[lld/LTO] Add support for optimization remarks

[davide]

Summary

Quite easy now that the yak is shaved.
For info on the feature: http://llvm.org/devmtg/2016-11/Slides/Nemet-Compiler-assistedPerformanceAnalysis.pdf
Makes much easier to visualize optimization decisions (loop vectorizations, inlining decisions etc..). I plan to use it for tuning profile-guided inlining.
This was only supported for Mach-O, but I happen to be interested in another object format.

[ruiu]

Generally looking file, but I'll let people who are familiar with this feature to sign off.

What is expected to happen if you pass only -opt-remarks-filename but no -opt-remarks-with-hotness?

[anemet]

Indeed, it looks very easy. LGTM from the opt-diat point of view.

[davide]

In D29878#675147, @ruiu wrote:

Generally looking file, but I'll let people who are familiar with this feature to sign off.

What is expected to happen if you pass only -opt-remarks-filename but no -opt-remarks-with-hotness?

It happens that you generate a remarks filename without hotness informations.

[davide]

In D29878#675162, @davide wrote:

In D29878#675147, @ruiu wrote:

Generally looking file, but I'll let people who are familiar with this feature to sign off.

What is expected to happen if you pass only -opt-remarks-filename but no -opt-remarks-with-hotness?

It happens that you generate a remarks filename without hotness informations.

*remarks file.

[anemet]

In D29878#675162, @davide wrote:

In D29878#675147, @ruiu wrote:

Generally looking file, but I'll let people who are familiar with this feature to sign off.

What is expected to happen if you pass only -opt-remarks-filename but no -opt-remarks-with-hotness?

It happens that you generate a remarks filename without hotness informations.

The idea being that without hotness the opt-diag emitter pass does not require BlockFrequencyInfo which may cost some comple-time.

[ruiu]

LGTM

ELF/Config.h

struct Configuration {
   llvm::StringRef Entry;
   llvm::StringRef Emulation;
   llvm::StringRef Fini;
   llvm::StringRef Init;
   llvm::StringRef LTOAAPipeline;
   llvm::StringRef LTONewPmPasses;
   llvm::StringRef MapFile;
   llvm::StringRef OutputFile;
+  llvm::StringRef OptRemarksFilename;
   llvm::StringRef SoName;
   llvm::StringRef Sysroot;
   std::string RPath;
   std::vector<VersionDefinition> VersionDefinitions;
   std::vector<llvm::StringRef> AuxiliaryList;
   std::vector<llvm::StringRef> SearchPaths;
   std::vector<llvm::StringRef> SymbolOrderingFile;
   std::vector<llvm::StringRef> Undefined;
struct Configuration {
   bool ICF;
   bool Mips64EL = false;
   bool MipsN32Abi = false;
   bool NoGnuUnique;
   bool NoUndefinedVersion;
   bool Nostdlib;
   bool OFormatBinary;
   bool OMagic;
+  bool OptRemarksWithHotness;
   bool Pic;
   bool Pie;
   bool PrintGcSections;
   bool Rela;
   bool Relocatable;
   bool SaveTemps;
   bool SingleRoRx;
   bool Shared;

ELF/Driver.cpp

       getArg(Args, OPT_fatal_warnings, OPT_no_fatal_warnings, false);
   Config->GcSections = getArg(Args, OPT_gc_sections, OPT_no_gc_sections, false);
   Config->GdbIndex = Args.hasArg(OPT_gdb_index);
   Config->ICF = Args.hasArg(OPT_icf);
   Config->NoGnuUnique = Args.hasArg(OPT_no_gnu_unique);
   Config->NoUndefinedVersion = Args.hasArg(OPT_no_undefined_version);
   Config->Nostdlib = Args.hasArg(OPT_nostdlib);
   Config->OMagic = Args.hasArg(OPT_omagic);
+  Config->OptRemarksWithHotness = Args.hasArg(OPT_opt_remarks_with_hotness);
   Config->Pie = getArg(Args, OPT_pie, OPT_nopie, false);
   Config->PrintGcSections = Args.hasArg(OPT_print_gc_sections);
   Config->Relocatable = Args.hasArg(OPT_relocatable);
   Config->DefineCommon = getArg(Args, OPT_define_common, OPT_no_define_common,
                                 !Config->Relocatable);
   Config->Discard = getDiscardOption(Args);
   Config->SaveTemps = Args.hasArg(OPT_save_temps);
   Config->SingleRoRx = Args.hasArg(OPT_no_rosegment);
   Config->Shared = Args.hasArg(OPT_shared);
   Config->Target1Rel = getArg(Args, OPT_target1_rel, OPT_target1_abs, false);
   Config->Threads = getArg(Args, OPT_threads, OPT_no_threads, true);
   Config->Trace = Args.hasArg(OPT_trace);
   Config->Verbose = Args.hasArg(OPT_verbose);
   Config->WarnCommon = Args.hasArg(OPT_warn_common);
 
   Config->DynamicLinker = getString(Args, OPT_dynamic_linker);
   Config->Entry = getString(Args, OPT_entry);
   Config->Fini = getString(Args, OPT_fini, "_fini");
   Config->Init = getString(Args, OPT_init, "_init");
   Config->LTOAAPipeline = getString(Args, OPT_lto_aa_pipeline);
   Config->LTONewPmPasses = getString(Args, OPT_lto_newpm_passes);
   Config->MapFile = getString(Args, OPT_Map);
+  Config->OptRemarksFilename = getString(Args, OPT_opt_remarks_filename);
   Config->OutputFile = getString(Args, OPT_o);
   Config->SoName = getString(Args, OPT_soname);
   Config->Sysroot = getString(Args, OPT_sysroot);
 
   Config->Optimize = getInteger(Args, OPT_O, 1);
   Config->LTOO = getInteger(Args, OPT_lto_O, 2);
   if (Config->LTOO > 3)
     error("invalid optimization level for LTO: " + getString(Args, OPT_lto_O));

ELF/LTO.cpp

static std::unique_ptr<lto::LTO> createLTO() {
   Conf.DisableVerify = Config->DisableVerify;
   Conf.DiagHandler = diagnosticHandler;
   Conf.OptLevel = Config->LTOO;
 
   // Set up a custom pipeline if we've been asked to.
   Conf.OptPipeline = Config->LTONewPmPasses;
   Conf.AAPipeline = Config->LTOAAPipeline;
 
+  // Set up optimization remarks if we've been asked to.
+  Conf.RemarksFilename = Config->OptRemarksFilename;
+  Conf.RemarksWithHotness = Config->OptRemarksWithHotness;
+
   if (Config->SaveTemps)
     checkError(Conf.addSaveTemps(std::string(Config->OutputFile) + ".",
                                  /*UseInputModulePath*/ true));
 
   lto::ThinBackend Backend;
   if (Config->ThinLTOJobs != -1u)
     Backend = lto::createInProcessThinBackend(Config->ThinLTOJobs);
   return llvm::make_unique<lto::LTO>(std::move(Conf), Backend,

ELF/Options.td

 def lto_aa_pipeline: J<"lto-aa-pipeline=">,
   HelpText<"AA pipeline to run during LTO. Used in conjunction with -lto-newpm-passes">;
 def lto_newpm_passes: J<"lto-newpm-passes=">,
   HelpText<"Passes to run during LTO">;
 def lto_partitions: J<"lto-partitions=">,
   HelpText<"Number of LTO codegen partitions">;
 def disable_verify: F<"disable-verify">;
 def mllvm: S<"mllvm">;
+def opt_remarks_filename: S<"opt-remarks-filename">,
+  HelpText<"YAML output file for optimization remarks">;
+def opt_remarks_with_hotness: F<"opt-remarks-with-hotness">,
+  HelpText<"Include hotness informations in the optimization remarks file">;
 def save_temps: F<"save-temps">;
 def thinlto_jobs: J<"thinlto-jobs=">, HelpText<"Number of ThinLTO jobs">;

test/ELF/lto/opt-remarks.ll

+; RUN: llvm-as %s -o %t.o
+
+; RUN: rm -f %t.yaml
+; RUN: ld.lld -opt-remarks-filename %t.yaml %t.o -o %t -shared -save-temps
+; RUN: llvm-dis %t.0.4.opt.bc -o - | FileCheck %s
+; RUN: ld.lld -opt-remarks-with-hotness -opt-remarks-filename %t.hot.yaml \
+; RUN:  %t.o -o %t -shared
+; RUN: cat %t.yaml | FileCheck %s -check-prefix=YAML
+; RUN: cat %t.hot.yaml | FileCheck %s -check-prefix=YAML-HOT
+
+; Check that @tinkywinky is inlined after optimizations.
+; CHECK-LABEL: define i32 @main
+; CHECK-NEXT:  %a.i = call i32 @patatino()
+; CHECK-NEXT:  ret i32 %a.i
+; CHECK-NEXT: }
+
+; YAML: --- !Analysis
+; YAML-NEXT: Pass:            inline
+; YAML-NEXT: Name:            CanBeInlined
+; YAML-NEXT: Function:        main
+; YAML-NEXT: Args:
+; YAML-NEXT:   - Callee:          tinkywinky
+; YAML-NEXT:   - String:          ' can be inlined into '
+; YAML-NEXT:   - Caller:          main
+; YAML-NEXT:   - String:          ' with cost='
+; YAML-NEXT:   - Cost:            '0'
+; YAML-NEXT:   - String:          ' (threshold='
+; YAML-NEXT:   - Threshold:       '337'
+; YAML-NEXT:   - String:          ')'
+; YAML-NEXT: ...
+; YAML-NEXT: --- !Passed
+; YAML-NEXT: Pass:            inline
+; YAML-NEXT: Name:            Inlined
+; YAML-NEXT: Function:        main
+; YAML-NEXT: Args:
+; YAML-NEXT:   - Callee:          tinkywinky
+; YAML-NEXT:   - String:          ' inlined into '
+; YAML-NEXT:   - Caller:          main
+; YAML-NEXT: ...
+
+; YAML-HOT: ...
+; YAML-HOT: --- !Passed
+; YAML-HOT: Pass:            inline
+; YAML-HOT-NEXT: Name:            Inlined
+; YAML-HOT-NEXT: Function:        main
+; YAML-HOT-NEXT: Hotness:         300
+; YAML-HOT-NEXT: Args:
+; YAML-HOT-NEXT:   - Callee:          tinkywinky
+; YAML-HOT-NEXT:   - String:          ' inlined into '
+; YAML-HOT-NEXT:   - Caller:          main
+; YAML-HOT-NEXT: ...
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-scei-ps4"
+
+declare i32 @patatino()
+
+define i32 @tinkywinky() {
+  %a = call i32 @patatino()
+  ret i32 %a
+}
+
+define i32 @main() !prof !0 {
+  %i = call i32 @tinkywinky()
+  ret i32 %i
+}
+
+!0 = !{!"function_entry_count", i64 300}