "" end, {bs_put_string,Len,{string,String}}; -resolve_inst({bs_need_buf,[{u,N}]},_,_,_) -> - {bs_need_buf,N}; %% %% Instructions for handling floating point numbers added in June 2001 (R8). @@ -961,9 +888,6 @@ resolve_inst({raise,[_Reg1,_Reg2]=Regs},_,_,_) -> resolve_inst({bs_init2,[Lbl,Arg2,{u,W},{u,R},{u,F},Arg6]},_,_,_) -> [A2,A6] = resolve_args([Arg2,Arg6]), {bs_init2,Lbl,A2,W,R,decode_field_flags(F),A6}; -resolve_inst({bs_bits_to_bytes,[Lbl,Arg2,Arg3]},_,_,_) -> - [A2,A3] = resolve_args([Arg2,Arg3]), - {bs_bits_to_bytes,Lbl,A2,A3}; resolve_inst({bs_add=I,[Lbl,Arg2,Arg3,Arg4,Arg5]},_,_,_) -> [A2,A3,A4,A5] = resolve_args([Arg2,Arg3,Arg4,Arg5]), {I,Lbl,[A2,A3,A4],A5}; @@ -1041,12 +965,6 @@ resolve_inst({gc_bif3,Args},Imports,_,_) -> {gc_bif,BifName,F,Live,[A1,A2,A3],Reg}; %% -%% New instructions for creating non-byte aligned binaries. -%% -resolve_inst({bs_final2,[X,Y]},_,_,_) -> - {bs_final2,X,Y}; - -%% %% R11B-5. %% resolve_inst({is_bitstr=I,Args0},_,_,_) -> @@ -1165,6 +1083,19 @@ resolve_inst({get_map_elements,Args0},_,_,_) -> {get_map_elements,FLbl,Src,{list,List}}; %% +%% OTP 21. +%% + +resolve_inst({build_stacktrace,[]},_,_,_) -> + build_stacktrace; +resolve_inst({raw_raise,[]},_,_,_) -> + raw_raise; +resolve_inst({get_hd,[Src,Dst]},_,_,_) -> + {get_hd,Src,Dst}; +resolve_inst({get_tl,[Src,Dst]},_,_,_) -> + {get_tl,Src,Dst}; + +%% %% Catches instructions that are not yet handled. %% resolve_inst(X,_,_,_) -> ?exit({resolve_inst,X}). diff --git a/lib/compiler/src/beam_disasm.hrl b/lib/compiler/src/beam_disasm.hrl index 30e3f22665..e8ebfc4cfc 100644 --- a/lib/compiler/src/beam_disasm.hrl +++ b/lib/compiler/src/beam_disasm.hrl @@ -2,7 +2,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2007-2009. All Rights Reserved. +%% Copyright Ericsson AB 2007-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -22,9 +22,12 @@ %% the system (e.g. in the translation from Beam to Icode). %% -%% XXX: THE FOLLOWING TYPE DECLARATION DOES NOT BELONG HERE... +%% XXX: THE FOLLOWING TYPE DECLARATION DOES NOT BELONG HERE. +%% IT SHOULD BE MOVED TO A FILE THAT DEFINES (AND EXPORTS) +%% PROPER TYPES FOR THE SET OF BEAM INSTRUCTIONS. %% --type beam_instr() :: 'bs_init_writable' | 'fclearerror' | 'if_end' +-type beam_instr() :: 'bs_init_writable' | 'build_stacktrace' + | 'fclearerror' | 'if_end' | 'raw_raise' | 'remove_message' | 'return' | 'send' | 'timeout' | tuple(). %% XXX: Very underspecified - FIX THIS @@ -34,7 +37,7 @@ -record(function, {name :: atom(), arity :: byte(), - entry, %% unused ?? + entry :: beam_lib:label(), %% unnecessary ? code = [] :: [beam_instr()]}). -record(beam_file, {module :: module(), diff --git a/lib/compiler/src/beam_except.erl b/lib/compiler/src/beam_except.erl index e33655281f..05c0f4fbc7 100644 --- a/lib/compiler/src/beam_except.erl +++ b/lib/compiler/src/beam_except.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2011-2013. All Rights Reserved. +%% Copyright Ericsson AB 2011-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -33,6 +33,9 @@ -import(lists, [reverse/1]). +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> Fs = [function(F) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. @@ -42,16 +45,15 @@ function({function,Name,Arity,CLabel,Is0}) -> Is = function_1(Is0), {function,Name,Arity,CLabel,Is} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. -record(st, - {lbl, %func_info label - loc, %location for func_info - arity %arity for function + {lbl :: beam_asm:label(), %func_info label + loc :: [_], %location for func_info + arity :: arity() %arity for function }). function_1(Is0) -> @@ -133,10 +135,12 @@ translate_exception(_, _, _, _) -> no. fix_block(Is, 0) -> reverse(Is); fix_block(Is, Words) -> - fix_block_1(reverse(Is), Words). + reverse(fix_block_1(Is, Words)). -fix_block_1([{set,[],[],{alloc,Live,{F1,F2,Needed,F3}}}|Is], Words) -> - [{set,[],[],{alloc,Live,{F1,F2,Needed-Words,F3}}}|Is]; +fix_block_1([{set,[],[],{alloc,Live,{F1,F2,Needed0,F3}}}|Is], Words) -> + Needed = Needed0 - Words, + true = Needed >= 0, %Assertion. + [{set,[],[],{alloc,Live,{F1,F2,Needed,F3}}}|Is]; fix_block_1([I|Is], Words) -> [I|fix_block_1(Is, Words)]. diff --git a/lib/compiler/src/beam_flatten.erl b/lib/compiler/src/beam_flatten.erl index 13c243b155..20bd23a912 100644 --- a/lib/compiler/src/beam_flatten.erl +++ b/lib/compiler/src/beam_flatten.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2013. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -25,6 +25,9 @@ -import(lists, [reverse/1,reverse/2]). +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs,Lc}, _Opt) -> {ok,{Mod,Exp,Attr,[function(F) || F <- Fs],Lc}}. @@ -47,6 +50,9 @@ norm_block([{set,[],[],{alloc,R,Alloc}}|Is], Acc0) -> Acc -> norm_block(Is, Acc) end; +norm_block([{set,[D1],[S],get_hd},{set,[D2],[S],get_tl}|Is], Acc) -> + I = {get_list,S,D1,D2}, + norm_block(Is, [I|Acc]); norm_block([I|Is], Acc) -> norm_block(Is, [norm(I)|Acc]); norm_block([], Acc) -> Acc. @@ -61,13 +67,14 @@ norm({set,[D],[],{put_tuple,A}}) -> {put_tuple,A,D}; norm({set,[],[S],put}) -> {put,S}; norm({set,[D],[S],{get_tuple_element,I}}) -> {get_tuple_element,S,I,D}; norm({set,[],[S,D],{set_tuple_element,I}}) -> {set_tuple_element,S,D,I}; -norm({set,[D1,D2],[S],get_list}) -> {get_list,S,D1,D2}; +norm({set,[D],[S],get_hd}) -> {get_hd,S,D}; +norm({set,[D],[S],get_tl}) -> {get_tl,S,D}; norm({set,[D],[S|Puts],{alloc,R,{put_map,Op,F}}}) -> {put_map,F,Op,S,D,R,{list,Puts}}; -%% get_map_elements is always handled in beam_split (moved out of block) norm({set,[],[],remove_message}) -> remove_message; norm({set,[],[],fclearerror}) -> fclearerror; -norm({set,[],[],fcheckerror}) -> {fcheckerror,{f,0}}. +norm({set,[],[],fcheckerror}) -> {fcheckerror,{f,0}}; +norm({set,[],[],{line,_}=Line}) -> Line. norm_allocate({_Zero,nostack,Nh,[]}, Regs) -> [{test_heap,Nh,Regs}]; diff --git a/lib/compiler/src/beam_jump.erl b/lib/compiler/src/beam_jump.erl index 5e58e0f6ac..22974da398 100644 --- a/lib/compiler/src/beam_jump.erl +++ b/lib/compiler/src/beam_jump.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2013. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -23,7 +23,7 @@ -export([module/2, is_unreachable_after/1,is_exit_instruction/1, - remove_unused_labels/1,is_label_used_in/2]). + remove_unused_labels/1]). %%% The following optimisations are done: %%% @@ -71,9 +71,9 @@ %%% %%% jump L2 %%% . . . -%%% L1: %%% L2: ... %%% +%%% and all preceding uses of L1 renamed to L2. %%% If the jump is unreachable, it will be removed according to (1). %%% %%% (5) In @@ -128,7 +128,12 @@ %%% on the program state. %%% --import(lists, [reverse/1,reverse/2,foldl/3]). +-import(lists, [dropwhile/2,reverse/1,reverse/2,foldl/3]). + +-type instruction() :: beam_utils:instruction(). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> Fs = [function(F) || F <- Fs0], @@ -155,9 +160,7 @@ share(Is0) -> Is = eliminate_fallthroughs(Is0, []), share_1(Is, #{}, [], []). -share_1([{label,_}=Lbl|Is], Dict, [], Acc) -> - share_1(Is, Dict, [], [Lbl|Acc]); -share_1([{label,L}=Lbl|Is], Dict0, Seq, Acc) -> +share_1([{label,L}=Lbl|Is], Dict0, [_|_]=Seq, Acc) -> case maps:find(Seq, Dict0) of error -> Dict = maps:put(Seq, L, Dict0), @@ -167,12 +170,18 @@ share_1([{label,L}=Lbl|Is], Dict0, Seq, Acc) -> end; share_1([{func_info,_,_,_}=I|Is], _, [], Acc) -> reverse(Is, [I|Acc]); +share_1([{'catch',_,_}=I|Is], Dict0, Seq, Acc) -> + Dict = clean_non_sharable(Dict0), + share_1(Is, Dict, [I|Seq], Acc); share_1([{'try',_,_}=I|Is], Dict0, Seq, Acc) -> Dict = clean_non_sharable(Dict0), share_1(Is, Dict, [I|Seq], Acc); share_1([{try_case,_}=I|Is], Dict0, Seq, Acc) -> Dict = clean_non_sharable(Dict0), share_1(Is, Dict, [I|Seq], Acc); +share_1([{catch_end,_}=I|Is], Dict0, Seq, Acc) -> + Dict = clean_non_sharable(Dict0), + share_1(Is, Dict, [I|Seq], Acc); share_1([I|Is], Dict, Seq, Acc) -> case is_unreachable_after(I) of false -> @@ -182,18 +191,18 @@ share_1([I|Is], Dict, Seq, Acc) -> end. clean_non_sharable(Dict) -> - %% We are passing in or out of a 'try' block. Remove - %% sequences that should not shared over the boundaries - %% of a 'try' block. Since the end of the sequence must match, - %% the only possible match between a sequence outside and - %% a sequence inside the 'try' block is a sequence that ends - %% with an instruction that causes an exception. Any sequence - %% that causes an exception must contain a line/1 instruction. + %% We are passing in or out of a 'catch' or 'try' block. Remove + %% sequences that should not be shared over the boundaries of the + %% block. Since the end of the sequence must match, the only + %% possible match between a sequence outside and a sequence inside + %% the 'catch'/'try' block is a sequence that ends with an + %% instruction that causes an exception. Any sequence that causes + %% an exception must contain a line/1 instruction. maps:filter(fun(K, _V) -> sharable_with_try(K) end, Dict). sharable_with_try([{line,_}|_]) -> %% This sequence may cause an exception and may potentially - %% match a sequence on the other side of the 'try' block + %% match a sequence on the other side of the 'catch'/'try' block %% boundary. false; sharable_with_try([_|Is]) -> @@ -202,21 +211,18 @@ sharable_with_try([]) -> true. %% Eliminate all fallthroughs. Return the result reversed. -eliminate_fallthroughs([I,{label,L}=Lbl|Is], Acc) -> - case is_unreachable_after(I) orelse is_label(I) of +eliminate_fallthroughs([{label,L}=Lbl|Is], [I|_]=Acc) -> + case is_unreachable_after(I) of false -> %% Eliminate fallthrough. - eliminate_fallthroughs(Is, [Lbl,{jump,{f,L}},I|Acc]); + eliminate_fallthroughs(Is, [Lbl,{jump,{f,L}}|Acc]); true -> - eliminate_fallthroughs(Is, [Lbl,I|Acc]) + eliminate_fallthroughs(Is, [Lbl|Acc]) end; eliminate_fallthroughs([I|Is], Acc) -> eliminate_fallthroughs(Is, [I|Acc]); eliminate_fallthroughs([], Acc) -> Acc. -is_label({label,_}) -> true; -is_label(_) -> false. - %%% %%% (2) Move short code sequences ending in an instruction that causes an exit %%% to the end of the function. @@ -266,17 +272,18 @@ extract_seq_1(_, _) -> no. %%% (3) (4) (5) (6) Jump and unreachable code optimizations. %%% --record(st, {fc, %Label for function class errors. - entry, %Entry label (must not be moved). - mlbl, %Moved labels. - labels :: cerl_sets:set() %Set of referenced labels. - }). +-record(st, + { + entry :: beam_asm:label(), %Entry label (must not be moved). + replace :: #{beam_asm:label() := beam_asm:label()}, %Labels to replace. + labels :: cerl_sets:set(), %Set of referenced labels. + index :: beam_utils:code_index() | {lazy,[beam_utils:instruction()]} %Index built lazily only if needed + }). -opt([{label,Fc}|_]=Is0, CLabel) -> - Lbls = initial_labels(Is0), +opt(Is0, CLabel) -> find_fixpoint(fun(Is) -> - St = #st{fc=Fc,entry=CLabel,mlbl=#{}, - labels=Lbls}, + Lbls = initial_labels(Is), + St = #st{entry=CLabel,replace=#{},labels=Lbls,index={lazy,Is}}, opt(Is, [], St) end, Is0). @@ -286,7 +293,7 @@ find_fixpoint(OptFun, Is0) -> Is -> find_fixpoint(OptFun, Is) end. -opt([{test,_,{f,L}=Lbl,_}=I|[{jump,{f,L}}|_]=Is], Acc, St) -> +opt([{test,_,{f,L}=Lbl,_}=I|[{jump,{f,L}}|_]=Is], Acc0, St0) -> %% We have %% Test Label Ops %% jump Label @@ -295,10 +302,23 @@ opt([{test,_,{f,L}=Lbl,_}=I|[{jump,{f,L}}|_]=Is], Acc, St) -> case beam_utils:is_pure_test(I) of false -> %% Test is not pure; we must keep it. - opt(Is, [I|Acc], label_used(Lbl, St)); + opt(Is, [I|Acc0], label_used(Lbl, St0)); true -> %% The test is pure and its failure label is the same %% as in the jump that follows -- thus it is not needed. + %% Check if any of the previous instructions could also be eliminated. + {Acc,St} = opt_useless_loads(Acc0, L, St0), + opt(Is, Acc, St) + end; +opt([{test,_,{f,L}=Lbl,_}=I|[{label,L}|_]=Is], Acc0, St0) -> + %% Similar to the above, except we have a fall-through rather than jump + %% Test Label Ops + %% label Label + case beam_utils:is_pure_test(I) of + false -> + opt(Is, [I|Acc0], label_used(Lbl, St0)); + true -> + {Acc,St} = opt_useless_loads(Acc0, L, St0), opt(Is, Acc, St) end; opt([{test,Test0,{f,L}=Lbl,Ops}=I|[{jump,To}|Is]=Is0], Acc, St) -> @@ -320,29 +340,16 @@ opt([{test,_,{f,_}=Lbl,_,_,_}=I|Is], Acc, St) -> opt(Is, [I|Acc], label_used(Lbl, St)); opt([{select,_,_R,Fail,Vls}=I|Is], Acc, St) -> skip_unreachable(Is, [I|Acc], label_used([Fail|Vls], St)); -opt([{label,Lbl}=I|Is], Acc, #st{mlbl=Mlbl}=St0) -> - case maps:find(Lbl, Mlbl) of - {ok,Lbls} -> - %% Essential to remove the list of labels from the dictionary, - %% since we will rescan the inserted labels. We MUST rescan. - St = St0#st{mlbl=maps:remove(Lbl, Mlbl)}, - insert_labels([Lbl|Lbls], Is, Acc, St); - error -> opt(Is, [I|Acc], St0) - end; +opt([{label,From}=I,{label,To}|Is], Acc, #st{replace=Replace}=St) -> + opt([I|Is], Acc, St#st{replace=Replace#{To => From}}); opt([{jump,{f,_}=X}|[{label,_},{jump,X}|_]=Is], Acc, St) -> opt(Is, Acc, St); opt([{jump,{f,Lbl}}|[{label,Lbl}|_]=Is], Acc, St) -> opt(Is, Acc, St); -opt([{jump,{f,L}=Lbl}=I|Is], Acc0, #st{mlbl=Mlbl0}=St0) -> - %% All labels before this jump instruction should now be - %% moved to the location of the jump's target. - {Lbls,Acc} = collect_labels(Acc0, St0), - St = case Lbls of - [] -> St0; - [_|_] -> - Mlbl = maps_append_list(L, Lbls, Mlbl0), - St0#st{mlbl=Mlbl} - end, +opt([{jump,{f,L}=Lbl}=I|Is], Acc0, St0) -> + %% Replace all labels before this jump instruction into the + %% location of the jump's target. + {Acc,St} = collect_labels(Acc0, L, St0), skip_unreachable(Is, [I|Acc], label_used(Lbl, St)); %% Optimization: quickly handle some common instructions that don't %% have any failure labels and where is_unreachable_after(I) =:= false. @@ -362,39 +369,77 @@ opt([I|Is], Acc, #st{labels=Used0}=St0) -> true -> skip_unreachable(Is, [I|Acc], St); false -> opt(Is, [I|Acc], St) end; -opt([], Acc, #st{fc=Fc,mlbl=Mlbl}) -> - Code = reverse(Acc), - case maps:find(Fc, Mlbl) of - {ok,Lbls} -> insert_fc_labels(Lbls, Mlbl, Code); - error -> Code - end. - -maps_append_list(K,Vs,M) -> - case M of - #{K:=Vs0} -> M#{K:=Vs0++Vs}; % same order as dict - _ -> M#{K => Vs} - end. - -insert_fc_labels([L|Ls], Mlbl, Acc0) -> - Acc = [{label,L}|Acc0], - case maps:find(L, Mlbl) of - error -> - insert_fc_labels(Ls, Mlbl, Acc); - {ok,Lbls} -> - insert_fc_labels(Lbls++Ls, Mlbl, Acc) +opt([], Acc, #st{replace=Replace0}) when Replace0 =/= #{} -> + Replace = normalize_replace(maps:to_list(Replace0), Replace0, []), + beam_utils:replace_labels(Acc, [], Replace, fun(Old) -> Old end); +opt([], Acc, #st{replace=Replace}) when Replace =:= #{} -> + reverse(Acc). + +normalize_replace([{From,To0}|Rest], Replace, Acc) -> + case Replace of + #{To0 := To} -> + normalize_replace([{From,To}|Rest], Replace, Acc); + _ -> + normalize_replace(Rest, Replace, [{From,To0}|Acc]) + end; +normalize_replace([], _Replace, Acc) -> + maps:from_list(Acc). + +%% After eliminating a test, it might happen, that a register was only used +%% in this test. Let's check if that was the case and if it was so, we can +%% eliminate the load into the register completely. +opt_useless_loads([{block,_}|_]=Is, L, #st{index={lazy,FIs}}=St) -> + opt_useless_loads(Is, L, St#st{index=beam_utils:index_labels(FIs)}); +opt_useless_loads([{block,Block0}|Is], L, #st{index=Index}=St) -> + case opt_useless_block_loads(Block0, L, Index) of + [] -> + opt_useless_loads(Is, L, St); + [_|_]=Block -> + {[{block,Block}|Is],St} + end; +%% After eliminating the test and useless blocks, it might happen, +%% that the previous test could also be eliminated. +%% It might be that the label was already marked as used, even if ultimately, +%% it never will be - we can't do much about it at that point, though +opt_useless_loads([{test,_,{f,L},_}=I|Is], L, St) -> + case beam_utils:is_pure_test(I) of + false -> + {[I|Is],St}; + true -> + opt_useless_loads(Is, L, St) end; -insert_fc_labels([], _, Acc) -> Acc. +opt_useless_loads(Is, _L, St) -> + {Is,St}. + +opt_useless_block_loads([{set,[Dst],_,_}=I|Is0], L, Index) -> + BlockJump = [{block,Is0},{jump,{f,L}}], + case beam_utils:is_killed(Dst, BlockJump, Index) of + true -> + %% The register is killed and not used, we can remove the load. + %% Remove any `put` instructions in case we just + %% removed a `put_tuple` instruction. + Is = dropwhile(fun({set,_,_,put}) -> true; + (_) -> false + end, Is0), + opt_useless_block_loads(Is, L, Index); + false -> + [I|opt_useless_block_loads(Is0, L, Index)] + end; +opt_useless_block_loads([I|Is], L, Index) -> + [I|opt_useless_block_loads(Is, L, Index)]; +opt_useless_block_loads([], _L, _Index) -> + []. -collect_labels(Is, #st{entry=Entry}) -> - collect_labels_1(Is, Entry, []). +collect_labels(Is, Label, #st{entry=Entry,replace=Replace} = St) -> + collect_labels_1(Is, Label, Entry, Replace, St). -collect_labels_1([{label,Entry}|_]=Is, Entry, Acc) -> +collect_labels_1([{label,Entry}|_]=Is, _Label, Entry, Acc, St) -> %% Never move the entry label. - {Acc,Is}; -collect_labels_1([{label,L}|Is], Entry, Acc) -> - collect_labels_1(Is, Entry, [L|Acc]); -collect_labels_1(Is, _Entry, Acc) -> - {Acc,Is}. + {Is,St#st{replace=Acc}}; +collect_labels_1([{label,L}|Is], Label, Entry, Acc, St) -> + collect_labels_1(Is, Label, Entry, Acc#{L => Label}, St); +collect_labels_1(Is, _Label, _Entry, Acc, St) -> + {Is,St#st{replace=Acc}}. %% label_defined(Is, Label) -> true | false. %% Test whether the label Label is defined at the start of the instruction @@ -414,13 +459,6 @@ invert_test(is_eq_exact) -> is_ne_exact; invert_test(is_ne_exact) -> is_eq_exact; invert_test(_) -> not_possible. -insert_labels([L|Ls], Is, [{jump,{f,L}}|Acc], St) -> - insert_labels(Ls, [{label,L}|Is], Acc, St); -insert_labels([L|Ls], Is, Acc, St) -> - insert_labels(Ls, [{label,L}|Is], Acc, St); -insert_labels([], Is, Acc, St) -> - opt(Is, Acc, St). - %% skip_unreachable([Instruction], St). %% Remove all instructions (including definitions of labels %% that have not been referenced yet) up to the next @@ -454,6 +492,8 @@ is_label_used(L, St) -> %% is_unreachable_after(Instruction) -> boolean() %% Test whether the code after Instruction is unreachable. +-spec is_unreachable_after(instruction()) -> boolean(). + is_unreachable_after({func_info,_M,_F,_A}) -> true; is_unreachable_after(return) -> true; is_unreachable_after({jump,_Lbl}) -> true; @@ -466,6 +506,8 @@ is_unreachable_after(I) -> is_exit_instruction(I). %% Test whether the instruction Instruction always %% causes an exit/failure. +-spec is_exit_instruction(instruction()) -> boolean(). + is_exit_instruction({call_ext,_,{extfunc,M,F,A}}) -> erl_bifs:is_exit_bif(M, F, A); is_exit_instruction(if_end) -> true; @@ -474,40 +516,12 @@ is_exit_instruction({try_case_end,_}) -> true; is_exit_instruction({badmatch,_}) -> true; is_exit_instruction(_) -> false. -%% is_label_used_in(LabelNumber, [Instruction]) -> boolean() -%% Check whether the label is used in the instruction sequence -%% (including inside blocks). - -is_label_used_in(Lbl, Is) -> - is_label_used_in_1(Is, Lbl, cerl_sets:new()). - -is_label_used_in_1([{block,Block}|Is], Lbl, Empty) -> - lists:any(fun(I) -> is_label_used_in_block(I, Lbl) end, Block) - orelse is_label_used_in_1(Is, Lbl, Empty); -is_label_used_in_1([I|Is], Lbl, Empty) -> - Used = ulbl(I, Empty), - cerl_sets:is_element(Lbl, Used) orelse is_label_used_in_1(Is, Lbl, Empty); -is_label_used_in_1([], _, _) -> false. - -is_label_used_in_block({set,_,_,Info}, Lbl) -> - case Info of - {bif,_,{f,F}} -> F =:= Lbl; - {alloc,_,{gc_bif,_,{f,F}}} -> F =:= Lbl; - {alloc,_,{put_map,_,{f,F}}} -> F =:= Lbl; - {get_map_elements,{f,F}} -> F =:= Lbl; - {'catch',{f,F}} -> F =:= Lbl; - {alloc,_,_} -> false; - {put_tuple,_} -> false; - {get_tuple_element,_} -> false; - {set_tuple_element,_} -> false; - {line,_} -> false; - _ when is_atom(Info) -> false - end. - %% remove_unused_labels(Instructions0) -> Instructions %% Remove all unused labels. Also remove unreachable %% instructions following labels that are removed. +-spec remove_unused_labels([instruction()]) -> [instruction()]. + remove_unused_labels(Is) -> Used0 = initial_labels(Is), Used = foldl(fun ulbl/2, Used0, Is), diff --git a/lib/compiler/src/beam_listing.erl b/lib/compiler/src/beam_listing.erl index fd5c0a042b..518b958794 100644 --- a/lib/compiler/src/beam_listing.erl +++ b/lib/compiler/src/beam_listing.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1997-2011. All Rights Reserved. +%% Copyright Ericsson AB 1997-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -21,24 +21,29 @@ -export([module/2]). --include("v3_life.hrl"). +-include("core_parse.hrl"). +-include("v3_kernel.hrl"). +-include("beam_disasm.hrl"). -import(lists, [foreach/2]). -module(File, Core) when element(1, Core) == c_module -> +-type code() :: cerl:c_module() + | beam_utils:module_code() + | #k_mdef{} + | [_]. %form-based format + +-spec module(file:io_device(), code()) -> 'ok'. + +module(File, #c_module{}=Core) -> %% This is a core module. io:put_chars(File, core_pp:format(Core)); -module(File, Kern) when element(1, Kern) == k_mdef -> +module(File, #k_mdef{}=Kern) -> %% This is a kernel module. io:put_chars(File, v3_kernel_pp:format(Kern)); %%io:put_chars(File, io_lib:format("~p~n", [Kern])); -module(File, {Mod,Exp,Attr,Kern}) -> - %% This is output from beam_life (v3). - io:fwrite(File, "~w.~n~p.~n~p.~n", [Mod,Exp,Attr]), - foreach(fun (F) -> function(File, F) end, Kern); module(Stream, {Mod,Exp,Attr,Code,NumLabels}) -> - %% This is output from beam_codegen. - io:format(Stream, "{module, ~p}. %% version = ~w\n", + %% This is output from v3_codegen. + io:format(Stream, "{module, ~p}. %% version = ~w\n", [Mod, beam_opcodes:format_number()]), io:format(Stream, "\n{exports, ~p}.\n", [Exp]), io:format(Stream, "\n{attributes, ~p}.\n", [Attr]), @@ -49,10 +54,6 @@ module(Stream, {Mod,Exp,Attr,Code,NumLabels}) -> [Name, Arity, Entry]), io:put_chars(Stream, format_asm(Asm)) end, Code); -module(Stream, {Mod,Exp,Inter}) -> - %% Other kinds of intermediate formats. - io:fwrite(Stream, "~w.~n~p.~n", [Mod,Exp]), - foreach(fun (F) -> io:format(Stream, "~p.\n", [F]) end, Inter); module(Stream, [_|_]=Fs) -> %% Form-based abstract format. foreach(fun (F) -> io:format(Stream, "~p.\n", [F]) end, Fs). @@ -62,60 +63,3 @@ format_asm([{label,L}|Is]) -> format_asm([I|Is]) -> [io_lib:format(" ~p", [I]),".\n"|format_asm(Is)]; format_asm([]) -> []. - -function(File, {function,Name,Arity,Args,Body,Vdb,_Anno}) -> - io:nl(File), - io:format(File, "function ~p/~p.\n", [Name,Arity]), - io:format(File, " ~p.\n", [Args]), - print_vdb(File, Vdb), - put(beam_listing_nl, false), - nl(File), - foreach(fun(F) -> format(File, F, []) end, Body), - nl(File), - erase(beam_listing_nl). - -format(File, #l{ke=Ke,i=I,vdb=Vdb}, Ind) -> - nl(File), - ind_format(File, Ind, "~p ", [I]), - print_vdb(File, Vdb), - nl(File), - format(File, Ke, Ind); -format(File, Tuple, Ind) when is_tuple(Tuple) -> - ind_format(File, Ind, "{", []), - format_list(File, tuple_to_list(Tuple), [$\s|Ind]), - ind_format(File, Ind, "}", []); -format(File, List, Ind) when is_list(List) -> - ind_format(File, Ind, "[", []), - format_list(File, List, [$\s|Ind]), - ind_format(File, Ind, "]", []); -format(File, F, Ind) -> - ind_format(File, Ind, "~p", [F]). - -format_list(File, [F], Ind) -> - format(File, F, Ind); -format_list(File, [F|Fs], Ind) -> - format(File, F, Ind), - ind_format(File, Ind, ",", []), - format_list(File, Fs, Ind); -format_list(_, [], _) -> ok. - - -print_vdb(File, [{Var,F,E}|Vs]) -> - io:format(File, "~p:~p..~p ", [Var,F,E]), - print_vdb(File, Vs); -print_vdb(_, []) -> ok. - -ind_format(File, Ind, Format, Args) -> - case get(beam_listing_nl) of - true -> - put(beam_listing_nl, false), - io:put_chars(File, Ind); - false -> ok - end, - io:format(File, Format, Args). - -nl(File) -> - case put(beam_listing_nl, true) of - true -> ok; - false -> io:nl(File) - end. diff --git a/lib/compiler/src/beam_peep.erl b/lib/compiler/src/beam_peep.erl index 17fd2e502a..2b8dd40e29 100644 --- a/lib/compiler/src/beam_peep.erl +++ b/lib/compiler/src/beam_peep.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2008-2013. All Rights Reserved. +%% Copyright Ericsson AB 2008-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -24,6 +24,9 @@ -import(lists, [reverse/1,member/2]). +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,_}, _Opts) -> %% First coalesce adjacent labels. {Fs1,Lc} = beam_clean:clean_labels(Fs0), @@ -38,8 +41,7 @@ function({function,Name,Arity,CLabel,Is0}) -> Is = beam_jump:remove_unused_labels(Is1), {function,Name,Arity,CLabel,Is} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -65,18 +67,6 @@ function({function,Name,Arity,CLabel,Is0}) -> %% InEncoding =:= latin1, OutEncoding =:= unicode; %% InEncoding =:= latin1, OutEncoding =:= utf8 -> %% -%% (2) A select_val/4 instruction that only verifies that -%% its argument is either 'true' or 'false' can be -%% be replaced with an is_boolean/2 instruction. That is: -%% -%% select_val Reg Fail [ true Next false Next ] -%% Next: ... -%% -%% can be rewritten to -%% -%% is_boolean Fail Reg -%% Next: ... -%% peep(Is) -> peep(Is, gb_sets:empty(), []). @@ -87,6 +77,12 @@ peep([{bif,tuple_size,_,[_]=Ops,Dst}=I|Is], SeenTests0, Acc) -> %% Kill all remembered tests that depend on the destination register. SeenTests = kill_seen(Dst, SeenTests1), peep(Is, SeenTests, [I|Acc]); +peep([{bif,map_get,_,[Key,Map],Dst}=I|Is], SeenTests0, Acc) -> + %% Pretend that we have seen {test,has_map_fields,_,[Map,Key]} + SeenTests1 = gb_sets:add({has_map_fields,[Map,Key]}, SeenTests0), + %% Kill all remembered tests that depend on the destination register. + SeenTests = kill_seen(Dst, SeenTests1), + peep(Is, SeenTests, [I|Acc]); peep([{bif,_,_,_,Dst}=I|Is], SeenTests0, Acc) -> %% Kill all remembered tests that depend on the destination register. SeenTests = kill_seen(Dst, SeenTests0), @@ -95,12 +91,41 @@ peep([{gc_bif,_,_,_,_,Dst}=I|Is], SeenTests0, Acc) -> %% Kill all remembered tests that depend on the destination register. SeenTests = kill_seen(Dst, SeenTests0), peep(Is, SeenTests, [I|Acc]); -peep([{test,is_boolean,{f,Fail},Ops}|_]=Is, SeenTests, - [{test,is_atom,{f,Fail},Ops}|Acc]) -> - %% The previous is_atom/2 test (with the same failure label) is redundant. - %% (If is_boolean(Src) is true, is_atom(Src) is also true, so it is - %% OK to still remember that we have seen is_atom/1.) - peep(Is, SeenTests, Acc); +peep([{jump,{f,L}},{label,L}=I|Is], _, Acc) -> + %% Sometimes beam_jump has missed this optimization. + peep(Is, gb_sets:empty(), [I|Acc]); +peep([{select,Op,R,F,Vls0}|Is], SeenTests0, Acc0) -> + case prune_redundant_values(Vls0, F) of + [] -> + %% No values left. Must convert to plain jump. + I = {jump,F}, + peep([I|Is], gb_sets:empty(), Acc0); + [{atom,_}=Value,Lbl] when Op =:= select_val -> + %% Single value left. Convert to regular test and pop redundant tests. + Is1 = [{test,is_eq_exact,F,[R,Value]},{jump,Lbl}|Is], + case Acc0 of + [{test,is_atom,F,[R]}|Acc] -> + peep(Is1, SeenTests0, Acc); + _ -> + peep(Is1, SeenTests0, Acc0) + end; + [{integer,_}=Value,Lbl] when Op =:= select_val -> + %% Single value left. Convert to regular test and pop redundant tests. + Is1 = [{test,is_eq_exact,F,[R,Value]},{jump,Lbl}|Is], + case Acc0 of + [{test,is_integer,F,[R]}|Acc] -> + peep(Is1, SeenTests0, Acc); + _ -> + peep(Is1, SeenTests0, Acc0) + end; + [Arity,Lbl] when Op =:= select_tuple_arity -> + %% Single value left. Convert to regular test + Is1 = [{test,test_arity,F,[R,Arity]},{jump,Lbl}|Is], + peep(Is1, SeenTests0, Acc0); + [_|_]=Vls -> + I = {select,Op,R,F,Vls}, + peep(Is, gb_sets:empty(), [I|Acc0]) + end; peep([{test,Op,_,Ops}=I|Is], SeenTests0, Acc) -> case beam_utils:is_pure_test(I) of false -> @@ -121,16 +146,6 @@ peep([{test,Op,_,Ops}=I|Is], SeenTests0, Acc) -> peep(Is, SeenTests, [I|Acc]) end end; -peep([{select,select_val,Src,Fail, - [{atom,false},{f,L},{atom,true},{f,L}]}| - [{label,L}|_]=Is], SeenTests, Acc) -> - I = {test,is_boolean,Fail,[Src]}, - peep([I|Is], SeenTests, Acc); -peep([{select,select_val,Src,Fail, - [{atom,true},{f,L},{atom,false},{f,L}]}| - [{label,L}|_]=Is], SeenTests, Acc) -> - I = {test,is_boolean,Fail,[Src]}, - peep([I|Is], SeenTests, Acc); peep([I|Is], _, Acc) -> %% An unknown instruction. Throw away all information we %% have collected about test instructions. @@ -155,3 +170,9 @@ kill_seen_1([{_,Ops}=Test|T], Dst) -> false -> [Test|kill_seen_1(T, Dst)] end; kill_seen_1([], _) -> []. + +prune_redundant_values([_Val,F|Vls], F) -> + prune_redundant_values(Vls, F); +prune_redundant_values([Val,Lbl|Vls], F) -> + [Val,Lbl|prune_redundant_values(Vls, F)]; +prune_redundant_values([], _) -> []. diff --git a/lib/compiler/src/beam_receive.erl b/lib/compiler/src/beam_receive.erl index 7276537949..ddbe67605a 100644 --- a/lib/compiler/src/beam_receive.erl +++ b/lib/compiler/src/beam_receive.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2010-2013. All Rights Reserved. +%% Copyright Ericsson AB 2010-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -65,6 +65,9 @@ %%% as the SomeUniqInteger. %%% +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,Lc}, _Opts) -> Fs = [function(F) || F <- Fs0], Code = {Mod,Exp,Attr,Fs,Lc}, @@ -79,8 +82,7 @@ function({function,Name,Arity,Entry,Is}) -> D = beam_utils:index_labels(Is), {function,Name,Arity,Entry,opt(Is, D, [])} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -177,7 +179,8 @@ opt_recv([I|Is], D, R0, L0, Acc) -> no; false -> opt_recv(Is, D, R, L, [I|Acc]) - end. + end; +opt_recv([], _, _, _, _) -> no. opt_update_regs({block,Bl}, R, L) -> {opt_update_regs_bl(Bl, R),L}; @@ -203,6 +206,8 @@ opt_update_regs({label,Lbl}, R, L) -> %% A catch label for a previously seen catch instruction is OK. {R,L} end; +opt_update_regs({'try',_,{f,Lbl}}, R, L) -> + {R,gb_sets:add(Lbl, L)}; opt_update_regs({try_end,_}, R, L) -> {R,L}; opt_update_regs({line,_}, R, L) -> diff --git a/lib/compiler/src/beam_record.erl b/lib/compiler/src/beam_record.erl new file mode 100644 index 0000000000..58a6de6775 --- /dev/null +++ b/lib/compiler/src/beam_record.erl @@ -0,0 +1,131 @@ +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2014-2017. All Rights Reserved. +%% +%% The contents of this file are subject to the Erlang Public License, +%% Version 1.1, (the "License"); you may not use this file except in +%% compliance with the License. You should have received a copy of the +%% Erlang Public License along with this software. If not, it can be +%% retrieved online at http://www.erlang.org/. +%% +%% Software distributed under the License is distributed on an "AS IS" +%% basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See +%% the License for the specific language governing rights and limitations +%% under the License. +%% +%% %CopyrightEnd% +%% + +%% Rewrite the instruction stream on tagged tuple tests. +%% Tagged tuples means a tuple of any arity with an atom as its +%% first element, such as records and error tuples. +%% +%% From: +%% ... +%% {test,is_tuple,Fail,[Src]}. +%% {test,test_arity,Fail,[Src,Sz]}. +%% ... +%% {get_tuple_element,Src,0,Dst}. +%% ... +%% {test,is_eq_exact,Fail,[Dst,Atom]}. +%% ... +%% To: +%% ... +%% {test,is_tagged_tuple,Fail,[Src,Sz,Atom]}. +%% ... +%% + +-module(beam_record). +-export([module/2]). + +-import(lists, [reverse/1,reverse/2]). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + +module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> + Fs = [function(F) || F <- Fs0], + {ok,{Mod,Exp,Attr,Fs,Lc}}. + +function({function,Name,Arity,CLabel,Is0}) -> + try + Is1 = beam_utils:anno_defs(Is0), + Idx = beam_utils:index_labels(Is1), + Is = rewrite(reverse(Is1), Idx), + {function,Name,Arity,CLabel,Is} + catch + Class:Error:Stack -> + io:fwrite("Function: ~w/~w\n", [Name,Arity]), + erlang:raise(Class, Error, Stack) + end. + +rewrite(Is, Idx) -> + rewrite(Is, Idx, 0, []). + +rewrite([{test,test_arity,Fail,[Src,N]}=TA, + {test,is_tuple,Fail,[Src]}=TT|Is], Idx, Def, Acc0) -> + case is_tagged_tuple(Acc0, Def, Fail, Src, Idx) of + no -> + rewrite(Is, Idx, 0, [TT,TA|Acc0]); + {yes,Atom,Acc} -> + I = {test,is_tagged_tuple,Fail,[Src,N,Atom]}, + rewrite(Is, Idx, Def, [I|Acc]) + end; +rewrite([{block,[{'%anno',{def,Def}}|Bl]}|Is], Idx, _Def, Acc) -> + rewrite(Is, Idx, Def, [{block,Bl}|Acc]); +rewrite([{label,L}=I|Is], Idx0, Def, Acc) -> + Idx = beam_utils:index_label(L, Acc, Idx0), + rewrite(Is, Idx, Def, [I|Acc]); +rewrite([I|Is], Idx, Def, Acc) -> + rewrite(Is, Idx, Def, [I|Acc]); +rewrite([], _, _, Acc) -> Acc. + +is_tagged_tuple([{block,Bl}, + {test,is_eq_exact,Fail,[Dst,{atom,_}=Atom]}|Is], + Def, Fail, Src, Idx) -> + case is_tagged_tuple_1(Bl, Is, Fail, Src, Dst, Idx, Def, []) of + no -> + no; + {yes,[]} -> + {yes,Atom,Is}; + {yes,[_|_]=Block} -> + {yes,Atom,[{block,Block}|Is]} + end; +is_tagged_tuple(_, _, _, _, _) -> + no. + +is_tagged_tuple_1([{set,[Dst],[Src],{get_tuple_element,0}}=I|Bl], + Is, Fail, Src, Dst, Idx, Def, Acc) -> + %% Check usage of Dst to find out whether the get_tuple_element + %% is needed. + case usage(Dst, Is, Fail, Idx) of + killed -> + %% Safe to remove the get_tuple_element instruction. + {yes,reverse(Acc, Bl)}; + used -> + %% Actively used. Must keep instruction. + {yes,reverse(Acc, [I|Bl])}; + not_used -> + %% Not actually used (but must be initialized). + case is_defined(Dst, Def) of + false -> + %% Dst must be initialized, but the + %% actual value does not matter. + Kill = {set,[Dst],[nil],move}, + {yes,reverse(Acc, [Kill|Bl])}; + true -> + %% The register is previously initialized. + %% We can remove the instruction. + {yes,reverse(Acc, Bl)} + end + end; +is_tagged_tuple_1([I|Bl], Is, Fail, Src, Dst, Idx, Def, Acc) -> + is_tagged_tuple_1(Bl, Is, Fail, Src, Dst, Idx, Def, [I|Acc]); +is_tagged_tuple_1(_, _, _, _, _, _, _, _) -> + no. + +usage(Dst, Is, Fail, Idx) -> + beam_utils:usage(Dst, [{test,is_number,Fail,[nil]}|Is], Idx). + +is_defined({x,X}, Def) -> + (Def bsr X) band 1 =:= 1. diff --git a/lib/compiler/src/beam_reorder.erl b/lib/compiler/src/beam_reorder.erl new file mode 100644 index 0000000000..8d2ef5a431 --- /dev/null +++ b/lib/compiler/src/beam_reorder.erl @@ -0,0 +1,150 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% + +-module(beam_reorder). + +-export([module/2]). +-import(lists, [member/2,reverse/1]). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + +module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> + Fs = [function(F) || F <- Fs0], + {ok,{Mod,Exp,Attr,Fs,Lc}}. + +function({function,Name,Arity,CLabel,Is0}) -> + try + Is = reorder(Is0), + {function,Name,Arity,CLabel,Is} + catch + Class:Error:Stack -> + io:fwrite("Function: ~w/~w\n", [Name,Arity]), + erlang:raise(Class, Error, Stack) + end. + +%% reorder(Instructions0) -> Instructions +%% Reorder instructions before the beam_block pass, because reordering +%% will be more cumbersome when the blocks are in place. +%% +%% Execution of get_tuple_element instructions can be delayed until +%% they are actually needed. Consider the sequence: +%% +%% get_tuple_element Tuple Pos Dst +%% test Test Fail Operands +%% +%% If Dst is killed at label Fail (and not referenced in Operands), +%% we can can swap the instructions: +%% +%% test Test Fail Operands +%% get_tuple_element Tuple Pos Dst +%% +%% That can be beneficial in two ways: Firstly, if the branch is taken +%% we have avoided execution of the get_tuple_element instruction. +%% Secondly, even if the branch is not taken, subsequent optimization +%% (opt_blocks/1) may be able to change Dst to the final destination +%% register and eliminate a 'move' instruction. + +reorder(Is) -> + D = beam_utils:index_labels(Is), + reorder_1(Is, D, []). + +reorder_1([{Op,_,_}=TryCatch|[I|Is]=Is0], D, Acc) + when Op =:= 'catch'; Op =:= 'try' -> + %% Don't allow 'try' or 'catch' instructions to split blocks if + %% it can be avoided. + case is_safe(I) of + false -> + reorder_1(Is0, D, [TryCatch|Acc]); + true -> + reorder_1([TryCatch|Is], D, [I|Acc]) + end; +reorder_1([{label,L}=I|_], D, Acc) -> + Is = beam_utils:code_at(L, D), + reorder_1(Is, D, [I|Acc]); +reorder_1([{test,is_nonempty_list,_,_}=I|Is], D, Acc) -> + %% The run-time system may combine the is_nonempty_list test with + %% the following get_list instruction. + reorder_1(Is, D, [I|Acc]); +reorder_1([{test,_,_,_}=I, + {select,_,_,_,_}=S|Is], D, Acc) -> + %% There is nothing to gain by inserting a get_tuple_element + %% instruction between the test instruction and the select + %% instruction. + reorder_1(Is, D, [S,I|Acc]); +reorder_1([{test,_,{f,_},[Src|_]}=I|Is], D, + [{get_tuple_element,Src,_,_}|_]=Acc) -> + %% We want to avoid code that can confuse beam_validator such as: + %% is_tuple Fail Src + %% test_arity Fail Src Arity + %% is_map Fail Src + %% get_tuple_element Src Pos Dst + %% Therefore, don't reorder the instructions in such cases. + reorder_1(Is, D, [I|Acc]); +reorder_1([{test,_,{f,L},Ss}=I|Is0], D0, + [{get_tuple_element,_,_,El}=G|Acc0]=Acc) -> + case member(El, Ss) of + true -> + reorder_1(Is0, D0, [I|Acc]); + false -> + case beam_utils:is_killed_at(El, L, D0) of + true -> + Is = [I,G|Is0], + reorder_1(Is, D0, Acc0); + false -> + case beam_utils:is_killed(El, Is0, D0) of + true -> + Code0 = beam_utils:code_at(L, D0), + Code = [G|Code0], + D = beam_utils:index_label(L, Code, D0), + Is = [I|Is0], + reorder_1(Is, D, Acc0); + false -> + reorder_1(Is0, D0, [I|Acc]) + end + end + end; +reorder_1([{allocate_zero,N,Live}=I0|Is], D, + [{get_tuple_element,{x,Tup},_,{x,Dst}}=G|Acc]=Acc0) -> + case Tup < Dst andalso Dst+1 =:= Live of + true -> + %% Move allocation instruction upwards past + %% get_tuple_element instructions to create more + %% opportunities for moving get_tuple_element + %% instructions. + I = {allocate_zero,N,Dst}, + reorder_1([I,G|Is], D, Acc); + false -> + reorder_1(Is, D, [I0|Acc0]) + end; +reorder_1([I|Is], D, Acc) -> + reorder_1(Is, D, [I|Acc]); +reorder_1([], _, Acc) -> reverse(Acc). + +%% is_safe(Instruction) -> true|false +%% Test whether an instruction is safe (cannot cause an exception). + +is_safe({kill,_}) -> true; +is_safe({move,_,_}) -> true; +is_safe({put,_}) -> true; +is_safe({put_list,_,_,_}) -> true; +is_safe({put_tuple,_,_}) -> true; +is_safe({test_heap,_,_}) -> true; +is_safe(_) -> false. diff --git a/lib/compiler/src/beam_split.erl b/lib/compiler/src/beam_split.erl index 3be9311080..809e49b3d0 100644 --- a/lib/compiler/src/beam_split.erl +++ b/lib/compiler/src/beam_split.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2011. All Rights Reserved. +%% Copyright Ericsson AB 2011-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -23,6 +23,9 @@ -import(lists, [reverse/1]). +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,Lc}, _Opts) -> Fs = [split_blocks(F) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. @@ -47,6 +50,9 @@ split_block([{set,[R],[_,_,_]=As,{bif,is_record,{f,Lbl}}}|Is], Bl, Acc) -> split_block(Is, [], [{bif,is_record,{f,Lbl},As,R}|make_block(Bl, Acc)]); split_block([{set,[R],As,{bif,N,{f,Lbl}=Fail}}|Is], Bl, Acc) when Lbl =/= 0 -> split_block(Is, [], [{bif,N,Fail,As,R}|make_block(Bl, Acc)]); +split_block([{set,[],[],{line,_}=Line}, + {set,[R],As,{bif,raise,{f,_}=Fail}}|Is], Bl, Acc) -> + split_block(Is, [], [{bif,raise,Fail,As,R},Line|make_block(Bl, Acc)]); split_block([{set,[R],As,{alloc,Live,{gc_bif,N,{f,Lbl}=Fail}}}|Is], Bl, Acc) when Lbl =/= 0 -> split_block(Is, [], [{gc_bif,N,Fail,Live,As,R}|make_block(Bl, Acc)]); @@ -54,13 +60,8 @@ split_block([{set,[D],[S|Puts],{alloc,R,{put_map,Op,{f,Lbl}=Fail}}}|Is], Bl, Acc) when Lbl =/= 0 -> split_block(Is, [], [{put_map,Fail,Op,S,D,R,{list,Puts}}| make_block(Bl, Acc)]); -split_block([{set,Ds,[S|Ss],{get_map_elements,Fail}}|Is], Bl, Acc) -> - Gets = beam_utils:join_even(Ss,Ds), - split_block(Is, [], [{get_map_elements,Fail,S,{list,Gets}}|make_block(Bl, Acc)]); -split_block([{set,[R],[],{'catch',L}}|Is], Bl, Acc) -> - split_block(Is, [], [{'catch',R,L}|make_block(Bl, Acc)]); -split_block([{set,[],[],{line,_}=Line}|Is], Bl, Acc) -> - split_block(Is, [], [Line|make_block(Bl, Acc)]); +split_block([{set,[R],[],{try_catch,Op,L}}|Is], Bl, Acc) -> + split_block(Is, [], [{Op,R,L}|make_block(Bl, Acc)]); split_block([I|Is], Bl, Acc) -> split_block(Is, [I|Bl], Acc); split_block([], Bl, Acc) -> make_block(Bl, Acc). diff --git a/lib/compiler/src/beam_trim.erl b/lib/compiler/src/beam_trim.erl index 509e013b62..4da0985085 100644 --- a/lib/compiler/src/beam_trim.erl +++ b/lib/compiler/src/beam_trim.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2007-2013. All Rights Reserved. +%% Copyright Ericsson AB 2007-2016. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -24,10 +24,13 @@ -import(lists, [reverse/1,reverse/2,splitwith/2,sort/1]). -record(st, - {safe, %Safe labels. - lbl %Code at each label. + {safe :: gb_sets:set(beam_asm:label()), %Safe labels. + lbl :: beam_utils:code_index() %Code at each label. }). +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. + module({Mod,Exp,Attr,Fs0,Lc}, _Opts) -> Fs = [function(F) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. @@ -230,7 +233,7 @@ safe_labels([], Acc) -> gb_sets:from_list(Acc). frame_layout(Is, Kills, #st{safe=Safe,lbl=D}) -> N = frame_size(Is, Safe), - IsKilled = fun(R) -> beam_utils:is_killed(R, Is, D) end, + IsKilled = fun(R) -> beam_utils:is_not_used(R, Is, D) end, {N,frame_layout_1(Kills, 0, N, IsKilled, [])}. frame_layout_1([{kill,{y,Y}}=I|Ks], Y, N, IsKilled, Acc) -> diff --git a/lib/compiler/src/beam_type.erl b/lib/compiler/src/beam_type.erl index 5298589f83..b5c979e529 100644 --- a/lib/compiler/src/beam_type.erl +++ b/lib/compiler/src/beam_type.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2013. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -17,13 +17,20 @@ %% %% %CopyrightEnd% %% -%% Purpose : Type-based optimisations. +%% Purpose: Type-based optimisations. See the comment for verified_type/1 +%% the very end of this file for a description of the types in the +%% type database. -module(beam_type). -export([module/2]). --import(lists, [foldl/3,reverse/1,filter/2]). +-import(lists, [foldl/3,member/2,reverse/1,reverse/2,sort/1]). + +-define(UNICODE_INT, {integer,{0,16#10FFFF}}). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. module({Mod,Exp,Attr,Fs0,Lc}, _Opts) -> Fs = [function(F) || F <- Fs0], @@ -33,11 +40,11 @@ function({function,Name,Arity,CLabel,Asm0}) -> try Asm1 = beam_utils:live_opt(Asm0), Asm2 = opt(Asm1, [], tdb_new()), - Asm = beam_utils:delete_live_annos(Asm2), + Asm3 = beam_utils:live_opt(Asm2), + Asm = beam_utils:delete_annos(Asm3), {function,Name,Arity,CLabel,Asm} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -74,75 +81,154 @@ simplify(Is0, TypeDb0) -> %% Basic simplification, mostly tuples, no floating point optimizations. simplify_basic(Is, Ts) -> - simplify_basic_1(Is, Ts, []). - -simplify_basic_1([{set,[D],[{integer,Index},Reg],{bif,element,_}}=I0|Is], Ts0, Acc) -> - I = case max_tuple_size(Reg, Ts0) of - Sz when 0 < Index, Index =< Sz -> - {set,[D],[Reg],{get_tuple_element,Index-1}}; - _Other -> I0 - end, - Ts = update(I, Ts0), - simplify_basic_1(Is, Ts, [I|Acc]); -simplify_basic_1([{set,[D],[TupleReg],{get_tuple_element,0}}=I|Is0], Ts0, Acc) -> - case tdb_find(TupleReg, Ts0) of - {tuple,_,[Contents]} -> - simplify_basic_1([{set,[D],[Contents],move}|Is0], Ts0, Acc); - _ -> - Ts = update(I, Ts0), - simplify_basic_1(Is0, Ts, [I|Acc]) + simplify_basic(Is, Ts, []). + +simplify_basic([I0|Is], Ts0, Acc) -> + case simplify_instr(I0, Ts0) of + [] -> + simplify_basic(Is, Ts0, Acc); + [I] -> + Ts = update(I, Ts0), + simplify_basic(Is, Ts, [I|Acc]) + end; +simplify_basic([], Ts, Acc) -> + {reverse(Acc),Ts}. + +%% simplify_instr(Instruction, Ts) -> [Instruction]. + +%% Simplify a simple instruction using type information. Return an +%% empty list if the instruction should be removed, or a list with +%% the original or modified instruction. + +simplify_instr({set,[D],[{integer,Index},Reg],{bif,element,_}}=I, Ts) -> + case max_tuple_size(Reg, Ts) of + Sz when 0 < Index, Index =< Sz -> + [{set,[D],[Reg],{get_tuple_element,Index-1}}]; + _ -> [I] end; -simplify_basic_1([{set,_,_,{'catch',_}}=I|Is], _Ts, Acc) -> - simplify_basic_1(Is, tdb_new(), [I|Acc]); -simplify_basic_1([{test,is_tuple,_,[R]}=I|Is], Ts, Acc) -> +simplify_instr({test,Test,Fail,[R]}=I, Ts) -> case tdb_find(R, Ts) of - {tuple,_,_} -> simplify_basic_1(Is, Ts, Acc); - _ -> simplify_basic_1(Is, Ts, [I|Acc]) + any -> + [I]; + Type -> + case will_succeed(Test, Type) of + yes -> []; + no -> [{jump,Fail}]; + maybe -> [I] + end end; -simplify_basic_1([{test,test_arity,_,[R,Arity]}=I|Is], Ts0, Acc) -> - case tdb_find(R, Ts0) of - {tuple,Arity,_} -> - simplify_basic_1(Is, Ts0, Acc); - _Other -> - Ts = update(I, Ts0), - simplify_basic_1(Is, Ts, [I|Acc]) +simplify_instr({set,[D],[TupleReg],{get_tuple_element,0}}=I, Ts) -> + case tdb_find(TupleReg, Ts) of + {tuple,_,_,[Contents]} -> + [{set,[D],[Contents],move}]; + _ -> + [I] end; -simplify_basic_1([{test,is_map,_,[R]}=I|Is], Ts0, Acc) -> - case tdb_find(R, Ts0) of - map -> simplify_basic_1(Is, Ts0, Acc); - _Other -> - Ts = update(I, Ts0), - simplify_basic_1(Is, Ts, [I|Acc]) +simplify_instr({test,test_arity,_,[R,Arity]}=I, Ts) -> + case tdb_find(R, Ts) of + {tuple,exact_size,Arity,_} -> []; + _ -> [I] end; -simplify_basic_1([{test,is_nonempty_list,_,[R]}=I|Is], Ts0, Acc) -> - case tdb_find(R, Ts0) of - nonempty_list -> simplify_basic_1(Is, Ts0, Acc); - _Other -> - Ts = update(I, Ts0), - simplify_basic_1(Is, Ts, [I|Acc]) - end; -simplify_basic_1([{test,is_eq_exact,Fail,[R,{atom,_}=Atom]}=I|Is0], Ts0, Acc0) -> - Acc = case tdb_find(R, Ts0) of - {atom,_}=Atom -> Acc0; - {atom,_} -> [{jump,Fail}|Acc0]; - _ -> [I|Acc0] - end, - Ts = update(I, Ts0), - simplify_basic_1(Is0, Ts, Acc); -simplify_basic_1([{test,is_record,_,[R,{atom,_}=Tag,{integer,Arity}]}=I|Is], Ts0, Acc) -> - case tdb_find(R, Ts0) of - {tuple,Arity,[Tag]} -> - simplify_basic_1(Is, Ts0, Acc); - _Other -> - Ts = update(I, Ts0), - simplify_basic_1(Is, Ts, [I|Acc]) +simplify_instr({test,is_eq_exact,Fail,[R,{atom,A}=Atom]}=I, Ts) -> + case tdb_find(R, Ts) of + {atom,_}=Atom -> []; + boolean when is_boolean(A) -> [I]; + any -> [I]; + _ -> [{jump,Fail}] end; -simplify_basic_1([I|Is], Ts0, Acc) -> - Ts = update(I, Ts0), - simplify_basic_1(Is, Ts, [I|Acc]); -simplify_basic_1([], Ts, Acc) -> - Is = reverse(Acc), - {Is,Ts}. +simplify_instr({test,is_record,_,[R,{atom,_}=Tag,{integer,Arity}]}=I, Ts) -> + case tdb_find(R, Ts) of + {tuple,exact_size,Arity,[Tag]} -> []; + _ -> [I] + end; +simplify_instr({select,select_val,Reg,_,_}=I, Ts) -> + [case tdb_find(Reg, Ts) of + {integer,Range} -> + simplify_select_val_int(I, Range); + boolean -> + simplify_select_val_bool(I); + _ -> + I + end]; +simplify_instr({test,bs_test_unit,_,[Src,Unit]}=I, Ts) -> + case tdb_find(Src, Ts) of + {binary,U} when U rem Unit =:= 0 -> []; + _ -> [I] + end; +simplify_instr(I, _) -> [I]. + +simplify_select_val_int({select,select_val,R,_,L0}=I, {Min,Max}) -> + Vs = sort([V || {integer,V} <- L0]), + case eq_ranges(Vs, Min, Max) of + false -> I; + true -> simplify_select_val_1(L0, {integer,Max}, R, []) + end. + +simplify_select_val_bool({select,select_val,R,_,L}=I) -> + Vs = sort([V || {atom,V} <- L]), + case Vs of + [false,true] -> + simplify_select_val_1(L, {atom,false}, R, []); + _ -> + I + end. + +simplify_select_val_1([Val,F|T], Val, R, Acc) -> + L = reverse(Acc, T), + {select,select_val,R,F,L}; +simplify_select_val_1([V,F|T], Val, R, Acc) -> + simplify_select_val_1(T, Val, R, [F,V|Acc]). + +eq_ranges([H], H, H) -> true; +eq_ranges([H|T], H, Max) -> eq_ranges(T, H+1, Max); +eq_ranges(_, _, _) -> false. + +%% will_succeed(TestOperation, Type) -> yes|no|maybe. +%% Test whether TestOperation applied to an argument of type Type +%% will succeed. Return yes, no, or maybe. +%% +%% Type is a type as described in the comment for verified_type/1 at +%% the very end of this file, but it will *never* be 'any'. + +will_succeed(is_atom, Type) -> + case Type of + {atom,_} -> yes; + boolean -> yes; + _ -> no + end; +will_succeed(is_binary, Type) -> + case Type of + {binary,U} when U rem 8 =:= 0 -> yes; + {binary,_} -> maybe; + _ -> no + end; +will_succeed(is_bitstr, Type) -> + case Type of + {binary,_} -> yes; + _ -> no + end; +will_succeed(is_integer, Type) -> + case Type of + integer -> yes; + {integer,_} -> yes; + _ -> no + end; +will_succeed(is_map, Type) -> + case Type of + map -> yes; + _ -> no + end; +will_succeed(is_nonempty_list, Type) -> + case Type of + nonempty_list -> yes; + _ -> no + end; +will_succeed(is_tuple, Type) -> + case Type of + {tuple,_,_,_} -> yes; + _ -> no + end; +will_succeed(_, _) -> maybe. %% simplify_float([Instruction], TypeDatabase) -> %% {[Instruction],TypeDatabase'} | not_possible @@ -173,7 +259,7 @@ simplify_float_1([{set,[D0],[A0],{alloc,_,{gc_bif,'-',{f,0}}}}=I|Is]=Is0, {D,Rs} = find_dest(D0, Rs1), Areg = fetch_reg(A, Rs), Acc = [{set,[D],[Areg],{bif,fnegate,{f,0}}}|clearerror(Acc1)], - Ts = tdb_update([{D0,float}], Ts0), + Ts = tdb_store(D0, float, Ts0), simplify_float_1(Is, Ts, Rs, Acc); _Other -> Ts = update(I, Ts0), @@ -196,10 +282,10 @@ simplify_float_1([{set,[D0],[A0,B0],{alloc,_,{gc_bif,Op0,{f,0}}}}=I|Is]=Is0, Areg = fetch_reg(A, Rs), Breg = fetch_reg(B, Rs), Acc = [{set,[D],[Areg,Breg],{bif,Op,{f,0}}}|clearerror(Acc2)], - Ts = tdb_update([{D0,float}], Ts0), + Ts = tdb_store(D0, float, Ts0), simplify_float_1(Is, Ts, Rs, Acc) end; -simplify_float_1([{set,_,_,{'catch',_}}=I|Is]=Is0, _Ts, Rs0, Acc0) -> +simplify_float_1([{set,_,_,{try_catch,_,_}}=I|Is]=Is0, _Ts, Rs0, Acc0) -> Acc = flush_all(Rs0, Is0, Acc0), simplify_float_1(Is, tdb_new(), Rs0, [I|Acc]); simplify_float_1([{set,_,_,{line,_}}=I|Is], Ts, Rs, Acc) -> @@ -247,7 +333,7 @@ clearerror([], OrigIs) -> [{set,[],[],fclearerror}|OrigIs]. %% Combine two blocks and eliminate any move instructions that assign %% to registers that are killed later in the block. %% -merge_blocks(B1, [{'%live',_,_}|B2]) -> +merge_blocks(B1, [{'%anno',_}|B2]) -> merge_blocks_1(B1++[{set,[],[],stop_here}|B2]). merge_blocks_1([{set,[],_,stop_here}|Is]) -> Is; @@ -296,27 +382,17 @@ flt_need_heap_2({set,_,_,{put_tuple,_}}, H, Fl) -> {[],H+1,Fl}; flt_need_heap_2({set,_,_,put}, H, Fl) -> {[],H+1,Fl}; -%% Then the "neutral" instructions. We just pass them. -flt_need_heap_2({set,[{fr,_}],_,_}, H, Fl) -> - {[],H,Fl}; -flt_need_heap_2({set,[],[],fclearerror}, H, Fl) -> - {[],H,Fl}; -flt_need_heap_2({set,[],[],fcheckerror}, H, Fl) -> - {[],H,Fl}; -flt_need_heap_2({set,_,_,{bif,_,_}}, H, Fl) -> - {[],H,Fl}; -flt_need_heap_2({set,_,_,move}, H, Fl) -> - {[],H,Fl}; -flt_need_heap_2({set,_,_,{get_tuple_element,_}}, H, Fl) -> - {[],H,Fl}; -flt_need_heap_2({set,_,_,get_list}, H, Fl) -> - {[],H,Fl}; -flt_need_heap_2({set,_,_,{'catch',_}}, H, Fl) -> - {[],H,Fl}; -%% All other instructions should cause the insertion of an allocation +%% The following instructions cause the insertion of an allocation %% instruction if needed. +flt_need_heap_2({set,_,_,{alloc,_,_}}, H, Fl) -> + {flt_alloc(H, Fl),0,0}; +flt_need_heap_2({set,_,_,{set_tuple_element,_}}, H, Fl) -> + {flt_alloc(H, Fl),0,0}; +flt_need_heap_2({'%anno',_}, H, Fl) -> + {flt_alloc(H, Fl),0,0}; +%% All other instructions are "neutral". We just pass them. flt_need_heap_2(_, H, Fl) -> - {flt_alloc(H, Fl),0,0}. + {[],H,Fl}. flt_alloc(0, 0) -> []; @@ -339,7 +415,7 @@ build_alloc(Words, Floats) -> {alloc,[{words,Words},{floats,Floats}]}. %% is not continous at an allocation function (e.g. if {x,0} and {x,2} %% are live, but not {x,1}). -flt_liveness([{'%live',_Live,Regs}=LiveInstr|Is]) -> +flt_liveness([{'%anno',{used,Regs}}=LiveInstr|Is]) -> flt_liveness_1(Is, Regs, [LiveInstr]). flt_liveness_1([{set,Ds,Ss,{alloc,Live0,Alloc}}|Is], Regs0, Acc) -> @@ -351,7 +427,7 @@ flt_liveness_1([{set,Ds,Ss,{alloc,Live0,Alloc}}|Is], Regs0, Acc) -> flt_liveness_1([{set,Ds,_,_}=I|Is], Regs0, Acc) -> Regs = x_live(Ds, Regs0), flt_liveness_1(Is, Regs, [I|Acc]); -flt_liveness_1([{'%live',_,_}], _Regs, Acc) -> +flt_liveness_1([{'%anno',_}], _Regs, Acc) -> reverse(Acc). init_regs(Live) -> @@ -375,88 +451,217 @@ x_live([], Regs) -> Regs. %% Update the type database to account for executing an instruction. %% %% First the cases for instructions inside basic blocks. -update({'%live',_,_}, Ts) -> Ts; +update({'%anno',_}, Ts) -> + Ts; update({set,[D],[S],move}, Ts) -> tdb_copy(S, D, Ts); -update({set,[D],[{integer,I},Reg],{bif,element,_}}, Ts0) -> - tdb_update([{Reg,{tuple,I,[]}},{D,kill}], Ts0); -update({set,[D],[_Index,Reg],{bif,element,_}}, Ts0) -> - tdb_update([{Reg,{tuple,0,[]}},{D,kill}], Ts0); -update({set,[D],[S],{get_tuple_element,0}}, Ts) -> - tdb_update([{D,{tuple_element,S,0}}], Ts); +update({set,[D],[Index,Reg],{bif,element,_}}, Ts0) -> + MinSize = case Index of + {integer,I} -> I; + _ -> 0 + end, + Ts = tdb_meet(Reg, {tuple,min_size,MinSize,[]}, Ts0), + tdb_store(D, any, Ts); +update({set,[D],[_Key,Map],{bif,map_get,_}}, Ts0) -> + Ts = tdb_meet(Map, map, Ts0), + tdb_store(D, any, Ts); +update({set,[D],Args,{bif,N,_}}, Ts) -> + Ar = length(Args), + BoolOp = erl_internal:new_type_test(N, Ar) orelse + erl_internal:comp_op(N, Ar) orelse + erl_internal:bool_op(N, Ar), + Type = case BoolOp of + true -> boolean; + false -> unary_op_type(N) + end, + tdb_store(D, Type, Ts); +update({set,[D],[S],{get_tuple_element,0}}, Ts0) -> + if + D =:= S -> + tdb_store(D, any, Ts0); + true -> + Ts = tdb_store(D, {tuple_element,S,0}, Ts0), + tdb_store(S, {tuple,min_size,1,[]}, Ts) + end; update({set,[D],[S],{alloc,_,{gc_bif,float,{f,0}}}}, Ts0) -> %% Make sure we reject non-numeric literal argument. case possibly_numeric(S) of - true -> tdb_update([{D,float}], Ts0); - false -> Ts0 + true -> tdb_store(D, float, Ts0); + false -> Ts0 end; -update({set,[D],[S1,S2],{alloc,_,{gc_bif,'/',{f,0}}}}, Ts0) -> +update({set,[D],[S1,S2],{alloc,_,{gc_bif,'band',{f,0}}}}, Ts) -> + Type = band_type(S1, S2, Ts), + tdb_store(D, Type, Ts); +update({set,[D],[S1,S2],{alloc,_,{gc_bif,'/',{f,0}}}}, Ts) -> %% Make sure we reject non-numeric literals. case possibly_numeric(S1) andalso possibly_numeric(S2) of - true -> tdb_update([{D,float}], Ts0); - false -> Ts0 + true -> tdb_store(D, float, Ts); + false -> Ts end; update({set,[D],[S1,S2],{alloc,_,{gc_bif,Op,{f,0}}}}, Ts0) -> - case arith_op(Op) of - no -> - tdb_update([{D,kill}], Ts0); - {yes,_} -> - case {tdb_find(S1, Ts0),tdb_find(S2, Ts0)} of - {float,_} -> tdb_update([{D,float}], Ts0); - {_,float} -> tdb_update([{D,float}], Ts0); - {_,_} -> tdb_update([{D,kill}], Ts0) - end - end; -update({set,[],_Src,_Op}, Ts0) -> Ts0; -update({set,[D],_Src,_Op}, Ts0) -> - tdb_update([{D,kill}], Ts0); -update({set,[D1,D2],_Src,_Op}, Ts0) -> - tdb_update([{D1,kill},{D2,kill}], Ts0); + case op_type(Op) of + integer -> + tdb_store(D, integer, Ts0); + {float,_} -> + case {tdb_find(S1, Ts0),tdb_find(S2, Ts0)} of + {float,_} -> tdb_store(D, float, Ts0); + {_,float} -> tdb_store(D, float, Ts0); + {_,_} -> tdb_store(D, any, Ts0) + end; + Type -> + tdb_store(D, Type, Ts0) + end; +update({set,[D],[_],{alloc,_,{gc_bif,Op,{f,0}}}}, Ts) -> + tdb_store(D, unary_op_type(Op), Ts); +update({set,[],_Src,_Op}, Ts) -> + Ts; +update({set,[D],_Src,_Op}, Ts) -> + tdb_store(D, any, Ts); update({kill,D}, Ts) -> - tdb_update([{D,kill}], Ts); + tdb_store(D, any, Ts); %% Instructions outside of blocks. -update({test,is_float,_Fail,[Src]}, Ts0) -> - tdb_update([{Src,float}], Ts0); -update({test,test_arity,_Fail,[Src,Arity]}, Ts0) -> - tdb_update([{Src,{tuple,Arity,[]}}], Ts0); -update({test,is_map,_Fail,[Src]}, Ts0) -> - tdb_update([{Src,map}], Ts0); -update({test,is_nonempty_list,_Fail,[Src]}, Ts0) -> - tdb_update([{Src,nonempty_list}], Ts0); -update({test,is_eq_exact,_,[Reg,{atom,_}=Atom]}, Ts) -> - case tdb_find(Reg, Ts) of - error -> - Ts; - {tuple_element,TupleReg,0} -> - tdb_update([{TupleReg,{tuple,1,[Atom]}}], Ts); - _ -> - Ts - end; +update({test,test_arity,_Fail,[Src,Arity]}, Ts) -> + tdb_meet(Src, {tuple,exact_size,Arity,[]}, Ts); +update({get_map_elements,_,Src,{list,Elems0}}, Ts0) -> + Ts1 = tdb_meet(Src, map, Ts0), + {_Ss,Ds} = beam_utils:split_even(Elems0), + foldl(fun(Dst, A) -> tdb_store(Dst, any, A) end, Ts1, Ds); +update({test,is_eq_exact,_,[Reg,{atom,_}=Atom]}, Ts0) -> + Ts = case tdb_find_source_tuple(Reg, Ts0) of + {source_tuple,TupleReg} -> + tdb_meet(TupleReg, {tuple,min_size,1,[Atom]}, Ts0); + none -> + Ts0 + end, + tdb_meet(Reg, Atom, Ts); update({test,is_record,_Fail,[Src,Tag,{integer,Arity}]}, Ts) -> - tdb_update([{Src,{tuple,Arity,[Tag]}}], Ts); + tdb_meet(Src, {tuple,exact_size,Arity,[Tag]}, Ts); + +%% Binaries and binary matching. + +update({test,bs_get_integer2,_,_,Args,Dst}, Ts) -> + tdb_store(Dst, get_bs_integer_type(Args), Ts); +update({test,bs_get_utf8,_,_,_,Dst}, Ts) -> + tdb_store(Dst, ?UNICODE_INT, Ts); +update({test,bs_get_utf16,_,_,_,Dst}, Ts) -> + tdb_store(Dst, ?UNICODE_INT, Ts); +update({test,bs_get_utf32,_,_,_,Dst}, Ts) -> + tdb_store(Dst, ?UNICODE_INT, Ts); +update({bs_init,_,{bs_init2,_,_},_,_,Dst}, Ts) -> + tdb_store(Dst, {binary,8}, Ts); +update({bs_init,_,_,_,_,Dst}, Ts) -> + tdb_store(Dst, {binary,1}, Ts); +update({bs_put,_,_,_}, Ts) -> + Ts; +update({bs_save2,_,_}, Ts) -> + Ts; +update({bs_restore2,_,_}, Ts) -> + Ts; +update({bs_context_to_binary,Dst}, Ts) -> + tdb_store(Dst, any, Ts); +update({test,bs_start_match2,_,_,[Src,_],Dst}, Ts0) -> + Ts = tdb_meet(Src, {binary,1}, Ts0), + tdb_copy(Src, Dst, Ts); +update({test,bs_get_binary2,_,_,[_,_,Unit,_],Dst}, Ts) -> + true = is_integer(Unit), %Assertion. + tdb_store(Dst, {binary,Unit}, Ts); +update({test,bs_get_float2,_,_,_,Dst}, Ts) -> + tdb_store(Dst, float, Ts); +update({test,bs_test_unit,_,[Src,Unit]}, Ts) -> + tdb_meet(Src, {binary,Unit}, Ts); + +%% Other test instructions +update({test,Test,_Fail,[Src]}, Ts) -> + Type = case Test of + is_binary -> {binary,8}; + is_bitstr -> {binary,1}; + is_boolean -> boolean; + is_float -> float; + is_integer -> integer; + is_map -> map; + is_nonempty_list -> nonempty_list; + _ -> any + end, + tdb_meet(Src, Type, Ts); update({test,_Test,_Fail,_Other}, Ts) -> Ts; + +%% Calls + update({call_ext,Ar,{extfunc,math,Math,Ar}}, Ts) -> case is_math_bif(Math, Ar) of - true -> tdb_update([{{x,0},float}], Ts); + true -> tdb_store({x,0}, float, Ts); false -> tdb_kill_xregs(Ts) end; update({call_ext,3,{extfunc,erlang,setelement,3}}, Ts0) -> - Op = case tdb_find({x,1}, Ts0) of - error -> kill; - Info -> Info - end, - Ts1 = tdb_kill_xregs(Ts0), - tdb_update([{{x,0},Op}], Ts1); + Ts = tdb_kill_xregs(Ts0), + case tdb_find({x,1}, Ts0) of + {tuple,SzKind,Sz,_}=T0 -> + T = case tdb_find({x,0}, Ts0) of + {integer,{I,I}} when I > 1 -> + %% First element is not changed. The result + %% will have the same type. + T0; + _ -> + %% Position is 1 or unknown. May change the + %% first element of the tuple. + {tuple,SzKind,Sz,[]} + end, + tdb_store({x,0}, T, Ts); + _ -> + Ts + end; update({call,_Arity,_Func}, Ts) -> tdb_kill_xregs(Ts); update({call_ext,_Arity,_Func}, Ts) -> tdb_kill_xregs(Ts); update({make_fun2,_,_,_,_}, Ts) -> tdb_kill_xregs(Ts); +update({call_fun, _}, Ts) -> tdb_kill_xregs(Ts); +update({apply, _}, Ts) -> tdb_kill_xregs(Ts); + update({line,_}, Ts) -> Ts; +update({'%',_}, Ts) -> Ts; %% The instruction is unknown. Kill all information. update(_I, _Ts) -> tdb_new(). +band_type({integer,Int}, Other, Ts) -> + band_type_1(Int, Other, Ts); +band_type(Other, {integer,Int}, Ts) -> + band_type_1(Int, Other, Ts); +band_type(_, _, _) -> integer. + +band_type_1(Int, OtherSrc, Ts) -> + Type = band_type_2(Int, 0), + OtherType = tdb_find(OtherSrc, Ts), + meet(Type, OtherType). + +band_type_2(N, Bits) when Bits < 64 -> + case 1 bsl Bits of + P when P =:= N + 1 -> + {integer,{0,N}}; + P when P > N + 1 -> + integer; + _ -> + band_type_2(N, Bits+1) + end; +band_type_2(_, _) -> + %% Negative or large positive number. Give up. + integer. + +get_bs_integer_type([_,{integer,N},U,{field_flags,Fl}]) + when N*U < 64 -> + NumBits = N*U, + case member(unsigned, Fl) of + true -> + {integer,{0,(1 bsl NumBits)-1}}; + false -> + %% Signed integer. Don't bother. + integer + end; +get_bs_integer_type(_) -> + %% Avoid creating ranges with a huge upper limit. + integer. + is_math_bif(cos, 1) -> true; is_math_bif(cosh, 1) -> true; is_math_bif(sin, 1) -> true; @@ -478,6 +683,9 @@ is_math_bif(log10, 1) -> true; is_math_bif(sqrt, 1) -> true; is_math_bif(atan2, 2) -> true; is_math_bif(pow, 2) -> true; +is_math_bif(ceil, 1) -> true; +is_math_bif(floor, 1) -> true; +is_math_bif(fmod, 2) -> true; is_math_bif(pi, 0) -> true; is_math_bif(_, _) -> false. @@ -490,7 +698,7 @@ possibly_numeric(_) -> false. max_tuple_size(Reg, Ts) -> case tdb_find(Reg, Ts) of - {tuple,Sz,_} -> Sz; + {tuple,_,Sz,_} -> Sz; _Other -> 0 end. @@ -545,12 +753,34 @@ load_reg(V, Ts, Rs0, Is0) -> {Rs,Is} end. -arith_op('+') -> {yes,fadd}; -arith_op('-') -> {yes,fsub}; -arith_op('*') -> {yes,fmul}; -arith_op('/') -> {yes,fdiv}; -arith_op(_) -> no. +arith_op(Op) -> + case op_type(Op) of + {float,Instr} -> {yes,Instr}; + _ -> no + end. +op_type('+') -> {float,fadd}; +op_type('-') -> {float,fsub}; +op_type('*') -> {float,fmul}; +%% '/' and 'band' are specially handled. +op_type('bor') -> integer; +op_type('bxor') -> integer; +op_type('bsl') -> integer; +op_type('bsr') -> integer; +op_type('div') -> integer; +op_type(_) -> any. + +unary_op_type(bit_size) -> integer; +unary_op_type(byte_size) -> integer; +unary_op_type(length) -> integer; +unary_op_type(map_size) -> integer; +unary_op_type(size) -> integer; +unary_op_type(tuple_size) -> integer; +unary_op_type(_) -> any. + +flush(Rs, [{set,[_],[_,_,_],{bif,is_record,_}}|_]=Is0, Acc0) -> + Acc = flush_all(Rs, Is0, Acc0), + {[],Acc}; flush(Rs, [{set,[_],[],{put_tuple,_}}|_]=Is0, Acc0) -> Acc = flush_all(Rs, Is0, Acc0), {[],Acc}; @@ -618,7 +848,6 @@ checkerror(Is) -> checkerror_1(Is, Is). checkerror_1([{set,[],[],fcheckerror}|_], OrigIs) -> OrigIs; -checkerror_1([{set,[],[],fclearerror}|_], OrigIs) -> OrigIs; checkerror_1([{set,_,_,{bif,fadd,_}}|_], OrigIs) -> checkerror_2(OrigIs); checkerror_1([{set,_,_,{bif,fsub,_}}|_], OrigIs) -> checkerror_2(OrigIs); checkerror_1([{set,_,_,{bif,fmul,_}}|_], OrigIs) -> checkerror_2(OrigIs); @@ -630,34 +859,39 @@ checkerror_1([], OrigIs) -> OrigIs. checkerror_2(OrigIs) -> [{set,[],[],fcheckerror}|OrigIs]. -%%% Routines for maintaining a type database. The type database +%%% Routines for maintaining a type database. The type database %%% associates type information with registers. %%% -%%% {tuple,Size,First} means that the corresponding register contains a -%%% tuple with *at least* Size elements. An tuple with unknown -%%% size is represented as {tuple,0}. First is either [] (meaning that -%%% the tuple's first element is unknown) or [FirstElement] (the contents -%%% of the first element). -%%% -%%% 'float' means that the register contains a float. +%%% See the comment for verified_type/1 at the end of module for +%%% a description of the possible types. %% tdb_new() -> EmptyDataBase %% Creates a new, empty type database. tdb_new() -> []. -%% tdb_find(Register, Db) -> Information|error +%% tdb_find(Register, Db) -> Type %% Returns type information or the atom error if there is no type %% information available for Register. +%% +%% See the comment for verified_type/1 at the end of module for +%% a description of the possible types. -tdb_find({x,_}=K, Ts) -> tdb_find_1(K, Ts); -tdb_find({y,_}=K, Ts) -> tdb_find_1(K, Ts); -tdb_find(_, _) -> error. +tdb_find(Reg, Ts) -> + case tdb_find_raw(Reg, Ts) of + {tuple_element,_,_} -> any; + Type -> Type + end. -tdb_find_1(K, Ts) -> - case orddict:find(K, Ts) of - {ok,Val} -> Val; - error -> error +%% tdb_find_source_tuple(Register, Ts) -> {source_tuple,Register} | 'none'. +%% Find the tuple whose first element was fetched to the register Register. + +tdb_find_source_tuple(Reg, Ts) -> + case tdb_find_raw(Reg, Ts) of + {tuple_element,Src,0} -> + {source_tuple,Src}; + _ -> + none end. %% tdb_copy(Source, Dest, Db) -> Db' @@ -665,48 +899,110 @@ tdb_find_1(K, Ts) -> %% as the Source. tdb_copy({Tag,_}=S, D, Ts) when Tag =:= x; Tag =:= y -> - case tdb_find(S, Ts) of - error -> orddict:erase(D, Ts); - Type -> orddict:store(D, Type, Ts) - end; -tdb_copy(Literal, D, Ts) -> orddict:store(D, Literal, Ts). - -%% tdb_update([UpdateOp], Db) -> NewDb -%% UpdateOp = {Register,kill}|{Register,NewInfo} -%% Updates a type database. If a 'kill' operation is given, the type -%% information for that register will be removed from the database. -%% A kill operation takes precedence over other operations for the same -%% register (i.e. [{{x,0},kill},{{x,0},{tuple,5}}] means that the -%% the existing type information, if any, will be discarded, and the -%% the '{tuple,5}' information ignored. + case tdb_find_raw(S, Ts) of + any -> orddict:erase(D, Ts); + Type -> orddict:store(D, Type, Ts) + end; +tdb_copy(Literal, D, Ts) -> + Type = case Literal of + {atom,_} -> Literal; + {float,_} -> float; + {integer,Int} -> {integer,{Int,Int}}; + {literal,[_|_]} -> nonempty_list; + {literal,#{}} -> map; + {literal,Tuple} when tuple_size(Tuple) >= 1 -> + Lit = tag_literal(element(1, Tuple)), + {tuple,exact_size,tuple_size(Tuple),[Lit]}; + _ -> any + end, + tdb_store(D, verified_type(Type), Ts). + +%% tdb_store(Register, Type, Ts0) -> Ts. +%% Store a new type for register Register. Return the update type +%% database. Use this function when a new value is assigned to +%% a register. +%% +%% See the comment for verified_type/1 at the end of module for +%% a description of the possible types. + +tdb_store(Reg, any, Ts) -> + erase(Reg, Ts); +tdb_store(Reg, Type, Ts) -> + store(Reg, verified_type(Type), Ts). + +store(Key, New, [{K,_}|_]=Dict) when Key < K -> + [{Key,New}|Dict]; +store(Key, New, [{K,Val}=E|Dict]) when Key > K -> + case Val of + {tuple_element,Key,_} -> store(Key, New, Dict); + _ -> [E|store(Key, New, Dict)] + end; +store(Key, New, [{_K,Old}|Dict]) -> %Key == K + case Old of + {tuple,_,_,_} -> + [{Key,New}|erase_tuple_element(Key, Dict)]; + _ -> + [{Key,New}|Dict] + end; +store(Key, New, []) -> [{Key,New}]. + +erase(Key, [{K,_}=E|Dict]) when Key < K -> + [E|Dict]; +erase(Key, [{K,Val}=E|Dict]) when Key > K -> + case Val of + {tuple_element,Key,_} -> erase(Key, Dict); + _ -> [E|erase(Key, Dict)] + end; +erase(Key, [{_K,Val}|Dict]) -> %Key == K + case Val of + {tuple,_,_,_} -> erase_tuple_element(Key, Dict); + _ -> Dict + end; +erase(_, []) -> []. + +erase_tuple_element(Key, [{_,{tuple_element,Key,_}}|Dict]) -> + erase_tuple_element(Key, Dict); +erase_tuple_element(Key, [E|Dict]) -> + [E|erase_tuple_element(Key, Dict)]; +erase_tuple_element(_Key, []) -> []. + +%% tdb_meet(Register, Type, Ts0) -> Ts. +%% Update information of a register that is used as the source for an +%% instruction. The type Type will be combined using the meet operation +%% with the previous type information for the register, resulting in +%% narrower (more specific) type. +%% +%% For example, if the previous type is {tuple,min_size,2,[]} and the +%% the new type is {tuple,exact_size,5,[]}, the meet of the types will +%% be {tuple,exact_size,5,[]}. %% -%% If NewInfo information is given and there exists information about -%% the register, the old and new type information will be merged. -%% For instance, {tuple,5} and {tuple,10} will be merged to produce -%% {tuple,10}. - -tdb_update(Uis0, Ts0) -> - Uis1 = filter(fun ({{x,_},_Op}) -> true; - ({{y,_},_Op}) -> true; - (_) -> false - end, Uis0), - tdb_update1(lists:sort(Uis1), Ts0). - -tdb_update1([{Key,kill}|Ops], [{K,_Old}|_]=Db) when Key < K -> - tdb_update1(remove_key(Key, Ops), Db); -tdb_update1([{Key,_New}=New|Ops], [{K,_Old}|_]=Db) when Key < K -> - [New|tdb_update1(Ops, Db)]; -tdb_update1([{Key,kill}|Ops], [{Key,_}|Db]) -> - tdb_update1(remove_key(Key, Ops), Db); -tdb_update1([{Key,NewInfo}|Ops], [{Key,OldInfo}|Db]) -> - [{Key,merge_type_info(NewInfo, OldInfo)}|tdb_update1(Ops, Db)]; -tdb_update1([{_,_}|_]=Ops, [Old|Db]) -> - [Old|tdb_update1(Ops, Db)]; -tdb_update1([{Key,kill}|Ops], []) -> - tdb_update1(remove_key(Key, Ops), []); -tdb_update1([{_,_}=New|Ops], []) -> - [New|tdb_update1(Ops, [])]; -tdb_update1([], Db) -> Db. +%% See the comment for verified_type/1 at the end of module for +%% a description of the possible types. + +tdb_meet(Reg, NewType, Ts) -> + Update = fun(Type0) -> meet(Type0, NewType) end, + orddict:update(Reg, Update, NewType, Ts). + +%%% +%%% Here follows internal helper functions for accessing and +%%% updating the type database. +%%% + +tdb_find_raw({x,_}=K, Ts) -> tdb_find_raw_1(K, Ts); +tdb_find_raw({y,_}=K, Ts) -> tdb_find_raw_1(K, Ts); +tdb_find_raw(_, _) -> any. + +tdb_find_raw_1(K, Ts) -> + case orddict:find(K, Ts) of + {ok,Val} -> Val; + error -> any + end. + +tag_literal(A) when is_atom(A) -> {atom,A}; +tag_literal(F) when is_float(F) -> {float,F}; +tag_literal(I) when is_integer(I) -> {integer,I}; +tag_literal([]) -> nil; +tag_literal(Lit) -> {literal,Lit}. %% tdb_kill_xregs(Db) -> NewDb %% Kill all information about x registers. Also kill all tuple_element @@ -716,26 +1012,106 @@ tdb_kill_xregs([{{x,_},_Type}|Db]) -> tdb_kill_xregs(Db); tdb_kill_xregs([{{y,_},{tuple_element,{x,_},_}}|Db]) -> tdb_kill_xregs(Db); tdb_kill_xregs([Any|Db]) -> [Any|tdb_kill_xregs(Db)]; tdb_kill_xregs([]) -> []. - -remove_key(Key, [{Key,_Op}|Ops]) -> remove_key(Key, Ops); -remove_key(_, Ops) -> Ops. - -merge_type_info(I, I) -> I; -merge_type_info({tuple,Sz1,Same}, {tuple,Sz2,Same}=Max) when Sz1 < Sz2 -> + +%% meet(Type1, Type2) -> Type +%% Returns the "meet" of Type1 and Type2. The meet is a narrower +%% type than Type1 and Type2. For example: +%% +%% meet(integer, {integer,{0,3}}) -> {integer,{0,3}} +%% +%% The meet for two different types result in 'none', which is +%% the bottom element for our type lattice: +%% +%% meet(integer, map) -> none + +meet(T, T) -> + T; +meet({integer,_}=T, integer) -> + T; +meet(integer, {integer,_}=T) -> + T; +meet({integer,{Min1,Max1}}, {integer,{Min2,Max2}}) -> + {integer,{max(Min1, Min2),min(Max1, Max2)}}; +meet({tuple,min_size,Sz1,Same}, {tuple,min_size,Sz2,Same}=Max) when Sz1 < Sz2 -> Max; -merge_type_info({tuple,Sz1,Same}=Max, {tuple,Sz2,Same}) when Sz1 > Sz2 -> +meet({tuple,min_size,Sz1,Same}=Max, {tuple,min_size,Sz2,Same}) when Sz1 > Sz2 -> Max; -merge_type_info({tuple,Sz1,[]}, {tuple,_Sz2,First}=Tuple2) -> - merge_type_info({tuple,Sz1,First}, Tuple2); -merge_type_info({tuple,_Sz1,First}=Tuple1, {tuple,Sz2,_}) -> - merge_type_info(Tuple1, {tuple,Sz2,First}); -merge_type_info(NewType, _) -> - verify_type(NewType), - NewType. - -verify_type(map) -> ok; -verify_type(nonempty_list) -> ok; -verify_type({tuple,Sz,[]}) when is_integer(Sz) -> ok; -verify_type({tuple,Sz,[_]}) when is_integer(Sz) -> ok; -verify_type({tuple_element,_,_}) -> ok; -verify_type(float) -> ok. +meet({tuple,exact_size,_,Same}=Exact, {tuple,_,_,Same}) -> + Exact; +meet({tuple,_,_,Same},{tuple,exact_size,_,Same}=Exact) -> + Exact; +meet({tuple,SzKind1,Sz1,[]}, {tuple,_SzKind2,_Sz2,First}=Tuple2) -> + meet({tuple,SzKind1,Sz1,First}, Tuple2); +meet({tuple,_SzKind1,_Sz1,First}=Tuple1, {tuple,SzKind2,Sz2,_}) -> + meet(Tuple1, {tuple,SzKind2,Sz2,First}); +meet({binary,U1}, {binary,U2}) -> + {binary,max(U1, U2)}; +meet(T1, T2) -> + case is_any(T1) of + true -> + verified_type(T2); + false -> + case is_any(T2) of + true -> + verified_type(T1); + false -> + none %The bottom element. + end + end. + +is_any(any) -> true; +is_any({tuple_element,_,_}) -> true; +is_any(_) -> false. + +%% verified_type(Type) -> Type +%% Returns the passed in type if it is one of the defined types. +%% Crashes if there is anything wrong with the type. +%% +%% Here are all possible types: +%% +%% any Any Erlang term (top element for the type lattice). +%% +%% {atom,Atom} The specific atom Atom. +%% {binary,Unit} Binary/bitstring aligned to unit Unit. +%% boolean 'true' | 'false' +%% float Floating point number. +%% integer Integer. +%% {integer,{Min,Max}} Integer in the inclusive range Min through Max. +%% map Map. +%% nonempty_list Nonempty list. +%% {tuple,_,_,_} Tuple (see below). +%% +%% none No type (bottom element for the type lattice). +%% +%% {tuple,min_size,Size,First} means that the corresponding register +%% contains a tuple with *at least* Size elements (conversely, +%% {tuple,exact_size,Size,First} means that it contains a tuple with +%% *exactly* Size elements). An tuple with unknown size is +%% represented as {tuple,min_size,0,[]}. First is either [] (meaning +%% that the tuple's first element is unknown) or [FirstElement] (the +%% contents of the first element). +%% +%% There is also a pseudo-type called {tuple_element,_,_}: +%% +%% {tuple_element,SrcTuple,ElementNumber} +%% +%% that does not provide any information about the type of the +%% register itself, but provides a link back to the source tuple that +%% the register got its value from. +%% +%% Note that {tuple_element,_,_} will *never* be returned by tdb_find/2. +%% Use tdb_find_source_tuple/2 to locate the source tuple for a register. + +verified_type(any=T) -> T; +verified_type({atom,_}=T) -> T; +verified_type({binary,U}=T) when is_integer(U) -> T; +verified_type(boolean=T) -> T; +verified_type(integer=T) -> T; +verified_type({integer,{Min,Max}}=T) + when is_integer(Min), is_integer(Max) -> T; +verified_type(map=T) -> T; +verified_type(nonempty_list=T) -> T; +verified_type({tuple,_,Sz,[]}=T) when is_integer(Sz) -> T; +verified_type({tuple,_,Sz,[_]}=T) when is_integer(Sz) -> T; +verified_type({tuple_element,_,_}=T) -> T; +verified_type(float=T) -> T. diff --git a/lib/compiler/src/beam_utils.erl b/lib/compiler/src/beam_utils.erl index 1945fadfd3..6e23003fc7 100644 --- a/lib/compiler/src/beam_utils.erl +++ b/lib/compiler/src/beam_utils.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2007-2013. All Rights Reserved. +%% Copyright Ericsson AB 2007-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -22,19 +22,62 @@ -module(beam_utils). -export([is_killed_block/2,is_killed/3,is_killed_at/3, - is_not_used/3,is_not_used_at/3, - empty_label_index/0,index_label/3,index_labels/1, + is_not_used/3,usage/3, + empty_label_index/0,index_label/3,index_labels/1,replace_labels/4, code_at/2,bif_to_test/3,is_pure_test/1, - live_opt/1,delete_live_annos/1,combine_heap_needs/2]). + live_opt/1,delete_annos/1,combine_heap_needs/2, + anno_defs/1, + split_even/1 + ]). --export([join_even/2,split_even/1]). +-export_type([code_index/0,module_code/0,instruction/0]). --import(lists, [member/2,sort/1,reverse/1,splitwith/2]). +-import(lists, [flatmap/2,map/2,member/2,sort/1,reverse/1,splitwith/2]). + +-define(is_const(Val), (Val =:= nil orelse + element(1, Val) =:= integer orelse + element(1, Val) =:= float orelse + element(1, Val) =:= atom orelse + element(1, Val) =:= literal)). + +%% instruction() describes all instructions that are used during optimzation +%% (from beam_a to beam_z). +-type instruction() :: atom() | tuple(). + +-type code_index() :: gb_trees:tree(beam_asm:label(), [instruction()]). + +-type int_function() :: {'function',beam_asm:function_name(),arity(), + beam_asm:label(),[instruction()]}. + +-type module_code() :: + {module(),[_],[_],[int_function()],pos_integer()}. + +%% Internal types. +-type fail() :: beam_asm:fail() | 'fail'. +-type test() :: {'test',atom(),fail(),[beam_asm:src()]} | + {'test',atom(),fail(),integer(),list(),beam_asm:reg()}. +-type result_cache() :: gb_trees:tree(beam_asm:label(), 'killed' | 'used'). -record(live, - {bl, %Block check fun. - lbl, %Label to code index. - res}). %Result cache for each label. + {lbl :: code_index(), %Label to code index. + res :: result_cache()}). %Result cache for each label. + +%% usage(Register, [Instruction], State) -> killed|not_used|used. +%% Determine the usage of Register in the instruction sequence. +%% The return value is one of: +%% +%% killed - The register is not used in any way. +%% not_used - The register is referenced only by an allocating instruction +%% (the actual value does not matter). +%% used - The register is used (its value do matter). + +-spec usage(beam_asm:reg(), [instruction()], code_index()) -> + 'killed' | 'not_used' | 'used'. + +usage(R, Is, D) -> + St = #live{lbl=D,res=gb_trees:empty()}, + {Usage,_} = check_liveness(R, Is, St), + Usage. %% is_killed_block(Register, [Instruction]) -> true|false @@ -46,12 +89,18 @@ %% i.e. it is OK to enter the instruction sequence with Register %% containing garbage. -is_killed_block(R, Is) -> - case check_killed_block(R, Is) of - killed -> true; - used -> false; - transparent -> false - end. +-spec is_killed_block(beam_asm:reg(), [instruction()]) -> boolean(). + +is_killed_block({x,X}, [{set,_,_,{alloc,Live,_}}|_]) -> + X >= Live; +is_killed_block(R, [{set,Ds,Ss,_Op}|Is]) -> + not member(R, Ss) andalso (member(R, Ds) orelse is_killed_block(R, Is)); +is_killed_block(R, [{'%anno',{used,Regs}}|Is]) -> + case R of + {x,X} when (Regs bsr X) band 1 =:= 0 -> true; + _ -> is_killed_block(R, Is) + end; +is_killed_block(_, []) -> false. %% is_killed(Register, [Instruction], State) -> true|false %% Determine whether a register is killed by the instruction sequence. @@ -63,23 +112,27 @@ is_killed_block(R, Is) -> %% The state (constructed by index_instructions/1) is used to allow us %% to determine the kill state across branches. +-spec is_killed(beam_asm:reg(), [instruction()], code_index()) -> boolean(). + is_killed(R, Is, D) -> - St = #live{bl=check_killed_block_fun(),lbl=D,res=gb_trees:empty()}, + St = #live{lbl=D,res=gb_trees:empty()}, case check_liveness(R, Is, St) of {killed,_} -> true; - {used,_} -> false; - {unknown,_} -> false + {exit_not_used,_} -> false; + {_,_} -> false end. %% is_killed_at(Reg, Lbl, State) -> true|false %% Determine whether Reg is killed at label Lbl. +-spec is_killed_at(beam_asm:reg(), beam_asm:label(), code_index()) -> boolean(). + is_killed_at(R, Lbl, D) when is_integer(Lbl) -> - St0 = #live{bl=check_killed_block_fun(),lbl=D,res=gb_trees:empty()}, + St0 = #live{lbl=D,res=gb_trees:empty()}, case check_liveness_at(R, Lbl, St0) of {killed,_} -> true; - {used,_} -> false; - {unknown,_} -> false + {exit_not_used,_} -> false; + {_,_} -> false end. %% is_not_used(Register, [Instruction], State) -> true|false @@ -89,45 +142,39 @@ is_killed_at(R, Lbl, D) when is_integer(Lbl) -> %% The state is used to allow us to determine the usage state %% across branches. +-spec is_not_used(beam_asm:reg(), [instruction()], code_index()) -> boolean(). + is_not_used(R, Is, D) -> - St = #live{bl=fun check_used_block/3,lbl=D,res=gb_trees:empty()}, + St = #live{lbl=D,res=gb_trees:empty()}, case check_liveness(R, Is, St) of - {killed,_} -> true; {used,_} -> false; - {unknown,_} -> false - end. - -%% is_not_used(Register, [Instruction], State) -> true|false -%% Determine whether a register is never used in the instruction sequence -%% (it could still be referenced by an allocate instruction, meaning that -%% it MUST be initialized, but that its value does not matter). -%% The state is used to allow us to determine the usage state -%% across branches. - -is_not_used_at(R, Lbl, D) -> - St = #live{bl=fun check_used_block/3,lbl=D,res=gb_trees:empty()}, - case check_liveness_at(R, Lbl, St) of - {killed,_} -> true; - {used,_} -> false; - {unknown,_} -> false + {exit_not_used,_} -> true; + {_,_} -> true end. %% index_labels(FunctionIs) -> State %% Index the instruction sequence so that we can quickly %% look up the instruction following a specific label. +-spec index_labels([instruction()]) -> code_index(). + index_labels(Is) -> index_labels_1(Is, []). %% empty_label_index() -> State %% Create an empty label index. +-spec empty_label_index() -> code_index(). + empty_label_index() -> gb_trees:empty(). %% index_label(Label, [Instruction], State) -> State %% Add an index for a label. +-spec index_label(beam_asm:label(), [instruction()], code_index()) -> + code_index(). + index_label(Lbl, Is0, Acc) -> Is = drop_labels(Is0), gb_trees:enter(Lbl, Is, Acc). @@ -136,15 +183,28 @@ index_label(Lbl, Is0, Acc) -> %% code_at(Label, State) -> [I]. %% Retrieve the code at the given label. +-spec code_at(beam_asm:label(), code_index()) -> [instruction()]. + code_at(L, Ll) -> - case gb_trees:lookup(L, Ll) of - {value,Code} -> Code; - none -> none - end. + gb_trees:get(L, Ll). + +%% replace_labels(FunctionIs, Tail, ReplaceDb, Fallback) -> FunctionIs. +%% Replace all labels in instructions according to the ReplaceDb. +%% If label is not found the Fallback is called with the label to +%% produce a new one. + +-spec replace_labels([instruction()], + [instruction()], + #{beam_asm:label() => beam_asm:label()}, + fun((beam_asm:label()) -> term())) -> [instruction()]. +replace_labels(Is, Acc, D, Fb) -> + replace_labels_1(Is, Acc, D, Fb). %% bif_to_test(Bif, [Op], Fail) -> {test,Test,Fail,[Op]} %% Convert a BIF to a test. Fail if not possible. +-spec bif_to_test(atom(), list(), fail()) -> test(). + bif_to_test(is_atom, [_]=Ops, Fail) -> {test,is_atom,Fail,Ops}; bif_to_test(is_boolean, [_]=Ops, Fail) -> {test,is_boolean,Fail,Ops}; bif_to_test(is_binary, [_]=Ops, Fail) -> {test,is_binary,Fail,Ops}; @@ -164,20 +224,32 @@ bif_to_test('=<', [A,B], Fail) -> {test,is_ge,Fail,[B,A]}; bif_to_test('>', [A,B], Fail) -> {test,is_lt,Fail,[B,A]}; bif_to_test('<', [_,_]=Ops, Fail) -> {test,is_lt,Fail,Ops}; bif_to_test('>=', [_,_]=Ops, Fail) -> {test,is_ge,Fail,Ops}; -bif_to_test('==', [A,[]], Fail) -> {test,is_nil,Fail,[A]}; +bif_to_test('==', [A,nil], Fail) -> {test,is_nil,Fail,[A]}; +bif_to_test('==', [nil,A], Fail) -> {test,is_nil,Fail,[A]}; +bif_to_test('==', [C,A], Fail) when ?is_const(C) -> + {test,is_eq,Fail,[A,C]}; bif_to_test('==', [_,_]=Ops, Fail) -> {test,is_eq,Fail,Ops}; +bif_to_test('/=', [C,A], Fail) when ?is_const(C) -> + {test,is_ne,Fail,[A,C]}; bif_to_test('/=', [_,_]=Ops, Fail) -> {test,is_ne,Fail,Ops}; -bif_to_test('=:=', [A,[]], Fail) -> {test,is_nil,Fail,[A]}; +bif_to_test('=:=', [A,nil], Fail) -> {test,is_nil,Fail,[A]}; +bif_to_test('=:=', [nil,A], Fail) -> {test,is_nil,Fail,[A]}; +bif_to_test('=:=', [C,A], Fail) when ?is_const(C) -> + {test,is_eq_exact,Fail,[A,C]}; bif_to_test('=:=', [_,_]=Ops, Fail) -> {test,is_eq_exact,Fail,Ops}; +bif_to_test('=/=', [C,A], Fail) when ?is_const(C) -> + {test,is_ne_exact,Fail,[A,C]}; bif_to_test('=/=', [_,_]=Ops, Fail) -> {test,is_ne_exact,Fail,Ops}; bif_to_test(is_record, [_,_,_]=Ops, Fail) -> {test,is_record,Fail,Ops}. %% is_pure_test({test,Op,Fail,Ops}) -> true|false. %% Return 'true' if the test instruction does not modify any -%% registers and/or bit syntax matching state, nor modifies -%% any bit syntax matching state. +%% registers and/or bit syntax matching state. %% + +-spec is_pure_test(test()) -> boolean(). + is_pure_test({test,is_eq,_,[_,_]}) -> true; is_pure_test({test,is_ne,_,[_,_]}) -> true; is_pure_test({test,is_eq_exact,_,[_,_]}) -> true; @@ -188,6 +260,8 @@ is_pure_test({test,is_nil,_,[_]}) -> true; is_pure_test({test,is_nonempty_list,_,[_]}) -> true; is_pure_test({test,test_arity,_,[_,_]}) -> true; is_pure_test({test,has_map_fields,_,[_|_]}) -> true; +is_pure_test({test,is_bitstr,_,[_]}) -> true; +is_pure_test({test,is_function2,_,[_,_]}) -> true; is_pure_test({test,Op,_,Ops}) -> erl_internal:new_type_test(Op, length(Ops)). @@ -196,9 +270,11 @@ is_pure_test({test,Op,_,Ops}) -> %% Go through the instruction sequence in reverse execution %% order, keep track of liveness and remove 'move' instructions %% whose destination is a register that will not be used. -%% Also insert {'%live',Live,Regs} annotations at the beginning +%% Also insert {used,Regs} annotations at the beginning %% and end of each block. -%% + +-spec live_opt([instruction()]) -> [instruction()]. + live_opt(Is0) -> {[{label,Fail}|_]=Bef,[Fi|Is]} = splitwith(fun({func_info,_,_,_}) -> false; @@ -209,51 +285,89 @@ live_opt(Is0) -> Bef ++ [Fi|live_opt(reverse(Is), 0, D, [])]. -%% delete_live_annos([Instruction]) -> [Instruction]. -%% Delete all live annotations. -%% -delete_live_annos([{block,Bl0}|Is]) -> - case delete_live_annos(Bl0) of - [] -> delete_live_annos(Is); - [_|_]=Bl -> [{block,Bl}|delete_live_annos(Is)] - end; -delete_live_annos([{'%live',_,_}|Is]) -> - delete_live_annos(Is); -delete_live_annos([I|Is]) -> - [I|delete_live_annos(Is)]; -delete_live_annos([]) -> []. - +%% delete_annos([Instruction]) -> [Instruction]. +%% Delete all annotations. + +-spec delete_annos([instruction()]) -> [instruction()]. + +delete_annos([{block,Bl0}|Is]) -> + case delete_annos(Bl0) of + [] -> delete_annos(Is); + [_|_]=Bl -> [{block,Bl}|delete_annos(Is)] + end; +delete_annos([{'%anno',_}|Is]) -> + delete_annos(Is); +delete_annos([I|Is]) -> + [I|delete_annos(Is)]; +delete_annos([]) -> []. + %% combine_heap_needs(HeapNeed1, HeapNeed2) -> HeapNeed %% Combine the heap need for two allocation instructions. -combine_heap_needs({alloc,Alloc1}, {alloc,Alloc2}) -> - {alloc,combine_alloc_lists(Alloc1, Alloc2)}; -combine_heap_needs({alloc,Alloc}, Words) when is_integer(Words) -> - {alloc,combine_alloc_lists(Alloc, [{words,Words}])}; -combine_heap_needs(Words, {alloc,Alloc}) when is_integer(Words) -> - {alloc,combine_alloc_lists(Alloc, [{words,Words}])}; +-type heap_need_tag() :: 'floats' | 'words'. +-type heap_need() :: non_neg_integer() | + {'alloc',[{heap_need_tag(),non_neg_integer()}]}. +-spec combine_heap_needs(heap_need(), heap_need()) -> heap_need(). + combine_heap_needs(H1, H2) when is_integer(H1), is_integer(H2) -> - H1+H2. + H1 + H2; +combine_heap_needs(H1, H2) -> + {alloc,combine_alloc_lists([H1,H2])}. + + +%% anno_defs(Instructions) -> Instructions' +%% Add {def,RegisterBitmap} annotations to the beginning of +%% each block. Iff bit X is set in the the bitmap, it means +%% that {x,X} is defined when the block is entered. + +-spec anno_defs([instruction()]) -> [instruction()]. + +anno_defs(Is0) -> + {Bef,[Fi|Is1]} = + splitwith(fun({func_info,_,_,_}) -> false; + (_) -> true + end, Is0), + {func_info,_,_,Arity} = Fi, + Regs = init_def_regs(Arity), + Is = defs(Is1, Regs, #{}), + Bef ++ [Fi|Is]. + +%% split_even/1 +%% [1,2,3,4,5,6] -> {[1,3,5],[2,4,6]} + +-spec split_even(list()) -> {list(),list()}. + +split_even(Rs) -> split_even(Rs, [], []). %%% %%% Local functions. %%% -%% check_liveness(Reg, [Instruction], {State,BlockCheckFun}) -> -%% {killed | used | unknown,UpdateState} -%% Finds out how Reg is used in the instruction sequence. Returns one of: -%% killed - Reg is assigned a new value or killed by an allocation instruction -%% used - Reg is used (or possibly referenced by an allocation instruction) -%% unknown - not possible to determine (perhaps because of an instruction -%% that we don't recognize) - -check_liveness(R, [{set,_,_,_}=I|_], St) -> - erlang:error(only_allowed_in_blocks, [R,I,St]); -check_liveness(R, [{block,Blk}|Is], #live{bl=BlockCheck}=St0) -> - case BlockCheck(R, Blk, St0) of - {transparent,St} -> check_liveness(R, Is, St); - {Other,_}=Res when is_atom(Other) -> Res +%% check_liveness(Reg, [Instruction], #live{}) -> +%% {killed | not_used | used, #live{}} +%% Find out whether Reg is used or killed in instruction sequence. +%% +%% killed - Reg is assigned or killed by an allocation instruction. +%% not_used - the value of Reg is not used, but Reg must not be garbage +%% exit_not_used - the value of Reg is not used, but must not be garbage +%% because the stack will be scanned because an +%% exit BIF will raise an exception +%% used - Reg is used + +check_liveness({fr,_}, _, St) -> + %% Conservatively always consider the floating point register used. + {used,St}; +check_liveness(R, [{block,Blk}|Is], St0) -> + case check_liveness_block(R, Blk, St0) of + {transparent,St1} -> + check_liveness(R, Is, St1); + {alloc_used,St1} -> + %% Used by an allocating instruction, but value not referenced. + %% Must check the rest of the instructions. + not_used(check_liveness(R, Is, St1)); + {Other,_}=Res when is_atom(Other) -> + Res end; check_liveness(R, [{label,_}|Is], St) -> check_liveness(R, Is, St); @@ -263,8 +377,14 @@ check_liveness(R, [{test,_,{f,Fail},As}|Is], St0) -> {used,St0}; false -> case check_liveness_at(R, Fail, St0) of - {killed,St} -> check_liveness(R, Is, St); - {_,_}=Other -> Other + {killed,St1} -> + check_liveness(R, Is, St1); + {exit_not_used,St1} -> + not_used(check_liveness(R, Is, St1)); + {not_used,St1} -> + not_used(check_liveness(R, Is, St1)); + {used,_}=Used -> + Used end end; check_liveness(R, [{test,Op,Fail,Live,Ss,Dst}|Is], St) -> @@ -278,12 +398,14 @@ check_liveness(R, [{select,_,_,Fail,Branches}|_], St) -> check_liveness_everywhere(R, [Fail|Branches], St); check_liveness(R, [{jump,{f,F}}|_], St) -> check_liveness_at(R, F, St); -check_liveness(R, [{case_end,Used}|_], St) -> - check_liveness_ret(R, Used, St); +check_liveness(R, [{case_end,Used}|_], St) -> + check_liveness_exit(R, Used, St); +check_liveness(R, [{try_case_end,Used}|_], St) -> + check_liveness_exit(R, Used, St); check_liveness(R, [{badmatch,Used}|_], St) -> - check_liveness_ret(R, Used, St); -check_liveness(_, [if_end|_], St) -> - {killed,St}; + check_liveness_exit(R, Used, St); +check_liveness(R, [if_end|_], St) -> + check_liveness_exit(R, ignore, St); check_liveness(R, [{func_info,_,_,Ar}|_], St) -> case R of {x,X} when X < Ar -> {used,St}; @@ -306,17 +428,27 @@ check_liveness(R, [{bs_init,_,_,none,Ss,Dst}|Is], St) -> check_liveness(R, [{bs_init,_,_,Live,Ss,Dst}|Is], St) -> case R of {x,X} -> - case X < Live orelse member(R, Ss) of - true -> {used,St}; - false -> {killed,St} + case member(R, Ss) of + true -> + {used,St}; + false -> + if + X < Live -> + not_used(check_liveness(R, Is, St)); + true -> + {killed,St} + end end; {y,_} -> case member(R, Ss) of true -> {used,St}; false -> + %% If the exception is taken, the stack may + %% be scanned. Therefore the register is not + %% guaranteed to be killed. if - R =:= Dst -> {killed,St}; - true -> check_liveness(R, Is, St) + R =:= Dst -> {not_used,St}; + true -> not_used(check_liveness(R, Is, St)) end end end; @@ -325,13 +457,16 @@ check_liveness(R, [{deallocate,_}|Is], St) -> {y,_} -> {killed,St}; _ -> check_liveness(R, Is, St) end; -check_liveness(R, [return|_], St) -> - check_liveness_live_ret(R, 1, St); +check_liveness({x,_}=R, [return|_], St) -> + case R of + {x,0} -> {used,St}; + {x,_} -> {killed,St} + end; check_liveness(R, [{call,Live,_}|Is], St) -> case R of {x,X} when X < Live -> {used,St}; {x,_} -> {killed,St}; - {y,_} -> check_liveness(R, Is, St) + {y,_} -> not_used(check_liveness(R, Is, St)) end; check_liveness(R, [{call_ext,Live,_}=I|Is], St) -> case R of @@ -342,55 +477,32 @@ check_liveness(R, [{call_ext,Live,_}=I|Is], St) -> {y,_} -> case beam_jump:is_exit_instruction(I) of false -> - check_liveness(R, Is, St); + not_used(check_liveness(R, Is, St)); true -> %% We must make sure we don't check beyond this %% instruction or we will fall through into random %% unrelated code and get stuck in a loop. - {killed,St} + {exit_not_used,St} end end; check_liveness(R, [{call_fun,Live}|Is], St) -> case R of {x,X} when X =< Live -> {used,St}; {x,_} -> {killed,St}; - {y,_} -> check_liveness(R, Is, St) + {y,_} -> not_used(check_liveness(R, Is, St)) end; check_liveness(R, [{apply,Args}|Is], St) -> case R of {x,X} when X < Args+2 -> {used,St}; {x,_} -> {killed,St}; - {y,_} -> check_liveness(R, Is, St) - end; -check_liveness(R, [{bif,Op,{f,Fail},Ss,D}|Is], St0) -> - case check_liveness_fail(R, Op, Ss, Fail, St0) of - {killed,St} = Killed -> - case member(R, Ss) of - true -> {used,St}; - false when R =:= D -> Killed; - false -> check_liveness(R, Is, St) - end; - Other -> - Other - end; -check_liveness(R, [{gc_bif,Op,{f,Fail},Live,Ss,D}|Is], St0) -> - case R of - {x,X} when X >= Live -> - {killed,St0}; - {x,_} -> - {used,St0}; - _ -> - case check_liveness_fail(R, Op, Ss, Fail, St0) of - {killed,St}=Killed -> - case member(R, Ss) of - true -> {used,St}; - false when R =:= D -> Killed; - false -> check_liveness(R, Is, St) - end; - Other -> - Other - end + {y,_} -> not_used(check_liveness(R, Is, St)) end; +check_liveness(R, [{bif,Op,Fail,Ss,D}|Is], St) -> + Set = {set,[D],Ss,{bif,Op,Fail}}, + check_liveness(R, [{block,[Set]}|Is], St); +check_liveness(R, [{gc_bif,Op,{f,Fail},Live,Ss,D}|Is], St) -> + Set = {set,[D],Ss,{alloc,Live,{gc_bif,Op,Fail}}}, + check_liveness(R, [{block,[Set]}|Is], St); check_liveness(R, [{bs_put,{f,0},_,Ss}|Is], St) -> case member(R, Ss) of true -> {used,St}; @@ -416,18 +528,14 @@ check_liveness(R, [{make_fun2,_,_,_,NumFree}|Is], St) -> case R of {x,X} when X < NumFree -> {used,St}; {x,_} -> {killed,St}; - _ -> check_liveness(R, Is, St) + {y,_} -> not_used(check_liveness(R, Is, St)) end; -check_liveness({x,_}=R, [{'catch',_,_}|Is], St) -> - %% All x registers will be killed if an exception occurs. - %% Therefore we only need to check the liveness for the - %% instructions following the catch instruction. - check_liveness(R, Is, St); -check_liveness({x,_}=R, [{'try',_,_}|Is], St) -> - %% All x registers will be killed if an exception occurs. - %% Therefore we only need to check the liveness for the - %% instructions inside the 'try' block. - check_liveness(R, Is, St); +check_liveness(R, [{'catch'=Op,Y,Fail}|Is], St) -> + Set = {set,[Y],[],{try_catch,Op,Fail}}, + check_liveness(R, [{block,[Set]}|Is], St); +check_liveness(R, [{'try'=Op,Y,Fail}|Is], St) -> + Set = {set,[Y],[],{try_catch,Op,Fail}}, + check_liveness(R, [{block,[Set]}|Is], St); check_liveness(R, [{try_end,Y}|Is], St) -> case R of Y -> @@ -461,8 +569,9 @@ check_liveness(R, [{loop_rec,{f,_},{x,0}}|_], St) -> {x,_} -> {killed,St}; _ -> - %% y register. Rarely happens. Be very conversative. - {unknown,St} + %% y register. Rarely happens. Be very conversative and + %% assume it's used. + {used,St} end; check_liveness(R, [{loop_rec_end,{f,Fail}}|_], St) -> check_liveness_at(R, Fail, St); @@ -484,23 +593,58 @@ check_liveness(R, [{get_map_elements,{f,Fail},S,{list,L}}|Is], St0) -> Other end end; +check_liveness(R, [{put_map,F,Op,S,D,Live,{list,Puts}}|Is], St) -> + Set = {set,[D],[S|Puts],{alloc,Live,{put_map,Op,F}}}, + check_liveness(R, [{block,[Set]}||Is], St); +check_liveness(R, [{put_tuple,Ar,D}|Is], St) -> + Set = {set,[D],[],{put_tuple,Ar}}, + check_liveness(R, [{block,[Set]}||Is], St); +check_liveness(R, [{put_list,S1,S2,D}|Is], St) -> + Set = {set,[D],[S1,S2],put_list}, + check_liveness(R, [{block,[Set]}||Is], St); +check_liveness(R, [{test_heap,N,Live}|Is], St) -> + I = {block,[{set,[],[],{alloc,Live,{nozero,nostack,N,[]}}}]}, + check_liveness(R, [I|Is], St); +check_liveness(R, [{allocate_zero,N,Live}|Is], St) -> + I = {block,[{set,[],[],{alloc,Live,{zero,N,0,[]}}}]}, + check_liveness(R, [I|Is], St); +check_liveness(R, [{get_hd,S,D}|Is], St) -> + I = {block,[{set,[D],[S],get_hd}]}, + check_liveness(R, [I|Is], St); +check_liveness(R, [{get_tl,S,D}|Is], St) -> + I = {block,[{set,[D],[S],get_tl}]}, + check_liveness(R, [I|Is], St); +check_liveness(R, [remove_message|Is], St) -> + check_liveness(R, Is, St); +check_liveness({x,X}, [build_stacktrace|_], St) when X > 0 -> + {killed,St}; +check_liveness(R, [{recv_mark,_}|Is], St) -> + check_liveness(R, Is, St); +check_liveness(R, [{recv_set,_}|Is], St) -> + check_liveness(R, Is, St); +check_liveness(R, [{'%',_}|Is], St) -> + check_liveness(R, Is, St); check_liveness(_R, Is, St) when is_list(Is) -> -%% case Is of -%% [I|_] -> -%% io:format("~p ~p\n", [_R,I]); -%% _ -> ok -%% end, - {unknown,St}. - -check_liveness_everywhere(R, [{f,Lbl}|T], St0) -> - case check_liveness_at(R, Lbl, St0) of - {killed,St} -> check_liveness_everywhere(R, T, St); - {_,_}=Other -> Other + %% Not implemented. Conservatively assume that the register is used. + {used,St}. + +check_liveness_everywhere(R, Lbls, St0) -> + check_liveness_everywhere_1(R, Lbls, killed, St0). + +check_liveness_everywhere_1(R, [{f,Lbl}|T], Res0, St0) -> + {Res1,St} = check_liveness_at(R, Lbl, St0), + Res = case Res1 of + killed -> Res0; + _ -> Res1 + end, + case Res of + used -> {used,St}; + _ -> check_liveness_everywhere_1(R, T, Res, St) end; -check_liveness_everywhere(R, [_|T], St) -> - check_liveness_everywhere(R, T, St); -check_liveness_everywhere(_, [], St) -> - {killed,St}. +check_liveness_everywhere_1(R, [_|T], Res, St) -> + check_liveness_everywhere_1(R, T, Res, St); +check_liveness_everywhere_1(_, [], Res, St) -> + {Res,St}. check_liveness_at(R, Lbl, #live{lbl=Ll,res=ResMemorized}=St0) -> case gb_trees:lookup(Lbl, ResMemorized) of @@ -509,119 +653,121 @@ check_liveness_at(R, Lbl, #live{lbl=Ll,res=ResMemorized}=St0) -> none -> {Res,St} = case gb_trees:lookup(Lbl, Ll) of {value,Is} -> check_liveness(R, Is, St0); - none -> {unknown,St0} + none -> {used,St0} end, {Res,St#live{res=gb_trees:insert(Lbl, Res, St#live.res)}} end. -check_liveness_ret(R, R, St) -> {used,St}; -check_liveness_ret(_, _, St) -> {killed,St}. - -check_liveness_live_ret({x,R}, Live, St) -> - if - R < Live -> {used,St}; - true -> {killed,St} - end; -check_liveness_live_ret({y,_}, _, St) -> - {killed,St}. +not_used({used,_}=Res) -> Res; +not_used({_,St}) -> {not_used,St}. -check_liveness_fail(_, _, _, 0, St) -> - {killed,St}; -check_liveness_fail(R, Op, Args, Fail, St) -> - Arity = length(Args), - case erl_internal:comp_op(Op, Arity) orelse - erl_internal:new_type_test(Op, Arity) of - true -> {killed,St}; - false -> check_liveness_at(R, Fail, St) - end. +check_liveness_exit(R, R, St) -> {used,St}; +check_liveness_exit({x,_}, _, St) -> {killed,St}; +check_liveness_exit({y,_}, _, St) -> {exit_not_used,St}. -%% check_killed_block(Reg, [Instruction], State) -> killed | transparent | used +%% check_liveness_block(Reg, [Instruction], State) -> +%% {killed | not_used | used | alloc_used | transparent,State'} %% Finds out how Reg is used in the instruction sequence inside a block. %% Returns one of: -%% killed - Reg is assigned a new value or killed by an allocation instruction +%% killed - Reg is assigned a new value or killed by an +%% allocation instruction +%% not_used - The value is not used, but the register is referenced +%% e.g. by an allocation instruction %% transparent - Reg is neither used nor killed -%% used - Reg is used or referenced by an allocation instruction. -%% -%% (Unknown instructions will cause an exception.) - -check_killed_block_fun() -> - fun(R, Is, St) -> {check_killed_block(R, Is),St} end. +%% alloc_used - Used only in an allocate instruction +%% used - Reg is explicitly used by an instruction +%% +%% Annotations are not allowed. +%% +%% (Unknown instructions will cause an exception.) -check_killed_block({x,X}, [{set,_,_,{alloc,Live,_}}|_]) -> +check_liveness_block({x,X}=R, [{set,Ds,Ss,{alloc,Live,Op}}|Is], St0) -> if - X >= Live -> killed; - true -> used - end; -check_killed_block(R, [{set,Ds,Ss,_Op}|Is]) -> - case member(R, Ss) of - true -> used; - false -> - case member(R, Ds) of - true -> killed; - false -> check_killed_block(R, Is) + X >= Live -> + {killed,St0}; + true -> + case check_liveness_block_1(R, Ss, Ds, Op, Is, St0) of + {transparent,St} -> {alloc_used,St}; + {_,_}=Res -> not_used(Res) end end; -check_killed_block(R, [{'%live',Live,_}|Is]) -> - case R of - {x,X} when X >= Live -> killed; - _ -> check_killed_block(R, Is) - end; -check_killed_block(_, []) -> transparent. - -%% check_used_block(Reg, [Instruction], State) -> killed | transparent | used -%% Finds out how Reg is used in the instruction sequence inside a block. -%% Returns one of: -%% killed - Reg is assigned a new value or killed by an allocation instruction -%% transparent - Reg is neither used nor killed -%% used - Reg is explicitly used by an instruction -%% -%% (Unknown instructions will cause an exception.) - -check_used_block({x,X}=R, [{set,Ds,Ss,{alloc,Live,Op}}|Is], St) -> - if - X >= Live -> {killed,St}; - true -> check_used_block_1(R, Ss, Ds, Op, Is, St) +check_liveness_block({y,_}=R, [{set,Ds,Ss,{alloc,_Live,Op}}|Is], St0) -> + case check_liveness_block_1(R, Ss, Ds, Op, Is, St0) of + {transparent,St} -> {alloc_used,St}; + {_,_}=Res -> not_used(Res) end; -check_used_block(R, [{set,Ds,Ss,Op}|Is], St) -> - check_used_block_1(R, Ss, Ds, Op, Is, St); -check_used_block(R, [{'%live',Live,_}|Is], St) -> - case R of - {x,X} when X >= Live -> {killed,St}; - _ -> check_used_block(R, Is, St) +check_liveness_block({y,_}=R, [{set,Ds,Ss,{try_catch,_,Op}}|Is], St0) -> + case Ds of + [R] -> + {killed,St0}; + _ -> + case check_liveness_block_1(R, Ss, Ds, Op, Is, St0) of + {exit_not_used,St} -> + {used,St}; + {transparent,St} -> + %% Conservatively assumed that it is used. + {used,St}; + {_,_}=Res -> + Res + end end; -check_used_block(_, [], St) -> {transparent,St}. +check_liveness_block(R, [{set,Ds,Ss,Op}|Is], St) -> + check_liveness_block_1(R, Ss, Ds, Op, Is, St); +check_liveness_block(_, [], St) -> {transparent,St}. -check_used_block_1(R, Ss, Ds, Op, Is, St0) -> +check_liveness_block_1(R, Ss, Ds, Op, Is, St0) -> case member(R, Ss) of true -> {used,St0}; false -> - case is_reg_used_at(R, Op, St0) of - {true,St} -> - {used,St}; - {false,St} -> + case check_liveness_block_2(R, Op, Ss, St0) of + {killed,St} -> case member(R, Ds) of true -> {killed,St}; - false -> check_used_block(R, Is, St) - end + false -> check_liveness_block(R, Is, St) + end; + {exit_not_used,St} -> + case member(R, Ds) of + true -> {exit_not_used,St}; + false -> check_liveness_block(R, Is, St) + end; + {not_used,St} -> + not_used(case member(R, Ds) of + true -> {killed,St}; + false -> check_liveness_block(R, Is, St) + end); + {used,St} -> + {used,St} end end. -is_reg_used_at(R, {gc_bif,_,{f,Lbl}}, St) -> - is_reg_used_at_1(R, Lbl, St); -is_reg_used_at(R, {bif,_,{f,Lbl}}, St) -> - is_reg_used_at_1(R, Lbl, St); -is_reg_used_at(_, _, St) -> - {false,St}. +check_liveness_block_2(R, {gc_bif,Op,{f,Lbl}}, Ss, St) -> + check_liveness_block_3(R, Lbl, {Op,length(Ss)}, St); +check_liveness_block_2(R, {bif,Op,{f,Lbl}}, Ss, St) -> + Arity = length(Ss), + case erl_internal:comp_op(Op, Arity) orelse + erl_internal:new_type_test(Op, Arity) of + true -> + {killed,St}; + false -> + check_liveness_block_3(R, Lbl, {Op,length(Ss)}, St) + end; +check_liveness_block_2(R, {put_map,_Op,{f,Lbl}}, _Ss, St) -> + check_liveness_block_3(R, Lbl, {unsafe,0}, St); +check_liveness_block_2(_, _, _, St) -> + {killed,St}. -is_reg_used_at_1(_, 0, St) -> - {false,St}; -is_reg_used_at_1(R, Lbl, St0) -> - case check_liveness_at(R, Lbl, St0) of - {killed,St} -> {false,St}; - {used,St} -> {true,St}; - {unknown,St} -> {true,St} - end. +check_liveness_block_3({x,_}, 0, _FA, St) -> + {killed,St}; +check_liveness_block_3({y,_}, 0, {F,A}, St) -> + %% If the exception is thrown, the stack may be scanned, + %% thus implicitly using the y register. + case erl_bifs:is_safe(erlang, F, A) of + true -> {killed,St}; + false -> {used,St} + end; +check_liveness_block_3(R, Lbl, _FA, St0) -> + check_liveness_at(R, Lbl, St0). index_labels_1([{label,Lbl}|Is0], Acc) -> Is = drop_labels(Is0), @@ -633,22 +779,74 @@ index_labels_1([], Acc) -> gb_trees:from_orddict(sort(Acc)). drop_labels([{label,_}|Is]) -> drop_labels(Is); drop_labels(Is) -> Is. -%% Help functions for combine_heap_needs. -combine_alloc_lists(Al1, Al2) -> - combine_alloc_lists_1(sort(Al1++Al2)). +replace_labels_1([{test,Test,{f,Lbl},Ops}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{test,Test,{f,label(Lbl, D, Fb)},Ops}|Acc], D, Fb); +replace_labels_1([{test,Test,{f,Lbl},Live,Ops,Dst}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{test,Test,{f,label(Lbl, D, Fb)},Live,Ops,Dst}|Acc], D, Fb); +replace_labels_1([{select,I,R,{f,Fail0},Vls0}|Is], Acc, D, Fb) -> + Vls = map(fun ({f,L}) -> {f,label(L, D, Fb)}; + (Other) -> Other + end, Vls0), + Fail = label(Fail0, D, Fb), + replace_labels_1(Is, [{select,I,R,{f,Fail},Vls}|Acc], D, Fb); +replace_labels_1([{'try',R,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{'try',R,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{'catch',R,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{'catch',R,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{jump,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{jump,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{loop_rec,{f,Lbl},R}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{loop_rec,{f,label(Lbl, D, Fb)},R}|Acc], D, Fb); +replace_labels_1([{loop_rec_end,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{loop_rec_end,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{wait,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{wait,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{wait_timeout,{f,Lbl},To}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{wait_timeout,{f,label(Lbl, D, Fb)},To}|Acc], D, Fb); +replace_labels_1([{recv_mark=Op,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{Op,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{recv_set=Op,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{Op,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{bif,Name,{f,Lbl},As,R}|Is], Acc, D, Fb) when Lbl =/= 0 -> + replace_labels_1(Is, [{bif,Name,{f,label(Lbl, D, Fb)},As,R}|Acc], D, Fb); +replace_labels_1([{gc_bif,Name,{f,Lbl},Live,As,R}|Is], Acc, D, Fb) when Lbl =/= 0 -> + replace_labels_1(Is, [{gc_bif,Name,{f,label(Lbl, D, Fb)},Live,As,R}|Acc], D, Fb); +replace_labels_1([{call,Ar,{f,Lbl}}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{call,Ar,{f,label(Lbl, D, Fb)}}|Acc], D, Fb); +replace_labels_1([{make_fun2,{f,Lbl},U1,U2,U3}|Is], Acc, D, Fb) -> + replace_labels_1(Is, [{make_fun2,{f,label(Lbl, D, Fb)},U1,U2,U3}|Acc], D, Fb); +replace_labels_1([{bs_init,{f,Lbl},Info,Live,Ss,Dst}|Is], Acc, D, Fb) when Lbl =/= 0 -> + replace_labels_1(Is, [{bs_init,{f,label(Lbl, D, Fb)},Info,Live,Ss,Dst}|Acc], D, Fb); +replace_labels_1([{bs_put,{f,Lbl},Info,Ss}|Is], Acc, D, Fb) when Lbl =/= 0 -> + replace_labels_1(Is, [{bs_put,{f,label(Lbl, D, Fb)},Info,Ss}|Acc], D, Fb); +replace_labels_1([{put_map=I,{f,Lbl},Op,Src,Dst,Live,List}|Is], Acc, D, Fb) + when Lbl =/= 0 -> + replace_labels_1(Is, [{I,{f,label(Lbl, D, Fb)},Op,Src,Dst,Live,List}|Acc], D, Fb); +replace_labels_1([{get_map_elements=I,{f,Lbl},Src,List}|Is], Acc, D, Fb) when Lbl =/= 0 -> + replace_labels_1(Is, [{I,{f,label(Lbl, D, Fb)},Src,List}|Acc], D, Fb); +replace_labels_1([I|Is], Acc, D, Fb) -> + replace_labels_1(Is, [I|Acc], D, Fb); +replace_labels_1([], Acc, _, _) -> Acc. + +label(Old, D, Fb) -> + case D of + #{Old := New} -> New; + _ -> Fb(Old) + end. + +%% Help function for combine_heap_needs. -combine_alloc_lists_1([{words,W1},{words,W2}|T]) - when is_integer(W1), is_integer(W2) -> - [{words,W1+W2}|combine_alloc_lists_1(T)]; -combine_alloc_lists_1([{floats,F1},{floats,F2}|T]) - when is_integer(F1), is_integer(F2) -> - [{floats,F1+F2}|combine_alloc_lists_1(T)]; -combine_alloc_lists_1([{words,_}=W|T]) -> - [W|combine_alloc_lists_1(T)]; -combine_alloc_lists_1([{floats,_}=F|T]) -> - [F|combine_alloc_lists_1(T)]; -combine_alloc_lists_1([]) -> []. +combine_alloc_lists(Al0) -> + Al1 = flatmap(fun(Words) when is_integer(Words) -> + [{words,Words}]; + ({alloc,List}) -> + List + end, Al0), + Al2 = sofs:relation(Al1), + Al3 = sofs:relation_to_family(Al2), + Al4 = sofs:to_external(Al3), + [{Tag,lists:sum(L)} || {Tag,L} <- Al4]. %% live_opt/4. @@ -687,10 +885,14 @@ live_opt([{test,bs_start_match2,Fail,Live,[Src,_],_}=I|Is], _, D, Acc) -> %% Other instructions. live_opt([{block,Bl0}|Is], Regs0, D, Acc) -> - Live0 = {'%live',live_regs(Regs0),Regs0}, + Live0 = make_anno({used,Regs0}), {Bl,Regs} = live_opt_block(reverse(Bl0), Regs0, D, [Live0]), - Live = {'%live',live_regs(Regs),Regs}, + Live = make_anno({used,Regs}), live_opt(Is, Regs, D, [{block,[Live|Bl]}|Acc]); +live_opt([build_stacktrace=I|Is], _, D, Acc) -> + live_opt(Is, live_call(1), D, [I|Acc]); +live_opt([raw_raise=I|Is], _, D, Acc) -> + live_opt(Is, live_call(3), D, [I|Acc]); live_opt([{label,L}=I|Is], Regs, D0, Acc) -> D = gb_trees:insert(L, Regs, D0), live_opt(Is, Regs, D, [I|Acc]); @@ -732,23 +934,49 @@ live_opt([{test,_,Fail,Live,Ss,_}=I|Is], _, D, Acc) -> Regs1 = x_live(Ss, Regs0), Regs = live_join_label(Fail, D, Regs1), live_opt(Is, Regs, D, [I|Acc]); -live_opt([{select,_,Src,Fail,List}=I|Is], Regs0, D, Acc) -> +live_opt([{select,_,Src,Fail,List}=I|Is], _, D, Acc) -> + Regs0 = 0, Regs1 = x_live([Src], Regs0), Regs = live_join_labels([Fail|List], D, Regs1), live_opt(Is, Regs, D, [I|Acc]); -live_opt([{'try',_,_}=I|Is], Regs, D, Acc) -> - %% If an exeption happens, all x registers will be killed. - %% Therefore, we should only base liveness of the code inside - %% the try. - live_opt(Is, Regs, D, [I|Acc]); -live_opt([{try_case,_}=I|Is], _, D, Acc) -> - live_opt(Is, live_call(1), D, [I|Acc]); +live_opt([{try_case,Y}=I|Is], Regs0, D, Acc) -> + Regs = live_call(1), + case Regs0 of + 0 -> + live_opt(Is, Regs, D, [{try_end,Y}|Acc]); + _ -> + live_opt(Is, live_call(1), D, [I|Acc]) + end; live_opt([{loop_rec,_Fail,_Dst}=I|Is], _, D, Acc) -> live_opt(Is, 0, D, [I|Acc]); live_opt([timeout=I|Is], _, D, Acc) -> live_opt(Is, 0, D, [I|Acc]); live_opt([{wait,_}=I|Is], _, D, Acc) -> live_opt(Is, 0, D, [I|Acc]); +live_opt([{get_map_elements,Fail,Src,{list,List}}=I|Is], Regs0, D, Acc) -> + {Ss,Ds} = split_even(List), + Regs1 = x_live([Src|Ss], x_dead(Ds, Regs0)), + Regs = live_join_label(Fail, D, Regs1), + live_opt(Is, Regs, D, [I|Acc]); +live_opt([{gc_bif,N,F,R,As,Dst}=I|Is], Regs0, D, Acc) -> + Bl = [{set,[Dst],As,{alloc,R,{gc_bif,N,F}}}], + {_,Regs} = live_opt_block(Bl, Regs0, D, []), + live_opt(Is, Regs, D, [I|Acc]); +live_opt([{bif,N,F,As,Dst}=I|Is], Regs0, D, Acc) -> + Bl = [{set,[Dst],As,{bif,N,F}}], + {_,Regs} = live_opt_block(Bl, Regs0, D, []), + live_opt(Is, Regs, D, [I|Acc]); +live_opt([{get_tuple_element,Src,Idx,Dst}=I|Is], Regs0, D, Acc) -> + Bl = [{set,[Dst],[Src],{get_tuple_element,Idx}}], + {_,Regs} = live_opt_block(Bl, Regs0, D, []), + live_opt(Is, Regs, D, [I|Acc]); +live_opt([{move,Src,Dst}=I|Is], Regs0, D, Acc) -> + Regs = x_live([Src], x_dead([Dst], Regs0)), + live_opt(Is, Regs, D, [I|Acc]); +live_opt([{put_map,F,Op,S,Dst,R,{list,Puts}}=I|Is], Regs0, D, Acc) -> + Bl = [{set,[Dst],[S|Puts],{alloc,R,{put_map,Op,F}}}], + {_,Regs} = live_opt_block(Bl, Regs0, D, []), + live_opt(Is, Regs, D, [I|Acc]); %% Transparent instructions - they neither use nor modify x registers. live_opt([{deallocate,_}=I|Is], Regs, D, Acc) -> @@ -765,6 +993,10 @@ live_opt([{wait_timeout,_,{Tag,_}}=I|Is], Regs, D, Acc) when Tag =/= x -> live_opt(Is, Regs, D, [I|Acc]); live_opt([{line,_}=I|Is], Regs, D, Acc) -> live_opt(Is, Regs, D, [I|Acc]); +live_opt([{'catch',_,_}=I|Is], Regs, D, Acc) -> + live_opt(Is, Regs, D, [I|Acc]); +live_opt([{'try',_,_}=I|Is], Regs, D, Acc) -> + live_opt(Is, Regs, D, [I|Acc]); %% The following instructions can occur if the "compilation" has been %% started from a .S file using the 'from_asm' option. @@ -779,46 +1011,52 @@ live_opt([{recv_mark,_}=I|Is], Regs, D, Acc) -> live_opt([], _, _, Acc) -> Acc. -live_opt_block([{set,Ds,Ss,Op0}|Is], Regs0, D, Acc) -> - Regs1 = x_live(Ss, x_dead(Ds, Regs0)), - {Op, Regs} = live_opt_block_op(Op0, Regs1, D), - I = {set, Ds, Ss, Op}, - - case Ds of - [{x,X}] -> - case (not is_live(X, Regs0)) andalso Op =:= move of - true -> - live_opt_block(Is, Regs0, D, Acc); - false -> - live_opt_block(Is, Regs, D, [I|Acc]) - end; - _ -> - live_opt_block(Is, Regs, D, [I|Acc]) +live_opt_block([{set,[{x,X}]=Ds,Ss,move}=I|Is], Regs0, D, Acc) -> + Regs = x_live(Ss, x_dead(Ds, Regs0)), + case is_live(X, Regs0) of + true -> + live_opt_block(Is, Regs, D, [I|Acc]); + false -> + %% Useless move, will never be used. + live_opt_block(Is, Regs, D, Acc) end; - -live_opt_block([], Regs, _, Acc) -> {Acc,Regs}. - -live_opt_block_op({alloc,Live0,AllocOp}, Regs0, D) -> - Regs = - case AllocOp of - {Kind, _N, Fail} when Kind =:= gc_bif; Kind =:= put_map -> - live_join_label(Fail, D, Regs0); - _ -> - Regs0 - end, +live_opt_block([{set,Ds,Ss,{alloc,Live0,AllocOp}}|Is], Regs0, D, Acc) -> + %% Calculate liveness from the point of view of the GC. + %% There will never be a GC if the instruction fails, so we should + %% ignore the failure branch. + GcRegs1 = x_dead(Ds, Regs0), + GcRegs = x_live(Ss, GcRegs1), + Live = live_regs(GcRegs), %% The life-time analysis used by the code generator is sometimes too %% conservative, so it may be possible to lower the number of live %% registers based on the exact liveness information. The main benefit is %% that more optimizations that depend on liveness information (such as the - %% beam_bool and beam_dead passes) may be applied. - Live = live_regs(Regs), - true = Live =< Live0, - {{alloc,Live,AllocOp}, live_call(Live)}; -live_opt_block_op({bif,_N,Fail} = Op, Regs, D) -> - {Op, live_join_label(Fail, D, Regs)}; -live_opt_block_op(Op, Regs, _D) -> - {Op, Regs}. + %% beam_dead pass) may be applied. + true = Live =< Live0, %Assertion. + I = {set,Ds,Ss,{alloc,Live,AllocOp}}, + + %% Calculate liveness from the point of view of the preceding instruction. + %% The liveness is the union of live registers in the GC and the live + %% registers at the failure label. + Regs1 = live_call(Live), + Regs = live_join_alloc(AllocOp, D, Regs1), + live_opt_block(Is, Regs, D, [I|Acc]); +live_opt_block([{set,Ds,Ss,{bif,_,Fail}}=I|Is], Regs0, D, Acc) -> + Regs1 = x_dead(Ds, Regs0), + Regs2 = x_live(Ss, Regs1), + Regs = live_join_label(Fail, D, Regs2), + live_opt_block(Is, Regs, D, [I|Acc]); +live_opt_block([{set,Ds,Ss,_}=I|Is], Regs0, D, Acc) -> + Regs = x_live(Ss, x_dead(Ds, Regs0)), + live_opt_block(Is, Regs, D, [I|Acc]); +live_opt_block([{'%anno',_}|Is], Regs, D, Acc) -> + live_opt_block(Is, Regs, D, Acc); +live_opt_block([], Regs, _, Acc) -> {Acc,Regs}. + +live_join_alloc({Kind,_Name,Fail}, D, Regs) when Kind =:= gc_bif; Kind =:= put_map -> + live_join_label(Fail, D, Regs); +live_join_alloc(_, _, Regs) -> Regs. live_join_labels([{f,L}|T], D, Regs0) when L =/= 0 -> Regs = gb_trees:get(L, D) bor Regs0, @@ -850,14 +1088,232 @@ x_live([], Regs) -> Regs. is_live(X, Regs) -> ((Regs bsr X) band 1) =:= 1. -%% split_even/1 -%% [1,2,3,4,5,6] -> {[1,3,5],[2,4,6]} -split_even(Rs) -> split_even(Rs,[],[]). -split_even([],Ss,Ds) -> {reverse(Ss),reverse(Ds)}; -split_even([S,D|Rs],Ss,Ds) -> - split_even(Rs,[S|Ss],[D|Ds]). - -%% join_even/1 -%% {[1,3,5],[2,4,6]} -> [1,2,3,4,5,6] -join_even([],[]) -> []; -join_even([S|Ss],[D|Ds]) -> [S,D|join_even(Ss,Ds)]. +split_even([], Ss, Ds) -> + {reverse(Ss),reverse(Ds)}; +split_even([S,D|Rs], Ss, Ds) -> + split_even(Rs, [S|Ss], [D|Ds]). + +%%% +%%% Add annotations for defined registers. +%%% +%%% This analysis is done by scanning the instructions in +%%% execution order. +%%% + +defs([{apply,_}=I|Is], _Regs, D) -> + [I|defs(Is, 1, D)]; +defs([{bif,_,{f,Fail},_Src,Dst}=I|Is], Regs0, D) -> + Regs = def_regs([Dst], Regs0), + [I|defs(Is, Regs, update_regs(Fail, Regs0, D))]; +defs([{block,Block0}|Is], Regs0, D0) -> + {Block,Regs,D} = defs_list(Block0, Regs0, D0), + [{block,[make_anno({def,Regs0})|Block]}|defs(Is, Regs, D)]; +defs([{bs_init,{f,L},_,Live,_,Dst}=I|Is], Regs0, D) -> + Regs1 = case Live of + none -> Regs0; + _ -> init_def_regs(Live) + end, + Regs = def_regs([Dst], Regs1), + [I|defs(Is, Regs, update_regs(L, Regs, D))]; +defs([{test,bs_start_match2,{f,L},Live,_,Dst}=I|Is], _Regs, D) -> + Regs0 = init_def_regs(Live), + Regs = def_regs([Dst], Regs0), + [I|defs(Is, Regs, update_regs(L, Regs0, D))]; +defs([{bs_put,{f,L},_,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, update_regs(L, Regs, D))]; +defs([build_stacktrace=I|Is], _Regs, D) -> + [I|defs(Is, 1, D)]; +defs([{call,_,_}=I|Is], _Regs, D) -> + [I|defs(Is, 1, D)]; +defs([{call_ext,_,{extfunc,M,F,A}}=I|Is], _Regs, D) -> + case erl_bifs:is_exit_bif(M, F, A) of + false -> + [I|defs(Is, 1, D)]; + true -> + [I|defs_unreachable(Is, D)] + end; +defs([{call_ext,_,_}=I|Is], _Regs, D) -> + [I|defs(Is, 1, D)]; +defs([{call_fun,_}=I|Is], _Regs, D) -> + [I|defs(Is, 1, D)]; +defs([{'catch',_,{f,L}}=I|Is], Regs, D) -> + RegsAtLabel = init_def_regs(1), + [I|defs(Is, Regs, update_regs(L, RegsAtLabel, D))]; +defs([{catch_end,_}=I|Is], _Regs, D) -> + Regs = init_def_regs(1), + [I|defs(Is, Regs, D)]; +defs([{gc_bif,_,{f,Fail},Live,_Src,Dst}=I|Is], Regs0, D) -> + true = all_defined(Live, Regs0), %Assertion. + Regs = def_regs([Dst], init_def_regs(Live)), + [I|defs(Is, Regs, update_regs(Fail, Regs0, D))]; +defs([{get_map_elements,{f,L},_Src,{list,DstList}}=I|Is], Regs0, D) -> + {_,Ds} = beam_utils:split_even(DstList), + Regs = def_regs(Ds, Regs0), + [I|defs(Is, Regs, update_regs(L, Regs0, D))]; +defs([{get_tuple_element,_,_,Dst}=I|Is], Regs0, D) -> + Regs = def_regs([Dst], Regs0), + [I|defs(Is, Regs, D)]; +defs([{jump,{f,L}}=I|Is], Regs, D) -> + [I|defs_unreachable(Is, update_regs(L, Regs, D))]; +defs([{label,L}=I|Is], Regs0, D) -> + case D of + #{L:=Regs1} -> + Regs = Regs0 band Regs1, + [I|defs(Is, Regs, D)]; + #{} -> + [I|defs(Is, Regs0, D)] + end; +defs([{loop_rec,{f,L},{x,0}}=I|Is], _Regs, D0) -> + RegsAtLabel = init_def_regs(0), + D = update_regs(L, RegsAtLabel, D0), + [I|defs(Is, init_def_regs(1), D)]; +defs([{loop_rec_end,_}=I|Is], _Regs, D) -> + [I|defs_unreachable(Is, D)]; +defs([{make_fun2,_,_,_,_}=I|Is], _Regs, D) -> + [I|defs(Is, 1, D)]; +defs([{move,_,Dst}=I|Is], Regs0, D) -> + Regs = def_regs([Dst], Regs0), + [I|defs(Is, Regs, D)]; +defs([{put_map,{f,Fail},_,_,Dst,_,_}=I|Is], Regs0, D) -> + Regs = def_regs([Dst], Regs0), + [I|defs(Is, Regs, update_regs(Fail, Regs0, D))]; +defs([raw_raise=I|Is], _Regs, D) -> + [I|defs(Is, 1, D)]; +defs([return=I|Is], _Regs, D) -> + [I|defs_unreachable(Is, D)]; +defs([{select,_,_Src,Fail,List}=I|Is], Regs, D0) -> + D = update_list([Fail|List], Regs, D0), + [I|defs_unreachable(Is, D)]; +defs([{test,_,{f,L},_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, update_regs(L, Regs, D))]; +defs([{test,_,{f,L},Live,_,Dst}=I|Is], Regs0, D) -> + true = all_defined(Live, Regs0), %Assertion. + Regs = def_regs([Dst], init_def_regs(Live)), + [I|defs(Is, Regs, update_regs(L, Regs0, D))]; +defs([{'try',_,{f,L}}=I|Is], Regs, D) -> + RegsAtLabel = init_def_regs(3), + [I|defs(Is, Regs, update_regs(L, RegsAtLabel, D))]; +defs([{try_case,_}=I|Is], _Regs, D) -> + [I|defs(Is, init_def_regs(3), D)]; +defs([{wait,_}=I|Is], _Regs, D) -> + [I|defs_unreachable(Is, D)]; +defs([{wait_timeout,_,_}=I|Is], _Regs, D) -> + [I|defs(Is, 0, D)]; + +%% Exceptions. +defs([{badmatch,_}=I|Is], _Regs, D) -> + [I|defs_unreachable(Is, D)]; +defs([{case_end,_}=I|Is], _Regs, D) -> + [I|defs_unreachable(Is, D)]; +defs([if_end=I|Is], _Regs, D) -> + [I|defs_unreachable(Is, D)]; +defs([{try_case_end,_}=I|Is], _Regs, D) -> + [I|defs_unreachable(Is, D)]; + +%% Neutral instructions +defs([{bs_context_to_binary,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{bs_restore2,_,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{bs_save2,_,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{deallocate,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{kill,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{line,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{recv_mark,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{recv_set,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([timeout=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{trim,_,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{try_end,_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([{'%',_}=I|Is], Regs, D) -> + [I|defs(Is, Regs, D)]; +defs([], _, _) -> []. + +defs_unreachable([{label,L}=I|Is], D) -> + case D of + #{L:=Regs} -> + [I|defs(Is, Regs, D)]; + #{} -> + defs_unreachable(Is, D) + end; +defs_unreachable([_|Is], D) -> + defs_unreachable(Is, D); +defs_unreachable([], _D) -> []. + +defs_list(Is, Regs, D) -> + defs_list(Is, Regs, D, []). + +defs_list([{set,Ds,_,{alloc,Live,Info}}=I|Is], Regs0, D0, Acc) -> + true = all_defined(Live, Regs0), %Assertion. + D = case Info of + {gc_bif,_,{f,Fail}} -> + update_regs(Fail, Regs0, D0); + {put_map,_,{f,Fail}} -> + update_regs(Fail, Regs0, D0); + _ -> + D0 + end, + Regs = def_regs(Ds, init_def_regs(Live)), + defs_list(Is, Regs, D, [I|Acc]); +defs_list([{set,Ds,_,Info}=I|Is], Regs0, D0, Acc) -> + D = case Info of + {bif,_,{f,Fail}} -> + update_regs(Fail, Regs0, D0); + {try_catch,'catch',{f,Fail}} -> + update_regs(Fail, init_def_regs(1), D0); + {try_catch,'try',{f,Fail}} -> + update_regs(Fail, init_def_regs(3), D0); + _ -> + D0 + end, + Regs = def_regs(Ds, Regs0), + defs_list(Is, Regs, D, [I|Acc]); +defs_list([], Regs, D, Acc) -> + {reverse(Acc),Regs,D}. + +init_def_regs(Arity) -> + (1 bsl Arity) - 1. + +def_regs([{x,X}|T], Regs) -> + def_regs(T, Regs bor (1 bsl X)); +def_regs([_|T], Regs) -> + def_regs(T, Regs); +def_regs([], Regs) -> Regs. + +update_list([{f,L}|T], Regs, D0) -> + D = update_regs(L, Regs, D0), + update_list(T, Regs, D); +update_list([_|T], Regs, D) -> + update_list(T, Regs, D); +update_list([], _Regs, D) -> D. + +update_regs(L, Regs0, D) -> + case D of + #{L:=Regs1} -> + Regs = Regs0 band Regs1, + D#{L:=Regs}; + #{} -> + D#{L=>Regs0} + end. + +all_defined(Live, Regs) -> + All = (1 bsl Live) - 1, + Regs band All =:= All. + +%%% +%%% Utilities. +%%% + +%% make_anno(Anno) -> WrappedAnno. +%% Wrap an annotation term. + +make_anno(Anno) -> + {'%anno',Anno}. diff --git a/lib/compiler/src/beam_validator.erl b/lib/compiler/src/beam_validator.erl index b18eb47d0e..fb2e7df65c 100644 --- a/lib/compiler/src/beam_validator.erl +++ b/lib/compiler/src/beam_validator.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2004-2014. All Rights Reserved. +%% Copyright Ericsson AB 2004-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -27,20 +27,13 @@ %% Interface for compiler. -export([module/2, format_error/1]). --include("beam_disasm.hrl"). +-import(lists, [any/2,dropwhile/2,foldl/3,foreach/2,reverse/1]). --import(lists, [reverse/1,foldl/3,foreach/2,dropwhile/2]). - --define(MAXREG, 1024). +%% To be called by the compiler. -%%-define(DEBUG, 1). --ifdef(DEBUG). --define(DBG_FORMAT(F, D), (io:format((F), (D)))). --else. --define(DBG_FORMAT(F, D), ok). --endif. +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_utils:module_code()}. -%% To be called by the compiler. module({Mod,Exp,Attr,Fs,Lc}=Code, _Opts) when is_atom(Mod), is_list(Exp), is_list(Attr), is_integer(Lc) -> case validate(Mod, Fs) of @@ -90,8 +83,6 @@ format_error(Error) -> %%% Things currently not checked. XXX %%% %%% - Heap allocation for binaries. -%%% - That put_tuple is followed by the correct number of -%%% put instructions. %%% %% validate(Module, [Function]) -> [] | [Error] @@ -135,9 +126,8 @@ validate_0(Module, [{function,Name,Ar,Entry,Code}|Fs], Ft) -> throw:Error -> %% Controlled error. [Error|validate_0(Module, Fs, Ft)]; - Class:Error -> + Class:Error:Stack -> %% Crash. - Stack = erlang:get_stacktrace(), io:fwrite("Function: ~w/~w\n", [Name,Ar]), erlang:raise(Class, Error, Stack) end. @@ -154,7 +144,8 @@ validate_0(Module, [{function,Name,Ar,Entry,Code}|Fs], Ft) -> hf=0, %Available heap size for floats. fls=undefined, %Floating point state. ct=[], %List of hot catch/try labels - setelem=false %Previous instruction was setelement/3. + setelem=false, %Previous instruction was setelement/3. + puts_left=none %put/1 instructions left. }). -type label() :: integer(). @@ -170,29 +161,25 @@ validate_0(Module, [{function,Name,Ar,Entry,Code}|Fs], Ft) -> % in the module (those that start with bs_start_match2). }). --ifdef(DEBUG). -print_st(#st{x=Xs,y=Ys,numy=NumY,h=H,ct=Ct}) -> - io:format(" #st{x=~p~n" - " y=~p~n" - " numy=~p,h=~p,ct=~w~n", - [gb_trees:to_list(Xs),gb_trees:to_list(Ys),NumY,H,Ct]). --endif. +%% Match context type. +-record(ms, + {id=make_ref() :: reference(), %Unique ID. + valid=0 :: non_neg_integer(), %Valid slots + slots=0 :: non_neg_integer() %Number of slots + }). validate_1(Is, Name, Arity, Entry, Ft) -> validate_2(labels(Is), Name, Arity, Entry, Ft). validate_2({Ls1,[{func_info,{atom,Mod},{atom,Name},Arity}=_F|Is]}, Name, Arity, Entry, Ft) -> - lists:foreach(fun (_L) -> ?DBG_FORMAT(" ~p.~n", [{label,_L}]) end, Ls1), - ?DBG_FORMAT(" ~p.~n", [_F]), validate_3(labels(Is), Name, Arity, Entry, Mod, Ls1, Ft); validate_2({Ls1,Is}, Name, Arity, _Entry, _Ft) -> error({{'_',Name,Arity},{first(Is),length(Ls1),illegal_instruction}}). validate_3({Ls2,Is}, Name, Arity, Entry, Mod, Ls1, Ft) -> - lists:foreach(fun (_L) -> ?DBG_FORMAT(" ~p.~n", [{label,_L}]) end, Ls2), Offset = 1 + length(Ls1) + 1 + length(Ls2), - EntryOK = (Entry =:= undefined) orelse lists:member(Entry, Ls2), + EntryOK = lists:member(Entry, Ls2), if EntryOK -> St = init_state(Arity), @@ -260,7 +247,6 @@ valfun([], MFA, _Offset, #vst{branched=Targets0,labels=Labels0}=Vst) -> error({MFA,Error}) end; valfun([I|Is], MFA, Offset, Vst0) -> - ?DBG_FORMAT(" ~p.\n", [I]), valfun(Is, MFA, Offset+1, try Vst = val_dsetel(I, Vst0), @@ -278,17 +264,20 @@ valfun_1({label,Lbl}, #vst{current=St0,branched=B,labels=Lbls}=Vst) -> valfun_1(_I, #vst{current=none}=Vst) -> %% Ignore instructions after erlang:error/1,2, which %% the original R10B compiler thought would return. - ?DBG_FORMAT("Ignoring ~p\n", [_I]), Vst; valfun_1({badmatch,Src}, Vst) -> assert_term(Src, Vst), + verify_y_init(Vst), kill_state(Vst); valfun_1({case_end,Src}, Vst) -> assert_term(Src, Vst), + verify_y_init(Vst), kill_state(Vst); valfun_1(if_end, Vst) -> + verify_y_init(Vst), kill_state(Vst); valfun_1({try_case_end,Src}, Vst) -> + verify_y_init(Vst), assert_term(Src, Vst), kill_state(Vst); %% Instructions that can not cause exceptions @@ -296,7 +285,7 @@ valfun_1({bs_context_to_binary,Ctx}, #vst{current=#st{x=Xs}}=Vst) -> case Ctx of {Tag,X} when Tag =:= x; Tag =:= y -> Type = case gb_trees:lookup(X, Xs) of - {value,{match_context,_,_}} -> term; + {value,#ms{}} -> term; _ -> get_term_type(Ctx, Vst) end, set_type_reg(Type, Ctx, Vst); @@ -305,6 +294,8 @@ valfun_1({bs_context_to_binary,Ctx}, #vst{current=#st{x=Xs}}=Vst) -> end; valfun_1(bs_init_writable=I, Vst) -> call(I, 1, Vst); +valfun_1(build_stacktrace=I, Vst) -> + call(I, 1, Vst); valfun_1({move,{y,_}=Src,{y,_}=Dst}, Vst) -> %% The stack trimming optimization may generate a move from an initialized %% but unassigned Y register to another Y register. @@ -350,11 +341,25 @@ valfun_1({put_list,A,B,Dst}, Vst0) -> Vst = eat_heap(2, Vst0), set_type_reg(cons, Dst, Vst); valfun_1({put_tuple,Sz,Dst}, Vst0) when is_integer(Sz) -> + Vst1 = eat_heap(1, Vst0), + Vst = set_type_reg(tuple_in_progress, Dst, Vst1), + #vst{current=St0} = Vst, + St = St0#st{puts_left={Sz,{Dst,{tuple,Sz}}}}, + Vst#vst{current=St}; +valfun_1({put,Src}, Vst0) -> + assert_term(Src, Vst0), Vst = eat_heap(1, Vst0), - set_type_reg({tuple,Sz}, Dst, Vst); -valfun_1({put,Src}, Vst) -> - assert_term(Src, Vst), - eat_heap(1, Vst); + #vst{current=St0} = Vst, + case St0 of + #st{puts_left=none} -> + error(not_building_a_tuple); + #st{puts_left={1,{Dst,Type}}} -> + St = St0#st{puts_left=none}, + set_type_reg(Type, Dst, Vst#vst{current=St}); + #st{puts_left={PutsLeft,Info}} when is_integer(PutsLeft) -> + St = St0#st{puts_left={PutsLeft-1,Info}}, + Vst#vst{current=St} + end; %% Instructions for optimization of selective receives. valfun_1({recv_mark,{f,Fail}}, Vst) when is_integer(Fail) -> Vst; @@ -362,7 +367,9 @@ valfun_1({recv_set,{f,Fail}}, Vst) when is_integer(Fail) -> Vst; %% Misc. valfun_1(remove_message, Vst) -> - Vst; + %% The message term is no longer fragile. It can be used + %% without restrictions. + remove_fragility(Vst); valfun_1({'%',_}, Vst) -> Vst; valfun_1({line,_}, Vst) -> @@ -372,6 +379,9 @@ valfun_1({call_ext,Live,Func}=I, Vst) -> case return_type(Func, Vst) of exception -> verify_live(Live, Vst), + %% The stack will be scanned, so Y registers + %% must be initialized. + verify_y_init(Vst), kill_state(Vst); _ -> valfun_2(I, Vst) @@ -441,6 +451,19 @@ valfun_1({try_case,Reg}, #vst{current=#st{ct=[Fail|Fails]}}=Vst0) -> Type -> error({bad_type,Type}) end; +valfun_1({get_list,Src,D1,D2}, Vst0) -> + assert_type(cons, Src, Vst0), + Vst = set_type_reg(term, Src, D1, Vst0), + set_type_reg(term, Src, D2, Vst); +valfun_1({get_hd,Src,Dst}, Vst) -> + assert_type(cons, Src, Vst), + set_type_reg(term, Src, Dst, Vst); +valfun_1({get_tl,Src,Dst}, Vst) -> + assert_type(cons, Src, Vst), + set_type_reg(term, Src, Dst, Vst); +valfun_1({get_tuple_element,Src,I,Dst}, Vst) -> + assert_type({tuple_element,I+1}, Src, Vst), + set_type_reg(term, Src, Dst, Vst); valfun_1(I, Vst) -> valfun_2(I, Vst). @@ -530,20 +553,40 @@ valfun_4({bif,element,{f,Fail},[Pos,Tuple],Dst}, Vst0) -> Vst1 = branch_state(Fail, Vst0), TupleType = upgrade_tuple_type({tuple,[get_tuple_size(PosType)]}, TupleType0), Vst = set_type(TupleType, Tuple, Vst1), - set_type_reg(term, Dst, Vst); + set_type_reg(term, Tuple, Dst, Vst); +valfun_4({bif,raise,{f,0},Src,_Dst}, Vst) -> + validate_src(Src, Vst), + kill_state(Vst); +valfun_4(raw_raise=I, Vst) -> + call(I, 3, Vst); +valfun_4({bif,map_get,{f,Fail},[_Key,Map]=Src,Dst}, Vst0) -> + validate_src(Src, Vst0), + Vst1 = branch_state(Fail, Vst0), + Vst = set_type(map, Map, Vst1), + Type = propagate_fragility(term, Src, Vst), + set_type_reg(Type, Dst, Vst); +valfun_4({bif,is_map_key,{f,Fail},[_Key,Map]=Src,Dst}, Vst0) -> + validate_src(Src, Vst0), + Vst1 = branch_state(Fail, Vst0), + Vst = set_type(map, Map, Vst1), + Type = propagate_fragility(bool, Src, Vst), + set_type_reg(Type, Dst, Vst); valfun_4({bif,Op,{f,Fail},Src,Dst}, Vst0) -> validate_src(Src, Vst0), Vst = branch_state(Fail, Vst0), - Type = bif_type(Op, Src, Vst), + Type0 = bif_type(Op, Src, Vst), + Type = propagate_fragility(Type0, Src, Vst), set_type_reg(Type, Dst, Vst); valfun_4({gc_bif,Op,{f,Fail},Live,Src,Dst}, #vst{current=St0}=Vst0) -> + verify_live(Live, Vst0), + verify_y_init(Vst0), St = kill_heap_allocation(St0), Vst1 = Vst0#vst{current=St}, - verify_live(Live, Vst1), Vst2 = branch_state(Fail, Vst1), Vst = prune_x_regs(Live, Vst2), validate_src(Src, Vst), - Type = bif_type(Op, Src, Vst), + Type0 = bif_type(Op, Src, Vst), + Type = propagate_fragility(Type0, Src, Vst), set_type_reg(Type, Dst, Vst); valfun_4(return, #vst{current=#st{numy=none}}=Vst) -> assert_term({x,0}, Vst), @@ -554,13 +597,20 @@ valfun_4({jump,{f,Lbl}}, Vst) -> kill_state(branch_state(Lbl, Vst)); valfun_4({loop_rec,{f,Fail},Dst}, Vst0) -> Vst = branch_state(Fail, Vst0), - set_type_reg(term, Dst, Vst); + %% This term may not be part of the root set until + %% remove_message/0 is executed. If control transfers + %% to the loop_rec_end/1 instruction, no part of + %% this term must be stored in a Y register. + set_type_reg({fragile,term}, Dst, Vst); valfun_4({wait,_}, Vst) -> + verify_y_init(Vst), kill_state(Vst); valfun_4({wait_timeout,_,Src}, Vst) -> assert_term(Src, Vst), - Vst; + verify_y_init(Vst), + prune_x_regs(0, Vst); valfun_4({loop_rec_end,_}, Vst) -> + verify_y_init(Vst), kill_state(Vst); valfun_4(timeout, #vst{current=St}=Vst) -> Vst#vst{current=St#st{x=init_regs(0, term)}}; @@ -578,13 +628,6 @@ valfun_4({select_val,Src,{f,Fail},{list,Choices}}, Vst) -> valfun_4({select_tuple_arity,Tuple,{f,Fail},{list,Choices}}, Vst) -> assert_type(tuple, Tuple, Vst), kill_state(branch_arities(Choices, Tuple, branch_state(Fail, Vst))); -valfun_4({get_list,Src,D1,D2}, Vst0) -> - assert_type(cons, Src, Vst0), - Vst = set_type_reg(term, D1, Vst0), - set_type_reg(term, D2, Vst); -valfun_4({get_tuple_element,Src,I,Dst}, Vst) -> - assert_type({tuple_element,I+1}, Src, Vst), - set_type_reg(term, Dst, Vst); %% New bit syntax matching instructions. valfun_4({test,bs_start_match2,{f,Fail},Live,[Ctx,NeedSlots],Ctx}, Vst0) -> @@ -592,9 +635,10 @@ valfun_4({test,bs_start_match2,{f,Fail},Live,[Ctx,NeedSlots],Ctx}, Vst0) -> %% is OK as input. CtxType = get_move_term_type(Ctx, Vst0), verify_live(Live, Vst0), + verify_y_init(Vst0), Vst1 = prune_x_regs(Live, Vst0), BranchVst = case CtxType of - {match_context,_,_} -> + #ms{} -> %% The failure branch will never be taken when Ctx %% is a match context. Therefore, the type for Ctx %% at the failure label must not be match_context @@ -608,9 +652,10 @@ valfun_4({test,bs_start_match2,{f,Fail},Live,[Ctx,NeedSlots],Ctx}, Vst0) -> valfun_4({test,bs_start_match2,{f,Fail},Live,[Src,Slots],Dst}, Vst0) -> assert_term(Src, Vst0), verify_live(Live, Vst0), + verify_y_init(Vst0), Vst1 = prune_x_regs(Live, Vst0), Vst = branch_state(Fail, Vst1), - set_type_reg(bsm_match_state(Slots), Dst, Vst); + set_type_reg(bsm_match_state(Slots), Src, Dst, Vst); valfun_4({test,bs_match_string,{f,Fail},[Ctx,_,_]}, Vst) -> bsm_validate_context(Ctx, Vst), branch_state(Fail, Vst); @@ -631,17 +676,18 @@ valfun_4({test,bs_skip_utf16,{f,Fail},[Ctx,Live,_]}, Vst) -> valfun_4({test,bs_skip_utf32,{f,Fail},[Ctx,Live,_]}, Vst) -> validate_bs_skip_utf(Fail, Ctx, Live, Vst); valfun_4({test,bs_get_integer2,{f,Fail},Live,[Ctx,_,_,_],Dst}, Vst) -> - validate_bs_get(Fail, Ctx, Live, Dst, Vst); + validate_bs_get(Fail, Ctx, Live, {integer, []}, Dst, Vst); valfun_4({test,bs_get_float2,{f,Fail},Live,[Ctx,_,_,_],Dst}, Vst) -> - validate_bs_get(Fail, Ctx, Live, Dst, Vst); + validate_bs_get(Fail, Ctx, Live, {float, []}, Dst, Vst); valfun_4({test,bs_get_binary2,{f,Fail},Live,[Ctx,_,_,_],Dst}, Vst) -> - validate_bs_get(Fail, Ctx, Live, Dst, Vst); + Type = propagate_fragility(term, [Ctx], Vst), + validate_bs_get(Fail, Ctx, Live, Type, Dst, Vst); valfun_4({test,bs_get_utf8,{f,Fail},Live,[Ctx,_],Dst}, Vst) -> - validate_bs_get(Fail, Ctx, Live, Dst, Vst); + validate_bs_get(Fail, Ctx, Live, {integer, []}, Dst, Vst); valfun_4({test,bs_get_utf16,{f,Fail},Live,[Ctx,_],Dst}, Vst) -> - validate_bs_get(Fail, Ctx, Live, Dst, Vst); + validate_bs_get(Fail, Ctx, Live, {integer, []}, Dst, Vst); valfun_4({test,bs_get_utf32,{f,Fail},Live,[Ctx,_],Dst}, Vst) -> - validate_bs_get(Fail, Ctx, Live, Dst, Vst); + validate_bs_get(Fail, Ctx, Live, {integer, []}, Dst, Vst); valfun_4({bs_save2,Ctx,SavePoint}, Vst) -> bsm_save(Ctx, SavePoint, Vst); valfun_4({bs_restore2,Ctx,SavePoint}, Vst) -> @@ -661,6 +707,9 @@ valfun_4({test,is_nonempty_list,{f,Lbl},[Cons]}, Vst) -> valfun_4({test,test_arity,{f,Lbl},[Tuple,Sz]}, Vst) when is_integer(Sz) -> assert_type(tuple, Tuple, Vst), set_type_reg({tuple,Sz}, Tuple, branch_state(Lbl, Vst)); +valfun_4({test,is_tagged_tuple,{f,Lbl},[Src,Sz,_Atom]}, Vst) -> + validate_src([Src], Vst), + set_type_reg({tuple, Sz}, Src, branch_state(Lbl, Vst)); valfun_4({test,has_map_fields,{f,Lbl},Src,{list,List}}, Vst) -> assert_type(map, Src, Vst), assert_unique_map_keys(List), @@ -670,8 +719,10 @@ valfun_4({test,is_map,{f,Lbl},[Src]}, Vst0) -> case Src of {Tag,_} when Tag =:= x; Tag =:= y -> set_type_reg(map, Src, Vst); + {literal,Map} when is_map(Map) -> + Vst; _ -> - Vst + kill_state(Vst) end; valfun_4({test,_Op,{f,Lbl},Src}, Vst) -> validate_src(Src, Vst), @@ -688,6 +739,7 @@ valfun_4({bs_utf16_size,{f,Fail},A,Dst}, Vst) -> set_type_reg({integer,[]}, Dst, branch_state(Fail, Vst)); valfun_4({bs_init2,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> verify_live(Live, Vst0), + verify_y_init(Vst0), if is_integer(Sz) -> ok; @@ -700,6 +752,7 @@ valfun_4({bs_init2,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> set_type_reg(binary, Dst, Vst); valfun_4({bs_init_bits,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> verify_live(Live, Vst0), + verify_y_init(Vst0), if is_integer(Sz) -> ok; @@ -712,6 +765,7 @@ valfun_4({bs_init_bits,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> set_type_reg(binary, Dst, Vst); valfun_4({bs_append,{f,Fail},Bits,Heap,Live,_Unit,Bin,_Flags,Dst}, Vst0) -> verify_live(Live, Vst0), + verify_y_init(Vst0), assert_term(Bits, Vst0), assert_term(Bin, Vst0), Vst1 = heap_alloc(Heap, Vst0), @@ -767,7 +821,7 @@ verify_get_map(Fail, Src, List, Vst0) -> Vst2 = branch_state(Fail, Vst1), Keys = extract_map_keys(List), assert_unique_map_keys(Keys), - verify_get_map_pair(List,Vst0,Vst2). + verify_get_map_pair(List, Src, Vst0, Vst2). extract_map_vals([_Key,Val|T]) -> [Val|extract_map_vals(T)]; @@ -777,10 +831,11 @@ extract_map_keys([Key,_Val|T]) -> [Key|extract_map_keys(T)]; extract_map_keys([]) -> []. -verify_get_map_pair([],_,Vst) -> Vst; -verify_get_map_pair([Src,Dst|Vs],Vst0,Vsti) -> +verify_get_map_pair([Src,Dst|Vs], Map, Vst0, Vsti0) -> assert_term(Src, Vst0), - verify_get_map_pair(Vs,Vst0,set_type_reg(term,Dst,Vsti)). + Vsti = set_type_reg(term, Map, Dst, Vsti0), + verify_get_map_pair(Vs, Map, Vst0, Vsti); +verify_get_map_pair([], _Map, _Vst0, Vst) -> Vst. verify_put_map(Fail, Src, Dst, Live, List, Vst0) -> assert_type(map, Src, Vst0), @@ -797,18 +852,20 @@ verify_put_map(Fail, Src, Dst, Live, List, Vst0) -> %% %% Common code for validating bs_get* instructions. %% -validate_bs_get(Fail, Ctx, Live, Dst, Vst0) -> +validate_bs_get(Fail, Ctx, Live, Type, Dst, Vst0) -> bsm_validate_context(Ctx, Vst0), verify_live(Live, Vst0), + verify_y_init(Vst0), Vst1 = prune_x_regs(Live, Vst0), Vst = branch_state(Fail, Vst1), - set_type_reg(term, Dst, Vst). + set_type_reg(Type, Dst, Vst). %% %% Common code for validating bs_skip_utf* instructions. %% validate_bs_skip_utf(Fail, Ctx, Live, Vst0) -> bsm_validate_context(Ctx, Vst0), + verify_y_init(Vst0), verify_live(Live, Vst0), Vst = prune_x_regs(Live, Vst0), branch_state(Fail, Vst). @@ -818,9 +875,11 @@ validate_bs_skip_utf(Fail, Ctx, Live, Vst0) -> %% A possibility for garbage collection must not occur between setelement/3 and %% set_tuple_element/3. %% +%% Note that #vst.current will be 'none' if the instruction is unreachable. +%% val_dsetel({move,_,_}, Vst) -> Vst; -val_dsetel({call_ext,3,{extfunc,erlang,setelement,3}}, #vst{current=St}=Vst) -> +val_dsetel({call_ext,3,{extfunc,erlang,setelement,3}}, #vst{current=#st{}=St}=Vst) -> Vst#vst{current=St#st{setelem=true}}; val_dsetel({set_tuple_element,_,_,_}, #vst{current=#st{setelem=false}}) -> error(illegal_context_for_set_tuple_element); @@ -847,7 +906,7 @@ kill_state_1(Vst) -> %% The stackframe must be initialized. %% The instruction will return to the instruction following the call. call(Name, Live, #vst{current=St}=Vst) -> - verify_live(Live, Vst), + verify_call_args(Name, Live, Vst), verify_y_init(Vst), case return_type(Name, Vst) of Type when Type =/= exception -> @@ -859,44 +918,74 @@ call(Name, Live, #vst{current=St}=Vst) -> %% Tail call. %% The stackframe must have a known size and be initialized. %% Does not return to the instruction following the call. -tail_call(Name, Live, Vst) -> +tail_call(Name, Live, Vst0) -> + verify_y_init(Vst0), + Vst = deallocate(Vst0), verify_call_args(Name, Live, Vst), - verify_y_init(Vst), verify_no_ct(Vst), kill_state(Vst). verify_call_args(_, 0, #vst{}) -> ok; verify_call_args({f,Lbl}, Live, Vst) when is_integer(Live)-> - Verify = fun(R) -> - case get_move_term_type(R, Vst) of - {match_context,_,_} -> - verify_call_match_context(Lbl, Vst); - _ -> - ok - end - end, - verify_call_args_1(Live, Verify, Vst); + verify_local_call(Lbl, Live, Vst); verify_call_args(_, Live, Vst) when is_integer(Live)-> - Verify = fun(R) -> get_term_type(R, Vst) end, - verify_call_args_1(Live, Verify, Vst); + verify_call_args_1(Live, Vst); verify_call_args(_, Live, _) -> error({bad_number_of_live_regs,Live}). -verify_call_args_1(0, _, _) -> ok; -verify_call_args_1(N, Verify, Vst) -> +verify_call_args_1(0, _) -> ok; +verify_call_args_1(N, Vst) -> X = N - 1, - Verify({x,X}), - verify_call_args_1(X, Verify, Vst). + get_term_type({x,X}, Vst), + verify_call_args_1(X, Vst). + +verify_local_call(Lbl, Live, Vst) -> + case all_ms_in_x_regs(Live, Vst) of + [{R,Ctx}] -> + %% Verify that there is a suitable bs_start_match2 instruction. + verify_call_match_context(Lbl, R, Vst), + + %% Since the callee has consumed the match context, + %% there must be no additional copies in Y registers. + #ms{id=Id} = Ctx, + case ms_in_y_regs(Id, Vst) of + [] -> + ok; + [_|_]=Ys -> + error({multiple_match_contexts,[R|Ys]}) + end; + [_,_|_]=Xs0 -> + Xs = [R || {R,_} <- Xs0], + error({multiple_match_contexts,Xs}); + [] -> + ok + end. + +all_ms_in_x_regs(0, _Vst) -> + []; +all_ms_in_x_regs(Live0, Vst) -> + Live = Live0 - 1, + R = {x,Live}, + case get_move_term_type(R, Vst) of + #ms{}=M -> + [{R,M}|all_ms_in_x_regs(Live, Vst)]; + _ -> + all_ms_in_x_regs(Live, Vst) + end. -verify_call_match_context(Lbl, #vst{ft=Ft}) -> +ms_in_y_regs(Id, #vst{current=#st{y=Ys0}}) -> + Ys = gb_trees:to_list(Ys0), + [{y,Y} || {Y,#ms{id=OtherId}} <- Ys, OtherId =:= Id]. + +verify_call_match_context(Lbl, Ctx, #vst{ft=Ft}) -> case gb_trees:lookup(Lbl, Ft) of none -> error(no_bs_start_match2); {value,[{test,bs_start_match2,_,_,[Ctx,_],Ctx}|_]} -> ok; - {value,[{test,bs_start_match2,_,_,[Bin,_,_],Ctx}|_]} -> - error({binary_and_context_regs_different,Bin,Ctx}) + {value,[{test,bs_start_match2,_,_,_,_}=I|_]} -> + error({unsuitable_bs_start_match2,I}) end. allocate(Zero, Stk, Heap, Live, #vst{current=#st{numy=none}}=Vst0) -> @@ -915,6 +1004,7 @@ deallocate(#vst{current=St}=Vst) -> test_heap(Heap, Live, Vst0) -> verify_live(Live, Vst0), + verify_y_init(Vst0), Vst = prune_x_regs(Live, Vst0), heap_alloc(Heap, Vst). @@ -980,9 +1070,9 @@ get_fls(#vst{current=#st{fls=Fls}}) when is_atom(Fls) -> Fls. init_fregs() -> 0. -set_freg({fr,Fr}, #vst{current=#st{f=Fregs0}=St}=Vst) +set_freg({fr,Fr}=Freg, #vst{current=#st{f=Fregs0}=St}=Vst) when is_integer(Fr), 0 =< Fr -> - limit_check(Fr), + check_limit(Freg), Bit = 1 bsl Fr, if Fregs0 band Bit =:= 0 -> @@ -995,9 +1085,10 @@ set_freg(Fr, _) -> error({bad_target,Fr}). assert_freg_set({fr,Fr}=Freg, #vst{current=#st{f=Fregs}}) when is_integer(Fr), 0 =< Fr -> if - Fregs band (1 bsl Fr) =/= 0 -> - limit_check(Fr); - true -> error({uninitialized_reg,Freg}) + (Fregs bsr Fr) band 1 =:= 0 -> + error({uninitialized_reg,Freg}); + true -> + ok end; assert_freg_set(Fr, _) -> error({bad_source,Fr}). @@ -1027,7 +1118,7 @@ assert_unique_map_keys([_,_|_]=Ls) -> %%% bsm_match_state(Slots) -> - {match_context,0,Slots}. + #ms{slots=Slots}. bsm_validate_context(Reg, Vst) -> _ = bsm_get_context(Reg, Vst), @@ -1035,11 +1126,12 @@ bsm_validate_context(Reg, Vst) -> bsm_get_context({x,X}=Reg, #vst{current=#st{x=Xs}}=_Vst) when is_integer(X) -> case gb_trees:lookup(X, Xs) of - {value,{match_context,_,_}=Ctx} -> Ctx; + {value,#ms{}=Ctx} -> Ctx; + {value,{fragile,#ms{}=Ctx}} -> Ctx; _ -> error({no_bsm_context,Reg}) end; bsm_get_context(Reg, _) -> error({bad_source,Reg}). - + bsm_save(Reg, {atom,start}, Vst) -> %% Save point refering to where the match started. %% It is always valid. But don't forget to validate the context register. @@ -1047,8 +1139,8 @@ bsm_save(Reg, {atom,start}, Vst) -> Vst; bsm_save(Reg, SavePoint, Vst) -> case bsm_get_context(Reg, Vst) of - {match_context,Bits,Slots} when SavePoint < Slots -> - Ctx = {match_context,Bits bor (1 bsl SavePoint),Slots}, + #ms{valid=Bits,slots=Slots}=Ctxt0 when SavePoint < Slots -> + Ctx = Ctxt0#ms{valid=Bits bor (1 bsl SavePoint),slots=Slots}, set_type_reg(Ctx, Reg, Vst); _ -> error({illegal_save,SavePoint}) end. @@ -1060,7 +1152,7 @@ bsm_restore(Reg, {atom,start}, Vst) -> Vst; bsm_restore(Reg, SavePoint, Vst) -> case bsm_get_context(Reg, Vst) of - {match_context,Bits,Slots} when SavePoint < Slots -> + #ms{valid=Bits,slots=Slots} when SavePoint < Slots -> case Bits band (1 bsl SavePoint) of 0 -> error({illegal_restore,SavePoint,not_set}); _ -> Vst @@ -1076,16 +1168,37 @@ set_type(Type, {x,_}=Reg, Vst) -> set_type_reg(Type, Reg, Vst); set_type(Type, {y,_}=Reg, Vst) -> set_type_y(Type, Reg, Vst); set_type(_, _, #vst{}=Vst) -> Vst. -set_type_reg(Type, {x,X}, #vst{current=#st{x=Xs}=St}=Vst) - when is_integer(X), 0 =< X -> - limit_check(X), - Vst#vst{current=St#st{x=gb_trees:enter(X, Type, Xs)}}; +set_type_reg(Type, Src, Dst, Vst) -> + case get_term_type_1(Src, Vst) of + {fragile,_} -> + set_type_reg(make_fragile(Type), Dst, Vst); + _ -> + set_type_reg(Type, Dst, Vst) + end. + +set_type_reg(Type, {x,_}=Reg, Vst) -> + set_type_x(Type, Reg, Vst); set_type_reg(Type, Reg, Vst) -> set_type_y(Type, Reg, Vst). +set_type_x(Type, {x,X}=Reg, #vst{current=#st{x=Xs0}=St}=Vst) + when is_integer(X), 0 =< X -> + check_limit(Reg), + Xs = case gb_trees:lookup(X, Xs0) of + none -> + gb_trees:insert(X, Type, Xs0); + {value,{fragile,_}} -> + gb_trees:update(X, make_fragile(Type), Xs0); + {value,_} -> + gb_trees:update(X, Type, Xs0) + end, + Vst#vst{current=St#st{x=Xs}}; +set_type_x(Type, Reg, #vst{}) -> + error({invalid_store,Reg,Type}). + set_type_y(Type, {y,Y}=Reg, #vst{current=#st{y=Ys0}=St}=Vst) when is_integer(Y), 0 =< Y -> - limit_check(Y), + check_limit(Reg), Ys = case gb_trees:lookup(Y, Ys0) of none -> error({invalid_store,Reg,Type}); @@ -1096,13 +1209,40 @@ set_type_y(Type, {y,Y}=Reg, #vst{current=#st{y=Ys0}=St}=Vst) {value,_} -> gb_trees:update(Y, Type, Ys0) end, + check_try_catch_tags(Type, Y, Ys0), Vst#vst{current=St#st{y=Ys}}; set_type_y(Type, Reg, #vst{}) -> error({invalid_store,Reg,Type}). +make_fragile({fragile,_}=Type) -> Type; +make_fragile(Type) -> {fragile,Type}. + set_catch_end({y,Y}, #vst{current=#st{y=Ys0}=St}=Vst) -> Ys = gb_trees:update(Y, initialized, Ys0), Vst#vst{current=St#st{y=Ys}}. +check_try_catch_tags(Type, LastY, Ys) -> + case is_try_catch_tag(Type) of + false -> + ok; + true -> + %% Every catch or try/catch must use a lower Y register + %% number than any enclosing catch or try/catch. That will + %% ensure that when the stack is scanned when an + %% exception occurs, the innermost try/catch tag is found + %% first. + Bad = [{{y,Y},Tag} || {Y,Tag} <- gb_trees:to_list(Ys), + Y < LastY, is_try_catch_tag(Tag)], + case Bad of + [] -> + ok; + [_|_] -> + error({bad_try_catch_nesting,{y,LastY},Bad}) + end + end. + +is_try_catch_tag({catchtag,_}) -> true; +is_try_catch_tag({trytag,_}) -> true; +is_try_catch_tag(_) -> false. is_reg_defined({x,_}=Reg, Vst) -> is_type_defined_x(Reg, Vst); is_reg_defined({y,_}=Reg, Vst) -> is_type_defined_y(Reg, Vst); @@ -1141,7 +1281,7 @@ assert_term(Src, Vst) -> %% Thus 'exception' is never stored as type descriptor %% for a register. %% -%% {match_context,_,_} A matching context for bit syntax matching. We do allow +%% #ms{} A match context for bit syntax matching. We do allow %% it to moved/to from stack, but otherwise it must only %% be accessed by bit syntax matching instructions. %% @@ -1176,19 +1316,41 @@ assert_term(Src, Vst) -> %% %% map Map. %% +%% +%% +%% FRAGILITY +%% --------- +%% +%% The loop_rec/2 instruction may return a reference to a term that is +%% not part of the root set. That term or any part of it must not be +%% included in a garbage collection. Therefore, the term (or any part +%% of it) must not be stored in an Y register. +%% +%% Such terms are wrapped in a {fragile,Type} tuple, where Type is one +%% of the types described above. assert_type(WantedType, Term, Vst) -> - assert_type(WantedType, get_term_type(Term, Vst)). + case get_term_type(Term, Vst) of + {fragile,Type} -> + assert_type(WantedType, Type); + Type -> + assert_type(WantedType, Type) + end. assert_type(Correct, Correct) -> ok; assert_type(float, {float,_}) -> ok; assert_type(tuple, {tuple,_}) -> ok; +assert_type(tuple, {literal,Tuple}) when is_tuple(Tuple) -> ok; assert_type({tuple_element,I}, {tuple,[Sz]}) when 1 =< I, I =< Sz -> ok; assert_type({tuple_element,I}, {tuple,Sz}) when is_integer(Sz), 1 =< I, I =< Sz -> ok; +assert_type({tuple_element,I}, {literal,Lit}) when I =< tuple_size(Lit) -> + ok; +assert_type(cons, {literal,[_|_]}) -> + ok; assert_type(Needed, Actual) -> error({bad_type,{needed,Needed},{actual,Actual}}). @@ -1199,14 +1361,19 @@ assert_type(Needed, Actual) -> %% is inconsistent, and we know that some instructions will never %% be executed at run-time. -upgrade_tuple_type({tuple,[Sz]}, {tuple,[OldSz]}=T) when Sz < OldSz -> +upgrade_tuple_type(NewType, {fragile,OldType}) -> + make_fragile(upgrade_tuple_type_1(NewType, OldType)); +upgrade_tuple_type(NewType, OldType) -> + upgrade_tuple_type_1(NewType, OldType). + +upgrade_tuple_type_1({tuple,[Sz]}, {tuple,[OldSz]}=T) when Sz < OldSz -> %% The old type has a higher value for the least tuple size. T; -upgrade_tuple_type({tuple,[Sz]}, {tuple,OldSz}=T) +upgrade_tuple_type_1({tuple,[Sz]}, {tuple,OldSz}=T) when is_integer(Sz), is_integer(OldSz), Sz =< OldSz -> %% The old size is exact, and the new size is smaller than the old size. T; -upgrade_tuple_type({tuple,_}=T, _) -> +upgrade_tuple_type_1({tuple,_}=T, _) -> %% The new type information is exact or has a higher value for %% the least tuple size. %% Note that inconsistencies are also handled in this @@ -1231,6 +1398,7 @@ get_move_term_type(Src, Vst) -> initialized -> error({unassigned,Src}); {catchtag,_} -> error({catchtag,Src}); {trytag,_} -> error({trytag,Src}); + tuple_in_progress -> error({tuple_in_progress,Src}); Type -> Type end. @@ -1239,11 +1407,8 @@ get_move_term_type(Src, Vst) -> %% a standard Erlang type (no catch/try tags or match contexts). get_term_type(Src, Vst) -> - case get_term_type_1(Src, Vst) of - initialized -> error({unassigned,Src}); - {catchtag,_} -> error({catchtag,Src}); - {trytag,_} -> error({trytag,Src}); - {match_context,_,_} -> error({match_context,Src}); + case get_move_term_type(Src, Vst) of + #ms{} -> error({match_context,Src}); Type -> Type end. @@ -1289,7 +1454,12 @@ branch_arities([Sz,{f,L}|T], Tuple, #vst{current=St}=Vst0) Vst = branch_state(L, Vst1), branch_arities(T, Tuple, Vst#vst{current=St}). -branch_state(0, #vst{}=Vst) -> Vst; +branch_state(0, #vst{}=Vst) -> + %% If the instruction fails, the stack may be scanned + %% looking for a catch tag. Therefore the Y registers + %% must be initialized at this point. + verify_y_init(Vst), + Vst; branch_state(L, #vst{current=St,branched=B}=Vst) -> Vst#vst{ branched=case gb_trees:is_defined(L, B) of @@ -1370,6 +1540,14 @@ merge_y_regs_1(_, _, Regs) -> Regs. %% merge_types(Type1, Type2) -> Type %% Return the most specific type possible. %% Note: Type1 must NOT be the same as Type2. +merge_types({fragile,Same}=Type, Same) -> + Type; +merge_types({fragile,T1}, T2) -> + make_fragile(merge_types(T1, T2)); +merge_types(Same, {fragile,Same}=Type) -> + Type; +merge_types(T1, {fragile,T2}) -> + make_fragile(merge_types(T1, T2)); merge_types(uninitialized=I, _) -> I; merge_types(_, uninitialized=I) -> I; merge_types(initialized=I, _) -> I; @@ -1395,12 +1573,13 @@ merge_types(bool, {atom,A}) -> merge_bool(A); merge_types({atom,A}, bool) -> merge_bool(A); -merge_types({match_context,B0,Slots},{match_context,B1,Slots}) -> - {match_context,B0 bor B1,Slots}; -merge_types({match_context,_,_}=M, _) -> - M; -merge_types(_, {match_context,_,_}=M) -> - M; +merge_types(#ms{id=Id1,valid=B1,slots=Slots1}, + #ms{id=Id2,valid=B2,slots=Slots2}) -> + Id = if + Id1 =:= Id2 -> Id1; + true -> make_ref() + end, + #ms{id=Id,valid=B1 band B2,slots=min(Slots1, Slots2)}; merge_types(T1, T2) when T1 =/= T2 -> %% Too different. All we know is that the type is a 'term'. term. @@ -1419,6 +1598,10 @@ verify_y_init(#vst{current=#st{y=Ys}}) -> verify_y_init_1([]) -> ok; verify_y_init_1([{Y,uninitialized}|_]) -> error({uninitialized_reg,{y,Y}}); +verify_y_init_1([{Y,{fragile,_}}|_]) -> + %% Unsafe. This term may be outside any heap belonging + %% to the process and would be corrupted by a GC. + error({fragile_message_reference,{y,Y}}); verify_y_init_1([{_,_}|Ys]) -> verify_y_init_1(Ys). @@ -1464,6 +1647,27 @@ eat_heap_float(#vst{current=#st{hf=HeapFloats0}=St}=Vst) -> Vst#vst{current=St#st{hf=HeapFloats}} end. +remove_fragility(#vst{current=#st{x=Xs0,y=Ys0}=St0}=Vst) -> + F = fun(_, {fragile,Type}) -> Type; + (_, Type) -> Type + end, + Xs = gb_trees:map(F, Xs0), + Ys = gb_trees:map(F, Ys0), + St = St0#st{x=Xs,y=Ys}, + Vst#vst{current=St}. + +propagate_fragility(Type, Ss, Vst) -> + F = fun(S) -> + case get_term_type_1(S, Vst) of + {fragile,_} -> true; + _ -> false + end + end, + case any(F, Ss) of + true -> make_fragile(Type); + false -> Type + end. + bif_type('-', Src, Vst) -> arith_type(Src, Vst); bif_type('+', Src, Vst) -> @@ -1479,7 +1683,9 @@ bif_type(abs, [Num], Vst) -> bif_type(float, _, _) -> {float,[]}; bif_type('/', _, _) -> {float,[]}; %% Integer operations. +bif_type(ceil, [_], _) -> {integer,[]}; bif_type('div', [_,_], _) -> {integer,[]}; +bif_type(floor, [_], _) -> {integer,[]}; bif_type('rem', [_,_], _) -> {integer,[]}; bif_type(length, [_], _) -> {integer,[]}; bif_type(size, [_], _) -> {integer,[]}; @@ -1523,7 +1729,6 @@ bif_type(node, [_], _) -> {atom,[]}; bif_type(hd, [_], _) -> term; bif_type(tl, [_], _) -> term; bif_type(get, [_], _) -> term; -bif_type(raise, [_,_], _) -> exception; bif_type(Bif, _, _) when is_atom(Bif) -> term. is_bif_safe('/=', 2) -> true; @@ -1537,6 +1742,7 @@ is_bif_safe('>=', 2) -> true; is_bif_safe(is_atom, 1) -> true; is_bif_safe(is_boolean, 1) -> true; is_bif_safe(is_binary, 1) -> true; +is_bif_safe(is_bitstring, 1) -> true; is_bif_safe(is_float, 1) -> true; is_bif_safe(is_function, 1) -> true; is_bif_safe(is_integer, 1) -> true; @@ -1567,8 +1773,12 @@ return_type_1(erlang, setelement, 3, Vst) -> Tuple = {x,1}, TupleType = case get_term_type(Tuple, Vst) of - {tuple,_}=TT -> TT; - _ -> {tuple,[0]} + {tuple,_}=TT -> + TT; + {literal,Lit} when is_tuple(Lit) -> + {tuple,tuple_size(Lit)}; + _ -> + {tuple,[0]} end, case get_term_type({x,0}, Vst) of {integer,[]} -> TupleType; @@ -1609,20 +1819,25 @@ return_type_math(log10, 1) -> {float,[]}; return_type_math(sqrt, 1) -> {float,[]}; return_type_math(atan2, 2) -> {float,[]}; return_type_math(pow, 2) -> {float,[]}; +return_type_math(ceil, 1) -> {float,[]}; +return_type_math(floor, 1) -> {float,[]}; +return_type_math(fmod, 2) -> {float,[]}; return_type_math(pi, 0) -> {float,[]}; return_type_math(F, A) when is_atom(F), is_integer(A), A >= 0 -> term. -limit_check(Num) when is_integer(Num), Num >= ?MAXREG -> - error(limit); -limit_check(_) -> ok. +check_limit({x,X}) when is_integer(X), X < 1023 -> + %% Note: x(1023) is reserved for use by the BEAM loader. + ok; +check_limit({y,Y}) when is_integer(Y), Y < 1024 -> + ok; +check_limit({fr,Fr}) when is_integer(Fr), Fr < 1024 -> + ok; +check_limit(_) -> + error(limit). min(A, B) when is_integer(A), is_integer(B), A < B -> A; min(A, B) when is_integer(A), is_integer(B) -> B. gb_trees_from_list(L) -> gb_trees:from_orddict(lists:sort(L)). --ifdef(DEBUG). -error(Error) -> exit(Error). --else. error(Error) -> throw(Error). --endif. diff --git a/lib/compiler/src/beam_z.erl b/lib/compiler/src/beam_z.erl index 8381578b68..1c9d762eb1 100644 --- a/lib/compiler/src/beam_z.erl +++ b/lib/compiler/src/beam_z.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2012. All Rights Reserved. +%% Copyright Ericsson AB 2012-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -24,17 +24,23 @@ -export([module/2]). -module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> - Fs = [function(F) || F <- Fs0], +-import(lists, [dropwhile/2,map/2]). + +-spec module(beam_utils:module_code(), [compile:option()]) -> + {'ok',beam_asm:module_code()}. + +module({Mod,Exp,Attr,Fs0,Lc}, Opts) -> + NoGetHdTl = proplists:get_bool(no_get_hd_tl, Opts), + Fs = [function(F, NoGetHdTl) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. -function({function,Name,Arity,CLabel,Is0}) -> +function({function,Name,Arity,CLabel,Is0}, NoGetHdTl) -> try - Is = undo_renames(Is0), + Is1 = undo_renames(Is0), + Is = maybe_eliminate_get_hd_tl(Is1, NoGetHdTl), {function,Name,Arity,CLabel,Is} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [Name,Arity]), erlang:raise(Class, Error, Stack) end. @@ -51,6 +57,20 @@ undo_renames([{call,A,F},return|Is]) -> [{call_only,A,F}|undo_renames(Is)]; undo_renames([{call_ext,A,F},return|Is]) -> [{call_ext_only,A,F}|undo_renames(Is)]; +undo_renames([{bif,raise,_,_,_}=I|Is0]) -> + %% A minor optimization. Done here because: + %% (1) beam_jump may move or share 'raise' instructions, and that + %% may confuse beam_validator. + %% (2) beam_trim cannot do its optimization if the 'deallocate' + %% instruction after 'raise' has been removed. + Is = dropwhile(fun({label,_}) -> false; + (_) -> true + end, Is0), + [I|undo_renames(Is)]; +undo_renames([{get_hd,Src,Dst1},{get_tl,Src,Dst2}|Is]) -> + [{get_list,Src,Dst1,Dst2}|undo_renames(Is)]; +undo_renames([{get_tl,Src,Dst2},{get_hd,Src,Dst1}|Is]) -> + [{get_list,Src,Dst1,Dst2}|undo_renames(Is)]; undo_renames([I|Is]) -> [undo_rename(I)|undo_renames(Is)]; undo_renames([]) -> []. @@ -93,3 +113,17 @@ undo_rename({get_map_elements,Fail,Src,{list,List}}) -> undo_rename({select,I,Reg,Fail,List}) -> {I,Reg,Fail,{list,List}}; undo_rename(I) -> I. + +%%% +%%% Eliminate get_hd/get_tl instructions if requested by +%%% the no_get_hd_tl option. +%%% + +maybe_eliminate_get_hd_tl(Is, true) -> + map(fun({get_hd,Cons,Hd}) -> + {get_list,Cons,Hd,{x,1022}}; + ({get_tl,Cons,Tl}) -> + {get_list,Cons,{x,1022},Tl}; + (I) -> I + end, Is); +maybe_eliminate_get_hd_tl(Is, false) -> Is. diff --git a/lib/compiler/src/cerl.erl b/lib/compiler/src/cerl.erl index e7a2b8177a..fce23bfd68 100644 --- a/lib/compiler/src/cerl.erl +++ b/lib/compiler/src/cerl.erl @@ -1,8 +1,3 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 2001-2010. All Rights Reserved. -%% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. %% You may obtain a copy of the License at @@ -15,9 +10,8 @@ %% See the License for the specific language governing permissions and %% limitations under the License. %% -%% %CopyrightEnd% - -%% ===================================================================== +%% @copyright 1999-2002 Richard Carlsson +%% @author Richard Carlsson <[email protected]> %% @doc Core Erlang abstract syntax trees. %% %% <p> This module defines an abstract data type for representing Core @@ -126,6 +120,7 @@ %% keep map exports here for now c_map_pattern/1, is_c_map/1, + is_c_map_pattern/1, map_es/1, map_arg/1, update_c_map/3, @@ -134,7 +129,7 @@ ann_c_map_pattern/2, map_pair_op/1,map_pair_key/1,map_pair_val/1, update_c_map_pair/4, - c_map_pair/2, + c_map_pair/2, c_map_pair_exact/2, ann_c_map_pair/4 ]). @@ -438,6 +433,8 @@ is_literal_term(T) when is_tuple(T) -> is_literal_term(B) when is_bitstring(B) -> true; is_literal_term(M) when is_map(M) -> is_literal_term_list(maps:to_list(M)); +is_literal_term(F) when is_function(F) -> + erlang:fun_info(F, type) =:= {type,external}; is_literal_term(_) -> false. @@ -1589,6 +1586,8 @@ ann_make_list(_, [], Node) -> %% @doc Returns <code>true</code> if <code>Node</code> is an abstract %% map constructor, otherwise <code>false</code>. +-type map_op() :: #c_literal{val::'assoc'} | #c_literal{val::'exact'}. + -spec is_c_map(cerl()) -> boolean(). is_c_map(#c_map{}) -> @@ -1636,6 +1635,11 @@ is_c_map_empty(#c_map{ es=[] }) -> true; is_c_map_empty(#c_literal{val=M}) when is_map(M),map_size(M) =:= 0 -> true; is_c_map_empty(_) -> false. +-spec is_c_map_pattern(c_map()) -> boolean(). + +is_c_map_pattern(#c_map{is_pat=IsPat}) -> + IsPat. + -spec ann_c_map([term()], [c_map_pair()]) -> c_map() | c_literal(). ann_c_map(As, Es) -> @@ -1679,8 +1683,16 @@ update_c_map(#c_map{is_pat=true}=Old, M, Es) -> update_c_map(#c_map{is_pat=false}=Old, M, Es) -> ann_c_map(get_ann(Old), M, Es). +-spec map_pair_key(c_map_pair()) -> cerl(). + map_pair_key(#c_map_pair{key=K}) -> K. + +-spec map_pair_val(c_map_pair()) -> cerl(). + map_pair_val(#c_map_pair{val=V}) -> V. + +-spec map_pair_op(c_map_pair()) -> map_op(). + map_pair_op(#c_map_pair{op=Op}) -> Op. -spec c_map_pair(cerl(), cerl()) -> c_map_pair(). @@ -1688,12 +1700,19 @@ map_pair_op(#c_map_pair{op=Op}) -> Op. c_map_pair(Key,Val) -> #c_map_pair{op=#c_literal{val=assoc},key=Key,val=Val}. +-spec c_map_pair_exact(cerl(), cerl()) -> c_map_pair(). + +c_map_pair_exact(Key,Val) -> + #c_map_pair{op=#c_literal{val=exact},key=Key,val=Val}. + -spec ann_c_map_pair([term()], cerl(), cerl(), cerl()) -> c_map_pair(). ann_c_map_pair(As,Op,K,V) -> #c_map_pair{op=Op, key = K, val=V, anno = As}. +-spec update_c_map_pair(c_map_pair(), map_op(), cerl(), cerl()) -> c_map_pair(). + update_c_map_pair(Old,Op,K,V) -> #c_map_pair{op=Op, key=K, val=V, anno = get_ann(Old)}. @@ -1944,7 +1963,7 @@ is_c_var(_) -> false. -%% @spec c_fname(Name::atom(), Arity::integer()) -> cerl() +%% @spec c_fname(Name::atom(), Arity::arity()) -> cerl() %% @equiv c_var({Name, Arity}) %% @see fname_id/1 %% @see fname_arity/1 @@ -1952,18 +1971,18 @@ is_c_var(_) -> %% @see ann_c_fname/3 %% @see update_c_fname/3 --spec c_fname(atom(), non_neg_integer()) -> c_var(). +-spec c_fname(atom(), arity()) -> c_var(). c_fname(Atom, Arity) -> c_var({Atom, Arity}). -%% @spec ann_c_fname(As::[term()], Name::atom(), Arity::integer()) -> +%% @spec ann_c_fname(As::[term()], Name::atom(), Arity::arity()) -> %% cerl() %% @equiv ann_c_var(As, {Atom, Arity}) %% @see c_fname/2 --spec ann_c_fname([term()], atom(), non_neg_integer()) -> c_var(). +-spec ann_c_fname([term()], atom(), arity()) -> c_var(). ann_c_fname(As, Atom, Arity) -> ann_c_var(As, {Atom, Arity}). @@ -1981,13 +2000,13 @@ update_c_fname(#c_var{name = {_, Arity}, anno = As}, Atom) -> #c_var{name = {Atom, Arity}, anno = As}. -%% @spec update_c_fname(Old::cerl(), Name::atom(), Arity::integer()) -> +%% @spec update_c_fname(Old::cerl(), Name::atom(), Arity::arity()) -> %% cerl() %% @equiv update_c_var(Old, {Atom, Arity}) %% @see update_c_fname/2 %% @see c_fname/2 --spec update_c_fname(c_var(), atom(), integer()) -> c_var(). +-spec update_c_fname(c_var(), atom(), arity()) -> c_var(). update_c_fname(Node, Atom, Arity) -> update_c_var(Node, {Atom, Arity}). @@ -2036,14 +2055,14 @@ fname_id(#c_var{name={A,_}}) -> A. -%% @spec fname_arity(cerl()) -> byte() +%% @spec fname_arity(cerl()) -> arity() %% %% @doc Returns the arity part of an abstract function name variable. %% %% @see fname_id/1 %% @see c_fname/2 --spec fname_arity(c_var()) -> byte(). +-spec fname_arity(c_var()) -> arity(). fname_arity(#c_var{name={_,N}}) -> N. @@ -2489,7 +2508,7 @@ fun_body(Node) -> Node#c_fun.body. -%% @spec fun_arity(Node::cerl()) -> integer() +%% @spec fun_arity(Node::cerl()) -> arity() %% %% @doc Returns the number of parameter subtrees of an abstract %% fun-expression. @@ -2500,7 +2519,7 @@ fun_body(Node) -> %% @see c_fun/2 %% @see fun_vars/1 --spec fun_arity(c_fun()) -> non_neg_integer(). +-spec fun_arity(c_fun()) -> arity(). fun_arity(Node) -> length(fun_vars(Node)). @@ -3407,7 +3426,7 @@ apply_args(Node) -> Node#c_apply.args. -%% @spec apply_arity(Node::cerl()) -> integer() +%% @spec apply_arity(Node::cerl()) -> arity() %% %% @doc Returns the number of argument subtrees of an abstract %% function application. @@ -3419,7 +3438,7 @@ apply_args(Node) -> %% @see c_apply/2 %% @see apply_args/1 --spec apply_arity(c_apply()) -> non_neg_integer(). +-spec apply_arity(c_apply()) -> arity(). apply_arity(Node) -> length(apply_args(Node)). @@ -3525,7 +3544,7 @@ call_args(Node) -> Node#c_call.args. -%% @spec call_arity(Node::cerl()) -> integer() +%% @spec call_arity(Node::cerl()) -> arity() %% %% @doc Returns the number of argument subtrees of an abstract %% inter-module call. @@ -3537,7 +3556,7 @@ call_args(Node) -> %% @see c_call/3 %% @see call_args/1 --spec call_arity(c_call()) -> non_neg_integer(). +-spec call_arity(c_call()) -> arity(). call_arity(Node) -> length(call_args(Node)). @@ -3629,7 +3648,7 @@ primop_args(Node) -> Node#c_primop.args. -%% @spec primop_arity(Node::cerl()) -> integer() +%% @spec primop_arity(Node::cerl()) -> arity() %% %% @doc Returns the number of argument subtrees of an abstract %% primitive operation call. @@ -3641,7 +3660,7 @@ primop_args(Node) -> %% @see c_primop/2 %% @see primop_args/1 --spec primop_arity(c_primop()) -> non_neg_integer(). +-spec primop_arity(c_primop()) -> arity(). primop_arity(Node) -> length(primop_args(Node)). diff --git a/lib/compiler/src/cerl_clauses.erl b/lib/compiler/src/cerl_clauses.erl index 4d2f1ebd1b..fa5104c01b 100644 --- a/lib/compiler/src/cerl_clauses.erl +++ b/lib/compiler/src/cerl_clauses.erl @@ -1,8 +1,3 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 2001-2010. All Rights Reserved. -%% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. %% You may obtain a copy of the License at @@ -15,8 +10,8 @@ %% See the License for the specific language governing permissions and %% limitations under the License. %% -%% %CopyrightEnd% - +%% @copyright 1999-2002 Richard Carlsson +%% @author Richard Carlsson <[email protected]> %% @doc Utility functions for Core Erlang case/receive clauses. %% %% <p>Syntax trees are defined in the module <a @@ -358,6 +353,8 @@ match(P, E, Bs) -> map -> %% The most we can do is to say "definitely no match" if a %% map pattern is matched against non-map data. + %% (Note: See the document internal_doc/cerl-notes.md for + %% information why we don't try to do more here.) case E of any -> {false, Bs}; diff --git a/lib/compiler/src/cerl_inline.erl b/lib/compiler/src/cerl_inline.erl index cbcacf9e8e..caff47dbcb 100644 --- a/lib/compiler/src/cerl_inline.erl +++ b/lib/compiler/src/cerl_inline.erl @@ -1,8 +1,3 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 2001-2012. All Rights Reserved. -%% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. %% You may obtain a copy of the License at @@ -15,9 +10,9 @@ %% See the License for the specific language governing permissions and %% limitations under the License. %% -%% %CopyrightEnd% -%% -%% Core Erlang inliner. +%% @copyright 1999-2002 Richard Carlsson +%% @author Richard Carlsson <[email protected]> +%% @doc Core Erlang inliner. %% ===================================================================== %% @@ -1827,6 +1822,14 @@ new_var(Env) -> Name = env__new_vname(Env), c_var(Name). +%% The way a template variable is used makes it necessary +%% to make sure that it is unique in the entire function. +%% Therefore, template variables are atoms with the prefix "@i". + +new_template_var(Env) -> + Name = env__new_tname(Env), + c_var(Name). + residualize_var(R, S) -> S1 = count_size(weight(var), S), {ref_to_var(R), st__set_var_referenced(R#ref.loc, S1)}. @@ -2188,7 +2191,7 @@ make_template(E, Vs0, Env0) -> T = make_data_skel(data_type(E), Ts), E1 = update_data(E, data_type(E), [hd(get_ann(T)) || T <- Ts]), - V = new_var(Env1), + V = new_template_var(Env1), Env2 = env__bind(var_name(V), E1, Env1), {set_ann(T, [V]), [V | Vs1], Env2}; false -> @@ -2203,7 +2206,7 @@ make_template(E, Vs0, Env0) -> Env2 = env__bind(V, E1, Env1), {T, Vs1, Env2}; _ -> - V = new_var(Env0), + V = new_template_var(Env0), Env1 = env__bind(var_name(V), E, Env0), {set_ann(V, [V]), [V | Vs0], Env1} end @@ -2569,6 +2572,11 @@ env__is_defined(Key, Env) -> env__new_vname(Env) -> rec_env:new_key(Env). +env__new_tname(Env) -> + rec_env:new_key(fun(I) -> + list_to_atom("@i"++integer_to_list(I)) + end, Env). + env__new_fname(A, N, Env) -> rec_env:new_key(fun (X) -> S = integer_to_list(X), diff --git a/lib/compiler/src/cerl_trees.erl b/lib/compiler/src/cerl_trees.erl index 58bb18e34a..533c984221 100644 --- a/lib/compiler/src/cerl_trees.erl +++ b/lib/compiler/src/cerl_trees.erl @@ -1,8 +1,3 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 2001-2010. All Rights Reserved. -%% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. %% You may obtain a copy of the License at @@ -15,8 +10,8 @@ %% See the License for the specific language governing permissions and %% limitations under the License. %% -%% %CopyrightEnd% - +%% @copyright 1999-2002 Richard Carlsson. +%% @author Richard Carlsson <[email protected]> %% @doc Basic functions on Core Erlang abstract syntax trees. %% %% <p>Syntax trees are defined in the module <a @@ -27,7 +22,8 @@ -module(cerl_trees). -export([depth/1, fold/3, free_variables/1, get_label/1, label/1, label/2, - map/2, mapfold/3, size/1, variables/1]). + map/2, mapfold/3, mapfold/4, next_free_variable_name/1, + size/1, variables/1]). -import(cerl, [alias_pat/1, alias_var/1, ann_c_alias/3, ann_c_apply/3, ann_c_binary/2, ann_c_bitstr/6, ann_c_call/4, @@ -61,6 +57,7 @@ map_arg/1, map_es/1, ann_c_map/3, update_c_map/3, + is_c_map_pattern/1, ann_c_map_pattern/2, map_pair_key/1,map_pair_val/1,map_pair_op/1, ann_c_map_pair/4, update_c_map_pair/4 @@ -340,136 +337,161 @@ fold_pairs(_, S, []) -> %% starting with the given value <code>Initial</code>, while doing a %% post-order traversal of the tree, much like <code>fold/3</code>. %% +%% This is the same as mapfold/4, with an identity function as the +%% pre-operation. +%% %% @see map/2 %% @see fold/3 +%% @see mapfold/4 -spec mapfold(fun((cerl:cerl(), term()) -> {cerl:cerl(), term()}), term(), cerl:cerl()) -> {cerl:cerl(), term()}. mapfold(F, S0, T) -> + mapfold(fun(T0, A) -> {T0, A} end, F, S0, T). + + +%% @spec mapfold(Pre, Post, Initial::term(), Tree::cerl()) -> {cerl(), term()} +%% Pre = (cerl(), term()) -> {cerl(), term()} +%% Post = (cerl(), term()) -> {cerl(), term()} +%% +%% @doc Does a combined map/fold operation on the nodes of the +%% tree. It begins by calling <code>Pre</code> on the tree, using the +%% <code>Initial</code> value. It then deconstructs the top node of +%% the returned tree and recurses on the children, using the returned +%% value as the new initial and carrying the returned values from one +%% call to the next. Finally it reassembles the top node from the +%% children, calls <code>Post</code> on it and returns the result. + +-spec mapfold(fun((cerl:cerl(), term()) -> {cerl:cerl(), term()}), + fun((cerl:cerl(), term()) -> {cerl:cerl(), term()}), + term(), cerl:cerl()) -> {cerl:cerl(), term()}. + +mapfold(Pre, Post, S00, T0) -> + {T, S0} = Pre(T0, S00), case type(T) of literal -> case concrete(T) of [_ | _] -> - {T1, S1} = mapfold(F, S0, cons_hd(T)), - {T2, S2} = mapfold(F, S1, cons_tl(T)), - F(update_c_cons(T, T1, T2), S2); + {T1, S1} = mapfold(Pre, Post, S0, cons_hd(T)), + {T2, S2} = mapfold(Pre, Post, S1, cons_tl(T)), + Post(update_c_cons(T, T1, T2), S2); V when tuple_size(V) > 0 -> - {Ts, S1} = mapfold_list(F, S0, tuple_es(T)), - F(update_c_tuple(T, Ts), S1); + {Ts, S1} = mapfold_list(Pre, Post, S0, tuple_es(T)), + Post(update_c_tuple(T, Ts), S1); _ -> - F(T, S0) + Post(T, S0) end; var -> - F(T, S0); + Post(T, S0); values -> - {Ts, S1} = mapfold_list(F, S0, values_es(T)), - F(update_c_values(T, Ts), S1); + {Ts, S1} = mapfold_list(Pre, Post, S0, values_es(T)), + Post(update_c_values(T, Ts), S1); cons -> - {T1, S1} = mapfold(F, S0, cons_hd(T)), - {T2, S2} = mapfold(F, S1, cons_tl(T)), - F(update_c_cons_skel(T, T1, T2), S2); + {T1, S1} = mapfold(Pre, Post, S0, cons_hd(T)), + {T2, S2} = mapfold(Pre, Post, S1, cons_tl(T)), + Post(update_c_cons_skel(T, T1, T2), S2); tuple -> - {Ts, S1} = mapfold_list(F, S0, tuple_es(T)), - F(update_c_tuple_skel(T, Ts), S1); + {Ts, S1} = mapfold_list(Pre, Post, S0, tuple_es(T)), + Post(update_c_tuple_skel(T, Ts), S1); map -> - {M , S1} = mapfold(F, S0, map_arg(T)), - {Ts, S2} = mapfold_list(F, S1, map_es(T)), - F(update_c_map(T, M, Ts), S2); + {M , S1} = mapfold(Pre, Post, S0, map_arg(T)), + {Ts, S2} = mapfold_list(Pre, Post, S1, map_es(T)), + Post(update_c_map(T, M, Ts), S2); map_pair -> - {Op, S1} = mapfold(F, S0, map_pair_op(T)), - {Key, S2} = mapfold(F, S1, map_pair_key(T)), - {Val, S3} = mapfold(F, S2, map_pair_val(T)), - F(update_c_map_pair(T,Op,Key,Val), S3); + {Op, S1} = mapfold(Pre, Post, S0, map_pair_op(T)), + {Key, S2} = mapfold(Pre, Post, S1, map_pair_key(T)), + {Val, S3} = mapfold(Pre, Post, S2, map_pair_val(T)), + Post(update_c_map_pair(T,Op,Key,Val), S3); 'let' -> - {Vs, S1} = mapfold_list(F, S0, let_vars(T)), - {A, S2} = mapfold(F, S1, let_arg(T)), - {B, S3} = mapfold(F, S2, let_body(T)), - F(update_c_let(T, Vs, A, B), S3); + {Vs, S1} = mapfold_list(Pre, Post, S0, let_vars(T)), + {A, S2} = mapfold(Pre, Post, S1, let_arg(T)), + {B, S3} = mapfold(Pre, Post, S2, let_body(T)), + Post(update_c_let(T, Vs, A, B), S3); seq -> - {A, S1} = mapfold(F, S0, seq_arg(T)), - {B, S2} = mapfold(F, S1, seq_body(T)), - F(update_c_seq(T, A, B), S2); + {A, S1} = mapfold(Pre, Post, S0, seq_arg(T)), + {B, S2} = mapfold(Pre, Post, S1, seq_body(T)), + Post(update_c_seq(T, A, B), S2); apply -> - {E, S1} = mapfold(F, S0, apply_op(T)), - {As, S2} = mapfold_list(F, S1, apply_args(T)), - F(update_c_apply(T, E, As), S2); + {E, S1} = mapfold(Pre, Post, S0, apply_op(T)), + {As, S2} = mapfold_list(Pre, Post, S1, apply_args(T)), + Post(update_c_apply(T, E, As), S2); call -> - {M, S1} = mapfold(F, S0, call_module(T)), - {N, S2} = mapfold(F, S1, call_name(T)), - {As, S3} = mapfold_list(F, S2, call_args(T)), - F(update_c_call(T, M, N, As), S3); + {M, S1} = mapfold(Pre, Post, S0, call_module(T)), + {N, S2} = mapfold(Pre, Post, S1, call_name(T)), + {As, S3} = mapfold_list(Pre, Post, S2, call_args(T)), + Post(update_c_call(T, M, N, As), S3); primop -> - {N, S1} = mapfold(F, S0, primop_name(T)), - {As, S2} = mapfold_list(F, S1, primop_args(T)), - F(update_c_primop(T, N, As), S2); + {N, S1} = mapfold(Pre, Post, S0, primop_name(T)), + {As, S2} = mapfold_list(Pre, Post, S1, primop_args(T)), + Post(update_c_primop(T, N, As), S2); 'case' -> - {A, S1} = mapfold(F, S0, case_arg(T)), - {Cs, S2} = mapfold_list(F, S1, case_clauses(T)), - F(update_c_case(T, A, Cs), S2); + {A, S1} = mapfold(Pre, Post, S0, case_arg(T)), + {Cs, S2} = mapfold_list(Pre, Post, S1, case_clauses(T)), + Post(update_c_case(T, A, Cs), S2); clause -> - {Ps, S1} = mapfold_list(F, S0, clause_pats(T)), - {G, S2} = mapfold(F, S1, clause_guard(T)), - {B, S3} = mapfold(F, S2, clause_body(T)), - F(update_c_clause(T, Ps, G, B), S3); + {Ps, S1} = mapfold_list(Pre, Post, S0, clause_pats(T)), + {G, S2} = mapfold(Pre, Post, S1, clause_guard(T)), + {B, S3} = mapfold(Pre, Post, S2, clause_body(T)), + Post(update_c_clause(T, Ps, G, B), S3); alias -> - {V, S1} = mapfold(F, S0, alias_var(T)), - {P, S2} = mapfold(F, S1, alias_pat(T)), - F(update_c_alias(T, V, P), S2); + {V, S1} = mapfold(Pre, Post, S0, alias_var(T)), + {P, S2} = mapfold(Pre, Post, S1, alias_pat(T)), + Post(update_c_alias(T, V, P), S2); 'fun' -> - {Vs, S1} = mapfold_list(F, S0, fun_vars(T)), - {B, S2} = mapfold(F, S1, fun_body(T)), - F(update_c_fun(T, Vs, B), S2); + {Vs, S1} = mapfold_list(Pre, Post, S0, fun_vars(T)), + {B, S2} = mapfold(Pre, Post, S1, fun_body(T)), + Post(update_c_fun(T, Vs, B), S2); 'receive' -> - {Cs, S1} = mapfold_list(F, S0, receive_clauses(T)), - {E, S2} = mapfold(F, S1, receive_timeout(T)), - {A, S3} = mapfold(F, S2, receive_action(T)), - F(update_c_receive(T, Cs, E, A), S3); + {Cs, S1} = mapfold_list(Pre, Post, S0, receive_clauses(T)), + {E, S2} = mapfold(Pre, Post, S1, receive_timeout(T)), + {A, S3} = mapfold(Pre, Post, S2, receive_action(T)), + Post(update_c_receive(T, Cs, E, A), S3); 'try' -> - {E, S1} = mapfold(F, S0, try_arg(T)), - {Vs, S2} = mapfold_list(F, S1, try_vars(T)), - {B, S3} = mapfold(F, S2, try_body(T)), - {Evs, S4} = mapfold_list(F, S3, try_evars(T)), - {H, S5} = mapfold(F, S4, try_handler(T)), - F(update_c_try(T, E, Vs, B, Evs, H), S5); + {E, S1} = mapfold(Pre, Post, S0, try_arg(T)), + {Vs, S2} = mapfold_list(Pre, Post, S1, try_vars(T)), + {B, S3} = mapfold(Pre, Post, S2, try_body(T)), + {Evs, S4} = mapfold_list(Pre, Post, S3, try_evars(T)), + {H, S5} = mapfold(Pre, Post, S4, try_handler(T)), + Post(update_c_try(T, E, Vs, B, Evs, H), S5); 'catch' -> - {B, S1} = mapfold(F, S0, catch_body(T)), - F(update_c_catch(T, B), S1); + {B, S1} = mapfold(Pre, Post, S0, catch_body(T)), + Post(update_c_catch(T, B), S1); binary -> - {Ds, S1} = mapfold_list(F, S0, binary_segments(T)), - F(update_c_binary(T, Ds), S1); + {Ds, S1} = mapfold_list(Pre, Post, S0, binary_segments(T)), + Post(update_c_binary(T, Ds), S1); bitstr -> - {Val, S1} = mapfold(F, S0, bitstr_val(T)), - {Size, S2} = mapfold(F, S1, bitstr_size(T)), - {Unit, S3} = mapfold(F, S2, bitstr_unit(T)), - {Type, S4} = mapfold(F, S3, bitstr_type(T)), - {Flags, S5} = mapfold(F, S4, bitstr_flags(T)), - F(update_c_bitstr(T, Val, Size, Unit, Type, Flags), S5); + {Val, S1} = mapfold(Pre, Post, S0, bitstr_val(T)), + {Size, S2} = mapfold(Pre, Post, S1, bitstr_size(T)), + {Unit, S3} = mapfold(Pre, Post, S2, bitstr_unit(T)), + {Type, S4} = mapfold(Pre, Post, S3, bitstr_type(T)), + {Flags, S5} = mapfold(Pre, Post, S4, bitstr_flags(T)), + Post(update_c_bitstr(T, Val, Size, Unit, Type, Flags), S5); letrec -> - {Ds, S1} = mapfold_pairs(F, S0, letrec_defs(T)), - {B, S2} = mapfold(F, S1, letrec_body(T)), - F(update_c_letrec(T, Ds, B), S2); + {Ds, S1} = mapfold_pairs(Pre, Post, S0, letrec_defs(T)), + {B, S2} = mapfold(Pre, Post, S1, letrec_body(T)), + Post(update_c_letrec(T, Ds, B), S2); module -> - {N, S1} = mapfold(F, S0, module_name(T)), - {Es, S2} = mapfold_list(F, S1, module_exports(T)), - {As, S3} = mapfold_pairs(F, S2, module_attrs(T)), - {Ds, S4} = mapfold_pairs(F, S3, module_defs(T)), - F(update_c_module(T, N, Es, As, Ds), S4) + {N, S1} = mapfold(Pre, Post, S0, module_name(T)), + {Es, S2} = mapfold_list(Pre, Post, S1, module_exports(T)), + {As, S3} = mapfold_pairs(Pre, Post, S2, module_attrs(T)), + {Ds, S4} = mapfold_pairs(Pre, Post, S3, module_defs(T)), + Post(update_c_module(T, N, Es, As, Ds), S4) end. -mapfold_list(F, S0, [T | Ts]) -> - {T1, S1} = mapfold(F, S0, T), - {Ts1, S2} = mapfold_list(F, S1, Ts), +mapfold_list(Pre, Post, S0, [T | Ts]) -> + {T1, S1} = mapfold(Pre, Post, S0, T), + {Ts1, S2} = mapfold_list(Pre, Post, S1, Ts), {[T1 | Ts1], S2}; -mapfold_list(_, S, []) -> +mapfold_list(_, _, S, []) -> {[], S}. -mapfold_pairs(F, S0, [{T1, T2} | Ps]) -> - {T3, S1} = mapfold(F, S0, T1), - {T4, S2} = mapfold(F, S1, T2), - {Ps1, S3} = mapfold_pairs(F, S2, Ps), +mapfold_pairs(Pre, Post, S0, [{T1, T2} | Ps]) -> + {T3, S1} = mapfold(Pre, Post, S0, T1), + {T4, S2} = mapfold(Pre, Post, S1, T2), + {Ps1, S3} = mapfold_pairs(Pre, Post, S2, Ps), {[{T3, T4} | Ps1], S3}; -mapfold_pairs(_, S, []) -> +mapfold_pairs(_, _, S, []) -> {[], S}. @@ -485,6 +507,7 @@ mapfold_pairs(_, S, []) -> %% well-formed Core Erlang syntax tree. %% %% @see free_variables/1 +%% @see next_free_variable_name/1 -spec variables(cerl:cerl()) -> [cerl:var_name()]. @@ -497,6 +520,7 @@ variables(T) -> %% @doc Like <code>variables/1</code>, but only includes variables %% that are free in the tree. %% +%% @see next_free_variable_name/1 %% @see variables/1 -spec free_variables(cerl:cerl()) -> [cerl:var_name()]. @@ -640,8 +664,8 @@ vars_in_list([], _, A) -> vars_in_defs(Ds, S) -> vars_in_defs(Ds, S, []). -vars_in_defs([{_, F} | Ds], S, A) -> - vars_in_defs(Ds, S, ordsets:union(variables(F, S), A)); +vars_in_defs([{_, Post} | Ds], S, A) -> + vars_in_defs(Ds, S, ordsets:union(variables(Post, S), A)); vars_in_defs([], _, A) -> A. @@ -656,6 +680,110 @@ var_list_names([V | Vs], A) -> var_list_names([], A) -> A. +%% --------------------------------------------------------------------- + +%% @spec next_free_variable_name(Tree::cerl()) -> var_name() +%% +%% var_name() = integer() +%% +%% @doc Returns a integer variable name higher than any other integer +%% variable name in the syntax tree. An exception is thrown if +%% <code>Tree</code> does not represent a well-formed Core Erlang +%% syntax tree. +%% +%% @see variables/1 +%% @see free_variables/1 + +-spec next_free_variable_name(cerl:cerl()) -> integer(). + +next_free_variable_name(T) -> + 1 + next_free(T, -1). + +next_free(T, Max) -> + case type(T) of + literal -> + Max; + var -> + case var_name(T) of + Int when is_integer(Int) -> + max(Int, Max); + _ -> + Max + end; + values -> + next_free_in_list(values_es(T), Max); + cons -> + next_free(cons_hd(T), next_free(cons_tl(T), Max)); + tuple -> + next_free_in_list(tuple_es(T), Max); + map -> + next_free_in_list([map_arg(T)|map_es(T)], Max); + map_pair -> + next_free_in_list([map_pair_op(T),map_pair_key(T), + map_pair_val(T)], Max); + 'let' -> + Max1 = next_free(let_body(T), Max), + Max2 = next_free_in_list(let_vars(T), Max1), + next_free(let_arg(T), Max2); + seq -> + next_free(seq_arg(T), + next_free(seq_body(T), Max)); + apply -> + next_free(apply_op(T), + next_free_in_list(apply_args(T), Max)); + call -> + next_free(call_module(T), + next_free(call_name(T), + next_free_in_list( + call_args(T), Max))); + primop -> + next_free_in_list(primop_args(T), Max); + 'case' -> + next_free(case_arg(T), + next_free_in_list(case_clauses(T), Max)); + clause -> + Max1 = next_free(clause_guard(T), + next_free(clause_body(T), Max)), + next_free_in_list(clause_pats(T), Max1); + alias -> + next_free(alias_var(T), + next_free(alias_pat(T), Max)); + 'fun' -> + next_free(fun_body(T), + next_free_in_list(fun_vars(T), Max)); + 'receive' -> + Max1 = next_free_in_list(receive_clauses(T), + next_free(receive_timeout(T), Max)), + next_free(receive_action(T), Max1); + 'try' -> + Max1 = next_free(try_body(T), Max), + Max2 = next_free_in_list(try_vars(T), Max1), + Max3 = next_free(try_handler(T), Max2), + Max4 = next_free_in_list(try_evars(T), Max3), + next_free(try_arg(T), Max4); + 'catch' -> + next_free(catch_body(T), Max); + binary -> + next_free_in_list(binary_segments(T), Max); + bitstr -> + next_free(bitstr_val(T), next_free(bitstr_size(T), Max)); + letrec -> + Max1 = next_free_in_defs(letrec_defs(T), Max), + Max2 = next_free(letrec_body(T), Max1), + next_free_in_list(letrec_vars(T), Max2); + module -> + next_free_in_defs(module_defs(T), Max) + end. + +next_free_in_list([H | T], Max) -> + next_free_in_list(T, next_free(H, Max)); +next_free_in_list([], Max) -> + Max. + +next_free_in_defs([{_, Post} | Ds], Max) -> + next_free_in_defs(Ds, next_free(Post, Max)); +next_free_in_defs([], Max) -> + Max. %% --------------------------------------------------------------------- @@ -703,13 +831,14 @@ label(T, N, Env) -> %% Constant literals are not labeled. {T, N}; var -> - case dict:find(var_name(T), Env) of - {ok, L} -> - {As, _} = label_ann(T, L), - N1 = N; - error -> - {As, N1} = label_ann(T, N) - end, + {As, N1} = + case dict:find(var_name(T), Env) of + {ok, L} -> + {A, _} = label_ann(T, L), + {A, N}; + error -> + label_ann(T, N) + end, {set_ann(T, As), N1}; values -> {Ts, N1} = label_list(values_es(T), N, Env), @@ -725,10 +854,17 @@ label(T, N, Env) -> {As, N2} = label_ann(T, N1), {ann_c_tuple_skel(As, Ts), N2}; map -> - {M, N1} = label(map_arg(T), N, Env), - {Ts, N2} = label_list(map_es(T), N1, Env), - {As, N3} = label_ann(T, N2), - {ann_c_map(As, M, Ts), N3}; + case is_c_map_pattern(T) of + false -> + {M, N1} = label(map_arg(T), N, Env), + {Ts, N2} = label_list(map_es(T), N1, Env), + {As, N3} = label_ann(T, N2), + {ann_c_map(As, M, Ts), N3}; + true -> + {Ts, N1} = label_list(map_es(T), N, Env), + {As, N2} = label_ann(T, N1), + {ann_c_map_pattern(As, Ts), N2} + end; map_pair -> {Op, N1} = label(map_pair_op(T), N, Env), {Key, N2} = label(map_pair_key(T), N1, Env), diff --git a/lib/compiler/src/compile.erl b/lib/compiler/src/compile.erl index e0a29fe9b1..6510571441 100644 --- a/lib/compiler/src/compile.erl +++ b/lib/compiler/src/compile.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1996-2015. All Rights Reserved. +%% Copyright Ericsson AB 1996-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -26,6 +26,7 @@ -export([forms/1,forms/2,noenv_forms/2]). -export([output_generated/1,noenv_output_generated/1]). -export([options/0]). +-export([env_compiler_options/0]). %% Erlc interface. -export([compile/3,compile_beam/3,compile_asm/3,compile_core/3]). @@ -40,6 +41,12 @@ %%---------------------------------------------------------------------- +-type abstract_code() :: [erl_parse:abstract_form()]. + +%% Internal representations used for 'from_asm' and 'from_beam' compilation can +%% also be valid, but have no relevant types defined. +-type forms() :: abstract_code() | cerl:c_module(). + -type option() :: atom() | {atom(), term()} | {'d', atom(), term()}. -type err_info() :: {erl_anno:line() | 'none', @@ -48,6 +55,9 @@ -type warnings() :: [{file:filename(), [err_info()]}]. -type mod_ret() :: {'ok', module()} | {'ok', module(), cerl:c_module()} %% with option 'to_core' + | {'ok', %% with option 'to_pp' + module() | [], %% module() if 'to_exp' + abstract_code()} | {'ok', module(), warnings()}. -type bin_ret() :: {'ok', module(), binary()} | {'ok', module(), binary(), warnings()}. @@ -78,7 +88,11 @@ file(File, Opts) when is_list(Opts) -> file(File, Opt) -> file(File, [Opt|?DEFAULT_OPTIONS]). -forms(File) -> forms(File, ?DEFAULT_OPTIONS). +-spec forms(abstract_code()) -> comp_ret(). + +forms(Forms) -> forms(Forms, ?DEFAULT_OPTIONS). + +-spec forms(forms(), [option()] | option()) -> comp_ret(). forms(Forms, Opts) when is_list(Opts) -> do_compile({forms,Forms}, [binary|Opts++env_default_opts()]); @@ -106,6 +120,8 @@ noenv_file(File, Opts) when is_list(Opts) -> noenv_file(File, Opt) -> noenv_file(File, [Opt|?DEFAULT_OPTIONS]). +-spec noenv_forms(forms(), [option()] | option()) -> comp_ret(). + noenv_forms(Forms, Opts) when is_list(Opts) -> do_compile({forms,Forms}, [binary|Opts]); noenv_forms(Forms, Opt) when is_atom(Opt) -> @@ -120,11 +136,19 @@ noenv_output_generated(Opts) -> end, Passes). %% +%% Retrieve ERL_COMPILER_OPTIONS as a list of terms +%% + +-spec env_compiler_options() -> [term()]. + +env_compiler_options() -> env_default_opts(). + +%% %% Local functions %% --define(pass(P), {P,fun P/1}). --define(pass(P,T), {P,fun T/1,fun P/1}). +-define(pass(P), {P,fun P/2}). +-define(pass(P,T), {P,fun T/1,fun P/2}). env_default_opts() -> Key = "ERL_COMPILER_OPTIONS", @@ -149,29 +173,42 @@ env_default_opts() -> do_compile(Input, Opts0) -> Opts = expand_opts(Opts0), - {Pid,Ref} = - spawn_monitor(fun() -> - exit(try - internal(Input, Opts) - catch - error:Reason -> - {error,Reason} - end) - end), - receive - {'DOWN',Ref,process,Pid,Rep} -> Rep + IntFun = fun() -> try + internal(Input, Opts) + catch + error:Reason -> + {error,Reason} + end + end, + %% Dialyzer has already spawned workers. + case lists:member(dialyzer, Opts) of + true -> + IntFun(); + false -> + {Pid,Ref} = + spawn_monitor(fun() -> + exit(IntFun()) + end), + receive + {'DOWN',Ref,process,Pid,Rep} -> Rep + end end. expand_opts(Opts0) -> %% {debug_info_key,Key} implies debug_info. Opts = case {proplists:get_value(debug_info_key, Opts0), proplists:get_value(encrypt_debug_info, Opts0), - proplists:get_bool(debug_info, Opts0)} of + proplists:get_value(debug_info, Opts0)} of {undefined,undefined,_} -> Opts0; - {_,_,false} -> [debug_info|Opts0]; + {_,_,undefined} -> [debug_info|Opts0]; {_,_,_} -> Opts0 end, - foldr(fun expand_opt/2, [], Opts). + %% iff,unless processing is to complex... + Opts1 = case proplists:is_defined(makedep_side_effect,Opts) of + true -> proplists:delete(makedep,Opts); + false -> Opts + end, + foldr(fun expand_opt/2, [], Opts1). expand_opt(basic_validation, Os) -> [no_code_generation,to_pp,binary|Os]; @@ -181,12 +218,16 @@ expand_opt(report, Os) -> [report_errors,report_warnings|Os]; expand_opt(return, Os) -> [return_errors,return_warnings|Os]; -expand_opt(r12, Os) -> - [no_recv_opt,no_line_info|Os]; -expand_opt(r13, Os) -> - [no_recv_opt,no_line_info|Os]; -expand_opt(r14, Os) -> - [no_line_info|Os]; +expand_opt(r16, Os) -> + [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; +expand_opt(r17, Os) -> + [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; +expand_opt(r18, Os) -> + [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; +expand_opt(r19, Os) -> + [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; +expand_opt(r20, Os) -> + [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; expand_opt({debug_info_key,_}=O, Os) -> [encrypt_debug_info,O|Os]; expand_opt(no_float_opt, Os) -> @@ -196,6 +237,8 @@ expand_opt(O, Os) -> [O|Os]. %% format_error(ErrorDescriptor) -> string() +-spec format_error(term()) -> iolist(). + format_error(no_native_support) -> "this system is not configured for native-code compilation."; format_error(no_crypto) -> @@ -216,6 +259,8 @@ format_error({epp,E}) -> epp:format_error(E); format_error(write_error) -> "error writing file"; +format_error({write_error, Error}) -> + io_lib:format("error writing file: ~ts", [file:format_error(Error)]); format_error({rename,From,To,Error}) -> io_lib:format("failed to rename ~ts to ~ts: ~ts", [From,To,file:format_error(Error)]); @@ -226,9 +271,9 @@ format_error({delete_temp,File,Error}) -> io_lib:format("failed to delete temporary file ~ts: ~ts", [File,file:format_error(Error)]); format_error({parse_transform,M,R}) -> - io_lib:format("error in parse transform '~s': ~tp", [M, R]); + io_lib:format("error in parse transform '~ts': ~tp", [M, R]); format_error({undef_parse_transform,M}) -> - io_lib:format("undefined parse transform '~s'", [M]); + io_lib:format("undefined parse transform '~ts'", [M]); format_error({core_transform,M,R}) -> io_lib:format("error in core transform '~s': ~tp", [M, R]); format_error({crash,Pass,Reason}) -> @@ -250,38 +295,45 @@ format_error_reason({Reason, Stack}) when is_list(Stack) -> end, FormatFun = fun (Term, _) -> io_lib:format("~tp", [Term]) end, [io_lib:format("~tp", [Reason]),"\n\n", - lib:format_stacktrace(1, Stack, StackFun, FormatFun)]; + erl_error:format_stacktrace(1, Stack, StackFun, FormatFun)]; format_error_reason(Reason) -> io_lib:format("~tp", [Reason]). +-type err_warn_info() :: tuple(). + %% The compile state record. -record(compile, {filename="" :: file:filename(), dir="" :: file:filename(), base="" :: file:filename(), ifile="" :: file:filename(), ofile="" :: file:filename(), - module=[], - code=[], - core_code=[], - abstract_code=[], %Abstract code for debugger. - options=[] :: [option()], %Options for compilation + module=[] :: module() | [], + core_code=[] :: cerl:c_module() | [], + abstract_code=[] :: abstract_code(), %Abstract code for debugger. + options=[] :: [option()], %Options for compilation mod_options=[] :: [option()], %Options for module_info encoding=none :: none | epp:source_encoding(), - errors=[], - warnings=[]}). + errors=[] :: [err_warn_info()], + warnings=[] :: [err_warn_info()], + extra_chunks=[] :: [{binary(), binary()}]}). internal({forms,Forms}, Opts0) -> {_,Ps} = passes(forms, Opts0), Source = proplists:get_value(source, Opts0, ""), Opts1 = proplists:delete(source, Opts0), - Compile = #compile{code=Forms,options=Opts1,mod_options=Opts1}, - internal_comp(Ps, Source, "", Compile); + Compile = build_compile(Opts1), + internal_comp(Ps, Forms, Source, "", Compile); internal({file,File}, Opts) -> {Ext,Ps} = passes(file, Opts), - Compile = #compile{options=Opts,mod_options=Opts}, - internal_comp(Ps, File, Ext, Compile). + Compile = build_compile(Opts), + internal_comp(Ps, none, File, Ext, Compile). -internal_comp(Passes, File, Suffix, St0) -> +build_compile(Opts0) -> + ExtraChunks = proplists:get_value(extra_chunks, Opts0, []), + Opts1 = proplists:delete(extra_chunks, Opts0), + #compile{options=Opts1,mod_options=Opts1,extra_chunks=ExtraChunks}. + +internal_comp(Passes, Code0, File, Suffix, St0) -> Dir = filename:dirname(File), Base = filename:basename(File, Suffix), St1 = St0#compile{filename=File, dir=Dir, base=Base, @@ -291,36 +343,41 @@ internal_comp(Passes, File, Suffix, St0) -> Run0 = case member(time, Opts) of true -> io:format("Compiling ~tp\n", [File]), - fun run_tc/2; - false -> fun({_Name,Fun}, St) -> catch Fun(St) end + fun run_tc/3; + false -> + fun({_Name,Fun}, Code, St) -> + catch Fun(Code, St) + end end, Run = case keyfind(eprof, 1, Opts) of {eprof,EprofPass} -> - fun(P, St) -> - run_eprof(P, EprofPass, St) + fun(P, Code, St) -> + run_eprof(P, Code, EprofPass, St) end; false -> Run0 end, - case fold_comp(Passes, Run, St1) of - {ok,St2} -> comp_ret_ok(St2); + case fold_comp(Passes, Run, Code0, St1) of + {ok,Code,St2} -> comp_ret_ok(Code, St2); {error,St2} -> comp_ret_err(St2) end. -fold_comp([{delay,Ps0}|Passes], Run, #compile{options=Opts}=St) -> +fold_comp([{delay,Ps0}|Passes], Run, Code, #compile{options=Opts}=St) -> Ps = select_passes(Ps0, Opts) ++ Passes, - fold_comp(Ps, Run, St); -fold_comp([{Name,Test,Pass}|Ps], Run, St) -> + fold_comp(Ps, Run, Code, St); +fold_comp([{Name,Test,Pass}|Ps], Run, Code, St) -> case Test(St) of false -> %Pass is not needed. - fold_comp(Ps, Run, St); + fold_comp(Ps, Run, Code, St); true -> %Run pass in the usual way. - fold_comp([{Name,Pass}|Ps], Run, St) + fold_comp([{Name,Pass}|Ps], Run, Code, St) end; -fold_comp([{Name,Pass}|Ps], Run, St0) -> - case Run({Name,Pass}, St0) of - {ok,St1} -> fold_comp(Ps, Run, St1); - {error,_St1} = Error -> Error; +fold_comp([{Name,Pass}|Ps], Run, Code0, St0) -> + case Run({Name,Pass}, Code0, St0) of + {ok,Code,St1} -> + fold_comp(Ps, Run, Code, St1); + {error,_St1}=Error -> + Error; {'EXIT',Reason} -> Es = [{St0#compile.ifile,[{none,?MODULE,{crash,Name,Reason}}]}], {error,St0#compile{errors=St0#compile.errors ++ Es}}; @@ -328,30 +385,30 @@ fold_comp([{Name,Pass}|Ps], Run, St0) -> Es = [{St0#compile.ifile,[{none,?MODULE,{bad_return,Name,Other}}]}], {error,St0#compile{errors=St0#compile.errors ++ Es}} end; -fold_comp([], _Run, St) -> {ok,St}. +fold_comp([], _Run, Code, St) -> {ok,Code,St}. -run_tc({Name,Fun}, St) -> +run_tc({Name,Fun}, Code, St) -> T1 = erlang:monotonic_time(), - Val = (catch Fun(St)), + Val = (catch Fun(Code, St)), T2 = erlang:monotonic_time(), - Elapsed = erlang:convert_time_unit(T2 - T1, native, milli_seconds), + Elapsed = erlang:convert_time_unit(T2 - T1, native, millisecond), Mem0 = erts_debug:flat_size(Val)*erlang:system_info(wordsize), Mem = lists:flatten(io_lib:format("~.1f kB", [Mem0/1024])), io:format(" ~-30s: ~10.3f s ~12s\n", [Name,Elapsed/1000,Mem]), Val. -run_eprof({Name,Fun}, Name, St) -> +run_eprof({Name,Fun}, Code, Name, St) -> io:format("~p: Running eprof\n", [Name]), c:appcall(tools, eprof, start_profiling, [[self()]]), - Val = (catch Fun(St)), + Val = (catch Fun(Code, St)), c:appcall(tools, eprof, stop_profiling, []), c:appcall(tools, eprof, analyze, []), Val; -run_eprof({_,Fun}, _, St) -> - catch Fun(St). +run_eprof({_,Fun}, Code, _, St) -> + catch Fun(Code, St). -comp_ret_ok(#compile{code=Code,warnings=Warn0,module=Mod,options=Opts}=St) -> +comp_ret_ok(Code, #compile{warnings=Warn0,module=Mod,options=Opts}=St) -> case werror(St) of true -> case member(report_warnings, Opts) of @@ -417,8 +474,10 @@ mpf(Ms) -> passes(Type, Opts) -> {Ext,Passes0} = passes_1(Opts), Passes1 = case Type of - file -> Passes0; - forms -> tl(Passes0) + file -> + Passes0; + forms -> + fix_first_pass(Passes0) end, Passes = select_passes(Passes1, Opts), @@ -455,6 +514,22 @@ pass(from_beam) -> {".beam",[?pass(read_beam_file)|binary_passes()]}; pass(_) -> none. +%% For compilation from forms, replace the first pass with a pass +%% that retrieves the module name. The module name is needed for +%% proper diagnostics and for compilation to native code. + +fix_first_pass([{parse_core,_}|Passes]) -> + [?pass(get_module_name_from_core)|Passes]; +fix_first_pass([{beam_consult_asm,_}|Passes]) -> + [?pass(get_module_name_from_asm)|Passes]; +fix_first_pass([{read_beam_file,_}|Passes]) -> + [?pass(get_module_name_from_beam)|Passes]; +fix_first_pass([_|Passes]) -> + %% When compiling from abstract code, the module name + %% will be set after running the v3_core pass. + Passes. + + %% select_passes([Command], Opts) -> [{Name,Function}] %% Interpret the lists of commands to return a pure list of passes. %% @@ -506,21 +581,21 @@ pass(_) -> none. %% select_passes([{pass,Mod}|Ps], Opts) -> - F = fun(St) -> - case catch Mod:module(St#compile.code, St#compile.options) of + F = fun(Code0, St) -> + case catch Mod:module(Code0, St#compile.options) of {ok,Code} -> - {ok,St#compile{code=Code}}; + {ok,Code,St}; {ok,Code,Ws} -> - {ok,St#compile{code=Code,warnings=St#compile.warnings++Ws}}; + {ok,Code,St#compile{warnings=St#compile.warnings++Ws}}; {error,Es} -> {error,St#compile{errors=St#compile.errors ++ Es}} end end, [{Mod,F}|select_passes(Ps, Opts)]; select_passes([{src_listing,Ext}|_], _Opts) -> - [{listing,fun (St) -> src_listing(Ext, St) end}]; + [{listing,fun (Code, St) -> src_listing(Ext, Code, St) end}]; select_passes([{listing,Ext}|_], _Opts) -> - [{listing,fun (St) -> listing(Ext, St) end}]; + [{listing,fun (Code, St) -> listing(Ext, Code, St) end}]; select_passes([done|_], _Opts) -> []; select_passes([{done,Ext}|_], Opts) -> @@ -606,6 +681,7 @@ select_list_passes_1([], _, Acc) -> standard_passes() -> [?pass(transform_module), + {iff,makedep_side_effect,?pass(makedep_and_output)}, {iff,makedep,[ ?pass(makedep), {unless,binary,?pass(makedep_output)} @@ -620,13 +696,13 @@ standard_passes() -> {iff,'dabstr',{listing,"abstr"}}, {iff,debug_info,?pass(save_abstract_code)}, - ?pass(expand_module), + ?pass(expand_records), {iff,'dexp',{listing,"expand"}}, {iff,'E',{src_listing,"E"}}, {iff,'to_exp',{done,"E"}}, %% Conversion to Core Erlang. - {pass,v3_core}, + ?pass(core), {iff,'dcore',{listing,"core"}}, {iff,'to_core0',{done,"core"}} | core_passes()]. @@ -636,33 +712,35 @@ core_passes() -> [{iff,clint0,?pass(core_lint_module)}, {delay, [{unless,no_copt, - [{core_old_inliner,fun test_old_inliner/1,fun core_old_inliner/1}, + [{core_old_inliner,fun test_old_inliner/1,fun core_old_inliner/2}, {iff,doldinline,{listing,"oldinline"}}, - {pass,sys_core_fold}, + {unless,no_fold,{pass,sys_core_fold}}, {iff,dcorefold,{listing,"corefold"}}, - {core_inline_module,fun test_core_inliner/1,fun core_inline_module/1}, + {core_inline_module,fun test_core_inliner/1,fun core_inline_module/2}, {iff,dinline,{listing,"inline"}}, {core_fold_after_inlining,fun test_any_inliner/1, - fun core_fold_module_after_inlining/1}, + fun core_fold_module_after_inlining/2}, + {iff,dcopt,{listing,"copt"}}, + {unless,no_alias,{pass,sys_core_alias}}, + {iff,dalias,{listing,"core_alias"}}, ?pass(core_transforms)]}, - {iff,dcopt,{listing,"copt"}}, {iff,'to_core',{done,"core"}}]} | kernel_passes()]. kernel_passes() -> - %% Destructive setelement/3 optimization and core lint. - [{pass,sys_core_dsetel}, + %% Optimizations that must be done after all other optimizations. + [{pass,sys_core_bsm}, + {iff,dcbsm,{listing,"core_bsm"}}, + {pass,sys_core_dsetel}, {iff,dsetel,{listing,"dsetel"}}, {iff,clint,?pass(core_lint_module)}, {iff,core,?pass(save_core_code)}, %% Kernel Erlang and code generation. - {pass,v3_kernel}, + ?pass(v3_kernel), {iff,dkern,{listing,"kernel"}}, {iff,'to_kernel',{done,"kernel"}}, - {pass,v3_life}, - {iff,dlife,{listing,"life"}}, {pass,v3_codegen}, {iff,dcg,{listing,"codegen"}} | asm_passes()]. @@ -671,13 +749,16 @@ asm_passes() -> %% Assembly level optimisations. [{delay, [{pass,beam_a}, + {iff,da,{listing,"a"}}, {unless,no_postopt, - [{pass,beam_block}, + [{unless,no_reorder,{pass,beam_reorder}}, + {iff,dre,{listing,"reorder"}}, + {pass,beam_block}, {iff,dblk,{listing,"block"}}, {unless,no_except,{pass,beam_except}}, {iff,dexcept,{listing,"except"}}, - {unless,no_bopt,{pass,beam_bool}}, - {iff,dbool,{listing,"bool"}}, + {unless,no_bs_opt,{pass,beam_bs}}, + {iff,dbs,{listing,"bs"}}, {unless,no_topt,{pass,beam_type}}, {iff,dtype,{listing,"type"}}, {pass,beam_split}, @@ -694,6 +775,10 @@ asm_passes() -> {iff,dbsm,{listing,"bsm"}}, {unless,no_recv_opt,{pass,beam_receive}}, {iff,drecv,{listing,"recv"}}, + {unless,no_record_opt,{pass,beam_record}}, + {iff,drecord,{listing,"record"}}, + {unless,no_blk2,?pass(block2)}, + {iff,dblk2,{listing,"block2"}}, {unless,no_stack_trimming,{pass,beam_trim}}, {iff,dtrim,{listing,"trim"}}, {pass,beam_flatten}]}, @@ -703,6 +788,7 @@ asm_passes() -> {iff,no_postopt,[{pass,beam_clean}]}, {pass,beam_z}, + {iff,dz,{listing,"z"}}, {iff,dopt,{listing,"optimize"}}, {iff,'S',{listing,"S"}}, {iff,'to_asm',{done,"S"}}]}, @@ -711,17 +797,19 @@ asm_passes() -> | binary_passes()]. binary_passes() -> - [{native_compile,fun test_native/1,fun native_compile/1}, - {unless,binary,?pass(save_binary,not_werror)}]. + [{iff,'to_dis',?pass(to_dis)}, + {native_compile,fun test_native/1,fun native_compile/2}, + {unless,binary,?pass(save_binary,not_werror)} + ]. %%% %%% Compiler passes. %%% %% Remove the target file so we don't have an old one if the compilation fail. -remove_file(St) -> +remove_file(Code, St) -> _ = file:delete(St#compile.ofile), - {ok,St}. + {ok,Code,St}. -record(asm_module, {module, exports, @@ -769,34 +857,50 @@ collect_asm([{attributes, Attr} | Rest], R) -> collect_asm([X | Rest], R) -> collect_asm(Rest, R#asm_module{code=R#asm_module.code++[X]}). -beam_consult_asm(St) -> +beam_consult_asm(_Code, St) -> case file:consult(St#compile.ifile) of - {ok, Forms0} -> + {ok,Forms0} -> Encoding = epp:read_encoding(St#compile.ifile), - {Module, Forms} = preprocess_asm_forms(Forms0), - {ok,St#compile{module=Module, code=Forms, encoding=Encoding}}; + {Module,Forms} = preprocess_asm_forms(Forms0), + {ok,Forms,St#compile{module=Module,encoding=Encoding}}; {error,E} -> Es = [{St#compile.ifile,[{none,?MODULE,{open,E}}]}], {error,St#compile{errors=St#compile.errors ++ Es}} end. -read_beam_file(St) -> +get_module_name_from_asm({Mod,_,_,_,_}=Asm, St) -> + {ok,Asm,St#compile{module=Mod}}; +get_module_name_from_asm(Asm, St) -> + %% Invalid Beam assembly code. Let it crash in a later pass. + {ok,Asm,St}. + +read_beam_file(_Code, St) -> case file:read_file(St#compile.ifile) of {ok,Beam} -> Infile = St#compile.ifile, case no_native_compilation(Infile, St) of true -> - {ok,St#compile{module=none,code=none}}; + {ok,none,St#compile{module=none}}; false -> Mod0 = filename:rootname(filename:basename(Infile)), Mod = list_to_atom(Mod0), - {ok,St#compile{module=Mod,code=Beam,ofile=Infile}} + {ok,Beam,St#compile{module=Mod,ofile=Infile}} end; {error,E} -> Es = [{St#compile.ifile,[{none,?MODULE,{open,E}}]}], {error,St#compile{errors=St#compile.errors ++ Es}} end. +get_module_name_from_beam(Beam, St) -> + case beam_lib:info(Beam) of + {error,beam_lib,Error} -> + Es = [{"((forms))",[{none,beam_lib,Error}]}], + {error,St#compile{errors=St#compile.errors ++ Es}}; + Info -> + {module,Mod} = keyfind(module, 1, Info), + {ok,Beam,St#compile{module=Mod}} + end. + no_native_compilation(BeamFile, #compile{options=Opts0}) -> case beam_lib:chunks(BeamFile, ["CInf"]) of {ok,{_,[{"CInf",Term0}]}} -> @@ -809,17 +913,17 @@ no_native_compilation(BeamFile, #compile{options=Opts0}) -> _ -> false end. -parse_module(St0) -> +parse_module(_Code, St0) -> case do_parse_module(utf8, St0) of - {ok,_}=Ret -> + {ok,_,_}=Ret -> Ret; {error,_}=Ret -> Ret; {invalid_unicode,File,Line} -> case do_parse_module(latin1, St0) of - {ok,St} -> + {ok,Code,St} -> Es = [{File,[{Line,?MODULE,reparsing_invalid_unicode}]}], - {ok,St#compile{warnings=Es++St#compile.warnings}}; + {ok,Code,St#compile{warnings=Es++St#compile.warnings}}; {error,St} -> Es = [{File,[{Line,?MODULE,reparsing_invalid_unicode}]}], {error,St#compile{errors=Es++St#compile.errors}} @@ -827,23 +931,29 @@ parse_module(St0) -> end. do_parse_module(DefEncoding, #compile{ifile=File,options=Opts,dir=Dir}=St) -> + SourceName0 = proplists:get_value(source, Opts, File), + SourceName = case member(deterministic, Opts) of + true -> filename:basename(SourceName0); + false -> SourceName0 + end, R = epp:parse_file(File, - [{includes,[".",Dir|inc_paths(Opts)]}, - {macros,pre_defs(Opts)}, - {default_encoding,DefEncoding}, - extra]), + [{includes,[".",Dir|inc_paths(Opts)]}, + {source_name, SourceName}, + {macros,pre_defs(Opts)}, + {default_encoding,DefEncoding}, + extra]), case R of {ok,Forms,Extra} -> Encoding = proplists:get_value(encoding, Extra), case find_invalid_unicode(Forms, File) of none -> - {ok,St#compile{code=Forms,encoding=Encoding}}; + {ok,Forms,St#compile{encoding=Encoding}}; {invalid_unicode,_,_}=Ret -> case Encoding of none -> Ret; _ -> - {ok,St#compile{code=Forms,encoding=Encoding}} + {ok,Forms,St#compile{encoding=Encoding}} end end; {error,E} -> @@ -862,7 +972,7 @@ find_invalid_unicode([H|T], File0) -> end; find_invalid_unicode([], _) -> none. -parse_core(St) -> +parse_core(_Code, St) -> case file:read_file(St#compile.ifile) of {ok,Bin} -> case core_scan:string(binary_to_list(Bin)) of @@ -870,7 +980,7 @@ parse_core(St) -> case core_parse:parse(Toks) of {ok,Mod} -> Name = (Mod#c_module.name)#c_literal.val, - {ok,St#compile{module=Name,code=Mod}}; + {ok,Mod,St#compile{module=Name}}; {error,E} -> Es = [{St#compile.ifile,[E]}], {error,St#compile{errors=St#compile.errors ++ Es}} @@ -884,6 +994,16 @@ parse_core(St) -> {error,St#compile{errors=St#compile.errors ++ Es}} end. +get_module_name_from_core(Core, St) -> + try + Mod = cerl:concrete(cerl:module_name(Core)), + {ok,Core,St#compile{module=Mod}} + catch + _:_ -> + %% Invalid Core Erlang code. Let it crash in a later pass. + {ok,Core,St} + end. + compile_options([{attribute,_L,compile,C}|Fs]) when is_list(C) -> C ++ compile_options(Fs); compile_options([{attribute,_L,compile,C}|Fs]) -> @@ -907,31 +1027,36 @@ clean_parse_transforms_1([], Acc) -> reverse(Acc). transforms(Os) -> [ M || {parse_transform,M} <- Os ]. -transform_module(#compile{options=Opt,code=Code0}=St0) -> +transform_module(Code0, #compile{options=Opt}=St) -> %% Extract compile options from code into options field. case transforms(Opt ++ compile_options(Code0)) of - [] -> {ok,St0}; %No parse transforms. + [] -> + %% No parse transforms. + {ok,Code0,St}; Ts -> %% Remove parse_transform attributes from the abstract code to %% prevent parse transforms to be run more than once. Code = clean_parse_transforms(Code0), - St = St0#compile{code=Code}, - foldl_transform(St, Ts) + foldl_transform(Ts, Code, St) end. -foldl_transform(St, [T|Ts]) -> +foldl_transform([T|Ts], Code0, St) -> Name = "transform " ++ atom_to_list(T), case code:ensure_loaded(T) =:= {module,T} andalso - erlang:function_exported(T, parse_transform, 2) of + erlang:function_exported(T, parse_transform, 2) of true -> - Fun = fun(S) -> - T:parse_transform(S#compile.code, S#compile.options) + Fun = fun(Code, S) -> + T:parse_transform(Code, S#compile.options) end, Run = case member(time, St#compile.options) of - true -> fun run_tc/2; - false -> fun({_Name,F}, S) -> catch F(S) end + true -> + fun run_tc/3; + false -> + fun({_Name,F}, Code, S) -> + catch F(Code, S) + end end, - case Run({Name, Fun}, St) of + case Run({Name, Fun}, Code0, St) of {error,Es,Ws} -> {error,St#compile{warnings=St#compile.warnings ++ Ws, errors=St#compile.errors ++ Es}}; @@ -940,41 +1065,44 @@ foldl_transform(St, [T|Ts]) -> {parse_transform,T,R}}]}], {error,St#compile{errors=St#compile.errors ++ Es}}; {warning, Forms, Ws} -> - foldl_transform( - St#compile{code=Forms, - warnings=St#compile.warnings ++ Ws}, Ts); + foldl_transform(Ts, Forms, + St#compile{warnings=St#compile.warnings ++ Ws}); Forms -> - foldl_transform(St#compile{code=Forms}, Ts) + foldl_transform(Ts, Forms, St) end; false -> Es = [{St#compile.ifile,[{none,compile, {undef_parse_transform,T}}]}], {error,St#compile{errors=St#compile.errors ++ Es}} end; -foldl_transform(St, []) -> {ok,St}. +foldl_transform([], Code, St) -> {ok,Code,St}. get_core_transforms(Opts) -> [M || {core_transform,M} <- Opts]. -core_transforms(St) -> +core_transforms(Code, St) -> %% The options field holds the complete list of options at this Ts = get_core_transforms(St#compile.options), - foldl_core_transforms(St, Ts). + foldl_core_transforms(Ts, Code, St). -foldl_core_transforms(St, [T|Ts]) -> +foldl_core_transforms([T|Ts], Code0, St) -> Name = "core transform " ++ atom_to_list(T), - Fun = fun(S) -> T:core_transform(S#compile.code, S#compile.options) end, + Fun = fun(Code, S) -> T:core_transform(Code, S#compile.options) end, Run = case member(time, St#compile.options) of - true -> fun run_tc/2; - false -> fun({_Name,F}, S) -> catch F(S) end + true -> + fun run_tc/3; + false -> + fun({_Name,F}, Code, S) -> + catch F(Code, S) + end end, - case Run({Name, Fun}, St) of + case Run({Name, Fun}, Code0, St) of {'EXIT',R} -> Es = [{St#compile.ifile,[{none,compile,{core_transform,T,R}}]}], {error,St#compile{errors=St#compile.errors ++ Es}}; Forms -> - foldl_core_transforms(St#compile{code=Forms}, Ts) + foldl_core_transforms(Ts, Forms, St) end; -foldl_core_transforms(St, []) -> {ok,St}. +foldl_core_transforms([], Code, St) -> {ok,Code,St}. %%% Fetches the module name from a list of forms. The module attribute must %%% be present. @@ -995,31 +1123,38 @@ add_default_base(St, Forms) -> St end. -lint_module(St) -> - case erl_lint:module(St#compile.code, - St#compile.ifile, St#compile.options) of +lint_module(Code, St) -> + case erl_lint:module(Code, St#compile.ifile, St#compile.options) of {ok,Ws} -> %% Insert name of module as base name, if needed. This is %% for compile:forms to work with listing files. - St1 = add_default_base(St, St#compile.code), - {ok,St1#compile{warnings=St1#compile.warnings ++ Ws}}; + St1 = add_default_base(St, Code), + {ok,Code,St1#compile{warnings=St1#compile.warnings ++ Ws}}; {error,Es,Ws} -> {error,St#compile{warnings=St#compile.warnings ++ Ws, errors=St#compile.errors ++ Es}} end. -core_lint_module(St) -> - case core_lint:module(St#compile.code, St#compile.options) of +core_lint_module(Code, St) -> + case core_lint:module(Code, St#compile.options) of {ok,Ws} -> - {ok,St#compile{warnings=St#compile.warnings ++ Ws}}; + {ok,Code,St#compile{warnings=St#compile.warnings ++ Ws}}; {error,Es,Ws} -> {error,St#compile{warnings=St#compile.warnings ++ Ws, errors=St#compile.errors ++ Es}} end. -makedep(#compile{code=Code,options=Opts}=St) -> - Ifile = St#compile.ifile, - Ofile = St#compile.ofile, +%% makedep + output and continue +makedep_and_output(Code0, St) -> + {ok,DepCode,St1} = makedep(Code0,St), + case makedep_output(DepCode, St1) of + {ok,_IgnoreCode,St2} -> + {ok,Code0,St2}; + {error,St2} -> + {error,St2} + end. + +makedep(Code0, #compile{ifile=Ifile,ofile=Ofile,options=Opts}=St) -> %% Get the target of the Makefile rule. Target0 = @@ -1051,7 +1186,7 @@ makedep(#compile{code=Code,options=Opts}=St) -> %% List the dependencies (includes) for this target. {MainRule,PhonyRules} = makedep_add_headers( Ifile, % The input file name. - Code, % The parsed source. + Code0, % The parsed source. [], % The list of dependencies already added. length(Target), % The current line length. Target, % The target. @@ -1071,7 +1206,8 @@ makedep(#compile{code=Code,options=Opts}=St) -> true -> MainRule ++ PhonyRules; _ -> MainRule end, - {ok,St#compile{code=iolist_to_binary([Makefile,"\n"])}}. + Code = iolist_to_binary([Makefile,"\n"]), + {ok,Code,St}. makedep_add_headers(Ifile, [{attribute,_,file,{File,_}}|Rest], Included, LineLen, MainTarget, Phony, Opts) -> @@ -1136,7 +1272,7 @@ makedep_add_header(Ifile, Included, LineLen, MainTarget, Phony, File) -> end end. -makedep_output(#compile{code=Code,options=Opts,ofile=Ofile}=St) -> +makedep_output(Code, #compile{options=Opts,ofile=Ofile}=St) -> %% Write this Makefile (Code) to the selected output. %% If no output is specified, the default is to write to a file named after %% the output file. @@ -1178,9 +1314,9 @@ makedep_output(#compile{code=Code,options=Opts,ofile=Ofile}=St) -> CloseOutput -> ok = file:close(Output1); true -> ok end, - {ok,St} + {ok,Code,St} catch - exit:_ -> + error:_ -> %% Couldn't write to output Makefile. Err = {St#compile.ifile,[{none,?MODULE,write_error}]}, {error,St#compile{errors=St#compile.errors++[Err]}} @@ -1195,19 +1331,38 @@ makedep_output(#compile{code=Code,options=Opts,ofile=Ofile}=St) -> {error,St#compile{errors=St#compile.errors++[Err]}} end. -%% expand_module(State) -> State' -%% Do the common preprocessing of the input forms. +expand_records(Code0, #compile{options=Opts}=St) -> + Code = erl_expand_records:module(Code0, Opts), + {ok,Code,St}. -expand_module(#compile{code=Code,options=Opts0}=St0) -> - {Mod,Exp,Forms,Opts1} = sys_pre_expand:module(Code, Opts0), +core(Forms, #compile{options=Opts0}=St) -> + Opts1 = lists:flatten([C || {attribute,_,compile,C} <- Forms] ++ Opts0), Opts = expand_opts(Opts1), - {ok,St0#compile{module=Mod,options=Opts,code={Mod,Exp,Forms}}}. + {ok,Core,Ws} = v3_core:module(Forms, Opts), + Mod = cerl:concrete(cerl:module_name(Core)), + {ok,Core,St#compile{module=Mod,options=Opts, + warnings=St#compile.warnings++Ws}}. -core_fold_module_after_inlining(#compile{code=Code0,options=Opts}=St) -> +core_fold_module_after_inlining(Code0, #compile{options=Opts}=St) -> %% Inlining may produce code that generates spurious warnings. %% Ignore all warnings. {ok,Code,_Ws} = sys_core_fold:module(Code0, Opts), - {ok,St#compile{code=Code}}. + {ok,Code,St}. + +v3_kernel(Code0, #compile{options=Opts,warnings=Ws0}=St) -> + {ok,Code,Ws} = v3_kernel:module(Code0, Opts), + case Ws =:= [] orelse test_core_inliner(St) of + false -> + {ok,Code,St#compile{warnings=Ws0++Ws}}; + true -> + %% cerl_inline may produce code that generates spurious + %% warnings. Ignore any such warnings. + {ok,Code,St} + end. + +block2(Code0, #compile{options=Opts}=St) -> + {ok,Code} = beam_block:module(Code0, [no_blockify|Opts]), + {ok,Code,St}. test_old_inliner(#compile{options=Opts}) -> %% The point of this test is to avoid loading the old inliner @@ -1230,52 +1385,51 @@ test_core_inliner(#compile{options=Opts}) -> test_any_inliner(St) -> test_old_inliner(St) orelse test_core_inliner(St). -core_old_inliner(#compile{code=Code0,options=Opts}=St) -> +core_old_inliner(Code0, #compile{options=Opts}=St) -> {ok,Code} = sys_core_inline:module(Code0, Opts), - {ok,St#compile{code=Code}}. + {ok,Code,St}. -core_inline_module(#compile{code=Code0,options=Opts}=St) -> +core_inline_module(Code0, #compile{options=Opts}=St) -> Code = cerl_inline:core_transform(Code0, Opts), - {ok,St#compile{code=Code}}. - -save_abstract_code(#compile{ifile=File}=St) -> - case abstract_code(St) of - {ok,Code} -> - {ok,St#compile{abstract_code=Code}}; - {error,Es} -> - {error,St#compile{errors=St#compile.errors ++ [{File,Es}]}} - end. + {ok,Code,St}. + +save_abstract_code(Code, St) -> + {ok,Code,St#compile{abstract_code=erl_parse:anno_to_term(Code)}}. + +debug_info(#compile{module=Module,mod_options=Opts0,ofile=OFile,abstract_code=Abst}) -> + AbstOpts = cleanup_compile_options(Opts0), + Opts1 = proplists:delete(debug_info, Opts0), + {Backend,Metadata,Opts2} = + case proplists:get_value(debug_info, Opts0, false) of + {OptBackend,OptMetadata} when is_atom(OptBackend) -> {OptBackend,OptMetadata,Opts1}; + false -> {erl_abstract_code,{none,AbstOpts},Opts1}; + true -> {erl_abstract_code,{Abst,AbstOpts},[debug_info | Opts1]} + end, + DebugInfo = erlang:term_to_binary({debug_info_v1,Backend,Metadata}, [compressed]), -abstract_code(#compile{code=Code0,options=Opts,ofile=OFile}) -> - Code = erl_parse:anno_to_term(Code0), - Abstr = erlang:term_to_binary({raw_abstract_v1,Code}, [compressed]), - case member(encrypt_debug_info, Opts) of + case member(encrypt_debug_info, Opts2) of true -> - case keyfind(debug_info_key, 1, Opts) of - {_,Key} -> - encrypt_abs_code(Abstr, Key); + case lists:keytake(debug_info_key, 1, Opts2) of + {value,{_, Key},Opts3} -> + encrypt_debug_info(DebugInfo, Key, [{debug_info_key,'********'} | Opts3]); false -> - %% Note: #compile.module has not been set yet. - %% Here is an approximation that should work for - %% all valid cases. - Module = list_to_atom(filename:rootname(filename:basename(OFile))), - Mode = proplists:get_value(crypto_mode, Opts, des3_cbc), + Mode = proplists:get_value(crypto_mode, Opts2, des3_cbc), case beam_lib:get_crypto_key({debug_info, Mode, Module, OFile}) of error -> {error, [{none,?MODULE,no_crypto_key}]}; Key -> - encrypt_abs_code(Abstr, {Mode, Key}) + encrypt_debug_info(DebugInfo, {Mode, Key}, Opts2) end end; false -> - {ok, Abstr} + {ok,DebugInfo,Opts2} end. -encrypt_abs_code(Abstr, Key0) -> +encrypt_debug_info(DebugInfo, Key, Opts) -> try - RealKey = generate_key(Key0), + RealKey = generate_key(Key), case start_crypto() of - ok -> {ok,encrypt(RealKey, Abstr)}; + ok -> {ok,encrypt(RealKey, DebugInfo),Opts}; {error,_}=E -> E end catch @@ -1283,6 +1437,32 @@ encrypt_abs_code(Abstr, Key0) -> {error,[{none,?MODULE,bad_crypto_key}]} end. +cleanup_compile_options(Opts) -> + IsDeterministic = lists:member(deterministic, Opts), + lists:filter(fun(Opt) -> + keep_compile_option(Opt, IsDeterministic) + end, Opts). + +%% Include paths and current directory don't affect compilation, but they might +%% be helpful so we include them unless we're doing a deterministic build. +keep_compile_option({i, _}, Deterministic) -> + not Deterministic; +keep_compile_option({cwd, _}, Deterministic) -> + not Deterministic; +%% We are storing abstract, not asm or core. +keep_compile_option(from_asm, _Deterministic) -> + false; +keep_compile_option(from_core, _Deterministic) -> + false; +%% Parse transform and macros have already been applied. +keep_compile_option({parse_transform, _}, _Deterministic) -> + false; +keep_compile_option({d, _, _}, _Deterministic) -> + false; +%% Do not affect compilation result on future calls. +keep_compile_option(Option, _Deterministic) -> + effects_code_generation(Option). + start_crypto() -> try crypto:start() of {error,{already_started,crypto}} -> ok; @@ -1300,33 +1480,53 @@ generate_key(String) when is_list(String) -> encrypt({des3_cbc=Type,Key,IVec,BlockSize}, Bin0) -> Bin1 = case byte_size(Bin0) rem BlockSize of 0 -> Bin0; - N -> list_to_binary([Bin0,random_bytes(BlockSize-N)]) + N -> list_to_binary([Bin0,crypto:strong_rand_bytes(BlockSize-N)]) end, Bin = crypto:block_encrypt(Type, Key, IVec, Bin1), TypeString = atom_to_list(Type), list_to_binary([0,length(TypeString),TypeString,Bin]). -random_bytes(N) -> - _ = random:seed(erlang:time_offset(), - erlang:monotonic_time(), - erlang:unique_integer()), - random_bytes_1(N, []). - -random_bytes_1(0, Acc) -> Acc; -random_bytes_1(N, Acc) -> random_bytes_1(N-1, [random:uniform(255)|Acc]). - -save_core_code(St) -> - {ok,St#compile{core_code=cerl:from_records(St#compile.code)}}. - -beam_asm(#compile{ifile=File,code=Code0, - abstract_code=Abst,mod_options=Opts0}=St) -> - Source = filename:absname(File), - Opts1 = lists:map(fun({debug_info_key,_}) -> {debug_info_key,'********'}; - (Other) -> Other - end, Opts0), - Opts2 = [O || O <- Opts1, effects_code_generation(O)], - case beam_asm:module(Code0, Abst, Source, Opts2) of - {ok,Code} -> {ok,St#compile{code=Code,abstract_code=[]}} +save_core_code(Code, St) -> + {ok,Code,St#compile{core_code=cerl:from_records(Code)}}. + +beam_asm(Code0, #compile{ifile=File,extra_chunks=ExtraChunks,options=CompilerOpts}=St) -> + case debug_info(St) of + {ok,DebugInfo,Opts0} -> + Opts1 = [O || O <- Opts0, effects_code_generation(O)], + Chunks = [{<<"Dbgi">>, DebugInfo} | ExtraChunks], + CompileInfo = compile_info(File, CompilerOpts, Opts1), + {ok,Code} = beam_asm:module(Code0, Chunks, CompileInfo, CompilerOpts), + {ok,Code,St#compile{abstract_code=[]}}; + {error,Es} -> + {error,St#compile{errors=St#compile.errors ++ [{File,Es}]}} + end. + +compile_info(File, CompilerOpts, Opts) -> + IsSlim = member(slim, CompilerOpts), + IsDeterministic = member(deterministic, CompilerOpts), + Info0 = proplists:get_value(compile_info, Opts, []), + Info1 = + case paranoid_absname(File) of + [_|_] = Source when not IsSlim, not IsDeterministic -> + [{source,Source} | Info0]; + _ -> + Info0 + end, + Info2 = + case IsDeterministic of + false -> [{options,proplists:delete(compile_info, Opts)} | Info1]; + true -> Info1 + end, + Info2. + +paranoid_absname(""=File) -> + File; +paranoid_absname(File) -> + case file:get_cwd() of + {ok,Cwd} -> + filename:absname(File, Cwd); + _ -> + File end. test_native(#compile{options=Opts}) -> @@ -1340,17 +1540,17 @@ is_native_enabled([no_native|_]) -> false; is_native_enabled([_|Opts]) -> is_native_enabled(Opts); is_native_enabled([]) -> false. -native_compile(#compile{code=none}=St) -> {ok,St}; -native_compile(St) -> +native_compile(none, St) -> {ok,none,St}; +native_compile(Code, St) -> case erlang:system_info(hipe_architecture) of undefined -> Ws = [{St#compile.ifile,[{none,compile,no_native_support}]}], - {ok,St#compile{warnings=St#compile.warnings ++ Ws}}; + {ok,Code,St#compile{warnings=St#compile.warnings ++ Ws}}; _ -> - native_compile_1(St) + native_compile_1(Code, St) end. -native_compile_1(St) -> +native_compile_1(Code, St) -> Opts0 = St#compile.options, IgnoreErrors = member(ignore_native_errors, Opts0), Opts = case keyfind(hipe, 1, Opts0) of @@ -1360,10 +1560,10 @@ native_compile_1(St) -> end, try hipe:compile(St#compile.module, St#compile.core_code, - St#compile.code, + Code, Opts) of {ok,{_Type,Bin}=T} when is_binary(Bin) -> - {ok,embed_native_code(St, T)}; + {ok,embed_native_code(Code, T),St}; {error,R} -> case IgnoreErrors of true -> @@ -1374,25 +1574,24 @@ native_compile_1(St) -> {error,St#compile{errors=St#compile.errors ++ Es}} end catch - Class:R -> - Stk = erlang:get_stacktrace(), + Class:R:Stack -> case IgnoreErrors of true -> Ws = [{St#compile.ifile, - [{none,?MODULE,{native_crash,R,Stk}}]}], + [{none,?MODULE,{native_crash,R,Stack}}]}], {ok,St#compile{warnings=St#compile.warnings ++ Ws}}; false -> - erlang:raise(Class, R, Stk) + erlang:raise(Class, R, Stack) end end. -embed_native_code(St, {Architecture,NativeCode}) -> - {ok, _, Chunks0} = beam_lib:all_chunks(St#compile.code), +embed_native_code(Code, {Architecture,NativeCode}) -> + {ok, _, Chunks0} = beam_lib:all_chunks(Code), ChunkName = hipe_unified_loader:chunk_name(Architecture), Chunks1 = lists:keydelete(ChunkName, 1, Chunks0), Chunks = Chunks1 ++ [{ChunkName,NativeCode}], - {ok, BeamPlusNative} = beam_lib:build_module(Chunks), - St#compile{code=BeamPlusNative}. + {ok,BeamPlusNative} = beam_lib:build_module(Chunks), + BeamPlusNative. %% effects_code_generation(Option) -> true|false. %% Determine whether the option could have any effect on the @@ -1400,30 +1599,31 @@ embed_native_code(St, {Architecture,NativeCode}) -> %% errors will be reported). effects_code_generation(Option) -> - case Option of + case Option of beam -> false; report_warnings -> false; report_errors -> false; return_errors-> false; return_warnings-> false; + warnings_as_errors -> false; binary -> false; verbose -> false; {cwd,_} -> false; + {outdir, _} -> false; _ -> true end. -save_binary(#compile{code=none}=St) -> {ok,St}; -save_binary(#compile{module=Mod,ofile=Outfile, - options=Opts}=St) -> +save_binary(none, St) -> {ok,none,St}; +save_binary(Code, #compile{module=Mod,ofile=Outfile,options=Opts}=St) -> %% Test that the module name and output file name match. case member(no_error_module_mismatch, Opts) of true -> - save_binary_1(St); + save_binary_1(Code, St); false -> Base = filename:rootname(filename:basename(Outfile)), case atom_to_list(Mod) of Base -> - save_binary_1(St); + save_binary_1(Code, St); _ -> Es = [{St#compile.ofile, [{none,?MODULE,{module_name,Mod,Base}}]}], @@ -1431,14 +1631,14 @@ save_binary(#compile{module=Mod,ofile=Outfile, end end. -save_binary_1(St) -> +save_binary_1(Code, St) -> Ofile = St#compile.ofile, Tfile = tmpfile(Ofile), %Temp working file - case write_binary(Tfile, St#compile.code, St) of + case write_binary(Tfile, Code, St) of ok -> case file:rename(Tfile, Ofile) of ok -> - {ok,St}; + {ok,none,St}; {error,RenameError} -> Es0 = [{Ofile,[{none,?MODULE,{rename,Tfile,Ofile, RenameError}}]}], @@ -1452,8 +1652,8 @@ save_binary_1(St) -> end, {error,St#compile{errors=St#compile.errors ++ Es}} end; - {error,_Error} -> - Es = [{Tfile,[{none,compile,write_error}]}], + {error,Error} -> + Es = [{Tfile,[{none,compile,{write_error,Error}}]}], {error,St#compile{errors=St#compile.errors ++ Es}} end. @@ -1538,6 +1738,9 @@ list_errors(_F, []) -> ok. %% tmpfile(ObjFile) -> TmpFile %% Work out the correct input and output file names. +-spec iofile(atom() | file:filename_all()) -> + {file:name_all(),file:name_all()}. + iofile(File) when is_atom(File) -> iofile(atom_to_list(File)); iofile(File) -> @@ -1578,45 +1781,57 @@ pre_defs([]) -> []. inc_paths(Opts) -> [ P || {i,P} <- Opts, is_list(P) ]. -src_listing(Ext, St) -> +src_listing(Ext, Code, St) -> listing(fun (Lf, {_Mod,_Exp,Fs}) -> do_src_listing(Lf, Fs); (Lf, Fs) -> do_src_listing(Lf, Fs) end, - Ext, St). + Ext, Code, St). do_src_listing(Lf, Fs) -> Opts = [lists:keyfind(encoding, 1, io:getopts(Lf))], foreach(fun (F) -> io:put_chars(Lf, [erl_pp:form(F, Opts),"\n"]) end, Fs). -listing(Ext, St0) -> +listing(Ext, Code, St0) -> St = St0#compile{encoding = none}, - listing(fun(Lf, Fs) -> beam_listing:module(Lf, Fs) end, Ext, St). + listing(fun(Lf, Fs) -> beam_listing:module(Lf, Fs) end, Ext, Code, St). -listing(LFun, Ext, St) -> +listing(LFun, Ext, Code, St) -> Lfile = outfile(St#compile.base, Ext, St#compile.options), case file:open(Lfile, [write,delayed_write]) of {ok,Lf} -> - Code = restore_expanded_types(Ext, St#compile.code), + Code = restore_expanded_types(Ext, Code), output_encoding(Lf, St), LFun(Lf, Code), ok = file:close(Lf), - {ok,St}; - {error,_Error} -> - Es = [{Lfile,[{none,compile,write_error}]}], + {ok,Code,St}; + {error,Error} -> + Es = [{Lfile,[{none,compile,{write_error,Error}}]}], {error,St#compile{errors=St#compile.errors ++ Es}} end. +to_dis(Code, #compile{module=Module,ofile=Outfile}=St) -> + Loaded = code:is_loaded(Module), + Sticky = code:is_sticky(Module), + _ = [code:unstick_mod(Module) || Sticky], + + {module,Module} = code:load_binary(Module, "", Code), + DestDir = filename:dirname(Outfile), + DisFile = filename:join(DestDir, atom_to_list(Module) ++ ".dis"), + ok = erts_debug:dis_to_file(Module, DisFile), + + %% Restore loaded module + _ = [{module, Module} = code:load_file(Module) || Loaded =/= false], + [code:stick_mod(Module) || Sticky], + {ok,Code,St}. + output_encoding(F, #compile{encoding = none}) -> ok = io:setopts(F, [{encoding, epp:default_encoding()}]); output_encoding(F, #compile{encoding = Encoding}) -> ok = io:setopts(F, [{encoding, Encoding}]), ok = io:fwrite(F, <<"%% ~s\n">>, [epp:encoding_to_string(Encoding)]). -restore_expanded_types("P", Fs) -> - epp:restore_typed_record_fields(Fs); restore_expanded_types("E", {M,I,Fs0}) -> - Fs1 = restore_expand_module(Fs0), - Fs = epp:restore_typed_record_fields(Fs1), + Fs = restore_expand_module(Fs0), {M,I,Fs}; restore_expanded_types(_Ext, Code) -> Code. @@ -1628,6 +1843,8 @@ restore_expand_module([{attribute,Line,spec,[Arg]}|Fs]) -> [{attribute,Line,spec,Arg}|restore_expand_module(Fs)]; restore_expand_module([{attribute,Line,callback,[Arg]}|Fs]) -> [{attribute,Line,callback,Arg}|restore_expand_module(Fs)]; +restore_expand_module([{attribute,Line,record,[R]}|Fs]) -> + [{attribute,Line,record,R}|restore_expand_module(Fs)]; restore_expand_module([F|Fs]) -> [F|restore_expand_module(Fs)]; restore_expand_module([]) -> []. @@ -1673,13 +1890,18 @@ help(_) -> %% compile(AbsFileName, Outfilename, Options) %% Compile entry point for erl_compile. +-spec compile(file:filename(), _, #options{}) -> 'ok' | 'error'. + compile(File0, _OutFile, Options) -> + pre_load(), File = shorten_filename(File0), case file(File, make_erl_options(Options)) of {ok,_Mod} -> ok; Other -> Other end. +-spec compile_beam(file:filename(), _, #options{}) -> 'ok' | 'error'. + compile_beam(File0, _OutFile, Opts) -> File = shorten_filename(File0), case file(File, [from_beam|make_erl_options(Opts)]) of @@ -1687,6 +1909,8 @@ compile_beam(File0, _OutFile, Opts) -> Other -> Other end. +-spec compile_asm(file:filename(), _, #options{}) -> 'ok' | 'error'. + compile_asm(File0, _OutFile, Opts) -> File = shorten_filename(File0), case file(File, [from_asm|make_erl_options(Opts)]) of @@ -1694,6 +1918,8 @@ compile_asm(File0, _OutFile, Opts) -> Other -> Other end. +-spec compile_core(file:filename(), _, #options{}) -> 'ok' | 'error'. + compile_core(File0, _OutFile, Opts) -> File = shorten_filename(File0), case file(File, [from_core|make_erl_options(Opts)]) of @@ -1738,3 +1964,47 @@ make_erl_options(Opts) -> end, Options ++ [report_errors, {cwd, Cwd}, {outdir, Outdir}| [{i, Dir} || Dir <- Includes]] ++ Specific. + +pre_load() -> + L = [beam_a, + beam_asm, + beam_block, + beam_bs, + beam_bsm, + beam_clean, + beam_dead, + beam_dict, + beam_except, + beam_flatten, + beam_jump, + beam_opcodes, + beam_peep, + beam_receive, + beam_record, + beam_reorder, + beam_split, + beam_trim, + beam_type, + beam_utils, + beam_validator, + beam_z, + cerl, + cerl_clauses, + cerl_sets, + cerl_trees, + core_lib, + epp, + erl_bifs, + erl_expand_records, + erl_lint, + erl_parse, + erl_scan, + sys_core_alias, + sys_core_bsm, + sys_core_dsetel, + sys_core_fold, + v3_codegen, + v3_core, + v3_kernel], + _ = code:ensure_modules_loaded(L), + ok. diff --git a/lib/compiler/src/compiler.app.src b/lib/compiler/src/compiler.app.src index afb85f4710..cf32fd251c 100644 --- a/lib/compiler/src/compiler.app.src +++ b/lib/compiler/src/compiler.app.src @@ -1,7 +1,7 @@ % This is an -*- erlang -*- file. %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1997-2013. All Rights Reserved. +%% Copyright Ericsson AB 1997-2017. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -24,7 +24,7 @@ beam_a, beam_asm, beam_block, - beam_bool, + beam_bs, beam_bsm, beam_clean, beam_dead, @@ -37,6 +37,8 @@ beam_opcodes, beam_peep, beam_receive, + beam_reorder, + beam_record, beam_split, beam_trim, beam_type, @@ -56,20 +58,20 @@ core_lib, erl_bifs, rec_env, + sys_core_alias, + sys_core_bsm, sys_core_dsetel, sys_core_fold, sys_core_fold_lists, sys_core_inline, sys_pre_attributes, - sys_pre_expand, v3_codegen, v3_core, v3_kernel, - v3_kernel_pp, - v3_life + v3_kernel_pp ]}, {registered, []}, {applications, [kernel, stdlib]}, {env, []}, - {runtime_dependencies, ["stdlib-2.5","kernel-4.0","hipe-3.12","erts-7.0", + {runtime_dependencies, ["stdlib-2.5","kernel-4.0","hipe-3.12","erts-9.0", "crypto-3.6"]}]}. diff --git a/lib/compiler/src/compiler.appup.src b/lib/compiler/src/compiler.appup.src index 3ada2e933f..e9db37e203 100644 --- a/lib/compiler/src/compiler.appup.src +++ b/lib/compiler/src/compiler.appup.src @@ -1,7 +1,7 @@ %% -*- erlang -*- %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2014. All Rights Reserved. +%% Copyright Ericsson AB 2014-2016. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. diff --git a/lib/compiler/src/core_lib.erl b/lib/compiler/src/core_lib.erl index 3abb520485..c1806272bd 100644 --- a/lib/compiler/src/core_lib.erl +++ b/lib/compiler/src/core_lib.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2000-2009. All Rights Reserved. +%% Copyright Ericsson AB 2000-2016. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -21,52 +21,16 @@ -module(core_lib). --deprecated({get_anno,1,next_major_release}). --deprecated({set_anno,2,next_major_release}). --deprecated({is_literal,1,next_major_release}). --deprecated({is_literal_list,1,next_major_release}). --deprecated({literal_value,1,next_major_release}). - --export([get_anno/1,set_anno/2]). --export([is_literal/1,is_literal_list/1]). --export([literal_value/1]). -export([make_values/1]). -export([is_var_used/2]). -include("core_parse.hrl"). -%% -%% Generic get/set annotation that should be used only with cerl() structures. -%% --spec get_anno(cerl:cerl()) -> term(). - -get_anno(C) -> cerl:get_ann(C). - --spec set_anno(cerl:cerl(), term()) -> cerl:cerl(). - -set_anno(C, A) -> cerl:set_ann(C, A). - --spec is_literal(cerl:cerl()) -> boolean(). - -is_literal(Cerl) -> - cerl:is_literal(cerl:fold_literal(Cerl)). - --spec is_literal_list([cerl:cerl()]) -> boolean(). - -is_literal_list(Es) -> lists:all(fun is_literal/1, Es). - -%% Return the value of LitExpr. --spec literal_value(cerl:c_literal() | cerl:c_binary() | - cerl:c_map() | cerl:c_cons() | cerl:c_tuple()) -> term(). - -literal_value(Cerl) -> - cerl:concrete(cerl:fold_literal(Cerl)). - %% Make a suitable values structure, expr or values, depending on Expr. -spec make_values([cerl:cerl()] | cerl:cerl()) -> cerl:cerl(). make_values([E]) -> E; -make_values([H|_]=Es) -> #c_values{anno=get_anno(H),es=Es}; +make_values([H|_]=Es) -> #c_values{anno=cerl:get_ann(H),es=Es}; make_values([]) -> #c_values{es=[]}; make_values(E) -> E. diff --git a/lib/compiler/src/core_lint.erl b/lib/compiler/src/core_lint.erl index cc54f6e411..3f69cb03a9 100644 --- a/lib/compiler/src/core_lint.erl +++ b/lib/compiler/src/core_lint.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2013. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -73,7 +73,7 @@ %% Define the lint state record. -record(lint, {module :: module(), % Current module - func :: fa(), % Current function + func :: fa() | 'undefined', % Current function errors = [] :: [error()], % Errors warnings= [] :: [warning()]}). % Warnings @@ -353,12 +353,6 @@ expr(#c_case{arg=Arg,clauses=Cs}, Def, Rt, St0) -> Pc = case_patcount(Cs), St1 = body(Arg, Def, Pc, St0), clauses(Cs, Def, Pc, Rt, St1); -expr(#c_receive{clauses=Cs,timeout=#c_literal{val=infinity}, - action=#c_literal{}}, - Def, Rt, St) -> - %% If the timeout is 'infinity', the after code can never - %% be reached. We don't care if the return count is wrong. - clauses(Cs, Def, 1, Rt, St); expr(#c_receive{clauses=Cs,timeout=T,action=A}, Def, Rt, St0) -> St1 = expr(T, Def, 1, St0), St2 = body(A, Def, Rt, St1), @@ -497,8 +491,10 @@ pattern(#c_tuple{es=Es}, Def, Ps, St) -> pattern_list(Es, Def, Ps, St); pattern(#c_map{es=Es}, Def, Ps, St) -> pattern_list(Es, Def, Ps, St); -pattern(#c_map_pair{op=#c_literal{val=exact},key=K,val=V},Def,Ps,St) -> - pattern_list([K,V],Def,Ps,St); +pattern(#c_map_pair{op=#c_literal{val=exact},key=K,val=V}, Def, Ps, St) -> + %% The key is an input. + pat_map_expr(K, Def, St), + pattern_list([V],Def,Ps,St); pattern(#c_binary{segments=Ss}, Def, Ps, St0) -> St = pat_bin_tail_check(Ss, St0), pat_bin(Ss, Def, Ps, St); @@ -561,6 +557,10 @@ pat_bit_expr(#c_binary{}, _, _Def, St) -> pat_bit_expr(_, _, _, St) -> add_error({illegal_expr,St#lint.func}, St). +pat_map_expr(#c_var{name=N}, Def, St) -> expr_var(N, Def, St); +pat_map_expr(#c_literal{}, _Def, St) -> St; +pat_map_expr(_, _, St) -> add_error({illegal_expr,St#lint.func}, St). + %% pattern_list([Var], Defined, State) -> {[PatVar],State}. %% pattern_list([Var], Defined, [PatVar], State) -> {[PatVar],State}. diff --git a/lib/compiler/src/core_parse.hrl b/lib/compiler/src/core_parse.hrl index ecf6cc9956..83a6f0179c 100644 --- a/lib/compiler/src/core_parse.hrl +++ b/lib/compiler/src/core_parse.hrl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2009. All Rights Reserved. +%% Copyright Ericsson AB 1999-2016. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. diff --git a/lib/compiler/src/core_parse.yrl b/lib/compiler/src/core_parse.yrl index 59cfc97ebd..10d8c44dd3 100644 --- a/lib/compiler/src/core_parse.yrl +++ b/lib/compiler/src/core_parse.yrl @@ -1,7 +1,7 @@ %% -*-Erlang-*- %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2009. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -36,7 +36,7 @@ other_pattern atomic_pattern tuple_pattern cons_pattern tail_pattern binary_pattern segment_patterns segment_pattern expression single_expression -literal literals atomic_literal tuple_literal cons_literal tail_literal +literal literals atomic_literal tuple_literal cons_literal tail_literal fun_literal nil tuple cons tail binary segments segment @@ -47,12 +47,14 @@ receive_expr timeout try_expr sequence catch_expr variable clause clause_pattern -map_expr map_pairs map_pair map_pair_assoc map_pair_exact +map_expr anno_map_expr map_pairs anno_map_pair map_pair map_pair_assoc map_pair_exact map_pattern map_pair_patterns map_pair_pattern -annotation anno_fun anno_expression anno_expressions +annotation anno_atom anno_fun anno_expression anno_expressions anno_variable anno_variables anno_pattern anno_patterns anno_function_name +anno_literal +anno_segment anno_segment_pattern anno_clause anno_clauses. Terminals @@ -90,7 +92,7 @@ module_definition -> module_definition -> '(' 'module' atom module_export module_attribute module_defs 'end' '-|' annotation ')' : - #c_module{anno='$9',name=tok_val('$3'),exports='$4', + #c_module{anno='$9',name=#c_literal{val=tok_val('$3')},exports='$4', attrs='$5',defs='$6'}. module_export -> '[' ']' : []. @@ -99,7 +101,7 @@ module_export -> '[' exported_names ']' : '$2'. exported_names -> exported_name ',' exported_names : ['$1' | '$3']. exported_names -> exported_name : ['$1']. -exported_name -> function_name : '$1'. +exported_name -> anno_function_name : '$1'. module_attribute -> 'attributes' '[' ']' : []. module_attribute -> 'attributes' '[' attribute_list ']' : '$3'. @@ -107,8 +109,16 @@ module_attribute -> 'attributes' '[' attribute_list ']' : '$3'. attribute_list -> attribute ',' attribute_list : ['$1' | '$3']. attribute_list -> attribute : ['$1']. -attribute -> atom '=' literal : - {#c_literal{val=tok_val('$1')},'$3'}. +attribute -> anno_atom '=' anno_literal : + {'$1','$3'}. + +anno_atom -> atom : + cerl:c_atom(tok_val('$1')). +anno_atom -> '(' atom '-|' annotation ')' : + cerl:ann_c_atom('$4', tok_val('$2')). + +anno_literal -> literal : '$1'. +anno_literal -> '(' literal '-|' annotation ')' : cerl:set_ann('$2', '$4'). module_defs -> function_definitions : '$1'. @@ -186,7 +196,9 @@ tuple_pattern -> '{' anno_patterns '}' : c_tuple('$2'). map_pattern -> '~' '{' '}' '~' : c_map_pattern([]). map_pattern -> '~' '{' map_pair_patterns '}' '~' : - c_map_pattern(lists:sort('$3')). + c_map_pattern('$3'). +map_pattern -> '~' '{' map_pair_patterns '|' anno_map_expr '}' '~' : + ann_c_map_pattern('$5', '$3'). map_pair_patterns -> map_pair_pattern : ['$1']. map_pair_patterns -> map_pair_pattern ',' map_pair_patterns : ['$1' | '$3']. @@ -194,6 +206,9 @@ map_pair_patterns -> map_pair_pattern ',' map_pair_patterns : ['$1' | '$3']. map_pair_pattern -> anno_expression ':=' anno_pattern : #c_map_pair{op=#c_literal{val=exact}, key='$1',val='$3'}. +map_pair_pattern -> '(' anno_expression ':=' anno_pattern '-|' annotation ')' : + #c_map_pair{anno='$6',op=#c_literal{val=exact}, + key='$2',val='$4'}. cons_pattern -> '[' anno_pattern tail_pattern : c_cons('$2', '$3'). @@ -206,8 +221,12 @@ tail_pattern -> ',' anno_pattern tail_pattern : binary_pattern -> '#' '{' '}' '#' : #c_binary{segments=[]}. binary_pattern -> '#' '{' segment_patterns '}' '#' : #c_binary{segments='$3'}. -segment_patterns -> segment_pattern ',' segment_patterns : ['$1' | '$3']. -segment_patterns -> segment_pattern : ['$1']. +segment_patterns -> anno_segment_pattern ',' segment_patterns : ['$1' | '$3']. +segment_patterns -> anno_segment_pattern : ['$1']. + +anno_segment_pattern -> segment_pattern : '$1'. +anno_segment_pattern -> '(' segment_pattern '-|' annotation ')' : + cerl:set_ann('$2', '$4'). segment_pattern -> '#' '<' anno_pattern '>' '(' anno_expressions ')': case '$6' of @@ -248,6 +267,7 @@ single_expression -> cons : '$1'. single_expression -> binary : '$1'. single_expression -> variable : '$1'. single_expression -> function_name : '$1'. +single_expression -> fun_literal : '$1'. single_expression -> fun_expr : '$1'. single_expression -> let_expr : '$1'. single_expression -> letrec_expr : '$1'. @@ -284,16 +304,25 @@ tail_literal -> ']' : #c_literal{val=[]}. tail_literal -> '|' literal ']' : '$2'. tail_literal -> ',' literal tail_literal : c_cons('$2', '$3'). +fun_literal -> 'fun' atom ':' atom '/' integer : + #c_literal{val = erlang:make_fun(tok_val('$2'), tok_val('$4'), tok_val('$6'))}. + tuple -> '{' '}' : c_tuple([]). tuple -> '{' anno_expressions '}' : c_tuple('$2'). map_expr -> '~' '{' '}' '~' : c_map([]). map_expr -> '~' '{' map_pairs '}' '~' : c_map('$3'). -map_expr -> '~' '{' map_pairs '|' variable '}' '~' : ann_c_map([], '$5', '$3'). -map_expr -> '~' '{' map_pairs '|' map_expr '}' '~' : ann_c_map([], '$5', '$3'). +map_expr -> '~' '{' map_pairs '|' anno_variable '}' '~' : ann_c_map([], '$5', '$3'). +map_expr -> '~' '{' map_pairs '|' anno_map_expr '}' '~' : ann_c_map([], '$5', '$3'). + +anno_map_expr -> map_expr : '$1'. +anno_map_expr -> '(' map_expr '-|' annotation ')' : cerl:set_ann('$2', '$4'). -map_pairs -> map_pair : ['$1']. -map_pairs -> map_pair ',' map_pairs : ['$1' | '$3']. +map_pairs -> anno_map_pair : ['$1']. +map_pairs -> anno_map_pair ',' map_pairs : ['$1' | '$3']. + +anno_map_pair -> map_pair : '$1'. +anno_map_pair -> '(' map_pair '-|' annotation ')' : cerl:set_ann('$2', '$4'). map_pair -> map_pair_assoc : '$1'. map_pair -> map_pair_exact : '$1'. @@ -312,8 +341,11 @@ tail -> ',' anno_expression tail : c_cons('$2', '$3'). binary -> '#' '{' '}' '#' : #c_literal{val = <<>>}. binary -> '#' '{' segments '}' '#' : make_binary('$3'). -segments -> segment ',' segments : ['$1' | '$3']. -segments -> segment : ['$1']. +segments -> anno_segment ',' segments : ['$1' | '$3']. +segments -> anno_segment : ['$1']. + +anno_segment -> segment : '$1'. +anno_segment -> '(' segment '-|' annotation ')' : cerl:set_ann('$2', '$4'). segment -> '#' '<' anno_expression '>' '(' anno_expressions ')': case '$6' of @@ -404,6 +436,21 @@ timeout -> %% ====================================================================== %% +Header +"%% This file was automatically generated from the file \"core_parse.yrl\"." +"%%" +"%% Copyright Ericsson AB 1999-2009. All Rights Reserved." +"%%" +"%% Licensed under the Apache License, Version 2.0 (the \"License\"); you may" +"%% not use this file except in compliance with the License. You may obtain" +"%% a copy of the License at <http://www.apache.org/licenses/LICENSE-2.0>" +"%%" +"%% Unless required by applicable law or agreed to in writing, software" +"%% distributed under the License is distributed on an \"AS IS\" BASIS," +"%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied." +"%% See the License for the specific language governing permissions and" +"%% limitations under the License." +"". Erlang code. @@ -413,7 +460,8 @@ Erlang code. -include("core_parse.hrl"). --import(cerl, [ann_c_map/3,c_cons/2,c_map/1,c_map_pattern/1,c_tuple/1]). +-import(cerl, [ann_c_map/3,ann_c_map_pattern/2,c_cons/2,c_map/1, + c_map_pattern/1,c_tuple/1]). tok_val(T) -> element(3, T). tok_line(T) -> element(2, T). @@ -452,7 +500,7 @@ make_lit_bin(Acc, [#c_bitstr{val=I0,size=Sz0,unit=U0,type=Type0,flags=F0}|T]) -> throw(impossible) end, if - Sz =< 8, T =:= [] -> + 0 =< Sz, Sz =< 8, T =:= [] -> <<Acc/binary,I:Sz>>; Sz =:= 8 -> make_lit_bin(<<Acc/binary,I:8>>, T); diff --git a/lib/compiler/src/core_pp.erl b/lib/compiler/src/core_pp.erl index c2a6a81d5e..cb3f24fd08 100644 --- a/lib/compiler/src/core_pp.erl +++ b/lib/compiler/src/core_pp.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2009. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -21,7 +21,7 @@ -module(core_pp). --export([format/1]). +-export([format/1,format_all/1]). -include("core_parse.hrl"). @@ -33,25 +33,35 @@ %% Prettyprint-formats (naively) an abstract Core Erlang syntax %% tree. --record(ctxt, {class = term :: 'clause' | 'def' | 'expr' | 'term', - indent = 0 :: integer(), +-record(ctxt, {indent = 0 :: integer(), item_indent = 2 :: integer(), body_indent = 4 :: integer(), - tab_width = 8 :: non_neg_integer(), - line = 0 :: integer()}). + line = 0 :: integer(), + clean = true :: boolean()}). + +-define(TAB_WIDTH, 8). -spec format(cerl:cerl()) -> iolist(). format(Node) -> format(Node, #ctxt{}). -maybe_anno(Node, Fun, Ctxt) -> +-spec format_all(cerl:cerl()) -> iolist(). + +format_all(Node) -> + format(Node, #ctxt{clean=false}). + +maybe_anno(Node, Fun, #ctxt{clean=false}=Ctxt) -> As = cerl:get_ann(Node), - case get_line(As) of + maybe_anno(Node, Fun, Ctxt, As); +maybe_anno(Node, Fun, #ctxt{clean=true}=Ctxt) -> + As0 = cerl:get_ann(Node), + case get_line(As0) of none -> - maybe_anno(Node, Fun, Ctxt, As); + maybe_anno(Node, Fun, Ctxt, As0); Line -> - if Line > Ctxt#ctxt.line -> + As = strip_line(As0), + if Line > Ctxt#ctxt.line -> [io_lib:format("%% Line ~w",[Line]), nl_indent(Ctxt), maybe_anno(Node, Fun, Ctxt#ctxt{line = Line}, As) @@ -61,22 +71,22 @@ maybe_anno(Node, Fun, Ctxt) -> end end. -maybe_anno(Node, Fun, Ctxt, As) -> - case strip_line(As) of - [] -> - Fun(Node, Ctxt); - List -> - Ctxt1 = add_indent(Ctxt, 2), - Ctxt2 = add_indent(Ctxt1, 3), - ["( ", - Fun(Node, Ctxt1), - nl_indent(Ctxt1), - "-| ",format_anno(List, Ctxt2)," )" - ] - end. +maybe_anno(Node, Fun, Ctxt, []) -> + Fun(Node, Ctxt); +maybe_anno(Node, Fun, Ctxt, List) -> + Ctxt1 = add_indent(Ctxt, 2), + Ctxt2 = add_indent(Ctxt1, 3), + ["( ", + Fun(Node, Ctxt1), + nl_indent(Ctxt1), + "-| ",format_anno(List, Ctxt2)," )" + ]. format_anno([_|_]=List, Ctxt) -> [$[,format_anno_list(List, Ctxt),$]]; +format_anno({file,Name}, _Ctxt) -> + %% Optimization: Reduces file size considerably. + io_lib:format("{'file',~p}", [Name]); format_anno(Tuple, Ctxt) when is_tuple(Tuple) -> [${,format_anno_list(tuple_to_list(Tuple), Ctxt),$}]; format_anno(Val, Ctxt) when is_atom(Val) -> @@ -121,14 +131,16 @@ format_1(#c_literal{anno=A,val=Bitstring}, Ctxt) when is_bitstring(Bitstring) -> format_1(#c_binary{anno=A,segments=Segs}, Ctxt); format_1(#c_literal{anno=A,val=M},Ctxt) when is_map(M) -> Pairs = maps:to_list(M), - Op = case Ctxt of - #ctxt{ class = clause } -> exact; - _ -> assoc - end, - Cpairs = [#c_map_pair{op=#c_literal{val=Op}, + Op = #c_literal{val=assoc}, + Cpairs = [#c_map_pair{op=Op, key=#c_literal{val=K}, val=#c_literal{val=V}} || {K,V} <- Pairs], - format_1(#c_map{anno=A,arg=#c_literal{val=#{}},es=Cpairs},Ctxt); + format_1(#c_map{anno=A,arg=#c_literal{val=#{}},es=Cpairs},Ctxt); +format_1(#c_literal{val=F},_Ctxt) when is_function(F) -> + {module,M} = erlang:fun_info(F, module), + {name,N} = erlang:fun_info(F, name), + {arity,A} = erlang:fun_info(F, arity), + ["fun ",core_atom(M),$:,core_atom(N),$/,integer_to_list(A)]; format_1(#c_var{name={I,A}}, _) -> [core_atom(I),$/,integer_to_list(A)]; format_1(#c_var{name=V}, _) -> @@ -172,7 +184,8 @@ format_1(#c_tuple{es=Es}, Ctxt) -> format_hseq(Es, ",", add_indent(Ctxt, 1), fun format/2), $} ]; -format_1(#c_map{arg=#c_literal{val=M},es=Es}, Ctxt) when is_map(M),map_size(M)=:=0 -> +format_1(#c_map{arg=#c_literal{val=M},es=Es}, Ctxt) + when is_map(M), map_size(M) =:= 0 -> ["~{", format_hseq(Es, ",", add_indent(Ctxt, 1), fun format/2), "}~" @@ -195,9 +208,16 @@ format_1(#c_values{es=Es}, Ctxt) -> format_1(#c_alias{var=V,pat=P}, Ctxt) -> Txt = [format(V, Ctxt)|" = "], [Txt|format(P, add_indent(Ctxt, width(Txt, Ctxt)))]; -format_1(#c_let{vars=Vs0,arg=A,body=B}, Ctxt) -> - Vs = [cerl:set_ann(V, []) || V <- Vs0], - case is_simple_term(A) of +format_1(#c_let{anno=Anno0,vars=Vs0,arg=A0,body=B}, #ctxt{clean=Clean}=Ctxt) -> + {Vs,A,Anno} = case Clean of + false -> + {Vs0,A0,Anno0}; + true -> + {[cerl:set_ann(V, []) || V <- Vs0], + cerl:set_ann(A0, []), + []} + end, + case is_simple_term(A) andalso Anno =:= [] of false -> Ctxt1 = add_indent(Ctxt, Ctxt#ctxt.body_indent), ["let ", @@ -214,7 +234,7 @@ format_1(#c_let{vars=Vs0,arg=A,body=B}, Ctxt) -> ["let ", format_values(Vs, add_indent(Ctxt, 4)), " = ", - format(cerl:set_ann(A, []), Ctxt1), + format(A, Ctxt1), nl_indent(Ctxt), "in " | format(B, add_indent(Ctxt, 4)) @@ -321,35 +341,30 @@ format_1(#c_module{name=N,exports=Es,attrs=As,defs=Ds}, Ctxt) -> [Mod," [", format_vseq(Es, "", ",", - add_indent(set_class(Ctxt, term), width(Mod, Ctxt)+2), + add_indent(Ctxt, width(Mod, Ctxt)+2), fun format/2), "]", nl_indent(Ctxt), " attributes [", format_vseq(As, "", ",", - add_indent(set_class(Ctxt, def), 16), + add_indent(Ctxt, 16), fun format_def/2), "]", nl_indent(Ctxt), format_funcs(Ds, Ctxt), nl_indent(Ctxt) | "end" - ]; -format_1(Type, _) -> - ["** Unsupported type: ", - io_lib:write(Type) - | " **" ]. format_funcs(Fs, Ctxt) -> format_vseq(Fs, "", "", - set_class(Ctxt, def), + Ctxt, fun format_def/2). format_def({N,V}, Ctxt0) -> - Ctxt1 = add_indent(set_class(Ctxt0, expr), Ctxt0#ctxt.body_indent), + Ctxt1 = add_indent(Ctxt0, Ctxt0#ctxt.body_indent), [format(N, Ctxt0), " =", nl_indent(Ctxt1) @@ -362,7 +377,10 @@ format_values(Vs, Ctxt) -> format_hseq(Vs, ",", add_indent(Ctxt, 1), fun format/2), $>]. -format_bitstr(#c_bitstr{val=V,size=S,unit=U,type=T,flags=Fs}, Ctxt0) -> +format_bitstr(Node, Ctxt) -> + maybe_anno(Node, fun do_format_bitstr/2, Ctxt). + +do_format_bitstr(#c_bitstr{val=V,size=S,unit=U,type=T,flags=Fs}, Ctxt0) -> Vs = [S, U, T, Fs], Ctxt1 = add_indent(Ctxt0, 2), Val = format(V, Ctxt1), @@ -370,8 +388,7 @@ format_bitstr(#c_bitstr{val=V,size=S,unit=U,type=T,flags=Fs}, Ctxt0) -> ["#<", Val, ">(", format_hseq(Vs,",", Ctxt2, fun format/2), $)]. format_clauses(Cs, Ctxt) -> - format_vseq(Cs, "", "", set_class(Ctxt, clause), - fun format_clause/2). + format_vseq(Cs, "", "", Ctxt, fun format_clause/2). format_clause(Node, Ctxt) -> maybe_anno(Node, fun format_clause_1/2, Ctxt). @@ -383,15 +400,13 @@ format_clause_1(#c_clause{pats=Ps,guard=G,body=B}, Ctxt) -> case is_trivial_guard(G) of true -> [" when ", - format_guard(G, add_indent(set_class(Ctxt, expr), - width(Ptxt, Ctxt) + 6))]; + format_guard(G, add_indent(Ctxt, width(Ptxt, Ctxt) + 6))]; false -> [nl_indent(Ctxt2), "when ", format_guard(G, add_indent(Ctxt2, 2))] end++ " ->", - nl_indent(Ctxt2) - | format(B, set_class(Ctxt2, expr)) + nl_indent(Ctxt2) | format(B, Ctxt2) ]. is_trivial_guard(#c_literal{val=Val}) when is_atom(Val) -> true; @@ -445,50 +460,53 @@ format_list_tail(Tail, Ctxt) -> format_map_pair(Op, K, V, Ctxt0) -> Ctxt1 = add_indent(Ctxt0, 1), - Txt = format(K, set_class(Ctxt1, expr)), + Txt = format(K, Ctxt1), Ctxt2 = add_indent(Ctxt0, width(Txt, Ctxt1)), [Txt,Op,format(V, Ctxt2)]. -indent(Ctxt) -> indent(Ctxt#ctxt.indent, Ctxt). - -indent(N, _) when N =< 0 -> ""; -indent(N, Ctxt) -> - T = Ctxt#ctxt.tab_width, - string:chars($\t, N div T, string:chars($\s, N rem T)). +indent(#ctxt{indent=N}) -> + if + N =< 0 -> + ""; + true -> + lists:duplicate(N div ?TAB_WIDTH, $\t) ++ spaces(N rem ?TAB_WIDTH) + end. nl_indent(Ctxt) -> [$\n|indent(Ctxt)]. +spaces(0) -> ""; +spaces(1) -> " "; +spaces(2) -> " "; +spaces(3) -> " "; +spaces(4) -> " "; +spaces(5) -> " "; +spaces(6) -> " "; +spaces(7) -> " ". +%% Undo indentation done by nl_indent/1. unindent(T, Ctxt) -> - unindent(T, Ctxt#ctxt.indent, Ctxt, []). + unindent(T, Ctxt#ctxt.indent, []). -unindent(T, N, _, C) when N =< 0 -> +unindent(T, N, C) when N =< 0 -> [T|C]; -unindent([$\s|T], N, Ctxt, C) -> - unindent(T, N - 1, Ctxt, C); -unindent([$\t|T], N, Ctxt, C) -> - Tab = Ctxt#ctxt.tab_width, +unindent([$\s|T], N, C) -> + unindent(T, N - 1, C); +unindent([$\t|T], N, C) -> + Tab = ?TAB_WIDTH, if N >= Tab -> - unindent(T, N - Tab, Ctxt, C); + unindent(T, N - Tab, C); true -> - unindent([string:chars($\s, Tab - N)|T], 0, Ctxt, C) + unindent([spaces(Tab - N)|T], 0, C) end; -unindent([L|T], N, Ctxt, C) when is_list(L) -> - unindent(L, N, Ctxt, [T|C]); -unindent([H|T], _, _, C) -> - [H|[T|C]]; -unindent([], N, Ctxt, [H|T]) -> - unindent(H, N, Ctxt, T); -unindent([], _, _, []) -> []. +unindent([L|T], N, C) when is_list(L) -> + unindent(L, N, [T|C]). width(Txt, Ctxt) -> - try width(Txt, 0, Ctxt, []) - catch error:_ -> exit({bad_text,Txt}) - end. + width(Txt, 0, Ctxt, []). width([$\t|T], A, Ctxt, C) -> - width(T, A + Ctxt#ctxt.tab_width, Ctxt, C); + width(T, A + ?TAB_WIDTH, Ctxt, C); width([$\n|T], _, Ctxt, C) -> width(unindent([T|C], Ctxt), Ctxt); width([H|T], A, Ctxt, C) when is_list(H) -> @@ -502,14 +520,9 @@ width([], A, _, []) -> A. add_indent(Ctxt, Dx) -> Ctxt#ctxt{indent = Ctxt#ctxt.indent + Dx}. -set_class(Ctxt, Class) -> - Ctxt#ctxt{class = Class}. - core_atom(A) -> io_lib:write_string(atom_to_list(A), $'). -is_simple_term(#c_values{es=Es}) -> - length(Es) < 3 andalso lists:all(fun is_simple_term/1, Es); is_simple_term(#c_tuple{es=Es}) -> length(Es) < 4 andalso lists:all(fun is_simple_term/1, Es); is_simple_term(#c_var{}) -> true; @@ -533,4 +546,3 @@ segs_from_bitstring(Bitstring) -> unit=#c_literal{val=1}, type=#c_literal{val=integer}, flags=#c_literal{val=[unsigned,big]}}]. - diff --git a/lib/compiler/src/core_scan.erl b/lib/compiler/src/core_scan.erl index 5e85bba2bd..a50a2ffa8d 100644 --- a/lib/compiler/src/core_scan.erl +++ b/lib/compiler/src/core_scan.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2000-2013. All Rights Reserved. +%% Copyright Ericsson AB 2000-2017. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -49,13 +49,37 @@ -import(lists, [reverse/1]). +-type location() :: integer(). +-type category() :: atom(). +-type symbol() :: atom() | float() | integer() | string(). +-type token() :: {category(), Anno :: location(), symbol()} + | {category(), Anno :: location()}. +-type tokens() :: [token()]. +-type error_description() :: term(). +-type error_info() :: {erl_anno:location(), module(), error_description()}. + %% string([Char]) -> %% string([Char], StartPos) -> %% {ok, [Tok], EndPos} | %% {error, {Pos,core_scan,What}, EndPos} +-spec string(String) -> Return when + String :: string(), + Return :: {'ok', Tokens :: tokens(), EndLocation} + | {'error', ErrorInfo :: error_info(), ErrorLocation}, + EndLocation :: location(), + ErrorLocation :: location(). + string(Cs) -> string(Cs, 1). +-spec string(String, StartLocation) -> Return when + String :: string(), + Return :: {'ok', Tokens :: tokens(), EndLocation} + | {'error', ErrorInfo :: error_info(), ErrorLocation}, + StartLocation :: location(), + EndLocation :: location(), + ErrorLocation :: location(). + string(Cs, Sp) -> %% Add an 'eof' to always get correct handling. case string_pre_scan(Cs, [], Sp) of @@ -176,8 +200,8 @@ pre_string(eof, Q, _, Sp, SoFar, Pos) -> pre_string_error(Q, Sp, SoFar, Pos). pre_string_error(Q, Sp, SoFar, Pos) -> - S = reverse(string:substr(SoFar, 1, string:chr(SoFar, Q)-1)), - pre_error({string,Q,string:substr(S, 1, 16)}, Sp, Pos). + [S,_] = string:split(SoFar, [Q]), + pre_error({string,Q,string:slice(string:reverse(S), 0, 16)}, Sp, Pos). pre_char([C|Cs], SoFar) -> pre_char(C, Cs, SoFar); pre_char([], _) -> more; @@ -259,10 +283,12 @@ scan1([$$|Cs0], Toks, Pos) -> %Character constant scan1(Cs, [{char,Pos,C}|Toks], Pos1); scan1([$'|Cs0], Toks, Pos) -> %Atom (always quoted) {S,Cs1,Pos1} = scan_string(Cs0, $', Pos), - case catch list_to_atom(S) of + try binary_to_atom(list_to_binary(S), utf8) of A when is_atom(A) -> - scan1(Cs1, [{atom,Pos,A}|Toks], Pos1); - _Error -> scan_error({illegal,atom}, Pos) + scan1(Cs1, [{atom,Pos,A}|Toks], Pos1) + catch + error:_ -> + scan_error({illegal,atom}, Pos) end; scan1([$"|Cs0], Toks, Pos) -> %String {S,Cs1,Pos1} = scan_string(Cs0, $", Pos), diff --git a/lib/compiler/src/erl_bifs.erl b/lib/compiler/src/erl_bifs.erl index c00f5eab70..68489a0122 100644 --- a/lib/compiler/src/erl_bifs.erl +++ b/lib/compiler/src/erl_bifs.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2001-2013. All Rights Reserved. +%% Copyright Ericsson AB 2001-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -75,11 +75,12 @@ is_pure(erlang, binary_to_list, 1) -> true; is_pure(erlang, binary_to_list, 3) -> true; is_pure(erlang, bit_size, 1) -> true; is_pure(erlang, byte_size, 1) -> true; +is_pure(erlang, ceil, 1) -> true; is_pure(erlang, element, 2) -> true; is_pure(erlang, float, 1) -> true; is_pure(erlang, float_to_list, 1) -> true; is_pure(erlang, float_to_binary, 1) -> true; -is_pure(erlang, hash, 2) -> false; +is_pure(erlang, floor, 1) -> true; is_pure(erlang, hd, 1) -> true; is_pure(erlang, integer_to_binary, 1) -> true; is_pure(erlang, integer_to_list, 1) -> true; @@ -93,6 +94,7 @@ is_pure(erlang, is_function, 1) -> true; is_pure(erlang, is_integer, 1) -> true; is_pure(erlang, is_list, 1) -> true; is_pure(erlang, is_map, 1) -> true; +is_pure(erlang, is_map_key, 2) -> true; is_pure(erlang, is_number, 1) -> true; is_pure(erlang, is_pid, 1) -> true; is_pure(erlang, is_port, 1) -> true; @@ -108,6 +110,8 @@ is_pure(erlang, list_to_integer, 1) -> true; is_pure(erlang, list_to_pid, 1) -> true; is_pure(erlang, list_to_tuple, 1) -> true; is_pure(erlang, max, 2) -> true; +is_pure(erlang, make_fun, 3) -> true; +is_pure(erlang, map_get, 2) -> true; is_pure(erlang, min, 2) -> true; is_pure(erlang, phash, 2) -> false; is_pure(erlang, pid_to_list, 1) -> true; @@ -129,11 +133,14 @@ is_pure(math, asinh, 1) -> true; is_pure(math, atan, 1) -> true; is_pure(math, atan2, 2) -> true; is_pure(math, atanh, 1) -> true; +is_pure(math, ceil, 1) -> true; is_pure(math, cos, 1) -> true; is_pure(math, cosh, 1) -> true; is_pure(math, erf, 1) -> true; is_pure(math, erfc, 1) -> true; is_pure(math, exp, 1) -> true; +is_pure(math, floor, 1) -> true; +is_pure(math, fmod, 2) -> true; is_pure(math, log, 1) -> true; is_pure(math, log2, 1) -> true; is_pure(math, log10, 1) -> true; @@ -192,6 +199,7 @@ is_safe(erlang, is_port, 1) -> true; is_safe(erlang, is_reference, 1) -> true; is_safe(erlang, is_tuple, 1) -> true; is_safe(erlang, make_ref, 0) -> true; +is_safe(erlang, make_fun, 3) -> true; is_safe(erlang, max, 2) -> true; is_safe(erlang, min, 2) -> true; is_safe(erlang, node, 0) -> true; @@ -203,7 +211,6 @@ is_safe(erlang, registered, 0) -> true; is_safe(erlang, self, 0) -> true; is_safe(erlang, term_to_binary, 1) -> true; is_safe(erlang, time, 0) -> true; -is_safe(error_logger, warning_map, 0) -> true; is_safe(_, _, _) -> false. diff --git a/lib/compiler/src/genop.tab b/lib/compiler/src/genop.tab index 3a877f2403..02dead9e92 100755 --- a/lib/compiler/src/genop.tab +++ b/lib/compiler/src/genop.tab @@ -1,7 +1,7 @@ # # %CopyrightBegin% # -# Copyright Ericsson AB 1998-2011. All Rights Reserved. +# Copyright Ericsson AB 1998-2018. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -537,3 +537,40 @@ BEAM_FORMAT_NUMBER=0 156: is_map/2 157: has_map_fields/3 158: get_map_elements/3 + +# OTP 20 + +## @spec is_tagged_tuple Lbl Reg N Atom +## @doc Test the type of Reg and jumps to Lbl if it is not a tuple. +## Test the arity of Reg and jumps to Lbl if it is not N. +## Test the first element of the tuple and jumps to Lbl if it is not Atom. +159: is_tagged_tuple/4 + +# OTP 21 + +## @spec build_stacktrace +## @doc Given the raw stacktrace in x(0), build a cooked stacktrace suitable +## for human consumption. Store it in x(0). Destroys all other registers. +## Do a garbage collection if necessary to allocate space on the heap +## for the result. +160: build_stacktrace/0 + +## @spec raw_raise +## @doc This instruction works like the erlang:raise/3 BIF, except that the +## stacktrace in x(2) must be a raw stacktrace. +## x(0) is the class of the exception (error, exit, or throw), +## x(1) is the exception term, and x(2) is the raw stackframe. +## If x(0) is not a valid class, the instruction will not throw an +## exception, but store the atom 'badarg' in x(0) and execute the +## next instruction. +161: raw_raise/0 + +## @spec get_hd Source Head +## @doc Get the head (or car) part of a list (a cons cell) from Source and +## put it into the register Head. +162: get_hd/2 + +## @spec get_tl Source Tail +## @doc Get the tail (or cdr) part of a list (a cons cell) from Source and +## put it into the register Tail. +163: get_tl/2 diff --git a/lib/compiler/src/rec_env.erl b/lib/compiler/src/rec_env.erl index 0e9e12d1ad..48d39776dc 100644 --- a/lib/compiler/src/rec_env.erl +++ b/lib/compiler/src/rec_env.erl @@ -1,8 +1,3 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 2001-2014. All Rights Reserved. -%% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. %% You may obtain a copy of the License at @@ -15,15 +10,12 @@ %% See the License for the specific language governing permissions and %% limitations under the License. %% -%% %CopyrightEnd% -%% -%% @author Richard Carlsson <[email protected]> %% @copyright 1999-2004 Richard Carlsson +%% @author Richard Carlsson <[email protected]> %% @doc Abstract environments, supporting self-referential bindings and %% automatic new-key generation. -%% The current implementation is based on Erlang standard library -%% dictionaries. +%% The current implementation is based on Erlang standard library maps. %%% -define(DEBUG, true). @@ -62,7 +54,7 @@ test_0(Type, N) -> io:fwrite("\ncalls: ~w.\n", [get(new_key_calls)]), io:fwrite("\nretries: ~w.\n", [get(new_key_retries)]), io:fwrite("\nmax: ~w.\n", [get(new_key_max)]), - dict:to_list(element(1,Env)). + maps:to_list(element(1,Env)). test_1(integer = Type, N, Env) when is_integer(N), N > 0 -> Key = new_key(Env), @@ -80,14 +72,13 @@ test_1(_,0, Env) -> %% %% environment() = [Mapping] %% -%% Mapping = {map, Dict} | {rec, Dict, Dict} -%% Dict = dict:dictionary() +%% Mapping = {map, map()} | {rec, map(), map()} %% -%% An empty environment is a list containing a single `{map, Dict}' +%% An empty environment is a list containing a single `{map, map()}' %% element - empty lists are not valid environments. To find a key in an %% environment, it is searched for in each mapping in the list, in %% order, until it the key is found in some mapping, or the end of the -%% list is reached. In a 'rec' mapping, we keep the original dictionary +%% list is reached. In a 'rec' mapping, we keep the original map %% together with a version where entries may have been deleted - this %% makes it possible to garbage collect the entire 'rec' mapping when %% all its entries are unused (for example, by being shadowed by later @@ -97,7 +88,7 @@ test_1(_,0, Env) -> %% ===================================================================== %% @type environment(). An abstract environment. --type mapping() :: {'map', dict:dict()} | {'rec', dict:dict(), dict:dict()}. +-type mapping() :: {'map', map()} | {'rec', map(), map()}. -type environment() :: [mapping(),...]. %% ===================================================================== @@ -108,7 +99,7 @@ test_1(_,0, Env) -> -spec empty() -> environment(). empty() -> - [{map, dict:new()}]. + [{map, #{}}]. %% ===================================================================== @@ -119,14 +110,14 @@ empty() -> -spec is_empty(environment()) -> boolean(). -is_empty([{map, Dict} | Es]) -> - N = dict:size(Dict), +is_empty([{map, Map} | Es]) -> + N = map_size(Map), if N =/= 0 -> false; Es =:= [] -> true; true -> is_empty(Es) end; -is_empty([{rec, Dict, _} | Es]) -> - N = dict:size(Dict), +is_empty([{rec, Map, _} | Es]) -> + N = map_size(Map), if N =/= 0 -> false; Es =:= [] -> true; true -> is_empty(Es) @@ -146,12 +137,12 @@ is_empty([{rec, Dict, _} | Es]) -> size(Env) -> env_size(Env). -env_size([{map, Dict}]) -> - dict:size(Dict); -env_size([{map, Dict} | Env]) -> - dict:size(Dict) + env_size(Env); -env_size([{rec, Dict, _Dict0} | Env]) -> - dict:size(Dict) + env_size(Env). +env_size([{map, Map}]) -> + map_size(Map); +env_size([{map, Map} | Env]) -> + map_size(Map) + env_size(Env); +env_size([{rec, Map, _Map0} | Env]) -> + map_size(Map) + env_size(Env). %% ===================================================================== @@ -165,8 +156,8 @@ env_size([{rec, Dict, _Dict0} | Env]) -> -spec is_defined(term(), environment()) -> boolean(). -is_defined(Key, [{map, Dict} | Env]) -> - case dict:is_key(Key, Dict) of +is_defined(Key, [{map, Map} | Env]) -> + case maps:is_key(Key, Map) of true -> true; false when Env =:= [] -> @@ -174,8 +165,8 @@ is_defined(Key, [{map, Dict} | Env]) -> false -> is_defined(Key, Env) end; -is_defined(Key, [{rec, Dict, _Dict0} | Env]) -> - dict:is_key(Key, Dict) orelse is_defined(Key, Env). +is_defined(Key, [{rec, Map, _Map0} | Env]) -> + maps:is_key(Key, Map) orelse is_defined(Key, Env). %% ===================================================================== @@ -188,12 +179,12 @@ is_defined(Key, [{rec, Dict, _Dict0} | Env]) -> keys(Env) -> lists:sort(keys(Env, [])). -keys([{map, Dict}], S) -> - dict:fetch_keys(Dict) ++ S; -keys([{map, Dict} | Env], S) -> - keys(Env, dict:fetch_keys(Dict) ++ S); -keys([{rec, Dict, _Dict0} | Env], S) -> - keys(Env, dict:fetch_keys(Dict) ++ S). +keys([{map, Map}], S) -> + maps:keys(Map) ++ S; +keys([{map, Map} | Env], S) -> + keys(Env, maps:keys(Map) ++ S); +keys([{rec, Map, _Map0} | Env], S) -> + keys(Env, maps:keys(Map) ++ S). %% ===================================================================== @@ -212,12 +203,12 @@ keys([{rec, Dict, _Dict0} | Env], S) -> to_list(Env) -> lists:sort(to_list(Env, [])). -to_list([{map, Dict}], S) -> - dict:to_list(Dict) ++ S; -to_list([{map, Dict} | Env], S) -> - to_list(Env, dict:to_list(Dict) ++ S); -to_list([{rec, Dict, _Dict0} | Env], S) -> - to_list(Env, dict:to_list(Dict) ++ S). +to_list([{map, Map}], S) -> + maps:to_list(Map) ++ S; +to_list([{map, Map} | Env], S) -> + to_list(Env, maps:to_list(Map) ++ S); +to_list([{rec, Map, _Map0} | Env], S) -> + to_list(Env, maps:to_list(Map) ++ S). %% ===================================================================== @@ -236,12 +227,12 @@ to_list([{rec, Dict, _Dict0} | Env], S) -> -spec bind(term(), term(), environment()) -> environment(). -bind(Key, Value, [{map, Dict}]) -> - [{map, dict:store(Key, Value, Dict)}]; -bind(Key, Value, [{map, Dict} | Env]) -> - [{map, dict:store(Key, Value, Dict)} | delete_any(Key, Env)]; +bind(Key, Value, [{map, Map}]) -> + [{map, maps:put(Key, Value, Map)}]; +bind(Key, Value, [{map, Map} | Env]) -> + [{map, maps:put(Key, Value, Map)} | delete_any(Key, Env)]; bind(Key, Value, Env) -> - [{map, dict:store(Key, Value, dict:new())} | delete_any(Key, Env)]. + [{map, maps:put(Key, Value, #{})} | delete_any(Key, Env)]. %% ===================================================================== @@ -259,17 +250,17 @@ bind(Key, Value, Env) -> -spec bind_list([term()], [term()], environment()) -> environment(). -bind_list(Ks, Vs, [{map, Dict}]) -> - [{map, store_list(Ks, Vs, Dict)}]; -bind_list(Ks, Vs, [{map, Dict} | Env]) -> - [{map, store_list(Ks, Vs, Dict)} | delete_list(Ks, Env)]; +bind_list(Ks, Vs, [{map, Map}]) -> + [{map, store_list(Ks, Vs, Map)}]; +bind_list(Ks, Vs, [{map, Map} | Env]) -> + [{map, store_list(Ks, Vs, Map)} | delete_list(Ks, Env)]; bind_list(Ks, Vs, Env) -> - [{map, store_list(Ks, Vs, dict:new())} | delete_list(Ks, Env)]. + [{map, store_list(Ks, Vs, #{})} | delete_list(Ks, Env)]. -store_list([K | Ks], [V | Vs], Dict) -> - store_list(Ks, Vs, dict:store(K, V, Dict)); -store_list([], _, Dict) -> - Dict. +store_list([K | Ks], [V | Vs], Map) -> + store_list(Ks, Vs, maps:put(K, V, Map)); +store_list([], _, Map) -> + Map. delete_list([K | Ks], Env) -> delete_list(Ks, delete_any(K, Env)); @@ -298,48 +289,40 @@ delete_any(Key, Env) -> -spec delete(term(), environment()) -> environment(). -delete(Key, [{map, Dict} = E | Env]) -> - case dict:is_key(Key, Dict) of - true -> - [{map, dict:erase(Key, Dict)} | Env]; - false -> +delete(Key, [{map, Map} = E | Env]) -> + case maps:take(Key, Map) of + {_, Map1} -> + [{map, Map1} | Env]; + error -> delete_1(Key, Env, E) end; -delete(Key, [{rec, Dict, Dict0} = E | Env]) -> - case dict:is_key(Key, Dict) of - true -> - %% The Dict0 component must be preserved as it is until all - %% keys in Dict have been deleted. - Dict1 = dict:erase(Key, Dict), - case dict:size(Dict1) of - 0 -> - Env; % the whole {rec,...} is now garbage - _ -> - [{rec, Dict1, Dict0} | Env] - end; - false -> +delete(Key, [{rec, Map, Map0} = E | Env]) -> + case maps:take(Key, Map) of + {_, Map1} when map_size(Map1) =:= 0 -> + Env; % the whole {rec,...} is now garbage + %% The Map0 component must be preserved as it is until all + %% keys in Map have been deleted. + {_, Map1} -> + [{rec, Map1, Map0} | Env]; + error -> [E | delete(Key, Env)] end. %% This is just like above, except we pass on the preceding 'map' %% mapping in the list to enable merging when removing 'rec' mappings. -delete_1(Key, [{rec, Dict, Dict0} = E | Env], E1) -> - case dict:is_key(Key, Dict) of - true -> - Dict1 = dict:erase(Key, Dict), - case dict:size(Dict1) of - 0 -> - concat(E1, Env); - _ -> - [E1, {rec, Dict1, Dict0} | Env] - end; - false -> +delete_1(Key, [{rec, Map, Map0} = E | Env], E1) -> + case maps:take(Key, Map) of + {_, Map1} when map_size(Map1) =:= 0 -> + concat(E1, Env); + {_, Map1} -> + [E1, {rec, Map1, Map0} | Env]; + error -> [E1, E | delete(Key, Env)] end. -concat({map, D1}, [{map, D2} | Env]) -> - [dict:merge(fun (_K, V1, _V2) -> V1 end, D1, D2) | Env]; +concat({map, M1}, [{map, M2} | Env]) -> + [maps:merge(M2, M1) | Env]; concat(E1, Env) -> [E1 | Env]. @@ -392,15 +375,15 @@ bind_recursive([], [], _, Env) -> Env; bind_recursive(Ks, Vs, F, Env) -> F1 = fun (V) -> - fun (Dict) -> F(V, [{rec, Dict, Dict} | Env]) end + fun (Map) -> F(V, [{rec, Map, Map} | Env]) end end, - Dict = bind_recursive_1(Ks, Vs, F1, dict:new()), - [{rec, Dict, Dict} | Env]. + Map = bind_recursive_1(Ks, Vs, F1, #{}), + [{rec, Map, Map} | Env]. -bind_recursive_1([K | Ks], [V | Vs], F, Dict) -> - bind_recursive_1(Ks, Vs, F, dict:store(K, F(V), Dict)); -bind_recursive_1([], [], _, Dict) -> - Dict. +bind_recursive_1([K | Ks], [V | Vs], F, Map) -> + bind_recursive_1(Ks, Vs, F, maps:put(K, F(V), Map)); +bind_recursive_1([], [], _, Map) -> + Map. %% ===================================================================== @@ -416,8 +399,8 @@ bind_recursive_1([], [], _, Dict) -> -spec lookup(term(), environment()) -> 'error' | {'ok', term()}. -lookup(Key, [{map, Dict} | Env]) -> - case dict:find(Key, Dict) of +lookup(Key, [{map, Map} | Env]) -> + case maps:find(Key, Map) of {ok, _}=Value -> Value; error when Env =:= [] -> @@ -425,10 +408,10 @@ lookup(Key, [{map, Dict} | Env]) -> error -> lookup(Key, Env) end; -lookup(Key, [{rec, Dict, Dict0} | Env]) -> - case dict:find(Key, Dict) of +lookup(Key, [{rec, Map, Map0} | Env]) -> + case maps:find(Key, Map) of {ok, F} -> - {ok, F(Dict0)}; + {ok, F(Map0)}; error -> lookup(Key, Env) end. @@ -598,7 +581,17 @@ start_range(Env) -> %% (pseudo-)randomly distributed over the range. generate(_N, Range) -> - random:uniform(Range). % works well + %% We must use the same sequence of random variables to ensure + %% that two compilations of the same source code generates the + %% same BEAM code. + case rand:export_seed() of + undefined -> + _ = rand:seed(exsplus, {1,42,2053}), + ok; + _ -> + ok + end, + rand:uniform(Range). % works well %% ===================================================================== diff --git a/lib/compiler/src/sys_core_alias.erl b/lib/compiler/src/sys_core_alias.erl new file mode 100644 index 0000000000..3326c6a2a8 --- /dev/null +++ b/lib/compiler/src/sys_core_alias.erl @@ -0,0 +1,307 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% +%% Purpose : Replace values by aliases from patterns optimisation for Core + +%% Replace expressions by aliases from patterns. For example: +%% +%% example({ok, Val}) -> +%% {ok, Val}. +%% +%% will become: +%% +%% example({ok, Val} = Tuple) -> +%% Tuple. +%% +%% Currently this pass aliases tuple and cons nodes made of literals, +%% variables and other cons. The tuple/cons may appear anywhere in the +%% pattern and it will be aliased if used later on. +%% +%% Notice a tuple/cons made only of literals is not aliased as it may +%% be part of the literal pool. + +-module(sys_core_alias). + +-export([module/2]). + +-include("core_parse.hrl"). + +-define(NOTSET, 0). + +-record(sub, {p=#{} :: #{term() => ?NOTSET | atom()}, %% Found pattern substitutions + v=cerl_sets:new() :: cerl_sets:set(cerl:var_name()), %% Variables used by patterns + t=undefined :: term()}). %% Temporary information from pre to post + +-type sub() :: #sub{}. + +-spec module(cerl:c_module(), [compile:option()]) -> + {'ok',cerl:c_module(),[]}. + +module(#c_module{defs=Ds0}=Mod, _Opts) -> + Ds1 = [def(D) || D <- Ds0], + {ok,Mod#c_module{defs=Ds1},[]}. + +def({#c_var{name={F,Arity}}=Name,B0}) -> + try + put(new_var_num, 0), + {B1,_} = cerl_trees:mapfold(fun pre/2, fun post/2, sub_new(undefined), B0), + erase(new_var_num), + {Name,B1} + catch + Class:Error:Stack -> + io:fwrite("Function: ~w/~w\n", [F,Arity]), + erlang:raise(Class, Error, Stack) + end. + +pre(#c_let{vars=Vars}=Node, Sub) -> + {Node,sub_fold(get_variables(Vars), Sub)}; + +pre(#c_fun{vars=Vars}=Node, Sub) -> + {Node,sub_fold(get_variables(Vars), Sub)}; + +pre(#c_clause{pats=Pats}=Node, Sub0) -> + VarNames = get_variables(Pats), + Sub1 = sub_fold(VarNames, Sub0), + Keys = get_pattern_keys(Pats), + Sub2 = sub_add_keys(Keys, Sub1), + + #sub{v=SubNames,t=Temp} = Sub2, + Sub3 = Sub2#sub{v=merge_variables(VarNames, SubNames), + t={clause,Pats,Keys,SubNames,Temp}}, + + {Node#c_clause{pats=[]},Sub3}; + +pre(Node, Sub0) -> + %% We cache only tuples and cons. + case cerl:is_data(Node) andalso not cerl:is_literal(Node) of + false -> + {Node,Sub0}; + true -> + Kind = cerl:data_type(Node), + Es = cerl:data_es(Node), + case sub_cache_nodes(Kind, Es, Sub0) of + {Name,Sub1} -> + {cerl:ann_c_var(cerl:get_ann(Node), Name),Sub1}; + error -> + {Node,Sub0} + end + end. + +post(#c_let{}=Node, Sub) -> + {Node,sub_unfold(Sub)}; + +post(#c_fun{}=Node, Sub) -> + {Node,sub_unfold(Sub)}; + +post(#c_clause{}=Node, #sub{t={clause,Pats0,Keys,V,T}}=Sub0) -> + {Sub1,PostKeys} = sub_take_keys(Keys, Sub0), + Pats1 = put_pattern_keys(Pats0, PostKeys), + Sub2 = sub_unfold(Sub1#sub{v=V,t=T}), + {Node#c_clause{pats=Pats1},Sub2}; + +post(Node, Sub) -> + {Node,Sub}. + +%% sub_new/1 +%% sub_add_keys/2 +%% sub_take_keys/3 +%% sub_cache_nodes/3 +%% +%% Manages the substitutions record. + +%% Builds a new sub. +-spec sub_new(term()) -> sub(). +sub_new(Temp) -> + #sub{t=Temp}. + +%% Folds the sub into a new one if the variables in nodes are not disjoint +sub_fold(VarNames, #sub{v=SubNames}=Sub) -> + case is_disjoint_variables(VarNames, SubNames) of + true -> Sub#sub{t={temp,Sub#sub.t}}; + false -> sub_new({sub,Sub}) + end. + +%% Unfolds the sub in case one was folded in the previous step +sub_unfold(#sub{t={temp,Temp}}=Sub) -> + Sub#sub{t=Temp}; +sub_unfold(#sub{t={sub,Sub}}) -> + Sub. + +%% Adds the keys extracted from patterns to the state. +-spec sub_add_keys([term()], sub()) -> sub(). +sub_add_keys(Keys, #sub{p=Pat0}=Sub) -> + Pat1 = + lists:foldl(fun(Key, Acc) -> + false = maps:is_key(Key, Acc), %Assertion. + maps:put(Key, ?NOTSET, Acc) + end, Pat0, Keys), + Sub#sub{p=Pat1}. + +%% Take the keys from the map taking into account the keys +%% that have changed as those must become aliases in the pattern. +-spec sub_take_keys([term()], sub()) -> {sub(), [{term(), atom()}]}. +sub_take_keys(Keys, #sub{p=Pat0}=Sub) -> + {Pat1,Acc} = sub_take_keys(Keys, Pat0, []), + {Sub#sub{p=Pat1},Acc}. + +sub_take_keys([K|T], Sub0, Acc) -> + case maps:take(K, Sub0) of + {?NOTSET,Sub1} -> + sub_take_keys(T, Sub1, Acc); + {Name,Sub1} -> + sub_take_keys(T, Sub1, [{K,Name}|Acc]) + end; +sub_take_keys([], Sub, Acc) -> + {Sub,Acc}. + +%% Check if the node can be cached based on the state information. +%% If it can be cached and it does not have an alias for it, we +%% build one. +-spec sub_cache_nodes(atom(), [cerl:cerl()], sub()) -> {atom(), sub()} | error. +sub_cache_nodes(Kind, Nodes, #sub{p=Pat}=Sub) -> + case nodes_to_key(Kind, Nodes) of + {ok, Key} -> + case Pat of + #{Key := ?NOTSET} -> + new_var_name(Key, Sub); + #{Key := Name} -> + {Name,Sub}; + #{} -> + error + end; + error -> + error + end. + +new_var_name(Key, #sub{p=Pat}=Sub) -> + Counter = get(new_var_num), + Name = list_to_atom("@r" ++ integer_to_list(Counter)), + put(new_var_num, Counter + 1), + {Name,Sub#sub{p=maps:put(Key, Name, Pat)}}. + +%% get_variables/1 +%% is_disjoint_variables/2 +%% merge_variables/2 + +get_variables(NodesList) -> + cerl_sets:from_list([Var || Node <- NodesList, Var <- cerl_trees:variables(Node)]). + +is_disjoint_variables(Vars1, Vars2) -> + cerl_sets:is_disjoint(Vars1, Vars2). + +merge_variables(Vars1, Vars2) -> + cerl_sets:union(Vars1, Vars2). + +%% get_pattern_keys/2 +%% put_pattern_keys/2 +%% +%% Gets keys from patterns or add them as aliases. + +get_pattern_keys(Patterns) -> + lists:foldl(fun get_pattern_keys/2, [], Patterns). + +get_pattern_keys(#c_tuple{es=Es}, Acc0) -> + Acc1 = accumulate_pattern_keys(tuple, Es, Acc0), + lists:foldl(fun get_pattern_keys/2, Acc1, Es); +get_pattern_keys(#c_cons{hd=Hd,tl=Tl}, Acc0) -> + Acc1 = accumulate_pattern_keys(cons, [Hd, Tl], Acc0), + get_pattern_keys(Tl, get_pattern_keys(Hd, Acc1)); +get_pattern_keys(#c_alias{pat=Pat}, Acc0) -> + get_pattern_keys(Pat, Acc0); +get_pattern_keys(#c_map{es=Es}, Acc0) -> + lists:foldl(fun get_pattern_keys/2, Acc0, Es); +get_pattern_keys(#c_map_pair{val=Val}, Acc0) -> + get_pattern_keys(Val, Acc0); +get_pattern_keys(_, Acc) -> + Acc. + +accumulate_pattern_keys(Kind, Nodes, Acc) -> + case nodes_to_key(Kind, Nodes) of + {ok,Key} -> [Key|Acc]; + error -> Acc + end. + +put_pattern_keys(Patterns, []) -> + Patterns; +put_pattern_keys(Patterns, Keys) -> + {NewPatterns,Map} = + lists:mapfoldl(fun alias_pattern_keys/2, maps:from_list(Keys), Patterns), + %% Check all aliases have been consumed from the map. + 0 = map_size(Map), + NewPatterns. + +alias_pattern_keys(#c_tuple{anno=Anno,es=Es0}=Node, Acc0) -> + {Es1,Acc1} = lists:mapfoldl(fun alias_pattern_keys/2, Acc0, Es0), + nodes_to_alias(tuple, Es0, Anno, Node#c_tuple{es=Es1}, Acc1); +alias_pattern_keys(#c_cons{anno=Anno,hd=Hd0,tl=Tl0}=Node, Acc0) -> + {Hd1,Acc1} = alias_pattern_keys(Hd0, Acc0), + {Tl1,Acc2} = alias_pattern_keys(Tl0, Acc1), + nodes_to_alias(cons, [Hd0, Tl0], Anno, Node#c_cons{hd=Hd1,tl=Tl1}, Acc2); +alias_pattern_keys(#c_alias{pat=Pat0}=Node, Acc0) -> + {Pat1,Acc1} = alias_pattern_keys(Pat0, Acc0), + {Node#c_alias{pat=Pat1}, Acc1}; +alias_pattern_keys(#c_map{es=Es0}=Node, Acc0) -> + {Es1,Acc1} = lists:mapfoldl(fun alias_pattern_keys/2, Acc0, Es0), + {Node#c_map{es=Es1}, Acc1}; +alias_pattern_keys(#c_map_pair{val=Val0}=Node, Acc0) -> + {Val1,Acc1} = alias_pattern_keys(Val0, Acc0), + {Node#c_map_pair{val=Val1}, Acc1}; +alias_pattern_keys(Pattern, Acc) -> + {Pattern,Acc}. + +%% Check if a node must become an alias because +%% its pattern was used later on as an expression. +nodes_to_alias(Kind, Inner, Anno, Node, Keys0) -> + case nodes_to_key(Kind, Inner) of + {ok,Key} -> + case maps:take(Key, Keys0) of + {Name,Keys1} -> + Var = cerl:ann_c_var(Anno, Name), + {cerl:ann_c_alias(Anno, Var, Node), Keys1}; + error -> + {Node,Keys0} + end; + error -> + {Node,Keys0} + end. + +%% Builds the key used to check if a value can be +%% replaced by an alias. It considers literals, +%% aliases, variables, tuples and cons recursively. +nodes_to_key(Kind, Nodes) -> + nodes_to_key(Nodes, [], Kind). + +nodes_to_key([#c_alias{var=Var}|T], Acc, Kind) -> + nodes_to_key([Var|T], Acc, Kind); +nodes_to_key([#c_var{name=Name}|T], Acc, Kind) -> + nodes_to_key(T, [[var,Name]|Acc], Kind); +nodes_to_key([Node|T], Acc0, Kind) -> + case cerl:is_data(Node) of + false -> + error; + true -> + case nodes_to_key(cerl:data_es(Node), [], cerl:data_type(Node)) of + {ok,Key} -> + nodes_to_key(T, [Key|Acc0], Kind); + error -> + error + end + end; +nodes_to_key([], Acc, Kind) -> + {ok,[Kind|Acc]}. diff --git a/lib/compiler/src/sys_core_bsm.erl b/lib/compiler/src/sys_core_bsm.erl new file mode 100644 index 0000000000..62657933ee --- /dev/null +++ b/lib/compiler/src/sys_core_bsm.erl @@ -0,0 +1,355 @@ +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 2017-2018. All Rights Reserved. +%% +%% Licensed under the Apache License, Version 2.0 (the "License"); +%% you may not use this file except in compliance with the License. +%% You may obtain a copy of the License at +%% +%% http://www.apache.org/licenses/LICENSE-2.0 +%% +%% Unless required by applicable law or agreed to in writing, software +%% distributed under the License is distributed on an "AS IS" BASIS, +%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +%% See the License for the specific language governing permissions and +%% limitations under the License. +%% +%% %CopyrightEnd% +%% +%% Purpose : Optimize bit syntax matching. + + +-module(sys_core_bsm). +-export([module/2,format_error/1]). + +-include("core_parse.hrl"). +-import(lists, [member/2,reverse/1,usort/1]). + +-spec module(cerl:c_module(), [compile:option()]) -> {'ok', cerl:c_module()}. + +module(#c_module{defs=Ds0}=Mod, Opts) -> + {Ds,Ws0} = function(Ds0, [], []), + case member(bin_opt_info, Opts) of + false -> + {ok,Mod#c_module{defs=Ds}}; + true -> + Ws1 = [make_warning(Where, What) || {Where,What} <- Ws0], + Ws = usort(Ws1), + {ok,Mod#c_module{defs=Ds},Ws} + end. + +function([{#c_var{name={F,Arity}}=Name,B0}|Fs], FsAcc, Ws0) -> + try cerl_trees:mapfold(fun bsm_an/2, Ws0, B0) of + {B,Ws} -> + function(Fs, [{Name,B}|FsAcc], Ws) + catch + throw:unsafe_bs_context_to_binary -> + %% Unsafe bs_context_to_binary (in the sense that the + %% contents of the binary will probably be wrong). + %% Disable binary optimizations for the entire function. + %% We don't generate an INFO message, because this happens + %% very infrequently and it would be hard to explain in + %% a comprehensible way in an INFO message. + function(Fs, [{Name,B0}|FsAcc], Ws0); + Class:Error:Stack -> + io:fwrite("Function: ~w/~w\n", [F,Arity]), + erlang:raise(Class, Error, Stack) + end; +function([], Fs, Ws) -> + {reverse(Fs),Ws}. + +-type error() :: atom(). +-spec format_error(error()) -> nonempty_string(). + +format_error(bin_opt_alias) -> + "INFO: the '=' operator will prevent delayed sub binary optimization"; +format_error(bin_partition) -> + "INFO: matching non-variables after a previous clause matching a variable " + "will prevent delayed sub binary optimization"; +format_error(bin_var_used) -> + "INFO: using a matched out sub binary will prevent " + "delayed sub binary optimization"; +format_error(orig_bin_var_used_in_guard) -> + "INFO: using the original binary variable in a guard will prevent " + "delayed sub binary optimization"; +format_error(bin_var_used_in_guard) -> + "INFO: using a matched out sub binary in a guard will prevent " + "delayed sub binary optimization". + + +%%% +%%% Annotate bit syntax matching to faciliate optimization in further passes. +%%% + +bsm_an(Core0, Ws0) -> + case bsm_an(Core0) of + {ok,Core} -> + {Core,Ws0}; + {ok,Core,W} -> + {Core,[W|Ws0]} + end. + +bsm_an(#c_case{arg=#c_var{}=V}=Case) -> + bsm_an_1([V], Case); +bsm_an(#c_case{arg=#c_values{es=Es}}=Case) -> + bsm_an_1(Es, Case); +bsm_an(Other) -> + {ok,Other}. + +bsm_an_1(Vs0, #c_case{clauses=Cs0}=Case) -> + case bsm_leftmost(Cs0) of + none -> + {ok,Case}; + 1 -> + bsm_an_2(Vs0, Cs0, Case); + Pos -> + Vs = move_from_col(Pos, Vs0), + Cs = [C#c_clause{pats=move_from_col(Pos, Ps)} || + #c_clause{pats=Ps}=C <- Cs0], + bsm_an_2(Vs, Cs, Case) + end. + +bsm_an_2(Vs, Cs, Case) -> + try + bsm_ensure_no_partition(Cs), + {ok,bsm_do_an(Vs, Cs, Case)} + catch + throw:{problem,Where,What} -> + {ok,Case,{Where,What}} + end. + +move_from_col(Pos, L) -> + {First,[Col|Rest]} = lists:split(Pos - 1, L), + [Col|First] ++ Rest. + +bsm_do_an([#c_var{name=Vname}=V0|Vs0], Cs0, Case) -> + bsm_inner_context_to_binary(Cs0), + Cs = bsm_do_an_var(Vname, Cs0), + V = bsm_annotate_for_reuse(V0), + Vs = core_lib:make_values([V|Vs0]), + Case#c_case{arg=Vs,clauses=Cs}; +bsm_do_an(_Vs, _Cs, Case) -> Case. + +bsm_inner_context_to_binary([#c_clause{body=B}|Cs]) -> + %% Consider: + %% + %% foo(<<Length, Data/binary>>) -> %Line 1 + %% case {Data, Length} of %Line 2 + %% {_, 0} -> Data; %Line 3 + %% {<<...>>, 4} -> ... %Line 4 + %% end. + %% + %% No sub binary will be created for Data in line 1. The match + %% context will be passed on to the `case` in line 2. In line 3, + %% this pass inserts a `bs_context_to_binary` instruction to + %% convert the match context representing Data to a binary before + %% returning it. The problem is that the binary created will be + %% the original binary (including the matched out Length field), + %% not the tail of the binary as it is supposed to be. + %% + %% Here follows a heuristic to disable the binary optimizations + %% for the entire function if this code kind of code is found. + + case cerl_trees:free_variables(B) of + [] -> + %% Since there are no free variables in the body of + %% this clause, there can't be any troublesome + %% bs_context_to_binary instructions. + bsm_inner_context_to_binary(Cs); + [_|_]=Free -> + %% One of the free variables could refer to a match context + %% created by the outer binary match. + F = fun(#c_primop{name=#c_literal{val=bs_context_to_binary}, + args=[#c_var{name=V}]}, _) -> + case member(V, Free) of + true -> + %% This bs_context_to_binary instruction will + %% make a binary of the match context from an + %% outer binary match. It is very likely that + %% the contents of the binary will be wrong + %% (the original binary as opposed to only + %% the tail binary). + throw(unsafe_bs_context_to_binary); + false -> + %% Safe. This bs_context_to_binary instruction + %% will make a binary from a match context + %% defined in the body of the clause. + ok + end; + (_, _) -> + ok + end, + cerl_trees:fold(F, ok, B) + end; +bsm_inner_context_to_binary([]) -> ok. + +bsm_do_an_var(V, [#c_clause{pats=[P|_],guard=G,body=B0}=C0|Cs]) -> + case P of + #c_var{name=VarName} -> + case core_lib:is_var_used(V, G) of + true -> bsm_problem(C0, orig_bin_var_used_in_guard); + false -> ok + end, + case core_lib:is_var_used(VarName, G) of + true -> bsm_problem(C0, bin_var_used_in_guard); + false -> ok + end, + B1 = bsm_maybe_ctx_to_binary(VarName, B0), + B = bsm_maybe_ctx_to_binary(V, B1), + C = C0#c_clause{body=B}, + [C|bsm_do_an_var(V, Cs)]; + #c_alias{} -> + case bsm_could_match_binary(P) of + false -> + [C0|bsm_do_an_var(V, Cs)]; + true -> + bsm_problem(C0, bin_opt_alias) + end; + _ -> + case bsm_could_match_binary(P) andalso bsm_is_var_used(V, G, B0) of + false -> + [C0|bsm_do_an_var(V, Cs)]; + true -> + bsm_problem(C0, bin_var_used) + end + end; +bsm_do_an_var(_, []) -> []. + +bsm_annotate_for_reuse(#c_var{anno=Anno}=Var) -> + Var#c_var{anno=[reuse_for_context|Anno]}. + +bsm_is_var_used(V, G, B) -> + core_lib:is_var_used(V, G) orelse core_lib:is_var_used(V, B). + +bsm_maybe_ctx_to_binary(V, B) -> + case core_lib:is_var_used(V, B) andalso not previous_ctx_to_binary(V, B) of + false -> + B; + true -> + #c_seq{arg=#c_primop{name=#c_literal{val=bs_context_to_binary}, + args=[#c_var{name=V}]}, + body=B} + end. + +previous_ctx_to_binary(V, Core) -> + case Core of + #c_seq{arg=#c_primop{name=#c_literal{val=bs_context_to_binary}, + args=[#c_var{name=V}]}} -> + true; + _ -> + false + end. + +%% bsm_leftmost(Cs) -> none | ArgumentNumber +%% Find the leftmost argument that matches a nonempty binary. +%% Return either 'none' or the argument number (1-N). + +bsm_leftmost(Cs) -> + bsm_leftmost_1(Cs, none). + +bsm_leftmost_1([_|_], 1) -> + 1; +bsm_leftmost_1([#c_clause{pats=Ps}|Cs], Pos) -> + bsm_leftmost_2(Ps, Cs, 1, Pos); +bsm_leftmost_1([], Pos) -> Pos. + +bsm_leftmost_2(_, Cs, Pos, Pos) -> + bsm_leftmost_1(Cs, Pos); +bsm_leftmost_2([#c_binary{segments=[_|_]}|_], Cs, N, _) -> + bsm_leftmost_1(Cs, N); +bsm_leftmost_2([_|Ps], Cs, N, Pos) -> + bsm_leftmost_2(Ps, Cs, N+1, Pos); +bsm_leftmost_2([], Cs, _, Pos) -> + bsm_leftmost_1(Cs, Pos). + +%% bsm_ensure_no_partition(Cs) -> ok (exception if problem) +%% There must only be a single bs_start_match2 instruction if we +%% are to reuse the binary variable for the match context. +%% +%% To make sure that there is only a single bs_start_match2 +%% instruction, we will check for partitions such as: +%% +%% foo(<<...>>) -> ... +%% foo(<Variable>) when ... -> ... +%% foo(<Non-variable pattern>) -> +%% +%% If there is such partition, we reject the optimization. + +bsm_ensure_no_partition(Cs) -> + bsm_ensure_no_partition_1(Cs, before). + +%% Loop through each clause. +bsm_ensure_no_partition_1([#c_clause{pats=Ps,guard=G}|Cs], State0) -> + State = bsm_ensure_no_partition_2(Ps, G, State0), + case State of + 'after' -> + bsm_ensure_no_partition_after(Cs); + _ -> + ok + end, + bsm_ensure_no_partition_1(Cs, State); +bsm_ensure_no_partition_1([], _) -> ok. + +bsm_ensure_no_partition_2([#c_binary{}|_], _, _State) -> + within; +bsm_ensure_no_partition_2([#c_alias{}=Alias|_], N, State) -> + %% Retrieve the real pattern that the alias refers to and check that. + P = bsm_real_pattern(Alias), + bsm_ensure_no_partition_2([P], N, State); +bsm_ensure_no_partition_2([_|_], _, before=State) -> + %% No binary matching yet - therefore no partition. + State; +bsm_ensure_no_partition_2([P|_], _, State) -> + case bsm_could_match_binary(P) of + false -> + State; + true -> + %% The pattern P *may* match a binary, so we must update the state. + %% (P must be a variable.) + 'after' + end. + +bsm_ensure_no_partition_after([#c_clause{pats=Ps}=C|Cs]) -> + case Ps of + [#c_var{}|_] -> + bsm_ensure_no_partition_after(Cs); + _ -> + bsm_problem(C, bin_partition) + end; +bsm_ensure_no_partition_after([]) -> ok. + +bsm_could_match_binary(#c_alias{pat=P}) -> bsm_could_match_binary(P); +bsm_could_match_binary(#c_cons{}) -> false; +bsm_could_match_binary(#c_tuple{}) -> false; +bsm_could_match_binary(#c_literal{val=Lit}) -> is_bitstring(Lit); +bsm_could_match_binary(_) -> true. + +bsm_real_pattern(#c_alias{pat=P}) -> bsm_real_pattern(P); +bsm_real_pattern(P) -> P. + +bsm_problem(Where, What) -> + throw({problem,Where,What}). + +make_warning(Core, Term) -> + case should_suppress_warning(Core) of + true -> + ok; + false -> + Anno = cerl:get_ann(Core), + Line = get_line(Anno), + File = get_file(Anno), + {File,[{Line,?MODULE,Term}]} + end. + +should_suppress_warning(Core) -> + Ann = cerl:get_ann(Core), + member(compiler_generated, Ann). + +get_line([Line|_]) when is_integer(Line) -> Line; +get_line([_|T]) -> get_line(T); +get_line([]) -> none. + +get_file([{file,File}|_]) -> File; +get_file([_|T]) -> get_file(T); +get_file([]) -> "no_file". % should not happen diff --git a/lib/compiler/src/sys_core_dsetel.erl b/lib/compiler/src/sys_core_dsetel.erl index ac32db10fe..9ab83c210f 100644 --- a/lib/compiler/src/sys_core_dsetel.erl +++ b/lib/compiler/src/sys_core_dsetel.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2002-2010. All Rights Reserved. +%% Copyright Ericsson AB 2002-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -72,7 +72,7 @@ module(M0, _Options) -> {ok,M}. visit_module(#c_module{defs=Ds0}=R) -> - Env = dict:new(), + Env = #{}, Ds = visit_module_1(Ds0, Env, []), R#c_module{defs=Ds}. @@ -81,8 +81,7 @@ visit_module_1([{Name,F0}|Fs], Env, Acc) -> {F,_} -> visit_module_1(Fs, Env, [{Name,F}|Acc]) catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> #c_var{name={Func,Arity}} = Name, io:fwrite("Function: ~w/~w\n", [Func,Arity]), erlang:raise(Class, Error, Stack) @@ -95,9 +94,11 @@ visit(Env, #c_var{name={_,_}}=R) -> {R, Env}; visit(Env0, #c_var{name=X}=R) -> %% There should not be any free variables. If there are, - %% the next line will cause an exception. - {ok, N} = dict:find(X, Env0), - {R, dict:store(X, N+1, Env0)}; + %% the case will fail with an exception. + case Env0 of + #{X:=N} -> + {R, Env0#{X:=N+1}} + end; visit(Env, #c_literal{}=R) -> {R, Env}; visit(Env0, #c_tuple{es=Es0}=R) -> @@ -203,7 +204,7 @@ bind_vars(Vs, Env) -> bind_vars(Vs, Env, []). bind_vars([#c_var{name=X}|Vs], Env0, Xs)-> - bind_vars(Vs, dict:store(X, 0, Env0), [X|Xs]); + bind_vars(Vs, Env0#{X=>0}, [X|Xs]); bind_vars([], Env,Xs) -> {Xs, Env}. @@ -217,7 +218,7 @@ visit_pats([], Env, Vs) -> {Vs, Env}. visit_pat(Env0, #c_var{name=V}, Vs) -> - {[V|Vs], dict:store(V, 0, Env0)}; + {[V|Vs], Env0#{V=>0}}; visit_pat(Env0, #c_tuple{es=Es}, Vs) -> visit_pats(Es, Env0, Vs); visit_pat(Env0, #c_map{es=Es}, Vs) -> @@ -235,23 +236,25 @@ visit_pat(Env0, #c_bitstr{val=Val,size=Sz}, Vs0) -> case Sz of #c_var{name=V} -> %% We don't tolerate free variables. - {ok, N} = dict:find(V, Env0), - {Vs0, dict:store(V, N+1, Env0)}; + case Env0 of + #{V:=N} -> + {Vs0, Env0#{V:=N+1}} + end; _ -> visit_pat(Env0, Sz, Vs0) end, visit_pat(Env1, Val, Vs1); visit_pat(Env0, #c_alias{pat=P,var=#c_var{name=V}}, Vs) -> - visit_pat(dict:store(V, 0, Env0), P, [V|Vs]); + visit_pat(Env0#{V=>0}, P, [V|Vs]); visit_pat(Env, #c_literal{}, Vs) -> {Vs, Env}. restore_vars([V|Vs], Env0, Env1) -> - case dict:find(V, Env0) of - {ok, N} -> - restore_vars(Vs, Env0, dict:store(V, N, Env1)); - error -> - restore_vars(Vs, Env0, dict:erase(V, Env1)) + case Env0 of + #{V:=N} -> + restore_vars(Vs, Env0, Env1#{V=>N}); + _ -> + restore_vars(Vs, Env0, maps:remove(V, Env1)) end; restore_vars([], _, Env1) -> Env1. @@ -349,8 +352,8 @@ is_safe(#c_literal{}) -> true; is_safe(_) -> false. is_single_use(V, Env) -> - case dict:find(V, Env) of - {ok, 1} -> + case Env of + #{V:=1} -> true; _ -> false diff --git a/lib/compiler/src/sys_core_fold.erl b/lib/compiler/src/sys_core_fold.erl index 65699ccda9..3a65b40fa5 100644 --- a/lib/compiler/src/sys_core_fold.erl +++ b/lib/compiler/src/sys_core_fold.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2013. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -71,7 +71,7 @@ -export([module/2,format_error/1]). -import(lists, [map/2,foldl/3,foldr/3,mapfoldl/3,all/2,any/2, - reverse/1,reverse/2,member/2,nth/2,flatten/1, + reverse/1,reverse/2,member/2,flatten/1, unzip/1,keyfind/3]). -import(cerl, [ann_c_cons/3,ann_c_map/3,ann_c_tuple/2]). @@ -83,10 +83,11 @@ -ifdef(DEBUG). -define(ASSERT(E), case E of - true -> ok; + true -> + ok; false -> io:format("~p, line ~p: assertion failed\n", [?MODULE,?LINE]), - exit(assertion_failed) + error(assertion_failed) end). -else. -define(ASSERT(E), ignore). @@ -106,29 +107,42 @@ {'ok', cerl:c_module(), [_]}. module(#c_module{defs=Ds0}=Mod, Opts) -> - put(bin_opt_info, member(bin_opt_info, Opts)), put(no_inline_list_funcs, not member(inline_list_funcs, Opts)), - case get(new_var_num) of - undefined -> put(new_var_num, 0); - _ -> ok - end, init_warnings(), Ds1 = [function_1(D) || D <- Ds0], + erase(new_var_num), erase(no_inline_list_funcs), - erase(bin_opt_info), {ok,Mod#c_module{defs=Ds1},get_warnings()}. function_1({#c_var{name={F,Arity}}=Name,B0}) -> try - B = expr(B0, value, sub_new()), %This must be a fun! + %% Find a suitable starting value for the variable + %% counter. Note that this pass assumes that new_var_name/1 + %% returns a variable name distinct from any variable used in + %% the entire body of the function. We use integers as + %% variable names to avoid filling up the atom table when + %% compiling huge functions. + Count = cerl_trees:next_free_variable_name(B0), + put(new_var_num, Count), + B = find_fixpoint(fun(Core) -> + %% This must be a fun! + expr(Core, value, sub_new()) + end, B0, 20), {Name,B} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [F,Arity]), erlang:raise(Class, Error, Stack) end. +find_fixpoint(_OptFun, Core, 0) -> + Core; +find_fixpoint(OptFun, Core0, Max) -> + case OptFun(Core0) of + Core0 -> Core0; + Core -> find_fixpoint(OptFun, Core, Max-1) + end. + %% body(Expr, Sub) -> Expr. %% body(Expr, Context, Sub) -> Expr. %% No special handling of anything except values. @@ -136,14 +150,9 @@ function_1({#c_var{name={F,Arity}}=Name,B0}) -> body(Body, Sub) -> body(Body, value, Sub). -body(#c_values{anno=A,es=Es0}, Ctxt, Sub) -> - Es1 = expr_list(Es0, Ctxt, Sub), - case Ctxt of - value -> - #c_values{anno=A,es=Es1}; - effect -> - make_effect_seq(Es1, Sub) - end; +body(#c_values{anno=A,es=Es0}, value, Sub) -> + Es1 = expr_list(Es0, value, Sub), + #c_values{anno=A,es=Es1}; body(E, Ctxt, Sub) -> ?ASSERT(verify_scope(E, Sub)), expr(E, Ctxt, Sub). @@ -160,13 +169,23 @@ guard(Expr, Sub) -> %% opt_guard_try(#c_seq{arg=Arg,body=Body0}=Seq) -> Body = opt_guard_try(Body0), - case {Arg,Body} of - {#c_call{module=#c_literal{val=Mod}, - name=#c_literal{val=Name}, - args=Args},#c_literal{val=false}} -> + WillFail = case Body of + #c_call{module=#c_literal{val=erlang}, + name=#c_literal{val=error}, + args=[_]} -> + true; + #c_literal{val=false} -> + true; + _ -> + false + end, + case Arg of + #c_call{module=#c_literal{val=Mod}, + name=#c_literal{val=Name}, + args=Args} when WillFail -> %% We have sequence consisting of a call (evaluated %% for a possible exception and/or side effect only), - %% followed by 'false'. + %% followed by 'false' or a call to error/1. %% Since the sequence is inside a try block that will %% default to 'false' if any exception occurs, not %% evalutating the call will not change the behaviour @@ -181,13 +200,15 @@ opt_guard_try(#c_seq{arg=Arg,body=Body0}=Seq) -> %% be safely removed. Body end; - {_,_} -> + _ -> Seq#c_seq{body=Body} end; opt_guard_try(#c_case{clauses=Cs}=Term) -> Term#c_case{clauses=opt_guard_try_list(Cs)}; opt_guard_try(#c_clause{body=B0}=Term) -> Term#c_clause{body=opt_guard_try(B0)}; +opt_guard_try(#c_let{vars=[],arg=#c_values{es=[]},body=B}) -> + B; opt_guard_try(#c_let{arg=Arg,body=B0}=Term) -> case opt_guard_try(B0) of #c_literal{}=B -> @@ -239,7 +260,7 @@ expr(#c_cons{anno=Anno,hd=H0,tl=T0}=Cons, Ctxt, Sub) -> case Ctxt of effect -> add_warning(Cons, useless_building), - expr(make_effect_seq([H1,T1], Sub), Ctxt, Sub); + make_effect_seq([H1,T1], Sub); value -> ann_c_cons(Anno, H1, T1) end; @@ -248,7 +269,7 @@ expr(#c_tuple{anno=Anno,es=Es0}=Tuple, Ctxt, Sub) -> case Ctxt of effect -> add_warning(Tuple, useless_building), - expr(make_effect_seq(Es, Sub), Ctxt, Sub); + make_effect_seq(Es, Sub); value -> ann_c_tuple(Anno, Es) end; @@ -257,7 +278,7 @@ expr(#c_map{anno=Anno,arg=V0,es=Es0}=Map, Ctxt, Sub) -> case Ctxt of effect -> add_warning(Map, useless_building), - expr(make_effect_seq(Es, Sub), Ctxt, Sub); + make_effect_seq(Es, Sub); value -> V = expr(V0, Ctxt, Sub), ann_c_map(Anno,V,Es) @@ -277,7 +298,7 @@ expr(#c_fun{}=Fun, effect, _) -> add_warning(Fun, useless_building), void(); expr(#c_fun{vars=Vs0,body=B0}=Fun, Ctxt0, Sub0) -> - {Vs1,Sub1} = pattern_list(Vs0, Sub0), + {Vs1,Sub1} = var_list(Vs0, Sub0), Ctxt = case Ctxt0 of {letrec,Ctxt1} -> Ctxt1; value -> value @@ -294,9 +315,15 @@ expr(#c_seq{arg=Arg0,body=B0}=Seq0, Ctxt, Sub) -> false -> %% Arg cannot be "values" here - only a single value %% make sense here. - case is_safe_simple(Arg, Sub) of - true -> B1; - false -> Seq0#c_seq{arg=Arg,body=B1} + case {Ctxt,is_safe_simple(Arg, Sub)} of + {effect,true} -> B1; + {effect,false} -> + case is_safe_simple(B1, Sub) of + true -> Arg; + false -> Seq0#c_seq{arg=Arg,body=B1} + end; + {value,true} -> B1; + {value,false} -> Seq0#c_seq{arg=Arg,body=B1} end end; expr(#c_let{}=Let0, Ctxt, Sub) -> @@ -310,7 +337,7 @@ expr(#c_let{}=Let0, Ctxt, Sub) -> Expr -> %% The let body was successfully moved into the let argument. %% Now recursively re-process the new expression. - expr(Expr, Ctxt, sub_new_preserve_types(Sub)) + Expr end; expr(#c_letrec{body=#c_var{}}=Letrec, effect, _Sub) -> %% This is named fun in an 'effect' context. Warn and ignore. @@ -318,7 +345,12 @@ expr(#c_letrec{body=#c_var{}}=Letrec, effect, _Sub) -> void(); expr(#c_letrec{defs=Fs0,body=B0}=Letrec, Ctxt, Sub) -> Fs1 = map(fun ({Name,Fb}) -> - {Name,expr(Fb, {letrec,Ctxt}, Sub)} + case Ctxt =:= effect andalso is_fun_effect_safe(Name, B0) of + true -> + {Name,expr(Fb, {letrec, effect}, Sub)}; + false -> + {Name,expr(Fb, {letrec, value}, Sub)} + end end, Fs0), B1 = body(B0, Ctxt, Sub), Letrec#c_letrec{defs=Fs1,body=B1}; @@ -351,7 +383,7 @@ expr(#c_case{}=Case0, Ctxt, Sub) -> %% (in addition to any warnings that may have been emitted %% according to the rules above). %% - case opt_bool_case(Case0) of + case opt_bool_case(Case0, Sub) of #c_case{arg=Arg0,clauses=Cs0}=Case1 -> Arg1 = body(Arg0, value, Sub), LitExpr = cerl:is_literal(Arg1), @@ -360,12 +392,7 @@ expr(#c_case{}=Case0, Ctxt, Sub) -> Case = Case1#c_case{arg=Arg2,clauses=Cs2}, warn_no_clause_match(Case1, Case), Expr = eval_case(Case, Sub), - case move_case_into_arg(Case, Sub) of - impossible -> - bsm_an(Expr); - Other -> - expr(Other, Ctxt, sub_new_preserve_types(Sub)) - end; + move_case_into_arg(Expr, Sub); Other -> expr(Other, Ctxt, Sub) end; @@ -374,10 +401,22 @@ expr(#c_receive{clauses=Cs0,timeout=T0,action=A0}=Recv, Ctxt, Sub) -> T1 = expr(T0, value, Sub), A1 = body(A0, Ctxt, Sub), Recv#c_receive{clauses=Cs1,timeout=T1,action=A1}; -expr(#c_apply{op=Op0,args=As0}=App, _, Sub) -> +expr(#c_apply{anno=Anno,op=Op0,args=As0}=Apply0, _, Sub) -> Op1 = expr(Op0, value, Sub), As1 = expr_list(As0, value, Sub), - App#c_apply{op=Op1,args=As1}; + case cerl:is_data(Op1) andalso not is_literal_fun(Op1) of + false -> + Apply = Apply0#c_apply{op=Op1,args=As1}, + fold_apply(Apply, Op1, As1); + true -> + add_warning(Apply0, invalid_call), + Err = #c_call{anno=Anno, + module=#c_literal{val=erlang}, + name=#c_literal{val=error}, + args=[#c_tuple{es=[#c_literal{val='badfun'}, + Op1]}]}, + make_effect_seq(As1++[Err], Sub) + end; expr(#c_call{module=M0,name=N0}=Call0, Ctxt, Sub) -> M1 = expr(M0, value, Sub), N1 = expr(N0, value, Sub), @@ -386,9 +425,20 @@ expr(#c_call{module=M0,name=N0}=Call0, Ctxt, Sub) -> no -> call(Call, M1, N1, Sub); {yes,Seq} -> expr(Seq, Ctxt, Sub) end; +expr(#c_primop{name=#c_literal{val=build_stacktrace}}, effect, _Sub) -> + void(); expr(#c_primop{args=As0}=Prim, _, Sub) -> As1 = expr_list(As0, value, Sub), Prim#c_primop{args=As1}; +expr(#c_catch{anno=Anno,body=B}, effect, Sub) -> + %% When the return value of the 'catch' is ignored, we can replace it + %% with a try/catch to avoid building a stack trace when an exception + %% occurs. + Var = #c_var{name='catch_value'}, + Evs = [#c_var{name='Class'},#c_var{name='Reason'},#c_var{name='Stk'}], + Try = #c_try{anno=Anno,arg=B,vars=[Var],body=Var, + evars=Evs,handler=void()}, + expr(Try, effect, Sub); expr(#c_catch{body=B0}=Catch, _, Sub) -> %% We can remove catch if the value is simple B1 = body(B0, value, Sub), @@ -420,17 +470,97 @@ expr(#c_try{anno=A,arg=E0,vars=Vs0,body=B0,evars=Evs0,handler=H0}=Try, _, Sub0) %% Here is the general try/catch construct outside of guards. %% We can remove try if the value is simple and replace it with a let. E1 = body(E0, value, Sub0), - {Vs1,Sub1} = pattern_list(Vs0, Sub0), + {Vs1,Sub1} = var_list(Vs0, Sub0), B1 = body(B0, value, Sub1), case is_safe_simple(E1, Sub0) of true -> expr(#c_let{anno=A,vars=Vs1,arg=E1,body=B1}, value, Sub0); false -> - {Evs1,Sub2} = pattern_list(Evs0, Sub0), + {Evs1,Sub2} = var_list(Evs0, Sub0), H1 = body(H0, value, Sub2), Try#c_try{arg=E1,vars=Vs1,body=B1,evars=Evs1,handler=H1} end. + +%% If a fun or its application is used as an argument, then it's unsafe to +%% handle it in effect context as the side-effects may rely on its return +%% value. The following is a minimal example of where it can go wrong: +%% +%% do letrec 'f'/0 = fun () -> ... whatever ... +%% in call 'side':'effect'(apply 'f'/0()) +%% 'ok' +%% +%% This function returns 'true' if Body definitely does not rely on a +%% value produced by FVar, or 'false' if Body depends on or might depend on +%% a value produced by FVar. + +is_fun_effect_safe(#c_var{}=FVar, Body) -> + ifes_1(FVar, Body, true). + +ifes_1(FVar, #c_alias{pat=Pat}, _Safe) -> + ifes_1(FVar, Pat, false); +ifes_1(FVar, #c_apply{op=Op,args=Args}, Safe) -> + %% FVar(...) is safe as long its return value is ignored, but it's never + %% okay to pass FVar as an argument. + ifes_list(FVar, Args, false) andalso ifes_1(FVar, Op, Safe); +ifes_1(FVar, #c_binary{segments=Segments}, _Safe) -> + ifes_list(FVar, Segments, false); +ifes_1(FVar, #c_bitstr{val=Val,size=Size,unit=Unit}, _Safe) -> + ifes_list(FVar, [Val, Size, Unit], false); +ifes_1(FVar, #c_call{args=Args}, _Safe) -> + ifes_list(FVar, Args, false); +ifes_1(FVar, #c_case{arg=Arg,clauses=Clauses}, Safe) -> + ifes_1(FVar, Arg, false) andalso ifes_list(FVar, Clauses, Safe); +ifes_1(FVar, #c_catch{body=Body}, _Safe) -> + ifes_1(FVar, Body, false); +ifes_1(FVar, #c_clause{pats=Pats,guard=Guard,body=Body}, Safe) -> + ifes_list(FVar, Pats, false) andalso + ifes_1(FVar, Guard, false) andalso + ifes_1(FVar, Body, Safe); +ifes_1(FVar, #c_cons{hd=Hd,tl=Tl}, _Safe) -> + ifes_1(FVar, Hd, false) andalso ifes_1(FVar, Tl, false); +ifes_1(FVar, #c_fun{body=Body}, _Safe) -> + ifes_1(FVar, Body, false); +ifes_1(FVar, #c_let{arg=Arg,body=Body}, Safe) -> + ifes_1(FVar, Arg, false) andalso ifes_1(FVar, Body, Safe); +ifes_1(FVar, #c_letrec{defs=Defs,body=Body}, Safe) -> + Funs = [Fun || {_,Fun} <- Defs], + ifes_list(FVar, Funs, false) andalso ifes_1(FVar, Body, Safe); +ifes_1(_FVar, #c_literal{}, _Safe) -> + true; +ifes_1(FVar, #c_map{arg=Arg,es=Elements}, _Safe) -> + ifes_1(FVar, Arg, false) andalso ifes_list(FVar, Elements, false); +ifes_1(FVar, #c_map_pair{key=Key,val=Val}, _Safe) -> + ifes_1(FVar, Key, false) andalso ifes_1(FVar, Val, false); +ifes_1(FVar, #c_primop{args=Args}, _Safe) -> + ifes_list(FVar, Args, false); +ifes_1(FVar, #c_receive{timeout=Timeout,action=Action,clauses=Clauses}, Safe) -> + ifes_1(FVar, Timeout, false) andalso + ifes_1(FVar, Action, Safe) andalso + ifes_list(FVar, Clauses, Safe); +ifes_1(FVar, #c_seq{arg=Arg,body=Body}, Safe) -> + %% Arg of a #c_seq{} has no effect so it's okay to use FVar there even if + %% Safe=false. + ifes_1(FVar, Arg, true) andalso ifes_1(FVar, Body, Safe); +ifes_1(FVar, #c_try{arg=Arg,handler=Handler,body=Body}, Safe) -> + ifes_1(FVar, Arg, false) andalso + ifes_1(FVar, Handler, Safe) andalso + ifes_1(FVar, Body, Safe); +ifes_1(FVar, #c_tuple{es=Elements}, _Safe) -> + ifes_list(FVar, Elements, false); +ifes_1(FVar, #c_values{es=Elements}, _Safe) -> + ifes_list(FVar, Elements, false); +ifes_1(#c_var{name=Name}, #c_var{name=Name}, Safe) -> + %% It's safe to return FVar if it's unused. + Safe; +ifes_1(_FVar, #c_var{}, _Safe) -> + true. + +ifes_list(FVar, [E|Es], Safe) -> + ifes_1(FVar, E, Safe) andalso ifes_list(FVar, Es, Safe); +ifes_list(_FVar, [], _Safe) -> + true. + expr_list(Es, Ctxt, Sub) -> [expr(E, Ctxt, Sub) || E <- Es]. @@ -450,6 +580,9 @@ bitstr_list(Es, Sub) -> bitstr(#c_bitstr{val=Val,size=Size}=BinSeg, Sub) -> BinSeg#c_bitstr{val=expr(Val, Sub),size=expr(Size, value, Sub)}. +is_literal_fun(#c_literal{val=F}) -> is_function(F); +is_literal_fun(_) -> false. + %% is_safe_simple(Expr, Sub) -> true | false. %% A safe simple cannot fail with badarg and is safe to use %% in a guard. @@ -457,7 +590,8 @@ bitstr(#c_bitstr{val=Val,size=Size}=BinSeg, Sub) -> %% Currently, we don't attempt to check binaries because they %% are difficult to check. -is_safe_simple(#c_var{}, _) -> true; +is_safe_simple(#c_var{}=Var, _) -> + not cerl:is_c_fname(Var); is_safe_simple(#c_cons{hd=H,tl=T}, Sub) -> is_safe_simple(H, Sub) andalso is_safe_simple(T, Sub); is_safe_simple(#c_tuple{es=Es}, Sub) -> is_safe_simple_list(Es, Sub); @@ -713,6 +847,25 @@ make_effect_seq([H|T], Sub) -> end; make_effect_seq([], _) -> void(). +%% fold_apply(Apply, LiteraFun, Args) -> Apply. +%% Replace an apply of a literal external fun with a call. + +fold_apply(Apply, #c_literal{val=Fun}, Args) when is_function(Fun) -> + {module,Mod} = erlang:fun_info(Fun, module), + {name,Name} = erlang:fun_info(Fun, name), + {arity,Arity} = erlang:fun_info(Fun, arity), + if + Arity =:= length(Args) -> + #c_call{anno=Apply#c_apply.anno, + module=#c_literal{val=Mod}, + name=#c_literal{val=Name}, + args=Args}; + true -> + Apply + end; +fold_apply(Apply, _, _) -> Apply. + + %% Handling remote calls. The module/name fields have been processed. call(#c_call{args=As}=Call, #c_literal{val=M}=M0, #c_literal{val=N}=N0, Sub) -> @@ -722,7 +875,7 @@ call(#c_call{args=As}=Call, #c_literal{val=M}=M0, #c_literal{val=N}=N0, Sub) -> false -> case sys_core_fold_lists:call(Call, M, N, As) of none -> - call_1(Call, M, N, As, Sub); + call_1(Call, M0, N0, As, Sub); Core -> expr(Core, Sub) end @@ -750,6 +903,8 @@ fold_call(Call, #c_literal{val=M}, #c_literal{val=F}, Args, Sub) -> fold_call_1(Call, M, F, Args, Sub); fold_call(Call, _M, _N, _Args, _Sub) -> Call. +fold_call_1(Call, erlang, apply, [Fun,Args], _) -> + simplify_fun_apply(Call, Fun, Args); fold_call_1(Call, erlang, apply, [Mod,Func,Args], _) -> simplify_apply(Call, Mod, Func, Args); fold_call_1(Call, Mod, Name, Args, Sub) -> @@ -775,7 +930,7 @@ fold_lit_args(Call, Module, Name, Args0) -> Val -> case cerl:is_literal_term(Val) of true -> - cerl:abstract(Val); + cerl:ann_abstract(cerl:get_ann(Call), Val); false -> %% Successful evaluation, but it was not possible %% to express the computed value as a literal. @@ -1058,35 +1213,62 @@ eval_failure(Call, Reason) -> %% Simplify an apply/3 to a call if the number of arguments %% are known at compile time. -simplify_apply(Call, Mod, Func, Args) -> +simplify_apply(Call, Mod, Func, Args0) -> case is_atom_or_var(Mod) andalso is_atom_or_var(Func) of - true -> simplify_apply_1(Args, Call, Mod, Func, []); - false -> Call + true -> + case get_fixed_args(Args0, []) of + error -> + Call; + {ok,Args} -> + Call#c_call{module=Mod,name=Func,args=Args} + end; + false -> + Call end. - -simplify_apply_1(#c_literal{val=MoreArgs0}, Call, Mod, Func, Args) - when length(MoreArgs0) >= 0 -> - MoreArgs = [#c_literal{val=Arg} || Arg <- MoreArgs0], - Call#c_call{module=Mod,name=Func,args=reverse(Args, MoreArgs)}; -simplify_apply_1(#c_cons{hd=Arg,tl=T}, Call, Mod, Func, Args) -> - simplify_apply_1(T, Call, Mod, Func, [Arg|Args]); -simplify_apply_1(_, Call, _, _, _) -> Call. - is_atom_or_var(#c_literal{val=Atom}) when is_atom(Atom) -> true; is_atom_or_var(#c_var{}) -> true; is_atom_or_var(_) -> false. +simplify_fun_apply(#c_call{anno=Anno}=Call, Fun, Args0) -> + case get_fixed_args(Args0, []) of + error -> + Call; + {ok,Args} -> + #c_apply{anno=Anno,op=Fun,args=Args} + end. + +get_fixed_args(#c_literal{val=MoreArgs0}, Args) + when length(MoreArgs0) >= 0 -> + MoreArgs = [#c_literal{val=Arg} || Arg <- MoreArgs0], + {ok,reverse(Args, MoreArgs)}; +get_fixed_args(#c_cons{hd=Arg,tl=T}, Args) -> + get_fixed_args(T, [Arg|Args]); +get_fixed_args(_, _) -> error. + %% clause(Clause, Cepxr, Context, Sub) -> Clause. -clause(#c_clause{pats=Ps0,guard=G0,body=B0}=Cl, Cexpr, Ctxt, Sub0) -> - {Ps1,Sub1} = pattern_list(Ps0, Sub0), +clause(#c_clause{pats=Ps0}=Cl, Cexpr, Ctxt, Sub0) -> + try pattern_list(Ps0, Sub0) of + {Ps1,Sub1} -> + clause_1(Cl, Ps1, Cexpr, Ctxt, Sub1) + catch + nomatch -> + Cl#c_clause{anno=[compiler_generated], + guard=#c_literal{val=false}} + end. + +clause_1(#c_clause{guard=G0,body=B0}=Cl, Ps1, Cexpr, Ctxt, Sub1) -> Sub2 = update_types(Cexpr, Ps1, Sub1), - GSub = case {Cexpr,Ps1} of - {#c_var{name='_'},_} -> + GSub = case {Cexpr,Ps1,G0} of + {_,_,#c_literal{}} -> + %% No need for substitution tricks when the guard + %% does not contain any variables. + Sub2; + {#c_var{name='_'},_,_} -> %% In a 'receive', Cexpr is the variable '_', which represents the %% message being matched. We must NOT do any extra substiutions. Sub2; - {#c_var{},[#c_var{}=Var]} -> + {#c_var{},[#c_var{}=Var],_} -> %% The idea here is to optimize expressions such as %% %% case A of A -> ... @@ -1105,7 +1287,13 @@ clause(#c_clause{pats=Ps0,guard=G0,body=B0}=Cl, Cexpr, Ctxt, Sub0) -> %% %% case A of NewVar when true -> ... %% - sub_set_var(Var, Cexpr, Sub2); + case cerl:is_c_fname(Cexpr) of + false -> + sub_set_var(Var, Cexpr, Sub2); + true -> + %% We must not copy funs, and especially not into guards. + Sub2 + end; _ -> Sub2 end, @@ -1120,7 +1308,7 @@ clause(#c_clause{pats=Ps0,guard=G0,body=B0}=Cl, Cexpr, Ctxt, Sub0) -> %% the unsubstituted variables and values. let_substs(Vs0, As0, Sub0) -> - {Vs1,Sub1} = pattern_list(Vs0, Sub0), + {Vs1,Sub1} = var_list(Vs0, Sub0), {Vs2,As1,Ss} = let_substs_1(Vs1, As0, Sub1), Sub2 = sub_add_scope([V || #c_var{name=V} <- Vs2], Sub1), {Vs2,As1, @@ -1165,13 +1353,18 @@ let_subst_list([], [], _) -> {[],[],[]}. %%pattern(Pat, Sub) -> pattern(Pat, Sub, Sub). pattern(#c_var{}=Pat, Isub, Osub) -> - case sub_is_val(Pat, Isub) of + case sub_is_in_scope(Pat, Isub) of true -> + %% This variable either has a substitution or is used in + %% the variable list of an enclosing `let`. In either + %% case, it must be renamed to an unused name to avoid + %% name capture problems. V1 = make_var_name(), Pat1 = #c_var{name=V1}, {Pat1,sub_set_var(Pat, Pat1, sub_add_scope([V1], Osub))}; false -> - {Pat,sub_del_var(Pat, Osub)} + %% This variable has never been used. Add it to the scope. + {Pat,sub_add_scope([Pat#c_var.name], Osub)} end; pattern(#c_literal{}=Pat, _, Osub) -> {Pat,Osub}; pattern(#c_cons{anno=Anno,hd=H0,tl=T0}, Isub, Osub0) -> @@ -1206,20 +1399,132 @@ bin_pattern_list(Ps0, Isub, Osub0) -> {Ps,{_,Osub}} = mapfoldl(fun bin_pattern/2, {Isub,Osub0}, Ps0), {Ps,Osub}. -bin_pattern(#c_bitstr{val=E0,size=Size0}=Pat, {Isub0,Osub0}) -> +bin_pattern(#c_bitstr{val=E0,size=Size0}=Pat0, {Isub0,Osub0}) -> Size1 = expr(Size0, Isub0), {E1,Osub} = pattern(E0, Isub0, Osub0), Isub = case E0 of #c_var{} -> sub_set_var(E0, E1, Isub0); _ -> Isub0 end, - {Pat#c_bitstr{val=E1,size=Size1},{Isub,Osub}}. + Pat = Pat0#c_bitstr{val=E1,size=Size1}, + bin_pat_warn(Pat), + {Pat,{Isub,Osub}}. pattern_list(Ps, Sub) -> pattern_list(Ps, Sub, Sub). pattern_list(Ps0, Isub, Osub0) -> mapfoldl(fun (P, Osub) -> pattern(P, Isub, Osub) end, Osub0, Ps0). +%% var_list([Var], InSub) -> {Pattern,OutSub}. +%% Works like pattern_list/2 but only accept variables and is +%% guaranteed not to throw an exception. + +var_list(Vs, Sub0) -> + mapfoldl(fun (#c_var{}=V, Sub) -> + pattern(V, Sub, Sub) + end, Sub0, Vs). + + +%%% +%%% Generate warnings for binary patterns that will not match. +%%% + +bin_pat_warn(#c_bitstr{type=#c_literal{val=Type}, + val=Val0, + size=#c_literal{val=Sz}, + unit=#c_literal{val=Unit}, + flags=Fl}=Pat) -> + case {Type,Sz} of + {_,_} when is_integer(Sz), Sz >= 0 -> ok; + {binary,all} -> ok; + {utf8,undefined} -> ok; + {utf16,undefined} -> ok; + {utf32,undefined} -> ok; + {_,_} -> + add_warning(Pat, {nomatch_bit_syntax_size,Sz}), + throw(nomatch) + end, + case {Type,Val0} of + {integer,#c_literal{val=Val}} when is_integer(Val) -> + Signedness = signedness(Fl), + TotalSz = Sz * Unit, + bit_pat_warn_int(Val, TotalSz, Signedness, Pat); + {float,#c_literal{val=Val}} when is_float(Val) -> + ok; + {utf8,#c_literal{val=Val}} when is_integer(Val) -> + bit_pat_warn_unicode(Val, Pat); + {utf16,#c_literal{val=Val}} when is_integer(Val) -> + bit_pat_warn_unicode(Val, Pat); + {utf32,#c_literal{val=Val}} when is_integer(Val) -> + bit_pat_warn_unicode(Val, Pat); + {_,#c_literal{val=Val}} -> + add_warning(Pat, {nomatch_bit_syntax_type,Val,Type}), + throw(nomatch); + {_,_} -> + ok + end; +bin_pat_warn(#c_bitstr{type=#c_literal{val=Type},val=Val0,flags=Fl}=Pat) -> + %% Size is variable. Not much that we can check. + case {Type,Val0} of + {integer,#c_literal{val=Val}} when is_integer(Val) -> + case signedness(Fl) of + unsigned when Val < 0 -> + add_warning(Pat, {nomatch_bit_syntax_unsigned,Val}), + throw(nomatch); + _ -> + ok + end; + {float,#c_literal{val=Val}} when is_float(Val) -> + ok; + {_,#c_literal{val=Val}} -> + add_warning(Pat, {nomatch_bit_syntax_type,Val,Type}), + throw(nomatch); + {_,_} -> + ok + end. + +bit_pat_warn_int(Val, 0, signed, Pat) -> + if + Val =:= 0 -> + ok; + true -> + add_warning(Pat, {nomatch_bit_syntax_truncated,signed,Val,0}), + throw(nomatch) + end; +bit_pat_warn_int(Val, Sz, signed, Pat) -> + if + Val < 0, Val bsr (Sz - 1) =/= -1 -> + add_warning(Pat, {nomatch_bit_syntax_truncated,signed,Val,Sz}), + throw(nomatch); + Val > 0, Val bsr (Sz - 1) =/= 0 -> + add_warning(Pat, {nomatch_bit_syntax_truncated,signed,Val,Sz}), + throw(nomatch); + true -> + ok + end; +bit_pat_warn_int(Val, _Sz, unsigned, Pat) when Val < 0 -> + add_warning(Pat, {nomatch_bit_syntax_unsigned,Val}), + throw(nomatch); +bit_pat_warn_int(Val, Sz, unsigned, Pat) -> + if + Val bsr Sz =:= 0 -> + ok; + true -> + add_warning(Pat, {nomatch_bit_syntax_truncated,unsigned,Val,Sz}), + throw(nomatch) + end. + +bit_pat_warn_unicode(U, _Pat) when 0 =< U, U =< 16#10FFFF -> + ok; +bit_pat_warn_unicode(U, Pat) -> + add_warning(Pat, {nomatch_bit_syntax_unicode,U}), + throw(nomatch). + +signedness(#c_literal{val=Flags}) -> + [S] = [F || F <- Flags, F =:= signed orelse F =:= unsigned], + S. + + %% is_subst(Expr) -> true | false. %% Test whether an expression is a suitable substitution. @@ -1238,8 +1543,8 @@ is_subst(_) -> false. %% sub_set_name(Name, Value, #sub{}) -> #sub{}. %% sub_del_var(Var, #sub{}) -> #sub{}. %% sub_subst_var(Var, Value, #sub{}) -> [{Name,Value}]. -%% sub_is_val(Var, #sub{}) -> boolean(). -%% sub_add_scope(#sub{}) -> #sub{} +%% sub_is_in_scope(Var, #sub{}) -> boolean(). +%% sub_add_scope([Var], #sub{}) -> #sub{} %% sub_subst_scope(#sub{}) -> #sub{} %% %% We use the variable name as key so as not have problems with @@ -1260,9 +1565,6 @@ sub_new() -> #sub{v=orddict:new(),s=cerl_sets:new(),t=#{}}. sub_new(#sub{}=Sub) -> Sub#sub{v=orddict:new(),t=#{}}. -sub_new_preserve_types(#sub{}=Sub) -> - Sub#sub{v=orddict:new()}. - sub_get_var(#c_var{name=V}=Var, #sub{v=S}) -> case orddict:find(V, S) of {ok,Val} -> Val; @@ -1277,18 +1579,6 @@ sub_set_name(V, Val, #sub{v=S,s=Scope,t=Tdb0}=Sub) -> Tdb = copy_type(V, Val, Tdb1), Sub#sub{v=orddict:store(V, Val, S),s=cerl_sets:add_element(V, Scope),t=Tdb}. -sub_del_var(#c_var{name=V}, #sub{v=S,s=Scope,t=Tdb}=Sub) -> - %% Profiling shows that for programs with many record operations, - %% sub_del_var/2 is a bottleneck. Since the scope contains all - %% variables that are live, we know that V cannot be present in S - %% if it is not in the scope. - case cerl_sets:is_element(V, Scope) of - false -> - Sub#sub{s=cerl_sets:add_element(V, Scope)}; - true -> - Sub#sub{v=orddict:erase(V, S),t=kill_types(V, Tdb)} - end. - sub_subst_var(#c_var{name=V}, Val, #sub{v=S0}) -> %% Fold chained substitutions. [{V,Val}] ++ [ {K,Val} || {K,#c_var{name=V1}} <- S0, V1 =:= V]. @@ -1300,19 +1590,22 @@ sub_add_scope(Vs, #sub{s=Scope0}=Sub) -> Sub#sub{s=Scope}. sub_subst_scope(#sub{v=S0,s=Scope}=Sub) -> - S = [{-1,#c_var{name=Sv}} || Sv <- cerl_sets:to_list(Scope)]++S0, - Sub#sub{v=S}. - -sub_is_val(#c_var{name=V}, #sub{v=S,s=Scope}) -> - %% When the bottleneck in sub_del_var/2 was eliminated, this - %% became the new bottleneck. Since the scope contains all - %% live variables, a variable V can only be the target for - %% a substitution if it is in the scope. - cerl_sets:is_element(V, Scope) andalso v_is_value(V, S). - -v_is_value(Var, [{_,#c_var{name=Var}}|_]) -> true; -v_is_value(Var, [_|T]) -> v_is_value(Var, T); -v_is_value(_, []) -> false. + Initial = case S0 of + [{NegInt,_}|_] when is_integer(NegInt), NegInt < 0 -> + NegInt - 1; + _ -> + -1 + end, + S = sub_subst_scope_1(cerl_sets:to_list(Scope), Initial, S0), + Sub#sub{v=orddict:from_list(S)}. + +%% The keys in an orddict must be unique. Make them so! +sub_subst_scope_1([H|T], Key, Acc) -> + sub_subst_scope_1(T, Key-1, [{Key,#c_var{name=H}}|Acc]); +sub_subst_scope_1([], _, Acc) -> Acc. + +sub_is_in_scope(#c_var{name=V}, #sub{s=Scope}) -> + cerl_sets:is_element(V, Scope). %% warn_no_clause_match(CaseOrig, CaseOpt) -> ok %% Generate a warning if none of the user-specified clauses @@ -1392,9 +1685,11 @@ will_match(E, [P]) -> will_match_1({false,_}) -> maybe; will_match_1({true,_}) -> yes. -%% opt_bool_case(CoreExpr) - CoreExpr'. -%% Do various optimizations to case statement that has a -%% boolean case expression. +%% opt_bool_case(CoreExpr, Sub) - CoreExpr'. +%% +%% In bodies, do various optimizations to case statements that have +%% boolean case expressions. We don't do the optimizations in guards, +%% because they would thwart the optimization in v3_kernel. %% %% We start with some simple optimizations and normalization %% to facilitate later optimizations. @@ -1403,7 +1698,7 @@ will_match_1({true,_}) -> yes. %% (or fail), we can remove any clause that cannot %% possibly match 'true' or 'false'. Also, any clause %% following both 'true' and 'false' clause can -%% be removed. If successful, we will end up this: +%% be removed. If successful, we will end up like this: %% %% case BoolExpr of case BoolExpr of %% true -> false -> @@ -1414,8 +1709,11 @@ will_match_1({true,_}) -> yes. %% %% We give up if there are clauses with guards, or if there %% is a variable clause that matches anything. -%% -opt_bool_case(#c_case{arg=Arg}=Case0) -> + +opt_bool_case(#c_case{}=Case, #sub{in_guard=true}) -> + %% v3_kernel does a better job without "help". + Case; +opt_bool_case(#c_case{arg=Arg}=Case0, #sub{in_guard=false}) -> case is_bool_expr(Arg) of false -> Case0; @@ -1427,8 +1725,7 @@ opt_bool_case(#c_case{arg=Arg}=Case0) -> impossible -> Case0 end - end; -opt_bool_case(Core) -> Core. + end. opt_bool_clauses(#c_case{clauses=Cs}=Case) -> Case#c_case{clauses=opt_bool_clauses(Cs, false, false)}. @@ -1444,16 +1741,14 @@ opt_bool_clauses(Cs, true, true) -> [] end; opt_bool_clauses([#c_clause{pats=[#c_literal{val=Lit}], - guard=#c_literal{val=true}, - body=B}=C0|Cs], SeenT, SeenF) -> + guard=#c_literal{val=true}}=C|Cs], SeenT, SeenF) -> case is_boolean(Lit) of false -> %% Not a boolean - this clause can't match. - add_warning(C0, nomatch_clause_type), + add_warning(C, nomatch_clause_type), opt_bool_clauses(Cs, SeenT, SeenF); true -> %% This clause will match. - C = C0#c_clause{body=opt_bool_case(B)}, case {Lit,SeenT,SeenF} of {false,_,false} -> [C|opt_bool_clauses(Cs, SeenT, true)]; @@ -1729,10 +2024,10 @@ case_opt_arg_1(E0, Cs0, LitExpr) -> true -> E = case_opt_compiler_generated(E0), Cs = case_opt_nomatch(E, Cs0, LitExpr), - case cerl:data_type(E) of - {atomic,_} -> + case cerl:is_literal(E) of + true -> case_opt_lit(E, Cs); - _ -> + false -> case_opt_data(E, Cs) end end. @@ -1880,10 +2175,10 @@ case_opt_lit_1(_, []) -> []. %% the clauses where it is actually needed. case_opt_data(E, Cs0) -> - Es = cerl:data_es(E), TypeSig = {cerl:data_type(E),cerl:data_arity(E)}, - try case_opt_data_1(Cs0, Es, TypeSig) of + try case_opt_data_1(Cs0, TypeSig) of Cs -> + Es = cerl:data_es(E), {ok,Es,Cs} catch throw:impossible -> @@ -1891,44 +2186,47 @@ case_opt_data(E, Cs0) -> {error,Cs0} end. -case_opt_data_1([{[P0|Ps0],C,PsAcc,Bs0}|Cs], Es, TypeSig) -> +case_opt_data_1([{[P0|Ps0],C,PsAcc,Bs0}|Cs], TypeSig) -> P = case_opt_compiler_generated(P0), - BindTo = #c_var{name=dummy}, - {Ps1,[{BindTo,_}|Bs1]} = case_data_pat_alias(P, BindTo, TypeSig, []), - [{Ps1++Ps0,C,PsAcc,Bs1++Bs0}|case_opt_data_1(Cs, Es, TypeSig)]; -case_opt_data_1([], _, _) -> []. + {Ps1,Bs} = case_opt_data_2(P, TypeSig, Bs0), + [{Ps1++Ps0,C,PsAcc,Bs}|case_opt_data_1(Cs, TypeSig)]; +case_opt_data_1([], _) -> []. -case_data_pat_alias(P, BindTo0, TypeSig, Bs0) -> - case cerl:type(P) of - alias -> - %% Recursively handle the pattern and bind to - %% the alias variable. - BindTo = cerl:alias_var(P), - Apat0 = cerl:alias_pat(P), - Ann = [compiler_generated], - Apat = cerl:set_ann(Apat0, Ann), - {Ps,Bs} = case_data_pat_alias(Apat, BindTo, TypeSig, Bs0), - {Ps,[{BindTo0,BindTo}|Bs]}; - var -> - %% Here we will need to actually build the data and bind - %% it to the variable. +case_opt_data_2(P, TypeSig, Bs0) -> + case case_analyze_pat(P) of + {[],Pat} when Pat =/= none -> + DataEs = cerl:data_es(P), + {DataEs,Bs0}; + {[V|Vs],none} -> {Type,Arity} = TypeSig, Ann = [compiler_generated], Vars = make_vars(Ann, Arity), Data = cerl:ann_make_data(Ann, Type, Vars), - Bs = [{BindTo0,P},{P,Data}|Bs0], + Bs = [{V,Data} | [{Var,V} || Var <- Vs] ++ Bs0], {Vars,Bs}; - _ -> - %% Since case_opt_nomatch/3 has removed all clauses that - %% cannot match, we KNOW that this clause must match and - %% that the pattern must be a data constructor. - %% Here we must build the data and bind it to the variable. + {[V|Vs],Pat} when Pat =/= none -> {Type,_} = TypeSig, - DataEs = cerl:data_es(P), + DataEs = cerl:data_es(Pat), Vars = pat_to_expr_list(DataEs), Ann = [compiler_generated], Data = cerl:ann_make_data(Ann, Type, Vars), - {DataEs,[{BindTo0,Data}]} + Bs = [{V,Data} | [{Var,V} || Var <- Vs] ++ Bs0], + {DataEs,Bs} + end. + +case_analyze_pat(P) -> + case_analyze_pat_1(P, [], none). + +case_analyze_pat_1(P, Vs, Pat) -> + case cerl:type(P) of + alias -> + V = cerl:alias_var(P), + Apat = cerl:alias_pat(P), + case_analyze_pat_1(Apat, [V|Vs], Pat); + var -> + {[P|Vs],Pat}; + _ -> + {Vs,P} end. %% pat_to_expr(Pattern) -> Expression. @@ -1972,7 +2270,7 @@ make_var(A) -> make_var_name() -> N = get(new_var_num), put(new_var_num, N+1), - list_to_atom("fol"++integer_to_list(N)). + N. letify(Bs, Body) -> Ann = cerl:get_ann(Body), @@ -1986,7 +2284,7 @@ letify(Bs, Body) -> -spec opt_not_in_let(cerl:c_let()) -> cerl:cerl(). opt_not_in_let(#c_let{vars=[_]=Vs0,arg=Arg0,body=Body0}=Let) -> - case opt_not_in_let(Vs0, Arg0, Body0) of + case opt_not_in_let_0(Vs0, Arg0, Body0) of {[],#c_values{es=[]},Body} -> Body; {Vs,Arg,Body} -> @@ -1994,13 +2292,7 @@ opt_not_in_let(#c_let{vars=[_]=Vs0,arg=Arg0,body=Body0}=Let) -> end; opt_not_in_let(Let) -> Let. -%% opt_not_in_let(Vs, Arg, Body) -> {Vs',Arg',Body'} -%% Try to optimize away a 'not' operator in a 'let'. - --spec opt_not_in_let([cerl:c_var()], cerl:cerl(), cerl:cerl()) -> - {[cerl:c_var()],cerl:cerl(),cerl:cerl()}. - -opt_not_in_let([#c_var{name=V}]=Vs0, Arg0, Body0) -> +opt_not_in_let_0([#c_var{name=V}]=Vs0, Arg0, Body0) -> case cerl:type(Body0) of call -> %% let <V> = Expr in not V ==> @@ -2031,33 +2323,29 @@ opt_not_in_let([#c_var{name=V}]=Vs0, Arg0, Body0) -> end; _ -> {Vs0,Arg0,Body0} - end; -opt_not_in_let(Vs, Arg, Body) -> - {Vs,Arg,Body}. + end. opt_not_in_let_1(V, Call, Body) -> case Call of #c_call{module=#c_literal{val=erlang}, name=#c_literal{val='not'}, args=[#c_var{name=V}]} -> - opt_not_in_let_2(Body); + opt_not_in_let_2(Body, Call); _ -> no end. -opt_not_in_let_2(#c_case{clauses=Cs0}=Case) -> +opt_not_in_let_2(#c_case{clauses=Cs0}=Case, NotCall) -> Vars = make_vars([], 1), - Body = #c_call{module=#c_literal{val=erlang}, - name=#c_literal{val='not'}, - args=Vars}, + Body = NotCall#c_call{args=Vars}, Cs = [begin Let = #c_let{vars=Vars,arg=B,body=Body}, C#c_clause{body=opt_not_in_let(Let)} end || #c_clause{body=B}=C <- Cs0], {yes,Case#c_case{clauses=Cs}}; -opt_not_in_let_2(#c_call{}=Call0) -> +opt_not_in_let_2(#c_call{}=Call0, _NotCall) -> invert_call(Call0); -opt_not_in_let_2(_) -> no. +opt_not_in_let_2(_, _) -> no. invert_call(#c_call{module=#c_literal{val=erlang}, name=#c_literal{val=Name0}, @@ -2081,24 +2369,24 @@ inverse_rel_op('=<') -> '>'; inverse_rel_op(_) -> no. -%% opt_bool_case_in_let(LetExpr, Sub) -> Core +%% opt_bool_case_in_let(LetExpr) -> Core opt_bool_case_in_let(#c_let{vars=Vs,arg=Arg,body=B}=Let, Sub) -> - opt_case_in_let_1(Vs, Arg, B, Let, Sub). + opt_bool_case_in_let_1(Vs, Arg, B, Let, Sub). -opt_case_in_let_1([#c_var{name=V}], Arg, +opt_bool_case_in_let_1([#c_var{name=V}], Arg, #c_case{arg=#c_var{name=V}}=Case0, Let, Sub) -> case is_simple_case_arg(Arg) of true -> - Case = opt_bool_case(Case0#c_case{arg=Arg}), + Case = opt_bool_case(Case0#c_case{arg=Arg}, Sub), case core_lib:is_var_used(V, Case) of - false -> expr(Case, sub_new(Sub)); + false -> Case; true -> Let end; false -> Let end; -opt_case_in_let_1(_, _, _, Let, _) -> Let. +opt_bool_case_in_let_1(_, _, _, Let, _) -> Let. %% is_simple_case_arg(Expr) -> true|false %% Determine whether the Expr is simple enough to be worth @@ -2231,9 +2519,7 @@ is_safe_bool_expr_list([], _, _) -> true. %% as a let or a sequence, move the original let body into the complex %% expression. -simplify_let(#c_let{arg=Arg0}=Let0, Sub) -> - Arg = opt_bool_case(Arg0), - Let = Let0#c_let{arg=Arg}, +simplify_let(#c_let{arg=Arg}=Let, Sub) -> move_let_into_expr(Let, Arg, Sub). move_let_into_expr(#c_let{vars=InnerVs0,body=InnerBody0}=Inner, @@ -2251,60 +2537,64 @@ move_let_into_expr(#c_let{vars=InnerVs0,body=InnerBody0}=Inner, %% Arg = body(Arg0, Sub0), ScopeSub0 = sub_subst_scope(Sub0#sub{t=#{}}), - {OuterVs,ScopeSub} = pattern_list(OuterVs0, ScopeSub0), + {OuterVs,ScopeSub} = var_list(OuterVs0, ScopeSub0), OuterBody = body(OuterBody0, ScopeSub), - {InnerVs,Sub} = pattern_list(InnerVs0, Sub0), + {InnerVs,Sub} = var_list(InnerVs0, Sub0), InnerBody = body(InnerBody0, Sub), Outer#c_let{vars=OuterVs,arg=Arg, body=Inner#c_let{vars=InnerVs,arg=OuterBody,body=InnerBody}}; move_let_into_expr(#c_let{vars=Lvs0,body=Lbody0}=Let, - #c_case{arg=Cexpr0,clauses=[Ca0,Cb0|Cs]}=Case, Sub0) -> - %% Test if there are no more clauses than Ca0 and Cb0, or if - %% Cb0 is guaranteed to match. - TwoClauses = Cs =:= [] orelse - case Cb0 of - #c_clause{pats=[#c_var{}],guard=#c_literal{val=true}} -> true; - _ -> false - end, - case {TwoClauses,is_failing_clause(Ca0),is_failing_clause(Cb0)} of - {true,false,true} -> + #c_case{arg=Cexpr0,clauses=[Ca0|Cs0]}=Case, Sub0) -> + case not is_failing_clause(Ca0) andalso + are_all_failing_clauses(Cs0) of + true -> %% let <Lvars> = case <Case-expr> of - %% <Cvars> -> <Clause-body>; - %% <OtherCvars> -> erlang:error(...) + %% <Cpats> -> <Clause-body>; + %% <OtherCpats> -> erlang:error(...) %% end %% in <Let-body> %% %% ==> %% %% case <Case-expr> of - %% <Cvars> -> + %% <Cpats> -> %% let <Lvars> = <Clause-body> %% in <Let-body>; - %% <OtherCvars> -> erlang:error(...) + %% <OtherCpats> -> erlang:error(...) %% end Cexpr = body(Cexpr0, Sub0), - CaVars0 = Ca0#c_clause.pats, + CaPats0 = Ca0#c_clause.pats, G0 = Ca0#c_clause.guard, B0 = Ca0#c_clause.body, ScopeSub0 = sub_subst_scope(Sub0#sub{t=#{}}), - {CaVars,ScopeSub} = pattern_list(CaVars0, ScopeSub0), - G = guard(G0, ScopeSub), - - B1 = body(B0, ScopeSub), - - {Lvs,B2,Sub1} = let_substs(Lvs0, B1, Sub0), - Sub2 = Sub1#sub{s=cerl_sets:union(ScopeSub#sub.s, - Sub1#sub.s)}, - Lbody = body(Lbody0, Sub2), - B = Let#c_let{vars=Lvs,arg=core_lib:make_values(B2),body=Lbody}, - - Ca = Ca0#c_clause{pats=CaVars,guard=G,body=B}, - Cb = clause(Cb0, Cexpr, value, Sub0), - Case#c_case{arg=Cexpr,clauses=[Ca,Cb]}; - {_,_,_} -> impossible + try pattern_list(CaPats0, ScopeSub0) of + {CaPats,ScopeSub} -> + G = guard(G0, ScopeSub), + + B1 = body(B0, ScopeSub), + + {Lvs,B2,Sub1} = let_substs(Lvs0, B1, Sub0), + Sub2 = Sub1#sub{s=cerl_sets:union(ScopeSub#sub.s, + Sub1#sub.s)}, + Lbody = body(Lbody0, Sub2), + B = Let#c_let{vars=Lvs, + arg=core_lib:make_values(B2), + body=Lbody}, + + Ca = Ca0#c_clause{pats=CaPats,guard=G,body=B}, + Cs = [clause(C, Cexpr, value, Sub0) || C <- Cs0], + Case#c_case{arg=Cexpr,clauses=[Ca|Cs]} + catch + nomatch -> + %% This is not a defeat. The code will eventually + %% be optimized to erlang:error(...) by the other + %% optimizations done in this module. + impossible + end; + false -> impossible end; move_let_into_expr(#c_let{vars=Lvs0,body=Lbody0}=Let, #c_seq{arg=Sarg0,body=Sbody0}=Seq, Sub0) -> @@ -2327,9 +2617,78 @@ move_let_into_expr(#c_let{vars=Lvs0,body=Lbody0}=Let, body=Lbody}}; move_let_into_expr(_Let, _Expr, _Sub) -> impossible. +are_all_failing_clauses(Cs) -> + all(fun is_failing_clause/1, Cs). + is_failing_clause(#c_clause{body=B}) -> will_fail(B). +%% opt_build_stacktrace(Let) -> Core. +%% If the stacktrace is *only* used in a call to erlang:raise/3, +%% there is no need to build a cooked stackframe using build_stacktrace/1. + +opt_build_stacktrace(#c_let{vars=[#c_var{name=Cooked}], + arg=#c_primop{name=#c_literal{val=build_stacktrace}, + args=[RawStk]}, + body=Body}=Let) -> + case Body of + #c_call{module=#c_literal{val=erlang}, + name=#c_literal{val=raise}, + args=[Class,Exp,#c_var{name=Cooked}]} -> + %% The stacktrace is only used in a call to erlang:raise/3. + %% There is no need to build the stacktrace. Replace the + %% call to erlang:raise/3 with the the raw_raise/3 instruction, + %% which will use a raw stacktrace. + #c_primop{name=#c_literal{val=raw_raise}, + args=[Class,Exp,RawStk]}; + #c_let{vars=[#c_var{name=V}],arg=Arg,body=B0} when V =/= Cooked -> + case core_lib:is_var_used(Cooked, Arg) of + false -> + %% The built stacktrace is not used in the argument, + %% so we can sink the building of the stacktrace into + %% the body of the let. + B = opt_build_stacktrace(Let#c_let{body=B0}), + Body#c_let{body=B}; + true -> + Let + end; + #c_seq{arg=Arg,body=B0} -> + case core_lib:is_var_used(Cooked, Arg) of + false -> + %% The built stacktrace is not used in the argument, + %% so we can sink the building of the stacktrace into + %% the body of the sequence. + B = opt_build_stacktrace(Let#c_let{body=B0}), + Body#c_seq{body=B}; + true -> + Let + end; + #c_case{arg=Arg,clauses=Cs0} -> + case core_lib:is_var_used(Cooked, Arg) orelse + is_used_in_any_guard(Cooked, Cs0) of + false -> + %% The built stacktrace is not used in the argument, + %% so we can sink the building of the stacktrace into + %% each arm of the case. + Cs = [begin + B = opt_build_stacktrace(Let#c_let{body=B0}), + C#c_clause{body=B} + end || #c_clause{body=B0}=C <- Cs0], + Body#c_case{clauses=Cs}; + true -> + Let + end; + _ -> + Let + end; +opt_build_stacktrace(Expr) -> + Expr. + +is_used_in_any_guard(V, Cs) -> + any(fun(#c_clause{guard=G}) -> + core_lib:is_var_used(V, G) + end, Cs). + %% opt_case_in_let(Let) -> Let' %% Try to avoid building tuples that are immediately matched. %% A common pattern is: @@ -2444,9 +2803,13 @@ delay_build_expr_1(#c_receive{clauses=Cs0, timeout=Timeout, action=A0}=Rec, TypeSig) -> Cs = delay_build_cs(Cs0, TypeSig), - A = case Timeout of - #c_literal{val=infinity} -> A0; - _ -> delay_build_expr(A0, TypeSig) + A = case {Timeout,A0} of + {#c_literal{val=infinity},#c_literal{}} -> + {_Type,Arity} = TypeSig, + Es = lists:duplicate(Arity, A0), + core_lib:make_values(Es); + _ -> + delay_build_expr(A0, TypeSig) end, Rec#c_receive{clauses=Cs,action=A}; delay_build_expr_1(#c_seq{body=B0}=Seq, TypeSig) -> @@ -2479,60 +2842,96 @@ opt_simple_let_0(#c_let{arg=Arg0}=Let, Ctxt, Sub) -> opt_simple_let_1(#c_let{vars=Vs0,body=B0}=Let, Arg0, Ctxt, Sub0) -> %% Optimise let and add new substitutions. - {Vs1,Args,Sub1} = let_substs(Vs0, Arg0, Sub0), - BodySub = update_let_types(Vs1, Args, Sub1), - B1 = body(B0, Ctxt, BodySub), - Arg1 = core_lib:make_values(Args), - {Vs,Arg,B} = opt_not_in_let(Vs1, Arg1, B1), - opt_simple_let_2(Let, Vs, Arg, B, B0, Ctxt, Sub1). - -opt_simple_let_2(Let0, Vs0, Arg0, Body, PrevBody, Ctxt, Sub) -> + {Vs,Args,Sub1} = let_substs(Vs0, Arg0, Sub0), + BodySub = update_let_types(Vs, Args, Sub1), + Sub = Sub1#sub{v=[],s=cerl_sets:new()}, + B = body(B0, Ctxt, BodySub), + Arg = core_lib:make_values(Args), + opt_simple_let_2(Let, Vs, Arg, B, B0, Sub). + + +%% opt_simple_let_2(Let0, Vs0, Arg0, Body, PrevBody, Ctxt, Sub) -> Core. +%% Do final simplifications of the let. +%% +%% Note that the substitutions and scope in Sub have been cleared +%% and should not be used. + +opt_simple_let_2(Let0, Vs0, Arg0, Body, PrevBody, Sub) -> case {Vs0,Arg0,Body} of - {[#c_var{name=N1}],Arg1,#c_var{name=N2}} -> - case N1 =:= N2 of - true -> - %% let <Var> = Arg in <Var> ==> Arg - Arg1; - false -> - %% let <Var> = Arg in <OtherVar> ==> seq Arg OtherVar - Arg = maybe_suppress_warnings(Arg1, Vs0, PrevBody), - expr(#c_seq{arg=Arg,body=Body}, Ctxt, - sub_new_preserve_types(Sub)) - end; + {[#c_var{name=V}],Arg1,#c_var{name=V}} -> + %% let <Var> = Arg in <Var> ==> Arg + Arg1; {[],#c_values{es=[]},_} -> %% No variables left. Body; - {Vs,Arg1,#c_literal{}} -> - Arg = maybe_suppress_warnings(Arg1, Vs, PrevBody), - E = case Ctxt of - effect -> - %% Throw away the literal body. - Arg; - value -> - %% Since the variable is not used in the body, we - %% can rewrite the let to a sequence. - %% let <Var> = Arg in Literal ==> seq Arg Literal - #c_seq{arg=Arg,body=Body} - end, - expr(E, Ctxt, sub_new_preserve_types(Sub)); - {Vs,Arg1,Body} -> - %% If none of the variables are used in the body, we can - %% rewrite the let to a sequence: - %% let <Var> = Arg in BodyWithoutVar ==> - %% seq Arg BodyWithoutVar - case is_any_var_used(Vs, Body) of - false -> - Arg = maybe_suppress_warnings(Arg1, Vs, PrevBody), - expr(#c_seq{arg=Arg,body=Body}, Ctxt, - sub_new_preserve_types(Sub)); - true -> - Let1 = Let0#c_let{vars=Vs,arg=Arg1,body=Body}, - Let2 = opt_bool_case_in_let(Let1, Sub), - opt_case_in_let_arg(Let2, Ctxt, Sub) + {[#c_var{name=V}=Var|Vars]=Vars0,Arg1,Body} -> + case core_lib:is_var_used(V, Body) of + false when Vars =:= [] -> + %% If the variable is not used in the body, we can + %% rewrite the let to a sequence: + %% let <Var> = Arg in BodyWithoutVar ==> + %% seq Arg BodyWithoutVar + Arg = maybe_suppress_warnings(Arg1, Var, PrevBody), + #c_seq{arg=Arg,body=Body}; + false -> + %% There are multiple values returned by the argument + %% and the first value is not used (this is a 'case' + %% with exported variables, but the return value is + %% ignored). We can remove the first variable and the + %% the first value returned from the 'let' argument. + Arg2 = remove_first_value(Arg1, Sub), + Let1 = Let0#c_let{vars=Vars,arg=Arg2,body=Body}, + post_opt_let(Let1, Sub); + true -> + Let1 = Let0#c_let{vars=Vars0,arg=Arg1,body=Body}, + post_opt_let(Let1, Sub) end end. -%% maybe_suppress_warnings(Arg, [#c_var{}], PreviousBody) -> Arg' +%% post_opt_let(Let, Sub) +%% Final optimizations of the let. +%% +%% Note that the substitutions and scope in Sub have been cleared +%% and should not be used. + +post_opt_let(Let0, Sub) -> + Let1 = opt_bool_case_in_let(Let0, Sub), + opt_build_stacktrace(Let1). + + +%% remove_first_value(Core0, Sub) -> Core. +%% Core0 is an expression that returns at least two values. +%% Remove the first value returned from Core0. + +remove_first_value(#c_values{es=[V|Vs]}, Sub) -> + Values = core_lib:make_values(Vs), + case is_safe_simple(V, Sub) of + false -> + #c_seq{arg=V,body=Values}; + true -> + Values + end; +remove_first_value(#c_case{clauses=Cs0}=Core, Sub) -> + Cs = remove_first_value_cs(Cs0, Sub), + Core#c_case{clauses=Cs}; +remove_first_value(#c_receive{clauses=Cs0,action=Act0}=Core, Sub) -> + Cs = remove_first_value_cs(Cs0, Sub), + Act = remove_first_value(Act0, Sub), + Core#c_receive{clauses=Cs,action=Act}; +remove_first_value(#c_let{body=B}=Core, Sub) -> + Core#c_let{body=remove_first_value(B, Sub)}; +remove_first_value(#c_seq{body=B}=Core, Sub) -> + Core#c_seq{body=remove_first_value(B, Sub)}; +remove_first_value(#c_primop{}=Core, _Sub) -> + Core; +remove_first_value(#c_call{}=Core, _Sub) -> + Core. + +remove_first_value_cs(Cs, Sub) -> + [C#c_clause{body=remove_first_value(B, Sub)} || + #c_clause{body=B}=C <- Cs]. + +%% maybe_suppress_warnings(Arg, #c_var{}, PreviousBody) -> Arg' %% Try to suppress false warnings when a variable is not used. %% For instance, we don't expect a warning for useless building in: %% @@ -2543,12 +2942,12 @@ opt_simple_let_2(Let0, Vs0, Arg0, Body, PrevBody, Ctxt, Sub) -> %% referenced in the original unoptimized code. If they were, we will %% consider the warning false and suppress it. -maybe_suppress_warnings(Arg, Vs, PrevBody) -> +maybe_suppress_warnings(Arg, #c_var{name=V}, PrevBody) -> case should_suppress_warning(Arg) of true -> Arg; %Already suppressed. false -> - case is_any_var_used(Vs, PrevBody) of + case core_lib:is_var_used(V, PrevBody) of true -> suppress_warning([Arg]); false -> @@ -2595,7 +2994,7 @@ move_case_into_arg(#c_case{arg=#c_let{vars=OuterVars0,arg=OuterArg, %% in case <InnerArg> of <InnerClauses> end %% ScopeSub0 = sub_subst_scope(Sub#sub{t=#{}}), - {OuterVars,ScopeSub} = pattern_list(OuterVars0, ScopeSub0), + {OuterVars,ScopeSub} = var_list(OuterVars0, ScopeSub0), InnerArg = body(InnerArg0, ScopeSub), Outer#c_let{vars=OuterVars,arg=OuterArg, body=Inner#c_case{arg=InnerArg,clauses=InnerClauses}}; @@ -2624,16 +3023,20 @@ move_case_into_arg(#c_case{arg=#c_case{arg=OuterArg, %% end %% ScopeSub0 = sub_subst_scope(Sub#sub{t=#{}}), - {OuterPats,ScopeSub} = pattern_list(OuterPats0, ScopeSub0), - OuterGuard = guard(OuterGuard0, ScopeSub), - InnerArg = body(InnerArg0, ScopeSub), - Inner = Inner0#c_case{arg=InnerArg,clauses=InnerClauses}, - OuterCa = OuterCa0#c_clause{pats=OuterPats,guard=OuterGuard, - body=Inner}, - Outer#c_case{arg=OuterArg, - clauses=[OuterCa,OuterCb]}; + + %% We KNOW that pattern_list/2 has already been called for OuterPats0; + %% therefore, it cannot throw an exception. + {OuterPats,ScopeSub} = pattern_list(OuterPats0, ScopeSub0), + OuterGuard = guard(OuterGuard0, ScopeSub), + InnerArg = body(InnerArg0, ScopeSub), + Inner = Inner0#c_case{arg=InnerArg,clauses=InnerClauses}, + OuterCa = OuterCa0#c_clause{pats=OuterPats, + guard=OuterGuard, + body=Inner}, + Outer#c_case{arg=OuterArg, + clauses=[OuterCa,OuterCb]}; false -> - impossible + Inner0 end; move_case_into_arg(#c_case{arg=#c_seq{arg=OuterArg,body=InnerArg}=Outer, clauses=InnerClauses}=Inner, _Sub) -> @@ -2649,57 +3052,8 @@ move_case_into_arg(#c_case{arg=#c_seq{arg=OuterArg,body=InnerArg}=Outer, %% Outer#c_seq{arg=OuterArg, body=Inner#c_case{arg=InnerArg,clauses=InnerClauses}}; -move_case_into_arg(_, _) -> - impossible. - -%% In guards only, rewrite a case in a let argument like -%% -%% let <Var> = case <> of -%% <> when AnyGuard -> Literal1; -%% <> when AnyGuard -> Literal2 -%% end -%% in LetBody -%% -%% to -%% -%% case <> of -%% <> when AnyGuard -> -%% let <Var> = Literal1 in LetBody -%% <> when 'true' -> -%% let <Var> = Literal2 in LetBody -%% end -%% -%% In the worst case, the size of the code could increase. -%% In practice, though, substituting the literals into -%% LetBody and doing constant folding will decrease the code -%% size. (Doing this transformation outside of guards could -%% lead to a substantational increase in code size.) -%% -opt_case_in_let_arg(#c_let{arg=#c_case{}=Case}=Let, Ctxt, - #sub{in_guard=true}=Sub) -> - opt_case_in_let_arg_1(Let, Case, Ctxt, Sub); -opt_case_in_let_arg(Let, _, _) -> Let. - -opt_case_in_let_arg_1(Let0, #c_case{arg=#c_values{es=[]}, - clauses=Cs}=Case0, Ctxt, Sub) -> - Let = mark_compiler_generated(Let0), - case Cs of - [#c_clause{body=#c_literal{}=BodyA}=Ca0, - #c_clause{body=#c_literal{}=BodyB}=Cb0] -> - Ca = Ca0#c_clause{body=Let#c_let{arg=BodyA}}, - Cb = Cb0#c_clause{body=Let#c_let{arg=BodyB}}, - Case = Case0#c_case{clauses=[Ca,Cb]}, - expr(Case, Ctxt, sub_new_preserve_types(Sub)); - _ -> Let - end; -opt_case_in_let_arg_1(Let, _, _, _) -> Let. - -is_any_var_used([#c_var{name=V}|Vs], Expr) -> - case core_lib:is_var_used(V, Expr) of - false -> is_any_var_used(Vs, Expr); - true -> true - end; -is_any_var_used([], _) -> false. +move_case_into_arg(Expr, _) -> + Expr. %%% %%% Retrieving information about types. @@ -2793,12 +3147,20 @@ extract_type_1(Expr, Sub) -> true -> bool end. +returns_integer('band', [_,_]) -> true; +returns_integer('bnot', [_]) -> true; +returns_integer('bor', [_,_]) -> true; +returns_integer('bxor', [_,_]) -> true; returns_integer(bit_size, [_]) -> true; returns_integer('bsl', [_,_]) -> true; returns_integer('bsr', [_,_]) -> true; returns_integer(byte_size, [_]) -> true; +returns_integer(ceil, [_]) -> true; +returns_integer('div', [_,_]) -> true; +returns_integer(floor, [_]) -> true; returns_integer(length, [_]) -> true; returns_integer('rem', [_,_]) -> true; +returns_integer('round', [_]) -> true; returns_integer(size, [_]) -> true; returns_integer(tuple_size, [_]) -> true; returns_integer(trunc, [_]) -> true; @@ -2813,15 +3175,8 @@ update_types(Expr, Pat, #sub{t=Tdb0}=Sub) -> Tdb = update_types_1(Expr, Pat, Tdb0), Sub#sub{t=Tdb}. -update_types_1(#c_var{name=V,anno=Anno}, Pat, Types) -> - case member(reuse_for_context, Anno) of - true -> - %% If a variable has been marked for reuse of binary context, - %% optimizations based on type information are unsafe. - kill_types(V, Types); - false -> - update_types_2(V, Pat, Types) - end; +update_types_1(#c_var{name=V}, Pat, Types) -> + update_types_2(V, Pat, Types); update_types_1(_, _, Types) -> Types. update_types_2(V, [#c_tuple{}=P], Types) -> @@ -2864,274 +3219,14 @@ copy_type(_, _, Tdb) -> Tdb. void() -> #c_literal{val=ok}. -%%% -%%% Annotate bit syntax matching to faciliate optimization in further passes. -%%% - -bsm_an(#c_case{arg=#c_var{}=V}=Case) -> - bsm_an_1([V], Case); -bsm_an(#c_case{arg=#c_values{es=Es}}=Case) -> - bsm_an_1(Es, Case); -bsm_an(Other) -> Other. - -bsm_an_1(Vs, #c_case{clauses=Cs}=Case) -> - case bsm_leftmost(Cs) of - none -> Case; - Pos -> bsm_an_2(Vs, Cs, Case, Pos) - end. - -bsm_an_2(Vs, Cs, Case, Pos) -> - case bsm_nonempty(Cs, Pos) of - true -> bsm_an_3(Vs, Cs, Case, Pos); - false -> Case - end. - -bsm_an_3(Vs, Cs, Case, Pos) -> - try - bsm_ensure_no_partition(Cs, Pos), - bsm_do_an(Vs, Pos, Cs, Case) - catch - throw:{problem,Where,What} -> - add_bin_opt_info(Where, What), - Case - end. - -bsm_do_an(Vs0, Pos, Cs0, Case) -> - case nth(Pos, Vs0) of - #c_var{name=Vname}=V0 -> - Cs = bsm_do_an_var(Vname, Pos, Cs0, []), - V = bsm_annotate_for_reuse(V0), - Bef = lists:sublist(Vs0, Pos-1), - Aft = lists:nthtail(Pos, Vs0), - case Bef ++ [V|Aft] of - [_] -> - Case#c_case{arg=V,clauses=Cs}; - Vs -> - Case#c_case{arg=#c_values{es=Vs},clauses=Cs} - end; - _ -> - Case - end. - -bsm_do_an_var(V, S, [#c_clause{pats=Ps,guard=G,body=B0}=C0|Cs], Acc) -> - case nth(S, Ps) of - #c_var{name=VarName} -> - case core_lib:is_var_used(V, G) of - true -> bsm_problem(C0, orig_bin_var_used_in_guard); - false -> ok - end, - case core_lib:is_var_used(VarName, G) of - true -> bsm_problem(C0, bin_var_used_in_guard); - false -> ok - end, - B1 = bsm_maybe_ctx_to_binary(VarName, B0), - B = bsm_maybe_ctx_to_binary(V, B1), - C = C0#c_clause{body=B}, - bsm_do_an_var(V, S, Cs, [C|Acc]); - #c_alias{}=P -> - case bsm_could_match_binary(P) of - false -> - bsm_do_an_var(V, S, Cs, [C0|Acc]); - true -> - bsm_problem(C0, bin_opt_alias) - end; - P -> - case bsm_could_match_binary(P) andalso bsm_is_var_used(V, G, B0) of - false -> - bsm_do_an_var(V, S, Cs, [C0|Acc]); - true -> - bsm_problem(C0, bin_var_used) - end - end; -bsm_do_an_var(_, _, [], Acc) -> reverse(Acc). - -bsm_annotate_for_reuse(#c_var{anno=Anno}=Var) -> - case member(reuse_for_context, Anno) of - false -> Var#c_var{anno=[reuse_for_context|Anno]}; - true -> Var - end. - -bsm_is_var_used(V, G, B) -> - core_lib:is_var_used(V, G) orelse core_lib:is_var_used(V, B). - -bsm_maybe_ctx_to_binary(V, B) -> - case core_lib:is_var_used(V, B) andalso not previous_ctx_to_binary(V, B) of - false -> - B; - true -> - #c_seq{arg=#c_primop{name=#c_literal{val=bs_context_to_binary}, - args=[#c_var{name=V}]}, - body=B} - end. - -previous_ctx_to_binary(V, Core) -> - case Core of - #c_seq{arg=#c_primop{name=#c_literal{val=bs_context_to_binary}, - args=[#c_var{name=V}]}} -> - true; - _ -> - false - end. - -%% bsm_leftmost(Cs) -> none | ArgumentNumber -%% Find the leftmost argument that does binary matching. Return -%% the number of the argument (1-N). - -bsm_leftmost(Cs) -> - bsm_leftmost_1(Cs, none). - -bsm_leftmost_1([#c_clause{pats=Ps}|Cs], Pos) -> - bsm_leftmost_2(Ps, Cs, 1, Pos); -bsm_leftmost_1([], Pos) -> Pos. - -bsm_leftmost_2(_, Cs, Pos, Pos) -> - bsm_leftmost_1(Cs, Pos); -bsm_leftmost_2([#c_binary{}|_], Cs, N, _) -> - bsm_leftmost_1(Cs, N); -bsm_leftmost_2([_|Ps], Cs, N, Pos) -> - bsm_leftmost_2(Ps, Cs, N+1, Pos); -bsm_leftmost_2([], Cs, _, Pos) -> - bsm_leftmost_1(Cs, Pos). - -%% bsm_nonempty(Cs, Pos) -> true|false -%% Check if at least one of the clauses matches a non-empty -%% binary in the given argument position. -%% -bsm_nonempty([#c_clause{pats=Ps}|Cs], Pos) -> - case nth(Pos, Ps) of - #c_binary{segments=[_|_]} -> - true; - _ -> - bsm_nonempty(Cs, Pos) - end; -bsm_nonempty([], _ ) -> false. - -%% bsm_ensure_no_partition(Cs, Pos) -> ok (exception if problem) -%% We must make sure that matching is not partitioned between -%% variables like this: -%% foo(<<...>>) -> ... -%% foo(<Variable>) when ... -> ... -%% foo(<Any non-variable pattern>) -> -%% If there is such partition, we are not allowed to reuse the binary variable -%% for the match context. -%% -%% Also, arguments to the left of the argument that is matched -%% against a binary, are only allowed to be simple variables, not -%% used in guards. The reason is that we must know that the binary is -%% only matched in one place (i.e. there must be only one bs_start_match2 -%% instruction emitted). - -bsm_ensure_no_partition(Cs, Pos) -> - bsm_ensure_no_partition_1(Cs, Pos, before). - -%% Loop through each clause. -bsm_ensure_no_partition_1([#c_clause{pats=Ps,guard=G}|Cs], Pos, State0) -> - State = bsm_ensure_no_partition_2(Ps, Pos, G, simple_vars, State0), - case State of - 'after' -> - bsm_ensure_no_partition_after(Cs, Pos); - _ -> - ok - end, - bsm_ensure_no_partition_1(Cs, Pos, State); -bsm_ensure_no_partition_1([], _, _) -> ok. - -%% Loop through each pattern for this clause. -bsm_ensure_no_partition_2([#c_binary{}=Where|_], 1, _, Vstate, State) -> - case State of - before when Vstate =:= simple_vars -> within; - before -> bsm_problem(Where, Vstate); - within when Vstate =:= simple_vars -> within; - within -> bsm_problem(Where, Vstate) - end; -bsm_ensure_no_partition_2([#c_alias{}=Alias|_], 1, N, Vstate, State) -> - %% Retrieve the real pattern that the alias refers to and check that. - P = bsm_real_pattern(Alias), - bsm_ensure_no_partition_2([P], 1, N, Vstate, State); -bsm_ensure_no_partition_2([_|_], 1, _, _Vstate, before=State) -> - %% No binary matching yet - therefore no partition. - State; -bsm_ensure_no_partition_2([P|_], 1, _, Vstate, State) -> - case bsm_could_match_binary(P) of - false -> - %% If clauses can be freely arranged (Vstate =:= simple_vars), - %% a clause that cannot match a binary will not partition the clause. - %% Example: - %% - %% a(Var, <<>>) -> ... - %% a(Var, []) -> ... - %% a(Var, <<B>>) -> ... - %% - %% But if the clauses can't be freely rearranged, as in - %% - %% b(Var, <<X>>) -> ... - %% b(1, 2) -> ... - %% - %% we do have a problem. - %% - case Vstate of - simple_vars -> State; - _ -> bsm_problem(P, Vstate) - end; - true -> - %% The pattern P *may* match a binary, so we must update the state. - %% (P must be a variable.) - case State of - within -> 'after'; - 'after' -> 'after' - end - end; -bsm_ensure_no_partition_2([#c_var{name=V}|Ps], N, G, Vstate, S) -> - case core_lib:is_var_used(V, G) of - false -> - bsm_ensure_no_partition_2(Ps, N-1, G, Vstate, S); - true -> - bsm_ensure_no_partition_2(Ps, N-1, G, bin_left_var_used_in_guard, S) - end; -bsm_ensure_no_partition_2([_|Ps], N, G, _, S) -> - bsm_ensure_no_partition_2(Ps, N-1, G, bin_argument_order, S). - -bsm_ensure_no_partition_after([#c_clause{pats=Ps}=C|Cs], Pos) -> - case nth(Pos, Ps) of - #c_var{} -> - bsm_ensure_no_partition_after(Cs, Pos); - _ -> - bsm_problem(C, bin_partition) - end; -bsm_ensure_no_partition_after([], _) -> ok. - -bsm_could_match_binary(#c_alias{pat=P}) -> bsm_could_match_binary(P); -bsm_could_match_binary(#c_cons{}) -> false; -bsm_could_match_binary(#c_tuple{}) -> false; -bsm_could_match_binary(#c_literal{val=Lit}) -> is_bitstring(Lit); -bsm_could_match_binary(_) -> true. - -bsm_real_pattern(#c_alias{pat=P}) -> bsm_real_pattern(P); -bsm_real_pattern(P) -> P. - -bsm_problem(Where, What) -> - throw({problem,Where,What}). %%% %%% Handling of warnings. %%% -mark_compiler_generated(Term) -> - cerl_trees:map(fun mark_compiler_generated_1/1, Term). - -mark_compiler_generated_1(#c_call{anno=Anno}=Term) -> - Term#c_call{anno=[compiler_generated|Anno--[compiler_generated]]}; -mark_compiler_generated_1(Term) -> Term. - init_warnings() -> put({?MODULE,warnings}, []). -add_bin_opt_info(Core, Term) -> - case get(bin_opt_info) of - true -> add_warning(Core, Term); - false -> ok - end. - add_warning(Core, Term) -> case should_suppress_warning(Core) of true -> @@ -3207,6 +3302,29 @@ format_error(nomatch_shadow) -> "this clause cannot match because a previous clause always matches"; format_error(nomatch_guard) -> "the guard for this clause evaluates to 'false'"; +format_error({nomatch_bit_syntax_truncated,Signess,Val,Sz}) -> + S = case Signess of + signed -> "a 'signed'"; + unsigned -> "an 'unsigned'" + end, + F = "this clause cannot match because the value ~P" + " will not fit in ~s binary segment of size ~p", + flatten(io_lib:format(F, [Val,10,S,Sz])); +format_error({nomatch_bit_syntax_unsigned,Val}) -> + F = "this clause cannot match because the negative value ~P" + " will never match the value of an 'unsigned' binary segment", + flatten(io_lib:format(F, [Val,10])); +format_error({nomatch_bit_syntax_size,Sz}) -> + F = "this clause cannot match because '~P' is not a valid size for a binary segment", + flatten(io_lib:format(F, [Sz,10])); +format_error({nomatch_bit_syntax_type,Val,Type}) -> + F = "this clause cannot match because '~P' is not of the" + " expected type '~p'", + flatten(io_lib:format(F, [Val,10,Type])); +format_error({nomatch_bit_syntax_unicode,Val}) -> + F = "this clause cannot match because the value ~p" + " is not a valid Unicode code point", + flatten(io_lib:format(F, [Val])); format_error(no_clause_match) -> "no clause will ever match"; format_error(nomatch_clause_type) -> @@ -3227,29 +3345,10 @@ format_error({no_effect,{erlang,F,A}}) -> format_error(result_ignored) -> "the result of the expression is ignored " "(suppress the warning by assigning the expression to the _ variable)"; +format_error(invalid_call) -> + "invalid function call"; format_error(useless_building) -> - "a term is constructed, but never used"; -format_error(bin_opt_alias) -> - "INFO: the '=' operator will prevent delayed sub binary optimization"; -format_error(bin_partition) -> - "INFO: matching non-variables after a previous clause matching a variable " - "will prevent delayed sub binary optimization"; -format_error(bin_left_var_used_in_guard) -> - "INFO: a variable to the left of the binary pattern is used in a guard; " - "will prevent delayed sub binary optimization"; -format_error(bin_argument_order) -> - "INFO: matching anything else but a plain variable to the left of " - "binary pattern will prevent delayed sub binary optimization; " - "SUGGEST changing argument order"; -format_error(bin_var_used) -> - "INFO: using a matched out sub binary will prevent " - "delayed sub binary optimization"; -format_error(orig_bin_var_used_in_guard) -> - "INFO: using the original binary variable in a guard will prevent " - "delayed sub binary optimization"; -format_error(bin_var_used_in_guard) -> - "INFO: using a matched out sub binary in a guard will prevent " - "delayed sub binary optimization". + "a term is constructed, but never used". -ifdef(DEBUG). %% In order for simplify_let/2 to work correctly, the list of @@ -3260,12 +3359,18 @@ format_error(bin_var_used_in_guard) -> verify_scope(E, #sub{s=Scope}) -> Free0 = cerl_trees:free_variables(E), Free = [V || V <- Free0, not is_tuple(V)], %Ignore function names. - case ordsets:is_subset(Free, cerl_sets:to_list(Scope)) of - true -> true; + case is_subset_of_scope(Free, Scope) of + true -> + true; false -> io:format("~p\n", [E]), io:format("~p\n", [Free]), - io:format("~p\n", [cerl_sets:to_list(Scope)]), + io:format("~p\n", [ordsets:from_list(cerl_sets:to_list(Scope))]), false end. + +is_subset_of_scope([V|Vs], Scope) -> + cerl_sets:is_element(V, Scope) andalso is_subset_of_scope(Vs, Scope); +is_subset_of_scope([], _) -> true. + -endif. diff --git a/lib/compiler/src/sys_core_inline.erl b/lib/compiler/src/sys_core_inline.erl index 838dda68c6..5a6cc45e4a 100644 --- a/lib/compiler/src/sys_core_inline.erl +++ b/lib/compiler/src/sys_core_inline.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2000-2010. All Rights Reserved. +%% Copyright Ericsson AB 2000-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -143,7 +143,7 @@ inline_inline(#ifun{body=B,weight=Iw}=If, Is) -> case find_inl(F, A, Is) of #ifun{vars=Vs,body=B2,weight=W} when W < Iw -> #c_let{vars=Vs, - arg=core_lib:make_values(As), + arg=kill_id_anns(core_lib:make_values(As)), body=kill_id_anns(B2)}; _Other -> Call end; @@ -160,7 +160,7 @@ inline_func(#fstat{def={Name,F0}}=Fstat, Is) -> case find_inl(F, A, Is) of #ifun{vars=Vs,body=B} -> {#c_let{vars=Vs, - arg=core_lib:make_values(As), + arg=kill_id_anns(core_lib:make_values(As)), body=kill_id_anns(B)}, true}; %Have modified _Other -> {Call,Mod} diff --git a/lib/compiler/src/sys_pre_attributes.erl b/lib/compiler/src/sys_pre_attributes.erl index f0cb630205..67adae5acf 100644 --- a/lib/compiler/src/sys_pre_attributes.erl +++ b/lib/compiler/src/sys_pre_attributes.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1998-2009. All Rights Reserved. +%% Copyright Ericsson AB 1998-2016. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -25,10 +25,10 @@ -define(OPTION_TAG, attributes). --record(state, {forms, - pre_ops = [], - post_ops = [], - options}). +-record(state, {forms :: [form()], + pre_ops = [] :: [op()], + post_ops = [] :: [op()], + options :: [option()]}). %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Inserts, deletes and replaces Erlang compiler attributes. @@ -59,9 +59,23 @@ %% due to that the pre_transform pass did not find the attribute plus %% all insert operations. +-type attribute() :: atom(). +-type value() :: term(). +-type form() :: {function, integer(), atom(), arity(), _} + | {attribute, integer(), attribute(), _}. +-type option() :: compile:option() + | {'attribute', 'insert', attribute(), value()} + | {'attribute', 'replace', attribute(), value()} + | {'attribute', 'delete', attribute()}. +-type op() :: {'insert', attribute(), value()} + | {'replace', attribute(), value()} + | {'delete', attribute()}. + +-spec parse_transform([form()], [option()]) -> [form()]. + parse_transform(Forms, Options) -> S = #state{forms = Forms, options = Options}, - S2 = init_transform(S), + S2 = init_transform(Options, S), report_verbose("Pre options: ~p~n", [S2#state.pre_ops], S2), report_verbose("Post options: ~p~n", [S2#state.post_ops], S2), S3 = pre_transform(S2), @@ -71,13 +85,6 @@ parse_transform(Forms, Options) -> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Computes the lists of pre_ops and post_ops that are %% used in the real transformation. -init_transform(S) -> - case S#state.options of - Options when is_list(Options) -> - init_transform(Options, S); - Option -> - init_transform([Option], S) - end. init_transform([{attribute, insert, Name, Val} | Tail], S) -> Op = {insert, Name, Val}, @@ -92,12 +99,9 @@ init_transform([{attribute, delete, Name} | Tail], S) -> Op = {delete, Name}, PreOps = [Op | S#state.pre_ops], init_transform(Tail, S#state{pre_ops = PreOps}); -init_transform([], S) -> - S; init_transform([_ | T], S) -> init_transform(T, S); -init_transform(BadOpt, S) -> - report_error("Illegal option (ignored): ~p~n", [BadOpt], S), +init_transform([], S) -> S. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% @@ -176,18 +180,9 @@ attrs([], _, _) -> %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Report functions. %% -%% Errors messages are controlled with the 'report_errors' compiler option %% Warning messages are controlled with the 'report_warnings' compiler option %% Verbose messages are controlled with the 'verbose' compiler option -report_error(Format, Args, S) -> - case is_error(S) of - true -> - io:format("~p: * ERROR * " ++ Format, [?MODULE | Args]); - false -> - ok - end. - report_warning(Format, Args, S) -> case is_warning(S) of true -> @@ -204,9 +199,6 @@ report_verbose(Format, Args, S) -> ok end. -is_error(S) -> - lists:member(report_errors, S#state.options) or is_verbose(S). - is_warning(S) -> lists:member(report_warnings, S#state.options) or is_verbose(S). diff --git a/lib/compiler/src/sys_pre_expand.erl b/lib/compiler/src/sys_pre_expand.erl deleted file mode 100644 index d9cc4b530c..0000000000 --- a/lib/compiler/src/sys_pre_expand.erl +++ /dev/null @@ -1,671 +0,0 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 1996-2015. All Rights Reserved. -%% -%% Licensed under the Apache License, Version 2.0 (the "License"); -%% you may not use this file except in compliance with the License. -%% You may obtain a copy of the License at -%% -%% http://www.apache.org/licenses/LICENSE-2.0 -%% -%% Unless required by applicable law or agreed to in writing, software -%% distributed under the License is distributed on an "AS IS" BASIS, -%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -%% See the License for the specific language governing permissions and -%% limitations under the License. -%% -%% %CopyrightEnd% -%% -%% Purpose : Expand some source Erlang constructions. This is part of the -%% pre-processing phase. - -%% N.B. Although structs (tagged tuples) are not yet allowed in the -%% language there is code included in pattern/2 and expr/3 (commented out) -%% that handles them by transforming them to tuples. - --module(sys_pre_expand). - -%% Main entry point. --export([module/2]). - --import(ordsets, [from_list/1,union/2]). --import(lists, [member/2,foldl/3,foldr/3]). - --include("../include/erl_bits.hrl"). - --type fa() :: {atom(), arity()}. - --record(expand, {module=[], %Module name - exports=[], %Exports - imports=[], %Imports - attributes=[], %Attributes - callbacks=[], %Callbacks - optional_callbacks=[] :: [fa()], %Optional callbacks - defined, %Defined functions (gb_set) - vcount=0, %Variable counter - func=[], %Current function - arity=[], %Arity for current function - fcount=0 %Local fun count - }). - -%% module(Forms, CompileOptions) -%% {ModuleName,Exports,TransformedForms,CompileOptions'} -%% Expand the forms in one module. N.B.: the lists of predefined -%% exports and imports are really ordsets! -%% CompileOptions is augmented with options from -compile attributes. - -module(Fs0, Opts0) -> - - %% Expand records. Normalise guard tests. - Fs = erl_expand_records:module(Fs0, Opts0), - - Opts = compiler_options(Fs) ++ Opts0, - - %% Set pre-defined exported functions. - PreExp = [{module_info,0},{module_info,1}], - - %% Build initial expand record. - St0 = #expand{exports=PreExp, - defined=PreExp - }, - %% Expand the functions. - {Tfs,St1} = forms(Fs, define_functions(Fs, St0)), - %% Get the correct list of exported functions. - Exports = case member(export_all, Opts) of - true -> gb_sets:to_list(St1#expand.defined); - false -> St1#expand.exports - end, - %% Generate all functions from stored info. - {Ats,St3} = module_attrs(St1#expand{exports = Exports}), - {Mfs,St4} = module_predef_funcs(St3), - {St4#expand.module, St4#expand.exports, Ats ++ Tfs ++ Mfs, - Opts}. - -compiler_options(Forms) -> - lists:flatten([C || {attribute,_,compile,C} <- Forms]). - -%% define_function(Form, State) -> State. -%% Add function to defined if form is a function. - -define_functions(Forms, #expand{defined=Predef}=St) -> - Fs = foldl(fun({function,_,N,A,_Cs}, Acc) -> [{N,A}|Acc]; - (_, Acc) -> Acc - end, Predef, Forms), - St#expand{defined=gb_sets:from_list(Fs)}. - -module_attrs(#expand{attributes=Attributes}=St) -> - Attrs = [{attribute,Line,Name,Val} || {Name,Line,Val} <- Attributes], - Callbacks = [Callback || {_,_,callback,_}=Callback <- Attrs], - OptionalCallbacks = get_optional_callbacks(Attrs), - {Attrs,St#expand{callbacks=Callbacks, - optional_callbacks=OptionalCallbacks}}. - -get_optional_callbacks(Attrs) -> - L = [O || - {attribute, _, optional_callbacks, O} <- Attrs, - is_fa_list(O)], - lists:append(L). - -is_fa_list([{FuncName, Arity}|L]) - when is_atom(FuncName), is_integer(Arity), Arity >= 0 -> - is_fa_list(L); -is_fa_list([]) -> true; -is_fa_list(_) -> false. - -module_predef_funcs(St) -> - {Mpf1,St1}=module_predef_func_beh_info(St), - {Mpf2,St2}=module_predef_funcs_mod_info(St1), - Mpf = [erl_parse:new_anno(F) || F <- Mpf1++Mpf2], - {Mpf,St2}. - -module_predef_func_beh_info(#expand{callbacks=[]}=St) -> - {[], St}; -module_predef_func_beh_info(#expand{callbacks=Callbacks, - optional_callbacks=OptionalCallbacks, - defined=Defined, - exports=Exports}=St) -> - PreDef=[{behaviour_info,1}], - PreExp=PreDef, - {[gen_beh_info(Callbacks, OptionalCallbacks)], - St#expand{defined=gb_sets:union(gb_sets:from_list(PreDef), Defined), - exports=union(from_list(PreExp), Exports)}}. - -gen_beh_info(Callbacks, OptionalCallbacks) -> - List = make_list(Callbacks), - OptionalList = make_optional_list(OptionalCallbacks), - {function,0,behaviour_info,1, - [{clause,0,[{atom,0,callbacks}],[], - [List]}, - {clause,0,[{atom,0,optional_callbacks}],[], - [OptionalList]}]}. - -make_list([]) -> {nil,0}; -make_list([{_,_,_,[{{Name,Arity},_}]}|Rest]) -> - {cons,0, - {tuple,0, - [{atom,0,Name}, - {integer,0,Arity}]}, - make_list(Rest)}. - -make_optional_list([]) -> {nil,0}; -make_optional_list([{Name,Arity}|Rest]) -> - {cons,0, - {tuple,0, - [{atom,0,Name}, - {integer,0,Arity}]}, - make_optional_list(Rest)}. - -module_predef_funcs_mod_info(St) -> - PreDef = [{module_info,0},{module_info,1}], - PreExp = PreDef, - {[{function,0,module_info,0, - [{clause,0,[],[], - [{call,0,{remote,0,{atom,0,erlang},{atom,0,get_module_info}}, - [{atom,0,St#expand.module}]}]}]}, - {function,0,module_info,1, - [{clause,0,[{var,0,'X'}],[], - [{call,0,{remote,0,{atom,0,erlang},{atom,0,get_module_info}}, - [{atom,0,St#expand.module},{var,0,'X'}]}]}]}], - St#expand{defined=gb_sets:union(gb_sets:from_list(PreDef), - St#expand.defined), - exports=union(from_list(PreExp), St#expand.exports)}}. - -%% forms(Forms, State) -> -%% {TransformedForms,State'} -%% Process the forms. Attributes are lost and just affect the state. -%% Ignore uninteresting forms like eof and type. - -forms([{attribute,_,file,_File}=F|Fs0], St0) -> - {Fs,St1} = forms(Fs0, St0), - {[F|Fs],St1}; -forms([{attribute,Line,Name,Val}|Fs0], St0) -> - St1 = attribute(Name, Val, Line, St0), - forms(Fs0, St1); -forms([{function,L,N,A,Cs}|Fs0], St0) -> - {Ff,St1} = function(L, N, A, Cs, St0), - {Fs,St2} = forms(Fs0, St1), - {[Ff|Fs],St2}; -forms([_|Fs], St) -> forms(Fs, St); -forms([], St) -> {[],St}. - -%% attribute(Attribute, Value, Line, State) -> State'. -%% Process an attribute, this just affects the state. - -attribute(module, Module, _L, St) -> - true = is_atom(Module), - St#expand{module=Module}; -attribute(export, Es, _L, St) -> - St#expand{exports=union(from_list(Es), St#expand.exports)}; -attribute(import, Is, _L, St) -> - import(Is, St); -attribute(compile, _C, _L, St) -> - St; -attribute(Name, Val, Line, St) when is_list(Val) -> - St#expand{attributes=St#expand.attributes ++ [{Name,Line,Val}]}; -attribute(Name, Val, Line, St) -> - St#expand{attributes=St#expand.attributes ++ [{Name,Line,[Val]}]}. - -function(L, N, A, Cs0, St0) -> - {Cs,St} = clauses(Cs0, St0#expand{func=N,arity=A,fcount=0}), - {{function,L,N,A,Cs},St}. - -%% clauses([Clause], State) -> -%% {[TransformedClause],State}. -%% Expand function clauses. - -clauses([{clause,Line,H0,G0,B0}|Cs0], St0) -> - {H,St1} = head(H0, St0), - {G,St2} = guard(G0, St1), - {B,St3} = exprs(B0, St2), - {Cs,St4} = clauses(Cs0, St3), - {[{clause,Line,H,G,B}|Cs],St4}; -clauses([], St) -> {[],St}. - -%% head(HeadPatterns, State) -> -%% {TransformedPatterns,Variables,UsedVariables,State'} - -head(As, St) -> pattern_list(As, St). - -%% pattern(Pattern, State) -> -%% {TransformedPattern,State'} -%% - -pattern({var,_,'_'}=Var, St) -> %Ignore anonymous variable. - {Var,St}; -pattern({var,_,_}=Var, St) -> - {Var,St}; -pattern({char,_,_}=Char, St) -> - {Char,St}; -pattern({integer,_,_}=Int, St) -> - {Int,St}; -pattern({float,_,_}=Float, St) -> - {Float,St}; -pattern({atom,_,_}=Atom, St) -> - {Atom,St}; -pattern({string,_,_}=String, St) -> - {String,St}; -pattern({nil,_}=Nil, St) -> - {Nil,St}; -pattern({cons,Line,H,T}, St0) -> - {TH,St1} = pattern(H, St0), - {TT,St2} = pattern(T, St1), - {{cons,Line,TH,TT},St2}; -pattern({tuple,Line,Ps}, St0) -> - {TPs,St1} = pattern_list(Ps, St0), - {{tuple,Line,TPs},St1}; -pattern({map,Line,Ps}, St0) -> - {TPs,St1} = pattern_list(Ps, St0), - {{map,Line,TPs},St1}; -pattern({map_field_exact,Line,K0,V0}, St0) -> - %% Key should be treated as an expression - %% but since expressions are not allowed yet, - %% process it through pattern .. and handle assoc - %% (normalise unary op integer -> integer) - {K,St1} = pattern(K0, St0), - {V,St2} = pattern(V0, St1), - {{map_field_exact,Line,K,V},St2}; -pattern({map_field_assoc,Line,K0,V0}, St0) -> - %% when keys are Maps - {K,St1} = pattern(K0, St0), - {V,St2} = pattern(V0, St1), - {{map_field_assoc,Line,K,V},St2}; -%%pattern({struct,Line,Tag,Ps}, St0) -> -%% {TPs,TPsvs,St1} = pattern_list(Ps, St0), -%% {{tuple,Line,[{atom,Line,Tag}|TPs]},TPsvs,St1}; -pattern({bin,Line,Es0}, St0) -> - {Es1,St1} = pattern_bin(Es0, St0), - {{bin,Line,Es1},St1}; -pattern({op,_,'++',{nil,_},R}, St) -> - pattern(R, St); -pattern({op,_,'++',{cons,Li,H,T},R}, St) -> - pattern({cons,Li,H,{op,Li,'++',T,R}}, St); -pattern({op,_,'++',{string,Li,L},R}, St) -> - pattern(string_to_conses(Li, L, R), St); -pattern({match,Line,Pat1, Pat2}, St0) -> - {TH,St1} = pattern(Pat2, St0), - {TT,St2} = pattern(Pat1, St1), - {{match,Line,TT,TH},St2}; -%% Compile-time pattern expressions, including unary operators. -pattern({op,_Line,_Op,_A}=Op, St) -> - {erl_eval:partial_eval(Op),St}; -pattern({op,_Line,_Op,_L,_R}=Op, St) -> - {erl_eval:partial_eval(Op),St}. - -pattern_list([P0|Ps0], St0) -> - {P,St1} = pattern(P0, St0), - {Ps,St2} = pattern_list(Ps0, St1), - {[P|Ps],St2}; -pattern_list([], St) -> {[],St}. - -%% guard(Guard, State) -> -%% {TransformedGuard,State'} -%% Transform a list of guard tests. We KNOW that this has been checked -%% and what the guards test are. Use expr for transforming the guard -%% expressions. - -guard([G0|Gs0], St0) -> - {G,St1} = guard_tests(G0, St0), - {Gs,St2} = guard(Gs0, St1), - {[G|Gs],St2}; -guard([], St) -> {[],St}. - -guard_tests([Gt0|Gts0], St0) -> - {Gt1,St1} = guard_test(Gt0, St0), - {Gts1,St2} = guard_tests(Gts0, St1), - {[Gt1|Gts1],St2}; -guard_tests([], St) -> {[],St}. - -guard_test(Test, St) -> - expr(Test, St). - -%% exprs(Expressions, State) -> -%% {TransformedExprs,State'} - -exprs([E0|Es0], St0) -> - {E,St1} = expr(E0, St0), - {Es,St2} = exprs(Es0, St1), - {[E|Es],St2}; -exprs([], St) -> {[],St}. - -%% expr(Expression, State) -> -%% {TransformedExpression,State'} - -expr({var,_,_}=Var, St) -> - {Var,St}; -expr({char,_,_}=Char, St) -> - {Char,St}; -expr({integer,_,_}=Int, St) -> - {Int,St}; -expr({float,_,_}=Float, St) -> - {Float,St}; -expr({atom,_,_}=Atom, St) -> - {Atom,St}; -expr({string,_,_}=String, St) -> - {String,St}; -expr({nil,_}=Nil, St) -> - {Nil,St}; -expr({cons,Line,H0,T0}, St0) -> - {H,St1} = expr(H0, St0), - {T,St2} = expr(T0, St1), - {{cons,Line,H,T},St2}; -expr({lc,Line,E0,Qs0}, St0) -> - {Qs1,St1} = lc_tq(Line, Qs0, St0), - {E1,St2} = expr(E0, St1), - {{lc,Line,E1,Qs1},St2}; -expr({bc,Line,E0,Qs0}, St0) -> - {Qs1,St1} = lc_tq(Line, Qs0, St0), - {E1,St2} = expr(E0, St1), - {{bc,Line,E1,Qs1},St2}; -expr({tuple,Line,Es0}, St0) -> - {Es1,St1} = expr_list(Es0, St0), - {{tuple,Line,Es1},St1}; -%%expr({struct,Line,Tag,Es0}, Vs, St0) -> -%% {Es1,Esvs,Esus,St1} = expr_list(Es0, Vs, St0), -%% {{tuple,Line,[{atom,Line,Tag}|Es1]},Esvs,Esus,St1}; -expr({map,Line,Es0}, St0) -> - {Es1,St1} = expr_list(Es0, St0), - {{map,Line,Es1},St1}; -expr({map,Line,E0,Es0}, St0) -> - {E1,St1} = expr(E0, St0), - {Es1,St2} = expr_list(Es0, St1), - {{map,Line,E1,Es1},St2}; -expr({map_field_assoc,Line,K0,V0}, St0) -> - {K,St1} = expr(K0, St0), - {V,St2} = expr(V0, St1), - {{map_field_assoc,Line,K,V},St2}; -expr({map_field_exact,Line,K0,V0}, St0) -> - {K,St1} = expr(K0, St0), - {V,St2} = expr(V0, St1), - {{map_field_exact,Line,K,V},St2}; -expr({bin,Line,Es0}, St0) -> - {Es1,St1} = expr_bin(Es0, St0), - {{bin,Line,Es1},St1}; -expr({block,Line,Es0}, St0) -> - {Es,St1} = exprs(Es0, St0), - {{block,Line,Es},St1}; -expr({'if',Line,Cs0}, St0) -> - {Cs,St1} = icr_clauses(Cs0, St0), - {{'if',Line,Cs},St1}; -expr({'case',Line,E0,Cs0}, St0) -> - {E,St1} = expr(E0, St0), - {Cs,St2} = icr_clauses(Cs0, St1), - {{'case',Line,E,Cs},St2}; -expr({'receive',Line,Cs0}, St0) -> - {Cs,St1} = icr_clauses(Cs0, St0), - {{'receive',Line,Cs},St1}; -expr({'receive',Line,Cs0,To0,ToEs0}, St0) -> - {To,St1} = expr(To0, St0), - {ToEs,St2} = exprs(ToEs0, St1), - {Cs,St3} = icr_clauses(Cs0, St2), - {{'receive',Line,Cs,To,ToEs},St3}; -expr({'fun',Line,Body}, St) -> - fun_tq(Line, Body, St); -expr({named_fun,Line,Name,Cs}, St) -> - fun_tq(Line, Cs, St, Name); -expr({call,Line,{atom,La,N}=Atom,As0}, St0) -> - {As,St1} = expr_list(As0, St0), - Ar = length(As), - case defined(N,Ar,St1) of - true -> - {{call,Line,Atom,As},St1}; - _ -> - case imported(N, Ar, St1) of - {yes,Mod} -> - {{call,Line,{remote,La,{atom,La,Mod},Atom},As},St1}; - no -> - case erl_internal:bif(N, Ar) of - true -> - {{call,Line,{remote,La,{atom,La,erlang},Atom},As},St1}; - false -> %% This should have been handled by erl_lint - {{call,Line,Atom,As},St1} - end - end - end; -expr({call,Line,{remote,Lr,M0,F},As0}, St0) -> - {[M1,F1|As1],St1} = expr_list([M0,F|As0], St0), - {{call,Line,{remote,Lr,M1,F1},As1},St1}; -expr({call,Line,F,As0}, St0) -> - {[Fun1|As1],St1} = expr_list([F|As0], St0), - {{call,Line,Fun1,As1},St1}; -expr({'try',Line,Es0,Scs0,Ccs0,As0}, St0) -> - {Es1,St1} = exprs(Es0, St0), - {Scs1,St2} = icr_clauses(Scs0, St1), - {Ccs1,St3} = icr_clauses(Ccs0, St2), - {As1,St4} = exprs(As0, St3), - {{'try',Line,Es1,Scs1,Ccs1,As1},St4}; -expr({'catch',Line,E0}, St0) -> - %% Catch exports no new variables. - {E,St1} = expr(E0, St0), - {{'catch',Line,E},St1}; -expr({match,Line,P0,E0}, St0) -> - {E,St1} = expr(E0, St0), - {P,St2} = pattern(P0, St1), - {{match,Line,P,E},St2}; -expr({op,Line,Op,A0}, St0) -> - {A,St1} = expr(A0, St0), - {{op,Line,Op,A},St1}; -expr({op,Line,Op,L0,R0}, St0) -> - {L,St1} = expr(L0, St0), - {R,St2} = expr(R0, St1), - {{op,Line,Op,L,R},St2}. - -expr_list([E0|Es0], St0) -> - {E,St1} = expr(E0, St0), - {Es,St2} = expr_list(Es0, St1), - {[E|Es],St2}; -expr_list([], St) -> {[],St}. - -%% icr_clauses([Clause], State) -> {[TransformedClause],State'} -%% Be very careful here to return the variables that are really used -%% and really new. - -icr_clauses([], St) -> {[],St}; -icr_clauses(Clauses, St) -> icr_clauses2(Clauses, St). - -icr_clauses2([{clause,Line,H0,G0,B0}|Cs0], St0) -> - {H,St1} = head(H0, St0), - {G,St2} = guard(G0, St1), - {B,St3} = exprs(B0, St2), - {Cs,St4} = icr_clauses2(Cs0, St3), - {[{clause,Line,H,G,B}|Cs],St4}; -icr_clauses2([], St) -> {[],St}. - -%% lc_tq(Line, Qualifiers, State) -> -%% {[TransQual],State'} - -lc_tq(Line, [{generate,Lg,P0,G0} | Qs0], St0) -> - {G1,St1} = expr(G0, St0), - {P1,St2} = pattern(P0, St1), - {Qs1,St3} = lc_tq(Line, Qs0, St2), - {[{generate,Lg,P1,G1} | Qs1],St3}; - -lc_tq(Line, [{b_generate,Lg,P0,G0}|Qs0], St0) -> - {G1,St1} = expr(G0, St0), - {P1,St2} = pattern(P0, St1), - {Qs1,St3} = lc_tq(Line, Qs0, St2), - {[{b_generate,Lg,P1,G1}|Qs1],St3}; -lc_tq(Line, [F0 | Qs0], St0) -> - case erl_lint:is_guard_test(F0) of - true -> - {F1,St1} = guard_test(F0, St0), - {Qs1,St2} = lc_tq(Line, Qs0, St1), - {[F1|Qs1],St2}; - false -> - {F1,St1} = expr(F0, St0), - {Qs1,St2} = lc_tq(Line, Qs0, St1), - {[F1 | Qs1],St2} - end; -lc_tq(_Line, [], St0) -> - {[],St0}. - - -%% fun_tq(Line, Body, State) -> -%% {Fun,State'} -%% Transform an "explicit" fun {'fun', Line, {clauses, Cs}} into an -%% extended form {'fun', Line, {clauses, Cs}, Info}, unless it is the -%% name of a BIF (erl_lint has checked that it is not an import). -%% "Implicit" funs {'fun', Line, {function, F, A}} are not changed. - -fun_tq(Lf, {function,F,A}=Function, St0) -> - case erl_internal:bif(F, A) of - true -> - {As,St1} = new_vars(A, Lf, St0), - Cs = [{clause,Lf,As,[],[{call,Lf,{atom,Lf,F},As}]}], - fun_tq(Lf, {clauses,Cs}, St1); - false -> - {Fname,St1} = new_fun_name(St0), - Index = Uniq = 0, - {{'fun',Lf,Function,{Index,Uniq,Fname}},St1} - end; -fun_tq(L, {function,M,F,A}, St) when is_atom(M), is_atom(F), is_integer(A) -> - %% This is the old format for external funs, generated by a pre-R15 - %% compiler. That means that a tool, such as the debugger or xref, - %% directly invoked this module with the abstract code from a - %% pre-R15 BEAM file. Be helpful, and translate it to the new format. - fun_tq(L, {function,{atom,L,M},{atom,L,F},{integer,L,A}}, St); -fun_tq(Lf, {function,_,_,_}=ExtFun, St) -> - {{'fun',Lf,ExtFun},St}; -fun_tq(Lf, {clauses,Cs0}, St0) -> - {Cs1,St1} = fun_clauses(Cs0, St0), - {Fname,St2} = new_fun_name(St1), - %% Set dummy values for Index and Uniq -- the real values will - %% be assigned by beam_asm. - Index = Uniq = 0, - {{'fun',Lf,{clauses,Cs1},{Index,Uniq,Fname}},St2}. - -fun_tq(Line, Cs0, St0, Name) -> - {Cs1,St1} = fun_clauses(Cs0, St0), - {Fname,St2} = new_fun_name(St1, Name), - {{named_fun,Line,Name,Cs1,{0,0,Fname}},St2}. - -fun_clauses([{clause,L,H0,G0,B0}|Cs0], St0) -> - {H,St1} = head(H0, St0), - {G,St2} = guard(G0, St1), - {B,St3} = exprs(B0, St2), - {Cs,St4} = fun_clauses(Cs0, St3), - {[{clause,L,H,G,B}|Cs],St4}; -fun_clauses([], St) -> {[],St}. - -%% new_fun_name(State) -> {FunName,State}. - -new_fun_name(St) -> - new_fun_name(St, 'fun'). - -new_fun_name(#expand{func=F,arity=A,fcount=I}=St, FName) -> - Name = "-" ++ atom_to_list(F) ++ "/" ++ integer_to_list(A) - ++ "-" ++ atom_to_list(FName) ++ "-" ++ integer_to_list(I) ++ "-", - {list_to_atom(Name),St#expand{fcount=I+1}}. - -%% pattern_bin([Element], State) -> {[Element],[Variable],[UsedVar],State}. - -pattern_bin(Es0, St) -> - Es1 = bin_expand_strings(Es0), - foldr(fun (E, Acc) -> pattern_element(E, Acc) end, {[],St}, Es1). - -pattern_element({bin_element,Line,Expr0,Size0,Type0}, {Es,St0}) -> - {Expr1,St1} = pattern(Expr0, St0), - {Size1,St2} = pat_bit_size(Size0, St1), - {Size,Type} = make_bit_type(Line, Size1, Type0), - Expr = coerce_to_float(Expr1, Type0), - {[{bin_element,Line,Expr,Size,Type}|Es],St2}. - -pat_bit_size(default, St) -> {default,St}; -pat_bit_size({atom,_La,all}=All, St) -> {All,St}; -pat_bit_size({var,_Lv,_V}=Var, St) -> {Var,St}; -pat_bit_size(Size, St) -> - Line = element(2, Size), - {value,Sz,_} = erl_eval:expr(Size, erl_eval:new_bindings()), - {{integer,Line,Sz},St}. - -make_bit_type(Line, default, Type0) -> - case erl_bits:set_bit_type(default, Type0) of - {ok,all,Bt} -> {{atom,Line,all},erl_bits:as_list(Bt)}; - {ok,undefined,Bt} -> {{atom,Line,undefined},erl_bits:as_list(Bt)}; - {ok,Size,Bt} -> {{integer,Line,Size},erl_bits:as_list(Bt)} - end; -make_bit_type(_Line, Size, Type0) -> %Integer or 'all' - {ok,Size,Bt} = erl_bits:set_bit_type(Size, Type0), - {Size,erl_bits:as_list(Bt)}. - -coerce_to_float({integer,L,I}=E, [float|_]) -> - try - {float,L,float(I)} - catch - error:badarg -> E; - error:badarith -> E - end; -coerce_to_float(E, _) -> E. - -%% expr_bin([Element], State) -> {[Element],State}. - -expr_bin(Es0, St) -> - Es1 = bin_expand_strings(Es0), - foldr(fun (E, Acc) -> bin_element(E, Acc) end, {[],St}, Es1). - -bin_element({bin_element,Line,Expr,Size,Type}, {Es,St0}) -> - {Expr1,St1} = expr(Expr, St0), - {Size1,St2} = if Size == default -> {default,St1}; - true -> expr(Size, St1) - end, - {Size2,Type1} = make_bit_type(Line, Size1, Type), - {[{bin_element,Line,Expr1,Size2,Type1}|Es],St2}. - -bin_expand_strings(Es) -> - foldr(fun ({bin_element,Line,{string,_,S},Sz,Ts}, Es1) -> - foldr(fun (C, Es2) -> - [{bin_element,Line,{char,Line,C},Sz,Ts}|Es2] - end, Es1, S); - (E, Es1) -> [E|Es1] - end, [], Es). - -%% new_var_name(State) -> {VarName,State}. - -new_var_name(St) -> - C = St#expand.vcount, - {list_to_atom("pre" ++ integer_to_list(C)),St#expand{vcount=C+1}}. - -%% new_var(Line, State) -> {Var,State}. - -new_var(L, St0) -> - {New,St1} = new_var_name(St0), - {{var,L,New},St1}. - -%% new_vars(Count, Line, State) -> {[Var],State}. -%% Make Count new variables. - -new_vars(N, L, St) -> new_vars(N, L, St, []). - -new_vars(N, L, St0, Vs) when N > 0 -> - {V,St1} = new_var(L, St0), - new_vars(N-1, L, St1, [V|Vs]); -new_vars(0, _L, St, Vs) -> {Vs,St}. - -string_to_conses(Line, Cs, Tail) -> - foldr(fun (C, T) -> {cons,Line,{char,Line,C},T} end, Tail, Cs). - - -%% import(Line, Imports, State) -> -%% State' -%% imported(Name, Arity, State) -> -%% {yes,Module} | no -%% Handle import declarations and test for imported functions. No need to -%% check when building imports as code is correct. - -import({Mod,Fs}, St) -> - true = is_atom(Mod), - Mfs = from_list(Fs), - St#expand{imports=add_imports(Mod, Mfs, St#expand.imports)}. - -add_imports(Mod, [F|Fs], Is) -> - add_imports(Mod, Fs, orddict:store(F, Mod, Is)); -add_imports(_, [], Is) -> Is. - -imported(F, A, St) -> - case orddict:find({F,A}, St#expand.imports) of - {ok,Mod} -> {yes,Mod}; - error -> no - end. - -defined(F, A, St) -> - gb_sets:is_element({F,A}, St#expand.defined). diff --git a/lib/compiler/src/v3_codegen.erl b/lib/compiler/src/v3_codegen.erl index 2a89305f4d..e9152ba88f 100644 --- a/lib/compiler/src/v3_codegen.erl +++ b/lib/compiler/src/v3_codegen.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2012. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -19,25 +19,6 @@ %% %% Purpose : Code generator for Beam. -%% The following assumptions have been made: -%% -%% 1. Matches, i.e. things with {match,M,Ret} wrappers, only return -%% values; no variables are exported. If the match would have returned -%% extra variables then these have been transformed to multiple return -%% values. -%% -%% 2. All BIF's called in guards are gc-safe so there is no need to -%% put thing on the stack in the guard. While this would in principle -%% work it would be difficult to keep track of the stack depth when -%% trimming. -%% -%% The code generation uses variable lifetime information added by -%% the v3_life module to save variables, allocate registers and -%% move registers to the stack when necessary. -%% -%% We try to use a consistent variable name scheme throughout. The -%% StackReg record is always called Bef,Int<n>,Aft. - -module(v3_codegen). %% The main interface. @@ -45,12 +26,14 @@ -import(lists, [member/2,keymember/3,keysort/2,keydelete/3, append/1,flatmap/2,filter/2,foldl/3,foldr/3,mapfoldl/3, - sort/1,reverse/1,reverse/2]). --import(v3_life, [vdb_find/2]). + sort/1,reverse/1,reverse/2,map/2]). +-import(ordsets, [add_element/2,intersection/2,union/2]). -%%-compile([export_all]). +-include("v3_kernel.hrl"). --include("v3_life.hrl"). +%% These are not defined in v3_kernel.hrl. +get_kanno(Kthing) -> element(2, Kthing). +set_kanno(Kthing, Anno) -> setelement(2, Kthing, Anno). %% Main codegen structure. -record(cg, {lcount=1, %Label counter @@ -61,34 +44,400 @@ functable=#{}, %Map of local functions: {Name,Arity}=>Label in_catch=false, %Inside a catch or not. need_frame, %Need a stack frame. - ultimate_failure %Label for ultimate match failure. - }). + ultimate_failure, %Label for ultimate match failure. + ctx %Match context. + }). %% Stack/register state record. -record(sr, {reg=[], %Register table stk=[], %Stack table - res=[]}). %Reserved regs: [{reserved,I,V}] + res=[]}). %Registers to reserve + +%% Internal records. +-record(cg_need_heap, {anno=[] :: term(), + h=0 :: integer()}). +-record(cg_block, {anno=[] :: term(), + es=[] :: [term()]}). + +-type vdb_entry() :: {atom(),non_neg_integer(),non_neg_integer()}. -module({Mod,Exp,Attr,Forms}, _Options) -> - {Fs,St} = functions(Forms, {atom,Mod}), - {ok,{Mod,Exp,Attr,Fs,St#cg.lcount}}. +-record(l, {i=0 :: non_neg_integer(), %Op number + vdb=[] :: [vdb_entry()], %Variable database + a=[] :: [term()]}). %Core annotation + +-spec module(#k_mdef{}, [compile:option()]) -> {'ok',beam_asm:module_code()}. + +module(#k_mdef{name=Mod,exports=Es,attributes=Attr,body=Forms}, _Opts) -> + {Asm,St} = functions(Forms, {atom,Mod}), + {ok,{Mod,Es,Attr,Asm,St#cg.lcount}}. functions(Forms, AtomMod) -> mapfoldl(fun (F, St) -> function(F, AtomMod, St) end, #cg{lcount=1}, Forms). -function({function,Name,Arity,Asm0,Vb,Vdb,Anno}, AtomMod, St0) -> +function(#k_fdef{anno=#k{a=Anno},func=Name,arity=Arity, + vars=As,body=Kb}, AtomMod, St0) -> try - {Asm,EntryLabel,St} = cg_fun(Vb, Asm0, Vdb, AtomMod, - {Name,Arity}, Anno, St0), - Func = {function,Name,Arity,EntryLabel,Asm}, - {Func,St} + #k_match{} = Kb, %Assertion. + + %% Try to suppress the stack frame unless it is + %% really needed. + Body0 = avoid_stack_frame(Kb), + + %% Annotate kernel records with variable usage. + Vdb0 = init_vars(As), + {Body,_,Vdb} = body(Body0, 1, Vdb0), + + %% Generate the BEAM assembly code. + {Asm,EntryLabel,St} = cg_fun(Body, As, Vdb, AtomMod, + {Name,Arity}, Anno, St0), + Func = {function,Name,Arity,EntryLabel,Asm}, + {Func,St} catch - Class:Error -> - Stack = erlang:get_stacktrace(), - io:fwrite("Function: ~w/~w\n", [Name,Arity]), - erlang:raise(Class, Error, Stack) + Class:Error:Stack -> + io:fwrite("Function: ~w/~w\n", [Name,Arity]), + erlang:raise(Class, Error, Stack) end. + +%% avoid_stack_frame(Kernel) -> Kernel' +%% If possible, avoid setting up a stack frame. Functions +%% that only do matching, calls to guard BIFs, and tail-recursive +%% calls don't need a stack frame. + +avoid_stack_frame(#k_match{body=Body}=M) -> + try + M#k_match{body=avoid_stack_frame_1(Body)} + catch + impossible -> + M + end. + +avoid_stack_frame_1(#k_alt{first=First0,then=Then0}=Alt) -> + First = avoid_stack_frame_1(First0), + Then = avoid_stack_frame_1(Then0), + Alt#k_alt{first=First,then=Then}; +avoid_stack_frame_1(#k_bif{op=Op}=Bif) -> + case Op of + #k_internal{} -> + %% Most internal BIFs clobber the X registers. + throw(impossible); + _ -> + Bif + end; +avoid_stack_frame_1(#k_break{anno=Anno,args=Args}) -> + #k_guard_break{anno=Anno,args=Args}; +avoid_stack_frame_1(#k_guard_break{}=Break) -> + Break; +avoid_stack_frame_1(#k_enter{}=Enter) -> + %% Tail-recursive calls don't need a stack frame. + Enter; +avoid_stack_frame_1(#k_guard{clauses=Cs0}=Guard) -> + Cs = avoid_stack_frame_list(Cs0), + Guard#k_guard{clauses=Cs}; +avoid_stack_frame_1(#k_guard_clause{guard=G0,body=B0}=C) -> + G = avoid_stack_frame_1(G0), + B = avoid_stack_frame_1(B0), + C#k_guard_clause{guard=G,body=B}; +avoid_stack_frame_1(#k_match{anno=A,vars=Vs,body=B0,ret=Ret}) -> + %% Use #k_guard_match{} instead to avoid saving the X registers + %% to the stack before matching. + B = avoid_stack_frame_1(B0), + #k_guard_match{anno=A,vars=Vs,body=B,ret=Ret}; +avoid_stack_frame_1(#k_guard_match{body=B0}=M) -> + B = avoid_stack_frame_1(B0), + M#k_guard_match{body=B}; +avoid_stack_frame_1(#k_protected{arg=Arg0}=Prot) -> + Arg = avoid_stack_frame_1(Arg0), + Prot#k_protected{arg=Arg}; +avoid_stack_frame_1(#k_put{}=Put) -> + Put; +avoid_stack_frame_1(#k_return{}=Ret) -> + Ret; +avoid_stack_frame_1(#k_select{var=#k_var{anno=Vanno},types=Types0}=Select) -> + case member(reuse_for_context, Vanno) of + false -> + Types = avoid_stack_frame_list(Types0), + Select#k_select{types=Types}; + true -> + %% Including binary patterns that overwrite the register containing + %% the binary with the match context may not be safe. For example, + %% bs_match_SUITE:bin_tail_e/1 with inlining will be rejected by + %% beam_validator. + %% + %% Essentially the following code is produced: + %% + %% bs_match {x,0} => {x,0} + %% ... + %% bs_match {x,0} => {x,1} %% ILLEGAL + %% + %% A bs_match instruction will only accept a match context as the + %% source operand if the source and destination registers are the + %% the same (as in the first bs_match instruction above). + %% The second bs_match instruction is therefore illegal. + %% + %% This situation is avoided if there is a stack frame: + %% + %% move {x,0} => {y,0} + %% bs_match {x,0} => {x,0} + %% ... + %% bs_match {y,0} => {x,1} %% LEGAL + %% + throw(impossible) + end; +avoid_stack_frame_1(#k_seq{arg=#k_call{anno=Anno,op=Op}=Call, + body=#k_break{args=BrArgs0}}=Seq) -> + case Op of + #k_remote{mod=#k_atom{val=Mod}, + name=#k_atom{val=Name}, + arity=Arity} -> + case erl_bifs:is_exit_bif(Mod, Name, Arity) of + false -> + %% Will clobber X registers. Must have a stack frame. + throw(impossible); + true -> + %% The call to this BIF will never return. It is safe + %% to suppress the stack frame. + Bif = #k_bif{anno=Anno, + op=#k_internal{name=guard_error,arity=1}, + args=[Call],ret=[]}, + BrArgs = lists:duplicate(length(BrArgs0), #k_nil{}), + GB = #k_guard_break{anno=#k{us=[],ns=[],a=[]},args=BrArgs}, + Seq#k_seq{arg=Bif,body=GB} + end; + _ -> + %% Will clobber X registers. Must have a stack frame. + throw(impossible) + end; +avoid_stack_frame_1(#k_seq{arg=A0,body=B0}=Seq) -> + A = avoid_stack_frame_1(A0), + B = avoid_stack_frame_1(B0), + Seq#k_seq{arg=A,body=B}; +avoid_stack_frame_1(#k_test{}=Test) -> + Test; +avoid_stack_frame_1(#k_type_clause{values=Values0}=TC) -> + Values = avoid_stack_frame_list(Values0), + TC#k_type_clause{values=Values}; +avoid_stack_frame_1(#k_val_clause{body=B0}=VC) -> + B = avoid_stack_frame_1(B0), + VC#k_val_clause{body=B}; +avoid_stack_frame_1(_Body) -> + throw(impossible). + +avoid_stack_frame_list([H|T]) -> + [avoid_stack_frame_1(H)|avoid_stack_frame_list(T)]; +avoid_stack_frame_list([]) -> []. + + +%% This pass creates beam format annotated with variable lifetime +%% information. Each thing is given an index and for each variable we +%% store the first and last index for its occurrence. The variable +%% database, VDB, attached to each thing is only relevant internally +%% for that thing. +%% +%% For nested things like matches the numbering continues locally and +%% the VDB for that thing refers to the variable usage within that +%% thing. Variables which live through a such a thing are internally +%% given a very large last index. Internally the indexes continue +%% after the index of that thing. This creates no problems as the +%% internal variable info never escapes and externally we only see +%% variable which are alive both before or after. +%% +%% This means that variables never "escape" from a thing and the only +%% way to get values from a thing is to "return" them, with 'break' or +%% 'return'. Externally these values become the return values of the +%% thing. This is no real limitation as most nested things have +%% multiple threads so working out a common best variable usage is +%% difficult. + +%% body(Kbody, I, Vdb) -> {[Expr],MaxI,Vdb}. +%% Handle a body. + +body(#k_seq{arg=Ke,body=Kb}, I, Vdb0) -> + %%ok = io:fwrite("life ~w:~p~n", [?LINE,{Ke,I,Vdb0}]), + A = get_kanno(Ke), + Vdb1 = use_vars(union(A#k.us, A#k.ns), I, Vdb0), + {Es,MaxI,Vdb2} = body(Kb, I+1, Vdb1), + E = expr(Ke, I, Vdb2), + {[E|Es],MaxI,Vdb2}; +body(Ke, I, Vdb0) -> + %%ok = io:fwrite("life ~w:~p~n", [?LINE,{Ke,I,Vdb0}]), + A = get_kanno(Ke), + Vdb1 = use_vars(union(A#k.us, A#k.ns), I, Vdb0), + E = expr(Ke, I, Vdb1), + {[E],I,Vdb1}. + +%% expr(Kexpr, I, Vdb) -> Expr. + +expr(#k_test{anno=A}=Test, I, _Vdb) -> + Test#k_test{anno=#l{i=I,a=A#k.a}}; +expr(#k_call{anno=A}=Call, I, _Vdb) -> + Call#k_call{anno=#l{i=I,a=A#k.a}}; +expr(#k_enter{anno=A}=Enter, I, _Vdb) -> + Enter#k_enter{anno=#l{i=I,a=A#k.a}}; +expr(#k_bif{anno=A}=Bif, I, _Vdb) -> + Bif#k_bif{anno=#l{i=I,a=A#k.a}}; +expr(#k_match{anno=A,body=Kb,ret=Rs}, I, Vdb) -> + %% Work out imported variables which need to be locked. + Mdb = vdb_sub(I, I+1, Vdb), + M = match(Kb, A#k.us, I+1, Mdb), + L = #l{i=I,vdb=use_vars(A#k.us, I+1, Mdb),a=A#k.a}, + #k_match{anno=L,body=M,ret=Rs}; +expr(#k_guard_match{anno=A,body=Kb,ret=Rs}, I, Vdb) -> + %% Work out imported variables which need to be locked. + Mdb = vdb_sub(I, I+1, Vdb), + M = match(Kb, A#k.us, I+1, Mdb), + L = #l{i=I,vdb=use_vars(A#k.us, I+1, Mdb),a=A#k.a}, + #k_guard_match{anno=L,body=M,ret=Rs}; +expr(#k_protected{}=Protected, I, Vdb) -> + protected(Protected, I, Vdb); +expr(#k_try{anno=A,arg=Ka,vars=Vs,body=Kb,evars=Evs,handler=Kh}=Try, I, Vdb) -> + %% Lock variables that are alive before the catch and used afterwards. + %% Don't lock variables that are only used inside the try. + Tdb0 = vdb_sub(I, I+1, Vdb), + %% This is the tricky bit. Lock variables in Arg that are used in + %% the body and handler. Add try tag 'variable'. + Ab = get_kanno(Kb), + Ah = get_kanno(Kh), + Tdb1 = use_vars(union(Ab#k.us, Ah#k.us), I+3, Tdb0), + Tdb2 = vdb_sub(I, I+2, Tdb1), + Vnames = fun (Kvar) -> Kvar#k_var.name end, %Get the variable names + {Aes,_,Adb} = body(Ka, I+2, add_var({catch_tag,I+1}, I+1, locked, Tdb2)), + {Bes,_,Bdb} = body(Kb, I+4, new_vars(sort(map(Vnames, Vs)), I+3, Tdb2)), + {Hes,_,Hdb} = body(Kh, I+4, new_vars(sort(map(Vnames, Evs)), I+3, Tdb2)), + L = #l{i=I,vdb=Tdb1,a=A#k.a}, + Try#k_try{anno=L, + arg=#cg_block{es=Aes,anno=#l{i=I+1,vdb=Adb,a=[]}}, + vars=Vs,body=#cg_block{es=Bes,anno=#l{i=I+3,vdb=Bdb,a=[]}}, + evars=Evs,handler=#cg_block{es=Hes,anno=#l{i=I+3,vdb=Hdb,a=[]}}}; +expr(#k_try_enter{anno=A,arg=Ka,vars=Vs,body=Kb,evars=Evs,handler=Kh}, I, Vdb) -> + %% Lock variables that are alive before the catch and used afterwards. + %% Don't lock variables that are only used inside the try. + Tdb0 = vdb_sub(I, I+1, Vdb), + %% This is the tricky bit. Lock variables in Arg that are used in + %% the body and handler. Add try tag 'variable'. + Ab = get_kanno(Kb), + Ah = get_kanno(Kh), + Tdb1 = use_vars(union(Ab#k.us, Ah#k.us), I+3, Tdb0), + Tdb2 = vdb_sub(I, I+2, Tdb1), + Vnames = fun (Kvar) -> Kvar#k_var.name end, %Get the variable names + {Aes,_,Adb} = body(Ka, I+2, add_var({catch_tag,I+1}, I+1, 1000000, Tdb2)), + {Bes,_,Bdb} = body(Kb, I+4, new_vars(sort(map(Vnames, Vs)), I+3, Tdb2)), + {Hes,_,Hdb} = body(Kh, I+4, new_vars(sort(map(Vnames, Evs)), I+3, Tdb2)), + L = #l{i=I,vdb=Tdb1,a=A#k.a}, + #k_try_enter{anno=L, + arg=#cg_block{es=Aes,anno=#l{i=I+1,vdb=Adb,a=[]}}, + vars=Vs,body=#cg_block{es=Bes,anno=#l{i=I+3,vdb=Bdb,a=[]}}, + evars=Evs,handler=#cg_block{es=Hes,anno=#l{i=I+3,vdb=Hdb,a=[]}}}; +expr(#k_catch{anno=A,body=Kb}=Catch, I, Vdb) -> + %% Lock variables that are alive before the catch and used afterwards. + %% Don't lock variables that are only used inside the catch. + %% Add catch tag 'variable'. + Cdb0 = vdb_sub(I, I+1, Vdb), + {Es,_,Cdb1} = body(Kb, I+1, add_var({catch_tag,I}, I, locked, Cdb0)), + L = #l{i=I,vdb=Cdb1,a=A#k.a}, + Catch#k_catch{anno=L,body=#cg_block{es=Es}}; +expr(#k_receive{anno=A,var=V,body=Kb,action=Ka}=Recv, I, Vdb) -> + %% Work out imported variables which need to be locked. + Rdb = vdb_sub(I, I+1, Vdb), + M = match(Kb, add_element(V#k_var.name, A#k.us), I+1, + new_vars([V#k_var.name], I, Rdb)), + {Tes,_,Adb} = body(Ka, I+1, Rdb), + Le = #l{i=I,vdb=use_vars(A#k.us, I+1, Vdb),a=A#k.a}, + Recv#k_receive{anno=Le,body=M, + action=#cg_block{anno=#l{i=I+1,vdb=Adb,a=[]},es=Tes}}; +expr(#k_receive_accept{anno=A}, I, _Vdb) -> + #k_receive_accept{anno=#l{i=I,a=A#k.a}}; +expr(#k_receive_next{anno=A}, I, _Vdb) -> + #k_receive_next{anno=#l{i=I,a=A#k.a}}; +expr(#k_put{anno=A}=Put, I, _Vdb) -> + Put#k_put{anno=#l{i=I,a=A#k.a}}; +expr(#k_break{anno=A}=Break, I, _Vdb) -> + Break#k_break{anno=#l{i=I,a=A#k.a}}; +expr(#k_guard_break{anno=A}=Break, I, _Vdb) -> + Break#k_guard_break{anno=#l{i=I,a=A#k.a}}; +expr(#k_return{anno=A}=Ret, I, _Vdb) -> + Ret#k_return{anno=#l{i=I,a=A#k.a}}. + +%% protected(Kprotected, I, Vdb) -> Protected. +%% Only used in guards. + +protected(#k_protected{anno=A,arg=Ts}=Prot, I, Vdb) -> + %% Lock variables that are alive before try and used afterwards. + %% Don't lock variables that are only used inside the protected + %% expression. + Pdb0 = vdb_sub(I, I+1, Vdb), + {T,MaxI,Pdb1} = body(Ts, I+1, Pdb0), + Pdb2 = use_vars(A#k.ns, MaxI+1, Pdb1), %Save "return" values + Prot#k_protected{arg=T,anno=#l{i=I,a=A#k.a,vdb=Pdb2}}. + +%% match(Kexpr, [LockVar], I, Vdb) -> Expr. +%% Convert match tree to old format. + +match(#k_alt{anno=A,first=Kf,then=Kt}, Ls, I, Vdb0) -> + Vdb1 = use_vars(union(A#k.us, Ls), I, Vdb0), + F = match(Kf, Ls, I+1, Vdb1), + T = match(Kt, Ls, I+1, Vdb1), + #k_alt{anno=[],first=F,then=T}; +match(#k_select{anno=A,types=Kts}=Select, Ls, I, Vdb0) -> + Vdb1 = use_vars(union(A#k.us, Ls), I, Vdb0), + Ts = [type_clause(Tc, Ls, I+1, Vdb1) || Tc <- Kts], + Select#k_select{anno=[],types=Ts}; +match(#k_guard{anno=A,clauses=Kcs}, Ls, I, Vdb0) -> + Vdb1 = use_vars(union(A#k.us, Ls), I, Vdb0), + Cs = [guard_clause(G, Ls, I+1, Vdb1) || G <- Kcs], + #k_guard{anno=[],clauses=Cs}; +match(Other, Ls, I, Vdb0) -> + Vdb1 = use_vars(Ls, I, Vdb0), + {B,_,Vdb2} = body(Other, I+1, Vdb1), + Le = #l{i=I,vdb=Vdb2,a=[]}, + #cg_block{anno=Le,es=B}. + +type_clause(#k_type_clause{anno=A,type=T,values=Kvs}, Ls, I, Vdb0) -> + %%ok = io:format("life ~w: ~p~n", [?LINE,{T,Kvs}]), + Vdb1 = use_vars(union(A#k.us, Ls), I+1, Vdb0), + Vs = [val_clause(Vc, Ls, I+1, Vdb1) || Vc <- Kvs], + #k_type_clause{anno=[],type=T,values=Vs}. + +val_clause(#k_val_clause{anno=A,val=V,body=Kb}, Ls0, I, Vdb0) -> + New = (get_kanno(V))#k.ns, + Bus = (get_kanno(Kb))#k.us, + %%ok = io:format("Ls0 = ~p, Used=~p\n New=~p, Bus=~p\n", [Ls0,Used,New,Bus]), + Ls1 = union(intersection(New, Bus), Ls0), %Lock for safety + Vdb1 = use_vars(union(A#k.us, Ls1), I+1, new_vars(New, I, Vdb0)), + B = match(Kb, Ls1, I+1, Vdb1), + Le = #l{i=I,vdb=use_vars(Bus, I+1, Vdb1),a=A#k.a}, + #k_val_clause{anno=Le,val=V,body=B}. + +guard_clause(#k_guard_clause{anno=A,guard=Kg,body=Kb}, Ls, I, Vdb0) -> + Vdb1 = use_vars(union(A#k.us, Ls), I+2, Vdb0), + Gdb = vdb_sub(I+1, I+2, Vdb1), + G = protected(Kg, I+1, Gdb), + B = match(Kb, Ls, I+2, Vdb1), + Le = #l{i=I,vdb=use_vars((get_kanno(Kg))#k.us, I+2, Vdb1),a=A#k.a}, + #k_guard_clause{anno=Le,guard=G,body=B}. + + +%% Here follows the code generator pass. +%% +%% The following assumptions have been made: +%% +%% 1. Matches, i.e. things with {match,M,Ret} wrappers, only return +%% values; no variables are exported. If the match would have returned +%% extra variables then these have been transformed to multiple return +%% values. +%% +%% 2. All BIF's called in guards are gc-safe so there is no need to +%% put thing on the stack in the guard. While this would in principle +%% work it would be difficult to keep track of the stack depth when +%% trimming. +%% +%% The code generation uses variable lifetime information added by +%% the previous pass to save variables, allocate registers and +%% move registers to the stack when necessary. +%% +%% We try to use a consistent variable name scheme throughout. The +%% StackReg record is always called Bef,Int<n>,Aft. + %% cg_fun([Lkexpr], [HeadVar], Vdb, State) -> {[Ainstr],State} cg_fun(Les, Hvs, Vdb, AtomMod, NameArity, Anno, St0) -> @@ -110,18 +459,18 @@ cg_fun(Les, Hvs, Vdb, AtomMod, NameArity, Anno, St0) -> %% Note that and 'if_end' instruction does not need any %% live x registers, so it will always be safe to jump to %% it. (We never ever expect the jump to be taken, and in - %% must functions there will never be any references to + %% most functions there will never be any references to %% the label in the first place.) %% {UltimateMatchFail,St3} = new_label(St2), %% Create initial stack/register state, clear unused arguments. - Bef = clear_dead(#sr{reg=foldl(fun ({var,V}, Reg) -> + Bef = clear_dead(#sr{reg=foldl(fun (#k_var{name=V}, Reg) -> put_reg(V, Reg) end, [], Hvs), stk=[]}, 0, Vdb), - {B,_Aft,St} = cg_list(Les, 0, Vdb, Bef, + {B,_Aft,St} = cg_list(Les, Vdb, Bef, St3#cg{bfail=0, ultimate_failure=UltimateMatchFail, is_top_block=true}), @@ -132,64 +481,64 @@ cg_fun(Les, Hvs, Vdb, AtomMod, NameArity, Anno, St0) -> %% cg(Lkexpr, Vdb, StackReg, State) -> {[Ainstr],StackReg,State}. %% Generate code for a kexpr. -%% Split function into two steps for clarity, not efficiency. - -cg(Le, Vdb, Bef, St) -> - cg(Le#l.ke, Le, Vdb, Bef, St). -cg({block,Es}, Le, Vdb, Bef, St) -> +cg(#cg_block{anno=Le,es=Es}, Vdb, Bef, St) -> block_cg(Es, Le, Vdb, Bef, St); -cg({match,M,Rs}, Le, Vdb, Bef, St) -> +cg(#k_match{anno=Le,body=M,ret=Rs}, Vdb, Bef, St) -> match_cg(M, Rs, Le, Vdb, Bef, St); -cg({guard_match,M,Rs}, Le, Vdb, Bef, St) -> +cg(#k_guard_match{anno=Le,body=M,ret=Rs}, Vdb, Bef, St) -> guard_match_cg(M, Rs, Le, Vdb, Bef, St); -cg({call,Func,As,Rs}, Le, Vdb, Bef, St) -> +cg(#k_call{anno=Le,op=Func,args=As,ret=Rs}, Vdb, Bef, St) -> call_cg(Func, As, Rs, Le, Vdb, Bef, St); -cg({enter,Func,As}, Le, Vdb, Bef, St) -> +cg(#k_enter{anno=Le,op=Func,args=As}, Vdb, Bef, St) -> enter_cg(Func, As, Le, Vdb, Bef, St); -cg({bif,Bif,As,Rs}, Le, Vdb, Bef, St) -> - bif_cg(Bif, As, Rs, Le, Vdb, Bef, St); -cg({gc_bif,Bif,As,Rs}, Le, Vdb, Bef, St) -> - gc_bif_cg(Bif, As, Rs, Le, Vdb, Bef, St); -cg({receive_loop,Te,Rvar,Rm,Tes,Rs}, Le, Vdb, Bef, St) -> +cg(#k_bif{anno=Le}=Bif, Vdb, Bef, St) -> + bif_cg(Bif, Le, Vdb, Bef, St); +cg(#k_receive{anno=Le,timeout=Te,var=Rvar,body=Rm,action=Tes,ret=Rs}, + Vdb, Bef, St) -> recv_loop_cg(Te, Rvar, Rm, Tes, Rs, Le, Vdb, Bef, St); -cg(receive_next, Le, Vdb, Bef, St) -> +cg(#k_receive_next{anno=Le}, Vdb, Bef, St) -> recv_next_cg(Le, Vdb, Bef, St); -cg(receive_accept, _Le, _Vdb, Bef, St) -> {[remove_message],Bef,St}; -cg({'try',Ta,Vs,Tb,Evs,Th,Rs}, Le, Vdb, Bef, St) -> +cg(#k_receive_accept{}, _Vdb, Bef, St) -> + {[remove_message],Bef,St}; +cg(#k_try{anno=Le,arg=Ta,vars=Vs,body=Tb,evars=Evs,handler=Th,ret=Rs}, + Vdb, Bef, St) -> try_cg(Ta, Vs, Tb, Evs, Th, Rs, Le, Vdb, Bef, St); -cg({try_enter,Ta,Vs,Tb,Evs,Th}, Le, Vdb, Bef, St) -> +cg(#k_try_enter{anno=Le,arg=Ta,vars=Vs,body=Tb,evars=Evs,handler=Th}, + Vdb, Bef, St) -> try_enter_cg(Ta, Vs, Tb, Evs, Th, Le, Vdb, Bef, St); -cg({'catch',Cb,R}, Le, Vdb, Bef, St) -> +cg(#k_catch{anno=Le,body=Cb,ret=[R]}, Vdb, Bef, St) -> catch_cg(Cb, R, Le, Vdb, Bef, St); -cg({set,Var,Con}, Le, Vdb, Bef, St) -> - set_cg(Var, Con, Le, Vdb, Bef, St); -cg({return,Rs}, Le, Vdb, Bef, St) -> return_cg(Rs, Le, Vdb, Bef, St); -cg({break,Bs}, Le, Vdb, Bef, St) -> break_cg(Bs, Le, Vdb, Bef, St); -cg({guard_break,Bs,N}, Le, Vdb, Bef, St) -> - guard_break_cg(Bs, N, Le, Vdb, Bef, St); -cg({need_heap,H}, _Le, _Vdb, Bef, St) -> +cg(#k_put{anno=Le,arg=Con,ret=Var}, Vdb, Bef, St) -> + put_cg(Var, Con, Le, Vdb, Bef, St); +cg(#k_return{anno=Le,args=Rs}, Vdb, Bef, St) -> + return_cg(Rs, Le, Vdb, Bef, St); +cg(#k_break{anno=Le,args=Bs}, Vdb, Bef, St) -> + break_cg(Bs, Le, Vdb, Bef, St); +cg(#k_guard_break{anno=Le,args=Bs}, Vdb, Bef, St) -> + guard_break_cg(Bs, Le, Vdb, Bef, St); +cg(#cg_need_heap{h=H}, _Vdb, Bef, St) -> {[{test_heap,H,max_reg(Bef#sr.reg)}],Bef,St}. %% cg_list([Kexpr], FirstI, Vdb, StackReg, St) -> {[Ainstr],StackReg,St}. -cg_list(Kes, I, Vdb, Bef, St0) -> +cg_list(Kes, Vdb, Bef, St0) -> {Keis,{Aft,St1}} = flatmapfoldl(fun (Ke, {Inta,Sta}) -> {Keis,Intb,Stb} = cg(Ke, Vdb, Inta, Sta), {Keis,{Intb,Stb}} - end, {Bef,St0}, need_heap(Kes, I)), + end, {Bef,St0}, need_heap(Kes)), {Keis,Aft,St1}. %% need_heap([Lkexpr], I, St) -> [Lkexpr]. %% Insert need_heap instructions in Kexpr list. Try to be smart and %% collect them together as much as possible. -need_heap(Kes0, I) -> +need_heap(Kes0) -> {Kes,H} = need_heap_0(reverse(Kes0), 0, []), %% Prepend need_heap if necessary. - need_heap_need(I, H) ++ Kes. + need_heap_need(H) ++ Kes. need_heap_0([Ke|Kes], H0, Acc) -> {Ns,H} = need_heap_1(Ke, H0), @@ -197,32 +546,56 @@ need_heap_0([Ke|Kes], H0, Acc) -> need_heap_0([], H, Acc) -> {Acc,H}. -need_heap_1(#l{ke={set,_,{binary,_}},i=I}, H) -> - {need_heap_need(I, H),0}; -need_heap_1(#l{ke={set,_,{map,_,_,_}},i=I}, H) -> - {need_heap_need(I, H),0}; -need_heap_1(#l{ke={set,_,Val}}, H) -> +need_heap_1(#k_put{arg=#k_binary{}}, H) -> + {need_heap_need(H),0}; +need_heap_1(#k_put{arg=#k_map{}}, H) -> + {need_heap_need(H),0}; +need_heap_1(#k_put{arg=Val}, H) -> %% Just pass through adding to needed heap. {[],H + case Val of - {cons,_} -> 2; - {tuple,Es} -> 1 + length(Es); + #k_cons{} -> 2; + #k_tuple{es=Es} -> 1 + length(Es); _Other -> 0 end}; -need_heap_1(#l{ke={bif,dsetelement,_As,_Rs},i=I}, H) -> - {need_heap_need(I, H),0}; -need_heap_1(#l{ke={bif,{make_fun,_,_,_,_},_As,_Rs},i=I}, H) -> - {need_heap_need(I, H),0}; -need_heap_1(#l{ke={bif,bs_init_writable,_As,_Rs},i=I}, H) -> - {need_heap_need(I, H),0}; -need_heap_1(#l{ke={bif,_Bif,_As,_Rs}}, H) -> - {[],H}; -need_heap_1(#l{i=I}, H) -> - {need_heap_need(I, H),0}. - -need_heap_need(_I, 0) -> []; -need_heap_need(I, H) -> [#l{ke={need_heap,H},i=I}]. - -%% match_cg(Match, [Ret], Le, Vdb, StackReg, State) -> +need_heap_1(#k_bif{}=Bif, H) -> + case is_gc_bif(Bif) of + false -> + {[],H}; + true -> + {need_heap_need(H),0} + end; +need_heap_1(_Ke, H) -> + %% Call or call-like instruction such as set_tuple_element/3. + {need_heap_need(H),0}. + +need_heap_need(0) -> []; +need_heap_need(H) -> [#cg_need_heap{h=H}]. + +%% is_gc_bif(#k_bif{}) -> true|false. +%% is_gc_bif(Name, Arity) -> true|false. +%% Determines whether the BIF Name/Arity might do a GC. + +is_gc_bif(#k_bif{op=#k_remote{name=#k_atom{val=Name}},args=Args}) -> + is_gc_bif(Name, length(Args)); +is_gc_bif(#k_bif{op=#k_internal{}}) -> + true. + +is_gc_bif(hd, 1) -> false; +is_gc_bif(tl, 1) -> false; +is_gc_bif(self, 0) -> false; +is_gc_bif(node, 0) -> false; +is_gc_bif(node, 1) -> false; +is_gc_bif(element, 2) -> false; +is_gc_bif(get, 1) -> false; +is_gc_bif(tuple_size, 1) -> false; +is_gc_bif(map_get, 2) -> false; +is_gc_bif(is_map_key, 2) -> false; +is_gc_bif(Bif, Arity) -> + not (erl_internal:bool_op(Bif, Arity) orelse + erl_internal:new_type_test(Bif, Arity) orelse + erl_internal:comp_op(Bif, Arity)). + +%% match_cg(Matc, [Ret], Le, Vdb, StackReg, State) -> %% {[Ainstr],StackReg,State}. %% Generate code for a match. First save all variables on the stack %% that are to survive after the match. We leave saved variables in @@ -243,7 +616,10 @@ match_cg(M, Rs, Le, Vdb, Bef, St0) -> guard_match_cg(M, Rs, Le, Vdb, Bef, St0) -> I = Le#l.i, {B,St1} = new_label(St0), - #cg{bfail=Fail} = St1, + Fail = case St0 of + #cg{bfail=0,ultimate_failure=Fail0} -> Fail0; + #cg{bfail=Fail0} -> Fail0 + end, {Mis,Aft,St2} = match_cg(M, Fail, Bef, St1#cg{break=B}), %% Update the register descriptors for the return registers. Reg = guard_match_regs(Aft#sr.reg, Rs), @@ -251,7 +627,7 @@ guard_match_cg(M, Rs, Le, Vdb, Bef, St0) -> clear_dead(Aft#sr{reg=Reg}, I, Vdb), St2#cg{break=St1#cg.break}}. -guard_match_regs([{I,gbreakvar}|Rs], [{var,V}|Vs]) -> +guard_match_regs([{I,gbreakvar}|Rs], [#k_var{name=V}|Vs]) -> [{I,V}|guard_match_regs(Rs, Vs)]; guard_match_regs([R|Rs], Vs) -> [R|guard_match_regs(Rs, Vs)]; @@ -263,17 +639,14 @@ guard_match_regs([], []) -> []. %% down as each level which uses this takes its own internal Vdb not %% the outer one. -match_cg(Le, Fail, Bef, St) -> - match_cg(Le#l.ke, Le, Fail, Bef, St). - -match_cg({alt,F,S}, _Le, Fail, Bef, St0) -> +match_cg(#k_alt{first=F,then=S}, Fail, Bef, St0) -> {Tf,St1} = new_label(St0), {Fis,Faft,St2} = match_cg(F, Tf, Bef, St1), {Sis,Saft,St3} = match_cg(S, Fail, Bef, St2), Aft = sr_merge(Faft, Saft), {Fis ++ [{label,Tf}] ++ Sis,Aft,St3}; -match_cg({select,{var,Vname}=V,Scs0}, #l{a=Anno}, Fail, Bef, St) -> - ReuseForContext = member(reuse_for_context, Anno) andalso +match_cg(#k_select{var=#k_var{anno=Vanno,name=Vname}=V,types=Scs0}, Fail, Bef, St) -> + ReuseForContext = member(reuse_for_context, Vanno) andalso find_reg(Vname, Bef#sr.reg) =/= error, Scs = case ReuseForContext of false -> Scs0; @@ -282,10 +655,10 @@ match_cg({select,{var,Vname}=V,Scs0}, #l{a=Anno}, Fail, Bef, St) -> match_fmf(fun (S, F, Sta) -> select_cg(S, V, F, Fail, Bef, Sta) end, Fail, St, Scs); -match_cg({guard,Gcs}, _Le, Fail, Bef, St) -> +match_cg(#k_guard{clauses=Gcs}, Fail, Bef, St) -> match_fmf(fun (G, F, Sta) -> guard_clause_cg(G, F, Bef, Sta) end, Fail, St, Gcs); -match_cg({block,Es}, Le, _Fail, Bef, St) -> +match_cg(#cg_block{anno=Le,es=Es}, _Fail, Bef, St) -> %% Must clear registers and stack of dead variables. Int = clear_dead(Bef, Le#l.i, Le#l.vdb), block_cg(Es, Le, Int, St). @@ -293,8 +666,8 @@ match_cg({block,Es}, Le, _Fail, Bef, St) -> %% bsm_rename_ctx([Clause], Var) -> [Clause] %% We know from an annotation that the register for a binary can %% be reused for the match context because the two are not truly -%% alive at the same time (even though the conservative life time -%% information calculated by v3_life says so). +%% alive at the same time (even though the life time information +%% says so). %% %% The easiest way to have those variables share the same register is %% to rename the variable with the shortest life-span (the match @@ -305,12 +678,14 @@ match_cg({block,Es}, Le, _Fail, Bef, St) -> %% We must also remove all information about the match context %% variable from all life-time information databases (Vdb). -bsm_rename_ctx([#l{ke={type_clause,binary, - [#l{ke={val_clause,{binary,{var,Old}},Ke0}}=L2]}}=L1|Cs], New) -> +bsm_rename_ctx([#k_type_clause{type=k_binary,values=Vcs}=TC|Cs], New) -> + [#k_val_clause{val=#k_binary{segs=#k_var{name=Old}}=Bin, + body=Ke0}=VC0] = Vcs, Ke = bsm_rename_ctx(Ke0, Old, New, false), - [L1#l{ke={type_clause,binary, - [L2#l{ke={val_clause,{binary,{var,New}},Ke}}]}}|bsm_rename_ctx(Cs, New)]; -bsm_rename_ctx([C|Cs], New) -> + VC = VC0#k_val_clause{val=Bin#k_binary{segs=#k_var{name=New}}, + body=Ke}, + [TC#k_type_clause{values=[VC]}|bsm_rename_ctx(Cs, New)]; +bsm_rename_ctx([C|Cs], New) -> [C|bsm_rename_ctx(Cs, New)]; bsm_rename_ctx([], _) -> []. @@ -320,34 +695,24 @@ bsm_rename_ctx([], _) -> []. %% only complicatate things to recurse into blocks not in a protected %% (the match context variable is not live inside them). -bsm_rename_ctx(#l{ke={select,{var,V},Cs0}}=L, Old, New, InProt) -> +bsm_rename_ctx(#k_select{var=#k_var{name=V},types=Cs0}=Sel, + Old, New, InProt) -> Cs = bsm_rename_ctx_list(Cs0, Old, New, InProt), - L#l{ke={select,{var,bsm_rename_var(V, Old, New)},Cs}}; -bsm_rename_ctx(#l{ke={type_clause,Type,Cs0}}=L, Old, New, InProt) -> + Sel#k_select{var=#k_var{name=bsm_rename_var(V, Old, New)},types=Cs}; +bsm_rename_ctx(#k_type_clause{values=Cs0}=TC, Old, New, InProt) -> Cs = bsm_rename_ctx_list(Cs0, Old, New, InProt), - L#l{ke={type_clause,Type,Cs}}; -bsm_rename_ctx(#l{ke={val_clause,{bin_end,V},Ke0}}=L, Old, New, InProt) -> - Ke = bsm_rename_ctx(Ke0, Old, New, InProt), - L#l{ke={val_clause,{bin_end,bsm_rename_var(V, Old, New)},Ke}}; -bsm_rename_ctx(#l{ke={val_clause,{bin_seg,V,Sz,U,Type,Fl,Vs},Ke0}}=L, - Old, New, InProt) -> - Ke = bsm_rename_ctx(Ke0, Old, New, InProt), - L#l{ke={val_clause,{bin_seg,bsm_rename_var(V, Old, New),Sz,U,Type,Fl,Vs},Ke}}; -bsm_rename_ctx(#l{ke={val_clause,{bin_int,V,Sz,U,Fl,Val,Vs},Ke0}}=L, - Old, New, InProt) -> + TC#k_type_clause{values=Cs}; +bsm_rename_ctx(#k_val_clause{body=Ke0}=VC, Old, New, InProt) -> Ke = bsm_rename_ctx(Ke0, Old, New, InProt), - L#l{ke={val_clause,{bin_int,bsm_rename_var(V, Old, New),Sz,U,Fl,Val,Vs},Ke}}; -bsm_rename_ctx(#l{ke={val_clause,Val,Ke0}}=L, Old, New, InProt) -> - Ke = bsm_rename_ctx(Ke0, Old, New, InProt), - L#l{ke={val_clause,Val,Ke}}; -bsm_rename_ctx(#l{ke={alt,F0,S0}}=L, Old, New, InProt) -> + VC#k_val_clause{body=Ke}; +bsm_rename_ctx(#k_alt{first=F0,then=S0}=Alt, Old, New, InProt) -> F = bsm_rename_ctx(F0, Old, New, InProt), S = bsm_rename_ctx(S0, Old, New, InProt), - L#l{ke={alt,F,S}}; -bsm_rename_ctx(#l{ke={guard,Gcs0}}=L, Old, New, InProt) -> + Alt#k_alt{first=F,then=S}; +bsm_rename_ctx(#k_guard{clauses=Gcs0}=Guard, Old, New, InProt) -> Gcs = bsm_rename_ctx_list(Gcs0, Old, New, InProt), - L#l{ke={guard,Gcs}}; -bsm_rename_ctx(#l{ke={guard_clause,G0,B0}}=L, Old, New, InProt) -> + Guard#k_guard{clauses=Gcs}; +bsm_rename_ctx(#k_guard_clause{guard=G0,body=B0}=GC, Old, New, InProt) -> G = bsm_rename_ctx(G0, Old, New, InProt), B = bsm_rename_ctx(B0, Old, New, InProt), %% A guard clause may cause unsaved variables to be saved on the stack. @@ -355,49 +720,45 @@ bsm_rename_ctx(#l{ke={guard_clause,G0,B0}}=L, Old, New, InProt) -> %% same register), it is neither in the stack nor register descriptor %% lists and we would crash when we didn't find it unless we remove %% it from the database. - bsm_forget_var(L#l{ke={guard_clause,G,B}}, Old); -bsm_rename_ctx(#l{ke={protected,Ts0,Rs}}=L, Old, New, _InProt) -> + bsm_forget_var(GC#k_guard_clause{guard=G,body=B}, Old); +bsm_rename_ctx(#k_protected{arg=Ts0}=Prot, Old, New, _InProt) -> InProt = true, Ts = bsm_rename_ctx_list(Ts0, Old, New, InProt), - bsm_forget_var(L#l{ke={protected,Ts,Rs}}, Old); -bsm_rename_ctx(#l{ke={match,Ms0,Rs}}=L, Old, New, InProt) -> - Ms = bsm_rename_ctx(Ms0, Old, New, InProt), - L#l{ke={match,Ms,Rs}}; -bsm_rename_ctx(#l{ke={guard_match,Ms0,Rs}}=L, Old, New, InProt) -> + bsm_forget_var(Prot#k_protected{arg=Ts}, Old); +bsm_rename_ctx(#k_guard_match{body=Ms0}=Match, Old, New, InProt) -> Ms = bsm_rename_ctx(Ms0, Old, New, InProt), - L#l{ke={guard_match,Ms,Rs}}; -bsm_rename_ctx(#l{ke={test,_,_}}=L, _, _, _) -> L; -bsm_rename_ctx(#l{ke={bif,_,_,_}}=L, _, _, _) -> L; -bsm_rename_ctx(#l{ke={gc_bif,_,_,_}}=L, _, _, _) -> L; -bsm_rename_ctx(#l{ke={set,_,_}}=L, _, _, _) -> L; -bsm_rename_ctx(#l{ke={call,_,_,_}}=L, _, _, _) -> L; -bsm_rename_ctx(#l{ke={block,_}}=L, Old, _, false) -> + Match#k_guard_match{body=Ms}; +bsm_rename_ctx(#k_test{}=Test, _, _, _) -> Test; +bsm_rename_ctx(#k_bif{}=Bif, _, _, _) -> Bif; +bsm_rename_ctx(#k_put{}=Put, _, _, _) -> Put; +bsm_rename_ctx(#k_call{}=Call, _, _, _) -> Call; +bsm_rename_ctx(#cg_block{}=Block, Old, _, false) -> %% This block is not inside a protected. The match context variable cannot %% possibly be live inside the block. - bsm_forget_var(L, Old); -bsm_rename_ctx(#l{ke={block,Bl0}}=L, Old, New, true) -> + bsm_forget_var(Block, Old); +bsm_rename_ctx(#cg_block{es=Es0}=Block, Old, New, true) -> %% A block in a protected. We must recursively rename the variable %% inside the block. - Bl = bsm_rename_ctx_list(Bl0, Old, New, true), - bsm_forget_var(L#l{ke={block,Bl}}, Old); -bsm_rename_ctx(#l{ke={guard_break,Bs,Locked0}}=L0, Old, _New, _InProt) -> - Locked = Locked0 -- [Old], - L = L0#l{ke={guard_break,Bs,Locked}}, - bsm_forget_var(L, Old). + Es = bsm_rename_ctx_list(Es0, Old, New, true), + bsm_forget_var(Block#cg_block{es=Es}, Old); +bsm_rename_ctx(#k_guard_break{}=Break, Old, _New, _InProt) -> + bsm_forget_var(Break, Old). bsm_rename_ctx_list([C|Cs], Old, New, InProt) -> [bsm_rename_ctx(C, Old, New, InProt)| bsm_rename_ctx_list(Cs, Old, New, InProt)]; bsm_rename_ctx_list([], _, _, _) -> []. - + bsm_rename_var(Old, Old, New) -> New; bsm_rename_var(V, _, _) -> V. %% bsm_forget_var(#l{}, Variable) -> #l{} %% Remove a variable from the variable life-time database. -bsm_forget_var(#l{vdb=Vdb}=L, V) -> - L#l{vdb=keydelete(V, 1, Vdb)}. +bsm_forget_var(Ke, V) -> + #l{vdb=Vdb} = L0 = get_kanno(Ke), + L = L0#l{vdb=keydelete(V, 1, Vdb)}, + set_kanno(Ke, L). %% block_cg([Kexpr], Le, Vdb, StackReg, St) -> {[Ainstr],StackReg,St}. %% block_cg([Kexpr], Le, StackReg, St) -> {[Ainstr],StackReg,St}. @@ -406,158 +767,227 @@ block_cg(Es, Le, _Vdb, Bef, St) -> block_cg(Es, Le, Bef, St). block_cg(Es, Le, Bef, #cg{is_top_block=false}=St) -> - cg_block(Es, Le#l.i, Le#l.vdb, Bef, St); -block_cg(Es, Le, Bef, St0) -> - {Is0,Aft,St} = cg_block(Es, Le#l.i, Le#l.vdb, Bef, - St0#cg{is_top_block=false,need_frame=false}), - Is = top_level_block(Is0, Aft, max_reg(Bef#sr.reg), St), - {Is,Aft,St#cg{is_top_block=true}}. - -cg_block([], _I, _Vdb, Bef, St0) -> + cg_block(Es, Le#l.vdb, Bef, St); +block_cg(Es, Le, Bef, #cg{is_top_block=true}=St0) -> + %% No stack frame has been established yet. Do we need one? + case need_stackframe(Es) of + true -> + %% We need a stack frame. Generate the code and add the + %% code for creating and deallocating the stack frame. + {Is0,Aft,St} = cg_block(Es, Le#l.vdb, Bef, + St0#cg{is_top_block=false,need_frame=false}), + Is = top_level_block(Is0, Aft, max_reg(Bef#sr.reg), St), + {Is,Aft,St#cg{is_top_block=true}}; + false -> + %% This sequence of instructions ending in a #k_match{} (a + %% 'case' or 'if') in the Erlang code does not need a + %% stack frame yet. Delay the creation (if a stack frame + %% is needed at all, it will be created inside the + %% #k_match{}). + cg_list(Es, Le#l.vdb, Bef, St0) + end. + +%% need_stackframe([Kexpr]) -> true|false. +%% Does this list of instructions need a stack frame? +%% +%% A sequence of instructions that don't clobber the X registers +%% followed by a single #k_match{} doesn't need a stack frame. + +need_stackframe([H|T]) -> + case H of + #k_bif{op=#k_internal{}} -> true; + #k_put{arg=#k_binary{}} -> true; + #k_bif{} -> need_stackframe(T); + #k_put{} -> need_stackframe(T); + #k_guard_match{} -> need_stackframe(T); + #k_match{} when T =:= [] -> false; + _ -> true + end; +need_stackframe([]) -> false. + +cg_block([], _Vdb, Bef, St0) -> {[],Bef,St0}; -cg_block(Kes0, I, Vdb, Bef, St0) -> +cg_block(Kes0, Vdb, Bef, St0) -> {Kes2,Int1,St1} = case basic_block(Kes0) of {Kes1,LastI,Args,Rest} -> - Ke = hd(Kes1), - Fb = Ke#l.i, - cg_basic_block(Kes1, Fb, LastI, Args, Vdb, Bef, St0); + cg_basic_block(Kes1, LastI, Args, Vdb, Bef, St0); {Kes1,Rest} -> - cg_list(Kes1, I, Vdb, Bef, St0) + cg_list(Kes1, Vdb, Bef, St0) end, - {Kes3,Int2,St2} = cg_block(Rest, I, Vdb, Int1, St1), + {Kes3,Int2,St2} = cg_block(Rest, Vdb, Int1, St1), {Kes2 ++ Kes3,Int2,St2}. basic_block(Kes) -> basic_block(Kes, []). -basic_block([Le|Les], Acc) -> - case collect_block(Le#l.ke) of - include -> basic_block(Les, [Le|Acc]); +basic_block([Ke|Kes], Acc) -> + case collect_block(Ke) of + include -> basic_block(Kes, [Ke|Acc]); {block_end,As} -> case Acc of [] -> - %% If the basic block does not contain any set instructions, + %% If the basic block does not contain any #k_put{} instructions, %% it serves no useful purpose to do basic block optimizations. - {[Le],Les}; + {[Ke],Kes}; _ -> - {reverse(Acc, [Le]),Le#l.i,As,Les} + #l{i=I} = get_kanno(Ke), + {reverse(Acc, [Ke]),I,As,Kes} end; - no_block -> {reverse(Acc, [Le]),Les} + no_block -> {reverse(Acc, [Ke]),Kes} end. -%% sets that may garbage collect are not allowed in basic blocks. - -collect_block({set,_,{binary,_}}) -> no_block; -collect_block({set,_,{map,_,_,_}}) -> no_block; -collect_block({set,_,_}) -> include; -collect_block({call,{var,_}=Var,As,_Rs}) -> {block_end,As++[Var]}; -collect_block({call,Func,As,_Rs}) -> {block_end,As++func_vars(Func)}; -collect_block({enter,{var,_}=Var,As})-> {block_end,As++[Var]}; -collect_block({enter,Func,As}) -> {block_end,As++func_vars(Func)}; -collect_block({return,Rs}) -> {block_end,Rs}; -collect_block({break,Bs}) -> {block_end,Bs}; -collect_block(_) -> no_block. - -func_vars({remote,M,F}) when element(1, M) =:= var; - element(1, F) =:= var -> +collect_block(#k_put{arg=Arg}) -> + %% #k_put{} instructions that may garbage collect are not allowed + %% in basic blocks. + case Arg of + #k_binary{} -> no_block; + #k_map{} -> no_block; + _ -> include + end; +collect_block(#k_call{op=Func,args=As}) -> + {block_end,As++func_vars(Func)}; +collect_block(#k_enter{op=Func,args=As}) -> + {block_end,As++func_vars(Func)}; +collect_block(#k_return{args=Rs}) -> + {block_end,Rs}; +collect_block(#k_break{args=Bs}) -> + {block_end,Bs}; +collect_block(_) -> no_block. + +func_vars(#k_var{}=Var) -> + [Var]; +func_vars(#k_remote{mod=M,name=F}) + when is_record(M, k_var); is_record(F, k_var) -> [M,F]; func_vars(_) -> []. -%% cg_basic_block([Kexpr], FirstI, LastI, As, Vdb, StackReg, State) -> +%% cg_basic_block([Kexpr], FirstI, LastI, Arguments, Vdb, StackReg, State) -> %% {[Ainstr],StackReg,State}. - -cg_basic_block(Kes, Fb, Lf, As, Vdb, Bef, St0) -> - Res = make_reservation(As, 0), - Regs0 = reserve(Res, Bef#sr.reg, Bef#sr.stk), - Stk = extend_stack(Bef, Lf, Lf+1, Vdb), - Int0 = Bef#sr{reg=Regs0,stk=Stk,res=Res}, - X0_v0 = x0_vars(As, Fb, Lf, Vdb), - {Keis,{Aft,_,St1}} = +%% +%% Do a specialized code generation for a basic block of #put{} +%% instructions (that don't do any garbage collection) followed by a +%% call, break, or return. +%% +%% 'Arguments' is a list of the variables that must be loaded into +%% consecutive X registers before the last instruction in the block. +%% The point of this specialized code generation is to try put the +%% all of the variables in 'Arguments' into the correct X register +%% to begin with, instead of putting them into the first available +%% X register and having to move them to the correct X register +%% later. +%% +%% To achieve that, we attempt to reserve the X registers that the +%% variables in 'Arguments' will need to be in when the block ends. +%% +%% To make it more likely that reservations will be successful, we +%% will try to save variables that need to be saved to the stack as +%% early as possible (if an X register needed by a variable in +%% Arguments is occupied by another variable, the value in the +%% X register can be evicted if it is saved on the stack). +%% +%% We will take care not to increase the size of stack frame compared +%% to what the standard code generator would have done (that is, to +%% save all X registers at the last possible moment). We will do that +%% by extending the stack frame to the minimal size needed to save +%% all that needs to be saved using extend_stack/4, and use +%% save_carefully/4 during code generation to only save the variables +%% that can be saved without growing the stack frame. + +cg_basic_block(Kes, Lf, As, Vdb, Bef, St0) -> + Int0 = reserve_arg_regs(As, Bef), + Int = extend_stack(Int0, Lf, Lf+1, Vdb), + {Keis,{Aft,St1}} = flatmapfoldl(fun(Ke, St) -> cg_basic_block(Ke, St, Lf, Vdb) end, - {Int0,X0_v0,St0}, need_heap(Kes, Fb)), + {Int,St0}, need_heap(Kes)), {Keis,Aft,St1}. -cg_basic_block(#l{ke={need_heap,_}}=Ke, {Inta,X0v,Sta}, _Lf, Vdb) -> - {Keis,Intb,Stb} = cg(Ke, Vdb, Inta, Sta), - {Keis, {Intb,X0v,Stb}}; -cg_basic_block(Ke, {Inta,X0_v1,Sta}, Lf, Vdb) -> - {Sis,Intb} = save_carefully(Inta, Ke#l.i, Lf+1, Vdb), - {X0_v2,Intc} = allocate_x0(X0_v1, Ke#l.i, Intb), - Intd = reserve(Intc), - {Keis,Inte,Stb} = cg(Ke, Vdb, Intd, Sta), - {Sis ++ Keis, {Inte,X0_v2,Stb}}. - -make_reservation([], _) -> []; -make_reservation([{var,V}|As], I) -> [{I,V}|make_reservation(As, I+1)]; -make_reservation([A|As], I) -> [{I,A}|make_reservation(As, I+1)]. - -reserve(Sr) -> Sr#sr{reg=reserve(Sr#sr.res, Sr#sr.reg, Sr#sr.stk)}. - -reserve([{I,V}|Rs], [free|Regs], Stk) -> [{reserved,I,V}|reserve(Rs, Regs, Stk)]; -reserve([{I,V}|Rs], [{I,V}|Regs], Stk) -> [{I,V}|reserve(Rs, Regs, Stk)]; -reserve([{I,V}|Rs], [{I,Var}|Regs], Stk) -> +cg_basic_block(#cg_need_heap{}=Ke, {Bef,St0}, _Lf, Vdb) -> + {Keis,Aft,St1} = cg(Ke, Vdb, Bef, St0), + {Keis,{Aft,St1}}; +cg_basic_block(Ke, {Bef,St0}, Lf, Vdb) -> + #l{i=I} = get_kanno(Ke), + + %% Save all we can to increase the possibility that reserving + %% registers will succeed. + {Sis,Int0} = save_carefully(Bef, I, Lf+1, Vdb), + Int1 = reserve(Int0), + {Keis,Aft,St1} = cg(Ke, Vdb, Int1, St0), + {Sis ++ Keis,{Aft,St1}}. + +%% reserve_arg_regs([Argument], Bef) -> Aft. +%% Try to reserve the X registers for all arguments. All registers +%% that we wish to reserve will be saved in Bef#sr.res. + +reserve_arg_regs(As, Bef) -> + Res = reserve_arg_regs_1(As, 0), + reserve(Bef#sr{res=Res}). + +reserve_arg_regs_1([#k_var{name=V}|As], I) -> + [{I,V}|reserve_arg_regs_1(As, I+1)]; +reserve_arg_regs_1([A|As], I) -> + [{I,A}|reserve_arg_regs_1(As, I+1)]; +reserve_arg_regs_1([], _) -> []. + +%% reserve(Bef) -> Aft. +%% Try to reserve more registers. The registers we wish to reserve +%% are found in Bef#sr.res. + +reserve(#sr{reg=Regs,stk=Stk,res=Res}=Sr) -> + Sr#sr{reg=reserve_1(Res, Regs, Stk)}. + +reserve_1([{I,V}|Rs], [free|Regs], Stk) -> + [{reserved,I,V}|reserve_1(Rs, Regs, Stk)]; +reserve_1([{I,V}|Rs], [{I,V}|Regs], Stk) -> + [{I,V}|reserve_1(Rs, Regs, Stk)]; +reserve_1([{I,V}|Rs], [{I,Var}|Regs], Stk) -> case on_stack(Var, Stk) of - true -> [{reserved,I,V}|reserve(Rs, Regs, Stk)]; - false -> [{I,Var}|reserve(Rs, Regs, Stk)] + true -> [{reserved,I,V}|reserve_1(Rs, Regs, Stk)]; + false -> [{I,Var}|reserve_1(Rs, Regs, Stk)] end; -reserve([{I,V}|Rs], [{reserved,I,_}|Regs], Stk) -> - [{reserved,I,V}|reserve(Rs, Regs, Stk)]; -%reserve([{I,V}|Rs], [Other|Regs], Stk) -> [Other|reserve(Rs, Regs, Stk)]; -reserve([{I,V}|Rs], [], Stk) -> [{reserved,I,V}|reserve(Rs, [], Stk)]; -reserve([], Regs, _) -> Regs. - -extend_stack(Bef, Fb, Lf, Vdb) -> - Stk0 = clear_dead_stk(Bef#sr.stk, Fb, Vdb), - Saves = [V || {V,F,L} <- Vdb, - F < Fb, - L >= Lf, - not on_stack(V, Stk0)], - Stk1 = foldl(fun (V, Stk) -> put_stack(V, Stk) end, Stk0, Saves), - Bef#sr.stk ++ lists:duplicate(length(Stk1) - length(Bef#sr.stk), free). - -save_carefully(Bef, Fb, Lf, Vdb) -> - Stk = Bef#sr.stk, - %% New variables that are in use but not on stack. - New = [VFL || {V,F,L} = VFL <- Vdb, - F < Fb, - L >= Lf, - not on_stack(V, Stk)], - Saves = [V || {V,_,_} <- keysort(2, New)], - save_carefully(Saves, Bef, []). - -save_carefully([], Bef, Acc) -> {reverse(Acc),Bef}; -save_carefully([V|Vs], Bef, Acc) -> - case put_stack_carefully(V, Bef#sr.stk) of - error -> {reverse(Acc),Bef}; +reserve_1([{I,V}|Rs], [{reserved,I,_}|Regs], Stk) -> + [{reserved,I,V}|reserve_1(Rs, Regs, Stk)]; +reserve_1([{I,V}|Rs], [], Stk) -> + [{reserved,I,V}|reserve_1(Rs, [], Stk)]; +reserve_1([], Regs, _) -> Regs. + +%% extend_stack(Bef, FirstBefore, LastFrom, Vdb) -> Aft. +%% Extend the stack enough to fit all variables alive past LastFrom +%% and not already on the stack. + +extend_stack(#sr{stk=Stk0}=Bef, Fb, Lf, Vdb) -> + Stk1 = clear_dead_stk(Stk0, Fb, Vdb), + New = new_not_on_stack(Stk1, Fb, Lf, Vdb), + Stk2 = foldl(fun ({V,_,_}, Stk) -> put_stack(V, Stk) end, Stk1, New), + Stk = Stk0 ++ lists:duplicate(length(Stk2) - length(Stk0), free), + Bef#sr{stk=Stk}. + +%% save_carefully(Bef, FirstBefore, LastFrom, Vdb) -> {[SaveVar],Aft}. +%% Save variables which are used past current point and which are not +%% already on the stack, but only if the variables can be saved without +%% growing the stack frame. + +save_carefully(#sr{stk=Stk}=Bef, Fb, Lf, Vdb) -> + New0 = new_not_on_stack(Stk, Fb, Lf, Vdb), + New = keysort(2, New0), + save_carefully_1(New, Bef, []). + +save_carefully_1([{V,_,_}|Vs], #sr{reg=Regs,stk=Stk0}=Bef, Acc) -> + case put_stack_carefully(V, Stk0) of + error -> + {reverse(Acc),Bef}; Stk1 -> - SrcReg = fetch_reg(V, Bef#sr.reg), + SrcReg = fetch_reg(V, Regs), Move = {move,SrcReg,fetch_stack(V, Stk1)}, {x,_} = SrcReg, %Assertion - must be X register. - save_carefully(Vs, Bef#sr{stk=Stk1}, [Move|Acc]) - end. + save_carefully_1(Vs, Bef#sr{stk=Stk1}, [Move|Acc]) + end; +save_carefully_1([], Bef, Acc) -> + {reverse(Acc),Bef}. -x0_vars([], _Fb, _Lf, _Vdb) -> []; -x0_vars([{var,V}|_], Fb, _Lf, Vdb) -> - {V,F,_L} = VFL = vdb_find(V, Vdb), - x0_vars1([VFL], Fb, F, Vdb); -x0_vars([X0|_], Fb, Lf, Vdb) -> - x0_vars1([{X0,Lf,Lf}], Fb, Lf, Vdb). - -x0_vars1(X0, Fb, Xf, Vdb) -> - Vs0 = [VFL || {_V,F,L}=VFL <- Vdb, - F >= Fb, - L < Xf], - Vs1 = keysort(3, Vs0), - keysort(2, X0++Vs1). - -allocate_x0([], _, Bef) -> {[],Bef#sr{res=[]}}; -allocate_x0([{_,_,L}|Vs], I, Bef) when L =< I -> - allocate_x0(Vs, I, Bef); -allocate_x0([{V,_F,_L}=VFL|Vs], _, Bef) -> - {[VFL|Vs],Bef#sr{res=reserve_x0(V, Bef#sr.res)}}. - -reserve_x0(V, [_|Res]) -> [{0,V}|Res]; -reserve_x0(V, []) -> [{0,V}]. +%% top_level_block([Instruction], Bef, MaxRegs, St) -> [Instruction]. +%% For the top-level block, allocate a stack frame a necessary, +%% adjust Y register numbering and instructions that return +%% from the function. top_level_block(Keis, #sr{stk=[]}, _MaxRegs, #cg{need_frame=false}) -> Keis; @@ -639,21 +1069,27 @@ turn_yreg(Other, _MaxY) -> %% wrong. These are different as in the second case there is no need %% to try the next type, it will always fail. -select_cg(#l{ke={type_clause,cons,[S]}}, {var,V}, Tf, Vf, Bef, St) -> +select_cg(#k_type_clause{type=Type,values=Vs}, Var, Tf, Vf, Bef, St) -> + #k_var{name=V} = Var, + select_cg(Type, Vs, V, Tf, Vf, Bef, St). + +select_cg(k_cons, [S], V, Tf, Vf, Bef, St) -> select_cons(S, V, Tf, Vf, Bef, St); -select_cg(#l{ke={type_clause,nil,[S]}}, {var,V}, Tf, Vf, Bef, St) -> +select_cg(k_nil, [S], V, Tf, Vf, Bef, St) -> select_nil(S, V, Tf, Vf, Bef, St); -select_cg(#l{ke={type_clause,binary,[S]}}, {var,V}, Tf, Vf, Bef, St) -> +select_cg(k_binary, [S], V, Tf, Vf, Bef, St) -> select_binary(S, V, Tf, Vf, Bef, St); -select_cg(#l{ke={type_clause,bin_seg,S}}, {var,V}, Tf, _Vf, Bef, St) -> +select_cg(k_bin_seg, S, V, Tf, _Vf, Bef, St) -> select_bin_segs(S, V, Tf, Bef, St); -select_cg(#l{ke={type_clause,bin_int,S}}, {var,V}, Tf, _Vf, Bef, St) -> +select_cg(k_bin_int, S, V, Tf, _Vf, Bef, St) -> select_bin_segs(S, V, Tf, Bef, St); -select_cg(#l{ke={type_clause,bin_end,[S]}}, {var,V}, Tf, _Vf, Bef, St) -> +select_cg(k_bin_end, [S], V, Tf, _Vf, Bef, St) -> select_bin_end(S, V, Tf, Bef, St); -select_cg(#l{ke={type_clause,map,S}}, {var,V}, Tf, Vf, Bef, St) -> +select_cg(k_map, S, V, Tf, Vf, Bef, St) -> select_map(S, V, Tf, Vf, Bef, St); -select_cg(#l{ke={type_clause,Type,Scs}}, {var,V}, Tf, Vf, Bef, St0) -> +select_cg(k_literal, S, V, Tf, Vf, Bef, St) -> + select_literal(S, V, Tf, Vf, Bef, St); +select_cg(Type, Scs, V, Tf, Vf, Bef, St0) -> {Vis,{Aft,St1}} = mapfoldl(fun (S, {Int,Sta}) -> {Val,Is,Inta,Stb} = select_val(S, V, Vf, Bef, Sta), @@ -663,22 +1099,29 @@ select_cg(#l{ke={type_clause,Type,Scs}}, {var,V}, Tf, Vf, Bef, St0) -> {Vls,Sis,St2} = select_labels(OptVls, St1, [], []), {select_val_cg(Type, fetch_var(V, Bef), Vls, Tf, Vf, Sis), Aft, St2}. -select_val_cg(tuple, R, [Arity,{f,Lbl}], Tf, Vf, [{label,Lbl}|Sis]) -> +select_val_cg(k_tuple, R, [Arity,{f,Lbl}], Tf, Vf, [{label,Lbl}|Sis]) -> [{test,is_tuple,{f,Tf},[R]},{test,test_arity,{f,Vf},[R,Arity]}|Sis]; -select_val_cg(tuple, R, Vls, Tf, Vf, Sis) -> +select_val_cg(k_tuple, R, Vls, Tf, Vf, Sis) -> [{test,is_tuple,{f,Tf},[R]},{select_tuple_arity,R,{f,Vf},{list,Vls}}|Sis]; select_val_cg(Type, R, [Val, {f,Lbl}], Fail, Fail, [{label,Lbl}|Sis]) -> - [{test,is_eq_exact,{f,Fail},[R,{Type,Val}]}|Sis]; + [{test,is_eq_exact,{f,Fail},[R,{type(Type),Val}]}|Sis]; select_val_cg(Type, R, [Val, {f,Lbl}], Tf, Vf, [{label,Lbl}|Sis]) -> [{test,select_type_test(Type),{f,Tf},[R]}, - {test,is_eq_exact,{f,Vf},[R,{Type,Val}]}|Sis]; + {test,is_eq_exact,{f,Vf},[R,{type(Type),Val}]}|Sis]; select_val_cg(Type, R, Vls0, Tf, Vf, Sis) -> - Vls1 = [case Value of {f,_Lbl} -> Value; _ -> {Type,Value} end || Value <- Vls0], + Vls1 = [case Value of + {f,_Lbl} -> Value; + _ -> {type(Type),Value} + end || Value <- Vls0], [{test,select_type_test(Type),{f,Tf},[R]}, {select_val,R,{f,Vf},{list,Vls1}}|Sis]. - -select_type_test(integer) -> is_integer; -select_type_test(atom) -> is_atom; -select_type_test(float) -> is_float. + +type(k_atom) -> atom; +type(k_float) -> float; +type(k_int) -> integer. + +select_type_test(k_int) -> is_integer; +select_type_test(k_atom) -> is_atom; +select_type_test(k_float) -> is_float. combine([{Is,Vs1}, {Is,Vs2}|Vis]) -> combine([{Is,Vs1 ++ Vs2}|Vis]); combine([V|Vis]) -> [V|combine(Vis)]; @@ -694,36 +1137,50 @@ add_vls([V|Vs], Lbl, Acc) -> add_vls(Vs, Lbl, [V, {f,Lbl}|Acc]); add_vls([], _, Acc) -> Acc. -select_cons(#l{ke={val_clause,{cons,Es},B},i=I,vdb=Vdb}, V, Tf, Vf, Bef, St0) -> +select_literal(S, V, Tf, Vf, Bef, St) -> + Reg = fetch_var(V, Bef), + F = fun(ValClause, Fail, St0) -> + {Val,Is,Aft,St1} = select_val(ValClause, V, Vf, Bef, St0), + Test = {test,is_eq_exact,{f,Fail},[Reg,{literal,Val}]}, + {[Test|Is],Aft,St1} + end, + match_fmf(F, Tf, St, S). + +select_cons(#k_val_clause{val=#k_cons{hd=Hd,tl=Tl},body=B,anno=#l{i=I,vdb=Vdb}}, + V, Tf, Vf, Bef, St0) -> + Es = [Hd,Tl], {Eis,Int,St1} = select_extract_cons(V, Es, I, Vdb, Bef, St0), {Bis,Aft,St2} = match_cg(B, Vf, Int, St1), {[{test,is_nonempty_list,{f,Tf},[fetch_var(V, Bef)]}] ++ Eis ++ Bis,Aft,St2}. -select_nil(#l{ke={val_clause,nil,B}}, V, Tf, Vf, Bef, St0) -> +select_nil(#k_val_clause{val=#k_nil{},body=B}, V, Tf, Vf, Bef, St0) -> {Bis,Aft,St1} = match_cg(B, Vf, Bef, St0), {[{test,is_nil,{f,Tf},[fetch_var(V, Bef)]}] ++ Bis,Aft,St1}. -select_binary(#l{ke={val_clause,{binary,{var,V}},B},i=I,vdb=Vdb}, - V, Tf, Vf, Bef, St0) -> +select_binary(#k_val_clause{val=#k_binary{segs=#k_var{name=V}},body=B, + anno=#l{i=I,vdb=Vdb}}, V, Tf, Vf, Bef, St0) -> + #cg{ctx=OldCtx} = St0, Int0 = clear_dead(Bef#sr{reg=Bef#sr.reg}, I, Vdb), - {Bis0,Aft,St1} = match_cg(B, Vf, Int0, St0), + {Bis0,Aft,St1} = match_cg(B, Vf, Int0, St0#cg{ctx=V}), CtxReg = fetch_var(V, Int0), Live = max_reg(Bef#sr.reg), - Bis1 = [{test,bs_start_match2,{f,Tf},Live,[CtxReg,V],CtxReg}, + Bis1 = [{test,bs_start_match2,{f,Tf},Live,[CtxReg,{context,V}],CtxReg}, {bs_save2,CtxReg,{V,V}}|Bis0], Bis = finish_select_binary(Bis1), - {Bis,Aft,St1}; -select_binary(#l{ke={val_clause,{binary,{var,Ivar}},B},i=I,vdb=Vdb}, - V, Tf, Vf, Bef, St0) -> + {Bis,Aft,St1#cg{ctx=OldCtx}}; +select_binary(#k_val_clause{val=#k_binary{segs=#k_var{name=Ivar}},body=B, + anno=#l{i=I,vdb=Vdb}}, V, Tf, Vf, Bef, St0) -> + #cg{ctx=OldCtx} = St0, Regs = put_reg(Ivar, Bef#sr.reg), Int0 = clear_dead(Bef#sr{reg=Regs}, I, Vdb), - {Bis0,Aft,St1} = match_cg(B, Vf, Int0, St0), + {Bis0,Aft,St1} = match_cg(B, Vf, Int0, St0#cg{ctx=Ivar}), CtxReg = fetch_var(Ivar, Int0), Live = max_reg(Bef#sr.reg), - Bis1 = [{test,bs_start_match2,{f,Tf},Live,[fetch_var(V, Bef),Ivar],CtxReg}, + Bis1 = [{test,bs_start_match2,{f,Tf},Live, + [fetch_var(V, Bef),{context,Ivar}],CtxReg}, {bs_save2,CtxReg,{Ivar,Ivar}}|Bis0], Bis = finish_select_binary(Bis1), - {Bis,Aft,St1}. + {Bis,Aft,St1#cg{ctx=OldCtx}}. finish_select_binary([{bs_save2,R,Point}=I,{bs_restore2,R,Point}|Is]) -> [I|finish_select_binary(Is)]; @@ -745,9 +1202,16 @@ select_bin_segs(Scs, Ivar, Tf, Bef, St) -> select_bin_seg(S, Ivar, Fail, Bef, Sta) end, Tf, St, Scs). -select_bin_seg(#l{ke={val_clause,{bin_seg,Ctx,Size,U,T,Fs0,Es},B},i=I,vdb=Vdb,a=A}, - Ivar, Fail, Bef, St0) -> +select_bin_seg(#k_val_clause{val=#k_bin_seg{size=Size,unit=U,type=T, + seg=Seg,flags=Fs0,next=Next}, + body=B, + anno=#l{i=I,vdb=Vdb,a=A}}, Ivar, Fail, Bef, St0) -> + Ctx = St0#cg.ctx, Fs = [{anno,A}|Fs0], + Es = case Next of + [] -> [Seg]; + _ -> [Seg,Next] + end, {Mis,Int,St1} = select_extract_bin(Es, Size, U, T, Fs, Fail, I, Vdb, Bef, Ctx, B, St0), {Bis,Aft,St2} = match_cg(B, Fail, Int, St1), @@ -760,9 +1224,12 @@ select_bin_seg(#l{ke={val_clause,{bin_seg,Ctx,Size,U,T,Fs0,Es},B},i=I,vdb=Vdb,a= [{bs_restore2,CtxReg,{Ctx,Ivar}}|Mis++Bis] end, {Is,Aft,St2}; -select_bin_seg(#l{ke={val_clause,{bin_int,Ctx,Sz,U,Fs,Val,Es},B},i=I,vdb=Vdb}, - Ivar, Fail, Bef, St0) -> - {Mis,Int,St1} = select_extract_int(Es, Val, Sz, U, Fs, Fail, +select_bin_seg(#k_val_clause{val=#k_bin_int{size=Sz,unit=U,flags=Fs, + val=Val,next=Next}, + body=B, + anno=#l{i=I,vdb=Vdb}}, Ivar, Fail, Bef, St0) -> + Ctx = St0#cg.ctx, + {Mis,Int,St1} = select_extract_int(Next, Val, Sz, U, Fs, Fail, I, Vdb, Bef, Ctx, St0), {Bis,Aft,St2} = match_cg(B, Fail, Int, St1), CtxReg = fetch_var(Ctx, Bef), @@ -783,7 +1250,7 @@ select_bin_seg(#l{ke={val_clause,{bin_int,Ctx,Sz,U,Fs,Val,Es},B},i=I,vdb=Vdb}, end, {[{bs_restore2,CtxReg,{Ctx,Ivar}}|Is],Aft,St2}. -select_extract_int([{var,Tl}], Val, {integer,Sz}, U, Fs, Vf, +select_extract_int(#k_var{name=Tl}, Val, #k_int{val=Sz}, U, Fs, Vf, I, Vdb, Bef, Ctx, St) -> Bits = U*Sz, Bin = case member(big, Fs) of @@ -804,7 +1271,7 @@ select_extract_int([{var,Tl}], Val, {integer,Sz}, U, Fs, Vf, end, {Is,clear_dead(Bef, I, Vdb),St}. -select_extract_bin([{var,Hd},{var,Tl}], Size0, Unit, Type, Flags, Vf, +select_extract_bin([#k_var{name=Hd},#k_var{name=Tl}], Size0, Unit, Type, Flags, Vf, I, Vdb, Bef, Ctx, _Body, St) -> SizeReg = get_bin_size_reg(Size0, Bef), {Es,Aft} = @@ -827,21 +1294,24 @@ select_extract_bin([{var,Hd},{var,Tl}], Size0, Unit, Type, Flags, Vf, {bs_save2,CtxReg,{Ctx,Tl}}],Int1} end, {Es,clear_dead(Aft, I, Vdb),St}; -select_extract_bin([{var,Hd}], Size0, Unit, binary, Flags, Vf, +select_extract_bin([#k_var{name=Hd}], Size, Unit, binary, Flags, Vf, I, Vdb, Bef, Ctx, Body, St) -> - SizeReg = get_bin_size_reg(Size0, Bef), + %% Match the last segment of a binary. We KNOW that the size + %% must be 'all'. + #k_atom{val=all} = Size, %Assertion. {Es,Aft} = case vdb_find(Hd, Vdb) of {_,_,Lhd} when Lhd =< I -> + %% The result will not be used. Furthermore, since we + %% we are at the end of the binary, the position will + %% not be used again; thus, it is safe to do a cheaper + %% test of the unit. CtxReg = fetch_var(Ctx, Bef), - {case SizeReg =:= {atom,all} andalso is_context_unused(Body) of - true when Unit =:= 1 -> + {case Unit of + 1 -> []; - true -> - [{test,bs_test_unit,{f,Vf},[CtxReg,Unit]}]; - false -> - [{test,bs_skip_bits2,{f,Vf}, - [CtxReg,SizeReg,Unit,{field_flags,Flags}]}] + _ -> + [{test,bs_test_unit,{f,Vf},[CtxReg,Unit]}] end,Bef}; {_,_,_} -> case is_context_unused(Body) of @@ -853,7 +1323,7 @@ select_extract_bin([{var,Hd}], Size0, Unit, binary, Flags, Vf, Name = bs_get_binary2, Live = max_reg(Bef#sr.reg), {[{test,Name,{f,Vf},Live, - [CtxReg,SizeReg,Unit,{field_flags,Flags}],Rhd}], + [CtxReg,atomic(Size),Unit,{field_flags,Flags}],Rhd}], Int1}; true -> %% Since the matching context will not be used again, @@ -868,36 +1338,42 @@ select_extract_bin([{var,Hd}], Size0, Unit, binary, Flags, Vf, Name = bs_get_binary2, Live = max_reg(Int1#sr.reg), {[{test,Name,{f,Vf},Live, - [CtxReg,SizeReg,Unit,{field_flags,Flags}],CtxReg}], + [CtxReg,atomic(Size),Unit,{field_flags,Flags}],CtxReg}], Int1} end end, {Es,clear_dead(Aft, I, Vdb),St}. %% is_context_unused(Ke) -> true | false -%% Simple heurististic to determine whether the code that follows will -%% use the current matching context again. (The information of liveness -%% calculcated by v3_life is too conservative to be useful for this purpose.) -%% 'true' means that the code that follows will definitely not use the context -%% again (because it is a block, not guard or matching code); 'false' that we -%% are not sure (there is either a guard, or more matching, either which may -%% reference the context again). - -is_context_unused(#l{ke=Ke}) -> is_context_unused(Ke); -is_context_unused({block,_}) -> true; -is_context_unused(_) -> false. - -select_bin_end(#l{ke={val_clause,{bin_end,Ctx},B}}, - Ivar, Tf, Bef, St0) -> +%% Simple heurististic to determine whether the code that follows +%% will use the current matching context again. (The liveness +%% information is too conservative to be useful for this purpose.) +%% 'true' means that the code that follows will definitely not use +%% the context again (because it is a block, not guard or matching +%% code); 'false' that we are not sure (there could be more +%% matching). + +is_context_unused(#k_alt{then=Then}) -> + %% #k_alt{} can be used for different purposes. If the Then part + %% is a block, it means that matching has finished and is used for a guard + %% to choose between the matched clauses. + is_context_unused(Then); +is_context_unused(#cg_block{}) -> + true; +is_context_unused(_) -> + false. + +select_bin_end(#k_val_clause{val=#k_bin_end{},body=B}, Ivar, Tf, Bef, St0) -> + Ctx = St0#cg.ctx, {Bis,Aft,St2} = match_cg(B, Tf, Bef, St0), CtxReg = fetch_var(Ctx, Bef), {[{bs_restore2,CtxReg,{Ctx,Ivar}}, {test,bs_test_tail2,{f,Tf},[CtxReg,0]}|Bis],Aft,St2}. -get_bin_size_reg({var,V}, Bef) -> +get_bin_size_reg(#k_var{name=V}, Bef) -> fetch_var(V, Bef); get_bin_size_reg(Literal, _Bef) -> - Literal. + atomic(Literal). build_bs_instr(Type, Vf, CtxReg, Live, SizeReg, Unit, Flags, Rhd) -> {Format,Name} = case Type of @@ -931,11 +1407,18 @@ build_skip_instr(Type, Vf, CtxReg, Live, SizeReg, Unit, Flags) -> {test,Name,{f,Vf},[CtxReg,Live,{field_flags,Flags}]} end. -select_val(#l{ke={val_clause,{tuple,Es},B},i=I,vdb=Vdb}, V, Vf, Bef, St0) -> +select_val(#k_val_clause{val=#k_tuple{es=Es},body=B,anno=#l{i=I,vdb=Vdb}}, + V, Vf, Bef, St0) -> {Eis,Int,St1} = select_extract_tuple(V, Es, I, Vdb, Bef, St0), {Bis,Aft,St2} = match_cg(B, Vf, Int, St1), {length(Es),Eis ++ Bis,Aft,St2}; -select_val(#l{ke={val_clause,{_,Val},B}}, _V, Vf, Bef, St0) -> +select_val(#k_val_clause{val=Val0,body=B}, _V, Vf, Bef, St0) -> + Val = case Val0 of + #k_atom{val=Lit} -> Lit; + #k_float{val=Lit} -> Lit; + #k_int{val=Lit} -> Lit; + #k_literal{val=Lit} -> Lit + end, {Bis,Aft,St1} = match_cg(B, Vf, Bef, St0), {Val,Bis,Aft,St1}. @@ -944,7 +1427,7 @@ select_val(#l{ke={val_clause,{_,Val},B}}, _V, Vf, Bef, St0) -> %% Extract tuple elements, but only if they do not immediately die. select_extract_tuple(Src, Vs, I, Vdb, Bef, St) -> - F = fun ({var,V}, {Int0,Elem}) -> + F = fun (#k_var{name=V}, {Int0,Elem}) -> case vdb_find(V, Vdb) of {V,_,L} when L =< I -> {[], {Int0,Elem+1}}; _Other -> @@ -961,9 +1444,10 @@ select_extract_tuple(Src, Vs, I, Vdb, Bef, St) -> select_map(Scs, V, Tf, Vf, Bef, St0) -> Reg = fetch_var(V, Bef), {Is,Aft,St1} = - match_fmf(fun(#l{ke={val_clause,{map,exact,_,Es},B},i=I,vdb=Vdb}, Fail, St1) -> - select_map_val(V, Es, B, Fail, I, Vdb, Bef, St1) - end, Vf, St0, Scs), + match_fmf(fun(#k_val_clause{val=#k_map{op=exact,es=Es}, + body=B,anno=#l{i=I,vdb=Vdb}}, Fail, St1) -> + select_map_val(V, Es, B, Fail, I, Vdb, Bef, St1) + end, Vf, St0, Scs), {[{test,is_map,{f,Tf},[Reg]}|Is],Aft,St1}. select_map_val(V, Es, B, Fail, I, Vdb, Bef, St0) -> @@ -980,29 +1464,32 @@ select_extract_map(Src, Vs, Fail, I, Vdb, Bef, St) -> %% Assume keys are term-sorted Rsrc = fetch_var(Src, Bef), - {{HasKs,GetVs,HasVarKs,GetVarVs},Aft} = lists:foldr(fun - ({map_pair,{var,K},{var,V}},{{HasKsi,GetVsi,HasVarVsi,GetVarVsi},Int0}) -> - case vdb_find(V, Vdb) of - {V,_,L} when L =< I -> - RK = fetch_var(K,Int0), - {{HasKsi,GetVsi,[RK|HasVarVsi],GetVarVsi},Int0}; - _Other -> - Reg1 = put_reg(V, Int0#sr.reg), - Int1 = Int0#sr{reg=Reg1}, - RK = fetch_var(K,Int0), - RV = fetch_reg(V,Reg1), - {{HasKsi,GetVsi,HasVarVsi,[[RK,RV]|GetVarVsi]},Int1} - end; - ({map_pair,Key,{var,V}},{{HasKsi,GetVsi,HasVarVsi,GetVarVsi},Int0}) -> - case vdb_find(V, Vdb) of - {V,_,L} when L =< I -> - {{[Key|HasKsi],GetVsi,HasVarVsi,GetVarVsi},Int0}; - _Other -> - Reg1 = put_reg(V, Int0#sr.reg), - Int1 = Int0#sr{reg=Reg1}, - {{HasKsi,[Key,fetch_reg(V, Reg1)|GetVsi],HasVarVsi,GetVarVsi},Int1} - end - end, {{[],[],[],[]},Bef}, Vs), + {{HasKs,GetVs,HasVarKs,GetVarVs},Aft} = + foldr(fun(#k_map_pair{key=#k_var{name=K},val=#k_var{name=V}}, + {{HasKsi,GetVsi,HasVarVsi,GetVarVsi},Int0}) -> + case vdb_find(V, Vdb) of + {V,_,L} when L =< I -> + RK = fetch_var(K,Int0), + {{HasKsi,GetVsi,[RK|HasVarVsi],GetVarVsi},Int0}; + _Other -> + Reg1 = put_reg(V, Int0#sr.reg), + Int1 = Int0#sr{reg=Reg1}, + RK = fetch_var(K,Int0), + RV = fetch_reg(V,Reg1), + {{HasKsi,GetVsi,HasVarVsi,[[RK,RV]|GetVarVsi]},Int1} + end; + (#k_map_pair{key=Key,val=#k_var{name=V}}, + {{HasKsi,GetVsi,HasVarVsi,GetVarVsi},Int0}) -> + case vdb_find(V, Vdb) of + {V,_,L} when L =< I -> + {{[atomic(Key)|HasKsi],GetVsi,HasVarVsi,GetVarVsi},Int0}; + _Other -> + Reg1 = put_reg(V, Int0#sr.reg), + Int1 = Int0#sr{reg=Reg1}, + {{HasKsi,[atomic(Key),fetch_reg(V, Reg1)|GetVsi], + HasVarVsi,GetVarVsi},Int1} + end + end, {{[],[],[],[]},Bef}, Vs), Code = [{test,has_map_fields,{f,Fail},Rsrc,{list,HasKs}} || HasKs =/= []] ++ [{test,has_map_fields,{f,Fail},Rsrc,{list,[K]}} || K <- HasVarKs] ++ @@ -1011,30 +1498,36 @@ select_extract_map(Src, Vs, Fail, I, Vdb, Bef, St) -> {Code, Aft, St}. -select_extract_cons(Src, [{var,Hd}, {var,Tl}], I, Vdb, Bef, St) -> - {Es,Aft} = case {vdb_find(Hd, Vdb), vdb_find(Tl, Vdb)} of - {{_,_,Lhd}, {_,_,Ltl}} when Lhd =< I, Ltl =< I -> - %% Both head and tail are dead. No need to generate - %% any instruction. - {[], Bef}; - _ -> - %% At least one of head and tail will be used, - %% but we must always fetch both. We will call - %% clear_dead/2 to allow reuse of the register - %% in case only of them is used. - - Reg0 = put_reg(Tl, put_reg(Hd, Bef#sr.reg)), - Int0 = Bef#sr{reg=Reg0}, - Rsrc = fetch_var(Src, Int0), - Rhd = fetch_reg(Hd, Reg0), - Rtl = fetch_reg(Tl, Reg0), - Int1 = clear_dead(Int0, I, Vdb), - {[{get_list,Rsrc,Rhd,Rtl}], Int1} - end, - {Es,Aft,St}. - +select_extract_cons(Src, [#k_var{name=Hd},#k_var{name=Tl}], I, Vdb, Bef, St) -> + Rsrc = fetch_var(Src, Bef), + Int = clear_dead(Bef, I, Vdb), + {{_,_,Lhd},{_,_,Ltl}} = {vdb_find(Hd, Vdb),vdb_find(Tl, Vdb)}, + case {Lhd =< I, Ltl =< I} of + {true,true} -> + %% Both dead. + {[],Bef,St}; + {true,false} -> + %% Head dead. + Reg0 = put_reg(Tl, Bef#sr.reg), + Aft = Int#sr{reg=Reg0}, + Rtl = fetch_reg(Tl, Reg0), + {[{get_tl,Rsrc,Rtl}],Aft,St}; + {false,true} -> + %% Tail dead. + Reg0 = put_reg(Hd, Bef#sr.reg), + Aft = Int#sr{reg=Reg0}, + Rhd = fetch_reg(Hd, Reg0), + {[{get_hd,Rsrc,Rhd}],Aft,St}; + {false,false} -> + %% Both used. + Reg0 = put_reg(Tl, put_reg(Hd, Bef#sr.reg)), + Aft = Bef#sr{reg=Reg0}, + Rhd = fetch_reg(Hd, Reg0), + Rtl = fetch_reg(Tl, Reg0), + {[{get_hd,Rsrc,Rhd},{get_tl,Rsrc,Rtl}],Aft,St} + end. -guard_clause_cg(#l{ke={guard_clause,G,B},vdb=Vdb}, Fail, Bef, St0) -> +guard_clause_cg(#k_guard_clause{anno=#l{vdb=Vdb},guard=G,body=B}, Fail, Bef, St0) -> {Gis,Int,St1} = guard_cg(G, Fail, Vdb, Bef, St0), {Bis,Aft,St} = match_cg(B, Fail, Int, St1), {Gis ++ Bis,Aft,St}. @@ -1047,18 +1540,37 @@ guard_clause_cg(#l{ke={guard_clause,G,B},vdb=Vdb}, Fail, Bef, St0) -> %% the correct exit point. Primops and tests all go to the next %% instruction on success or jump to a failure label. -guard_cg(#l{ke={protected,Ts,Rs},i=I,vdb=Pdb}, Fail, _Vdb, Bef, St) -> - protected_cg(Ts, Rs, Fail, I, Pdb, Bef, St); -guard_cg(#l{ke={block,Ts},i=I,vdb=Bdb}, Fail, _Vdb, Bef, St) -> - guard_cg_list(Ts, Fail, I, Bdb, Bef, St); -guard_cg(#l{ke={test,Test,As},i=I,vdb=_Tdb}, Fail, Vdb, Bef, St) -> - test_cg(Test, As, Fail, I, Vdb, Bef, St); +guard_cg(#k_protected{arg=Ts,ret=Rs,anno=#l{vdb=Pdb}}, Fail, _Vdb, Bef, St) -> + protected_cg(Ts, Rs, Fail, Pdb, Bef, St); +guard_cg(#k_test{anno=#l{i=I},op=Test0,args=As,inverted=Inverted}, + Fail, Vdb, Bef, St0) -> + #k_remote{mod=#k_atom{val=erlang},name=#k_atom{val=Test}} = Test0, + case Inverted of + false -> + test_cg(Test, As, Fail, I, Vdb, Bef, St0); + true -> + {Psucc,St1} = new_label(St0), + {Is,Aft,St2} = test_cg(Test, As, Psucc, I, Vdb, Bef, St1), + {Is++[{jump,{f,Fail}},{label,Psucc}],Aft,St2} + end; guard_cg(G, _Fail, Vdb, Bef, St) -> %%ok = io:fwrite("cg ~w: ~p~n", [?LINE,{G,Fail,Vdb,Bef}]), {Gis,Aft,St1} = cg(G, Vdb, Bef, St), %%ok = io:fwrite("cg ~w: ~p~n", [?LINE,{Aft}]), {Gis,Aft,St1}. +%% guard_cg_list([Kexpr], Fail, I, Vdb, StackReg, St) -> +%% {[Ainstr],StackReg,St}. + +guard_cg_list(Kes, Fail, Vdb, Bef, St0) -> + {Keis,{Aft,St1}} = + flatmapfoldl(fun (Ke, {Inta,Sta}) -> + {Keis,Intb,Stb} = + guard_cg(Ke, Fail, Vdb, Inta, Sta), + {Keis,{Intb,Stb}} + end, {Bef,St0}, need_heap(Kes)), + {Keis,Aft,St1}. + %% protected_cg([Kexpr], [Ret], Fail, I, Vdb, Bef, St) -> {[Ainstr],Aft,St}. %% Do a protected. Protecteds without return values are just done %% for effect, the return value is not checked, success passes on to @@ -1066,19 +1578,18 @@ guard_cg(G, _Fail, Vdb, Bef, St) -> %% return values then these must be set to 'false' on failure, %% control always passes to the next instruction. -protected_cg(Ts, [], Fail, I, Vdb, Bef, St0) -> +protected_cg(Ts, [], Fail, Vdb, Bef, St0) -> %% Protect these calls, revert when done. - {Tis,Aft,St1} = guard_cg_list(Ts, Fail, I, Vdb, Bef, - St0#cg{bfail=Fail}), + {Tis,Aft,St1} = guard_cg_list(Ts, Fail, Vdb, Bef, St0#cg{bfail=Fail}), {Tis,Aft,St1#cg{bfail=St0#cg.bfail}}; -protected_cg(Ts, Rs, _Fail, I, Vdb, Bef, St0) -> +protected_cg(Ts, Rs, _Fail, Vdb, Bef, St0) -> {Pfail,St1} = new_label(St0), {Psucc,St2} = new_label(St1), - {Tis,Aft,St3} = guard_cg_list(Ts, Pfail, I, Vdb, Bef, + {Tis,Aft,St3} = guard_cg_list(Ts, Pfail, Vdb, Bef, St2#cg{bfail=Pfail}), %%ok = io:fwrite("cg ~w: ~p~n", [?LINE,{Rs,I,Vdb,Aft}]), %% Set return values to false. - Mis = [{move,{atom,false},fetch_var(V,Aft)}||{var,V} <- Rs], + Mis = [{move,{atom,false},fetch_var(V,Aft)}||#k_var{name=V} <- Rs], {Tis ++ [{jump,{f,Psucc}}, {label,Pfail}] ++ Mis ++ [{label,Psucc}], Aft,St3#cg{bfail=St0#cg.bfail}}. @@ -1086,23 +1597,35 @@ protected_cg(Ts, Rs, _Fail, I, Vdb, Bef, St0) -> %% test_cg(TestName, Args, Fail, I, Vdb, Bef, St) -> {[Ainstr],Aft,St}. %% Generate test instruction. Use explicit fail label here. +test_cg(is_map, [A], Fail, I, Vdb, Bef, St) -> + %% We must avoid creating code like this: + %% + %% move x(0) y(0) + %% is_map Fail [x(0)] + %% make_fun => x(0) %% Overwrite x(0) + %% put_map_assoc y(0) ... + %% + %% The code is safe, but beam_validator does not understand that. + %% Extending beam_validator to handle such (rare) code as the + %% above would make it slower for all programs. Instead, change + %% the code generator to always prefer the Y register for is_map() + %% and put_map_assoc() instructions, ensuring that they use the + %% same register. + Arg = cg_reg_arg_prefer_y(A, Bef), + Aft = clear_dead(Bef, I, Vdb), + {[{test,is_map,{f,Fail},[Arg]}],Aft,St}; +test_cg(is_boolean, [#k_atom{val=Val}], Fail, I, Vdb, Bef, St) -> + Aft = clear_dead(Bef, I, Vdb), + Is = case is_boolean(Val) of + true -> []; + false -> [{jump,{f,Fail}}] + end, + {Is,Aft,St}; test_cg(Test, As, Fail, I, Vdb, Bef, St) -> Args = cg_reg_args(As, Bef), Aft = clear_dead(Bef, I, Vdb), {[beam_utils:bif_to_test(Test, Args, {f,Fail})],Aft,St}. -%% guard_cg_list([Kexpr], Fail, I, Vdb, StackReg, St) -> -%% {[Ainstr],StackReg,St}. - -guard_cg_list(Kes, Fail, I, Vdb, Bef, St0) -> - {Keis,{Aft,St1}} = - flatmapfoldl(fun (Ke, {Inta,Sta}) -> - {Keis,Intb,Stb} = - guard_cg(Ke, Fail, Vdb, Inta, Sta), - {Keis,{Intb,Stb}} - end, {Bef,St0}, need_heap(Kes, I)), - {Keis,Aft,St1}. - %% match_fmf(Fun, LastFail, State, [Clause]) -> {Is,Aft,State}. %% This is a special flatmapfoldl for match code gen where we %% generate a "failure" label for each clause. The last clause uses @@ -1125,7 +1648,7 @@ match_fmf(F, LastFail, St0, [H|T]) -> %% frame size. Finally the actual call is made. Call then needs the %% return values filled in. -call_cg({var,_V} = Var, As, Rs, Le, Vdb, Bef, St0) -> +call_cg(#k_var{}=Var, As, Rs, Le, Vdb, Bef, St0) -> {Sis,Int} = cg_setup_call(As++[Var], Bef, Le#l.i, Vdb), %% Put return values in registers. Reg = load_vars(Rs, clear_regs(Int#sr.reg)), @@ -1134,9 +1657,8 @@ call_cg({var,_V} = Var, As, Rs, Le, Vdb, Bef, St0) -> {Frees,Aft} = free_dead(clear_dead(Int#sr{reg=Reg}, Le#l.i, Vdb)), {Sis ++ Frees ++ [line(Le),{call_fun,Arity}],Aft, need_stack_frame(St0)}; -call_cg({remote,Mod,Name}, As, Rs, Le, Vdb, Bef, St0) - when element(1, Mod) =:= var; - element(1, Name) =:= var -> +call_cg(#k_remote{mod=Mod,name=Name}, As, Rs, Le, Vdb, Bef, St0) + when is_record(Mod, k_var); is_record(Name, k_var) -> {Sis,Int} = cg_setup_call(As++[Mod,Name], Bef, Le#l.i, Vdb), %% Put return values in registers. Reg = load_vars(Rs, clear_regs(Int#sr.reg)), @@ -1152,19 +1674,16 @@ call_cg(Func, As, Rs, Le, Vdb, Bef, St0) -> %% Inside a guard. The only allowed function call is to %% erlang:error/1,2. We will generate the following code: %% - %% jump FailureLabel %% move {atom,ok} DestReg - %% - %% The 'move' instruction will never be executed, but we - %% generate it anyway in case the beam_validator is run - %% on unoptimized code. - {remote,{atom,erlang},{atom,error}} = Func, %Assertion. - [{var,DestVar}] = Rs, + %% jump FailureLabel + #k_remote{mod=#k_atom{val=erlang}, + name=#k_atom{val=error}} = Func, %Assertion. + [#k_var{name=DestVar}] = Rs, Int0 = clear_dead(Bef, Le#l.i, Vdb), Reg = put_reg(DestVar, Int0#sr.reg), Int = Int0#sr{reg=Reg}, Dst = fetch_reg(DestVar, Reg), - {[{jump,{f,Fail}},{move,{atom,ok},Dst}], + {[{move,{atom,ok},Dst},{jump,{f,Fail}}], clear_dead(Int, Le#l.i, Vdb),St0}; #cg{} -> %% Ordinary function call in a function body. @@ -1178,11 +1697,11 @@ call_cg(Func, As, Rs, Le, Vdb, Bef, St0) -> {Sis ++ Frees ++ [line(Le)|Call],Aft,St1} end. -build_call({remote,{atom,erlang},{atom,'!'}}, 2, St0) -> +build_call(#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val='!'}}, 2, St0) -> {[send],need_stack_frame(St0)}; -build_call({remote,{atom,Mod},{atom,Name}}, Arity, St0) -> +build_call(#k_remote{mod=#k_atom{val=Mod},name=#k_atom{val=Name}}, Arity, St0) -> {[{call_ext,Arity,{extfunc,Mod,Name,Arity}}],need_stack_frame(St0)}; -build_call(Name, Arity, St0) when is_atom(Name) -> +build_call(#k_local{name=Name}, Arity, St0) when is_atom(Name) -> {Lbl,St1} = local_func_label(Name, Arity, need_stack_frame(St0)), {[{call,Arity,{f,Lbl}}],St1}. @@ -1198,16 +1717,15 @@ free_dead([Any|Stk], Y, Instr, StkAcc) -> free_dead(Stk, Y+1, Instr, [Any|StkAcc]); free_dead([], _, Instr, StkAcc) -> {Instr,reverse(StkAcc)}. -enter_cg({var,_V} = Var, As, Le, Vdb, Bef, St0) -> +enter_cg(#k_var{} = Var, As, Le, Vdb, Bef, St0) -> {Sis,Int} = cg_setup_call(As++[Var], Bef, Le#l.i, Vdb), %% Build complete code and final stack/register state. Arity = length(As), {Sis ++ [line(Le),{call_fun,Arity},return], clear_dead(Int#sr{reg=clear_regs(Int#sr.reg)}, Le#l.i, Vdb), need_stack_frame(St0)}; -enter_cg({remote,Mod,Name}, As, Le, Vdb, Bef, St0) - when element(1, Mod) =:= var; - element(1, Name) =:= var -> +enter_cg(#k_remote{mod=Mod,name=Name}, As, Le, Vdb, Bef, St0) + when is_record(Mod, k_var); is_record(Name, k_var) -> {Sis,Int} = cg_setup_call(As++[Mod,Name], Bef, Le#l.i, Vdb), %% Build complete code and final stack/register state. Arity = length(As), @@ -1225,19 +1743,19 @@ enter_cg(Func, As, Le, Vdb, Bef, St0) -> clear_dead(Int#sr{reg=clear_regs(Int#sr.reg)}, Le#l.i, Vdb), St1}. -build_enter({remote,{atom,erlang},{atom,'!'}}, 2, St0) -> +build_enter(#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val='!'}}, 2, St0) -> {[send,return],need_stack_frame(St0)}; -build_enter({remote,{atom,Mod},{atom,Name}}, Arity, St0) -> +build_enter(#k_remote{mod=#k_atom{val=Mod},name=#k_atom{val=Name}}, Arity, St0) -> St1 = case trap_bif(Mod, Name, Arity) of true -> need_stack_frame(St0); false -> St0 end, {[{call_ext_only,Arity,{extfunc,Mod,Name,Arity}}],St1}; -build_enter(Name, Arity, St0) when is_atom(Name) -> +build_enter(#k_local{name=Name}, Arity, St0) when is_atom(Name) -> {Lbl,St1} = local_func_label(Name, Arity, St0), {[{call_only,Arity,{f,Lbl}}],St1}. -enter_line({remote,{atom,Mod},{atom,Name}}, Arity, Le) -> +enter_line(#k_remote{mod=#k_atom{val=Mod},name=#k_atom{val=Name}}, Arity, Le) -> case erl_bifs:is_safe(Mod, Name, Arity) of false -> %% Tail-recursive call, possibly to a BIF. @@ -1285,37 +1803,78 @@ trap_bif(erlang, group_leader, 2) -> true; trap_bif(erlang, exit, 2) -> true; trap_bif(_, _, _) -> false. -%% bif_cg(Bif, [Arg], [Ret], Le, Vdb, StackReg, State) -> +%% bif_cg(#k_bif{}, Le, Vdb, StackReg, State) -> %% {[Ainstr],StackReg,State}. - -bif_cg(bs_context_to_binary=Instr, [Src0], [], Le, Vdb, Bef, St0) -> - [Src] = cg_reg_args([Src0], Bef), - case is_register(Src) of +%% Generate code a BIF. + +bif_cg(#k_bif{op=#k_internal{name=Name},args=As,ret=Rs}, Le, Vdb, Bef, St) -> + internal_cg(Name, As, Rs, Le, Vdb, Bef, St); +bif_cg(#k_bif{op=#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val=Name}}, + args=As,ret=Rs}, Le, Vdb, Bef, St) -> + Ar = length(As), + case is_gc_bif(Name, Ar) of false -> - {[],clear_dead(Bef, Le#l.i, Vdb), St0}; + bif_cg(Name, As, Rs, Le, Vdb, Bef, St); true -> - {[{Instr,Src}],clear_dead(Bef, Le#l.i, Vdb), St0} - end; -bif_cg(dsetelement, [Index0,Tuple0,New0], _Rs, Le, Vdb, Bef, St0) -> + gc_bif_cg(Name, As, Rs, Le, Vdb, Bef, St) + end. + +%% internal_cg(Bif, [Arg], [Ret], Le, Vdb, StackReg, State) -> +%% {[Ainstr],StackReg,State}. + +internal_cg(bs_context_to_binary=Instr, [Src0], [], Le, Vdb, Bef, St0) -> + [Src] = cg_reg_args([Src0], Bef), + {[{Instr,Src}],clear_dead(Bef, Le#l.i, Vdb), St0}; +internal_cg(dsetelement, [Index0,Tuple0,New0], _Rs, Le, Vdb, Bef, St0) -> [New,Tuple,{integer,Index1}] = cg_reg_args([New0,Tuple0,Index0], Bef), Index = Index1-1, {[{set_tuple_element,New,Tuple,Index}], clear_dead(Bef, Le#l.i, Vdb), St0}; -bif_cg({make_fun,Func,Arity,Index,Uniq}, As, Rs, Le, Vdb, Bef, St0) -> +internal_cg(make_fun, [Func0,Arity0|As], Rs, Le, Vdb, Bef, St0) -> %% This behaves more like a function call. + #k_atom{val=Func} = Func0, + #k_int{val=Arity} = Arity0, {Sis,Int} = cg_setup_call(As, Bef, Le#l.i, Vdb), Reg = load_vars(Rs, clear_regs(Int#sr.reg)), {FuncLbl,St1} = local_func_label(Func, Arity, St0), - MakeFun = {make_fun2,{f,FuncLbl},Index,Uniq,length(As)}, + MakeFun = {make_fun2,{f,FuncLbl},0,0,length(As)}, {Sis ++ [MakeFun], clear_dead(Int#sr{reg=Reg}, Le#l.i, Vdb), St1}; -bif_cg(bs_init_writable=I, As, Rs, Le, Vdb, Bef, St) -> +internal_cg(bs_init_writable=I, As, Rs, Le, Vdb, Bef, St) -> + %% This behaves like a function call. + {Sis,Int} = cg_setup_call(As, Bef, Le#l.i, Vdb), + Reg = load_vars(Rs, clear_regs(Int#sr.reg)), + {Sis++[I],clear_dead(Int#sr{reg=Reg}, Le#l.i, Vdb),St}; +internal_cg(build_stacktrace=I, As, Rs, Le, Vdb, Bef, St) -> %% This behaves like a function call. {Sis,Int} = cg_setup_call(As, Bef, Le#l.i, Vdb), Reg = load_vars(Rs, clear_regs(Int#sr.reg)), {Sis++[I],clear_dead(Int#sr{reg=Reg}, Le#l.i, Vdb),St}; -bif_cg(Bif, As, [{var,V}], Le, Vdb, Bef, St0) -> +internal_cg(raise, As, Rs, Le, Vdb, Bef, St) -> + %% raise can be treated like a guard BIF. + bif_cg(raise, As, Rs, Le, Vdb, Bef, St); +internal_cg(guard_error, [ExitCall], _Rs, Le, Vdb, Bef, St) -> + %% A call an exit BIF from inside a #k_guard_match{}. + %% Generate a standard call, but leave the register descriptors + %% alone, effectively pretending that there was no call. + #k_call{op=#k_remote{mod=#k_atom{val=Mod},name=#k_atom{val=Name}}, + args=As} = ExitCall, + Arity = length(As), + {Ms,_} = cg_call_args(As, Bef, Le#l.i, Vdb), + Call = {call_ext,Arity,{extfunc,Mod,Name,Arity}}, + Is = Ms++[line(Le),Call], + {Is,Bef,St}; +internal_cg(raw_raise=I, As, Rs, Le, Vdb, Bef, St) -> + %% This behaves like a function call. + {Sis,Int} = cg_setup_call(As, Bef, Le#l.i, Vdb), + Reg = load_vars(Rs, clear_regs(Int#sr.reg)), + {Sis++[I],clear_dead(Int#sr{reg=Reg}, Le#l.i, Vdb),St}. + +%% bif_cg(Bif, [Arg], [Ret], Le, Vdb, StackReg, State) -> +%% {[Ainstr],StackReg,State}. + +bif_cg(Bif, As, [#k_var{name=V}], Le, Vdb, Bef, St0) -> Ars = cg_reg_args(As, Bef), %% If we are inside a catch and in a body (not in guard) and the @@ -1353,7 +1912,7 @@ bif_cg(Bif, As, [{var,V}], Le, Vdb, Bef, St0) -> %% gc_bif_cg(Bif, [Arg], [Ret], Le, Vdb, StackReg, State) -> %% {[Ainstr],StackReg,State}. -gc_bif_cg(Bif, As, [{var,V}], Le, Vdb, Bef, St0) -> +gc_bif_cg(Bif, As, [#k_var{name=V}], Le, Vdb, Bef, St0) -> Ars = cg_reg_args(As, Bef), %% If we are inside a catch and in a body (not in guard) and the @@ -1399,7 +1958,7 @@ recv_loop_cg(Te, Rvar, Rm, Tes, Rs, Le, Vdb, Bef, St0) -> %% cg_recv_mesg( ) -> {[Ainstr],Aft,St}. -cg_recv_mesg({var,R}, Rm, Tl, Bef, St0) -> +cg_recv_mesg(#k_var{name=R}, Rm, Tl, Bef, St0) -> Int0 = Bef#sr{reg=put_reg(R, Bef#sr.reg)}, Ret = fetch_reg(R, Int0#sr.reg), %% Int1 = clear_dead(Int0, I, Rm#l.vdb), @@ -1409,22 +1968,22 @@ cg_recv_mesg({var,R}, Rm, Tl, Bef, St0) -> %% cg_recv_wait(Te, Tes, I, Vdb, Int2, St3) -> {[Ainstr],Aft,St}. -cg_recv_wait({atom,infinity}, Tes, I, Bef, St0) -> +cg_recv_wait(#k_atom{val=infinity}, #cg_block{anno=Le,es=Tes}, I, Bef, St0) -> %% We know that the 'after' body will never be executed. %% But to keep the stack and register information up to date, %% we will generate the code for the 'after' body, and then discard it. - Int1 = clear_dead(Bef, I, Tes#l.vdb), - {_,Int2,St1} = cg_block(Tes#l.ke, Tes#l.i, Tes#l.vdb, - Int1#sr{reg=clear_regs(Int1#sr.reg)}, St0), + Int1 = clear_dead(Bef, I, Le#l.vdb), + {_,Int2,St1} = cg_block(Tes, Le#l.vdb, + Int1#sr{reg=clear_regs(Int1#sr.reg)}, St0), {[{wait,{f,St1#cg.recv}}],Int2,St1}; -cg_recv_wait({integer,0}, Tes, _I, Bef, St0) -> - {Tis,Int,St1} = cg_block(Tes#l.ke, Tes#l.i, Tes#l.vdb, Bef, St0), +cg_recv_wait(#k_int{val=0}, #cg_block{anno=Le,es=Tes}, _I, Bef, St0) -> + {Tis,Int,St1} = cg_block(Tes, Le#l.vdb, Bef, St0), {[timeout|Tis],Int,St1}; -cg_recv_wait(Te, Tes, I, Bef, St0) -> +cg_recv_wait(Te, #cg_block{anno=Le,es=Tes}, I, Bef, St0) -> Reg = cg_reg_arg(Te, Bef), %% Must have empty registers here! Bug if anything in registers. - Int0 = clear_dead(Bef, I, Tes#l.vdb), - {Tis,Int,St1} = cg_block(Tes#l.ke, Tes#l.i, Tes#l.vdb, + Int0 = clear_dead(Bef, I, Le#l.vdb), + {Tis,Int,St1} = cg_block(Tes, Le#l.vdb, Int0#sr{reg=clear_regs(Int0#sr.reg)}, St0), {[{wait_timeout,{f,St1#cg.recv},Reg},timeout] ++ Tis,Int,St1}. @@ -1442,7 +2001,7 @@ try_cg(Ta, Vs, Tb, Evs, Th, Rs, Le, Vdb, Bef, St0) -> {B,St1} = new_label(St0), %Body label {H,St2} = new_label(St1), %Handler label {E,St3} = new_label(St2), %End label - TryTag = Ta#l.i, + #l{i=TryTag} = get_kanno(Ta), Int1 = Bef#sr{stk=put_catch(TryTag, Bef#sr.stk)}, TryReg = fetch_stack({catch_tag,TryTag}, Int1#sr.stk), {Ais,Int2,St4} = cg(Ta, Vdb, Int1, St3#cg{break=B,in_catch=true}), @@ -1462,7 +2021,7 @@ try_cg(Ta, Vs, Tb, Evs, Th, Rs, Le, Vdb, Bef, St0) -> try_enter_cg(Ta, Vs, Tb, Evs, Th, Le, Vdb, Bef, St0) -> {B,St1} = new_label(St0), %Body label {H,St2} = new_label(St1), %Handler label - TryTag = Ta#l.i, + #l{i=TryTag} = get_kanno(Ta), Int1 = Bef#sr{stk=put_catch(TryTag, Bef#sr.stk)}, TryReg = fetch_stack({catch_tag,TryTag}, Int1#sr.stk), {Ais,Int2,St3} = cg(Ta, Vdb, Int1, St2#cg{break=B,in_catch=true}), @@ -1480,12 +2039,12 @@ try_enter_cg(Ta, Vs, Tb, Evs, Th, Le, Vdb, Bef, St0) -> %% catch_cg(CatchBlock, Ret, Le, Vdb, Bef, St) -> {[Ainstr],Aft,St}. -catch_cg(C, {var,R}, Le, Vdb, Bef, St0) -> +catch_cg(#cg_block{es=C}, #k_var{name=R}, Le, Vdb, Bef, St0) -> {B,St1} = new_label(St0), CatchTag = Le#l.i, Int1 = Bef#sr{stk=put_catch(CatchTag, Bef#sr.stk)}, CatchReg = fetch_stack({catch_tag,CatchTag}, Int1#sr.stk), - {Cis,Int2,St2} = cg_block(C, Le#l.i, Le#l.vdb, Int1, + {Cis,Int2,St2} = cg_block(C, Le#l.vdb, Int1, St1#cg{break=B,in_catch=true}), [] = Int2#sr.reg, %Assertion. Aft = Int2#sr{reg=[{0,R}],stk=drop_catch(CatchTag, Int2#sr.stk)}, @@ -1494,8 +2053,8 @@ catch_cg(C, {var,R}, Le, Vdb, Bef, St0) -> clear_dead(Aft, Le#l.i, Vdb), St2#cg{break=St1#cg.break,in_catch=St1#cg.in_catch}}. -%% set_cg([Var], Constr, Le, Vdb, Bef, St) -> {[Ainstr],Aft,St}. -%% We have to be careful how a 'set' works. First the structure is +%% put_cg([Var], Constr, Le, Vdb, Bef, St) -> {[Ainstr],Aft,St}. +%% We have to be careful how a 'put' works. First the structure is %% built, then it is filled and finally things can be cleared. The %% annotation must reflect this and make sure that the return %% variable is allocated first. @@ -1503,13 +2062,14 @@ catch_cg(C, {var,R}, Le, Vdb, Bef, St0) -> %% put_list and put_map are atomic instructions, both of %% which can safely resuse one of the source registers as target. -set_cg([{var,R}], {cons,Es}, Le, Vdb, Bef, St) -> - [S1,S2] = cg_reg_args(Es, Bef), +put_cg([#k_var{name=R}], #k_cons{hd=Hd,tl=Tl}, Le, Vdb, Bef, St) -> + [S1,S2] = cg_reg_args([Hd,Tl], Bef), Int0 = clear_dead(Bef, Le#l.i, Vdb), Int1 = Int0#sr{reg=put_reg(R, Int0#sr.reg)}, Ret = fetch_reg(R, Int1#sr.reg), {[{put_list,S1,S2,Ret}], Int1, St}; -set_cg([{var,R}], {binary,Segs}, Le, Vdb, Bef, #cg{bfail=Bfail}=St) -> +put_cg([#k_var{name=R}], #k_binary{segs=Segs}, Le, Vdb, Bef, + #cg{bfail=Bfail}=St) -> %% At run-time, binaries are constructed in three stages: %% 1) First the size of the binary is calculated. %% 2) Then the binary is allocated. @@ -1535,14 +2095,14 @@ set_cg([{var,R}], {binary,Segs}, Le, Vdb, Bef, #cg{bfail=Bfail}=St) -> %% Now generate the complete code for constructing the binary. Code = cg_binary(PutCode, Target, Temp, Fail, MaxRegs, Le#l.a), {Sis++Code,Aft,St}; -% Map single variable key -set_cg([{var,R}], {map,Op,Map,[{map_pair,{var,_}=K,V}]}, Le, Vdb, Bef, - #cg{bfail=Bfail}=St) -> - Fail = {f,Bfail}, - {Sis,Int0} = maybe_adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb, St), +%% Map: single variable key. +put_cg([#k_var{name=R}], #k_map{op=Op,var=Map, + es=[#k_map_pair{key=#k_var{}=K,val=V}]}, + Le, Vdb, Bef, St0) -> + {Sis,Int0} = maybe_adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb, St0), - SrcReg = cg_reg_arg(Map,Int0), + SrcReg = cg_reg_arg_prefer_y(Map, Int0), Line = line(Le#l.a), List = [cg_reg_arg(K,Int0),cg_reg_arg(V,Int0)], @@ -1554,27 +2114,23 @@ set_cg([{var,R}], {map,Op,Map,[{map_pair,{var,_}=K,V}]}, Le, Vdb, Bef, Aft = Aft0#sr{reg=put_reg(R, Aft0#sr.reg)}, Target = fetch_reg(R, Aft#sr.reg), - I = case Op of - assoc -> put_map_assoc; - exact -> put_map_exact - end, - {Sis++[Line]++[{I,Fail,SrcReg,Target,Live,{list,List}}],Aft,St}; + {Is,St1} = put_cg_map(Line, Op, SrcReg, Target, Live, List, St0), + {Sis++Is,Aft,St1}; -% Map (possibly) multiple literal keys -set_cg([{var,R}], {map,Op,Map,Es}, Le, Vdb, Bef, - #cg{bfail=Bfail}=St) -> +%% Map: (possibly) multiple literal keys. +put_cg([#k_var{name=R}], #k_map{op=Op,var=Map,es=Es}, Le, Vdb, Bef, St0) -> %% assert key literals - [] = [Var||{map_pair,{var,_}=Var,_} <- Es], + [] = [Var || #k_map_pair{key=#k_var{}=Var} <- Es], - Fail = {f,Bfail}, - {Sis,Int0} = maybe_adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb, St), - SrcReg = cg_reg_arg(Map,Int0), + {Sis,Int0} = maybe_adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb, St0), + SrcReg = cg_reg_arg_prefer_y(Map, Int0), Line = line(Le#l.a), %% fetch registers for values to be put into the map - Pairs = [{K,V} || {_,K,V} <- Es], - List = flatmap(fun({K,V}) -> [K,cg_reg_arg(V,Int0)] end, Pairs), + List = flatmap(fun(#k_map_pair{key=K,val=V}) -> + [atomic(K),cg_reg_arg(V, Int0)] + end, Es), Live = max_reg(Bef#sr.reg), @@ -1583,23 +2139,50 @@ set_cg([{var,R}], {map,Op,Map,Es}, Le, Vdb, Bef, Aft = Aft0#sr{reg=put_reg(R, Aft0#sr.reg)}, Target = fetch_reg(R, Aft#sr.reg), - I = case Op of - assoc -> put_map_assoc; - exact -> put_map_exact - end, - {Sis++[Line]++[{I,Fail,SrcReg,Target,Live,{list,List}}],Aft,St}; -set_cg([{var,R}], Con, Le, Vdb, Bef, St) -> + {Is,St1} = put_cg_map(Line, Op, SrcReg, Target, Live, List, St0), + {Sis++Is,Aft,St1}; + +%% Everything else. +put_cg([#k_var{name=R}], Con, Le, Vdb, Bef, St) -> %% Find a place for the return register first. Int = Bef#sr{reg=put_reg(R, Bef#sr.reg)}, Ret = fetch_reg(R, Int#sr.reg), Ais = case Con of - {tuple,Es} -> + #k_tuple{es=Es} -> [{put_tuple,length(Es),Ret}] ++ cg_build_args(Es, Bef); Other -> [{move,cg_reg_arg(Other, Int),Ret}] end, {Ais,clear_dead(Int, Le#l.i, Vdb),St}. + +put_cg_map(Line, Op0, SrcReg, Target, Live, List, St0) -> + Bfail = St0#cg.bfail, + Fail = {f,St0#cg.bfail}, + Op = case Op0 of + assoc -> put_map_assoc; + exact -> put_map_exact + end, + {OkLbl,St1} = new_label(St0), + {BadLbl,St2} = new_label(St1), + Is = if + Bfail =:= 0 orelse Op =:= put_map_assoc -> + [Line,{Op,{f,0},SrcReg,Target,Live,{list,List}}]; + true -> + %% Ensure that Target is always set, even if + %% the map update operation fails. That is necessary + %% because Target may be included in a test_heap + %% instruction. + [Line, + {Op,{f,BadLbl},SrcReg,Target,Live,{list,List}}, + {jump,{f,OkLbl}}, + {label,BadLbl}, + {move,{atom,ok},Target}, + {jump,Fail}, + {label,OkLbl}] + end, + {Is,St2}. + %%% %%% Code generation for constructing binaries. %%% @@ -1751,24 +2334,44 @@ cg_gen_binsize([], _, _, _, _, Acc) -> Acc. %% cg_bin_opt(Code0) -> Code %% Optimize the size calculations for binary construction. -cg_bin_opt([{move,Size,D},{bs_append,Fail,D,Extra,Regs,U,Bin,Flags,D}|Is]) -> - cg_bin_opt([{bs_append,Fail,Size,Extra,Regs,U,Bin,Flags,D}|Is]); -cg_bin_opt([{move,Size,D},{bs_private_append,Fail,D,U,Bin,Flags,D}|Is]) -> - cg_bin_opt([{bs_private_append,Fail,Size,U,Bin,Flags,D}|Is]); -cg_bin_opt([{move,{integer,0},D},{bs_add,_,[D,{integer,_}=S,1],Dst}|Is]) -> - cg_bin_opt([{move,S,Dst}|Is]); -cg_bin_opt([{move,{integer,0},D},{bs_add,Fail,[D,S,U],Dst}|Is]) -> - cg_bin_opt([{bs_add,Fail,[{integer,0},S,U],Dst}|Is]); -cg_bin_opt([{move,{integer,Bytes},D},{Op,Fail,D,Extra,Regs,Flags,D}|Is]) +cg_bin_opt([{move,S1,{x,X}=D},{gc_bif,Op,Fail,Live0,As,Dst}|Is]) -> + Live = if + X + 1 =:= Live0 -> X; + true -> Live0 + end, + [{gc_bif,Op,Fail,Live,As,D}|cg_bin_opt([{move,S1,Dst}|Is])]; +cg_bin_opt([{move,_,_}=I1,{Op,_,_,_}=I2|Is]) + when Op =:= bs_utf8_size orelse Op =:= bs_utf16_size -> + [I2|cg_bin_opt([I1|Is])]; +cg_bin_opt([{bs_add,_,[{integer,0},Src,1],Dst}|Is]) -> + cg_bin_opt_1([{move,Src,Dst}|Is]); +cg_bin_opt([{bs_add,_,[Src,{integer,0},_],Dst}|Is]) -> + cg_bin_opt_1([{move,Src,Dst}|Is]); +cg_bin_opt(Is) -> + cg_bin_opt_1(Is). + +cg_bin_opt_1([{move,Size,D},{bs_append,Fail,D,Extra,Regs,U,Bin,Flags,D}|Is]) -> + [{bs_append,Fail,Size,Extra,Regs,U,Bin,Flags,D}|cg_bin_opt(Is)]; +cg_bin_opt_1([{move,Size,D},{bs_private_append,Fail,D,U,Bin,Flags,D}|Is]) -> + [{bs_private_append,Fail,Size,U,Bin,Flags,D}|cg_bin_opt(Is)]; +cg_bin_opt_1([{move,Size,D},{Op,Fail,D,Extra,Regs,Flags,D}|Is]) when Op =:= bs_init2; Op =:= bs_init_bits -> - cg_bin_opt([{Op,Fail,Bytes,Extra,Regs,Flags,D}|Is]); -cg_bin_opt([{move,Src1,Dst},{bs_add,Fail,[Dst,Src2,U],Dst}|Is]) -> - cg_bin_opt([{bs_add,Fail,[Src1,Src2,U],Dst}|Is]); -cg_bin_opt([I|Is]) -> + Bytes = case Size of + {integer,Int} -> Int; + _ -> Size + end, + [{Op,Fail,Bytes,Extra,Regs,Flags,D}|cg_bin_opt(Is)]; +cg_bin_opt_1([{move,S1,D},{bs_add,Fail,[D,S2,U],Dst}|Is]) -> + cg_bin_opt([{bs_add,Fail,[S1,S2,U],Dst}|Is]); +cg_bin_opt_1([{move,S1,D},{bs_add,Fail,[S2,D,U],Dst}|Is]) -> + cg_bin_opt([{bs_add,Fail,[S2,S1,U],Dst}|Is]); +cg_bin_opt_1([I|Is]) -> [I|cg_bin_opt(Is)]; -cg_bin_opt([]) -> []. +cg_bin_opt_1([]) -> + []. -cg_bin_put({bin_seg,[],S0,U,T,Fs,[E0,Next]}, Fail, Bef) -> +cg_bin_put(#k_bin_seg{size=S0,unit=U,type=T,flags=Fs,seg=E0,next=Next}, + Fail, Bef) -> S1 = cg_reg_arg(S0, Bef), E1 = cg_reg_arg(E0, Bef), {Format,Op} = case T of @@ -1785,7 +2388,7 @@ cg_bin_put({bin_seg,[],S0,U,T,Fs,[E0,Next]}, Fail, Bef) -> utf -> [{Op,Fail,{field_flags,Fs},E1}|cg_bin_put(Next, Fail, Bef)] end; -cg_bin_put({bin_end,[]}, _, _) -> []. +cg_bin_put(#k_bin_end{}, _, _) -> []. cg_build_args(As, Bef) -> [{put,cg_reg_arg(A, Bef)} || A <- As]. @@ -1805,13 +2408,12 @@ break_cg(Bs, Le, Vdb, Bef, St) -> {Ms ++ [{jump,{f,St#cg.break}}], Int#sr{reg=clear_regs(Int#sr.reg)},St}. -guard_break_cg(Bs, Locked, #l{i=I}, Vdb, #sr{reg=Reg0}=Bef, St) -> - RegLocked = get_locked_regs(Reg0, Locked), - #sr{reg=Reg1} = Int = clear_dead(Bef#sr{reg=RegLocked}, I, Vdb), +guard_break_cg(Bs, #l{i=I}, Vdb, #sr{reg=Reg0}=Bef, St) -> + #sr{reg=Reg1} = Int = clear_dead(Bef, I, Vdb), Reg2 = trim_free(Reg1), NumLocked = length(Reg2), Moves0 = gen_moves(Bs, Bef, NumLocked, []), - Moves = order_moves(Moves0, find_scratch_reg(RegLocked)), + Moves = order_moves(Moves0, find_scratch_reg(Reg0)), {BreakVars,_} = mapfoldl(fun(_, RegNum) -> {{RegNum,gbreakvar},RegNum+1} end, length(Reg2), Bs), @@ -1819,28 +2421,17 @@ guard_break_cg(Bs, Locked, #l{i=I}, Vdb, #sr{reg=Reg0}=Bef, St) -> Aft = Int#sr{reg=Reg}, {Moves ++ [{jump,{f,St#cg.break}}],Aft,St}. -get_locked_regs([R|Rs0], Preserve) -> - case {get_locked_regs(Rs0, Preserve),R} of - {[],{_,V}} -> - case lists:member(V, Preserve) of - true -> [R]; - false -> [] - end; - {[],_} -> - []; - {Rs,_} -> - [R|Rs] - end; -get_locked_regs([], _) -> []. - %% cg_reg_arg(Arg0, Info) -> Arg %% cg_reg_args([Arg0], Info) -> [Arg] %% Convert argument[s] into registers. Literal values are returned unchanged. cg_reg_args(As, Bef) -> [cg_reg_arg(A, Bef) || A <- As]. -cg_reg_arg({var,V}, Bef) -> fetch_var(V, Bef); -cg_reg_arg(Literal, _) -> Literal. +cg_reg_arg(#k_var{name=V}, Bef) -> fetch_var(V, Bef); +cg_reg_arg(Literal, _) -> atomic(Literal). + +cg_reg_arg_prefer_y(#k_var{name=V}, Bef) -> fetch_var_prefer_y(V, Bef); +cg_reg_arg_prefer_y(Literal, _) -> atomic(Literal). %% cg_setup_call([Arg], Bef, Cur, Vdb) -> {[Instr],Aft}. %% Do the complete setup for a call/enter. @@ -1878,9 +2469,9 @@ cg_call_args(As, Bef, I, Vdb) -> load_arg_regs(Regs, As) -> load_arg_regs(Regs, As, 0). -load_arg_regs([_|Rs], [{var,V}|As], I) -> [{I,V}|load_arg_regs(Rs, As, I+1)]; +load_arg_regs([_|Rs], [#k_var{name=V}|As], I) -> [{I,V}|load_arg_regs(Rs, As, I+1)]; load_arg_regs([_|Rs], [A|As], I) -> [{I,A}|load_arg_regs(Rs, As, I+1)]; -load_arg_regs([], [{var,V}|As], I) -> [{I,V}|load_arg_regs([], As, I+1)]; +load_arg_regs([], [#k_var{name=V}|As], I) -> [{I,V}|load_arg_regs([], As, I+1)]; load_arg_regs([], [A|As], I) -> [{I,A}|load_arg_regs([], As, I+1)]; load_arg_regs(Rs, [], _) -> Rs. @@ -1916,12 +2507,13 @@ move_unsaved([], _, Regs, Acc) -> {Acc,Regs}. gen_moves(As, Sr) -> gen_moves(As, Sr, 0, []). -gen_moves([{var,V}|As], Sr, I, Acc) -> +gen_moves([#k_var{name=V}|As], Sr, I, Acc) -> case fetch_var(V, Sr) of {x,I} -> gen_moves(As, Sr, I+1, Acc); Reg -> gen_moves(As, Sr, I+1, [{move,Reg,{x,I}}|Acc]) end; -gen_moves([A|As], Sr, I, Acc) -> +gen_moves([A0|As], Sr, I, Acc) -> + A = atomic(A0), gen_moves(As, Sr, I+1, [{move,A,{x,I}}|Acc]); gen_moves([], _, _, Acc) -> lists:keysort(3, Acc). @@ -1969,21 +2561,21 @@ break_up_cycle1(Dst, [M|Path], LastMove) -> %% clear_dead(Sr, Until, Vdb) -> Aft. %% Remove all variables in Sr which have died AT ALL so far. -clear_dead(Sr, Until, Vdb) -> - Sr#sr{reg=clear_dead_reg(Sr, Until, Vdb), - stk=clear_dead_stk(Sr#sr.stk, Until, Vdb)}. +clear_dead(#sr{stk=Stk}=Sr0, Until, Vdb) -> + Sr = Sr0#sr{reg=clear_dead_reg(Sr0, Until, Vdb), + stk=clear_dead_stk(Stk, Until, Vdb)}, + reserve(Sr). clear_dead_reg(Sr, Until, Vdb) -> - Reg = [case R of - {_I,V} = IV -> - case vdb_find(V, Vdb) of - {V,_,L} when L > Until -> IV; - _ -> free %Remove anything else - end; - {reserved,_I,_V} = Reserved -> Reserved; - free -> free - end || R <- Sr#sr.reg], - reserve(Sr#sr.res, Reg, Sr#sr.stk). + [case R of + {_I,V} = IV -> + case vdb_find(V, Vdb) of + {V,_,L} when L > Until -> IV; + _ -> free %Remove anything else + end; + {reserved,_I,_V}=Reserved -> Reserved; + free -> free + end || R <- Sr#sr.reg]. clear_dead_stk(Stk, Until, Vdb) -> [case S of @@ -2055,16 +2647,25 @@ adjust_stack(Bef, Fb, Lf, Vdb) -> save_stack(Stk0, Fb, Lf, Vdb) -> %% New variables that are in use but not on stack. - New = [VFL || {V,F,L} = VFL <- Vdb, - F < Fb, - L >= Lf, - not on_stack(V, Stk0)], + New = new_not_on_stack(Stk0, Fb, Lf, Vdb), + %% Add new variables that are not just dropped immediately. %% N.B. foldr works backwards from the end!! Saves = [V || {V,_,_} <- keysort(3, New)], Stk1 = foldr(fun (V, Stk) -> put_stack(V, Stk) end, Stk0, Saves), {Stk1,Saves}. +%% new_not_on_stack(Stack, FirstBefore, LastFrom, Vdb) -> +%% [{Variable,First,Last}] +%% Return information about all variables that are used past current +%% point and that are not already on the stack. + +new_not_on_stack(Stk, Fb, Lf, Vdb) -> + [VFL || {V,F,L} = VFL <- Vdb, + F < Fb, + L >= Lf, + not on_stack(V, Stk)]. + %% saves([SaveVar], Reg, Stk) -> [{move,Reg,Stk}]. %% Generate move instructions to save variables onto stack. The %% stack/reg info used is that after the new stack has been made. @@ -2083,8 +2684,14 @@ fetch_var(V, Sr) -> error -> fetch_stack(V, Sr#sr.stk) end. +fetch_var_prefer_y(V, #sr{reg=Reg,stk=Stk}) -> + case find_stack(V, Stk) of + {ok,R} -> R; + error -> fetch_reg(V, Reg) + end. + load_vars(Vs, Regs) -> - foldl(fun ({var,V}, Rs) -> put_reg(V, Rs) end, Regs, Vs). + foldl(fun (#k_var{name=V}, Rs) -> put_reg(V, Rs) end, Regs, Vs). %% put_reg(Val, Regs) -> Regs. %% find_reg(Val, Regs) -> {ok,r{R}} | error. @@ -2156,18 +2763,14 @@ fetch_stack(Var, Stk) -> fetch_stack(Var, Stk, 0). fetch_stack(V, [{V}|_], I) -> {yy,I}; fetch_stack(V, [_|Stk], I) -> fetch_stack(V, Stk, I+1). -% find_stack(Var, Stk) -> find_stack(Var, Stk, 0). +find_stack(Var, Stk) -> find_stack(Var, Stk, 0). -% find_stack(V, [{V}|Stk], I) -> {ok,{yy,I}}; -% find_stack(V, [O|Stk], I) -> find_stack(V, Stk, I+1); -% find_stack(V, [], I) -> error. +find_stack(V, [{V}|_], I) -> {ok,{yy,I}}; +find_stack(V, [_|Stk], I) -> find_stack(V, Stk, I+1); +find_stack(_, [], _) -> error. on_stack(V, Stk) -> keymember(V, 1, Stk). -is_register({x,_}) -> true; -is_register({yy,_}) -> true; -is_register(_) -> false. - %% put_catch(CatchTag, Stack) -> Stack' %% drop_catch(CatchTag, Stack) -> Stack' %% Special interface for putting and removing catch tags, to ensure that @@ -2185,6 +2788,16 @@ put_catch(Tag, [Other|Stk], Acc) -> drop_catch(Tag, [{{catch_tag,Tag}}|Stk]) -> [free|Stk]; drop_catch(Tag, [Other|Stk]) -> [Other|drop_catch(Tag, Stk)]. +%% atomic(Klit) -> Lit. +%% atomic_list([Klit]) -> [Lit]. + +atomic(#k_literal{val=V}) -> {literal,V}; +atomic(#k_int{val=I}) -> {integer,I}; +atomic(#k_float{val=F}) -> {float,F}; +atomic(#k_atom{val=A}) -> {atom,A}; +%%atomic(#k_char{val=C}) -> {char,C}; +atomic(#k_nil{}) -> nil. + %% new_label(St) -> {L,St}. new_label(#cg{lcount=Next}=St) -> @@ -2227,3 +2840,86 @@ flatmapfoldl(F, Accu0, [Hd|Tail]) -> {Rs,Accu2} = flatmapfoldl(F, Accu1, Tail), {R++Rs,Accu2}; flatmapfoldl(_, Accu, []) -> {[],Accu}. + +%% Keep track of life time for variables. +%% +%% init_vars([{var,VarName}]) -> Vdb. +%% new_vars([VarName], I, Vdb) -> Vdb. +%% use_vars([VarName], I, Vdb) -> Vdb. +%% add_var(VarName, F, L, Vdb) -> Vdb. +%% +%% The list of variable names for new_vars/3 and use_vars/3 +%% must be sorted. + +init_vars(Vs) -> + vdb_new(Vs). + +new_vars([], _, Vdb) -> Vdb; +new_vars([V], I, Vdb) -> vdb_store_new(V, {V,I,I}, Vdb); +new_vars(Vs, I, Vdb) -> vdb_update_vars(Vs, Vdb, I). + +use_vars([], _, Vdb) -> + Vdb; +use_vars([V], I, Vdb) -> + case vdb_find(V, Vdb) of + {V,F,L} when I > L -> vdb_update(V, {V,F,I}, Vdb); + {V,_,_} -> Vdb; + error -> vdb_store_new(V, {V,I,I}, Vdb) + end; +use_vars(Vs, I, Vdb) -> vdb_update_vars(Vs, Vdb, I). + +add_var(V, F, L, Vdb) -> + vdb_store_new(V, {V,F,L}, Vdb). + +%% vdb + +vdb_new(Vs) -> + ordsets:from_list([{V,0,0} || #k_var{name=V} <- Vs]). + +-type var() :: atom(). + +-spec vdb_find(var(), [vdb_entry()]) -> 'error' | vdb_entry(). + +vdb_find(V, Vdb) -> + case lists:keyfind(V, 1, Vdb) of + false -> error; + Vd -> Vd + end. + +vdb_update(V, Update, [{V,_,_}|Vdb]) -> + [Update|Vdb]; +vdb_update(V, Update, [Vd|Vdb]) -> + [Vd|vdb_update(V, Update, Vdb)]. + +vdb_store_new(V, New, [{V1,_,_}=Vd|Vdb]) when V > V1 -> + [Vd|vdb_store_new(V, New, Vdb)]; +vdb_store_new(V, New, [{V1,_,_}|_]=Vdb) when V < V1 -> + [New|Vdb]; +vdb_store_new(_, New, []) -> [New]. + +vdb_update_vars([V|_]=Vs, [{V1,_,_}=Vd|Vdb], I) when V > V1 -> + [Vd|vdb_update_vars(Vs, Vdb, I)]; +vdb_update_vars([V|Vs], [{V1,_,_}|_]=Vdb, I) when V < V1 -> + %% New variable. + [{V,I,I}|vdb_update_vars(Vs, Vdb, I)]; +vdb_update_vars([V|Vs], [{_,F,L}=Vd|Vdb], I) -> + %% Existing variable. + if + I > L -> [{V,F,I}|vdb_update_vars(Vs, Vdb, I)]; + true -> [Vd|vdb_update_vars(Vs, Vdb, I)] + end; +vdb_update_vars([V|Vs], [], I) -> + %% New variable. + [{V,I,I}|vdb_update_vars(Vs, [], I)]; +vdb_update_vars([], Vdb, _) -> Vdb. + +%% vdb_sub(Min, Max, Vdb) -> Vdb. +%% Extract variables which are used before and after Min. Lock +%% variables alive after Max. + +vdb_sub(Min, Max, Vdb) -> + [ if L >= Max -> {V,F,locked}; + true -> Vd + end || {V,F,L}=Vd <- Vdb, + F < Min, + L >= Min ]. diff --git a/lib/compiler/src/v3_core.erl b/lib/compiler/src/v3_core.erl index 7d93e2ae16..c9517c3e51 100644 --- a/lib/compiler/src/v3_core.erl +++ b/lib/compiler/src/v3_core.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2015. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -137,11 +137,13 @@ -record(core, {vcount=0 :: non_neg_integer(), %Variable counter fcount=0 :: non_neg_integer(), %Function counter + function={none,0} :: fa(), %Current function. in_guard=false :: boolean(), %In guard or not. wanted=true :: boolean(), %Result wanted or not. opts :: [compile:option()], %Options. ws=[] :: [warning()], %Warnings. - file=[{file,""}]}). %File + file=[{file,""}] %File. + }). %% XXX: The following type declarations do not belong in this module -type fa() :: {atom(), arity()}. @@ -149,38 +151,73 @@ -type form() :: {function, integer(), atom(), arity(), _} | {attribute, integer(), attribute(), _}. --spec module({module(), [fa()], [form()]}, [compile:option()]) -> +-record(imodule, {name = [], + exports = ordsets:new(), + attrs = [], + defs = [], + file = [], + opts = [], + ws = []}). + +-spec module([form()], [compile:option()]) -> {'ok',cerl:c_module(),[warning()]}. -module({Mod,Exp,Forms}, Opts) -> - Cexp = map(fun ({_N,_A} = NA) -> #c_var{name=NA} end, Exp), - {Kfs0,As0,Ws,_File} = foldl(fun (F, Acc) -> - form(F, Acc, Opts) - end, {[],[],[],[]}, Forms), - Kfs = reverse(Kfs0), +module(Forms0, Opts) -> + Forms = erl_internal:add_predefined_functions(Forms0), + Module = foldl(fun (F, Acc) -> + form(F, Acc, Opts) + end, #imodule{}, Forms), + #imodule{name=Mod,exports=Exp0,attrs=As0,defs=Kfs0,ws=Ws} = Module, + Exp = case member(export_all, Opts) of + true -> defined_functions(Forms); + false -> Exp0 + end, + Cexp = [#c_var{name=FA} || {_,_}=FA <- Exp], As = reverse(As0), + Kfs = reverse(Kfs0), {ok,#c_module{name=#c_literal{val=Mod},exports=Cexp,attrs=As,defs=Kfs},Ws}. -form({function,_,_,_,_}=F0, {Fs,As,Ws0,File}, Opts) -> +form({function,_,_,_,_}=F0, Module, Opts) -> + #imodule{file=File,defs=Defs,ws=Ws0} = Module, {F,Ws} = function(F0, Ws0, File, Opts), - {[F|Fs],As,Ws,File}; -form({attribute,_,file,{File,_Line}}, {Fs,As,Ws,_}, _Opts) -> - {Fs,As,Ws,File}; -form({attribute,_,_,_}=F, {Fs,As,Ws,File}, _Opts) -> - {Fs,[attribute(F)|As],Ws,File}. - -attribute(Attribute) -> - Fun = fun(A) -> [erl_anno:location(A)] end, - {attribute,Line,Name,Val} = erl_parse:map_anno(Fun, Attribute), + Module#imodule{defs=[F|Defs],ws=Ws}; +form({attribute,_,module,Mod}, Module, _Opts) -> + true = is_atom(Mod), + Module#imodule{name=Mod}; +form({attribute,_,file,{File,_Line}}=F, #imodule{attrs=As}=Module, _Opts) -> + Module#imodule{file=File, attrs=[attribute(F)|As]}; +form({attribute,_,import,_}, Module, _Opts) -> + %% Ignore. We have no futher use for imports. + Module; +form({attribute,_,export,Es}, #imodule{exports=Exp0}=Module, _Opts) -> + Exp = ordsets:union(ordsets:from_list(Es), Exp0), + Module#imodule{exports=Exp}; +form({attribute,_,_,_}=F, #imodule{attrs=As}=Module, _Opts) -> + Module#imodule{attrs=[attribute(F)|As]}; +form(_, Module, _Opts) -> + %% Ignore uninteresting forms such as 'eof'. + Module. + +attribute({attribute,A,Name,Val0}) -> + Line = [erl_anno:location(A)], + Val = if + is_list(Val0) -> Val0; + true -> [Val0] + end, {#c_literal{val=Name, anno=Line}, #c_literal{val=Val, anno=Line}}. +defined_functions(Forms) -> + Fs = [{Name,Arity} || {function,_,Name,Arity,_} <- Forms], + ordsets:from_list(Fs). + %% function_dump(module_info,_,_,_) -> ok; %% function_dump(Name,Arity,Format,Terms) -> %% io:format("~w/~w " ++ Format,[Name,Arity]++Terms), %% ok. function({function,_,Name,Arity,Cs0}, Ws0, File, Opts) -> - St0 = #core{vcount=0,opts=Opts,ws=Ws0,file=[{file,File}]}, + St0 = #core{vcount=0,function={Name,Arity},opts=Opts, + ws=Ws0,file=[{file,File}]}, {B0,St1} = body(Cs0, Name, Arity, St0), %% ok = function_dump(Name,Arity,"body:~n~p~n",[B0]), {B1,St2} = ubody(B0, St1), @@ -291,14 +328,16 @@ gexpr({protect,Line,Arg}, Bools0, St0) -> Anno = lineno_anno(Line, St), {#iprotect{anno=#a{anno=Anno},body=Eps++[E]},[],Bools0,St} end; -gexpr({op,L,'andalso',E1,E2}, Bools, St0) -> +gexpr({op,_,'andalso',_,_}=E0, Bools, St0) -> + {op,L,'andalso',E1,E2} = right_assoc(E0, 'andalso', St0), Anno = lineno_anno(L, St0), {#c_var{name=V0},St} = new_var(Anno, St0), V = {var,L,V0}, False = {atom,L,false}, E = make_bool_switch_guard(L, E1, V, E2, False), gexpr(E, Bools, St); -gexpr({op,L,'orelse',E1,E2}, Bools, St0) -> +gexpr({op,_,'orelse',_,_}=E0, Bools, St0) -> + {op,L,'orelse',E1,E2} = right_assoc(E0, 'orelse', St0), Anno = lineno_anno(L, St0), {#c_var{name=V0},St} = new_var(Anno, St0), V = {var,L,V0}, @@ -469,7 +508,8 @@ unforce_tree([#iset{var=#c_var{name=V},arg=Arg0}|Es], D0) -> unforce_tree(Es, D); unforce_tree([#icall{}=Call], D) -> unforce_tree_subst(Call, D); -unforce_tree([Top], _) -> Top. +unforce_tree([#c_var{name=V}], D) -> + gb_trees:get(V, D). unforce_tree_subst(#icall{module=#c_literal{val=erlang}, name=#c_literal{val='=:='}, @@ -631,9 +671,11 @@ expr({'catch',L,E0}, St0) -> {E1,Eps,St1} = expr(E0, St0), Lanno = lineno_anno(L, St1), {#icatch{anno=#a{anno=Lanno},body=Eps ++ [E1]},[],St1}; -expr({'fun',L,{function,F,A},{_,_,_}=Id}, St) -> - Lanno = full_anno(L, St), - {#c_var{anno=Lanno++[{id,Id}],name={F,A}},[],St}; +expr({'fun',L,{function,F,A}}, St0) -> + {Fname,St1} = new_fun_name(St0), + Lanno = full_anno(L, St1), + Id = {0,0,Fname}, + {#c_var{anno=Lanno++[{id,Id}],name={F,A}},[],St1}; expr({'fun',L,{function,M,F,A}}, St0) -> {As,Aps,St1} = safe_list([M,F,A], St0), Lanno = full_anno(L, St1), @@ -641,12 +683,12 @@ expr({'fun',L,{function,M,F,A}}, St0) -> module=#c_literal{val=erlang}, name=#c_literal{val=make_fun}, args=As},Aps,St1}; -expr({'fun',L,{clauses,Cs},Id}, St) -> - fun_tq(Id, Cs, L, St, unnamed); -expr({named_fun,L,'_',Cs,Id}, St) -> - fun_tq(Id, Cs, L, St, unnamed); -expr({named_fun,L,Name,Cs,Id}, St) -> - fun_tq(Id, Cs, L, St, {named,Name}); +expr({'fun',L,{clauses,Cs}}, St) -> + fun_tq(Cs, L, St, unnamed); +expr({named_fun,L,'_',Cs}, St) -> + fun_tq(Cs, L, St, unnamed); +expr({named_fun,L,Name,Cs}, St) -> + fun_tq(Cs, L, St, {named,Name}); expr({call,L,{remote,_,M,F},As0}, St0) -> {[M1,F1|As1],Aps,St1} = safe_list([M,F|As0], St0), Anno = full_anno(L, St1), @@ -680,9 +722,36 @@ expr({match,L,P0,E0}, St0) -> Fc = fail_clause([Fpat], Lanno, c_tuple([#c_literal{val=badmatch},Fpat])), case P2 of nomatch -> - St = add_warning(L, nomatch, St5), - {#icase{anno=#a{anno=Lanno}, - args=[E2],clauses=[],fc=Fc},Eps1++Eps2,St}; + %% The pattern will not match. We must take care here to + %% bind all variables that the pattern would have bound + %% so that subsequent expressions do not refer to unbound + %% variables. + %% + %% As an example, this code: + %% + %% [X] = {Y} = E, + %% X + Y. + %% + %% will be rewritten to: + %% + %% error({badmatch,E}), + %% case E of + %% {[X],{Y}} -> + %% X + Y; + %% Other -> + %% error({badmatch,Other}) + %% end. + %% + St6 = add_warning(L, nomatch, St5), + {Expr,Eps3,St7} = safe(E1, St6), + SanPat0 = sanitize(P1), + {SanPat,Eps4,St} = pattern(SanPat0, St7), + Badmatch = c_tuple([#c_literal{val=badmatch},Expr]), + Fail = #iprimop{anno=#a{anno=Lanno}, + name=#c_literal{val=match_fail}, + args=[Badmatch]}, + Eps = Eps3 ++ Eps4 ++ [Fail], + {#imatch{anno=#a{anno=Lanno},pat=SanPat,arg=Expr,fc=Fc},Eps,St}; Other when not is_atom(Other) -> {#imatch{anno=#a{anno=Lanno},pat=P2,arg=E2,fc=Fc},Eps1++Eps2,St5} end; @@ -724,6 +793,32 @@ expr({op,L,Op,L0,R0}, St0) -> module=#c_literal{anno=LineAnno,val=erlang}, name=#c_literal{anno=LineAnno,val=Op},args=As},Aps,St1}. + +%% sanitize(Pat) -> SanitizedPattern +%% Rewrite Pat so that it will be accepted by pattern/2 and will +%% bind the same variables as the original pattern. +%% +%% Here is an example of a pattern that would cause a pattern/2 +%% to generate a 'nomatch' exception: +%% +%% #{k:=X,k:=Y} = [Z] +%% +%% The sanitized pattern will look like: +%% +%% {{X,Y},[Z]} + +sanitize({match,L,P1,P2}) -> + {tuple,L,[sanitize(P1),sanitize(P2)]}; +sanitize({cons,L,H,T}) -> + {cons,L,sanitize(H),sanitize(T)}; +sanitize({tuple,L,Ps0}) -> + Ps = [sanitize(P) || P <- Ps0], + {tuple,L,Ps}; +sanitize({map,L,Ps0}) -> + Ps = [sanitize(V) || {map_field_exact,_,_,V} <- Ps0], + {tuple,L,Ps}; +sanitize(P) -> P. + make_bool_switch(L, E, V, T, F, #core{in_guard=true}) -> make_bool_switch_guard(L, E, V, T, F); make_bool_switch(L, E, V, T, F, #core{}) -> @@ -783,22 +878,36 @@ badmap_term(_Map, #core{in_guard=true}) -> %% since it is not user-visible. #c_literal{val=badmap}; badmap_term(Map, #core{in_guard=false}) -> - #c_tuple{es=[#c_literal{val=badmap},Map]}. + c_tuple([#c_literal{val=badmap},Map]). map_build_pairs(Map, Es0, Ann, St0) -> - {Es,Pre,St1} = map_build_pairs_1(Es0, St0), + {Es,Pre,_,St1} = map_build_pairs_1(Es0, cerl_sets:new(), St0), {ann_c_map(Ann, Map, Es),Pre,St1}. -map_build_pairs_1([{Op0,L,K0,V0}|Es], St0) -> +map_build_pairs_1([{Op0,L,K0,V0}|Es], Used0, St0) -> {K,Pre0,St1} = safe(K0, St0), {V,Pre1,St2} = safe(V0, St1), - {Pairs,Pre2,St3} = map_build_pairs_1(Es, St2), + {Pairs,Pre2,Used1,St3} = map_build_pairs_1(Es, Used0, St2), As = lineno_anno(L, St3), Op = map_op(Op0), + {Used2,St4} = maybe_warn_repeated_keys(K, L, Used1, St3), Pair = cerl:ann_c_map_pair(As, Op, K, V), - {[Pair|Pairs],Pre0++Pre1++Pre2,St3}; -map_build_pairs_1([], St) -> - {[],[],St}. + {[Pair|Pairs],Pre0++Pre1++Pre2,Used2,St4}; +map_build_pairs_1([], Used, St) -> + {[],[],Used,St}. + +maybe_warn_repeated_keys(Ck,Line,Used,St) -> + case cerl:is_literal(Ck) of + false -> {Used,St}; + true -> + K = cerl:concrete(Ck), + case cerl_sets:is_element(K,Used) of + true -> + {Used, add_warning(Line, {map_key_repeated,K}, St)}; + false -> + {cerl_sets:add_element(K,Used), St} + end + end. map_op(map_field_assoc) -> #c_literal{val=assoc}; map_op(map_field_exact) -> #c_literal{val=exact}. @@ -813,13 +922,18 @@ try_exception(Ecs0, St0) -> %% Note that Tag is not needed for rethrow - it is already in Info. {Evs,St1} = new_vars(3, St0), % Tag, Value, Info {Ecs1,Ceps,St2} = clauses(Ecs0, St1), + Ecs2 = try_build_stacktrace(Ecs1, hd(Evs)), [_,Value,Info] = Evs, - Ec = #iclause{anno=#a{anno=[compiler_generated]}, + LA = case Ecs2 of + [] -> []; + [C|_] -> get_lineno_anno(C) + end, + Ec = #iclause{anno=#a{anno=[compiler_generated|LA]}, pats=[c_tuple(Evs)],guard=[#c_literal{val=true}], body=[#iprimop{anno=#a{}, %Must have an #a{} name=#c_literal{val=raise}, args=[Info,Value]}]}, - Hs = [#icase{anno=#a{},args=[c_tuple(Evs)],clauses=Ecs1,fc=Ec}], + Hs = [#icase{anno=#a{anno=LA},args=[c_tuple(Evs)],clauses=Ecs2,fc=Ec}], {Evs,Ceps++Hs,St2}. try_after(As, St0) -> @@ -835,20 +949,54 @@ try_after(As, St0) -> Hs = [#icase{anno=#a{},args=[c_tuple(Evs)],clauses=[],fc=Ec}], {Evs,Hs,St1}. +try_build_stacktrace([#iclause{pats=Ps0,body=B0}=C0|Cs], RawStk) -> + [#c_tuple{es=[Class,Exc,Stk]}=Tup] = Ps0, + case Stk of + #c_var{name='_'} -> + %% Stacktrace variable is not used. Nothing to do. + [C0|try_build_stacktrace(Cs, RawStk)]; + _ -> + %% Add code to build the stacktrace. + Ps = [Tup#c_tuple{es=[Class,Exc,RawStk]}], + Call = #iprimop{anno=#a{}, + name=#c_literal{val=build_stacktrace}, + args=[RawStk]}, + Iset = #iset{var=Stk,arg=Call}, + B = [Iset|B0], + C = C0#iclause{pats=Ps,body=B}, + [C|try_build_stacktrace(Cs, RawStk)] + end; +try_build_stacktrace([], _) -> []. + %% expr_bin([ArgExpr], St) -> {[Arg],[PreExpr],St}. %% Flatten the arguments of a bin. Do this straight left to right! %% Note that ibinary needs to have its annotation wrapped in a #a{} %% record whereas c_literal should not have a wrapped annotation expr_bin(Es0, Anno, St0) -> - case constant_bin(Es0) of + Es1 = [bin_element(E) || E <- Es0], + case constant_bin(Es1) of error -> - {Es,Eps,St} = expr_bin_1(Es0, St0), + {Es,Eps,St} = expr_bin_1(bin_expand_strings(Es1), St0), {#ibinary{anno=#a{anno=Anno},segments=Es},Eps,St}; Bin -> {#c_literal{anno=Anno,val=Bin},[],St0} end. +bin_element({bin_element,Line,Expr,Size0,Type0}) -> + {Size,Type} = make_bit_type(Line, Size0, Type0), + {bin_element,Line,Expr,Size,Type}. + +make_bit_type(Line, default, Type0) -> + case erl_bits:set_bit_type(default, Type0) of + {ok,all,Bt} -> {{atom,Line,all},erl_bits:as_list(Bt)}; + {ok,undefined,Bt} -> {{atom,Line,undefined},erl_bits:as_list(Bt)}; + {ok,Size,Bt} -> {{integer,Line,Size},erl_bits:as_list(Bt)} + end; +make_bit_type(_Line, Size, Type0) -> %Integer or 'all' + {ok,Size,Bt} = erl_bits:set_bit_type(Size, Type0), + {Size,erl_bits:as_list(Bt)}. + %% constant_bin([{bin_element,_,_,_,_}]) -> binary() | error %% If the binary construction is truly constant (no variables, %% no native fields), and does not contain fields whose expansion @@ -865,7 +1013,8 @@ constant_bin(Es) -> constant_bin_1(Es) -> verify_suitable_fields(Es), EmptyBindings = erl_eval:new_bindings(), - EvalFun = fun({integer,_,I}, B) -> {value,I,B}; + EvalFun = fun({string,_,S}, B) -> {value,S,B}; + ({integer,_,I}, B) -> {value,I,B}; ({char,_,C}, B) -> {value,C,B}; ({float,_,F}, B) -> {value,F,B}; ({atom,_,undefined}, B) -> {value,undefined,B} @@ -886,6 +1035,9 @@ verify_suitable_fields([{bin_element,_,Val,SzTerm,Opts}|Es]) -> end, {unit,Unit} = keyfind(unit, 1, Opts), case {SzTerm,Val} of + {{atom,_,undefined},{string,_,_}} -> + %% UTF-8/16/32. + ok; {{atom,_,undefined},{char,_,_}} -> %% UTF-8/16/32. ok; @@ -925,6 +1077,31 @@ count_bits(Int) -> count_bits_1(0, Bits) -> Bits; count_bits_1(Int, Bits) -> count_bits_1(Int bsr 64, Bits+64). +bin_expand_strings(Es0) -> + foldr(fun ({bin_element,Line,{string,_,S},{integer,_,8},_}, Es) -> + bin_expand_string(S, Line, 0, 0) ++ Es; + ({bin_element,Line,{string,_,S},Sz,Ts}, Es1) -> + foldr( + fun (C, Es) -> + [{bin_element,Line,{char,Line,C},Sz,Ts}|Es] + end, Es1, S); + (E, Es) -> + [E|Es] + end, [], Es0). + +bin_expand_string(S, Line, Val, Size) when Size >= 2048 -> + Combined = make_combined(Line, Val, Size), + [Combined|bin_expand_string(S, Line, 0, 0)]; +bin_expand_string([H|T], Line, Val, Size) -> + bin_expand_string(T, Line, (Val bsl 8) bor H, Size+8); +bin_expand_string([], Line, Val, Size) -> + [make_combined(Line, Val, Size)]. + +make_combined(Line, Val, Size) -> + {bin_element,Line,{integer,Line,Val}, + {integer,Line,Size}, + [integer,{unit,1},unsigned,big]}. + expr_bin_1(Es, St) -> foldr(fun (E, {Ces,Esp,St0}) -> {Ce,Ep,St1} = bitstr(E, St0), @@ -960,22 +1137,24 @@ bitstr({bin_element,_,E0,Size0,[Type,{unit,Unit}|Flags]}, St0) -> %% fun_tq(Id, [Clauses], Line, State, NameInfo) -> {Fun,[PreExp],State}. -fun_tq({_,_,Name}=Id, Cs0, L, St0, NameInfo) -> +fun_tq(Cs0, L, St0, NameInfo) -> Arity = clause_arity(hd(Cs0)), {Cs1,Ceps,St1} = clauses(Cs0, St0), {Args,St2} = new_vars(Arity, St1), {Ps,St3} = new_vars(Arity, St2), %Need new variables here Anno = full_anno(L, St3), + {Name,St4} = new_fun_name(St3), Fc = function_clause(Ps, Anno, {Name,Arity}), + Id = {0,0,Name}, Fun = #ifun{anno=#a{anno=Anno}, id=[{id,Id}], %We KNOW! vars=Args,clauses=Cs1,fc=Fc,name=NameInfo}, - {Fun,Ceps,St3}. + {Fun,Ceps,St4}. %% lc_tq(Line, Exp, [Qualifier], Mc, State) -> {LetRec,[PreExp],State}. %% This TQ from Simon PJ pp 127-138. -lc_tq(Line, E, [#igen{anno=GAnno,ceps=Ceps, +lc_tq(Line, E, [#igen{anno=#a{anno=GA}=GAnno,ceps=Ceps, acc_pat=AccPat,acc_guard=AccGuard, skip_pat=SkipPat,tail=Tail,tail_pat=TailPat, arg={Pre,Arg}}|Qs], Mc, St0) -> @@ -985,7 +1164,7 @@ lc_tq(Line, E, [#igen{anno=GAnno,ceps=Ceps, F = #c_var{anno=LA,name={Name,1}}, Nc = #iapply{anno=GAnno,op=F,args=[Tail]}, {Var,St2} = new_var(St1), - Fc = function_clause([Var], LA, {Name,1}), + Fc = function_clause([Var], GA, {Name,1}), TailClause = #iclause{anno=LAnno,pats=[TailPat],guard=[],body=[Mc]}, Cs0 = case {AccPat,AccGuard} of {SkipPat,[]} -> @@ -1008,9 +1187,9 @@ lc_tq(Line, E, [#igen{anno=GAnno,ceps=Ceps, body=Lps ++ [Lc]}|Cs0], St3} end, - Fun = #ifun{anno=LAnno,id=[],vars=[Var],clauses=Cs,fc=Fc}, - {#iletrec{anno=LAnno#a{anno=[list_comprehension|LA]},defs=[{{Name,1},Fun}], - body=Pre ++ [#iapply{anno=LAnno,op=F,args=[Arg]}]}, + Fun = #ifun{anno=GAnno,id=[],vars=[Var],clauses=Cs,fc=Fc}, + {#iletrec{anno=GAnno#a{anno=[list_comprehension|GA]},defs=[{{Name,1},Fun}], + body=Pre ++ [#iapply{anno=GAnno,op=F,args=[Arg]}]}, Ceps,St4}; lc_tq(Line, E, [#ifilter{}=Filter|Qs], Mc, St) -> filter_tq(Line, E, Filter, Mc, St, Qs, fun lc_tq/5); @@ -1079,13 +1258,39 @@ bc_tq1(Line, E, [#igen{anno=GAnno,ceps=Ceps, bc_tq1(Line, E, [#ifilter{}=Filter|Qs], Mc, St) -> filter_tq(Line, E, Filter, Mc, St, Qs, fun bc_tq1/5); bc_tq1(_, {bin,Bl,Elements}, [], AccVar, St0) -> - {E,Pre,St} = expr({bin,Bl,[{bin_element,Bl, - {var,Bl,AccVar#c_var.name}, - {atom,Bl,all}, - [binary,{unit,1}]}|Elements]}, St0), + bc_tq_build(Bl, [], AccVar, Elements, St0); +bc_tq1(Line, E0, [], AccVar, St0) -> + BsFlags = [binary,{unit,1}], + BsSize = {atom,Line,all}, + {E1,Pre0,St1} = safe(E0, St0), + case E1 of + #c_var{name=VarName} -> + Var = {var,Line,VarName}, + Els = [{bin_element,Line,Var,BsSize,BsFlags}], + bc_tq_build(Line, Pre0, AccVar, Els, St1); + #c_literal{val=Val} when is_bitstring(Val) -> + Bits = bit_size(Val), + <<Int0:Bits>> = Val, + Int = {integer,Line,Int0}, + Sz = {integer,Line,Bits}, + Els = [{bin_element,Line,Int,Sz,[integer,{unit,1},big]}], + bc_tq_build(Line, Pre0, AccVar, Els, St1); + _ -> + %% Any other safe (cons, tuple, literal) is not a + %% bitstring. Force the evaluation to fail (and + %% generate a warning). + Els = [{bin_element,Line,{atom,Line,bad_value},BsSize,BsFlags}], + bc_tq_build(Line, Pre0, AccVar, Els, St1) + end. + +bc_tq_build(Line, Pre0, #c_var{name=AccVar}, Elements0, St0) -> + Elements = [{bin_element,Line,{var,Line,AccVar},{atom,Line,all}, + [binary,{unit,1}]}|Elements0], + {E,Pre,St} = expr({bin,Line,Elements}, St0), #a{anno=A} = Anno0 = get_anno(E), Anno = Anno0#a{anno=[compiler_generated,single_use|A]}, - {set_anno(E, Anno),Pre,St}. + {set_anno(E, Anno),Pre0++Pre,St}. + %% filter_tq(Line, Expr, Filter, Mc, State, [Qualifier], TqFun) -> %% {Case,[PreExpr],State}. @@ -1270,8 +1475,9 @@ list_gen_pattern(P0, Line, St) -> %%% the result binary in a binary comprehension. %%% -bc_initial_size(E, Q, St0) -> +bc_initial_size(E0, Q, St0) -> try + E = bin_bin_element(E0), {ElemSzExpr,ElemSzPre,EVs,St1} = bc_elem_size(E, St0), {V,St2} = new_var(St1), {GenSzExpr,GenSzPre,St3} = bc_gen_size(Q, EVs, St2), @@ -1297,7 +1503,7 @@ bc_initial_size(E, Q, St0) -> end. bc_elem_size({bin,_,El}, St0) -> - case bc_elem_size_1(El, 0, []) of + case bc_elem_size_1(El, ordsets:new(), 0, []) of {Bits,[]} -> {#c_literal{val=Bits},[],[],St0}; {Bits,Vars0} -> @@ -1306,18 +1512,38 @@ bc_elem_size({bin,_,El}, St0) -> Vs = [V || {_,#c_var{name=V}} <- Vars0], {E,Pre,St} = bc_mul_pairs(F, #c_literal{val=Bits}, [], St0), {E,Pre,Vs,St} - end. + end; +bc_elem_size(_, _) -> + throw(impossible). -bc_elem_size_1([{bin_element,_,_,{integer,_,N},Flags}|Es], Bits, Vars) -> - {unit,U} = keyfind(unit, 1, Flags), - bc_elem_size_1(Es, Bits+U*N, Vars); -bc_elem_size_1([{bin_element,_,_,{var,_,Var},Flags}|Es], Bits, Vars) -> - {unit,U} = keyfind(unit, 1, Flags), - bc_elem_size_1(Es, Bits, [{U,#c_var{name=Var}}|Vars]); -bc_elem_size_1([_|_], _, _) -> +bc_elem_size_1([{bin_element,_,{string,_,String},{integer,_,N},_}=El|Es], + DefVars, Bits, SizeVars) -> + U = get_unit(El), + bc_elem_size_1(Es, DefVars, Bits+U*N*length(String), SizeVars); +bc_elem_size_1([{bin_element,_,Expr,{integer,_,N},_}=El|Es], + DefVars0, Bits, SizeVars) -> + U = get_unit(El), + DefVars = bc_elem_size_def_var(Expr, DefVars0), + bc_elem_size_1(Es, DefVars, Bits+U*N, SizeVars); +bc_elem_size_1([{bin_element,_,Expr,{var,_,Src},_}=El|Es], + DefVars0, Bits, SizeVars) -> + case ordsets:is_element(Src, DefVars0) of + false -> + U = get_unit(El), + DefVars = bc_elem_size_def_var(Expr, DefVars0), + bc_elem_size_1(Es, DefVars, Bits, [{U,#c_var{name=Src}}|SizeVars]); + true -> + throw(impossible) + end; +bc_elem_size_1([_|_], _, _, _) -> throw(impossible); -bc_elem_size_1([], Bits, Vars) -> - {Bits,Vars}. +bc_elem_size_1([], _DefVars, Bits, SizeVars) -> + {Bits,SizeVars}. + +bc_elem_size_def_var({var,_,Var}, DefVars) -> + ordsets:add_element(Var, DefVars); +bc_elem_size_def_var(_Expr, DefVars) -> + DefVars. bc_elem_size_combine([{U,V}|T], U, UVars, Acc) -> bc_elem_size_combine(T, U, [V|UVars], Acc); @@ -1369,7 +1595,9 @@ bc_gen_size_1([{generate,L,El,Gen}|Qs], EVs, E0, Pre0, St0) -> {E,Pre,St} = bc_gen_size_mul(E0, #c_literal{val=Len}, Pre0, St0), bc_gen_size_1(Qs, EVs, E, Pre, St) end; -bc_gen_size_1([{b_generate,_,El,Gen}|Qs], EVs, E0, Pre0, St0) -> +bc_gen_size_1([{b_generate,_,El0,Gen0}|Qs], EVs, E0, Pre0, St0) -> + El = bin_bin_element(El0), + Gen = bin_bin_element(Gen0), bc_verify_non_filtering(El, EVs), {MatchSzExpr,Pre1,_,St1} = bc_elem_size(El, St0), Pre2 = reverse(Pre1, Pre0), @@ -1385,6 +1613,10 @@ bc_gen_size_1([], _, E, Pre, St) -> bc_gen_size_1(_, _, _, _, _) -> throw(impossible). +bin_bin_element({bin,L,El}) -> + {bin,L,[bin_element(E) || E <- El]}; +bin_bin_element(Other) -> Other. + bc_gen_bit_size({var,L,V}, Pre0, St0) -> Lanno = lineno_anno(L, St0), {SzVar,St} = new_var(St0), @@ -1427,8 +1659,11 @@ bc_list_length(_, _) -> bc_bin_size({bin,_,Els}) -> bc_bin_size_1(Els, 0). -bc_bin_size_1([{bin_element,_,_,{integer,_,Sz},Flags}|Els], N) -> - {unit,U} = keyfind(unit, 1, Flags), +bc_bin_size_1([{bin_element,_,{string,_,String},{integer,_,Sz},_}=El|Els], N) -> + U = get_unit(El), + bc_bin_size_1(Els, N+U*Sz*length(String)); +bc_bin_size_1([{bin_element,_,_,{integer,_,Sz},_}=El|Els], N) -> + U = get_unit(El), bc_bin_size_1(Els, N+U*Sz); bc_bin_size_1([], N) -> N; bc_bin_size_1(_, _) -> throw(impossible). @@ -1463,11 +1698,24 @@ bc_bsr(E1, E2) -> name=#c_literal{val='bsr'}, args=[E1,E2]}. -%% is_guard_test(Expression) -> true | false. -%% Test if a general expression is a guard test. Use erl_lint here -%% as it now allows sys_pre_expand transformed source. +get_unit({bin_element,_,_,_,Flags}) -> + {unit,U} = keyfind(unit, 1, Flags), + U. -is_guard_test(E) -> erl_lint:is_guard_test(E). +%% is_guard_test(Expression) -> true | false. +%% Test if a general expression is a guard test. +%% +%% Note that a local function overrides a BIF with the same name. +%% For example, if there is a local function named is_list/1, +%% any unqualified call to is_list/1 will be to the local function. +%% The guard function must be explicitly called as erlang:is_list/1. + +is_guard_test(E) -> + %% erl_expand_records has added a module prefix to any call + %% to a BIF or imported function. Any call without a module + %% prefix that remains must therefore be to a local function. + IsOverridden = fun({_,_}) -> true end, + erl_lint:is_guard_test(E, [], IsOverridden). %% novars(Expr, State) -> {Novars,[PreExpr],State}. %% Generate a novars expression, basically a call or a safe. At this @@ -1610,7 +1858,18 @@ pattern({bin,L,Ps}, St) -> pattern({match,_,P1,P2}, St) -> {Cp1,Eps1,St1} = pattern(P1,St), {Cp2,Eps2,St2} = pattern(P2,St1), - {pat_alias(Cp1,Cp2),Eps1++Eps2,St2}. + {pat_alias(Cp1,Cp2),Eps1++Eps2,St2}; +%% Evaluate compile-time expressions. +pattern({op,_,'++',{nil,_},R}, St) -> + pattern(R, St); +pattern({op,_,'++',{cons,Li,H,T},R}, St) -> + pattern({cons,Li,H,{op,Li,'++',T,R}}, St); +pattern({op,_,'++',{string,Li,L},R}, St) -> + pattern(string_to_conses(Li, L, R), St); +pattern({op,_Line,_Op,_A}=Op, St) -> + pattern(erl_eval:partial_eval(Op), St); +pattern({op,_Line,_Op,_L,_R}=Op, St) -> + pattern(erl_eval:partial_eval(Op), St). %% pattern_map_pairs([MapFieldExact],State) -> [#c_map_pairs{}] pattern_map_pairs(Ps, St) -> @@ -1650,16 +1909,29 @@ pat_alias_map_pairs_1([]) -> []. %% pat_bin([BinElement], State) -> [BinSeg]. -pat_bin(Ps, St) -> [pat_segment(P, St) || P <- Ps]. - -pat_segment({bin_element,_,Val,Size,[Type,{unit,Unit}|Flags]}, St) -> - {Pval,[],St1} = pattern(Val,St), - {Psize,[],_St2} = pattern(Size,St1), - #c_bitstr{val=Pval,size=Psize, +pat_bin(Ps, St) -> [pat_segment(P, St) || P <- bin_expand_strings(Ps)]. + +pat_segment({bin_element,L,Val,Size0,Type0}, St) -> + {Size,Type1} = make_bit_type(L, Size0, Type0), + [Type,{unit,Unit}|Flags] = Type1, + Anno = lineno_anno(L, St), + {Pval0,[],St1} = pattern(Val, St), + Pval = coerce_to_float(Pval0, Type0), + {Psize,[],_St2} = pattern(Size, St1), + #c_bitstr{anno=Anno, + val=Pval,size=Psize, unit=#c_literal{val=Unit}, type=#c_literal{val=Type}, flags=#c_literal{val=Flags}}. +coerce_to_float(#c_literal{val=Int}=E, [float|_]) when is_integer(Int) -> + try + E#c_literal{val=float(Int)} + catch + error:badarg -> E + end; +coerce_to_float(E, _) -> E. + %% pat_alias(CorePat, CorePat) -> AliasPat. %% Normalise aliases. Trap bad aliases by throwing 'nomatch'. @@ -1729,11 +2001,18 @@ pattern_list([P0|Ps0], St0) -> pattern_list([], St) -> {[],[],St}. +string_to_conses(Line, Cs, Tail) -> + foldr(fun (C, T) -> {cons,Line,{char,Line,C},T} end, Tail, Cs). %% make_vars([Name]) -> [{Var,Name}]. make_vars(Vs) -> [ #c_var{name=V} || V <- Vs ]. +new_fun_name(#core{function={F,A},fcount=I}=St) -> + Name = "-" ++ atom_to_list(F) ++ "/" ++ integer_to_list(A) + ++ "-fun-" ++ integer_to_list(I) ++ "-", + {list_to_atom(Name),St#core{fcount=I+1}}. + %% new_fun_name(Type, State) -> {FunName,State}. new_fun_name(Type, #core{fcount=C}=St) -> @@ -1742,7 +2021,7 @@ new_fun_name(Type, #core{fcount=C}=St) -> %% new_var_name(State) -> {VarName,State}. new_var_name(#core{vcount=C}=St) -> - {list_to_atom("cor" ++ integer_to_list(C)),St#core{vcount=C + 1}}. + {C,St#core{vcount=C + 1}}. %% new_var(State) -> {{var,Name},State}. %% new_var(LineAnno, State) -> {{var,Name},State}. @@ -1777,6 +2056,19 @@ fail_clause(Pats, Anno, Arg) -> body=[#iprimop{anno=#a{anno=Anno},name=#c_literal{val=match_fail}, args=[Arg]}]}. +%% Optimization for Dialyzer. +right_assoc(E, Op, St) -> + case member(dialyzer, St#core.opts) of + true -> + right_assoc2(E, Op); + false -> + E + end. + +right_assoc2({op,L1,Op,{op,L2,Op,E1,E2},E3}, Op) -> + right_assoc2({op,L2,Op,E1,{op,L1,Op,E2,E3}}, Op); +right_assoc2(E, _Op) -> E. + annotate_tuple(A, Es, St) -> case member(dialyzer, St#core.opts) of true -> @@ -1852,27 +2144,22 @@ uguard(Pg, Gs0, Ks, St0) -> %% uexprs([Kexpr], [KnownVar], State) -> {[Kexpr],State}. uexprs([#imatch{anno=A,pat=P0,arg=Arg,fc=Fc}|Les], Ks, St0) -> - %% Optimise for simple set of unbound variable. - case upattern(P0, Ks, St0) of - {#c_var{},[],_Pvs,_Pus,_} -> - %% Throw our work away and just set to iset. + case upat_is_new_var(P0, Ks) of + true -> + %% Assignment to a new variable. uexprs([#iset{var=P0,arg=Arg}|Les], Ks, St0); - _Other -> - %% Throw our work away and set to icase. - if - Les =:= [] -> - %% Need to explicitly return match "value", make - %% safe for efficiency. - {La0,Lps,St1} = force_safe(Arg, St0), - La = mark_compiler_generated(La0), - Mc = #iclause{anno=A,pats=[P0],guard=[],body=[La]}, - uexprs(Lps ++ [#icase{anno=A, - args=[La0],clauses=[Mc],fc=Fc}], Ks, St1); - true -> - Mc = #iclause{anno=A,pats=[P0],guard=[],body=Les}, - uexprs([#icase{anno=A,args=[Arg], - clauses=[Mc],fc=Fc}], Ks, St0) - end + false when Les =:= [] -> + %% Need to explicitly return match "value", make + %% safe for efficiency. + {La0,Lps,St1} = force_safe(Arg, St0), + La = mark_compiler_generated(La0), + Mc = #iclause{anno=A,pats=[P0],guard=[],body=[La]}, + uexprs(Lps ++ [#icase{anno=A, + args=[La0],clauses=[Mc],fc=Fc}], Ks, St1); + false -> + Mc = #iclause{anno=A,pats=[P0],guard=[],body=Les}, + uexprs([#icase{anno=A,args=[Arg], + clauses=[Mc],fc=Fc}], Ks, St0) end; uexprs([Le0|Les0], Ks, St0) -> {Le1,St1} = uexpr(Le0, Ks, St0), @@ -1880,6 +2167,15 @@ uexprs([Le0|Les0], Ks, St0) -> {[Le1|Les1],St2}; uexprs([], _, St) -> {[],St}. +%% upat_is_new_var(Pattern, [KnownVar]) -> true|false. +%% Test whether the pattern is a single, previously unknown +%% variable. + +upat_is_new_var(#c_var{name=V}, Ks) -> + not is_element(V, Ks); +upat_is_new_var(_, _) -> + false. + %% Mark a "safe" as compiler-generated. mark_compiler_generated(#c_cons{anno=A,hd=H,tl=T}) -> ann_c_cons([compiler_generated|A], mark_compiler_generated(H), @@ -2010,7 +2306,8 @@ upattern(#c_var{name=V}=Var, Ks, St0) -> true -> {N,St1} = new_var_name(St0), New = #c_var{name=N}, - Test = #icall{anno=#a{us=add_element(N, [V])}, + LA = get_lineno_anno(Var), + Test = #icall{anno=#a{anno=LA,us=add_element(N, [V])}, module=#c_literal{val=erlang}, name=#c_literal{val='=:='}, args=[New,Var]}, @@ -2214,9 +2511,11 @@ cexpr(#icase{anno=A,args=Largs,clauses=Lcs,fc=Lfc}, As, St0) -> cexpr(#ireceive1{anno=A,clauses=Lcs}, As, St0) -> Exp = intersection(A#a.ns, As), %Exports {Ccs,St1} = cclauses(Lcs, Exp, St0), + True = #c_literal{val=true}, + Action = core_lib:make_values(lists:duplicate(1+length(Exp), True)), {#c_receive{anno=A#a.anno, clauses=Ccs, - timeout=#c_literal{val=infinity},action=#c_literal{val=true}}, + timeout=#c_literal{val=infinity},action=Action}, Exp,A#a.us,St1}; cexpr(#ireceive2{anno=A,clauses=Lcs,timeout=Lto,action=Les}, As, St0) -> Exp = intersection(A#a.ns, As), %Exports @@ -2257,8 +2556,46 @@ cexpr(#ifun{anno=#a{us=Us0}=A0,name={named,Name},fc=#iclause{pats=Ps}}=Fun0, end; cexpr(#iapply{anno=A,op=Op,args=Args}, _As, St) -> {#c_apply{anno=A#a.anno,op=Op,args=Args},[],A#a.us,St}; -cexpr(#icall{anno=A,module=Mod,name=Name,args=Args}, _As, St) -> - {#c_call{anno=A#a.anno,module=Mod,name=Name,args=Args},[],A#a.us,St}; +cexpr(#icall{anno=A,module=Mod,name=Name,args=Args}, _As, St0) -> + Anno = A#a.anno, + case (not cerl:is_c_atom(Mod)) andalso member(tuple_calls, St0#core.opts) of + true -> + GenAnno = [compiler_generated|Anno], + + %% Generate the clause that matches on the tuple + {TupleVar,St1} = new_var(GenAnno, St0), + {TupleSizeVar, St2} = new_var(GenAnno, St1), + {TupleModVar, St3} = new_var(GenAnno, St2), + {TupleArgsVar, St4} = new_var(GenAnno, St3), + TryVar = cerl:c_var('Try'), + + TupleGuardExpr = + cerl:c_let([TupleSizeVar], + c_call_erl(tuple_size, [TupleVar]), + c_call_erl('>', [TupleSizeVar, cerl:c_int(0)])), + + TupleGuard = + cerl:c_try(TupleGuardExpr, [TryVar], TryVar, + [cerl:c_var('T'),cerl:c_var('R')], cerl:c_atom(false)), + + TupleApply = + cerl:c_let([TupleModVar], + c_call_erl(element, [cerl:c_int(1),TupleVar]), + cerl:c_let([TupleArgsVar], + cerl:make_list(Args ++ [TupleVar]), + c_call_erl(apply, [TupleModVar,Name,TupleArgsVar]))), + + TupleClause = cerl:ann_c_clause(GenAnno, [TupleVar], TupleGuard, TupleApply), + + %% Generate the fallback clause + {OtherVar,St5} = new_var(GenAnno, St4), + OtherApply = cerl:ann_c_call(GenAnno, OtherVar, Name, Args), + OtherClause = cerl:ann_c_clause(GenAnno, [OtherVar], OtherApply), + + {cerl:ann_c_case(GenAnno, Mod, [TupleClause,OtherClause]),[],A#a.us,St5}; + false -> + {#c_call{anno=Anno,module=Mod,name=Name,args=Args},[],A#a.us,St0} + end; cexpr(#iprimop{anno=A,name=Name,args=Args}, _As, St) -> {#c_primop{anno=A#a.anno,name=Name,args=Args},[],A#a.us,St}; cexpr(#iprotect{anno=A,body=Es}, _As, St0) -> @@ -2288,6 +2625,9 @@ cfun(#ifun{anno=A,id=Id,vars=Args,clauses=Lcs,fc=Lfc}, _As, St0) -> clauses=Ccs ++ [Cfc]}}, [],A#a.us,St2}. +c_call_erl(Fun, Args) -> + cerl:c_call(cerl:c_atom(erlang), cerl:c_atom(Fun), Args). + %% lit_vars(Literal) -> [Var]. lit_vars(Lit) -> lit_vars(Lit, []). @@ -2382,7 +2722,11 @@ format_error(nomatch) -> format_error(bad_binary) -> "binary construction will fail because of a type mismatch"; format_error(badmap) -> - "map construction will fail because of a type mismatch". + "map construction will fail because of a type mismatch"; +format_error({map_key_repeated,Key}) when is_atom(Key) -> + io_lib:format("key '~w' will be overridden in expression", [Key]); +format_error({map_key_repeated,Key}) -> + io_lib:format("key ~p will be overridden in expression", [Key]). add_warning(Anno, Term, #core{ws=Ws,file=[{file,File}]}=St) -> case erl_anno:generated(Anno) of diff --git a/lib/compiler/src/v3_kernel.erl b/lib/compiler/src/v3_kernel.erl index 7ee564683b..aef0b6cc9f 100644 --- a/lib/compiler/src/v3_kernel.erl +++ b/lib/compiler/src/v3_kernel.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2013. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -82,7 +82,8 @@ -export([module/2,format_error/1]). -import(lists, [map/2,foldl/3,foldr/3,mapfoldl/3,splitwith/2,member/2, - keymember/3,keyfind/3,partition/2,droplast/1,last/1]). + keymember/3,keyfind/3,partition/2,droplast/1,last/1,sort/1, + reverse/1]). -import(ordsets, [add_element/2,del_element/2,union/2,union/1,subtract/2]). -import(cerl, [c_tuple/1]). @@ -107,6 +108,7 @@ copy_anno(Kdst, Ksrc) -> -record(iclause, {anno=[],isub,osub,pats,guard,body}). -record(ireceive_accept, {anno=[],arg}). -record(ireceive_next, {anno=[],arg}). +-record(ignored, {anno=[]}). -type warning() :: term(). % XXX: REFINE @@ -117,7 +119,7 @@ copy_anno(Kdst, Ksrc) -> fcount=0, %Fun counter ds=cerl_sets:new() :: cerl_sets:set(), %Defined variables funs=[], %Fun functions - free=[], %Free variables + free=#{}, %Free variables ws=[] :: [warning()], %Warnings. guard_refc=0}). %> 0 means in guard @@ -143,26 +145,35 @@ attributes([]) -> []. include_attribute(type) -> false; include_attribute(spec) -> false; +include_attribute(callback) -> false; include_attribute(opaque) -> false; include_attribute(export_type) -> false; +include_attribute(record) -> false; +include_attribute(optional_callbacks) -> false; +include_attribute(file) -> false; +include_attribute(compile) -> false; include_attribute(_) -> true. function({#c_var{name={F,Arity}=FA},Body}, St0) -> + %%io:format("~w/~w~n", [F,Arity]), try - St1 = St0#kern{func=FA,ff=undefined,vcount=0,fcount=0,ds=cerl_sets:new()}, + %% Find a suitable starting value for the variable counter. Note + %% that this pass assumes that new_var_name/1 returns a variable + %% name distinct from any variable used in the entire body of + %% the function. We use integers as variable names to avoid + %% filling up the atom table when compiling huge functions. + Count = cerl_trees:next_free_variable_name(Body), + St1 = St0#kern{func=FA,ff=undefined,vcount=Count,fcount=0,ds=cerl_sets:new()}, {#ifun{anno=Ab,vars=Kvs,body=B0},[],St2} = expr(Body, new_sub(), St1), {B1,_,St3} = ubody(B0, return, St2), %%B1 = B0, St3 = St2, %Null second pass - {#k_fdef{anno=#k{us=[],ns=[],a=Ab}, - func=F,arity=Arity,vars=Kvs,body=B1},St3} + {make_fdef(#k{us=[],ns=[],a=Ab}, F, Arity, Kvs, B1),St3} catch - Class:Error -> - Stack = erlang:get_stacktrace(), + Class:Error:Stack -> io:fwrite("Function: ~w/~w\n", [F,Arity]), erlang:raise(Class, Error, Stack) end. - %% body(Cexpr, Sub, State) -> {Kexpr,[PreKepxr],State}. %% Do the main sequence of a body. A body ends in an atomic value or %% values. Must check if vector first so do expr. @@ -187,9 +198,479 @@ body(Ce, Sub, St0) -> guard(G0, Sub, St0) -> {G1,St1} = wrap_guard(G0, St0), {Ge0,Pre,St2} = expr(G1, Sub, St1), - {Ge,St} = gexpr_test(Ge0, St2), + {Ge1,St3} = gexpr_test(Ge0, St2), + {Ge,St} = guard_opt(Ge1, St3), {pre_seq(Pre, Ge),St}. +%% guard_opt(Kexpr, State) -> {Kexpr,State}. +%% Optimize the Kexpr for the guard. Instead of evaluating a boolean +%% expression comparing it to 'true' in a final #k_test{}, +%% replace BIF calls with #k_test{} in the expression. +%% +%% As an example, take the guard: +%% +%% when is_integer(V0), is_atom(V1) -> +%% +%% The unoptimized Kexpr translated to pseudo BEAM assembly +%% code would look like: +%% +%% bif is_integer V0 => Bool0 +%% bif is_atom V1 => Bool1 +%% bif and Bool0 Bool1 => Bool +%% test Bool =:= true else goto Fail +%% ... +%% Fail: +%% ... +%% +%% The optimized code would look like: +%% +%% test is_integer V0 else goto Fail +%% test is_atom V1 else goto Fail +%% ... +%% Fail: +%% ... +%% +%% An 'or' operation is only slightly more complicated: +%% +%% test is_integer V0 else goto NotFailedYet +%% goto Success +%% +%% NotFailedYet: +%% test is_atom V1 else goto Fail +%% +%% Success: +%% ... +%% Fail: +%% ... + +guard_opt(G, St0) -> + {Root,Forest0,St1} = make_forest(G, St0), + {Exprs,Forest,St} = rewrite_bool(Root, Forest0, false, St1), + E = forest_pre_seq(Exprs, Forest), + {G#k_try{arg=E},St}. + +%% rewrite_bool(Kexpr, Forest, Inv, St) -> {[Kexpr],Forest,St}. +%% Rewrite Kexpr to use #k_test{} operations instead of comparison +%% and type test BIFs. +%% +%% If Kexpr is a #k_test{} operation, the call will always +%% succeed. Otherwise, a 'not_possible' exception will be +%% thrown if Kexpr cannot be rewritten. + +rewrite_bool(#k_test{op=#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val='=:='}}, + args=[#k_var{}=V,#k_atom{val=true}]}=Test, Forest0, Inv, St0) -> + try rewrite_bool_var(V, Forest0, Inv, St0) of + {_,_,_}=Res -> + Res + catch + throw:not_possible -> + {[Test],Forest0,St0} + end; +rewrite_bool(#k_test{op=#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val='=:='}}, + args=[#k_var{}=V,#k_atom{val=false}]}=Test, Forest0, Inv, St0) -> + try rewrite_bool_var(V, Forest0, not Inv, St0) of + {_,_,_}=Res -> + Res + catch + throw:not_possible -> + {[Test],Forest0,St0} + end; +rewrite_bool(#k_test{op=#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val='=:='}}, + args=[#k_atom{val=V1},#k_atom{val=V2}]}, Forest0, false, St0) -> + case V1 =:= V2 of + true -> + {[make_test(is_boolean, [#k_atom{val=true}])],Forest0,St0}; + false -> + {[make_failing_test()],Forest0,St0} + end; +rewrite_bool(#k_test{}=Test, Forest, false, St) -> + {[Test],Forest,St}; +rewrite_bool(#k_try{vars=[#k_var{name=X}],body=#k_var{name=X}, + handler=#k_atom{val=false},ret=[]}=Prot, + Forest0, Inv, St0) -> + {Root,Forest1,St1} = make_forest(Prot, Forest0, St0), + {Exprs,Forest2,St} = rewrite_bool(Root, Forest1, Inv, St1), + InnerForest = maps:without(maps:keys(Forest0), Forest2), + Forest = maps:without(maps:keys(InnerForest), Forest2), + E = forest_pre_seq(Exprs, InnerForest), + {[Prot#k_try{arg=E}],Forest,St}; +rewrite_bool(#k_match{body=Body,ret=[]}, Forest, Inv, St) -> + rewrite_match(Body, Forest, Inv, St); +rewrite_bool(Other, Forest, Inv, St) -> + case extract_bif(Other) of + {Name,Args} -> + rewrite_bif(Name, Args, Forest, Inv, St); + error -> + throw(not_possible) + end. + +%% rewrite_bool_var(Var, Forest, Inv, St) -> {[Kexpr],Forest,St}. +%% Rewrite the boolean expression whose key in Forest is +%% given by Var. Throw a 'not_possible' expression if something +%% prevents the rewriting. + +rewrite_bool_var(Arg, Forest0, Inv, St) -> + {Expr,Forest} = forest_take_expr(Arg, Forest0), + rewrite_bool(Expr, Forest, Inv, St). + +%% rewrite_bool_args([Kexpr], Forest, Inv, St) -> {[[Kexpr]],Forest,St}. +%% Rewrite each Kexpr in the list. The input Kexpr should be variables +%% or boolean values. Throw a 'not_possible' expression if something +%% prevents the rewriting. +%% +%% This function is suitable for handling the arguments for both +%% 'and' and 'or'. + +rewrite_bool_args([#k_atom{val=B}=A|Vs], Forest0, false=Inv, St0) when is_boolean(B) -> + {Tail,Forest1,St1} = rewrite_bool_args(Vs, Forest0, Inv, St0), + Bif = make_bif('=:=', [A,#k_atom{val=true}]), + {Exprs,Forest,St} = rewrite_bool(Bif, Forest1, Inv, St1), + {[Exprs|Tail],Forest,St}; +rewrite_bool_args([#k_var{}=Var|Vs], Forest0, false=Inv, St0) -> + {Tail,Forest1,St1} = rewrite_bool_args(Vs, Forest0, Inv, St0), + {Exprs,Forest,St} = + case is_bool_expr(Var, Forest0) of + true -> + rewrite_bool_var(Var, Forest1, Inv, St1); + false -> + Bif = make_bif('=:=', [Var,#k_atom{val=true}]), + rewrite_bool(Bif, Forest1, Inv, St1) + end, + {[Exprs|Tail],Forest,St}; +rewrite_bool_args([_|_], _Forest, _Inv, _St) -> + throw(not_possible); +rewrite_bool_args([], Forest, _Inv, St) -> + {[],Forest,St}. + +%% rewrite_bif(Name, [Kexpr], Forest, Inv, St) -> {[Kexpr],Forest,St}. +%% Rewrite a BIF. Throw a 'not_possible' expression if something +%% prevents the rewriting. + +rewrite_bif('or', Args, Forest, true, St) -> + rewrite_not_args('and', Args, Forest, St); +rewrite_bif('and', Args, Forest, true, St) -> + rewrite_not_args('or', Args, Forest, St); +rewrite_bif('and', [#k_atom{val=Val},Arg], Forest0, Inv, St0) -> + false = Inv, %Assertion. + case Val of + true -> + %% The result only depends on Arg. + rewrite_bool_var(Arg, Forest0, Inv, St0); + _ -> + %% Will fail. There is no need to evalute the expression + %% represented by Arg. Take it out from the forest and + %% discard the expression. + Failing = make_failing_test(), + try rewrite_bool_var(Arg, Forest0, Inv, St0) of + {_,Forest,St} -> + {[Failing],Forest,St} + catch + throw:not_possible -> + try forest_take_expr(Arg, Forest0) of + {_,Forest} -> + {[Failing],Forest,St0} + catch + throw:not_possible -> + %% Arg is probably a variable bound in an + %% outer scope. + {[Failing],Forest0,St0} + end + end + end; +rewrite_bif('and', [Arg,#k_atom{}=Atom], Forest, Inv, St) -> + false = Inv, %Assertion. + rewrite_bif('and', [Atom,Arg], Forest, Inv, St); +rewrite_bif('and', Args, Forest0, Inv, St0) -> + false = Inv, %Assertion. + {[Es1,Es2],Forest,St} = rewrite_bool_args(Args, Forest0, Inv, St0), + {Es1 ++ Es2,Forest,St}; +rewrite_bif('or', Args, Forest0, Inv, St0) -> + false = Inv, %Assertion. + {[First,Then],Forest,St} = rewrite_bool_args(Args, Forest0, Inv, St0), + Alt = make_alt(First, Then), + {[Alt],Forest,St}; +rewrite_bif('xor', [_,_], _Forest, _Inv, _St) -> + %% Rewriting 'xor' is not practical. Fortunately, 'xor' is + %% almost never used in practice. + throw(not_possible); +rewrite_bif('not', [Arg], Forest0, Inv, St) -> + {Expr,Forest} = forest_take_expr(Arg, Forest0), + rewrite_bool(Expr, Forest, not Inv, St); +rewrite_bif(Op, Args, Forest, Inv, St) -> + case is_test(Op, Args) of + true -> + rewrite_bool(make_test(Op, Args, Inv), Forest, false, St); + false -> + throw(not_possible) + end. + +rewrite_not_args(Op, [A0,B0], Forest0, St0) -> + {A,Forest1,St1} = rewrite_not_args_1(A0, Forest0, St0), + {B,Forest2,St2} = rewrite_not_args_1(B0, Forest1, St1), + rewrite_bif(Op, [A,B], Forest2, false, St2). + +rewrite_not_args_1(Arg, Forest, St) -> + Not = make_bif('not', [Arg]), + forest_add_expr(Not, Forest, St). + +%% rewrite_match(Kvar, TypeClause, Forest, Inv, St) -> +%% {[Kexpr],Forest,St}. +%% Try to rewrite a #k_match{} originating from an 'andalso' or an 'orelse'. + +rewrite_match(#k_alt{first=First,then=Then}, Forest, Inv, St) -> + case {First,Then} of + {#k_select{var=#k_var{name=V}=Var,types=[TypeClause]},#k_var{name=V}} -> + rewrite_match_1(Var, TypeClause, Forest, Inv, St); + {_,_} -> + throw(not_possible) + end. + +rewrite_match_1(Var, #k_type_clause{values=Cs0}, Forest0, Inv, St0) -> + Cs = sort([{Val,B} || #k_val_clause{val=#k_atom{val=Val},body=B} <- Cs0]), + case Cs of + [{false,False},{true,True}] -> + rewrite_match_2(Var, False, True, Forest0, Inv, St0); + _ -> + throw(not_possible) + end. + +rewrite_match_2(Var, False, #k_atom{val=true}, Forest0, Inv, St0) -> + %% Originates from an 'orelse'. + case False of + #k_atom{val=NotBool} when not is_boolean(NotBool) -> + rewrite_bool(Var, Forest0, Inv, St0); + _ -> + {CodeVar,Forest1,St1} = add_protected_expr(False, Forest0, St0), + rewrite_bif('or', [Var,CodeVar], Forest1, Inv, St1) + end; +rewrite_match_2(Var, #k_atom{val=false}, True, Forest0, Inv, St0) -> + %% Originates from an 'andalso'. + {CodeVar,Forest1,St1} = add_protected_expr(True, Forest0, St0), + rewrite_bif('and', [Var,CodeVar], Forest1, Inv, St1); +rewrite_match_2(_V, _, _, _Forest, _Inv, _St) -> + throw(not_possible). + +%% is_bool_expr(#k_var{}, Forest) -> true|false. +%% Return true if the variable refers to a boolean expression +%% that does not need an explicit '=:= true' test. + +is_bool_expr(V, Forest) -> + case forest_peek_expr(V, Forest) of + error -> + %% Defined outside of the guard. We can't know. + false; + Expr -> + case extract_bif(Expr) of + {Name,Args} -> + is_test(Name, Args) orelse + erl_internal:bool_op(Name, length(Args)); + error -> + %% Not a BIF. Should be possible to rewrite + %% to a boolean. Definitely does not need + %% a '=:= true' test. + true + end + end. + +make_bif(Op, Args) -> + #k_bif{op=#k_remote{mod=#k_atom{val=erlang}, + name=#k_atom{val=Op}, + arity=length(Args)}, + args=Args}. + +extract_bif(#k_bif{op=#k_remote{mod=#k_atom{val=erlang}, + name=#k_atom{val=Name}}, + args=Args}) -> + {Name,Args}; +extract_bif(_) -> + error. + +%% make_alt(First, Then) -> KMatch. +%% Make a #k_alt{} within a #k_match{} to implement +%% 'or' or 'orelse'. + +make_alt(First0, Then0) -> + First1 = pre_seq(droplast(First0), last(First0)), + Then1 = pre_seq(droplast(Then0), last(Then0)), + First2 = make_protected(First1), + Then2 = make_protected(Then1), + Body = #ignored{}, + First3 = #k_guard_clause{guard=First2,body=Body}, + Then3 = #k_guard_clause{guard=Then2,body=Body}, + First = #k_guard{clauses=[First3]}, + Then = #k_guard{clauses=[Then3]}, + Alt = #k_alt{first=First,then=Then}, + #k_match{vars=[],body=Alt}. + +add_protected_expr(#k_atom{}=Atom, Forest, St) -> + {Atom,Forest,St}; +add_protected_expr(#k_var{}=Var, Forest, St) -> + {Var,Forest,St}; +add_protected_expr(E0, Forest, St) -> + E = make_protected(E0), + forest_add_expr(E, Forest, St). + +make_protected(#k_try{}=Try) -> + Try; +make_protected(B) -> + #k_try{arg=B,vars=[#k_var{name=''}],body=#k_var{name=''}, + handler=#k_atom{val=false}}. + +make_failing_test() -> + make_test(is_boolean, [#k_atom{val=fail}]). + +make_test(Op, Args) -> + make_test(Op, Args, false). + +make_test(Op, Args, Inv) -> + Remote = #k_remote{mod=#k_atom{val=erlang}, + name=#k_atom{val=Op}, + arity=length(Args)}, + #k_test{op=Remote,args=Args,inverted=Inv}. + +is_test(Op, Args) -> + A = length(Args), + erl_internal:new_type_test(Op, A) orelse erl_internal:comp_op(Op, A). + +%% make_forest(Kexpr, St) -> {RootKexpr,Forest,St}. +%% Build a forest out of Kexpr. RootKexpr is the final expression +%% nested inside Kexpr. + +make_forest(G, St) -> + make_forest_1(G, #{}, 0, St). + +%% make_forest(Kexpr, St) -> {RootKexpr,Forest,St}. +%% Add to Forest from Kexpr. RootKexpr is the final expression +%% nested inside Kexpr. + +make_forest(G, Forest0, St) -> + N = forest_next_index(Forest0), + make_forest_1(G, Forest0, N, St). + +make_forest_1(#k_try{arg=B}, Forest, I, St) -> + make_forest_1(B, Forest, I, St); +make_forest_1(#iset{vars=[]}=Iset0, Forest, I, St0) -> + {UnrefVar,St} = new_var(St0), + Iset = Iset0#iset{vars=[UnrefVar]}, + make_forest_1(Iset, Forest, I, St); +make_forest_1(#iset{vars=[#k_var{name=V}],arg=Arg,body=B}, Forest0, I, St) -> + Forest = Forest0#{V => {I,Arg}, {untaken,V} => true}, + make_forest_1(B, Forest, I+1, St); +make_forest_1(Innermost, Forest, _I, St) -> + {Innermost,Forest,St}. + +%% forest_take_expr(Kexpr, Forest) -> {Expr,Forest}. +%% If Kexpr is a variable, take out the expression corresponding +%% to variable in Forest. Expressions that have been taken out +%% of the forest will not be included the Kexpr returned +%% by forest_pre_seq/2. +%% +%% Throw a 'not_possible' exception if Kexpr is not a variable or +%% if the name of the variable is not a key in Forest. + +forest_take_expr(#k_var{name=V}, Forest0) -> + %% v3_core currently always generates guard expressions that can + %% be represented as a tree. Other code generators (such as LFE) + %% could generate guard expressions that can only be represented + %% as a DAG (i.e. some nodes are referenced more than once). To + %% handle DAGs, we must never remove a node from the forest, but + %% just remove the {untaken,V} marker. That will effectively convert + %% the DAG to a tree by duplicating the shared nodes and their + %% descendants. + + case maps:find(V, Forest0) of + {ok,{_,Expr}} -> + Forest = maps:remove({untaken,V}, Forest0), + {Expr,Forest}; + error -> + throw(not_possible) + end; +forest_take_expr(_, _) -> + throw(not_possible). + +%% forest_peek_expr(Kvar, Forest) -> Kexpr | error. +%% Return the expression corresponding to Kvar in Forest or +%% return 'error' if there is a corresponding expression. + +forest_peek_expr(#k_var{name=V}, Forest0) -> + case maps:find(V, Forest0) of + {ok,{_,Expr}} -> Expr; + error -> error + end. + +%% forest_add_expr(Kexpr, Forest, St) -> {Kvar,Forest,St}. +%% Add a new expression to Forest. + +forest_add_expr(Expr, Forest0, St0) -> + {#k_var{name=V}=Var,St} = new_var(St0), + N = forest_next_index(Forest0), + Forest = Forest0#{V => {N,Expr}}, + {Var,Forest,St}. + +forest_next_index(Forest) -> + 1 + lists:max([N || {N,_} <- maps:values(Forest), + is_integer(N)] ++ [0]). + +%% forest_pre_seq([Kexpr], Forest) -> Kexpr. +%% Package the list of Kexprs into a nested Kexpr, prepending all +%% expressions in Forest that have not been taken out using +%% forest_take_expr/2. + +forest_pre_seq(Exprs, Forest) -> + Es0 = [#k_var{name=V} || {untaken,V} <- maps:keys(Forest)], + Es = Es0 ++ Exprs, + Vs = extract_all_vars(Es, Forest, []), + Pre0 = sort([{maps:get(V, Forest),V} || V <- Vs]), + Pre = [#iset{vars=[#k_var{name=V}],arg=A} || + {{_,A},V} <- Pre0], + pre_seq(Pre++droplast(Exprs), last(Exprs)). + +extract_all_vars(Es, Forest, Acc0) -> + case extract_var_list(Es) of + [] -> + Acc0; + [_|_]=Vs0 -> + Vs = [V || V <- Vs0, maps:is_key(V, Forest)], + NewVs = ordsets:subtract(Vs, Acc0), + NewEs = [begin + {_,E} = maps:get(V, Forest), + E + end || V <- NewVs], + Acc = union(NewVs, Acc0), + extract_all_vars(NewEs, Forest, Acc) + end. + +extract_vars(#iset{arg=A,body=B}) -> + union(extract_vars(A), extract_vars(B)); +extract_vars(#k_bif{args=Args}) -> + ordsets:from_list(lit_list_vars(Args)); +extract_vars(#k_call{}) -> + []; +extract_vars(#k_test{args=Args}) -> + ordsets:from_list(lit_list_vars(Args)); +extract_vars(#k_match{body=Body}) -> + extract_vars(Body); +extract_vars(#k_alt{first=First,then=Then}) -> + union(extract_vars(First), extract_vars(Then)); +extract_vars(#k_guard{clauses=Cs}) -> + extract_var_list(Cs); +extract_vars(#k_guard_clause{guard=G}) -> + extract_vars(G); +extract_vars(#k_select{var=Var,types=Types}) -> + union(ordsets:from_list(lit_vars(Var)), + extract_var_list(Types)); +extract_vars(#k_type_clause{values=Values}) -> + extract_var_list(Values); +extract_vars(#k_val_clause{body=Body}) -> + extract_vars(Body); +extract_vars(#k_try{arg=Arg}) -> + extract_vars(Arg); +extract_vars(Lit) -> + ordsets:from_list(lit_vars(Lit)). + +extract_var_list(L) -> + union([extract_vars(E) || E <- L]). + %% Wrap the entire guard in a try/catch if needed. wrap_guard(#c_try{}=Try, St) -> {Try,St}; @@ -239,7 +720,7 @@ gexpr_test_add(Ke, St0) -> expr(#c_var{anno=A,name={_Name,Arity}}=Fname, Sub, St) -> %% A local in an expression. %% For now, these are wrapped into a fun by reverse - %% etha-conversion, but really, there should be exactly one + %% eta-conversion, but really, there should be exactly one %% such "lambda function" for each escaping local name, %% instead of one for each occurrence as done now. Vs = [#c_var{name=list_to_atom("V" ++ integer_to_list(V))} || @@ -399,7 +880,7 @@ expr(#c_call{anno=A,module=M0,name=F0,args=Cargs}, Sub, St0) -> Call = #c_call{anno=A, module=#c_literal{val=erlang}, name=#c_literal{val=apply}, - args=[M0,F0,make_list(Cargs)]}, + args=[M0,F0,cerl:make_list(Cargs)]}, expr(Call, Sub, St1); _ -> {[M1,F1|Kargs],Ap,St} = atomic_list([M0,F0|Cargs], Sub, St1), @@ -494,7 +975,7 @@ translate_match_fail_1(Anno, As, Sub, #kern{ff=FF}) -> end. translate_fc(Args) -> - [#c_literal{val=function_clause},make_list(Args)]. + [#c_literal{val=function_clause},cerl:make_list(Args)]. expr_map(A,Var0,Ces,Sub,St0) -> {Var,Mps,St1} = expr(Var0, Sub, St0), @@ -837,23 +1318,26 @@ get_vsub(V, Vsub) -> set_vsub(V, S, Vsub) -> orddict:store(V, S, Vsub). -subst_vsub(Key, New, [{K,Key}|Dict]) -> +subst_vsub(Key, New, Vsub) -> + orddict:from_list(subst_vsub_1(Key, New, Vsub)). + +subst_vsub_1(Key, New, [{K,Key}|Dict]) -> %% Fold chained substitution. - [{K,New}|subst_vsub(Key, New, Dict)]; -subst_vsub(Key, New, [{K,_}|_]=Dict) when Key < K -> + [{K,New}|subst_vsub_1(Key, New, Dict)]; +subst_vsub_1(Key, New, [{K,_}|_]=Dict) when Key < K -> %% Insert the new substitution here, and continue %% look for chained substitutions. - [{Key,New}|subst_vsub_1(Key, New, Dict)]; -subst_vsub(Key, New, [{K,_}=E|Dict]) when Key > K -> - [E|subst_vsub(Key, New, Dict)]; -subst_vsub(Key, New, []) -> [{Key,New}]. + [{Key,New}|subst_vsub_2(Key, New, Dict)]; +subst_vsub_1(Key, New, [{K,_}=E|Dict]) when Key > K -> + [E|subst_vsub_1(Key, New, Dict)]; +subst_vsub_1(Key, New, []) -> [{Key,New}]. -subst_vsub_1(V, S, [{K,V}|Dict]) -> +subst_vsub_2(V, S, [{K,V}|Dict]) -> %% Fold chained substitution. - [{K,S}|subst_vsub_1(V, S, Dict)]; -subst_vsub_1(V, S, [E|Dict]) -> - [E|subst_vsub_1(V, S, Dict)]; -subst_vsub_1(_, _, []) -> []. + [{K,S}|subst_vsub_2(V, S, Dict)]; +subst_vsub_2(V, S, [E|Dict]) -> + [E|subst_vsub_2(V, S, Dict)]; +subst_vsub_2(_, _, []) -> []. get_fsub(F, A, Fsub) -> case orddict:find({F,A}, Fsub) of @@ -877,7 +1361,7 @@ new_fun_name(Type, #kern{func={F,Arity},fcount=C}=St) -> %% new_var_name(State) -> {VarName,State}. new_var_name(#kern{vcount=C}=St) -> - {list_to_atom("ker" ++ integer_to_list(C)),St#kern{vcount=C+1}}. + {C,St#kern{vcount=C+1}}. %% new_var(State) -> {#k_var{},State}. @@ -1110,23 +1594,18 @@ match_var([U|Us], Cs0, Def, St) -> %% according to type, the order is really irrelevant but tries to be %% smart. -match_con(Us, [C], Def, St) -> - %% There is only one clause. We can keep literal tuples and - %% lists, but we must convert []/integer/float/atom literals - %% to the proper record (#k_nil{} and so on). - Cs = [expand_pat_lit_clause(C, false)], - match_con_1(Us, Cs, Def, St); match_con(Us, Cs0, Def, St) -> - %% More than one clause. Remove literals at the top level. - Cs = [expand_pat_lit_clause(C, true) || C <- Cs0], + %% Expand literals at the top level. + Cs = [expand_pat_lit_clause(C) || C <- Cs0], match_con_1(Us, Cs, Def, St). match_con_1([U|_Us] = L, Cs, Def, St0) -> %% Extract clauses for different constructors (types). %%ok = io:format("match_con ~p~n", [Cs]), - Ttcs = select_types([k_binary], Cs) ++ select_bin_con(Cs) ++ - select_types([k_cons,k_tuple,k_map,k_atom,k_float,k_int, - k_nil,k_literal], Cs), + Ttcs0 = select_types([k_binary], Cs) ++ select_bin_con(Cs) ++ + select_types([k_cons,k_tuple,k_map,k_atom,k_float, + k_int,k_nil], Cs), + Ttcs = opt_single_valued(Ttcs0), %%ok = io:format("ttcs = ~p~n", [Ttcs]), {Scs,St1} = mapfoldl(fun ({T,Tcs}, St) -> @@ -1139,28 +1618,14 @@ match_con_1([U|_Us] = L, Cs, Def, St0) -> select_types(Types, Cs) -> [{T,Tcs} || T <- Types, begin Tcs = select(T, Cs), Tcs =/= [] end]. - -expand_pat_lit_clause(#iclause{pats=[#ialias{pat=#k_literal{anno=A,val=Val}}=Alias|Ps]}=C, B) -> - P = case B of - true -> expand_pat_lit(Val, A); - false -> literal(Val, A) - end, + +expand_pat_lit_clause(#iclause{pats=[#ialias{pat=#k_literal{anno=A,val=Val}}=Alias|Ps]}=C) -> + P = expand_pat_lit(Val, A), C#iclause{pats=[Alias#ialias{pat=P}|Ps]}; -expand_pat_lit_clause(#iclause{pats=[#k_literal{anno=A,val=Val}|Ps]}=C, B) -> - P = case B of - true -> expand_pat_lit(Val, A); - false -> literal(Val, A) - end, +expand_pat_lit_clause(#iclause{pats=[#k_literal{anno=A,val=Val}|Ps]}=C) -> + P = expand_pat_lit(Val, A), C#iclause{pats=[P|Ps]}; -expand_pat_lit_clause(#iclause{pats=[#k_binary{anno=A,segs=#k_bin_end{}}|Ps]}=C, B) -> - case B of - true -> - C; - false -> - P = #k_literal{anno=A,val = <<>>}, - C#iclause{pats=[P|Ps]} - end; -expand_pat_lit_clause(C, _) -> C. +expand_pat_lit_clause(C) -> C. expand_pat_lit([H|T], A) -> #k_cons{anno=A,hd=literal(H, A),tl=literal(T, A)}; @@ -1180,6 +1645,107 @@ literal(Val, A) when is_atom(Val) -> literal(Val, A) when is_list(Val); is_tuple(Val) -> #k_literal{anno=A,val=Val}. +%% opt_singled_valued([{Type,Clauses}]) -> [{Type,Clauses}]. +%% If a type only has one clause and if the pattern is literal, +%% the matching can be done more efficiently by directly comparing +%% with the literal (that is especially true for binaries). + +opt_single_valued(Ttcs) -> + opt_single_valued(Ttcs, [], []). + +opt_single_valued([{_,[#iclause{pats=[P0|Ps]}=Tc]}=Ttc|Ttcs], TtcAcc, LitAcc) -> + try combine_lit_pat(P0) of + P -> + LitTtc = Tc#iclause{pats=[P|Ps]}, + opt_single_valued(Ttcs, TtcAcc, [LitTtc|LitAcc]) + catch + not_possible -> + opt_single_valued(Ttcs, [Ttc|TtcAcc], LitAcc) + end; +opt_single_valued([Ttc|Ttcs], TtcAcc, LitAcc) -> + opt_single_valued(Ttcs, [Ttc|TtcAcc], LitAcc); +opt_single_valued([], TtcAcc, []) -> + reverse(TtcAcc); +opt_single_valued([], TtcAcc, LitAcc) -> + Literals = {k_literal,reverse(LitAcc)}, + %% Test the literals as early as possible. + case reverse(TtcAcc) of + [{k_binary,_}=Bin|Ttcs] -> + %% The delayed creation of sub binaries requires + %% bs_start_match2 to be the first instruction in the + %% function. + [Bin,Literals|Ttcs]; + Ttcs -> + [Literals|Ttcs] + end. + +combine_lit_pat(#ialias{pat=Pat0}=Alias) -> + Pat = combine_lit_pat(Pat0), + Alias#ialias{pat=Pat}; +combine_lit_pat(Pat) -> + case do_combine_lit_pat(Pat) of + #k_literal{val=Val} when is_atom(Val) -> + throw(not_possible); + #k_literal{val=Val} when is_number(Val) -> + throw(not_possible); + #k_literal{val=[]} -> + throw(not_possible); + #k_literal{}=Lit -> + Lit + end. + +do_combine_lit_pat(#k_atom{anno=A,val=Val}) -> + #k_literal{anno=A,val=Val}; +do_combine_lit_pat(#k_float{anno=A,val=Val}) -> + #k_literal{anno=A,val=Val}; +do_combine_lit_pat(#k_int{anno=A,val=Val}) -> + #k_literal{anno=A,val=Val}; +do_combine_lit_pat(#k_nil{anno=A}) -> + #k_literal{anno=A,val=[]}; +do_combine_lit_pat(#k_binary{anno=A,segs=Segs}) -> + Bin = combine_bin_segs(Segs), + #k_literal{anno=A,val=Bin}; +do_combine_lit_pat(#k_cons{anno=A,hd=Hd0,tl=Tl0}) -> + #k_literal{val=Hd} = do_combine_lit_pat(Hd0), + #k_literal{val=Tl} = do_combine_lit_pat(Tl0), + #k_literal{anno=A,val=[Hd|Tl]}; +do_combine_lit_pat(#k_literal{}=Lit) -> + Lit; +do_combine_lit_pat(#k_tuple{anno=A,es=Es0}) -> + Es = [begin + #k_literal{val=Lit} = do_combine_lit_pat(El), + Lit + end || El <- Es0], + #k_literal{anno=A,val=list_to_tuple(Es)}; +do_combine_lit_pat(_) -> + throw(not_possible). + +combine_bin_segs(#k_bin_seg{size=Size0,unit=Unit,type=integer, + flags=[unsigned,big],seg=Seg,next=Next}) -> + #k_literal{val=Size1} = do_combine_lit_pat(Size0), + #k_literal{val=Int} = do_combine_lit_pat(Seg), + Size = Size1 * Unit, + if + 0 < Size, Size < 64 -> + Bin = <<Int:Size>>, + case Bin of + <<Int:Size>> -> + NextBin = combine_bin_segs(Next), + <<Bin/bits,NextBin/bits>>; + _ -> + %% The integer Int does not fit in the segment, + %% thus it will not match. + throw(not_possible) + end; + true -> + %% Avoid creating huge binary literals. + throw(not_possible) + end; +combine_bin_segs(#k_bin_end{}) -> + <<>>; +combine_bin_segs(_) -> + throw(not_possible). + %% select_bin_con([Clause]) -> [{Type,[Clause]}]. %% Extract clauses for the k_bin_seg constructor. As k_bin_seg %% matching can overlap, the k_bin_seg constructors cannot be @@ -1347,10 +1913,70 @@ select(T, Cs) -> [ C || C <- Cs, clause_con(C) =:= T ]. %% At this point all the clauses have the same constructor, we must %% now separate them according to value. -match_value(Us, T, Cs0, Def, St0) -> - Css = group_value(T, Cs0), +match_value(Us0, T, Cs0, Def, St0) -> + {Us1,Cs1,St1} = partition_intersection(T, Us0, Cs0, St0), + UCss = group_value(T, Us1, Cs1), %%ok = io:format("match_value ~p ~p~n", [T, Css]), - mapfoldl(fun (Cs, St) -> match_clause(Us, Cs, Def, St) end, St0, Css). + mapfoldl(fun ({Us,Cs}, St) -> match_clause(Us, Cs, Def, St) end, St1, UCss). + +%% partition_intersection +%% Partitions a map into two maps with the most common keys to the first map. +%% case <M> of +%% <#{a}> +%% <#{a,b}> +%% <#{a,c}> +%% <#{c}> +%% end +%% becomes +%% case <M,M> of +%% <#{a}, #{ }> +%% <#{a}, #{b}> +%% <#{ }, #{c}> +%% <#{a}, #{c}> +%% end +%% The intention is to group as many keys together as possible and thus +%% reduce the number of lookups to that key. +partition_intersection(k_map, [U|_]=Us0, [_,_|_]=Cs0,St0) -> + Ps = [clause_val(C) || C <- Cs0], + case find_key_partition(Ps) of + no_partition -> + {Us0,Cs0,St0}; + Ks -> + {Cs1,St1} = mapfoldl(fun(#iclause{pats=[Arg|Args]}=C, Sti) -> + {{Arg1,Arg2},St} = partition_key_intersection(Arg, Ks, Sti), + {C#iclause{pats=[Arg1,Arg2|Args]}, St} + end, St0, Cs0), + {[U|Us0],Cs1,St1} + end; +partition_intersection(_, Us, Cs, St) -> + {Us,Cs,St}. + +partition_key_intersection(#k_map{es=Pairs}=Map,Ks,St0) -> + F = fun(#k_map_pair{key=Key}) -> member(map_key_clean(Key), Ks) end, + {Ps1,Ps2} = partition(F, Pairs), + {{Map#k_map{es=Ps1},Map#k_map{es=Ps2}},St0}; +partition_key_intersection(#ialias{pat=Map}=Alias,Ks,St0) -> + %% only alias one of them + {{Map1,Map2},St1} = partition_key_intersection(Map, Ks, St0), + {{Map1,Alias#ialias{pat=Map2}},St1}. + +% Only check for the complete intersection of keys and not commonality +find_key_partition(Ps) -> + Sets = [sets:from_list(Ks)||Ks <- Ps], + Is = sets:intersection(Sets), + case sets:to_list(Is) of + [] -> no_partition; + KeyIntersection -> + %% Check if the intersection are all keys in all clauses. + %% Don't split if they are since this will only + %% infer extra is_map instructions with no gain. + All = foldl(fun (Kset, Bool) -> + Bool andalso sets:is_subset(Kset, Is) + end, true, Sets), + if All -> no_partition; + true -> KeyIntersection + end + end. %% group_value([Clause]) -> [[Clause]]. %% Group clauses according to value. Here we know that @@ -1358,30 +1984,30 @@ match_value(Us, T, Cs0, Def, St0) -> %% 2. The clauses in bin_segs cannot be reordered only grouped %% 3. Other types are disjoint and can be reordered -group_value(k_cons, Cs) -> [Cs]; %These are single valued -group_value(k_nil, Cs) -> [Cs]; -group_value(k_binary, Cs) -> [Cs]; -group_value(k_bin_end, Cs) -> [Cs]; -group_value(k_bin_seg, Cs) -> group_bin_seg(Cs); -group_value(k_bin_int, Cs) -> [Cs]; -group_value(k_map, Cs) -> group_map(Cs); -group_value(_, Cs) -> +group_value(k_cons, Us, Cs) -> [{Us,Cs}]; %These are single valued +group_value(k_nil, Us, Cs) -> [{Us,Cs}]; +group_value(k_binary, Us, Cs) -> [{Us,Cs}]; +group_value(k_bin_end, Us, Cs) -> [{Us,Cs}]; +group_value(k_bin_seg, Us, Cs) -> group_bin_seg(Us,Cs); +group_value(k_bin_int, Us, Cs) -> [{Us,Cs}]; +group_value(k_map, Us, Cs) -> group_map(Us,Cs); +group_value(_, Us, Cs) -> %% group_value(Cs). Cd = foldl(fun (C, Gcs0) -> dict:append(clause_val(C), C, Gcs0) end, dict:new(), Cs), - dict:fold(fun (_, Vcs, Css) -> [Vcs|Css] end, [], Cd). + dict:fold(fun (_, Vcs, Css) -> [{Us,Vcs}|Css] end, [], Cd). -group_bin_seg([C1|Cs]) -> +group_bin_seg(Us, [C1|Cs]) -> V1 = clause_val(C1), {More,Rest} = splitwith(fun (C) -> clause_val(C) == V1 end, Cs), - [[C1|More]|group_bin_seg(Rest)]; -group_bin_seg([]) -> []. + [{Us,[C1|More]}|group_bin_seg(Us,Rest)]; +group_bin_seg(_, []) -> []. -group_map([C1|Cs]) -> +group_map(Us, [C1|Cs]) -> V1 = clause_val(C1), {More,Rest} = splitwith(fun (C) -> clause_val(C) =:= V1 end, Cs), - [[C1|More]|group_map(Rest)]; -group_map([]) -> []. + [{Us,[C1|More]}|group_map(Us,Rest)]; +group_map(_, []) -> []. %% Profiling shows that this quadratic implementation account for a big amount %% of the execution time if there are many values. @@ -1604,7 +2230,9 @@ ubody(E, return, St0) -> {Ea,Pa,St1} = force_atomic(E, St0), ubody(pre_seq(Pa, #ivalues{args=[Ea]}), return, St1) end; -ubody(E, {break,_Rs} = Break, St0) -> +ubody(#ignored{}, {break,_} = Break, St) -> + ubody(#ivalues{args=[]}, Break, St); +ubody(E, {break,[_]} = Break, St0) -> %%ok = io:fwrite("ubody ~w:~p~n", [?LINE,{E,Br}]), %% Exiting expressions need no trailing break. case is_exit_expr(E) of @@ -1612,6 +2240,16 @@ ubody(E, {break,_Rs} = Break, St0) -> false -> {Ea,Pa,St1} = force_atomic(E, St0), ubody(pre_seq(Pa, #ivalues{args=[Ea]}), Break, St1) + end; +ubody(E, {break,Rs}=Break, St0) -> + case is_exit_expr(E) of + true -> + uexpr(E, return, St0); + false -> + {Vs,St1} = new_vars(length(Rs), St0), + Iset = #iset{vars=Vs,arg=E}, + PreSeq = pre_seq([Iset], #ivalues{args=Vs}), + ubody(PreSeq, Break, St1) end. iletrec_funs(#iletrec{defs=Fs}, St0) -> @@ -1640,9 +2278,8 @@ iletrec_funs_gen(Fs, FreeVs, St) -> Arity0 = length(Vs), {Fb1,_,Lst1} = ubody(Fb0, return, Lst0#kern{ff={N,Arity0}}), Arity = Arity0 + length(FreeVs), - Fun = #k_fdef{anno=#k{us=[],ns=[],a=Fa}, - func=N,arity=Arity, - vars=Vs ++ FreeVs,body=Fb1}, + Fun = make_fdef(#k{us=[],ns=[],a=Fa}, N, Arity, + Vs++FreeVs, Fb1), Lst1#kern{funs=[Fun|Lst1#kern.funs]} end, St, Fs). @@ -1731,26 +2368,25 @@ uexpr(#k_receive_accept{anno=A}, _, St) -> {#k_receive_accept{anno=#k{us=[],ns=[],a=A}},[],St}; uexpr(#k_receive_next{anno=A}, _, St) -> {#k_receive_next{anno=#k{us=[],ns=[],a=A}},[],St}; -uexpr(#k_try{anno=A,arg=A0,vars=Vs,body=B0,evars=Evs,handler=H0}=Try, +uexpr(#k_try{anno=A,arg=A0,vars=Vs,body=B0,evars=Evs,handler=H0}, {break,Rs0}=Br, St0) -> case is_in_guard(St0) of true -> {[#k_var{name=X}],#k_var{name=X}} = {Vs,B0}, %Assertion. #k_atom{val=false} = H0, %Assertion. {A1,Bu,St1} = uexpr(A0, Br, St0), - {Try#k_try{anno=#k{us=Bu,ns=lit_list_vars(Rs0),a=A}, - arg=A1,ret=Rs0},Bu,St1}; + {#k_protected{anno=#k{us=Bu,ns=lit_list_vars(Rs0),a=A}, + arg=A1,ret=Rs0},Bu,St1}; false -> {Avs,St1} = new_vars(length(Vs), St0), {A1,Au,St2} = ubody(A0, {break,Avs}, St1), {B1,Bu,St3} = ubody(B0, Br, St2), {H1,Hu,St4} = ubody(H0, Br, St3), - {Rs1,St5} = ensure_return_vars(Rs0, St4), Used = union([Au,subtract(Bu, lit_list_vars(Vs)), subtract(Hu, lit_list_vars(Evs))]), - {#k_try{anno=#k{us=Used,ns=lit_list_vars(Rs1),a=A}, - arg=A1,vars=Vs,body=B1,evars=Evs,handler=H1,ret=Rs1}, - Used,St5} + {#k_try{anno=#k{us=Used,ns=lit_list_vars(Rs0),a=A}, + arg=A1,vars=Vs,body=B1,evars=Evs,handler=H1,ret=Rs0}, + Used,St4} end; uexpr(#k_try{anno=A,arg=A0,vars=Vs,body=B0,evars=Evs,handler=H0}, return, St0) -> @@ -1758,13 +2394,11 @@ uexpr(#k_try{anno=A,arg=A0,vars=Vs,body=B0,evars=Evs,handler=H0}, {A1,Au,St2} = ubody(A0, {break,Avs}, St1), %Must break to clean up here! {B1,Bu,St3} = ubody(B0, return, St2), {H1,Hu,St4} = ubody(H0, return, St3), - NumNew = 1, - {Ns,St5} = new_vars(NumNew, St4), Used = union([Au,subtract(Bu, lit_list_vars(Vs)), subtract(Hu, lit_list_vars(Evs))]), - {#k_try_enter{anno=#k{us=Used,ns=Ns,a=A}, + {#k_try_enter{anno=#k{us=Used,ns=[],a=A}, arg=A1,vars=Vs,body=B1,evars=Evs,handler=H1}, - Used,St5}; + Used,St4}; uexpr(#k_catch{anno=A,body=B0}, {break,Rs0}, St0) -> {Rb,St1} = new_var(St0), {B1,Bu,St2} = ubody(B0, {break,[Rb]}, St1), @@ -1786,15 +2420,10 @@ uexpr(#ifun{anno=A,vars=Vs,body=B0}, {break,Rs}, St0) -> %% No id annotation. Must invent a fun name. new_fun_name(St1) end, - Fun = #k_fdef{anno=#k{us=[],ns=[],a=A},func=Fname,arity=Arity, - vars=Vs ++ Fvs,body=B1}, - %% Set dummy values for Index and Uniq -- the real values will - %% be assigned by beam_asm. - Index = Uniq = 0, + Fun = make_fdef(#k{us=[],ns=[],a=A}, Fname, Arity, Vs++Fvs, B1), {#k_bif{anno=#k{us=Free,ns=lit_list_vars(Rs),a=A}, - op=#k_internal{name=make_fun,arity=length(Free)+3}, - args=[#k_atom{val=Fname},#k_int{val=Arity}, - #k_int{val=Index},#k_int{val=Uniq}|Fvs], + op=#k_internal{name=make_fun,arity=length(Free)+2}, + args=[#k_atom{val=Fname},#k_int{val=Arity}|Fvs], ret=Rs}, Free,add_local_function(Fun, St)}; uexpr(Lit, {break,Rs0}, St0) -> @@ -1808,6 +2437,16 @@ uexpr(Lit, {break,Rs0}, St0) -> add_local_function(_, #kern{funs=ignore}=St) -> St; add_local_function(F, #kern{funs=Funs}=St) -> St#kern{funs=[F|Funs]}. +%% Make a #k_fdef{}, making sure that the body is always a #k_match{}. +make_fdef(Anno, Name, Arity, Vs, #k_match{}=Body) -> + #k_fdef{anno=Anno,func=Name,arity=Arity,vars=Vs,body=Body}; +make_fdef(Anno, Name, Arity, Vs, Body) -> + Ka = get_kanno(Body), + Match = #k_match{anno=#k{us=Ka#k.us,ns=[],a=Ka#k.a}, + vars=Vs,body=Body,ret=[]}, + #k_fdef{anno=Anno,func=Name,arity=Arity,vars=Vs,body=Match}. + + %% handle_reuse_annos([#k_var{}], State) -> State. %% In general, it is only safe to reuse a variable for a match context %% if the original value of the variable will no longer be needed. @@ -1837,14 +2476,17 @@ handle_reuse_anno_1(V, _St) -> V. %% get_free(Name, Arity, State) -> [Free]. %% store_free(Name, Arity, [Free], State) -> State. -get_free(F, A, St) -> - case orddict:find({F,A}, St#kern.free) of - {ok,Val} -> Val; - error -> [] +get_free(F, A, #kern{free=FreeMap}) -> + Key = {F,A}, + case FreeMap of + #{Key:=Val} -> Val; + _ -> [] end. -store_free(F, A, Free, St) -> - St#kern{free=orddict:store({F,A}, Free, St#kern.free)}. +store_free(F, A, Free, #kern{free=FreeMap0}=St) -> + Key = {F,A}, + FreeMap = FreeMap0#{Key=>Free}, + St#kern{free=FreeMap}. break_rets({break,Rs}) -> Rs; break_rets(return) -> []. @@ -1983,11 +2625,6 @@ pat_list_vars(Ps) -> {union(Used0, Used),union(New0, New)} end, {[],[]}, Ps). -make_list(Es) -> - foldr(fun(E, Acc) -> - #c_cons{hd=E,tl=Acc} - end, #c_literal{val=[]}, Es). - %% List of integers in interval [N,M]. Empty list if N > M. integers(N, M) when N =< M -> diff --git a/lib/compiler/src/v3_kernel.hrl b/lib/compiler/src/v3_kernel.hrl index 03b2eae006..e6f0d3c1f7 100644 --- a/lib/compiler/src/v3_kernel.hrl +++ b/lib/compiler/src/v3_kernel.hrl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2012. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -58,7 +58,7 @@ -record(k_seq, {anno=[],arg,body}). -record(k_put, {anno=[],arg,ret=[]}). -record(k_bif, {anno=[],op,args,ret=[]}). --record(k_test, {anno=[],op,args}). +-record(k_test, {anno=[],op,args,inverted=false}). -record(k_call, {anno=[],op,args,ret=[]}). -record(k_enter, {anno=[],op,args}). -record(k_receive, {anno=[],var,body,timeout,action,ret=[]}). @@ -66,6 +66,7 @@ -record(k_receive_next, {anno=[]}). -record(k_try, {anno=[],arg,vars,body,evars,handler,ret=[]}). -record(k_try_enter, {anno=[],arg,vars,body,evars,handler}). +-record(k_protected, {anno=[],arg,ret=[]}). -record(k_catch, {anno=[],body,ret=[]}). -record(k_guard_match, {anno=[],vars,body,ret=[]}). diff --git a/lib/compiler/src/v3_kernel_pp.erl b/lib/compiler/src/v3_kernel_pp.erl index 03b034ae98..c12c301ee2 100644 --- a/lib/compiler/src/v3_kernel_pp.erl +++ b/lib/compiler/src/v3_kernel_pp.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2011. All Rights Reserved. +%% Copyright Ericsson AB 1999-2018. All Rights Reserved. %% %% Licensed under the Apache License, Version 2.0 (the "License"); %% you may not use this file except in compliance with the License. @@ -47,7 +47,7 @@ canno(Cthing) -> element(2, Cthing). --spec format(cerl:cerl()) -> iolist(). +-spec format(#k_mdef{}) -> iolist(). format(Node) -> format(Node, #ctxt{}). @@ -145,7 +145,7 @@ format_1(#k_local{name=N,arity=A}, Ctxt) -> "local " ++ format_fa_pair({N,A}, Ctxt); format_1(#k_remote{mod=M,name=N,arity=A}, _Ctxt) -> %% This is for our internal translator. - io_lib:format("remote ~s:~s/~w", [format(M),format(N),A]); + io_lib:format("remote ~ts:~ts/~w", [format(M),format(N),A]); format_1(#k_internal{name=N,arity=A}, Ctxt) -> "internal " ++ format_fa_pair({N,A}, Ctxt); format_1(#k_seq{arg=A,body=B}, Ctxt) -> @@ -235,15 +235,20 @@ format_1(#k_bif{op=Op,args=As,ret=Rs}, Ctxt) -> [Txt,format_args(As, Ctxt1), format_ret(Rs, Ctxt1) ]; -format_1(#k_test{op=Op,args=As}, Ctxt) -> - Txt = ["test (",format(Op, ctxt_bump_indent(Ctxt, 6)),$)], +format_1(#k_test{op=Op,args=As,inverted=Inverted}, Ctxt) -> + Txt = case Inverted of + false -> + ["test (",format(Op, ctxt_bump_indent(Ctxt, 6)),$)]; + true -> + ["inverted_test (",format(Op, ctxt_bump_indent(Ctxt, 6)),$)] + end, Ctxt1 = ctxt_bump_indent(Ctxt, 2), [Txt,format_args(As, Ctxt1)]; format_1(#k_put{arg=A,ret=Rs}, Ctxt) -> [format(A, Ctxt), format_ret(Rs, ctxt_bump_indent(Ctxt, 1)) ]; -format_1(#k_try{arg=A,vars=Vs,body=B,evars=Evs,handler=H}, Ctxt) -> +format_1(#k_try{arg=A,vars=Vs,body=B,evars=Evs,handler=H,ret=Rs}, Ctxt) -> Ctxt1 = ctxt_bump_indent(Ctxt, Ctxt#ctxt.body_indent), ["try", nl_indent(Ctxt1), @@ -259,7 +264,8 @@ format_1(#k_try{arg=A,vars=Vs,body=B,evars=Evs,handler=H}, Ctxt) -> nl_indent(Ctxt1), format(H, Ctxt1), nl_indent(Ctxt), - "end" + "end", + format_ret(Rs, Ctxt) ]; format_1(#k_try_enter{arg=A,vars=Vs,body=B,evars=Evs,handler=H}, Ctxt) -> Ctxt1 = ctxt_bump_indent(Ctxt, Ctxt#ctxt.body_indent), @@ -279,6 +285,15 @@ format_1(#k_try_enter{arg=A,vars=Vs,body=B,evars=Evs,handler=H}, Ctxt) -> nl_indent(Ctxt), "end" ]; +format_1(#k_protected{arg=A,ret=Rs}, Ctxt) -> + Ctxt1 = ctxt_bump_indent(Ctxt, Ctxt#ctxt.body_indent), + ["protected", + nl_indent(Ctxt1), + format(A, Ctxt1), + nl_indent(Ctxt), + "end", + format_ret(Rs, ctxt_bump_indent(Ctxt, 1)) + ]; format_1(#k_catch{body=B,ret=Rs}, Ctxt) -> Ctxt1 = ctxt_bump_indent(Ctxt, Ctxt#ctxt.body_indent), ["catch", @@ -477,7 +492,7 @@ indent(Ctxt) -> indent(Ctxt#ctxt.indent, Ctxt). indent(N, _Ctxt) when N =< 0 -> ""; indent(N, Ctxt) -> T = Ctxt#ctxt.tab_width, - string:chars($\t, N div T, string:chars($\s, N rem T)). + lists:duplicate(N div T, $\t) ++ lists:duplicate(N rem T, $\s). nl_indent(Ctxt) -> [$\n|indent(Ctxt)]. @@ -494,7 +509,7 @@ unindent([$\t|T], N, Ctxt, C) -> if N >= Tab -> unindent(T, N - Tab, Ctxt, C); true -> - unindent([string:chars($\s, Tab - N)|T], 0, Ctxt, C) + unindent([lists:duplicate(Tab - N, $\s)|T], 0, Ctxt, C) end; unindent([L|T], N, Ctxt, C) when is_list(L) -> unindent(L, N, Ctxt, [T|C]); diff --git a/lib/compiler/src/v3_life.erl b/lib/compiler/src/v3_life.erl deleted file mode 100644 index fa057ae211..0000000000 --- a/lib/compiler/src/v3_life.erl +++ /dev/null @@ -1,484 +0,0 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 1999-2012. All Rights Reserved. -%% -%% Licensed under the Apache License, Version 2.0 (the "License"); -%% you may not use this file except in compliance with the License. -%% You may obtain a copy of the License at -%% -%% http://www.apache.org/licenses/LICENSE-2.0 -%% -%% Unless required by applicable law or agreed to in writing, software -%% distributed under the License is distributed on an "AS IS" BASIS, -%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -%% See the License for the specific language governing permissions and -%% limitations under the License. -%% -%% %CopyrightEnd% -%% -%% Purpose : Convert annotated kernel expressions to annotated beam format. - -%% This module creates beam format annotated with variable lifetime -%% information. Each thing is given an index and for each variable we -%% store the first and last index for its occurrence. The variable -%% database, VDB, attached to each thing is only relevant internally -%% for that thing. -%% -%% For nested things like matches the numbering continues locally and -%% the VDB for that thing refers to the variable usage within that -%% thing. Variables which live through a such a thing are internally -%% given a very large last index. Internally the indexes continue -%% after the index of that thing. This creates no problems as the -%% internal variable info never escapes and externally we only see -%% variable which are alive both before or after. -%% -%% This means that variables never "escape" from a thing and the only -%% way to get values from a thing is to "return" them, with 'break' or -%% 'return'. Externally these values become the return values of the -%% thing. This is no real limitation as most nested things have -%% multiple threads so working out a common best variable usage is -%% difficult. - --module(v3_life). - --export([module/2]). - --export([vdb_find/2]). - --import(lists, [member/2,map/2,reverse/1,sort/1]). --import(ordsets, [add_element/2,intersection/2,union/2]). - --include("v3_kernel.hrl"). --include("v3_life.hrl"). - -%% These are not defined in v3_kernel.hrl. -get_kanno(Kthing) -> element(2, Kthing). -%%set_kanno(Kthing, Anno) -> setelement(2, Kthing, Anno). - -module(#k_mdef{name=M,exports=Es,attributes=As,body=Fs0}, _Opts) -> - Fs1 = functions(Fs0, []), - {ok,{M,Es,As,Fs1}}. - -functions([F|Fs], Acc) -> - functions(Fs, [function(F)|Acc]); -functions([], Acc) -> reverse(Acc). - -%% function(Kfunc) -> Func. - -function(#k_fdef{anno=#k{a=Anno},func=F,arity=Ar,vars=Vs,body=Kb}) -> - try - As = var_list(Vs), - Vdb0 = init_vars(As), - %% Force a top-level match! - B0 = case Kb of - #k_match{} -> Kb; - _ -> - Ka = get_kanno(Kb), - #k_match{anno=#k{us=Ka#k.us,ns=[],a=Ka#k.a}, - vars=Vs,body=Kb,ret=[]} - end, - put(guard_refc, 0), - {B1,_,Vdb1} = body(B0, 1, Vdb0), - erase(guard_refc), - {function,F,Ar,As,B1,Vdb1,Anno} - catch - Class:Error -> - Stack = erlang:get_stacktrace(), - io:fwrite("Function: ~w/~w\n", [F,Ar]), - erlang:raise(Class, Error, Stack) - end. - -%% body(Kbody, I, Vdb) -> {[Expr],MaxI,Vdb}. -%% Handle a body. - -body(#k_seq{arg=Ke,body=Kb}, I, Vdb0) -> - %%ok = io:fwrite("life ~w:~p~n", [?LINE,{Ke,I,Vdb0}]), - A = get_kanno(Ke), - Vdb1 = use_vars(union(A#k.us, A#k.ns), I, Vdb0), - {Es,MaxI,Vdb2} = body(Kb, I+1, Vdb1), - E = expr(Ke, I, Vdb2), - {[E|Es],MaxI,Vdb2}; -body(Ke, I, Vdb0) -> - %%ok = io:fwrite("life ~w:~p~n", [?LINE,{Ke,I,Vdb0}]), - A = get_kanno(Ke), - Vdb1 = use_vars(union(A#k.us, A#k.ns), I, Vdb0), - E = expr(Ke, I, Vdb1), - {[E],I,Vdb1}. - -%% guard(Kguard, I, Vdb) -> Guard. - -guard(#k_try{anno=A,arg=Ts,vars=[#k_var{name=X}],body=#k_var{name=X}, - handler=#k_atom{val=false},ret=Rs}, I, Vdb) -> - %% Lock variables that are alive before try and used afterwards. - %% Don't lock variables that are only used inside the try expression. - Pdb0 = vdb_sub(I, I+1, Vdb), - {T,MaxI,Pdb1} = body(Ts, I+1, Pdb0), - Pdb2 = use_vars(A#k.ns, MaxI+1, Pdb1), %Save "return" values - #l{ke={protected,T,var_list(Rs)},i=I,a=A#k.a,vdb=Pdb2}. - -%% expr(Kexpr, I, Vdb) -> Expr. - -expr(#k_test{anno=A,op=Op,args=As}, I, _Vdb) -> - #l{ke={test,test_op(Op),atomic_list(As)},i=I,a=A#k.a}; -expr(#k_call{anno=A,op=Op,args=As,ret=Rs}, I, _Vdb) -> - #l{ke={call,call_op(Op),atomic_list(As),var_list(Rs)},i=I,a=A#k.a}; -expr(#k_enter{anno=A,op=Op,args=As}, I, _Vdb) -> - #l{ke={enter,call_op(Op),atomic_list(As)},i=I,a=A#k.a}; -expr(#k_bif{anno=A,op=Op,args=As,ret=Rs}, I, _Vdb) -> - Bif = k_bif(A, Op, As, Rs), - #l{ke=Bif,i=I,a=A#k.a}; -expr(#k_match{anno=A,body=Kb,ret=Rs}, I, Vdb) -> - %% Work out imported variables which need to be locked. - Mdb = vdb_sub(I, I+1, Vdb), - M = match(Kb, A#k.us, I+1, [], Mdb), - #l{ke={match,M,var_list(Rs)},i=I,vdb=use_vars(A#k.us, I+1, Mdb),a=A#k.a}; -expr(#k_guard_match{anno=A,body=Kb,ret=Rs}, I, Vdb) -> - %% Work out imported variables which need to be locked. - Mdb = vdb_sub(I, I+1, Vdb), - M = match(Kb, A#k.us, I+1, [], Mdb), - #l{ke={guard_match,M,var_list(Rs)},i=I,vdb=use_vars(A#k.us, I+1, Mdb),a=A#k.a}; -expr(#k_try{}=Try, I, Vdb) -> - case is_in_guard() of - false -> body_try(Try, I, Vdb); - true -> guard(Try, I, Vdb) - end; -expr(#k_try_enter{anno=A,arg=Ka,vars=Vs,body=Kb,evars=Evs,handler=Kh}, I, Vdb) -> - %% Lock variables that are alive before the catch and used afterwards. - %% Don't lock variables that are only used inside the try. - Tdb0 = vdb_sub(I, I+1, Vdb), - %% This is the tricky bit. Lock variables in Arg that are used in - %% the body and handler. Add try tag 'variable'. - Ab = get_kanno(Kb), - Ah = get_kanno(Kh), - Tdb1 = use_vars(union(Ab#k.us, Ah#k.us), I+3, Tdb0), - Tdb2 = vdb_sub(I, I+2, Tdb1), - Vnames = fun (Kvar) -> Kvar#k_var.name end, %Get the variable names - {Aes,_,Adb} = body(Ka, I+2, add_var({catch_tag,I+1}, I+1, 1000000, Tdb2)), - {Bes,_,Bdb} = body(Kb, I+4, new_vars(sort(map(Vnames, Vs)), I+3, Tdb2)), - {Hes,_,Hdb} = body(Kh, I+4, new_vars(sort(map(Vnames, Evs)), I+3, Tdb2)), - #l{ke={try_enter,#l{ke={block,Aes},i=I+1,vdb=Adb,a=[]}, - var_list(Vs),#l{ke={block,Bes},i=I+3,vdb=Bdb,a=[]}, - var_list(Evs),#l{ke={block,Hes},i=I+3,vdb=Hdb,a=[]}}, - i=I,vdb=Tdb1,a=A#k.a}; -expr(#k_catch{anno=A,body=Kb,ret=[R]}, I, Vdb) -> - %% Lock variables that are alive before the catch and used afterwards. - %% Don't lock variables that are only used inside the catch. - %% Add catch tag 'variable'. - Cdb0 = vdb_sub(I, I+1, Vdb), - {Es,_,Cdb1} = body(Kb, I+1, add_var({catch_tag,I}, I, locked, Cdb0)), - #l{ke={'catch',Es,variable(R)},i=I,vdb=Cdb1,a=A#k.a}; -expr(#k_receive{anno=A,var=V,body=Kb,timeout=T,action=Ka,ret=Rs}, I, Vdb) -> - %% Work out imported variables which need to be locked. - Rdb = vdb_sub(I, I+1, Vdb), - M = match(Kb, add_element(V#k_var.name, A#k.us), I+1, [], - new_vars([V#k_var.name], I, Rdb)), - {Tes,_,Adb} = body(Ka, I+1, Rdb), - #l{ke={receive_loop,atomic(T),variable(V),M, - #l{ke=Tes,i=I+1,vdb=Adb,a=[]},var_list(Rs)}, - i=I,vdb=use_vars(A#k.us, I+1, Vdb),a=A#k.a}; -expr(#k_receive_accept{anno=A}, I, _Vdb) -> - #l{ke=receive_accept,i=I,a=A#k.a}; -expr(#k_receive_next{anno=A}, I, _Vdb) -> - #l{ke=receive_next,i=I,a=A#k.a}; -expr(#k_put{anno=A,arg=Arg,ret=Rs}, I, _Vdb) -> - #l{ke={set,var_list(Rs),literal(Arg, [])},i=I,a=A#k.a}; -expr(#k_break{anno=A,args=As}, I, _Vdb) -> - #l{ke={break,atomic_list(As)},i=I,a=A#k.a}; -expr(#k_guard_break{anno=A,args=As}, I, Vdb) -> - Locked = [V || {V,_,_} <- Vdb], - #l{ke={guard_break,atomic_list(As),Locked},i=I,a=A#k.a}; -expr(#k_return{anno=A,args=As}, I, _Vdb) -> - #l{ke={return,atomic_list(As)},i=I,a=A#k.a}. - -body_try(#k_try{anno=A,arg=Ka,vars=Vs,body=Kb,evars=Evs,handler=Kh,ret=Rs}, - I, Vdb) -> - %% Lock variables that are alive before the catch and used afterwards. - %% Don't lock variables that are only used inside the try. - Tdb0 = vdb_sub(I, I+1, Vdb), - %% This is the tricky bit. Lock variables in Arg that are used in - %% the body and handler. Add try tag 'variable'. - Ab = get_kanno(Kb), - Ah = get_kanno(Kh), - Tdb1 = use_vars(union(Ab#k.us, Ah#k.us), I+3, Tdb0), - Tdb2 = vdb_sub(I, I+2, Tdb1), - Vnames = fun (Kvar) -> Kvar#k_var.name end, %Get the variable names - {Aes,_,Adb} = body(Ka, I+2, add_var({catch_tag,I+1}, I+1, locked, Tdb2)), - {Bes,_,Bdb} = body(Kb, I+4, new_vars(sort(map(Vnames, Vs)), I+3, Tdb2)), - {Hes,_,Hdb} = body(Kh, I+4, new_vars(sort(map(Vnames, Evs)), I+3, Tdb2)), - #l{ke={'try',#l{ke={block,Aes},i=I+1,vdb=Adb,a=[]}, - var_list(Vs),#l{ke={block,Bes},i=I+3,vdb=Bdb,a=[]}, - var_list(Evs),#l{ke={block,Hes},i=I+3,vdb=Hdb,a=[]}, - var_list(Rs)}, - i=I,vdb=Tdb1,a=A#k.a}. - -%% call_op(Op) -> Op. -%% bif_op(Op) -> Op. -%% test_op(Op) -> Op. -%% Do any necessary name translations here to munge into beam format. - -call_op(#k_local{name=N}) -> N; -call_op(#k_remote{mod=M,name=N}) -> {remote,atomic(M),atomic(N)}; -call_op(Other) -> variable(Other). - -bif_op(#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val=N}}) -> N; -bif_op(#k_internal{name=N}) -> N. - -test_op(#k_remote{mod=#k_atom{val=erlang},name=#k_atom{val=N}}) -> N. - -%% k_bif(Anno, Op, [Arg], [Ret], Vdb) -> Expr. -%% Build bifs, do special handling of internal some calls. - -k_bif(_A, #k_internal{name=dsetelement,arity=3}, As, []) -> - {bif,dsetelement,atomic_list(As),[]}; -k_bif(_A, #k_internal{name=bs_context_to_binary=Op,arity=1}, As, []) -> - {bif,Op,atomic_list(As),[]}; -k_bif(_A, #k_internal{name=bs_init_writable=Op,arity=1}, As, Rs) -> - {bif,Op,atomic_list(As),var_list(Rs)}; -k_bif(_A, #k_internal{name=make_fun}, - [#k_atom{val=Fun},#k_int{val=Arity}, - #k_int{val=Index},#k_int{val=Uniq}|Free], - Rs) -> - {bif,{make_fun,Fun,Arity,Index,Uniq},var_list(Free),var_list(Rs)}; -k_bif(_A, Op, As, Rs) -> - %% The general case. - Name = bif_op(Op), - Ar = length(As), - case is_gc_bif(Name, Ar) of - false -> - {bif,Name,atomic_list(As),var_list(Rs)}; - true -> - {gc_bif,Name,atomic_list(As),var_list(Rs)} - end. - -%% match(Kexpr, [LockVar], I, Vdb) -> Expr. -%% Convert match tree to old format. - -match(#k_alt{anno=A,first=Kf,then=Kt}, Ls, I, Ctxt, Vdb0) -> - Vdb1 = use_vars(union(A#k.us, Ls), I, Vdb0), - F = match(Kf, Ls, I+1, Ctxt, Vdb1), - T = match(Kt, Ls, I+1, Ctxt, Vdb1), - #l{ke={alt,F,T},i=I,vdb=Vdb1,a=A#k.a}; -match(#k_select{anno=A,var=V,types=Kts}, Ls0, I, Ctxt, Vdb0) -> - Vanno = get_kanno(V), - Ls1 = case member(no_usage, Vanno) of - false -> add_element(V#k_var.name, Ls0); - true -> Ls0 - end, - Anno = case member(reuse_for_context, Vanno) of - true -> [reuse_for_context|A#k.a]; - false -> A#k.a - end, - Vdb1 = use_vars(union(A#k.us, Ls1), I, Vdb0), - Ts = [type_clause(Tc, Ls1, I+1, Ctxt, Vdb1) || Tc <- Kts], - #l{ke={select,literal(V, Ctxt),Ts},i=I,vdb=Vdb1,a=Anno}; -match(#k_guard{anno=A,clauses=Kcs}, Ls, I, Ctxt, Vdb0) -> - Vdb1 = use_vars(union(A#k.us, Ls), I, Vdb0), - Cs = [guard_clause(G, Ls, I+1, Ctxt, Vdb1) || G <- Kcs], - #l{ke={guard,Cs},i=I,vdb=Vdb1,a=A#k.a}; -match(Other, Ls, I, _Ctxt, Vdb0) -> - Vdb1 = use_vars(Ls, I, Vdb0), - {B,_,Vdb2} = body(Other, I+1, Vdb1), - #l{ke={block,B},i=I,vdb=Vdb2,a=[]}. - -type_clause(#k_type_clause{anno=A,type=T,values=Kvs}, Ls, I, Ctxt, Vdb0) -> - %%ok = io:format("life ~w: ~p~n", [?LINE,{T,Kvs}]), - Vdb1 = use_vars(union(A#k.us, Ls), I+1, Vdb0), - Vs = [val_clause(Vc, Ls, I+1, Ctxt, Vdb1) || Vc <- Kvs], - #l{ke={type_clause,type(T),Vs},i=I,vdb=Vdb1,a=A#k.a}. - -val_clause(#k_val_clause{anno=A,val=V,body=Kb}, Ls0, I, Ctxt0, Vdb0) -> - New = (get_kanno(V))#k.ns, - Bus = (get_kanno(Kb))#k.us, - %%ok = io:format("Ls0 = ~p, Used=~p\n New=~p, Bus=~p\n", [Ls0,Used,New,Bus]), - Ls1 = union(intersection(New, Bus), Ls0), %Lock for safety - Vdb1 = use_vars(union(A#k.us, Ls1), I+1, new_vars(New, I, Vdb0)), - Ctxt = case V of - #k_binary{segs=#k_var{name=C0}} -> C0; - _ -> Ctxt0 - end, - B = match(Kb, Ls1, I+1, Ctxt, Vdb1), - #l{ke={val_clause,literal(V, Ctxt),B},i=I,vdb=use_vars(Bus, I+1, Vdb1),a=A#k.a}. - -guard_clause(#k_guard_clause{anno=A,guard=Kg,body=Kb}, Ls, I, Ctxt, Vdb0) -> - Vdb1 = use_vars(union(A#k.us, Ls), I+2, Vdb0), - Gdb = vdb_sub(I+1, I+2, Vdb1), - OldRefc = put(guard_refc, get(guard_refc)+1), - G = guard(Kg, I+1, Gdb), - put(guard_refc, OldRefc), - B = match(Kb, Ls, I+2, Ctxt, Vdb1), - #l{ke={guard_clause,G,B}, - i=I,vdb=use_vars((get_kanno(Kg))#k.us, I+2, Vdb1), - a=A#k.a}. - -%% type(Ktype) -> Type. - -type(k_literal) -> literal; -type(k_int) -> integer; -%%type(k_char) -> integer; %Hhhmmm??? -type(k_float) -> float; -type(k_atom) -> atom; -type(k_nil) -> nil; -type(k_cons) -> cons; -type(k_tuple) -> tuple; -type(k_binary) -> binary; -type(k_bin_seg) -> bin_seg; -type(k_bin_int) -> bin_int; -type(k_bin_end) -> bin_end; -type(k_map) -> map. - -%% variable(Klit) -> Lit. -%% var_list([Klit]) -> [Lit]. - -variable(#k_var{name=N}) -> {var,N}. - -var_list(Ks) -> [variable(K) || K <- Ks]. - -%% atomic(Klit) -> Lit. -%% atomic_list([Klit]) -> [Lit]. - -atomic(#k_literal{val=V}) -> {literal,V}; -atomic(#k_var{name=N}) -> {var,N}; -atomic(#k_int{val=I}) -> {integer,I}; -atomic(#k_float{val=F}) -> {float,F}; -atomic(#k_atom{val=N}) -> {atom,N}; -%%atomic(#k_char{val=C}) -> {char,C}; -atomic(#k_nil{}) -> nil. - -atomic_list(Ks) -> [atomic(K) || K <- Ks]. - -%% literal(Klit) -> Lit. -%% literal_list([Klit]) -> [Lit]. - -literal(#k_var{name=N}, _) -> {var,N}; -literal(#k_literal{val=I}, _) -> {literal,I}; -literal(#k_int{val=I}, _) -> {integer,I}; -literal(#k_float{val=F}, _) -> {float,F}; -literal(#k_atom{val=N}, _) -> {atom,N}; -%%literal(#k_char{val=C}, _) -> {char,C}; -literal(#k_nil{}, _) -> nil; -literal(#k_cons{hd=H,tl=T}, Ctxt) -> - {cons,[literal(H, Ctxt),literal(T, Ctxt)]}; -literal(#k_binary{segs=V}, Ctxt) -> - {binary,literal(V, Ctxt)}; -literal(#k_bin_seg{size=S,unit=U,type=T,flags=Fs,seg=Seg,next=[]}, Ctxt) -> - %% Only occurs in patterns. - {bin_seg,Ctxt,literal(S, Ctxt),U,T,Fs,[literal(Seg, Ctxt)]}; -literal(#k_bin_seg{size=S,unit=U,type=T,flags=Fs,seg=Seg,next=N}, Ctxt) -> - {bin_seg,Ctxt,literal(S, Ctxt),U,T,Fs, - [literal(Seg, Ctxt),literal(N, Ctxt)]}; -literal(#k_bin_int{size=S,unit=U,flags=Fs,val=Int,next=N}, Ctxt) -> - %% Only occurs in patterns. - {bin_int,Ctxt,literal(S, Ctxt),U,Fs,Int, - [literal(N, Ctxt)]}; -literal(#k_bin_end{}, Ctxt) -> - {bin_end,Ctxt}; -literal(#k_tuple{es=Es}, Ctxt) -> - {tuple,literal_list(Es, Ctxt)}; -literal(#k_map{op=Op,var=Var,es=Es0}, Ctxt) -> - {map,Op,literal(Var, Ctxt),literal_list(Es0, Ctxt)}; -literal(#k_map_pair{key=K,val=V}, Ctxt) -> - {map_pair,literal(K, Ctxt),literal(V, Ctxt)}. - -literal_list(Ks, Ctxt) -> - [literal(K, Ctxt) || K <- Ks]. - - -%% is_gc_bif(Name, Arity) -> true|false -%% Determines whether the BIF Name/Arity might do a GC. - -is_gc_bif(hd, 1) -> false; -is_gc_bif(tl, 1) -> false; -is_gc_bif(self, 0) -> false; -is_gc_bif(node, 0) -> false; -is_gc_bif(node, 1) -> false; -is_gc_bif(element, 2) -> false; -is_gc_bif(get, 1) -> false; -is_gc_bif(raise, 2) -> false; -is_gc_bif(tuple_size, 1) -> false; -is_gc_bif(Bif, Arity) -> - not (erl_internal:bool_op(Bif, Arity) orelse - erl_internal:new_type_test(Bif, Arity) orelse - erl_internal:comp_op(Bif, Arity)). - -%% Keep track of life time for variables. -%% -%% init_vars([{var,VarName}]) -> Vdb. -%% new_vars([VarName], I, Vdb) -> Vdb. -%% use_vars([VarName], I, Vdb) -> Vdb. -%% add_var(VarName, F, L, Vdb) -> Vdb. -%% -%% The list of variable names for new_vars/3 and use_vars/3 -%% must be sorted. - -init_vars(Vs) -> - vdb_new(Vs). - -new_vars([], _, Vdb) -> Vdb; -new_vars([V], I, Vdb) -> vdb_store_new(V, {V,I,I}, Vdb); -new_vars(Vs, I, Vdb) -> vdb_update_vars(Vs, Vdb, I). - -use_vars([], _, Vdb) -> - Vdb; -use_vars([V], I, Vdb) -> - case vdb_find(V, Vdb) of - {V,F,L} when I > L -> vdb_update(V, {V,F,I}, Vdb); - {V,_,_} -> Vdb; - error -> vdb_store_new(V, {V,I,I}, Vdb) - end; -use_vars(Vs, I, Vdb) -> vdb_update_vars(Vs, Vdb, I). - -add_var(V, F, L, Vdb) -> - vdb_store_new(V, {V,F,L}, Vdb). - -%% is_in_guard() -> true|false. - -is_in_guard() -> - get(guard_refc) > 0. - -%% vdb - -vdb_new(Vs) -> - sort([{V,0,0} || {var,V} <- Vs]). - -vdb_find(V, Vdb) -> - case lists:keyfind(V, 1, Vdb) of - false -> error; - Vd -> Vd - end. - -vdb_update(V, Update, [{V,_,_}|Vdb]) -> - [Update|Vdb]; -vdb_update(V, Update, [Vd|Vdb]) -> - [Vd|vdb_update(V, Update, Vdb)]. - -vdb_store_new(V, New, [{V1,_,_}=Vd|Vdb]) when V > V1 -> - [Vd|vdb_store_new(V, New, Vdb)]; -vdb_store_new(V, New, [{V1,_,_}|_]=Vdb) when V < V1 -> - [New|Vdb]; -vdb_store_new(_, New, []) -> [New]. - -vdb_update_vars([V|_]=Vs, [{V1,_,_}=Vd|Vdb], I) when V > V1 -> - [Vd|vdb_update_vars(Vs, Vdb, I)]; -vdb_update_vars([V|Vs], [{V1,_,_}|_]=Vdb, I) when V < V1 -> - %% New variable. - [{V,I,I}|vdb_update_vars(Vs, Vdb, I)]; -vdb_update_vars([V|Vs], [{_,F,L}=Vd|Vdb], I) -> - %% Existing variable. - if - I > L -> [{V,F,I}|vdb_update_vars(Vs, Vdb, I)]; - true -> [Vd|vdb_update_vars(Vs, Vdb, I)] - end; -vdb_update_vars([V|Vs], [], I) -> - %% New variable. - [{V,I,I}|vdb_update_vars(Vs, [], I)]; -vdb_update_vars([], Vdb, _) -> Vdb. - -%% vdb_sub(Min, Max, Vdb) -> Vdb. -%% Extract variables which are used before and after Min. Lock -%% variables alive after Max. - -vdb_sub(Min, Max, Vdb) -> - [ if L >= Max -> {V,F,locked}; - true -> Vd - end || {V,F,L}=Vd <- Vdb, F < Min, L >= Min ]. diff --git a/lib/compiler/src/v3_life.hrl b/lib/compiler/src/v3_life.hrl deleted file mode 100644 index 29a08f8c72..0000000000 --- a/lib/compiler/src/v3_life.hrl +++ /dev/null @@ -1,27 +0,0 @@ -%% -%% %CopyrightBegin% -%% -%% Copyright Ericsson AB 1999-2009. All Rights Reserved. -%% -%% Licensed under the Apache License, Version 2.0 (the "License"); -%% you may not use this file except in compliance with the License. -%% You may obtain a copy of the License at -%% -%% http://www.apache.org/licenses/LICENSE-2.0 -%% -%% Unless required by applicable law or agreed to in writing, software -%% distributed under the License is distributed on an "AS IS" BASIS, -%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -%% See the License for the specific language governing permissions and -%% limitations under the License. -%% -%% %CopyrightEnd% -%% -%% This record contains variable life-time annotation for a -%% kernel expression. Added by v3_life, used by v3_codegen. - --record(l, {ke, %Kernel expression - i=0, %Op number - vdb=[], %Variable database - a}). %Core annotation - |