diff options
Diffstat (limited to 'lib/compiler/src')
53 files changed, 2391 insertions, 1476 deletions
diff --git a/lib/compiler/src/Makefile b/lib/compiler/src/Makefile index 299b2892fc..518c89d044 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 1996-2016. 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. @@ -50,6 +50,7 @@ MODULES = \ beam_asm \ beam_block \ beam_bool \ + beam_bs \ beam_bsm \ beam_clean \ beam_dead \ @@ -62,6 +63,7 @@ MODULES = \ beam_opcodes \ beam_peep \ beam_receive \ + beam_reorder \ beam_split \ beam_trim \ beam_type \ diff --git a/lib/compiler/src/beam_a.erl b/lib/compiler/src/beam_a.erl index f0f2ee08c2..91e6d80da3 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 2012-2016. 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/beam_asm.erl b/lib/compiler/src/beam_asm.erl index a3201b0f4a..f6ca7a0afb 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 1996-2016. 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. @@ -30,11 +30,12 @@ module(Code, Abst, SourceFile, Opts) -> {ok,assemble(Code, Abst, SourceFile, Opts)}. -assemble({Mod,Exp,Attr0,Asm0,NumLabels}, Abst, SourceFile, Opts) -> +assemble({Mod,Exp0,Attr0,Asm0,NumLabels}, Abst, SourceFile, Opts) -> {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). @@ -61,7 +62,7 @@ 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 -> @@ -224,10 +225,12 @@ 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) -> + Misc0 = case SourceFile of + [] -> []; + [_|_] -> [{source,SourceFile}] + end, Misc = case member(slim, Opts) of - false -> - {{Y,Mo,D},{H,Mi,S}} = erlang:universaltime(), - [{time,{Y,Mo,D,H,Mi,S}},{source,SourceFile}]; + false -> Misc0; true -> [] end, Compile = [{options,Opts},{version,?COMPILER_VSN}|Misc], diff --git a/lib/compiler/src/beam_block.erl b/lib/compiler/src/beam_block.erl index 2def3de7f3..6a35191f6e 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 1999-2016. 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. @@ -23,14 +23,13 @@ -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,foldl/3,member/2]). -module({Mod,Exp,Attr,Fs0,Lc0}, _Opt) -> - {Fs,Lc} = mapfoldl(fun function/2, Lc0, Fs0), +module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> + Fs = [function(F) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. -function({function,Name,Arity,CLabel,Is0}, Lc0) -> +function({function,Name,Arity,CLabel,Is0}) -> try %% Collect basic blocks and optimize them. Is1 = blockify(Is0), @@ -40,11 +39,8 @@ function({function,Name,Arity,CLabel,Is0}, Lc0) -> 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} + {function,Name,Arity,CLabel,Is6} catch Class:Error -> Stack = erlang:get_stacktrace(), @@ -62,56 +58,15 @@ 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 +81,9 @@ 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,[]}}}; @@ -146,10 +103,10 @@ collect({get_list,S,D1,D2}) -> {set,[D1,D2],[S],get_list}; 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}; @@ -183,7 +140,9 @@ opt_blocks([I|Is]) -> opt_blocks([]) -> []. opt_block(Is0) -> - Is = find_fixpoint(fun opt/1, Is0), + Is = find_fixpoint(fun(Is) -> + opt_tuple_element(opt(Is)) + end, Is0), opt_alloc(Is). find_fixpoint(OptFun, Is0) -> @@ -193,14 +152,43 @@ 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: 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, in general it is not safe to transform the following +%% instructions: +%% +%% get_tuple_element x(0) Element => x(1) +%% allocate_zero StackSize 3 %x(0), x(1), x(2) are live %% -%% 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. +%% 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. Unfortunately, beam_reorder may have moved +%% a get_tuple_element instruction so that x(1) is not always +%% initialized when this code is reached. To find whether or not x(1) +%% is initialized, we would need to analyze all code preceding these +%% two instructions (across branches). Since we currently don't have +%% any practical mechanism for doing that, we will have to +%% conservatively assume that the transformation is unsafe. move_allocates([{block,Bl0}|Is]) -> Bl = move_allocates_1(reverse(Bl0), []), @@ -209,38 +197,26 @@ 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(Is, Acc); +move_allocates_1([I|Is], [{set,[],[],{alloc,Live0,Info}}|Acc]=Acc0) -> + case {alloc_may_pass(I),alloc_live_regs(I, Live0)} of + {false,_} -> + move_allocates_1(Is, [I|Acc0]); + {true,not_possible} -> + move_allocates_1(Is, [I|Acc0]); + {true,Live} when is_integer(Live) -> + A = {set,[],[],{alloc,Live,Info}}, + move_allocates_1(Is, [A,I|Acc]) + end; 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) -> - 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]) - end; -move_allocates_2(Alloc, [], Acc) -> - {[],[{set,[],[],Alloc}|Acc]}. +move_allocates_1([], Acc) -> Acc. 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}. - %% opt([Instruction]) -> [Instruction] %% Optimize the instruction stream inside a basic block. @@ -251,7 +227,7 @@ opt([{set,_,_,{line,_}}=Line1, {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,Ds0,Ss,Op}|Is0]) -> +opt([{set,Ds0,Ss,Op}|Is0]) -> {Ds,Is} = opt_moves(Ds0, Is0), [{set,Ds,Ss,Op}|opt(Is)]; opt([{'%live',_,_}=I|Is]) -> @@ -277,78 +253,167 @@ opt_moves([X0,Y0], Is0) -> 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 +%% -%% Reverse the instructions, while checking that there are no instructions that -%% would interfere with using the new destination register chosen. +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,Is} -> + {set,[S],Ss,Op} = I0, + I = {set,[D],Ss,Op}, + {yes,reverse(Acc, [I|Is])} + 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. -opt_move_2(D, [I|Is], Acc) -> - case is_transparent(D, I) of - false -> not_possible; - true -> opt_move_2(D, Is, [I|Acc]) +%% Reverse the instructions, while checking that there are no +%% instructions that would interfere with using the new destination +%% register (D). + +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([{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) -> + 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 -> + eliminate_use_of_from_reg(Is, From, To, [I|Acc]) 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. %% opt_alloc(Instructions) -> Instructions' %% Optimises all allocate instructions. +opt_alloc([{set,[],[],{alloc,Live0,Info0}}, + {set,[],[],{alloc,Live,Info}}|Is]) -> + Live = Live0, %Assertion. + Alloc = combine_alloc(Info0, Info), + I = {set,[],[],{alloc,Live,Alloc}}, + opt_alloc([I|Is]); opt_alloc([{set,[],[],{alloc,R,{_,Ns,Nh,[]}}}|Is]) -> - [{set,[],[],opt_alloc(Is, Ns, Nh, R)}|opt(Is)]; + [{set,[],[],opt_alloc(Is, Ns, Nh, R)}|Is]; opt_alloc([I|Is]) -> [I|opt_alloc(Is)]; 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 @@ -397,13 +462,14 @@ count_ones(Bits, Acc) -> alloc_live_regs({set,Ds,Ss,_}, Regs0) -> Rset = x_live(Ss, x_dead(Ds, (1 bsl Regs0)-1)), - live_regs(Rset). - -live_regs(Regs) -> - live_regs_1(0, Regs). + live_regs(0, Rset). -live_regs_1(N, 0) -> N; -live_regs_1(N, Regs) -> live_regs_1(N+1, Regs bsr 1). +live_regs(N, 0) -> + N; +live_regs(N, Regs) when Regs band 1 =:= 1 -> + live_regs(N+1, Regs bsr 1); +live_regs(_, _) -> + not_possible. x_dead([{x,N}|Rs], Regs) -> x_dead(Rs, Regs band (bnot (1 bsl N))); x_dead([_|Rs], Regs) -> x_dead(Rs, Regs); @@ -412,234 +478,3 @@ x_dead([], Regs) -> Regs. 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. - - -%%% -%%% 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. -%%% - -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) - 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_bool.erl b/lib/compiler/src/beam_bool.erl index 14b6381230..99e4ccb1e9 100644 --- a/lib/compiler/src/beam_bool.erl +++ b/lib/compiler/src/beam_bool.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2004-2013. All Rights Reserved. +%% Copyright Ericsson AB 2004-2016. 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. @@ -25,8 +25,6 @@ -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. @@ -142,11 +140,6 @@ bopt_block(Reg, Fail, OldIs, [{block,Bl0}|Acc0], St0) -> throw:not_boolean_expr -> failed; - %% The block contains a 'move' instruction that could - %% not be handled. - throw:move -> - failed; - %% The optimization is not safe. (A register %% used by the instructions following the %% optimized code is either not assigned a @@ -215,37 +208,14 @@ ensure_opt_safe(Bl, NewCode, OldIs, Fail, PrecedingCode, St) -> false -> throw(all_registers_not_killed); true -> ok end, - Same = assigned_same_value(Bl, NewCode), MustBeUnused = ordsets:subtract(ordsets:union(NotSet, NewDst), - ordsets:union(MustBeKilled, Same)), + MustBeKilled), case none_used(MustBeUnused, OldIs, Fail, St) of false -> throw(registers_used); true -> ok end, ok. -%% assigned_same_value(OldCode, NewCodeReversed) -> [DestinationRegs] -%% Return an ordset with a list of all y registers that are always -%% assigned the same value in the old and new code. Currently, we -%% are very conservative in that we only consider identical move -%% instructions in the same order. -%% -assigned_same_value(Old, New) -> - case reverse(New) of - [{block,Bl}|_] -> - assigned_same_value(Old, Bl, []); - _ -> - ordsets:new() - end. - -assigned_same_value([{set,[{y,_}=D],[S],move}|T1], - [{set,[{y,_}=D],[S],move}|T2], Acc) -> - assigned_same_value(T1, T2, [D|Acc]); -assigned_same_value(_, _, Acc) -> - ordsets:from_list(Acc). - -update_fail_label([{set,_,_,move}=I|Is], Fail, Acc) -> - update_fail_label(Is, Fail, [I|Acc]); 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) -> @@ -268,9 +238,9 @@ extend_block(BlAcc0, Fail, [{block,Is0}|OldAcc]) -> end; extend_block(BlAcc, _, OldAcc) -> {BlAcc,OldAcc}. -extend_block_1([{set,[_],_,{bif,_,{f,Fail}}}=I|Is], Fail, Acc) -> +extend_block_1([{set,[{x,_}],_,{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) -> +extend_block_1([{set,[{x,_}],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]) @@ -314,8 +284,6 @@ split_block_1(Is, Fail, ProhibitFailLabel) -> end end. -split_block_2([{set,_,_,move}=I|Is], Fail, Acc) -> - split_block_2(Is, Fail, [I|Acc]); 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) -> @@ -343,8 +311,8 @@ 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([{set,[D],_,move}|Is], Acc) -> - dst_regs(Is, [D|Acc]); +dst_regs([{protected,_,Bl,_}|Is], Acc) -> + dst_regs(Bl, dst_regs(Is, Acc)); dst_regs([_|Is], Acc) -> dst_regs(Is, Acc); dst_regs([], Acc) -> ordsets:from_list(Acc). @@ -411,13 +379,6 @@ bopt_tree([{protected,[Dst],Code,_}|Is], Forest0, Pre) -> _Res -> throw(not_boolean_expr) end; -bopt_tree([{set,[Dst],[Src],move}=Move|Is], Forest, Pre) -> - case {Src,Dst} of - {{tmp,_},_} -> throw(move); - {_,{tmp,_}} -> throw(move); - _ -> ok - end, - bopt_tree(Is, Forest, [Move|Pre]); bopt_tree([{set,[Dst],As,{bif,N,_}}=Bif|Is], Forest0, Pre) -> Ar = length(As), case safe_bool_op(N, Ar) of @@ -589,10 +550,6 @@ free_variables(Is) -> E = gb_sets:empty(), free_vars_1(Is, E, E, E). -free_vars_1([{set,Ds,As,move}|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,{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)), @@ -632,8 +589,6 @@ free_vars_regs(X) -> [{x,X-1}|free_vars_regs(X-1)]. rename_regs(Is, Regs) -> rename_regs(Is, Regs, []). -rename_regs([{set,_,_,move}=I|Is], Regs, Acc) -> - rename_regs(Is, Regs, [I|Acc]); rename_regs([{set,[Dst0],Ss0,{alloc,_,Info}}|Is], Regs0, Acc) -> Live = live_regs(Regs0), Ss = rename_sources(Ss0, Regs0), @@ -737,8 +692,7 @@ ssa_assign({x,_}=R, #ssa{sub=Sub0}=Ssa0) -> Sub1 = gb_trees:update(R, NewReg, Sub0), Sub = gb_trees:insert(NewReg, NewReg, Sub1), Ssa#ssa{sub=Sub} - end; -ssa_assign(_, Ssa) -> Ssa. + end. ssa_sub_list(List, Sub) -> [ssa_sub(E, Sub) || E <- List]. diff --git a/lib/compiler/src/beam_bs.erl b/lib/compiler/src/beam_bs.erl new file mode 100644 index 0000000000..2aed98d4e7 --- /dev/null +++ b/lib/compiler/src/beam_bs.erl @@ -0,0 +1,278 @@ +%% +%% %CopyrightBegin% +%% +%% 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. +%% 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]). + +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 = erlang:get_stacktrace(), + 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 4f76350269..ae1b34ba49 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 2007-2016. 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. @@ -205,8 +205,15 @@ 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 +367,10 @@ 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 +380,15 @@ 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 @@ -421,7 +432,8 @@ btb_follow_branches([], _, D) -> D. btb_follow_branch(0, _Regs, D) -> D; btb_follow_branch(Lbl, Regs, #btb{ok_br=Br0,index=Li}=D) -> - case gb_sets:is_member(Lbl, Br0) of + Key = {Lbl,Regs}, + case gb_sets:is_member(Key, Br0) of true -> %% We have already followed this branch and it was OK. D; @@ -432,7 +444,7 @@ btb_follow_branch(Lbl, Regs, #btb{ok_br=Br0,index=Li}=D) -> btb_reaches_match_1(Is, Regs, D), %% Since we got back, this branch is OK. - D#btb{ok_br=gb_sets:insert(Lbl, Br),must_not_save=MustNotSave, + D#btb{ok_br=gb_sets:insert(Key, Br),must_not_save=MustNotSave, must_save=MustSave} end. diff --git a/lib/compiler/src/beam_clean.erl b/lib/compiler/src/beam_clean.erl index 919ee3ee7d..10805a3c36 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 2000-2016. 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. @@ -141,7 +141,7 @@ 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; @@ -190,17 +190,11 @@ replace([{test,Test,{f,Lbl},Ops}|Is], 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), + Vls = 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(Is, [{select,I,R,{f,Fail},Vls}|Acc], D); 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) -> @@ -241,12 +235,6 @@ label(Old, D) -> {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 diff --git a/lib/compiler/src/beam_dead.erl b/lib/compiler/src/beam_dead.erl index ead88b57e9..6f6d742293 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 2002-2016. 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. @@ -239,15 +239,30 @@ 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; @@ -295,7 +310,28 @@ 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), @@ -348,6 +384,12 @@ 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: %% @@ -375,7 +417,7 @@ comp_op_find_shortcut(To0, Reg, Val, D) -> 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 @@ -818,3 +860,17 @@ 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..9565ab74c4 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. @@ -44,7 +44,7 @@ locals = [] :: [{label(), arity(), label()}], imports = gb_trees:empty() :: import_tab(), strings = <<>> :: binary(), %String pool - lambdas = [], %[{...}] + lambdas = {0,[]}, %[{...}] literals = dict:new() :: literal_tab(), fnames = #{} :: fname_tab(), lines = #{} :: line_tab(), @@ -145,15 +145,14 @@ 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}}. + {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'} @@ -236,13 +235,13 @@ 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), 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}. + {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..c699672db1 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 2000-2016. 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. @@ -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, diff --git a/lib/compiler/src/beam_disasm.hrl b/lib/compiler/src/beam_disasm.hrl index 30e3f22665..d968cd9587 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 2007-2016. 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. @@ -22,7 +22,9 @@ %% 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' | 'remove_message' | 'return' | 'send' | 'timeout' @@ -34,7 +36,7 @@ -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..4a181c1923 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 2011-2016. 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. @@ -133,10 +133,12 @@ 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..c9ff07b496 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 1999-2016. 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. @@ -64,7 +64,6 @@ 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|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}}. diff --git a/lib/compiler/src/beam_jump.erl b/lib/compiler/src/beam_jump.erl index 5e58e0f6ac..48b5a32814 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 1999-2016. 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. @@ -167,12 +167,18 @@ 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 +188,18 @@ 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]) -> @@ -266,17 +272,17 @@ 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, %Entry label (must not be moved). + mlbl, %Moved labels. + labels :: cerl_sets:set() %Set of referenced labels. + }). -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,mlbl=#{},labels=Lbls}, opt(Is, [], St) end, Is0). @@ -327,7 +333,8 @@ opt([{label,Lbl}=I|Is], Acc, #st{mlbl=Mlbl}=St0) -> %% 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) + error -> + opt(Is, [I|Acc], St0) end; opt([{jump,{f,_}=X}|[{label,_},{jump,X}|_]=Is], Acc, St) -> opt(Is, Acc, St); @@ -362,12 +369,19 @@ 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}) -> +opt([], Acc, #st{mlbl=Mlbl}) -> Code = reverse(Acc), - case maps:find(Fc, Mlbl) of - {ok,Lbls} -> insert_fc_labels(Lbls, Mlbl, Code); - error -> Code - end. + insert_fc_labels(Code, Mlbl). + +insert_fc_labels([{label,L}=I|Is0], Mlbl) -> + case maps:find(L, Mlbl) of + error -> + [I|insert_fc_labels(Is0, Mlbl)]; + {ok,Lbls} -> + Is = [{label,Lb} || Lb <- Lbls] ++ Is0, + [I|insert_fc_labels(Is, maps:remove(L, Mlbl))] + end; +insert_fc_labels([_|_]=Is, _) -> Is. maps_append_list(K,Vs,M) -> case M of @@ -375,16 +389,6 @@ maps_append_list(K,Vs,M) -> _ -> 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) - end; -insert_fc_labels([], _, Acc) -> Acc. - collect_labels(Is, #st{entry=Entry}) -> collect_labels_1(Is, Entry, []). @@ -495,7 +499,7 @@ is_label_used_in_block({set,_,_,Info}, 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; + {try_catch,_,{f,F}} -> F =:= Lbl; {alloc,_,_} -> false; {put_tuple,_} -> false; {get_tuple_element,_} -> false; diff --git a/lib/compiler/src/beam_listing.erl b/lib/compiler/src/beam_listing.erl index fd5c0a042b..ce566373bb 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 1997-2016. 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/beam_peep.erl b/lib/compiler/src/beam_peep.erl index 17fd2e502a..c8bef31824 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 2008-2016. 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. @@ -65,18 +65,6 @@ 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(), []). @@ -95,12 +83,19 @@ 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], _, Acc) -> + case prune_redundant_values(Vls0, F) of + [] -> + %% No values left. Must convert to plain jump. + I = {jump,F}, + peep(Is, gb_sets:empty(), [I|Acc]); + [_|_]=Vls -> + I = {select,Op,R,F,Vls}, + peep(Is, gb_sets:empty(), [I|Acc]) + end; peep([{test,Op,_,Ops}=I|Is], SeenTests0, Acc) -> case beam_utils:is_pure_test(I) of false -> @@ -121,16 +116,6 @@ 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 +140,9 @@ 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..89cafe27ce 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 2010-2016. 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. @@ -177,7 +177,8 @@ 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}; diff --git a/lib/compiler/src/beam_reorder.erl b/lib/compiler/src/beam_reorder.erl new file mode 100644 index 0000000000..6a7c033ec6 --- /dev/null +++ b/lib/compiler/src/beam_reorder.erl @@ -0,0 +1,148 @@ +%% +%% %CopyrightBegin% +%% +%% 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. +%% 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]). + +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 = erlang:get_stacktrace(), + 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..feeab0af50 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 2011-2016. 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. @@ -47,6 +47,8 @@ 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,[R],As,{bif,raise,{f,_}=Fail}}|Is], Bl, Acc) -> + split_block(Is, [], [{bif,raise,Fail,As,R}|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)]); @@ -54,11 +56,8 @@ 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,[R],[],{try_catch,Op,L}}|Is], Bl, Acc) -> + split_block(Is, [], [{Op,R,L}|make_block(Bl, Acc)]); split_block([{set,[],[],{line,_}=Line}|Is], Bl, Acc) -> split_block(Is, [], [Line|make_block(Bl, Acc)]); split_block([I|Is], Bl, Acc) -> diff --git a/lib/compiler/src/beam_trim.erl b/lib/compiler/src/beam_trim.erl index 509e013b62..a8dc6805bc 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. diff --git a/lib/compiler/src/beam_type.erl b/lib/compiler/src/beam_type.erl index 5298589f83..acaf3ede66 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 1999-2016. 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. @@ -23,7 +23,8 @@ -export([module/2]). --import(lists, [foldl/3,reverse/1,filter/2]). +-import(lists, [filter/2,foldl/3,keyfind/3,member/2, + reverse/1,reverse/2,sort/1]). module({Mod,Exp,Attr,Fs0,Lc}, _Opts) -> Fs = [function(F) || F <- Fs0], @@ -92,8 +93,19 @@ simplify_basic_1([{set,[D],[TupleReg],{get_tuple_element,0}}=I|Is0], Ts0, Acc) - Ts = update(I, Ts0), simplify_basic_1(Is0, Ts, [I|Acc]) end; -simplify_basic_1([{set,_,_,{'catch',_}}=I|Is], _Ts, Acc) -> +simplify_basic_1([{set,_,_,{try_catch,_,_}}=I|Is], _Ts, Acc) -> simplify_basic_1(Is, tdb_new(), [I|Acc]); +simplify_basic_1([{test,is_atom,_,[R]}=I|Is], Ts, Acc) -> + case tdb_find(R, Ts) of + boolean -> simplify_basic_1(Is, Ts, Acc); + _ -> simplify_basic_1(Is, Ts, [I|Acc]) + end; +simplify_basic_1([{test,is_integer,_,[R]}=I|Is], Ts, Acc) -> + case tdb_find(R, Ts) of + integer -> simplify_basic_1(Is, Ts, Acc); + {integer,_} -> simplify_basic_1(Is, Ts, Acc); + _ -> simplify_basic_1(Is, Ts, [I|Acc]) + end; simplify_basic_1([{test,is_tuple,_,[R]}=I|Is], Ts, Acc) -> case tdb_find(R, Ts) of {tuple,_,_} -> simplify_basic_1(Is, Ts, Acc); @@ -137,6 +149,16 @@ simplify_basic_1([{test,is_record,_,[R,{atom,_}=Tag,{integer,Arity}]}=I|Is], Ts0 Ts = update(I, Ts0), simplify_basic_1(Is, Ts, [I|Acc]) end; +simplify_basic_1([{select,select_val,Reg,_,_}=I0|Is], Ts, Acc) -> + I = case tdb_find(Reg, Ts) of + {integer,Range} -> + simplify_select_val_int(I0, Range); + boolean -> + simplify_select_val_bool(I0); + _ -> + I0 + end, + simplify_basic_1(Is, tdb_new(), [I|Acc]); simplify_basic_1([I|Is], Ts0, Acc) -> Ts = update(I, Ts0), simplify_basic_1(Is, Ts, [I|Acc]); @@ -144,6 +166,32 @@ simplify_basic_1([], Ts, Acc) -> Is = reverse(Acc), {Is,Ts}. +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. + %% simplify_float([Instruction], TypeDatabase) -> %% {[Instruction],TypeDatabase'} | not_possible %% Simplify floating point operations in blocks. @@ -199,7 +247,7 @@ simplify_float_1([{set,[D0],[A0,B0],{alloc,_,{gc_bif,Op0,{f,0}}}}=I|Is]=Is0, Ts = tdb_update([{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) -> @@ -311,7 +359,7 @@ 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) -> +flt_need_heap_2({set,_,_,{try_catch,_,_}}, H, Fl) -> {[],H,Fl}; %% All other instructions should cause the insertion of an allocation %% instruction if needed. @@ -382,6 +430,17 @@ 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],Args,{bif,N,_}}, Ts0) -> + 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), + case BoolOp of + true -> + tdb_update([{D,boolean}], Ts0); + false -> + tdb_update([{D,kill}], Ts0) + end; update({set,[D],[S],{get_tuple_element,0}}, Ts) -> tdb_update([{D,{tuple_element,S,0}}], Ts); update({set,[D],[S],{alloc,_,{gc_bif,float,{f,0}}}}, Ts0) -> @@ -390,6 +449,13 @@ update({set,[D],[S],{alloc,_,{gc_bif,float,{f,0}}}}, Ts0) -> true -> tdb_update([{D,float}], Ts0); false -> Ts0 end; +update({set,[D],[S1,S2],{alloc,_,{gc_bif,'band',{f,0}}}}, Ts) -> + case keyfind(integer, 1, [S1,S2]) of + {integer,N} -> + update_band(N, D, Ts); + false -> + tdb_update([{D,integer}], Ts) + end; update({set,[D],[S1,S2],{alloc,_,{gc_bif,'/',{f,0}}}}, Ts0) -> %% Make sure we reject non-numeric literals. case possibly_numeric(S1) andalso possibly_numeric(S2) of @@ -397,15 +463,17 @@ update({set,[D],[S1,S2],{alloc,_,{gc_bif,'/',{f,0}}}}, Ts0) -> false -> Ts0 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 op_type(Op) of + integer -> + tdb_update([{D,integer}], Ts0); + {float,_} -> 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; + unknown -> + tdb_update([{D,kill}], Ts0) end; update({set,[],_Src,_Op}, Ts0) -> Ts0; update({set,[D],_Src,_Op}, Ts0) -> @@ -437,26 +505,72 @@ update({test,is_record,_Fail,[Src,Tag,{integer,Arity}]}, Ts) -> tdb_update([{Src,{tuple,Arity,[Tag]}}], Ts); update({test,_Test,_Fail,_Other}, Ts) -> Ts; +update({test,bs_get_integer2,_,_,Args,Dst}, Ts) -> + tdb_update([{Dst,get_bs_integer_type(Args)}], Ts); update({call_ext,Ar,{extfunc,math,Math,Ar}}, Ts) -> case is_math_bif(Math, Ar) of true -> tdb_update([{{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,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,Sz,[]} + end, + tdb_update([{{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({line,_}, Ts) -> Ts; +update({bs_save2,_,_}, Ts) -> Ts; +update({bs_restore2,_,_}, Ts) -> Ts; %% The instruction is unknown. Kill all information. update(_I, _Ts) -> tdb_new(). +update_band(N, Reg, Ts) -> + Type = update_band_1(N, 0), + tdb_update([{Reg,Type}], Ts). + +update_band_1(N, Bits) when Bits < 64 -> + case 1 bsl Bits of + P when P =:= N + 1 -> + {integer,{0,N}}; + P when P > N + 1 -> + integer; + _ -> + update_band_1(N, Bits+1) + end; +update_band_1(_, _) -> + %% 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; @@ -545,11 +659,22 @@ 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(_) -> unknown. flush(Rs, [{set,[_],[],{put_tuple,_}}|_]=Is0, Acc0) -> Acc = flush_all(Rs, Is0, Acc0), @@ -618,7 +743,6 @@ 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); @@ -635,11 +759,14 @@ checkerror_2(OrigIs) -> [{set,[],[],fcheckerror}|OrigIs]. %%% %%% {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 +%%% 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. +%%% +%%% 'integer' or {integer,{Min,Max}} that the register contains an +%%% integer. %% tdb_new() -> EmptyDataBase %% Creates a new, empty type database. @@ -669,21 +796,45 @@ tdb_copy({Tag,_}=S, D, Ts) when Tag =:= x; Tag =:= y -> error -> orddict:erase(D, Ts); Type -> orddict:store(D, Type, Ts) end; -tdb_copy(Literal, D, Ts) -> orddict:store(D, Literal, Ts). +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,tuple_size(Tuple),[Lit]}; + _ -> term + end, + if + Type =:= term -> + orddict:erase(D, Ts); + true -> + verify_type(Type), + orddict:store(D, Type, Ts) + 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_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 +%% 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. +%% the '{tuple,5,[]}' information ignored. %% %% 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}. +%% For instance, {tuple,5,_} and {tuple,10,_} will be merged to produce +%% {tuple,10,_}. tdb_update(Uis0, Ts0) -> Uis1 = filter(fun ({{x,_},_Op}) -> true; @@ -694,7 +845,8 @@ tdb_update(Uis0, 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 -> +tdb_update1([{Key,Type}=New|Ops], [{K,_Old}|_]=Db) when Key < K -> + verify_type(Type), [New|tdb_update1(Ops, Db)]; tdb_update1([{Key,kill}|Ops], [{Key,_}|Db]) -> tdb_update1(remove_key(Key, Ops), Db); @@ -704,7 +856,8 @@ 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], []) -> +tdb_update1([{_,Type}=New|Ops], []) -> + verify_type(Type), [New|tdb_update1(Ops, [])]; tdb_update1([], Db) -> Db. @@ -729,10 +882,21 @@ 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(integer, {integer,_}=Int) -> + Int; +merge_type_info({integer,_}=Int, integer) -> + Int; +merge_type_info({integer,{Min1,Max1}}, {integer,{Min2,Max2}}) -> + {integer,{max(Min1, Min2),min(Max1, Max2)}}; merge_type_info(NewType, _) -> verify_type(NewType), NewType. +verify_type({atom,_}) -> ok; +verify_type(boolean) -> ok; +verify_type(integer) -> ok; +verify_type({integer,{Min,Max}}) + when is_integer(Min), is_integer(Max) -> ok; verify_type(map) -> ok; verify_type(nonempty_list) -> ok; verify_type({tuple,Sz,[]}) when is_integer(Sz) -> ok; diff --git a/lib/compiler/src/beam_utils.erl b/lib/compiler/src/beam_utils.erl index fbcd5de1bb..249d9395ca 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 2007-2016. 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. @@ -25,9 +25,8 @@ is_not_used/3,is_not_used_at/3, empty_label_index/0,index_label/3,index_labels/1, code_at/2,bif_to_test/3,is_pure_test/1, - live_opt/1,delete_live_annos/1,combine_heap_needs/2]). - --export([join_even/2,split_even/1]). + live_opt/1,delete_live_annos/1,combine_heap_needs/2, + split_even/1]). -import(lists, [member/2,sort/1,reverse/1,splitwith/2]). @@ -67,8 +66,7 @@ is_killed(R, Is, D) -> St = #live{bl=check_killed_block_fun(),lbl=D,res=gb_trees:empty()}, case check_liveness(R, Is, St) of {killed,_} -> true; - {used,_} -> false; - {unknown,_} -> false + {used,_} -> false end. %% is_killed_at(Reg, Lbl, State) -> true|false @@ -78,8 +76,7 @@ is_killed_at(R, Lbl, D) when is_integer(Lbl) -> St0 = #live{bl=check_killed_block_fun(),lbl=D,res=gb_trees:empty()}, case check_liveness_at(R, Lbl, St0) of {killed,_} -> true; - {used,_} -> false; - {unknown,_} -> false + {used,_} -> false end. %% is_not_used(Register, [Instruction], State) -> true|false @@ -93,8 +90,7 @@ is_not_used(R, Is, D) -> St = #live{bl=fun check_used_block/3,lbl=D,res=gb_trees:empty()}, case check_liveness(R, Is, St) of {killed,_} -> true; - {used,_} -> false; - {unknown,_} -> false + {used,_} -> false end. %% is_not_used(Register, [Instruction], State) -> true|false @@ -108,8 +104,7 @@ 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 + {used,_} -> false end. %% index_labels(FunctionIs) -> State @@ -137,10 +132,7 @@ index_label(Lbl, Is0, Acc) -> %% Retrieve the code at the given label. code_at(L, Ll) -> - case gb_trees:lookup(L, Ll) of - {value,Code} -> Code; - none -> none - end. + gb_trees:get(L, Ll). %% bif_to_test(Bif, [Op], Fail) -> {test,Test,Fail,[Op]} %% Convert a BIF to a test. Fail if not possible. @@ -164,10 +156,10 @@ 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('==', [_,_]=Ops, Fail) -> {test,is_eq,Fail,Ops}; 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('=:=', [_,_]=Ops, Fail) -> {test,is_eq_exact,Fail,Ops}; bif_to_test('=/=', [_,_]=Ops, Fail) -> {test,is_ne_exact,Fail,Ops}; bif_to_test(is_record, [_,_,_]=Ops, Fail) -> {test,is_record,Fail,Ops}. @@ -175,8 +167,7 @@ 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. %% is_pure_test({test,is_eq,_,[_,_]}) -> true; is_pure_test({test,is_ne,_,[_,_]}) -> true; @@ -188,6 +179,8 @@ 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)). @@ -235,21 +228,23 @@ combine_heap_needs(Words, {alloc,Alloc}) when is_integer(Words) -> combine_heap_needs(H1, H2) when is_integer(H1), is_integer(H2) -> H1+H2. +%% split_even/1 +%% [1,2,3,4,5,6] -> {[1,3,5],[2,4,6]} + +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(Reg, [Instruction], #live{}) -> +%% {killed | used, #live{}} +%% Find out whether Reg is used or killed in instruction sequence. +%% 'killed' means that Reg is assigned a new value or killed by an +%% allocation instruction. 'used' means that Reg is used in some way. -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); @@ -325,8 +320,11 @@ 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}; @@ -461,8 +459,9 @@ 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); @@ -484,13 +483,30 @@ check_liveness(R, [{get_map_elements,{f,Fail},S,{list,L}}|Is], St0) -> Other end end; +check_liveness(R, [{put_map,{f,_},_,Src,_D,Live,{list,_}}|_], St0) -> + case R of + Src -> + {used,St0}; + {x,X} when X < Live -> + {used,St0}; + {x,_} -> + {killed,St0}; + {y,_} -> + %% Conservatively mark it as used. + {used,St0} + end; +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_list,S,D1,D2}|Is], St) -> + I = {block,[{set,[D1,D2],[S],get_list}]}, + check_liveness(R, [I|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}. + %% Not implemented. Conservatively assume that the register is used. + {used,St}. check_liveness_everywhere(R, [{f,Lbl}|T], St0) -> case check_liveness_at(R, Lbl, St0) of @@ -509,7 +525,7 @@ 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. @@ -517,14 +533,6 @@ check_liveness_at(R, Lbl, #live{lbl=Ll,res=ResMemorized}=St0) -> 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}. - check_liveness_fail(_, _, _, 0, St) -> {killed,St}; check_liveness_fail(R, Op, Args, Fail, St) -> @@ -561,9 +569,9 @@ check_killed_block(R, [{set,Ds,Ss,_Op}|Is]) -> false -> check_killed_block(R, Is) end end; -check_killed_block(R, [{'%live',Live,_}|Is]) -> +check_killed_block(R, [{'%live',_,Regs}|Is]) -> case R of - {x,X} when X >= Live -> killed; + {x,X} when (Regs bsr X) band 1 =:= 0 -> killed; _ -> check_killed_block(R, Is) end; check_killed_block(_, []) -> transparent. @@ -574,8 +582,10 @@ check_killed_block(_, []) -> transparent. %% 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.) +%% +%% '%live' annotations are not allowed. +%% +%% (Unknown instructions will cause an exception.) check_used_block({x,X}=R, [{set,Ds,Ss,{alloc,Live,Op}}|Is], St) -> if @@ -584,11 +594,6 @@ check_used_block({x,X}=R, [{set,Ds,Ss,{alloc,Live,Op}}|Is], St) -> 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) - end; check_used_block(_, [], St) -> {transparent,St}. check_used_block_1(R, Ss, Ds, Op, Is, St0) -> @@ -619,8 +624,7 @@ is_reg_used_at_1(_, 0, 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} + {used,St} -> {true,St} end. index_labels_1([{label,Lbl}|Is0], Acc) -> @@ -736,11 +740,6 @@ live_opt([{select,_,Src,Fail,List}=I|Is], Regs0, D, Acc) -> 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([{loop_rec,_Fail,_Dst}=I|Is], _, D, Acc) -> @@ -749,6 +748,11 @@ 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]); %% Transparent instructions - they neither use nor modify x registers. live_opt([{deallocate,_}=I|Is], Regs, D, Acc) -> @@ -840,14 +844,7 @@ 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]). diff --git a/lib/compiler/src/beam_validator.erl b/lib/compiler/src/beam_validator.erl index 942d69a756..16dba35adc 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 2004-2016. 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. @@ -31,15 +31,6 @@ -import(lists, [reverse/1,foldl/3,foreach/2,dropwhile/2]). --define(MAXREG, 1024). - -%%-define(DEBUG, 1). --ifdef(DEBUG). --define(DBG_FORMAT(F, D), (io:format((F), (D)))). --else. --define(DBG_FORMAT(F, D), ok). --endif. - %% 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) -> @@ -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), @@ -278,7 +264,6 @@ 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), @@ -296,7 +281,7 @@ 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); @@ -531,6 +516,9 @@ valfun_4({bif,element,{f,Fail},[Pos,Tuple],Dst}, Vst0) -> TupleType = upgrade_tuple_type({tuple,[get_tuple_size(PosType)]}, TupleType0), Vst = set_type(TupleType, Tuple, Vst1), set_type_reg(term, Dst, Vst); +valfun_4({bif,raise,{f,0},Src,_Dst}, Vst) -> + validate_src(Src, Vst), + kill_state(Vst); valfun_4({bif,Op,{f,Fail},Src,Dst}, Vst0) -> validate_src(Src, Vst0), Vst = branch_state(Fail, Vst0), @@ -594,7 +582,7 @@ valfun_4({test,bs_start_match2,{f,Fail},Live,[Ctx,NeedSlots],Ctx}, Vst0) -> verify_live(Live, 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 @@ -670,8 +658,10 @@ 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), @@ -758,10 +748,20 @@ valfun_4(_, _) -> verify_get_map(Fail, Src, List, Vst0) -> assert_type(map, Src, Vst0), - Vst1 = branch_state(Fail, Vst0), + Vst1 = foldl(fun(D, Vsti) -> + case is_reg_defined(D,Vsti) of + true -> set_type_reg(term,D,Vsti); + false -> Vsti + end + end, Vst0, extract_map_vals(List)), + Vst2 = branch_state(Fail, Vst1), Keys = extract_map_keys(List), assert_unique_map_keys(Keys), - verify_get_map_pair(List,Vst0,Vst1). + verify_get_map_pair(List,Vst0,Vst2). + +extract_map_vals([_Key,Val|T]) -> + [Val|extract_map_vals(T)]; +extract_map_vals([]) -> []. extract_map_keys([Key,_Val|T]) -> [Key|extract_map_keys(T)]; @@ -808,9 +808,11 @@ 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); @@ -837,7 +839,7 @@ 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 -> @@ -849,44 +851,74 @@ 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. -verify_call_match_context(Lbl, #vst{ft=Ft}) -> +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. + +ms_in_y_regs(Id, #vst{current=#st{y=Ys0}}) -> + Ys = gb_trees:to_list(Ys0), + [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}=St}=Vst0) -> @@ -970,9 +1002,9 @@ 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 -> @@ -985,9 +1017,10 @@ 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}). @@ -1017,7 +1050,7 @@ 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), @@ -1025,7 +1058,7 @@ 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; _ -> error({no_bsm_context,Reg}) end; bsm_get_context(Reg, _) -> error({bad_source,Reg}). @@ -1037,8 +1070,8 @@ 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. @@ -1050,7 +1083,7 @@ 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 @@ -1066,16 +1099,16 @@ 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) +set_type_reg(Type, {x,X}=Reg, #vst{current=#st{x=Xs}=St}=Vst) when is_integer(X), 0 =< X -> - limit_check(X), + check_limit(Reg), Vst#vst{current=St#st{x=gb_trees:enter(X, Type, Xs)}}; set_type_reg(Type, Reg, Vst) -> set_type_y(Type, Reg, Vst). 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}); @@ -1093,6 +1126,17 @@ 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}}. + +is_reg_defined({x,_}=Reg, Vst) -> is_type_defined_x(Reg, Vst); +is_reg_defined({y,_}=Reg, Vst) -> is_type_defined_y(Reg, Vst); +is_reg_defined(V, #vst{}) -> error({not_a_register, V}). + +is_type_defined_x({x,X}, #vst{current=#st{x=Xs}}) -> + gb_trees:is_defined(X,Xs). + +is_type_defined_y({y,Y}, #vst{current=#st{y=Ys}}) -> + gb_trees:is_defined(Y,Ys). + assert_term(Src, Vst) -> get_term_type(Src, Vst), ok. @@ -1120,7 +1164,7 @@ 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. %% @@ -1162,12 +1206,17 @@ assert_type(WantedType, Term, Vst) -> 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}}). @@ -1222,7 +1271,7 @@ get_term_type(Src, Vst) -> initialized -> error({unassigned,Src}); {catchtag,_} -> error({catchtag,Src}); {trytag,_} -> error({trytag,Src}); - {match_context,_,_} -> error({match_context,Src}); + #ms{} -> error({match_context,Src}); Type -> Type end. @@ -1374,11 +1423,12 @@ 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, _) -> +merge_types(#ms{id=Id,valid=B0,slots=Slots}=M, + #ms{id=Id,valid=B1,slots=Slots}) -> + M#ms{valid=B0 bor B1,slots=Slots}; +merge_types(#ms{}=M, _) -> M; -merge_types(_, {match_context,_,_}=M) -> +merge_types(_, #ms{}=M) -> M; merge_types(T1, T2) when T1 =/= T2 -> %% Too different. All we know is that the type is a 'term'. @@ -1502,7 +1552,6 @@ 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; @@ -1516,6 +1565,7 @@ 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; @@ -1546,8 +1596,12 @@ 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; @@ -1591,17 +1645,19 @@ return_type_math(pow, 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..6c7f8543c2 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 2012-2016. 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. @@ -24,6 +24,8 @@ -export([module/2]). +-import(lists, [dropwhile/2]). + module({Mod,Exp,Attr,Fs0,Lc}, _Opt) -> Fs = [function(F) || F <- Fs0], {ok,{Mod,Exp,Attr,Fs,Lc}}. @@ -51,6 +53,16 @@ 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([I|Is]) -> [undo_rename(I)|undo_renames(Is)]; undo_renames([]) -> []. diff --git a/lib/compiler/src/cerl.erl b/lib/compiler/src/cerl.erl index 010327b5e3..61abae344c 100644 --- a/lib/compiler/src/cerl.erl +++ b/lib/compiler/src/cerl.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2001-2010. All Rights Reserved. +%% Copyright Ericsson AB 2001-2016. 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. @@ -126,6 +126,7 @@ %% 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 +135,7 @@ 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 ]). @@ -1598,13 +1599,20 @@ is_c_map(#c_literal{val = V}) when is_map(V) -> is_c_map(_) -> false. --spec map_es(c_map()) -> [c_map_pair()]. +-spec map_es(c_map() | c_literal()) -> [c_map_pair()]. +map_es(#c_literal{anno=As,val=M}) when is_map(M) -> + [ann_c_map_pair(As, + #c_literal{anno=As,val='assoc'}, + #c_literal{anno=As,val=K}, + #c_literal{anno=As,val=V}) || {K,V} <- maps:to_list(M)]; map_es(#c_map{es = Es}) -> Es. --spec map_arg(c_map()) -> c_map() | c_literal(). +-spec map_arg(c_map() | c_literal()) -> c_map() | c_literal(). +map_arg(#c_literal{anno=As,val=M}) when is_map(M) -> + #c_literal{anno=As,val=#{}}; map_arg(#c_map{arg=M}) -> M. @@ -1629,6 +1637,11 @@ 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) -> @@ -1681,6 +1694,11 @@ 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(). @@ -1937,7 +1955,7 @@ 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 @@ -1945,18 +1963,18 @@ 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}). @@ -1974,13 +1992,13 @@ 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}). @@ -2029,14 +2047,14 @@ 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. @@ -2482,7 +2500,7 @@ 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. @@ -2493,7 +2511,7 @@ 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)). @@ -3400,7 +3418,7 @@ 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. @@ -3412,7 +3430,7 @@ 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)). @@ -3518,7 +3536,7 @@ 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. @@ -3530,7 +3548,7 @@ 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)). @@ -3622,7 +3640,7 @@ 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. @@ -3634,7 +3652,7 @@ 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..805095e30c 100644 --- a/lib/compiler/src/cerl_clauses.erl +++ b/lib/compiler/src/cerl_clauses.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2001-2010. All Rights Reserved. +%% Copyright Ericsson AB 2001-2016. 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/cerl_inline.erl b/lib/compiler/src/cerl_inline.erl index cbcacf9e8e..2a8cf2e758 100644 --- a/lib/compiler/src/cerl_inline.erl +++ b/lib/compiler/src/cerl_inline.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2001-2012. All Rights Reserved. +%% Copyright Ericsson AB 2001-2016. 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/cerl_trees.erl b/lib/compiler/src/cerl_trees.erl index 2c9b72a30b..b3decbec1f 100644 --- a/lib/compiler/src/cerl_trees.erl +++ b/lib/compiler/src/cerl_trees.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2001-2010. All Rights Reserved. +%% Copyright Ericsson AB 2001-2016. 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. @@ -27,7 +27,7 @@ -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, 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 +61,7 @@ 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 +341,162 @@ 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 = 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}. @@ -640,8 +667,8 @@ 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. @@ -703,13 +730,14 @@ 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), @@ -725,14 +753,21 @@ 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), - {Val, N2} = label(map_pair_key(T), N1, Env), - {Key, N3} = label(map_pair_val(T), N2, Env), + {Key, N2} = label(map_pair_key(T), N1, Env), + {Val, N3} = label(map_pair_val(T), N2, Env), {As, N4} = label_ann(T, N3), {ann_c_map_pair(As,Op,Key,Val), N4}; 'let' -> diff --git a/lib/compiler/src/compile.erl b/lib/compiler/src/compile.erl index e0a29fe9b1..97d63d399a 100644 --- a/lib/compiler/src/compile.erl +++ b/lib/compiler/src/compile.erl @@ -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) -> @@ -120,6 +136,14 @@ 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 %% @@ -216,6 +240,8 @@ 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)]); @@ -658,7 +684,7 @@ kernel_passes() -> {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}, @@ -671,11 +697,16 @@ 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_bs_opt,{pass,beam_bs}}, + {iff,dbs,{listing,"bs"}}, {unless,no_bopt,{pass,beam_bool}}, {iff,dbool,{listing,"bool"}}, {unless,no_topt,{pass,beam_type}}, @@ -703,6 +734,7 @@ 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"}}]}, @@ -1180,7 +1212,7 @@ makedep_output(#compile{code=Code,options=Opts,ofile=Ofile}=St) -> end, {ok,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]}} @@ -1209,6 +1241,17 @@ core_fold_module_after_inlining(#compile{code=Code0,options=Opts}=St) -> {ok,Code,_Ws} = sys_core_fold:module(Code0, Opts), {ok,St#compile{code=Code}}. +v3_kernel(#compile{code=Code0,options=Opts,warnings=Ws0}=St) -> + {ok,Code,Ws} = v3_kernel:module(Code0, Opts), + case Ws =:= [] orelse test_core_inliner(St) of + false -> + {ok,St#compile{code=Code,warnings=Ws0++Ws}}; + true -> + %% cerl_inline may produce code that generates spurious + %% warnings. Ignore any such warnings. + {ok,St#compile{code=Code}} + end. + test_old_inliner(#compile{options=Opts}) -> %% The point of this test is to avoid loading the old inliner %% if we know that it will not be used. @@ -1300,27 +1343,18 @@ 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), + Source = paranoid_absname(File), Opts1 = lists:map(fun({debug_info_key,_}) -> {debug_info_key,'********'}; (Other) -> Other end, Opts0), @@ -1329,6 +1363,16 @@ beam_asm(#compile{ifile=File,code=Code0, {ok,Code} -> {ok,St#compile{code=Code,abstract_code=[]}} end. +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}) -> %% This test is done late, in case some other option has turned off native. %% 'native' given on the command line can be overridden by @@ -1452,8 +1496,8 @@ 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. @@ -1601,8 +1645,8 @@ listing(LFun, Ext, St) -> LFun(Lf, Code), ok = file:close(Lf), {ok,St}; - {error,_Error} -> - Es = [{Lfile,[{none,compile,write_error}]}], + {error,Error} -> + Es = [{Lfile,[{none,compile,{write_error,Error}}]}], {error,St#compile{errors=St#compile.errors ++ Es}} end. @@ -1612,11 +1656,8 @@ 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 +1669,8 @@ 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([]) -> []. @@ -1674,6 +1717,7 @@ help(_) -> %% Compile entry point for erl_compile. compile(File0, _OutFile, Options) -> + pre_load(), File = shorten_filename(File0), case file(File, make_erl_options(Options)) of {ok,_Mod} -> ok; @@ -1738,3 +1782,47 @@ 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_bool, + beam_bs, + beam_bsm, + beam_clean, + beam_dead, + beam_dict, + beam_except, + beam_flatten, + beam_jump, + beam_opcodes, + beam_peep, + beam_receive, + 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_dsetel, + sys_core_fold, + sys_pre_expand, + v3_codegen, + v3_core, + v3_kernel, + v3_life], + _ = code:ensure_modules_loaded(L), + ok. diff --git a/lib/compiler/src/compiler.app.src b/lib/compiler/src/compiler.app.src index afb85f4710..1fd7800e85 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 1997-2016. 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. @@ -25,6 +25,7 @@ beam_asm, beam_block, beam_bool, + beam_bs, beam_bsm, beam_clean, beam_dead, @@ -37,6 +38,7 @@ beam_opcodes, beam_peep, beam_receive, + beam_reorder, beam_split, beam_trim, beam_type, 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..7d3513c0ba 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 1999-2015. 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. @@ -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 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..8028aa99bb 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 1999-2016. 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. @@ -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 @@ -289,11 +308,17 @@ 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'). -map_pairs -> map_pair : ['$1']. -map_pairs -> map_pair ',' map_pairs : ['$1' | '$3']. +anno_map_expr -> map_expr : '$1'. +anno_map_expr -> '(' map_expr '-|' annotation ')' : cerl:set_ann('$2', '$4'). + +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 +337,11 @@ 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 @@ -413,7 +441,8 @@ 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). diff --git a/lib/compiler/src/core_pp.erl b/lib/compiler/src/core_pp.erl index c2a6a81d5e..67209d06be 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 1999-2016. 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. @@ -21,7 +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),$}]; format_anno(Val, Ctxt) when is_atom(Val) -> @@ -121,14 +131,11 @@ format_1(#c_literal{anno=A,val=Bitstring}, Ctxt) when is_bitstring(Bitstring) -> format_1(#c_binary{anno=A,segments=Segs}, Ctxt); format_1(#c_literal{anno=A,val=M},Ctxt) when is_map(M) -> Pairs = maps:to_list(M), - Op = case Ctxt of - #ctxt{ class = clause } -> exact; - _ -> assoc - end, - Cpairs = [#c_map_pair{op=#c_literal{val=Op}, + Op = #c_literal{val=assoc}, + Cpairs = [#c_map_pair{op=Op, key=#c_literal{val=K}, val=#c_literal{val=V}} || {K,V} <- Pairs], - format_1(#c_map{anno=A,arg=#c_literal{val=#{}},es=Cpairs},Ctxt); + format_1(#c_map{anno=A,arg=#c_literal{val=#{}},es=Cpairs},Ctxt); format_1(#c_var{name={I,A}}, _) -> [core_atom(I),$/,integer_to_list(A)]; format_1(#c_var{name=V}, _) -> @@ -172,7 +179,8 @@ 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{anno=[],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 +203,16 @@ 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 +229,7 @@ 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 +336,30 @@ 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 +372,10 @@ 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 +383,7 @@ format_bitstr(#c_bitstr{val=V,size=S,unit=U,type=T,flags=Fs}, Ctxt0) -> ["#<", Val, ">(", format_hseq(Vs,",", Ctxt2, fun format/2), $)]. format_clauses(Cs, Ctxt) -> - format_vseq(Cs, "", "", set_class(Ctxt, clause), - fun format_clause/2). + format_vseq(Cs, "", "", Ctxt, fun format_clause/2). format_clause(Node, Ctxt) -> maybe_anno(Node, fun format_clause_1/2, Ctxt). @@ -383,15 +395,13 @@ format_clause_1(#c_clause{pats=Ps,guard=G,body=B}, Ctxt) -> case is_trivial_guard(G) of true -> [" when ", - format_guard(G, add_indent(set_class(Ctxt, expr), - width(Ptxt, Ctxt) + 6))]; + format_guard(G, add_indent(Ctxt, width(Ptxt, Ctxt) + 6))]; false -> [nl_indent(Ctxt2), "when ", format_guard(G, add_indent(Ctxt2, 2))] end++ " ->", - nl_indent(Ctxt2) - | format(B, set_class(Ctxt2, expr)) + nl_indent(Ctxt2) | format(B, Ctxt2) ]. is_trivial_guard(#c_literal{val=Val}) when is_atom(Val) -> true; @@ -445,50 +455,53 @@ format_list_tail(Tail, Ctxt) -> format_map_pair(Op, K, V, Ctxt0) -> Ctxt1 = add_indent(Ctxt0, 1), - Txt = format(K, set_class(Ctxt1, expr)), + Txt = format(K, Ctxt1), Ctxt2 = add_indent(Ctxt0, width(Txt, Ctxt1)), [Txt,Op,format(V, Ctxt2)]. -indent(Ctxt) -> indent(Ctxt#ctxt.indent, Ctxt). - -indent(N, _) when N =< 0 -> ""; -indent(N, Ctxt) -> - T = Ctxt#ctxt.tab_width, - string:chars($\t, N div T, string:chars($\s, N rem T)). +indent(#ctxt{indent=N}) -> + if + N =< 0 -> + ""; + true -> + string:chars($\t, N div ?TAB_WIDTH, 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 +515,9 @@ 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; diff --git a/lib/compiler/src/core_scan.erl b/lib/compiler/src/core_scan.erl index 5e85bba2bd..11b52f6c5f 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 2000-2016. 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/erl_bifs.erl b/lib/compiler/src/erl_bifs.erl index c00f5eab70..6b2d781a76 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 2001-2016. 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/genop.tab b/lib/compiler/src/genop.tab index 8124729b35..dcbdeb32e6 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 1998-2016. 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. @@ -45,7 +45,7 @@ BEAM_FORMAT_NUMBER=0 ## Save the next instruction as the return address in the CP register. 4: call/2 -## @spec call_last Arity Label Dellocate +## @spec call_last Arity Label Deallocate ## @doc Deallocate and do a tail recursive call to the function at Label. ## Do not update the CP register. ## Before the call deallocate Deallocate words of stack. @@ -137,7 +137,7 @@ BEAM_FORMAT_NUMBER=0 # Sending & receiving. # ## @spec send -## @doc Send argument in x(0) as a message to the destination process in x(0). +## @doc Send argument in x(1) as a message to the destination process in x(0). ## The message in x(1) ends up as the result of the send in x(0). 20: send/0 @@ -164,12 +164,12 @@ BEAM_FORMAT_NUMBER=0 25: wait/1 ## @spec wait_timeout Lable Time -## @doc Sets up a timeout of Time milllisecons and saves the address of the +## @doc Sets up a timeout of Time milliseconds and saves the address of the ## following instruction as the entry point if the timeout triggers. 26: wait_timeout/2 # -# Arithmethic opcodes. +# Arithmetic opcodes. # 27: -m_plus/4 28: -m_minus/4 @@ -316,7 +316,7 @@ BEAM_FORMAT_NUMBER=0 66: get_tuple_element/3 ## @spec set_tuple_element NewElement Tuple Position -## @doc Update the element at postition Position of the tuple Tuple +## @doc Update the element at position Position of the tuple Tuple ## with the new element NewElement. 67: set_tuple_element/3 diff --git a/lib/compiler/src/rec_env.erl b/lib/compiler/src/rec_env.erl index 0e9e12d1ad..cdc513e57c 100644 --- a/lib/compiler/src/rec_env.erl +++ b/lib/compiler/src/rec_env.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2001-2014. All Rights Reserved. +%% Copyright Ericsson AB 2001-2016. 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. @@ -22,8 +22,7 @@ %% @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 +61,7 @@ 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 +79,13 @@ 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 +95,7 @@ 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 +106,7 @@ test_1(_,0, Env) -> -spec empty() -> environment(). empty() -> - [{map, dict:new()}]. + [{map, #{}}]. %% ===================================================================== @@ -119,14 +117,14 @@ 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 +144,12 @@ 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 +163,8 @@ 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 +172,8 @@ 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 +186,12 @@ 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 +210,12 @@ 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 +234,12 @@ 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 +257,17 @@ 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 +296,40 @@ 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 +382,15 @@ 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 +406,8 @@ 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 +415,10 @@ 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 +588,17 @@ 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_dsetel.erl b/lib/compiler/src/sys_core_dsetel.erl index ac32db10fe..bd3eeae238 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 2002-2016. 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. @@ -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}. @@ -95,9 +95,11 @@ 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 +205,7 @@ 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 +219,7 @@ 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 +237,25 @@ 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 +353,8 @@ 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 27d023d067..4922953407 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 1999-2016. 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. @@ -277,7 +277,7 @@ 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 @@ -374,10 +374,21 @@ 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}=App, _, Sub) -> Op1 = expr(Op0, value, Sub), As1 = expr_list(As0, value, Sub), - App#c_apply{op=Op1,args=As1}; + case Op1 of + #c_var{} -> + App#c_apply{op=Op1,args=As1}; + _ -> + add_warning(App, 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), @@ -420,13 +431,13 @@ 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. @@ -457,7 +468,8 @@ 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); @@ -722,7 +734,7 @@ 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 @@ -775,7 +787,7 @@ 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. @@ -1078,15 +1090,28 @@ is_atom_or_var(_) -> false. %% 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 +1130,13 @@ 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 +1151,7 @@ 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, @@ -1206,20 +1237,132 @@ 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. @@ -2040,24 +2183,22 @@ opt_not_in_let_1(V, Call, Body) -> #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}, @@ -2251,11 +2392,11 @@ 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}}; @@ -2271,39 +2412,49 @@ move_let_into_expr(#c_let{vars=Lvs0,body=Lbody0}=Let, case {TwoClauses,is_failing_clause(Ca0),is_failing_clause(Cb0)} of {true,false,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]}; + 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}, + Cb = clause(Cb0, Cexpr, value, Sub0), + Case#c_case{arg=Cexpr,clauses=[Ca,Cb]} + 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; {_,_,_} -> impossible end; move_let_into_expr(#c_let{vars=Lvs0,body=Lbody0}=Let, @@ -2595,7 +2746,7 @@ 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}}; @@ -2624,14 +2775,18 @@ 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 end; @@ -2793,12 +2948,18 @@ 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('div', [_,_]) -> 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; @@ -3091,12 +3252,12 @@ bsm_ensure_no_partition_2([#c_var{name=V}|Ps], N, G, Vstate, S) -> 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}|Cs], Pos) -> +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); - P -> - bsm_problem(P, bin_partition) + _ -> + bsm_problem(C, bin_partition) end; bsm_ensure_no_partition_after([], _) -> ok. @@ -3207,6 +3368,29 @@ 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) -> @@ -3227,6 +3411,8 @@ 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) -> diff --git a/lib/compiler/src/sys_core_inline.erl b/lib/compiler/src/sys_core_inline.erl index 838dda68c6..8c1f69d5de 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 2000-2016. 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/sys_pre_attributes.erl b/lib/compiler/src/sys_pre_attributes.erl index f0cb630205..bc93c85989 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. diff --git a/lib/compiler/src/sys_pre_expand.erl b/lib/compiler/src/sys_pre_expand.erl index d9cc4b530c..7ab4e1845c 100644 --- a/lib/compiler/src/sys_pre_expand.erl +++ b/lib/compiler/src/sys_pre_expand.erl @@ -29,30 +29,26 @@ %% 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 + fcount=0, %Local fun count + ctype %Call type map }). %% 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! +%% Expand the forms in one module. +%% %% CompileOptions is augmented with options from -compile attributes. module(Fs0, Opts0) -> @@ -65,19 +61,28 @@ module(Fs0, Opts0) -> %% Set pre-defined exported functions. PreExp = [{module_info,0},{module_info,1}], + %% Build the set of defined functions and the initial call + %% type map. + Defined = defined_functions(Fs, PreExp), + Ctype = maps:from_list([{K,local} || K <- Defined]), + %% Build initial expand record. St0 = #expand{exports=PreExp, - defined=PreExp + ctype=Ctype }, + %% Expand the functions. - {Tfs,St1} = forms(Fs, define_functions(Fs, St0)), + {Tfs,St1} = forms(Fs, St0), + %% Get the correct list of exported functions. Exports = case member(export_all, Opts) of - true -> gb_sets:to_list(St1#expand.defined); + true -> Defined; false -> St1#expand.exports end, + St2 = St1#expand{exports=Exports,ctype=undefined}, + %% Generate all functions from stored info. - {Ats,St3} = module_attrs(St1#expand{exports = Exports}), + {Ats,St3} = module_attrs(St2), {Mfs,St4} = module_predef_funcs(St3), {St4#expand.module, St4#expand.exports, Ats ++ Tfs ++ Mfs, Opts}. @@ -85,14 +90,14 @@ module(Fs0, Opts0) -> compiler_options(Forms) -> lists:flatten([C || {attribute,_,compile,C} <- Forms]). -%% define_function(Form, State) -> State. +%% defined_function(Forms, Predef) -> Functions. %% Add function to defined if form is a function. -define_functions(Forms, #expand{defined=Predef}=St) -> +defined_functions(Forms, Predef) -> Fs = foldl(fun({function,_,N,A,_Cs}, Acc) -> [{N,A}|Acc]; (_, Acc) -> Acc end, Predef, Forms), - St#expand{defined=gb_sets:from_list(Fs)}. + ordsets:from_list(Fs). module_attrs(#expand{attributes=Attributes}=St) -> Attrs = [{attribute,Line,Name,Val} || {Name,Line,Val} <- Attributes], @@ -113,23 +118,21 @@ is_fa_list([{FuncName, Arity}|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), +module_predef_funcs(St0) -> + {Mpf1,St1} = module_predef_func_beh_info(St0), + Mpf2 = module_predef_funcs_mod_info(St1), Mpf = [erl_parse:new_anno(F) || F <- Mpf1++Mpf2], - {Mpf,St2}. + {Mpf,St1}. 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, + PreDef0 = [{behaviour_info,1}], + PreDef = ordsets:from_list(PreDef0), {[gen_beh_info(Callbacks, OptionalCallbacks)], - St#expand{defined=gb_sets:union(gb_sets:from_list(PreDef), Defined), - exports=union(from_list(PreExp), Exports)}}. + St#expand{exports=ordsets:union(PreDef, Exports)}}. gen_beh_info(Callbacks, OptionalCallbacks) -> List = make_list(Callbacks), @@ -156,20 +159,16 @@ make_optional_list([{Name,Arity}|Rest]) -> {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,[],[], +module_predef_funcs_mod_info(#expand{module=Mod}) -> + ModAtom = {atom,0,Mod}, + [{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'}],[], + [ModAtom]}]}]}, + {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)}}. + [ModAtom,{var,0,'X'}]}]}]}]. %% forms(Forms, State) -> %% {TransformedForms,State'} @@ -196,7 +195,8 @@ 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)}; + St#expand{exports=ordsets:union(ordsets:from_list(Es), + St#expand.exports)}; attribute(import, Is, _L, St) -> import(Is, St); attribute(compile, _C, _L, St) -> @@ -231,8 +231,6 @@ head(As, St) -> pattern_list(As, St). %% {TransformedPattern,State'} %% -pattern({var,_,'_'}=Var, St) -> %Ignore anonymous variable. - {Var,St}; pattern({var,_,_}=Var, St) -> {Var,St}; pattern({char,_,_}=Char, St) -> @@ -385,19 +383,19 @@ 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), + {Cs,St1} = clauses(Cs0, St0), {{'if',Line,Cs},St1}; expr({'case',Line,E0,Cs0}, St0) -> {E,St1} = expr(E0, St0), - {Cs,St2} = icr_clauses(Cs0, St1), + {Cs,St2} = clauses(Cs0, St1), {{'case',Line,E,Cs},St2}; expr({'receive',Line,Cs0}, St0) -> - {Cs,St1} = icr_clauses(Cs0, St0), + {Cs,St1} = 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), + {Cs,St3} = clauses(Cs0, St2), {{'receive',Line,Cs,To,ToEs},St3}; expr({'fun',Line,Body}, St) -> fun_tq(Line, Body, St); @@ -406,21 +404,15 @@ expr({named_fun,Line,Name,Cs}, St) -> 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 -> + Key = {N,Ar}, + case St1#expand.ctype of + #{Key:=local} -> {{call,Line,Atom,As},St1}; + #{Key:={imported,Mod}} -> + {{call,Line,{remote,La,{atom,La,Mod},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 + true = erl_internal:bif(N, Ar), + {{call,Line,{remote,La,{atom,La,erlang},Atom},As},St1} end; expr({call,Line,{remote,Lr,M0,F},As0}, St0) -> {[M1,F1|As1],St1} = expr_list([M0,F|As0], St0), @@ -430,12 +422,11 @@ expr({call,Line,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), + {Scs1,St2} = clauses(Scs0, St1), + {Ccs1,St3} = 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) -> @@ -456,21 +447,6 @@ expr_list([E0|Es0], St0) -> {[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'} @@ -486,16 +462,9 @@ lc_tq(Line, [{b_generate,Lg,P0,G0}|Qs0], St0) -> {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; + {F1,St1} = expr(F0, St0), + {Qs1,St2} = lc_tq(Line, Qs0, St1), + {[F1|Qs1],St2}; lc_tq(_Line, [], St0) -> {[],St0}. @@ -527,7 +496,7 @@ fun_tq(L, {function,M,F,A}, St) when is_atom(M), is_atom(F), is_integer(A) -> fun_tq(Lf, {function,_,_,_}=ExtFun, St) -> {{'fun',Lf,ExtFun},St}; fun_tq(Lf, {clauses,Cs0}, St0) -> - {Cs1,St1} = fun_clauses(Cs0, St0), + {Cs1,St1} = 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. @@ -535,18 +504,10 @@ fun_tq(Lf, {clauses,Cs0}, St0) -> {{'fun',Lf,{clauses,Cs1},{Index,Uniq,Fname}},St2}. fun_tq(Line, Cs0, St0, Name) -> - {Cs1,St1} = fun_clauses(Cs0, St0), + {Cs1,St1} = 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) -> @@ -571,7 +532,6 @@ pattern_element({bin_element,Line,Expr0,Size0,Type0}, {Es,St0}) -> {[{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), @@ -592,8 +552,7 @@ coerce_to_float({integer,L,I}=E, [float|_]) -> try {float,L,float(I)} catch - error:badarg -> E; - error:badarith -> E + error:badarg -> E end; coerce_to_float(E, _) -> E. @@ -647,25 +606,11 @@ string_to_conses(Line, Cs, Tail) -> %% 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. +%% Handle import declarations. -import({Mod,Fs}, St) -> +import({Mod,Fs}, #expand{ctype=Ctype0}=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). + Ctype = foldl(fun(F, A) -> + A#{F=>{imported,Mod}} + end, Ctype0, Fs), + St#expand{ctype=Ctype}. diff --git a/lib/compiler/src/v3_codegen.erl b/lib/compiler/src/v3_codegen.erl index 34c67b16ca..4df1aadd0a 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 1999-2016. 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. @@ -827,21 +827,24 @@ 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([{var,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'. + Size = {atom,all}, %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 +856,7 @@ 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,Size,Unit,{field_flags,Flags}],Rhd}], Int1}; true -> %% Since the matching context will not be used again, @@ -868,7 +871,7 @@ 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,Size,Unit,{field_flags,Flags}],CtxReg}], Int1} end end, @@ -1086,6 +1089,23 @@ 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(Test, As, Fail, I, Vdb, Bef, St) -> Args = cg_reg_args(As, Bef), Aft = clear_dead(Bef, I, Vdb), @@ -1152,19 +1172,15 @@ 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. + %% jump FailureLabel {remote,{atom,erlang},{atom,error}} = Func, %Assertion. [{var,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. @@ -1327,12 +1343,13 @@ bif_cg(Bif, As, [{var,V}], Le, Vdb, Bef, St0) -> %% that we save any variable that will be live after this BIF call. MayFail = not erl_bifs:is_safe(erlang, Bif, length(As)), - {Sis,Int0} = case St0#cg.in_catch andalso - St0#cg.bfail =:= 0 andalso - MayFail of - true -> adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb); - false -> {[],Bef} - end, + {Sis,Int0} = + case MayFail of + true -> + maybe_adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb, St0); + false -> + {[],Bef} + end, Int1 = clear_dead(Int0, Le#l.i, Vdb), Reg = put_reg(V, Int1#sr.reg), Int = Int1#sr{reg=Reg}, @@ -1363,11 +1380,7 @@ gc_bif_cg(Bif, As, [{var,V}], Le, Vdb, Bef, St0) -> %% Currently, we are somewhat pessimistic in %% that we save any variable that will be live after this BIF call. - {Sis,Int0} = - case St0#cg.in_catch andalso St0#cg.bfail =:= 0 of - true -> adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb); - false -> {[],Bef} - end, + {Sis,Int0} = maybe_adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb, St0), Int1 = clear_dead(Int0, Le#l.i, Vdb), Reg = put_reg(V, Int1#sr.reg), @@ -1512,8 +1525,7 @@ set_cg([{var,R}], {cons,Es}, Le, Vdb, Bef, St) -> 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{in_catch=InCatch, bfail=Bfail}=St) -> +set_cg([{var,R}], {binary,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. @@ -1532,28 +1544,19 @@ set_cg([{var,R}], {binary,Segs}, Le, Vdb, Bef, %% First generate the code that constructs each field. Fail = {f,Bfail}, PutCode = cg_bin_put(Segs, Fail, Bef), - {Sis,Int1} = - case InCatch of - true -> adjust_stack(Int0, Le#l.i, Le#l.i+1, Vdb); - false -> {[],Int0} - end, + {Sis,Int1} = maybe_adjust_stack(Int0, Le#l.i, Le#l.i+1, Vdb, St), MaxRegs = max_reg(Bef#sr.reg), Aft = clear_dead(Int1, Le#l.i, Vdb), %% 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{in_catch=InCatch,bfail=Bfail}=St) -> - Fail = {f,Bfail}, - {Sis,Int0} = - case InCatch of - true -> adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb); - false -> {[],Bef} - end, - SrcReg = cg_reg_arg(Map,Int0), +%% Map: single variable key. +set_cg([{var,R}], {map,Op,Map,[{map_pair,{var,_}=K,V}]}, Le, Vdb, Bef, St0) -> + {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), List = [cg_reg_arg(K,Int0),cg_reg_arg(V,Int0)], @@ -1565,26 +1568,17 @@ 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} = set_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{in_catch=InCatch,bfail=Bfail}=St) -> +%% Map: (possibly) multiple literal keys. +set_cg([{var,R}], {map,Op,Map,Es}, Le, Vdb, Bef, St0) -> %% assert key literals [] = [Var||{map_pair,{var,_}=Var,_} <- Es], - Fail = {f,Bfail}, - {Sis,Int0} = - case InCatch of - true -> adjust_stack(Bef, Le#l.i, Le#l.i+1, Vdb); - false -> {[],Bef} - end, - 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 @@ -1598,11 +1592,10 @@ 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}; + {Is,St1} = set_cg_map(Line, Op, SrcReg, Target, Live, List, St0), + {Sis++Is,Aft,St1}; + +%% Everything else. set_cg([{var,R}], Con, Le, Vdb, Bef, St) -> %% Find a place for the return register first. Int = Bef#sr{reg=put_reg(R, Bef#sr.reg)}, @@ -1615,6 +1608,34 @@ set_cg([{var,R}], Con, Le, Vdb, Bef, St) -> end, {Ais,clear_dead(Int, Le#l.i, Vdb),St}. + +set_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. %%% @@ -1857,6 +1878,9 @@ 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_prefer_y({var,V}, Bef) -> fetch_var_prefer_y(V, Bef); +cg_reg_arg_prefer_y(Literal, _) -> Literal. + %% cg_setup_call([Arg], Bef, Cur, Vdb) -> {[Instr],Aft}. %% Do the complete setup for a call/enter. @@ -2038,6 +2062,19 @@ trim_free([R|Rs0]) -> end; trim_free([]) -> []. +%% maybe_adjust_stack(Bef, FirstBefore, LastFrom, Vdb, St) -> {[Ainstr],Aft}. +%% Adjust the stack, but only if the code is inside a catch and not +%% inside a guard. Use this funtion before instructions that may +%% cause an exception. + +maybe_adjust_stack(Bef, Fb, Lf, Vdb, St) -> + case St of + #cg{in_catch=true,bfail=0} -> + adjust_stack(Bef, Fb, Lf, Vdb); + #cg{} -> + {[],Bef} + end. + %% adjust_stack(Bef, FirstBefore, LastFrom, Vdb) -> {[Ainstr],Aft}. %% Do complete stack adjustment by compressing stack and adding %% variables to be saved. Try to optimise ordering on stack by @@ -2085,6 +2122,12 @@ 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). @@ -2158,11 +2201,11 @@ 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). diff --git a/lib/compiler/src/v3_core.erl b/lib/compiler/src/v3_core.erl index 0941ad5dd5..d71411de80 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 1999-2016. 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. @@ -469,7 +469,8 @@ 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='=:='}, @@ -680,9 +681,36 @@ 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 +752,32 @@ 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{}) -> @@ -783,7 +837,7 @@ 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), @@ -804,7 +858,7 @@ map_op(map_field_assoc) -> #c_literal{val=assoc}; map_op(map_field_exact) -> #c_literal{val=exact}. is_valid_map_src(#c_literal{val = M}) when is_map(M) -> true; -is_valid_map_src(#c_var{}) -> true; +is_valid_map_src(#c_var{}=Var) -> not cerl:is_c_fname(Var); is_valid_map_src(_) -> false. %% try_exception([ExcpClause], St) -> {[ExcpVar],Handler,St}. @@ -814,12 +868,16 @@ try_exception(Ecs0, St0) -> {Evs,St1} = new_vars(3, St0), % Tag, Value, Info {Ecs1,Ceps,St2} = clauses(Ecs0, St1), [_,Value,Info] = Evs, - Ec = #iclause{anno=#a{anno=[compiler_generated]}, + LA = case Ecs1 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=Ecs1,fc=Ec}], {Evs,Ceps++Hs,St2}. try_after(As, St0) -> @@ -1079,13 +1137,39 @@ 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}. @@ -1306,7 +1390,9 @@ 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), @@ -1652,10 +1738,12 @@ pat_alias_map_pairs_1([]) -> []. pat_bin(Ps, St) -> [pat_segment(P, St) || P <- Ps]. -pat_segment({bin_element,_,Val,Size,[Type,{unit,Unit}|Flags]}, St) -> +pat_segment({bin_element,L,Val,Size,[Type,{unit,Unit}|Flags]}, St) -> + Anno = lineno_anno(L, St), {Pval,[],St1} = pattern(Val,St), {Psize,[],_St2} = pattern(Size,St1), - #c_bitstr{val=Pval,size=Psize, + #c_bitstr{anno=Anno, + val=Pval,size=Psize, unit=#c_literal{val=Unit}, type=#c_literal{val=Type}, flags=#c_literal{val=Flags}}. @@ -1852,27 +1940,22 @@ 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 +1963,15 @@ 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 +2102,8 @@ 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]}, diff --git a/lib/compiler/src/v3_kernel.erl b/lib/compiler/src/v3_kernel.erl index 7ee564683b..b4bbc5e739 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 1999-2016. 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. @@ -117,7 +117,7 @@ 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 @@ -143,8 +143,11 @@ 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(_) -> true. function({#c_var{name={F,Arity}=FA},Body}, St0) -> @@ -239,7 +242,7 @@ 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 +402,7 @@ 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 +497,7 @@ 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), @@ -1837,14 +1840,17 @@ 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 +1989,6 @@ 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..5216a1a620 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 1999-2016. 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/v3_kernel_pp.erl b/lib/compiler/src/v3_kernel_pp.erl index 03b034ae98..0b90f0a1e0 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 1999-2016. 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/v3_life.erl b/lib/compiler/src/v3_life.erl index fa057ae211..1452b78d1d 100644 --- a/lib/compiler/src/v3_life.erl +++ b/lib/compiler/src/v3_life.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2012. 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/v3_life.hrl b/lib/compiler/src/v3_life.hrl index 29a08f8c72..9d03a86ccd 100644 --- a/lib/compiler/src/v3_life.hrl +++ b/lib/compiler/src/v3_life.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. |