This is an archive of the discontinued LLVM Phabricator instance.

Differential D80646

[PGO] Guard the memcmp/bcmp size value profiling instrumentation behind flag.
ClosedPublic

Authored by hjyamauchi on May 27 2020, 10:18 AM.

Details

Reviewers
davidxl
Commits
rGf0c2cfe4d044: [PGO] Guard the memcmp/bcmp size value profiling instrumentation behind flag.
Summary

Follow up D79751 and put the instrumentation / value collection side (in
addition to the optimization side) behind the flag as well.

Diff Detail

Event Timeline

hjyamauchi created this revision.May 27 2020, 10:18 AM
Herald added a project: Restricted Project. · View Herald TranscriptMay 27 2020, 10:18 AM
Herald added a subscriber: hiraditya. · View Herald Transcript
hjyamauchi mentioned this in D80578: [PGO] Enable memcmp/bcmp size value profiling..May 27 2020, 10:22 AM
davidxl accepted this revision.May 27 2020, 10:58 AM

lgtm

This revision is now accepted and ready to land.May 27 2020, 10:58 AM
Closed by commit rGf0c2cfe4d044: [PGO] Guard the memcmp/bcmp size value profiling instrumentation behind flag. (authored by hjyamauchi). · Explain WhyMay 28 2020, 10:25 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
llvm/
lib/
Transforms/
Instrumentation/
2 lines
4 lines
test/
Transforms/
PGOProfile/
8 lines

Diff 266591

llvm/lib/Transforms/Instrumentation/PGOMemOPSizeOpt.cpp

Show First 20 Lines • Show All 89 Lines • ▼ Show 20 Lines MemOPScaleCount("pgo-memop-scale-count", cl::init(true), cl::Hidden,
" block count value")); " block count value"));
// This option sets the rangge of precise profile memop sizes. // This option sets the rangge of precise profile memop sizes.
extern cl::opt<std::string> MemOPSizeRange; extern cl::opt<std::string> MemOPSizeRange;
// This option sets the value that groups large memop sizes // This option sets the value that groups large memop sizes
extern cl::opt<unsigned> MemOPSizeLarge; extern cl::opt<unsigned> MemOPSizeLarge;
static cl::opt<bool> cl::opt<bool>
MemOPOptMemcmpBcmp("pgo-memop-optimize-memcmp-bcmp", cl::init(false), MemOPOptMemcmpBcmp("pgo-memop-optimize-memcmp-bcmp", cl::init(false),
cl::Hidden, cl::Hidden,
cl::desc("Size-specialize memcmp and bcmp calls")); cl::desc("Size-specialize memcmp and bcmp calls"));
namespace { namespace {
class PGOMemOPSizeOptLegacyPass : public FunctionPass { class PGOMemOPSizeOptLegacyPass : public FunctionPass {
public: public:
static char ID; static char ID;
▲ Show 20 Lines • Show All 441 Lines • Show Last 20 Lines

llvm/lib/Transforms/Instrumentation/ValueProfilePlugins.inc

Show All 14 Lines
#include "ValueProfileCollector.h" #include "ValueProfileCollector.h"
#include "llvm/Analysis/IndirectCallVisitor.h" #include "llvm/Analysis/IndirectCallVisitor.h"
#include "llvm/IR/InstVisitor.h" #include "llvm/IR/InstVisitor.h"
using namespace llvm; using namespace llvm;
using CandidateInfo = ValueProfileCollector::CandidateInfo; using CandidateInfo = ValueProfileCollector::CandidateInfo;
extern cl::opt<bool> MemOPOptMemcmpBcmp;
///--------------------------- MemIntrinsicPlugin ------------------------------ ///--------------------------- MemIntrinsicPlugin ------------------------------
class MemIntrinsicPlugin : public InstVisitor<MemIntrinsicPlugin> { class MemIntrinsicPlugin : public InstVisitor<MemIntrinsicPlugin> {
Function &F; Function &F;
TargetLibraryInfo &TLI; TargetLibraryInfo &TLI;
std::vector<CandidateInfo> *Candidates; std::vector<CandidateInfo> *Candidates;
public: public:
static constexpr InstrProfValueKind Kind = IPVK_MemOPSize; static constexpr InstrProfValueKind Kind = IPVK_MemOPSize;
Show All 12 Lines void visitMemIntrinsic(MemIntrinsic &MI) {
if (dyn_cast<ConstantInt>(Length)) if (dyn_cast<ConstantInt>(Length))
return; return;
Instruction *InsertPt = &MI; Instruction *InsertPt = &MI;
Instruction *AnnotatedInst = &MI; Instruction *AnnotatedInst = &MI;
Candidates->emplace_back(CandidateInfo{Length, InsertPt, AnnotatedInst}); Candidates->emplace_back(CandidateInfo{Length, InsertPt, AnnotatedInst});
} }
void visitCallInst(CallInst &CI) { void visitCallInst(CallInst &CI) {
if (!MemOPOptMemcmpBcmp)
return;
auto *F = CI.getCalledFunction(); auto *F = CI.getCalledFunction();
if (!F) if (!F)
return; return;
LibFunc Func; LibFunc Func;
if (TLI.getLibFunc(CI, Func) && if (TLI.getLibFunc(CI, Func) &&
(Func == LibFunc_memcmp || Func == LibFunc_bcmp)) { (Func == LibFunc_memcmp || Func == LibFunc_bcmp)) {
Value *Length = CI.getArgOperand(2); Value *Length = CI.getArgOperand(2);
// Not instrument constant length calls. // Not instrument constant length calls.
Show All 35 Lines

llvm/test/Transforms/PGOProfile/memop_size_annotation.ll

; RUN: llvm-profdata merge %S/Inputs/memop_size_annotation.proftext -o %t.profdata ; RUN: llvm-profdata merge %S/Inputs/memop_size_annotation.proftext -o %t.profdata
; RUN: opt < %s -pgo-instr-use -memop-max-annotations=9 -pgo-test-profile-file=%t.profdata -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION9 ; RUN: opt < %s -pgo-instr-use -memop-max-annotations=9 -pgo-test-profile-file=%t.profdata -pgo-memop-optimize-memcmp-bcmp -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION9
; RUN: opt < %s -passes=pgo-instr-use -memop-max-annotations=9 -pgo-test-profile-file=%t.profdata -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION9 ; RUN: opt < %s -passes=pgo-instr-use -memop-max-annotations=9 -pgo-test-profile-file=%t.profdata -pgo-memop-optimize-memcmp-bcmp -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION9
; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%t.profdata -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION4 ; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%t.profdata -pgo-memop-optimize-memcmp-bcmp -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION4
; RUN: opt < %s -passes=pgo-instr-use -pgo-test-profile-file=%t.profdata -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION4 ; RUN: opt < %s -passes=pgo-instr-use -pgo-test-profile-file=%t.profdata -pgo-memop-optimize-memcmp-bcmp -S | FileCheck %s --check-prefixes=MEMOP_ANNOTATION,MEMOP_ANNOTATION4
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu" target triple = "x86_64-unknown-linux-gnu"
define void @foo(i8* %dst, i8* %src, i32* %a, i32 %n) { define void @foo(i8* %dst, i8* %src, i32* %a, i32 %n) {
entry: entry:
br label %for.cond br label %for.cond
▲ Show 20 Lines • Show All 55 Lines • Show Last 20 Lines