diff options
Diffstat (limited to 'lib/compiler/src/compile.erl')
| -rw-r--r-- | lib/compiler/src/compile.erl | 230 |
1 files changed, 179 insertions, 51 deletions
diff --git a/lib/compiler/src/compile.erl b/lib/compiler/src/compile.erl index 6510571441..0325c714d0 100644 --- a/lib/compiler/src/compile.erl +++ b/lib/compiler/src/compile.erl @@ -31,6 +31,9 @@ %% Erlc interface. -export([compile/3,compile_beam/3,compile_asm/3,compile_core/3]). +%% Utility functions for compiler passes. +-export([run_sub_passes/2]). + -export_type([option/0]). -include("erl_compile.hrl"). @@ -39,6 +42,8 @@ -import(lists, [member/2,reverse/1,reverse/2,keyfind/3,last/1, map/2,flatmap/2,foreach/2,foldr/3,any/2]). +-define(SUB_PASS_TIMES, compile__sub_pass_times). + %%---------------------------------------------------------------------- -type abstract_code() :: [erl_parse:abstract_form()]. @@ -64,6 +69,7 @@ -type err_ret() :: 'error' | {'error', errors(), warnings()}. -type comp_ret() :: mod_ret() | bin_ret() | err_ret(). + %%---------------------------------------------------------------------- %% @@ -143,6 +149,30 @@ noenv_output_generated(Opts) -> env_compiler_options() -> env_default_opts(). + +%%% +%%% Run sub passes from a compiler pass. +%%% + +-spec run_sub_passes([term()], term()) -> term(). + +run_sub_passes(Ps, St) -> + case get(?SUB_PASS_TIMES) of + undefined -> + Runner = fun(_Name, Run, S) -> Run(S) end, + run_sub_passes_1(Ps, Runner, St); + Times when is_list(Times) -> + Runner = fun(Name, Run, S0) -> + T1 = erlang:monotonic_time(), + S = Run(S0), + T2 = erlang:monotonic_time(), + put(?SUB_PASS_TIMES, + [{Name,T2-T1}|get(?SUB_PASS_TIMES)]), + S + end, + run_sub_passes_1(Ps, Runner, St) + end. + %% %% Local functions %% @@ -180,8 +210,11 @@ do_compile(Input, Opts0) -> {error,Reason} end end, - %% Dialyzer has already spawned workers. - case lists:member(dialyzer, Opts) of + %% Some tools, like Dialyzer, has already spawned workers + %% and spawning extra workers actually slow the compilation + %% down instead of speeding it up, so we provide a mechanism + %% to bypass the compiler process. + case lists:member(no_spawn_compiler_process, Opts) of true -> IntFun(); false -> @@ -218,23 +251,36 @@ expand_opt(report, Os) -> [report_errors,report_warnings|Os]; expand_opt(return, Os) -> [return_errors,return_warnings|Os]; +expand_opt(no_bsm3, Os) -> + %% The new bsm pass requires bsm3 instructions. + [no_bsm3,no_bsm_opt|Os]; expand_opt(r16, Os) -> - [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; + expand_opt_before_21(Os); expand_opt(r17, Os) -> - [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; + expand_opt_before_21(Os); expand_opt(r18, Os) -> - [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; + expand_opt_before_21(Os); expand_opt(r19, Os) -> - [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; + expand_opt_before_21(Os); expand_opt(r20, Os) -> - [no_get_hd_tl,no_record_opt,no_utf8_atoms|Os]; + expand_opt_before_21(Os); +expand_opt(r21, Os) -> + [no_put_tuple2 | expand_opt(no_bsm3, Os)]; expand_opt({debug_info_key,_}=O, Os) -> [encrypt_debug_info,O|Os]; -expand_opt(no_float_opt, Os) -> - %%Turn off the entire type optimization pass. - [no_topt|Os]; +expand_opt(no_type_opt=O, Os) -> + %% Be sure to keep the no_type_opt option so that it will + %% be recorded in the BEAM file, allowing the test suites + %% to recompile the file with this option. + [O,no_ssa_opt_type_start, + no_ssa_opt_type_continue, + no_ssa_opt_type_finish | Os]; expand_opt(O, Os) -> [O|Os]. +expand_opt_before_21(Os) -> + [no_put_tuple2, no_get_hd_tl, no_ssa_opt_record, + no_utf8_atoms | expand_opt(no_bsm3, Os)]. + %% format_error(ErrorDescriptor) -> string() -spec format_error(term()) -> iolist(). @@ -247,6 +293,10 @@ format_error(bad_crypto_key) -> "invalid crypto key."; format_error(no_crypto_key) -> "no crypto key supplied."; +format_error({unimplemented_instruction,Instruction}) -> + io_lib:fwrite("native-code compilation failed because of an " + "unimplemented instruction (~s).", + [Instruction]); format_error({native, E}) -> io_lib:fwrite("native-code compilation failed with reason: ~tP.", [E, 25]); @@ -387,17 +437,57 @@ fold_comp([{Name,Pass}|Ps], Run, Code0, St0) -> end; fold_comp([], _Run, Code, St) -> {ok,Code,St}. +run_sub_passes_1([{Name,Run}|Ps], Runner, St0) + when is_atom(Name), is_function(Run, 1) -> + try Runner(Name, Run, St0) of + St -> + run_sub_passes_1(Ps, Runner, St) + catch + C:E:Stk -> + io:format("Sub pass ~s\n", [Name]), + erlang:raise(C, E, Stk) + end; +run_sub_passes_1([], _, St) -> St. + run_tc({Name,Fun}, Code, St) -> + put(?SUB_PASS_TIMES, []), T1 = erlang:monotonic_time(), Val = (catch Fun(Code, St)), T2 = erlang:monotonic_time(), - Elapsed = erlang:convert_time_unit(T2 - T1, native, millisecond), + Times = erase(?SUB_PASS_TIMES), + Elapsed = erlang:convert_time_unit(T2 - T1, native, microsecond), Mem0 = erts_debug:flat_size(Val)*erlang:system_info(wordsize), Mem = lists:flatten(io_lib:format("~.1f kB", [Mem0/1024])), io:format(" ~-30s: ~10.3f s ~12s\n", - [Name,Elapsed/1000,Mem]), + [Name,Elapsed/1000000,Mem]), + print_times(Times, Name), Val. +print_times(Times0, Name) -> + Fam0 = sofs:relation(Times0), + Fam1 = sofs:rel2fam(Fam0), + Fam2 = sofs:to_external(Fam1), + Fam3 = [{W,lists:sum(Times)} || {W,Times} <- Fam2], + Fam = reverse(lists:keysort(2, Fam3)), + Total = case lists:sum([T || {_,T} <- Fam]) of + 0 -> 1; + Total0 -> Total0 + end, + case Fam of + [] -> + ok; + [_|_] -> + io:format(" %% Sub passes of ~s from slowest to fastest:\n", [Name]), + print_times_1(Fam, Total) + end. + +print_times_1([{Name,T}|Ts], Total) -> + Elapsed = erlang:convert_time_unit(T, native, microsecond), + io:format(" ~-27s: ~10.3f s ~3w %\n", + [Name,Elapsed/1000000,round(100*T/Total)]), + print_times_1(Ts, Total); +print_times_1([], _Total) -> ok. + run_eprof({Name,Fun}, Code, Name, St) -> io:format("~p: Running eprof\n", [Name]), c:appcall(tools, eprof, start_profiling, [[self()]]), @@ -731,8 +821,6 @@ kernel_passes() -> %% Optimizations that must be done after all other optimizations. [{pass,sys_core_bsm}, {iff,dcbsm,{listing,"core_bsm"}}, - {pass,sys_core_dsetel}, - {iff,dsetel,{listing,"dsetel"}}, {iff,clint,?pass(core_lint_module)}, {iff,core,?pass(save_core_code)}, @@ -741,8 +829,30 @@ kernel_passes() -> ?pass(v3_kernel), {iff,dkern,{listing,"kernel"}}, {iff,'to_kernel',{done,"kernel"}}, - {pass,v3_codegen}, - {iff,dcg,{listing,"codegen"}} + {pass,beam_kernel_to_ssa}, + {iff,dssa,{listing,"ssa"}}, + {iff,ssalint,{pass,beam_ssa_lint}}, + {delay, + [{unless,no_share_opt,{pass,beam_ssa_share}}, + {iff,dssashare,{listing,"ssashare"}}, + {iff,ssalint,{pass,beam_ssa_lint}}, + {unless,no_bsm_opt,{pass,beam_ssa_bsm}}, + {iff,dssabsm,{listing,"ssabsm"}}, + {iff,ssalint,{pass,beam_ssa_lint}}, + {unless,no_fun_opt,{pass,beam_ssa_funs}}, + {iff,dssafuns,{listing,"ssafuns"}}, + {iff,ssalint,{pass,beam_ssa_lint}}, + {unless,no_ssa_opt,{pass,beam_ssa_opt}}, + {iff,dssaopt,{listing,"ssaopt"}}, + {iff,ssalint,{pass,beam_ssa_lint}}, + {unless,no_recv_opt,{pass,beam_ssa_recv}}, + {iff,drecv,{listing,"recv"}}]}, + {pass,beam_ssa_pre_codegen}, + {iff,dprecg,{listing,"precodegen"}}, + {iff,ssalint,{pass,beam_ssa_lint}}, + {pass,beam_ssa_codegen}, + {iff,dcg,{listing,"codegen"}}, + {iff,doldcg,{listing,"codegen"}} | asm_passes()]. asm_passes() -> @@ -751,34 +861,16 @@ asm_passes() -> [{pass,beam_a}, {iff,da,{listing,"a"}}, {unless,no_postopt, - [{unless,no_reorder,{pass,beam_reorder}}, - {iff,dre,{listing,"reorder"}}, - {pass,beam_block}, + [{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_topt,{pass,beam_type}}, - {iff,dtype,{listing,"type"}}, - {pass,beam_split}, - {iff,dsplit,{listing,"split"}}, - {unless,no_dead,{pass,beam_dead}}, - {iff,ddead,{listing,"dead"}}, {unless,no_jopt,{pass,beam_jump}}, {iff,djmp,{listing,"jump"}}, {unless,no_peep_opt,{pass,beam_peep}}, {iff,dpeep,{listing,"peep"}}, {pass,beam_clean}, {iff,dclean,{listing,"clean"}}, - {unless,no_bsm_opt,{pass,beam_bsm}}, - {iff,dbsm,{listing,"bsm"}}, - {unless,no_recv_opt,{pass,beam_receive}}, - {iff,drecv,{listing,"recv"}}, - {unless,no_record_opt,{pass,beam_record}}, - {iff,drecord,{listing,"record"}}, - {unless,no_blk2,?pass(block2)}, - {iff,dblk2,{listing,"block2"}}, {unless,no_stack_trimming,{pass,beam_trim}}, {iff,dtrim,{listing,"trim"}}, {pass,beam_flatten}]}, @@ -787,7 +879,9 @@ asm_passes() -> %% need to do a few clean-ups to code. {iff,no_postopt,[{pass,beam_clean}]}, + {iff,diffable,?pass(diffable)}, {pass,beam_z}, + {iff,diffable,{listing,"S"}}, {iff,dz,{listing,"z"}}, {iff,dopt,{listing,"optimize"}}, {iff,'S',{listing,"S"}}, @@ -1360,10 +1454,6 @@ v3_kernel(Code0, #compile{options=Opts,warnings=Ws0}=St) -> {ok,Code,St} end. -block2(Code0, #compile{options=Opts}=St) -> - {ok,Code} = beam_block:module(Code0, [no_blockify|Opts]), - {ok,Code,St}. - test_old_inliner(#compile{options=Opts}) -> %% The point of this test is to avoid loading the old inliner %% if we know that it will not be used. @@ -1568,18 +1658,22 @@ native_compile_1(Code, St) -> case IgnoreErrors of true -> Ws = [{St#compile.ifile,[{none,?MODULE,{native,R}}]}], - {ok,St#compile{warnings=St#compile.warnings ++ Ws}}; + {ok,Code,St#compile{warnings=St#compile.warnings ++ Ws}}; false -> Es = [{St#compile.ifile,[{none,?MODULE,{native,R}}]}], {error,St#compile{errors=St#compile.errors ++ Es}} end catch + exit:{unimplemented_instruction,_}=Unimplemented -> + Ws = [{St#compile.ifile, + [{none,?MODULE,Unimplemented}]}], + {ok,Code,St#compile{warnings=St#compile.warnings ++ Ws}}; Class:R:Stack -> case IgnoreErrors of true -> Ws = [{St#compile.ifile, [{none,?MODULE,{native_crash,R,Stack}}]}], - {ok,St#compile{warnings=St#compile.warnings ++ Ws}}; + {ok,Code,St#compile{warnings=St#compile.warnings ++ Ws}}; false -> erlang:raise(Class, R, Stack) end @@ -1849,6 +1943,39 @@ restore_expand_module([F|Fs]) -> [F|restore_expand_module(Fs)]; restore_expand_module([]) -> []. +%%% +%%% Transform the BEAM code to make it more friendly for +%%% diffing: using function names instead of labels for +%%% local calls and number labels relative to each function. +%%% + +diffable(Code0, St) -> + {Mod,Exp,Attr,Fs0,NumLabels} = Code0, + EntryLabels0 = [{Entry,{Name,Arity}} || + {function,Name,Arity,Entry,_} <- Fs0], + EntryLabels = maps:from_list(EntryLabels0), + Fs = [diffable_fix_function(F, EntryLabels) || F <- Fs0], + Code = {Mod,Exp,Attr,Fs,NumLabels}, + {ok,Code,St}. + +diffable_fix_function({function,Name,Arity,Entry0,Is0}, LabelMap0) -> + Entry = maps:get(Entry0, LabelMap0), + {Is1,LabelMap} = diffable_label_map(Is0, 1, LabelMap0, []), + Fb = fun(Old) -> error({no_fb,Old}) end, + Is = beam_utils:replace_labels(Is1, [], LabelMap, Fb), + {function,Name,Arity,Entry,Is}. + +diffable_label_map([{label,Old}|Is], New, Map, Acc) -> + case Map of + #{Old:=NewLabel} -> + diffable_label_map(Is, New, Map, [{label,NewLabel}|Acc]); + #{} -> + diffable_label_map(Is, New+1, Map#{Old=>New}, [{label,New}|Acc]) + end; +diffable_label_map([I|Is], New, Map, Acc) -> + diffable_label_map(Is, New, Map, [I|Acc]); +diffable_label_map([], _New, Map, Acc) -> + {Acc,Map}. -spec options() -> 'ok'. @@ -1969,22 +2096,25 @@ pre_load() -> L = [beam_a, beam_asm, beam_block, - beam_bs, - beam_bsm, beam_clean, - beam_dead, beam_dict, beam_except, beam_flatten, beam_jump, + beam_kernel_to_ssa, beam_opcodes, beam_peep, - beam_receive, - beam_record, - beam_reorder, - beam_split, + beam_ssa, + beam_ssa_bsm, + beam_ssa_codegen, + beam_ssa_dead, + beam_ssa_funs, + beam_ssa_opt, + beam_ssa_pre_codegen, + beam_ssa_recv, + beam_ssa_share, + beam_ssa_type, beam_trim, - beam_type, beam_utils, beam_validator, beam_z, @@ -2001,9 +2131,7 @@ pre_load() -> erl_scan, sys_core_alias, sys_core_bsm, - sys_core_dsetel, sys_core_fold, - v3_codegen, v3_core, v3_kernel], _ = code:ensure_modules_loaded(L), |