hipe.erl

来自「OTP是开放电信平台的简称」· ERL 代码 · 共 1,594 行 · 第 1/4 页

option_text(pp_rtl_ssapre) ->
  "Display the intermediate HiPE-RTL A-SSAPRE sets";
option_text(pp_native) ->
  "Display the generated (back-end specific) native code";
option_text(regalloc) ->
  "Select register allocation algorithm. Used as {regalloc, METHOD}.\n" ++
  "  Currently available methods:\n" ++
  "    naive - spills everything (for debugging and testing)\n" ++
  "    linear_scan - fast; not so good if few registers available\n" ++
  "    graph_color - slow, but gives OK performance\n" ++
  "    coalescing - slower, tries hard to use registers\n" ++
  "    optimistic - another variant of a coalescing allocator";
option_text(remove_comments) ->
  "Strip comments from intermediate code";
option_text(rtl_ssa) ->
  "Perform SSA conversion on the RTL level -- default starting at O2";
option_text(rtl_ssa_const_prop) ->
  "Performs sparse conditional constant propagation on RTL SSA";
option_text(rtl_lcm) ->
  "Perform Lazy Code Motion on RTL";
option_text(rtl_ssapre) ->
  "Perform A-SSAPRE on RTL";
option_text(sparc_peephole) ->
  "Perform Sparc peephole optimization";
option_text(sparc_prop) ->
  "Perform Sparc-level optimization";
option_text(time) ->
  "Reports the compilation times for the different stages\n" ++
  "of the compiler.\n" ++
  "    {time, Module}       reports timings for the module Module.\n" ++
  "    {time, [M1, M2, M3]} reports timings for the specified modules.\n" ++
  "    {time, all}          reports timings all modules.\n" ++
  "    time                 reports timings for the main module.\n";
option_text(timeout) ->
  "Specify compilation time limit in ms. Used as {timeout, LIMIT}.\n" ++
  "    The limit must be a non-negative integer or the atom 'infinity'.\n" ++
  "    The current default limit is 15 minutes (900000 ms).";
option_text(type_warnings) ->
  "Turns on warnings from the icode type propagator";
option_text(use_indexing) ->
  "Use indexing for multiple-choice branch selection.";
option_text(use_callgraph) ->
  "Compile the functions in a module according to a reversed topological " ++
  "sorted order to gain more information when using a persistent lookup " ++
  "table for storing intra-modular type information.";
option_text(verbose) ->
  "Output information about what is being done";
option_text(Opt) when is_atom(Opt) ->
  "".

%% @spec (option()) -> ok
%% @doc Prints documentation about a specific option to the standard
%% output.
help_option(Opt) ->
  set_architecture([]), %% needed for target-specific option expansion
  case expand_options([Opt]) of
    [Opt] ->
      Name = if is_atom(Opt) -> Opt;
		is_tuple(Opt), size(Opt) =:= 2 -> element(1, Opt)
	     end,
      case option_text(Name) of
	"" ->  
	  case lists:member(Name, opt_keys()) of
	    true ->
	      io:format("~w - Sorry, this option is not documented yet.\n",
			[Name]);
	    _ -> 
	      io:format("Unknown option ~p.\n", [Name])
	  end;
	Txt ->
	  io:fwrite("~w - ~s\n", [Name, Txt])
      end;
    Opts ->
      io:fwrite("This is an alias for: ~p.\n", [Opts])
  end,
  ok.

%% @spec () -> ok
%% @doc Prints documentation about debugging options to the standard
%% output.
help_debug_options() ->
  io:format("HiPE compiler debug options:\n" ++
	    "  Might require that some modules have been compiled " ++ 
	    "with the debug flag.\n" ++
	    "    rtl_show_translation - Prints each step in the\n" ++
	    "                           translation from Icode to RTL\n",
	    []),
  ok.

%% @spec (Name) -> bool() 
%% Name = mod() | string()
%% @doc Returns true if the module or file contains native HiPE code.
has_hipe_code(Atom) when is_atom(Atom) ->
  has_hipe_code(atom_to_list(Atom));
has_hipe_code(File) ->
  case catch 
    lists:member($H, 
		 [ hd(ChunkName) ||
		   {ChunkName,_} <- element(3,beam_lib:all_chunks(File))]
		) of
    {'EXIT',_} ->
      exit({bad_beam_file,File});
    R -> R
  end.


%%
hipe_timers() ->
  [time_ra].

%% ____________________________________________________________________
%% 
%% Option expansion

%% These are currently in use, but not documented:
%%
%%     count_instrs:
%%     fill_delayslot:
%%     {hot, Functions}:
%%     icode_type:
%%     icode_range:
%%     {ls_order, Order}:
%%     {regalloc, Algorithm}:
%%     remove_comments
%%     sparc_estimate_block_time
%%     sparc_peephole:
%%     sparc_post_schedule
%%     sparc_profile:
%%     sparc_rename:
%%     sparc_schedule
%%     timeregalloc:
%%     timers
%%     use_indexing

%% Valid option keys. (Don't list aliases or negations - the check is
%% done after the options have been expanded to normal form.)

opt_keys() ->
    ['O',
     bitlevel_binaries,
     concurrent_comp,
     check_for_inlining,
     core,
     core_transform,
     counters,
     count_instrs,
     count_spills,
     count_temps,
     dialyzer,
     debug,
     fill_delayslot,
     frame_x86,
     get_called_modules,
     hot,
     split_arith,
     split_arith_unsafe,
     icode_inline_bifs,
     icode_ssa_check,
     icode_ssa_copy_prop,
     icode_ssa_const_prop,
     icode_ssa_struct_reuse,
     icode_type,
     icode_range,
     icode_multret,
     inline_fp,
     ls_order,
     load,
     measure_regalloc,
     peephole,
     pmatch,
     pp_asm,
     pp_beam,
     pp_icode,
     pp_icode_ssa,
     pp_icode_split_arith,
     pp_opt_icode,
     pp_range_icode,
     pp_typed_icode,
     pp_icode_liveness,
     pp_native,
     pp_rtl,
     pp_rtl_liveness,
     pp_rtl_ssa,
     pp_rtl_lcm,
     pp_rtl_ssapre,
     pp_rtl_linear,
     regalloc,
     remove_comments,
     rtl_ssa,
     rtl_ssa_const_prop,
     rtl_lcm,
     rtl_ssapre,
     rtl_show_translation,
     sparc_estimate_block_time,
     sparc_peephole,
     sparc_post_schedule,
     sparc_profile,
     sparc_prop,
     sparc_schedule,
     sparc_rename,
     spillmin_color,
     target,
     time,
     timeout,
     timeregalloc,
     timers,
     to_rtl,
     type_warnings,
     use_indexing,
     use_inline_atom_search,
     use_callgraph,
     use_clusters,
     use_jumptable,
     verbose,
     %% verbose_spills,
     x87].

%% Definitions: 

o1_opts() ->
  Common = [inline_fp, pmatch, peephole],
  case get(hipe_target_arch) of
    ultrasparc ->
      [sparc_peephole, fill_delayslot | Common];
    powerpc ->
      Common;
    arm ->
      Common -- [inline_fp]; % Pointless optimising for absent hardware
    x86 ->
      [x87 | Common];        % XXX: Temporary until x86 has sse2
    amd64 ->
      Common;
    Arch ->
      ?EXIT({executing_on_an_unsupported_architecture,Arch})
  end.

o2_opts() ->
  Common = [icode_ssa_const_prop, icode_ssa_copy_prop, % icode_ssa_struct_reuse,
	    icode_type, icode_inline_bifs,
	    rtl_ssa, rtl_ssa_const_prop,
	    spillmin_color, use_indexing, remove_comments, 
	    concurrent_comp | o1_opts()],
  case get(hipe_target_arch) of
    ultrasparc ->
      [sparc_prop | Common];	% no rtl_lcm here; untagged values over GC...
    powerpc ->
      [rtl_lcm | Common];
    arm ->
      [rtl_lcm | Common];
    x86 ->
      [rtl_lcm | Common];
      % [rtl_ssapre | Common];
    amd64 ->
      [rtl_lcm, icode_range | Common]; % range analysis is effective on 64 bits
    Arch ->
      ?EXIT({executing_on_an_unsupported_architecture,Arch})
  end.

o3_opts() ->
  Common = [icode_range, {regalloc,coalescing} | o2_opts()],
  case get(hipe_target_arch) of
    ultrasparc ->
      Common;
    powerpc ->
      Common;
    arm ->
      Common;
    x86 ->
      Common;
    amd64 ->
      Common;
    Arch ->
      ?EXIT({executing_on_an_unsupported_architecture,Arch})
  end.

%% Note that in general, the normal form for options should be positive.
%% This is a good programming convention, so that tests in the code say
%% "if 'x' ..." instead of "if not 'no_x' ...".

opt_negations() ->
  [{no_bitlevel_binaries, bitlevel_binaries},
   {no_core, core},
   {no_debug, debug},
   {no_fill_delayslot, fill_delayslot},
   {no_get_called_modules, get_called_modules},
   {no_split_arith, split_arith},
   {no_concurrent_comp, concurrent_comp},
   {no_icode_inline_bifs, icode_inline_bifs},
   {no_icode_range, icode_range},
   {no_icode_split_arith, icode_split_arith},
   {no_icode_ssa_check, icode_ssa_check},
   {no_icode_ssa_copy_prop, icode_ssa_copy_prop},
   {no_icode_ssa_const_prop, icode_ssa_const_prop},
   {no_icode_ssa_struct_reuse, icode_ssa_struct_reuse},
   {no_icode_type, icode_type},
   {no_icode_range, icode_range},
   {no_inline_fp, inline_fp},
   {no_load, load},
   {no_peephole, peephole},
   {no_pmatch, pmatch},
   {no_pp_beam, pp_beam},
   {no_pp_icode, pp_icode},
   {no_pp_icode_ssa, pp_icode_ssa},
   {no_pp_opt_icode, pp_opt_icode},
   {no_pp_typed_icode, pp_typed_icode},
   {no_pp_rtl, pp_rtl},
   {no_pp_native, pp_native},
   {no_pp_rtl_lcm, pp_rtl_lcm},
   {no_pp_rtl_ssapre, pp_rtl_ssapre},
   {no_remove_comments, remove_comments},
   {no_rtl_ssa, rtl_ssa},
   {no_rtl_ssa_const_prop, rtl_ssa_const_prop},
   {no_rtl_lcm, rtl_lcm},
   {no_rtl_ssapre, rtl_ssapre},
   {no_rtl_show_translation, rtl_show_translation},
   {no_sparc_estimate_block_time, sparc_estimate_block_time},
   {no_sparc_peephole, sparc_peephole},
   {no_sparc_post_schedule, sparc_post_schedule},
   {no_sparc_profile, sparc_profile},
   {no_sparc_prop, sparc_prop},
   {no_sparc_schedule, sparc_schedule},
   {no_time, time},
   {no_type_warnings, type_warnings},
   {no_use_callgraph, use_callgraph},
   {no_use_clusters, use_clusters},
   {no_use_inline_atom_search, use_inline_atom_search},
   {no_use_indexing, use_indexing}].

%% Don't use negative forms in right-hand sides of aliases and expansions!
%% We only expand negations once, before the other expansions are done.

opt_aliases() ->
  [{'O0', o0},
   {'O1', o1},
   {'O2', o2},
   {'O3', o3},
   {pp_sparc, pp_native},
   {pp_x86, pp_native},
   {pp_amd64, pp_native},
   {pp_ppc, pp_native}].

opt_basic_expansions() ->
  [{pp_all, [pp_beam, pp_icode, pp_rtl, pp_native]}].

opt_pre_expansions() ->
  [{o0, [{'O', 0}]},
   {o1, [{'O', 1}]},
   {o2, [{'O', 2}]},
   {o3, [{'O', 3}]}].

opt_expansions() ->
  [{{'O', 1}, [{'O', 1} | o1_opts()]},
   {{'O', 2}, [{'O', 2} | o2_opts()]},
   {{'O', 3}, [{'O', 3} | o3_opts()]},
   {x87, [x87, inline_fp]},
   {inline_fp, case get(hipe_target_arch) of %% XXX: Temporary until x86
		 x86 -> [x87, inline_fp];    %%       has sse2
		 _ -> [inline_fp] end}].

%% This expands "basic" options, which may be tested early and cannot be
%% in conflict with options found in the source code.

expand_basic_options(Opts) ->
  proplists:normalize(Opts, [{negations, opt_negations()},
			     {aliases, opt_aliases()},
			     {expand, opt_basic_expansions()}]).

expand_kt2(Opts) -> 
  proplists:normalize(Opts, [{expand, [{kt2_type,
					[{use_callgraph, fixpoint}, core, 
					 {core_transform, cerl_typean}]}]}]).

%% Note that set_architecture/1 must be called first, and that the given
%% list should contain the total set of options, since things like 'o2'
%% are expanded here. Basic expansions are processed here also, since
%% this function is called from the help-functions.

expand_options(Opts) ->
  proplists:normalize(Opts, [{negations, opt_negations()},
			     {aliases, opt_aliases()},
			     {expand, opt_basic_expansions()},
			     {expand, opt_pre_expansions()},
			     {expand, opt_expansions()}]).

check_options(Opts) ->
  Keys = ordsets:from_list(opt_keys()),
  Used = ordsets:from_list(proplists:get_keys(Opts)),
  case ordsets:subtract(Used, Keys) of
    [] ->
      ok;
    L ->
      ?WARNING_MSG("Unknown options: ~p.\n", [L]),
      ok
  end.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%