[Remarks][2/2] Expand remarks hotness threshold option support in more tools

apolloww · apolloww · commit 93dc1b5b8cb2 · 2020-11-30T21:55:50.000-08:00
This is the #2 of 2 changes that make remarks hotness threshold option available in more tools. The changes also allow the threshold to sync with hotness threshold from profile summary with special value 'auto'. This change expands remarks hotness threshold option -fdiagnostics-hotness-threshold in clang and *-remarks-hotness-threshold in other tools to utilize hotness threshold from profile summary. Remarks hotness filtering relies on several driver options. Table below lists how different options are correlated and affect final remarks outputs: | profile | hotness | threshold | remarks printed | |---------|---------|-----------|-----------------| | No | No | No | All | | No | No | Yes | None | | No | Yes | No | All | | No | Yes | Yes | None | | Yes | No | No | All | | Yes | No | Yes | None | | Yes | Yes | No | All | | Yes | Yes | Yes | >=threshold | In the presence of profile summary, it is often more desirable to directly use the hotness threshold from profile summary. The new argument value 'auto' indicates threshold will be synced with hotness threshold from profile summary during compilation. The "auto" threshold relies on the availability of profile summary. In case of missing such information, no remarks will be generated. Differential Revision: https://reviews.llvm.org/D85808
diff --git a/clang/include/clang/Basic/CodeGenOptions.def b/clang/include/clang/Basic/CodeGenOptions.def
@@ -366,10 +366,6 @@ VALUE_CODEGENOPT(EmitCheckPathComponentsToStrip, 32, 0)
 /// Whether to report the hotness of the code region for optimization remarks.
 CODEGENOPT(DiagnosticsWithHotness, 1, 0)
 
-/// The minimum hotness value a diagnostic needs in order to be included in
-/// optimization diagnostics.
-VALUE_CODEGENOPT(DiagnosticsHotnessThreshold, 32, 0)
-
 /// Whether copy relocations support is available when building as PIE.
 CODEGENOPT(PIECopyRelocations, 1, 0)
 
diff --git a/clang/include/clang/Basic/CodeGenOptions.h b/clang/include/clang/Basic/CodeGenOptions.h
@@ -346,6 +346,21 @@ class CodeGenOptions : public CodeGenOptionsBase {
   const char *Argv0 = nullptr;
   ArrayRef<const char *> CommandLineArgs;
 
+  /// The minimum hotness value a diagnostic needs in order to be included in
+  /// optimization diagnostics.
+  ///
+  /// The threshold is an Optional value, which maps to one of the 3 states:
+  /// 1. 0            => threshold disabled. All remarks will be printed.
+  /// 2. positive int => manual threshold by user. Remarks with hotness exceed
+  ///                    threshold will be printed.
+  /// 3. None         => 'auto' threshold by user. The actual value is not
+  ///                    available at command line, but will be synced with
+  ///                    hotness threshold from profile summary during
+  ///                    compilation.
+  ///
+  /// If threshold option is not specified, it is disabled by default.
+  Optional<uint64_t> DiagnosticsHotnessThreshold = 0;
+
 public:
   // Define accessors/mutators for code generation options of enumeration type.
 #define CODEGENOPT(Name, Bits, Default)
diff --git a/clang/include/clang/Basic/DiagnosticDriverKinds.td b/clang/include/clang/Basic/DiagnosticDriverKinds.td
@@ -117,6 +117,8 @@ def err_drv_command_failure : Error<
   "unable to execute command: %0">;
 def err_drv_invalid_darwin_version : Error<
   "invalid Darwin version number: %0">;
+def err_drv_invalid_diagnotics_hotness_threshold : Error<
+  "invalid argument in '%0', only integer or 'auto' is supported">;
 def err_drv_missing_argument : Error<
   "argument to '%0' is missing (expected %1 value%s1)">;
 def err_drv_invalid_Xarch_argument_with_args : Error<
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
@@ -984,8 +984,9 @@ def fdiagnostics_print_source_range_info : Flag<["-"], "fdiagnostics-print-sourc
 def fdiagnostics_show_hotness : Flag<["-"], "fdiagnostics-show-hotness">, Group<f_Group>,
     Flags<[CC1Option]>, HelpText<"Enable profile hotness information in diagnostic line">;
 def fdiagnostics_hotness_threshold_EQ : Joined<["-"], "fdiagnostics-hotness-threshold=">,
-    Group<f_Group>, Flags<[CC1Option]>, MetaVarName<"<number>">,
-    HelpText<"Prevent optimization remarks from being output if they do not have at least this profile count">;
+    Group<f_Group>, Flags<[CC1Option]>, MetaVarName<"<value>">,
+    HelpText<"Prevent optimization remarks from being output if they do not have at least this profile count. "
+    "Use 'auto' to apply the threshold from profile summary">;
 def fdiagnostics_show_option : Flag<["-"], "fdiagnostics-show-option">, Group<f_Group>,
     HelpText<"Print option name with mappable diagnostics">;
 def fdiagnostics_show_note_include_stack : Flag<["-"], "fdiagnostics-show-note-include-stack">,
diff --git a/clang/lib/Frontend/CompilerInvocation.cpp b/clang/lib/Frontend/CompilerInvocation.cpp
@@ -14,6 +14,7 @@
 #include "clang/Basic/CommentOptions.h"
 #include "clang/Basic/DebugInfoOptions.h"
 #include "clang/Basic/Diagnostic.h"
+#include "clang/Basic/DiagnosticDriver.h"
 #include "clang/Basic/DiagnosticOptions.h"
 #include "clang/Basic/FileSystemOptions.h"
 #include "clang/Basic/LLVM.h"
@@ -66,6 +67,7 @@
 #include "llvm/Option/OptTable.h"
 #include "llvm/Option/Option.h"
 #include "llvm/ProfileData/InstrProfReader.h"
+#include "llvm/Remarks/HotnessThresholdParser.h"
 #include "llvm/Support/CodeGen.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Error.h"
@@ -1501,11 +1503,24 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
     Diags.Report(diag::warn_drv_diagnostics_hotness_requires_pgo)
         << "-fdiagnostics-show-hotness";
 
-  Opts.DiagnosticsHotnessThreshold = getLastArgUInt64Value(
-      Args, options::OPT_fdiagnostics_hotness_threshold_EQ, 0);
-  if (Opts.DiagnosticsHotnessThreshold > 0 && !UsingProfile)
-    Diags.Report(diag::warn_drv_diagnostics_hotness_requires_pgo)
-        << "-fdiagnostics-hotness-threshold=";
+  // Parse remarks hotness threshold. Valid value is either integer or 'auto'.
+  if (auto *arg =
+          Args.getLastArg(options::OPT_fdiagnostics_hotness_threshold_EQ)) {
+    auto ResultOrErr =
+        llvm::remarks::parseHotnessThresholdOption(arg->getValue());
+
+    if (!ResultOrErr) {
+      Diags.Report(diag::err_drv_invalid_diagnotics_hotness_threshold)
+          << "-fdiagnostics-hotness-threshold=";
+    } else {
+      Opts.DiagnosticsHotnessThreshold = *ResultOrErr;
+      if ((!Opts.DiagnosticsHotnessThreshold.hasValue() ||
+           Opts.DiagnosticsHotnessThreshold.getValue() > 0) &&
+          !UsingProfile)
+        Diags.Report(diag::warn_drv_diagnostics_hotness_requires_pgo)
+            << "-fdiagnostics-hotness-threshold=";
+    }
+  }
 
   // If the user requested to use a sample profile for PGO, then the
   // backend will need to track source location information so the profile
diff --git a/clang/test/Driver/opt-record.c b/clang/test/Driver/opt-record.c
@@ -51,6 +51,7 @@
 // RUN: %clang -target x86_64-linux -### -o FOO -fuse-ld=gold -flto -fdiagnostics-hotness-threshold=100 -fsave-optimization-record -foptimization-record-passes=inline %s 2>&1 | FileCheck %s -check-prefix=CHECK-PASS
 // RUN: %clang -target x86_64-linux -### -o FOO -fuse-ld=lld -flto=thin -fdiagnostics-hotness-threshold=100 -fsave-optimization-record=some-format -foptimization-record-file=FOO.txt %s 2>&1 | FileCheck %s -check-prefix=CHECK-PASS-CUSTOM
 // RUN: %clang -target x86_64-linux -### -o FOO -fuse-ld=lld -flto=thin -fdiagnostics-hotness-threshold=100 -Rpass=inline -Rpass-missed=inline -Rpass-analysis=inline %s 2>&1 | FileCheck %s -check-prefix=CHECK-PASS-RPASS
+// RUN: %clang -target x86_64-linux -### -o FOO -fuse-ld=lld -flto=thin -fdiagnostics-hotness-threshold=auto -Rpass=inline -Rpass-missed=inline -Rpass-analysis=inline %s 2>&1 | FileCheck %s -check-prefix=CHECK-PASS-AUTO
 
 // CHECK-NOPASS-NOT: "--plugin-opt=opt-remarks-filename="
 // CHECK-NOPASS-NOT: "--plugin-opt=opt-remarks-passes=inline"
@@ -75,3 +76,5 @@
 // CHECK-PASS-RPASS-SAME: "--plugin-opt=-pass-remarks-missed=inline"
 // CHECK-PASS-RPASS-SAME: "--plugin-opt=-pass-remarks-analysis=inline"
 // CHECK-PASS-RPASS-SAME: "--plugin-opt=opt-remarks-hotness-threshold=100"
+
+// CHECK-PASS-AUTO:   "--plugin-opt=opt-remarks-hotness-threshold=auto"
diff --git a/clang/test/Frontend/Inputs/remarks-hotness.prof b/clang/test/Frontend/Inputs/remarks-hotness.prof
@@ -0,0 +1,8 @@
+_Z7callee1v:600:600
+ 1: 600
+_Z7callee2v:1:1
+ 1: 1
+_Z7caller1v:400:400
+ 1: 400
+_Z7caller2v:1:1
+ 1: 1
diff --git a/clang/test/Frontend/remarks-hotness.cpp b/clang/test/Frontend/remarks-hotness.cpp
@@ -0,0 +1,34 @@
+// Without hotness threshold, print both hot and cold remarks.
+// RUN: %clang_cc1 -triple x86_64-linux %s -emit-llvm-only -O3 \
+// RUN:     -fprofile-sample-use=%S/Inputs/remarks-hotness.prof \
+// RUN:     -Rpass=inline -Rpass-analysis=inline -Rpass-missed=inline \
+// RUN:     -fexperimental-new-pass-manager -fdiagnostics-show-hotness 2>&1 \
+// RUN:     | FileCheck -check-prefix=REMARKS %s
+
+// With auto hotness threshold, only print hot remarks.
+// RUN: %clang_cc1 -triple x86_64-linux %s -emit-llvm-only -O3 \
+// RUN:     -fprofile-sample-use=%S/Inputs/remarks-hotness.prof \
+// RUN:     -Rpass=inline -Rpass-analysis=inline -Rpass-missed=inline \
+// RUN:     -fexperimental-new-pass-manager -fdiagnostics-show-hotness \
+// RUN:     -fdiagnostics-hotness-threshold=auto 2>&1 \
+// RUN:     | FileCheck -check-prefix=HOT_CALL %s
+
+int callee1() {
+  return 1;
+}
+
+__attribute__((noinline)) int callee2() {
+  return 2;
+}
+
+// REMARKS: _Z7callee1v inlined into _Z7caller1v
+// HOT_CALL: _Z7callee1v inlined into _Z7caller1v
+int caller1() {
+  return callee1();
+}
+
+// REMARKS: _Z7callee2v not inlined into _Z7caller2v
+// HOT_CALL-NOT: _Z7callee2v not inlined into _Z7caller2v
+int caller2() {
+  return callee2();
+}
diff --git a/llvm/include/llvm/Analysis/ProfileSummaryInfo.h b/llvm/include/llvm/Analysis/ProfileSummaryInfo.h
@@ -38,7 +38,7 @@ class Function;
 // units. This would require making this depend on BFI.
 class ProfileSummaryInfo {
 private:
-  Module &M;
+  const Module &M;
   std::unique_ptr<ProfileSummary> Summary;
   void computeThresholds();
   // Count thresholds to answer isHotCount and isColdCount queries.
@@ -58,7 +58,8 @@ class ProfileSummaryInfo {
   mutable DenseMap<int, uint64_t> ThresholdCache;
 
 public:
-  ProfileSummaryInfo(Module &M) : M(M) { refresh(); }
+  ProfileSummaryInfo(const Module &M) : M(M) { refresh(); }
+
   ProfileSummaryInfo(ProfileSummaryInfo &&Arg) = default;
 
   /// If no summary is present, attempt to refresh.
diff --git a/llvm/include/llvm/IR/LLVMContext.h b/llvm/include/llvm/IR/LLVMContext.h
@@ -237,6 +237,9 @@ class LLVMContext {
   /// included in optimization diagnostics.
   void setDiagnosticsHotnessThreshold(Optional<uint64_t> Threshold);
 
+  /// Return if hotness threshold is requested from PSI.
+  bool isDiagnosticsHotnessThresholdSetFromPSI() const;
+
   /// The "main remark streamer" used by all the specialized remark streamers.
   /// This streamer keeps generic remark metadata in memory throughout the life
   /// of the LLVMContext. This metadata may be emitted in a section in object
diff --git a/llvm/include/llvm/IR/Module.h b/llvm/include/llvm/IR/Module.h
@@ -850,7 +850,7 @@ class Module {
 
   /// Returns profile summary metadata. When IsCS is true, use the context
   /// sensitive profile summary.
-  Metadata *getProfileSummary(bool IsCS);
+  Metadata *getProfileSummary(bool IsCS) const;
   /// @}
 
   /// Returns whether semantic interposition is to be respected.
diff --git a/llvm/lib/Analysis/OptimizationRemarkEmitter.cpp b/llvm/lib/Analysis/OptimizationRemarkEmitter.cpp
@@ -15,6 +15,7 @@
 #include "llvm/Analysis/BranchProbabilityInfo.h"
 #include "llvm/Analysis/LazyBlockFrequencyInfo.h"
 #include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Analysis/ProfileSummaryInfo.h"
 #include "llvm/IR/DiagnosticInfo.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/LLVMContext.h"
@@ -96,9 +97,17 @@ OptimizationRemarkEmitterWrapperPass::OptimizationRemarkEmitterWrapperPass()
 bool OptimizationRemarkEmitterWrapperPass::runOnFunction(Function &Fn) {
   BlockFrequencyInfo *BFI;
 
-  if (Fn.getContext().getDiagnosticsHotnessRequested())
+  auto &Context = Fn.getContext();
+  if (Context.getDiagnosticsHotnessRequested()) {
     BFI = &getAnalysis<LazyBlockFrequencyInfoPass>().getBFI();
-  else
+    // Get hotness threshold from PSI. This should only happen once.
+    if (Context.isDiagnosticsHotnessThresholdSetFromPSI()) {
+      if (ProfileSummaryInfo *PSI =
+              &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI())
+        Context.setDiagnosticsHotnessThreshold(
+            PSI->getOrCompHotCountThreshold());
+    }
+  } else
     BFI = nullptr;
 
   ORE = std::make_unique<OptimizationRemarkEmitter>(&Fn, BFI);
@@ -108,6 +117,7 @@ bool OptimizationRemarkEmitterWrapperPass::runOnFunction(Function &Fn) {
 void OptimizationRemarkEmitterWrapperPass::getAnalysisUsage(
     AnalysisUsage &AU) const {
   LazyBlockFrequencyInfoPass::getLazyBFIAnalysisUsage(AU);
+  AU.addRequired<ProfileSummaryInfoWrapperPass>();
   AU.setPreservesAll();
 }
 
@@ -117,10 +127,19 @@ OptimizationRemarkEmitter
 OptimizationRemarkEmitterAnalysis::run(Function &F,
                                        FunctionAnalysisManager &AM) {
   BlockFrequencyInfo *BFI;
+  auto &Context = F.getContext();
 
-  if (F.getContext().getDiagnosticsHotnessRequested())
+  if (Context.getDiagnosticsHotnessRequested()) {
     BFI = &AM.getResult<BlockFrequencyAnalysis>(F);
-  else
+    // Get hotness threshold from PSI. This should only happen once.
+    if (Context.isDiagnosticsHotnessThresholdSetFromPSI()) {
+      auto &MAMProxy = AM.getResult<ModuleAnalysisManagerFunctionProxy>(F);
+      if (ProfileSummaryInfo *PSI =
+              MAMProxy.getCachedResult<ProfileSummaryAnalysis>(*F.getParent()))
+        Context.setDiagnosticsHotnessThreshold(
+            PSI->getOrCompHotCountThreshold());
+    }
+  } else
     BFI = nullptr;
 
   return OptimizationRemarkEmitter(&F, BFI);
@@ -133,5 +152,6 @@ static const char ore_name[] = "Optimization Remark Emitter";
 INITIALIZE_PASS_BEGIN(OptimizationRemarkEmitterWrapperPass, ORE_NAME, ore_name,
                       false, true)
 INITIALIZE_PASS_DEPENDENCY(LazyBFIPass)
+INITIALIZE_PASS_DEPENDENCY(ProfileSummaryInfoWrapperPass)
 INITIALIZE_PASS_END(OptimizationRemarkEmitterWrapperPass, ORE_NAME, ore_name,
                     false, true)
diff --git a/llvm/lib/IR/LLVMContext.cpp b/llvm/lib/IR/LLVMContext.cpp
@@ -149,10 +149,15 @@ bool LLVMContext::getDiagnosticsHotnessRequested() const {
 void LLVMContext::setDiagnosticsHotnessThreshold(Optional<uint64_t> Threshold) {
   pImpl->DiagnosticsHotnessThreshold = Threshold;
 }
+
 uint64_t LLVMContext::getDiagnosticsHotnessThreshold() const {
   return pImpl->DiagnosticsHotnessThreshold.getValueOr(UINT64_MAX);
 }
 
+bool LLVMContext::isDiagnosticsHotnessThresholdSetFromPSI() const {
+  return !pImpl->DiagnosticsHotnessThreshold.hasValue();
+}
+
 remarks::RemarkStreamer *LLVMContext::getMainRemarkStreamer() {
   return pImpl->MainRemarkStreamer.get();
 }
diff --git a/llvm/lib/IR/LLVMRemarkStreamer.cpp b/llvm/lib/IR/LLVMRemarkStreamer.cpp
@@ -96,8 +96,7 @@ Expected<std::unique_ptr<ToolOutputFile>> llvm::setupLLVMOptimizationRemarks(
   if (RemarksWithHotness)
     Context.setDiagnosticsHotnessRequested(true);
 
-  if (RemarksHotnessThreshold)
-    Context.setDiagnosticsHotnessThreshold(RemarksHotnessThreshold);
+  Context.setDiagnosticsHotnessThreshold(RemarksHotnessThreshold);
 
   if (RemarksFilename.empty())
     return nullptr;
@@ -144,8 +143,7 @@ Error llvm::setupLLVMOptimizationRemarks(
   if (RemarksWithHotness)
     Context.setDiagnosticsHotnessRequested(true);
 
-  if (RemarksHotnessThreshold)
-    Context.setDiagnosticsHotnessThreshold(RemarksHotnessThreshold);
+  Context.setDiagnosticsHotnessThreshold(RemarksHotnessThreshold);
 
   Expected<remarks::Format> Format = remarks::parseFormat(RemarksFormat);
   if (Error E = Format.takeError())
diff --git a/llvm/lib/IR/Module.cpp b/llvm/lib/IR/Module.cpp
@@ -582,7 +582,7 @@ void Module::setProfileSummary(Metadata *M, ProfileSummary::Kind Kind) {
     setModuleFlag(ModFlagBehavior::Error, "ProfileSummary", M);
 }
 
-Metadata *Module::getProfileSummary(bool IsCS) {
+Metadata *Module::getProfileSummary(bool IsCS) const {
   return (IsCS ? getModuleFlag("CSProfileSummary")
                : getModuleFlag("ProfileSummary"));
 }
diff --git a/llvm/test/Other/optimization-remarks-auto.ll b/llvm/test/Other/optimization-remarks-auto.ll
@@ -0,0 +1,85 @@
+;; This test verifies 'auto' hotness threshold when profile summary is available.
+;;
+;; new PM
+; RUN: rm -f %t.yaml %t.hot.yaml
+; RUN: opt < %s --disable-output --enable-new-pm \
+; RUN: --passes='inline' \
+; RUN: --pass-remarks-output=%t.yaml --pass-remarks-filter='inline' \
+; RUN: --pass-remarks-with-hotness
+; RUN: FileCheck %s -check-prefix=YAML-PASS < %t.yaml
+; RUN: FileCheck %s -check-prefix=YAML-MISS < %t.yaml
+
+;; test 'auto' threshold
+; RUN: opt < %s --disable-output --enable-new-pm \
+; RUN: --passes='module(print-profile-summary,cgscc(inline))' \
+; RUN: --pass-remarks-output=%t.hot.yaml --pass-remarks-filter='inline' \
+; RUN: --pass-remarks-with-hotness --pass-remarks-hotness-threshold=auto 2>&1 | FileCheck %s
+; RUN: FileCheck %s -check-prefix=YAML-PASS < %t.hot.yaml
+; RUN: not FileCheck %s -check-prefix=YAML-MISS < %t.hot.yaml
+
+; RUN: opt < %s --disable-output --enable-new-pm \
+; RUN: --passes='module(print-profile-summary,cgscc(inline))' \
+; RUN: --pass-remarks=inline --pass-remarks-missed=inline --pass-remarks-analysis=inline \
+; RUN: --pass-remarks-with-hotness --pass-remarks-hotness-threshold=auto 2>&1 | FileCheck %s -check-prefix=CHECK-RPASS
+
+; YAML-PASS:      --- !Passed
+; YAML-PASS-NEXT: Pass:            inline
+; YAML-PASS-NEXT: Name:            Inlined
+; YAML-PASS-NEXT: Function:        caller1
+; YAML-PASS-NEXT: Hotness:         400
+
+; YAML-MISS:      --- !Missed
+; YAML-MISS-NEXT: Pass:            inline
+; YAML-MISS-NEXT: Name:            NeverInline
+; YAML-MISS-NEXT: Function:        caller2
+; YAML-MISS-NEXT: Hotness:         1
+
+; CHECK-RPASS: callee1 inlined into caller1 with (cost=-30, threshold=4500) (hotness: 400)
+; CHECK-RPASS-NOT: callee2 not inlined into caller2 because it should never be inlined (cost=never): noinline function attribute (hotness: 1)
+
+define void @callee1() !prof !20 {
+; CHECK: callee1 :hot
+entry:
+  ret void
+}
+
+; Function Attrs: noinline
+define void @callee2() noinline !prof !21 {
+; CHECK: callee2 :cold
+entry:
+  ret void
+}
+
+define void @caller1() !prof !20 {
+; CHECK: caller1 :hot
+entry:
+  call void @callee1()
+  ret void
+}
+
+define void @caller2() !prof !21 {
+; CHECK: caller2 :cold
+entry:
+  call void @callee2()
+  ret void
+}
+
+!llvm.module.flags = !{!1}
+!20 = !{!"function_entry_count", i64 400}
+!21 = !{!"function_entry_count", i64 1}
+
+!1 = !{i32 1, !"ProfileSummary", !2}
+!2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
+!3 = !{!"ProfileFormat", !"InstrProf"}
+!4 = !{!"TotalCount", i64 10000}
+!5 = !{!"MaxCount", i64 10}
+!6 = !{!"MaxInternalCount", i64 1}
+!7 = !{!"MaxFunctionCount", i64 1000}
+!8 = !{!"NumCounts", i64 3}
+!9 = !{!"NumFunctions", i64 3}
+!10 = !{!"DetailedSummary", !11}
+!11 = !{!12, !13, !14}
+!12 = !{i32 10000, i64 100, i32 1}
+!13 = !{i32 999000, i64 100, i32 1}
+!14 = !{i32 999999, i64 1, i32 2}
+
diff --git a/llvm/test/Transforms/SampleProfile/Inputs/remarks-hotness.prof b/llvm/test/Transforms/SampleProfile/Inputs/remarks-hotness.prof
diff --git a/llvm/test/Transforms/SampleProfile/remarks-hotness.ll b/llvm/test/Transforms/SampleProfile/remarks-hotness.ll
