[profcheck] Fix profile metadata in IntegerDivision/ExpandIRinsts #173114
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Jin Huang (jinhuang1102) ChangesThis PR enhance the profile metadata in the ExpandIRinsts pass to ensure accurate profiling information is included for better performance analysis. The following chagnes were make:
Patch is 37.41 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/173114.diff 6 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/IntegerDivision.cpp b/llvm/lib/Transforms/Utils/IntegerDivision.cpp
index e95a7a9ae525a..115c254df251e 100644
--- a/llvm/lib/Transforms/Utils/IntegerDivision.cpp
+++ b/llvm/lib/Transforms/Utils/IntegerDivision.cpp
@@ -16,13 +16,22 @@
#include "llvm/Transforms/Utils/IntegerDivision.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/MDBuilder.h"
+#include "llvm/IR/ProfDataUtils.h"
+#include "llvm/Support/Casting.h"
+#include "llvm/Support/CommandLine.h"
using namespace llvm;
#define DEBUG_TYPE "integer-division"
+namespace llvm {
+extern cl::opt<bool> ProfcheckDisableMetadataFixes;
+}
+
/// Generate code to compute the remainder of two signed integers. Returns the
/// remainder, which will have the sign of the dividend. Builder's insert point
/// should be pointing where the caller wants code generated, e.g. at the srem
@@ -235,11 +244,53 @@ static Value *generateUnsignedDivisionCode(Value *Dividend, Value *Divisor,
Value *Tmp1 = Builder.CreateCall(CTLZ, {Dividend, True});
Value *SR = Builder.CreateSub(Tmp0, Tmp1);
Value *Ret0_4 = Builder.CreateICmpUGT(SR, MSB);
+
+ // Add 'unlikely' branch weights to the select instruction ('Ret0') generated
+ // by 'CreateLogicalOr'. We assume the 'true' path, where either the divisor
+ // or dividend is zero ('Ret0_3'), is less likely than the 'false' path,
+ // which corresponds to the |divisor| > |dividend| ('Ret0_4').
Value *Ret0 = Builder.CreateLogicalOr(Ret0_3, Ret0_4);
+ if (!ProfcheckDisableMetadataFixes) {
+ if (Instruction *Ret0SelectI = dyn_cast<Instruction>(Ret0)) {
+ Ret0SelectI->setMetadata(
+ LLVMContext::MD_prof,
+ MDBuilder(Ret0SelectI->getContext()).createUnlikelyBranchWeights());
+ }
+ }
Value *RetDividend = Builder.CreateICmpEQ(SR, MSB);
- Value *RetVal = Builder.CreateSelect(Ret0, Zero, Dividend);
- Value *EarlyRet = Builder.CreateLogicalOr(Ret0, RetDividend);
- Builder.CreateCondBr(EarlyRet, End, BB1);
+
+ // This select instruction (RetVal) immediately follows the Ret0 check. Since
+ // Ret0_4 determines if the divisor is greater then the dividend so the
+ // unknown (50/50) branch weights are fairly accurate for RetVal.
+ Value *RetVal;
+ if (!ProfcheckDisableMetadataFixes) {
+ RetVal = Builder.CreateSelectWithUnknownProfile(Ret0, Zero, Dividend,
+ DEBUG_TYPE);
+ } else {
+ RetVal = Builder.CreateSelect(Ret0, Zero, Dividend);
+ }
+ // This select instruction (EarlyRet) is used to check another edge case, and
+ // it share the same branch weights as RetVal so the unknown (50/50)branch
+ // weights are also accurate for EarlyRet.
+ Value *EarlyRet;
+ if (!ProfcheckDisableMetadataFixes) {
+ EarlyRet = Builder.CreateSelectWithUnknownProfile(
+ Ret0, ConstantInt::getAllOnesValue(Ret0->getType()), RetDividend,
+ DEBUG_TYPE);
+ } else {
+ EarlyRet = Builder.CreateLogicalOr(Ret0, RetDividend);
+ }
+
+ // The condition of this branch is based on `EarlyRet`. `EarlyRet` is true
+ // only for special cases like dividend or divisor being zero, or the divisor
+ // being greater than the dividend. Thus, the branch to `End` is unlikely,
+ // and we expect to more frequently enter `BB1`.
+ Instruction *ConBrSpecialCases = Builder.CreateCondBr(EarlyRet, End, BB1);
+ if (!ProfcheckDisableMetadataFixes) {
+ ConBrSpecialCases->setMetadata(LLVMContext::MD_prof,
+ MDBuilder(ConBrSpecialCases->getContext())
+ .createUnlikelyBranchWeights());
+ }
// ; bb1: ; preds = %special-cases
// ; %sr_1 = add i32 %sr, 1
@@ -252,7 +303,15 @@ static Value *generateUnsignedDivisionCode(Value *Dividend, Value *Divisor,
Value *Tmp2 = Builder.CreateSub(MSB, SR);
Value *Q = Builder.CreateShl(Dividend, Tmp2);
Value *SkipLoop = Builder.CreateICmpEQ(SR_1, Zero);
- Builder.CreateCondBr(SkipLoop, LoopExit, Preheader);
+ // This branch is highly unlikely. SR_1 (SR + 1) is expected to be in [1, 129]
+ // because the case where |divisor| > |dividend| (which would make SR
+ // negative) has alredy been intercepted in the previous SpecialCases BB.
+ Instruction *ConBrBB1 = Builder.CreateCondBr(SkipLoop, LoopExit, Preheader);
+ if (!ProfcheckDisableMetadataFixes) {
+ ConBrBB1->setMetadata(LLVMContext::MD_prof,
+ MDBuilder(ConBrBB1->getContext())
+ .createUnlikelyBranchWeights());
+ }
// ; preheader: ; preds = %bb1
// ; %tmp3 = lshr i32 %dividend, %sr_1
@@ -298,7 +357,13 @@ static Value *generateUnsignedDivisionCode(Value *Dividend, Value *Divisor,
Value *R = Builder.CreateSub(Tmp7, Tmp11);
Value *SR_2 = Builder.CreateAdd(SR_3, NegOne);
Value *Tmp12 = Builder.CreateICmpEQ(SR_2, Zero);
- Builder.CreateCondBr(Tmp12, LoopExit, DoWhile);
+ // The loop implements the core bit-by-bit binary long division algorithm.
+ // Since each iteration generates one bit of the quotient until covering all
+ // significant bits so we would label it as unknown (50/50) branch weights.
+ Instruction *ConBrDoWhile = Builder.CreateCondBr(Tmp12, LoopExit, DoWhile);
+ if (!ProfcheckDisableMetadataFixes) {
+ setExplicitlyUnknownBranchWeightsIfProfiled(*ConBrDoWhile, DEBUG_TYPE, F);
+ }
// ; loop-exit: ; preds = %do-while, %bb1
// ; %carry_2 = phi i32 [ 0, %bb1 ], [ %carry, %do-while ]
diff --git a/llvm/test/Transforms/ExpandIRInsts/X86/sdiv129.ll b/llvm/test/Transforms/ExpandIRInsts/X86/sdiv129.ll
index fc823cd543144..01b50da066d88 100644
--- a/llvm/test/Transforms/ExpandIRInsts/X86/sdiv129.ll
+++ b/llvm/test/Transforms/ExpandIRInsts/X86/sdiv129.ll
@@ -2,8 +2,9 @@
; RUN: opt -S -mtriple=x86_64-- -expand-ir-insts -expand-div-rem-bits 128 < %s | FileCheck %s
; RUN: opt -S -mtriple=x86_64-- -passes='require<libcall-lowering-info>,expand-ir-insts' -expand-div-rem-bits 128 < %s | FileCheck %s
-define void @sdiv129(ptr %ptr, ptr %out) nounwind {
+define void @sdiv129(ptr %ptr, ptr %out) nounwind !prof !0 {
; CHECK-LABEL: @sdiv129(
+; CHECK: !prof [[PROF_0:![0-9]+]] {
; CHECK-NEXT: _udiv-special-cases:
; CHECK-NEXT: [[A:%.*]] = load i129, ptr [[PTR:%.*]], align 16
; CHECK-NEXT: [[TMP0:%.*]] = freeze i129 [[A]]
@@ -24,11 +25,11 @@ define void @sdiv129(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP15:%.*]] = call i129 @llvm.ctlz.i129(i129 [[TMP10]], i1 true)
; CHECK-NEXT: [[TMP16:%.*]] = sub i129 [[TMP14]], [[TMP15]]
; CHECK-NEXT: [[TMP17:%.*]] = icmp ugt i129 [[TMP16]], 128
-; CHECK-NEXT: [[TMP18:%.*]] = select i1 [[TMP13]], i1 true, i1 [[TMP17]]
+; CHECK-NEXT: [[TMP18:%.*]] = select i1 [[TMP13]], i1 true, i1 [[TMP17]], !prof [[PROF_1:![0-9]+]]
; CHECK-NEXT: [[TMP19:%.*]] = icmp eq i129 [[TMP16]], 128
-; CHECK-NEXT: [[TMP20:%.*]] = select i1 [[TMP18]], i129 0, i129 [[TMP10]]
-; CHECK-NEXT: [[TMP21:%.*]] = select i1 [[TMP18]], i1 true, i1 [[TMP19]]
-; CHECK-NEXT: br i1 [[TMP21]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]]
+; CHECK-NEXT: [[TMP20:%.*]] = select i1 [[TMP18]], i129 0, i129 [[TMP10]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: [[TMP21:%.*]] = select i1 [[TMP18]], i1 true, i1 [[TMP19]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP21]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-loop-exit:
; CHECK-NEXT: [[TMP22:%.*]] = phi i129 [ 0, [[UDIV_BB1]] ], [ [[TMP37:%.*]], [[UDIV_DO_WHILE:%.*]] ]
; CHECK-NEXT: [[TMP23:%.*]] = phi i129 [ [[TMP46:%.*]], [[UDIV_BB1]] ], [ [[TMP34:%.*]], [[UDIV_DO_WHILE]] ]
@@ -52,7 +53,7 @@ define void @sdiv129(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP39]] = sub i129 [[TMP32]], [[TMP38]]
; CHECK-NEXT: [[TMP40]] = add i129 [[TMP27]], -1
; CHECK-NEXT: [[TMP41:%.*]] = icmp eq i129 [[TMP40]], 0
-; CHECK-NEXT: br i1 [[TMP41]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]]
+; CHECK-NEXT: br i1 [[TMP41]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]], !prof [[PROF_2:![0-9]+]]
; CHECK: udiv-preheader:
; CHECK-NEXT: [[TMP42]] = lshr i129 [[TMP10]], [[TMP44]]
; CHECK-NEXT: [[TMP43]] = add i129 [[TMP9]], -1
@@ -62,7 +63,7 @@ define void @sdiv129(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP45:%.*]] = sub i129 128, [[TMP16]]
; CHECK-NEXT: [[TMP46]] = shl i129 [[TMP10]], [[TMP45]]
; CHECK-NEXT: [[TMP47:%.*]] = icmp eq i129 [[TMP44]], 0
-; CHECK-NEXT: br i1 [[TMP47]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]]
+; CHECK-NEXT: br i1 [[TMP47]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-end:
; CHECK-NEXT: [[TMP48:%.*]] = phi i129 [ [[TMP25]], [[UDIV_LOOP_EXIT]] ], [ [[TMP20]], [[_UDIV_SPECIAL_CASES:%.*]] ]
; CHECK-NEXT: [[TMP49:%.*]] = xor i129 [[TMP48]], [[TMP8]]
@@ -75,3 +76,8 @@ define void @sdiv129(ptr %ptr, ptr %out) nounwind {
store i129 %res, ptr %out
ret void
}
+
+!0 = !{!"function_entry_count", i64 1000}
+; CHECK: [[PROF_0]] = !{!"function_entry_count", i64 1000}
+; CHECK: [[PROF_1]] = !{!"branch_weights", i32 1, i32 1048575}
+; CHECK: [[PROF_2]] = !{!"unknown", !"integer-division"}
\ No newline at end of file
diff --git a/llvm/test/Transforms/ExpandIRInsts/X86/srem129.ll b/llvm/test/Transforms/ExpandIRInsts/X86/srem129.ll
index 667152228d258..0485c1ac5607e 100644
--- a/llvm/test/Transforms/ExpandIRInsts/X86/srem129.ll
+++ b/llvm/test/Transforms/ExpandIRInsts/X86/srem129.ll
@@ -2,8 +2,9 @@
; RUN: opt -S -mtriple=x86_64-- -expand-ir-insts -expand-div-rem-bits 128 < %s | FileCheck %s
; RUN: opt -S -mtriple=x86_64-- -passes='require<libcall-lowering-info>,expand-ir-insts' -expand-div-rem-bits 128 < %s | FileCheck %s
-define void @test(ptr %ptr, ptr %out) nounwind {
+define void @test(ptr %ptr, ptr %out) nounwind !prof !0 {
; CHECK-LABEL: @test(
+; CHECK: !prof [[PROF_0:![0-9]+]] {
; CHECK-NEXT: _udiv-special-cases:
; CHECK-NEXT: [[A:%.*]] = load i129, ptr [[PTR:%.*]], align 16
; CHECK-NEXT: [[TMP0:%.*]] = freeze i129 [[A]]
@@ -25,11 +26,11 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP16:%.*]] = call i129 @llvm.ctlz.i129(i129 [[TMP11]], i1 true)
; CHECK-NEXT: [[TMP17:%.*]] = sub i129 [[TMP15]], [[TMP16]]
; CHECK-NEXT: [[TMP18:%.*]] = icmp ugt i129 [[TMP17]], 128
-; CHECK-NEXT: [[TMP19:%.*]] = select i1 [[TMP14]], i1 true, i1 [[TMP18]]
+; CHECK-NEXT: [[TMP19:%.*]] = select i1 [[TMP14]], i1 true, i1 [[TMP18]], !prof [[PROF_1:![0-9]+]]
; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i129 [[TMP17]], 128
-; CHECK-NEXT: [[TMP21:%.*]] = select i1 [[TMP19]], i129 0, i129 [[TMP11]]
-; CHECK-NEXT: [[TMP22:%.*]] = select i1 [[TMP19]], i1 true, i1 [[TMP20]]
-; CHECK-NEXT: br i1 [[TMP22]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]]
+; CHECK-NEXT: [[TMP21:%.*]] = select i1 [[TMP19]], i129 0, i129 [[TMP11]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: [[TMP22:%.*]] = select i1 [[TMP19]], i1 true, i1 [[TMP20]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP22]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-loop-exit:
; CHECK-NEXT: [[TMP23:%.*]] = phi i129 [ 0, [[UDIV_BB1]] ], [ [[TMP38:%.*]], [[UDIV_DO_WHILE:%.*]] ]
; CHECK-NEXT: [[TMP24:%.*]] = phi i129 [ [[TMP47:%.*]], [[UDIV_BB1]] ], [ [[TMP35:%.*]], [[UDIV_DO_WHILE]] ]
@@ -53,7 +54,7 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP40]] = sub i129 [[TMP33]], [[TMP39]]
; CHECK-NEXT: [[TMP41]] = add i129 [[TMP28]], -1
; CHECK-NEXT: [[TMP42:%.*]] = icmp eq i129 [[TMP41]], 0
-; CHECK-NEXT: br i1 [[TMP42]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]]
+; CHECK-NEXT: br i1 [[TMP42]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]], !prof [[PROF_2:![0-9]+]]
; CHECK: udiv-preheader:
; CHECK-NEXT: [[TMP43]] = lshr i129 [[TMP11]], [[TMP45]]
; CHECK-NEXT: [[TMP44]] = add i129 [[TMP10]], -1
@@ -63,7 +64,7 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP46:%.*]] = sub i129 128, [[TMP17]]
; CHECK-NEXT: [[TMP47]] = shl i129 [[TMP11]], [[TMP46]]
; CHECK-NEXT: [[TMP48:%.*]] = icmp eq i129 [[TMP45]], 0
-; CHECK-NEXT: br i1 [[TMP48]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]]
+; CHECK-NEXT: br i1 [[TMP48]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-end:
; CHECK-NEXT: [[TMP49:%.*]] = phi i129 [ [[TMP26]], [[UDIV_LOOP_EXIT]] ], [ [[TMP21]], [[_UDIV_SPECIAL_CASES:%.*]] ]
; CHECK-NEXT: [[TMP50:%.*]] = mul i129 [[TMP9]], [[TMP49]]
@@ -78,3 +79,8 @@ define void @test(ptr %ptr, ptr %out) nounwind {
store i129 %res, ptr %out
ret void
}
+
+!0 = !{!"function_entry_count", i64 1000}
+; CHECK: [[PROF_0]] = !{!"function_entry_count", i64 1000}
+; CHECK: [[PROF_1]] = !{!"branch_weights", i32 1, i32 1048575}
+; CHECK: [[PROF_2]] = !{!"unknown", !"integer-division"}
diff --git a/llvm/test/Transforms/ExpandIRInsts/X86/udiv129.ll b/llvm/test/Transforms/ExpandIRInsts/X86/udiv129.ll
index b2b83815f79b0..c3a21dbd145d8 100644
--- a/llvm/test/Transforms/ExpandIRInsts/X86/udiv129.ll
+++ b/llvm/test/Transforms/ExpandIRInsts/X86/udiv129.ll
@@ -2,8 +2,9 @@
; RUN: opt -S -mtriple=x86_64-- -expand-ir-insts -expand-div-rem-bits 128 < %s | FileCheck %s
; RUN: opt -S -mtriple=x86_64-- -passes='require<libcall-lowering-info>,expand-ir-insts' -expand-div-rem-bits 128 < %s | FileCheck %s
-define void @test(ptr %ptr, ptr %out) nounwind {
+define void @test(ptr %ptr, ptr %out) nounwind !prof !0 {
; CHECK-LABEL: @test(
+; CHECK: !prof [[PROF_0:![0-9]+]] {
; CHECK-NEXT: _udiv-special-cases:
; CHECK-NEXT: [[A:%.*]] = load i129, ptr [[PTR:%.*]], align 16
; CHECK-NEXT: [[TMP0:%.*]] = freeze i129 3
@@ -15,11 +16,11 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP6:%.*]] = call i129 @llvm.ctlz.i129(i129 [[TMP1]], i1 true)
; CHECK-NEXT: [[TMP7:%.*]] = sub i129 [[TMP5]], [[TMP6]]
; CHECK-NEXT: [[TMP8:%.*]] = icmp ugt i129 [[TMP7]], 128
-; CHECK-NEXT: [[TMP9:%.*]] = select i1 [[TMP4]], i1 true, i1 [[TMP8]]
+; CHECK-NEXT: [[TMP9:%.*]] = select i1 [[TMP4]], i1 true, i1 [[TMP8]], !prof [[PROF_1:![0-9]+]]
; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i129 [[TMP7]], 128
-; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP9]], i129 0, i129 [[TMP1]]
-; CHECK-NEXT: [[TMP12:%.*]] = select i1 [[TMP9]], i1 true, i1 [[TMP10]]
-; CHECK-NEXT: br i1 [[TMP12]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]]
+; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP9]], i129 0, i129 [[TMP1]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: [[TMP12:%.*]] = select i1 [[TMP9]], i1 true, i1 [[TMP10]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP12]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-loop-exit:
; CHECK-NEXT: [[TMP13:%.*]] = phi i129 [ 0, [[UDIV_BB1]] ], [ [[TMP28:%.*]], [[UDIV_DO_WHILE:%.*]] ]
; CHECK-NEXT: [[TMP14:%.*]] = phi i129 [ [[TMP37:%.*]], [[UDIV_BB1]] ], [ [[TMP25:%.*]], [[UDIV_DO_WHILE]] ]
@@ -43,7 +44,7 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP30]] = sub i129 [[TMP23]], [[TMP29]]
; CHECK-NEXT: [[TMP31]] = add i129 [[TMP18]], -1
; CHECK-NEXT: [[TMP32:%.*]] = icmp eq i129 [[TMP31]], 0
-; CHECK-NEXT: br i1 [[TMP32]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]]
+; CHECK-NEXT: br i1 [[TMP32]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]], !prof [[PROF_2:![0-9]+]]
; CHECK: udiv-preheader:
; CHECK-NEXT: [[TMP33]] = lshr i129 [[TMP1]], [[TMP35]]
; CHECK-NEXT: [[TMP34]] = add i129 [[TMP0]], -1
@@ -53,7 +54,7 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP36:%.*]] = sub i129 128, [[TMP7]]
; CHECK-NEXT: [[TMP37]] = shl i129 [[TMP1]], [[TMP36]]
; CHECK-NEXT: [[TMP38:%.*]] = icmp eq i129 [[TMP35]], 0
-; CHECK-NEXT: br i1 [[TMP38]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]]
+; CHECK-NEXT: br i1 [[TMP38]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-end:
; CHECK-NEXT: [[TMP39:%.*]] = phi i129 [ [[TMP16]], [[UDIV_LOOP_EXIT]] ], [ [[TMP11]], [[_UDIV_SPECIAL_CASES:%.*]] ]
; CHECK-NEXT: store i129 [[TMP39]], ptr [[OUT:%.*]], align 16
@@ -64,3 +65,8 @@ define void @test(ptr %ptr, ptr %out) nounwind {
store i129 %res, ptr %out
ret void
}
+
+!0 = !{!"function_entry_count", i64 1000}
+; CHECK: [[PROF_0]] = !{!"function_entry_count", i64 1000}
+; CHECK: [[PROF_1]] = !{!"branch_weights", i32 1, i32 1048575}
+; CHECK: [[PROF_2]] = !{!"unknown", !"integer-division"}
diff --git a/llvm/test/Transforms/ExpandIRInsts/X86/urem129.ll b/llvm/test/Transforms/ExpandIRInsts/X86/urem129.ll
index 46e72001b2c2d..0809b70b55a42 100644
--- a/llvm/test/Transforms/ExpandIRInsts/X86/urem129.ll
+++ b/llvm/test/Transforms/ExpandIRInsts/X86/urem129.ll
@@ -2,8 +2,9 @@
; RUN: opt -S -mtriple=x86_64-- -expand-ir-insts -expand-div-rem-bits 128 < %s | FileCheck %s
; RUN: opt -S -mtriple=x86_64-- -passes='require<libcall-lowering-info>,expand-ir-insts' -expand-div-rem-bits 128 < %s | FileCheck %s
-define void @test(ptr %ptr, ptr %out) nounwind {
+define void @test(ptr %ptr, ptr %out) nounwind !prof !0 {
; CHECK-LABEL: @test(
+; CHECK: !prof [[PROF_0:![0-9]+]] {
; CHECK-NEXT: _udiv-special-cases:
; CHECK-NEXT: [[A:%.*]] = load i129, ptr [[PTR:%.*]], align 16
; CHECK-NEXT: [[TMP0:%.*]] = freeze i129 [[A]]
@@ -17,11 +18,11 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP8:%.*]] = call i129 @llvm.ctlz.i129(i129 [[TMP3]], i1 true)
; CHECK-NEXT: [[TMP9:%.*]] = sub i129 [[TMP7]], [[TMP8]]
; CHECK-NEXT: [[TMP10:%.*]] = icmp ugt i129 [[TMP9]], 128
-; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP6]], i1 true, i1 [[TMP10]]
+; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP6]], i1 true, i1 [[TMP10]], !prof [[PROF_1:![0-9]+]]
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i129 [[TMP9]], 128
-; CHECK-NEXT: [[TMP13:%.*]] = select i1 [[TMP11]], i129 0, i129 [[TMP3]]
-; CHECK-NEXT: [[TMP14:%.*]] = select i1 [[TMP11]], i1 true, i1 [[TMP12]]
-; CHECK-NEXT: br i1 [[TMP14]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]]
+; CHECK-NEXT: [[TMP13:%.*]] = select i1 [[TMP11]], i129 0, i129 [[TMP3]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: [[TMP14:%.*]] = select i1 [[TMP11]], i1 true, i1 [[TMP12]], !prof [[PROF_2:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP14]], label [[UDIV_END:%.*]], label [[UDIV_BB1:%.*]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-loop-exit:
; CHECK-NEXT: [[TMP15:%.*]] = phi i129 [ 0, [[UDIV_BB1]] ], [ [[TMP30:%.*]], [[UDIV_DO_WHILE:%.*]] ]
; CHECK-NEXT: [[TMP16:%.*]] = phi i129 [ [[TMP39:%.*]], [[UDIV_BB1]] ], [ [[TMP27:%.*]], [[UDIV_DO_WHILE]] ]
@@ -45,7 +46,7 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP32]] = sub i129 [[TMP25]], [[TMP31]]
; CHECK-NEXT: [[TMP33]] = add i129 [[TMP20]], -1
; CHECK-NEXT: [[TMP34:%.*]] = icmp eq i129 [[TMP33]], 0
-; CHECK-NEXT: br i1 [[TMP34]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]]
+; CHECK-NEXT: br i1 [[TMP34]], label [[UDIV_LOOP_EXIT:%.*]], label [[UDIV_DO_WHILE]], !prof [[PROF_2:![0-9]+]]
; CHECK: udiv-preheader:
; CHECK-NEXT: [[TMP35]] = lshr i129 [[TMP3]], [[TMP37]]
; CHECK-NEXT: [[TMP36]] = add i129 [[TMP2]], -1
@@ -55,7 +56,7 @@ define void @test(ptr %ptr, ptr %out) nounwind {
; CHECK-NEXT: [[TMP38:%.*]] = sub i129 128, [[TMP9]]
; CHECK-NEXT: [[TMP39]] = shl i129 [[TMP3]], [[TMP38]]
; CHECK-NEXT: [[TMP40:%.*]] = icmp eq i129 [[TMP37]], 0
-; CHECK-NEXT: br i1 [[TMP40]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]]
+; CHECK-NEXT: br i1 [[TMP40]], label [[UDIV_LOOP_EXIT]], label [[UDIV_PREHEADER]], !prof [[PROF_1:![0-9]+]]
; CHECK: udiv-end:
; CHECK-NEXT: [[TMP41:%.*]] = phi i129 [ [[TMP18]], [[UDIV_LOOP_EXIT]] ], [ [[TMP13]], [[_UDIV_SPECIAL_CASES:%.*]] ]
; CHECK-NEXT: [[TMP42:%.*]] = mul i129 [[TMP1]], [[TMP41]]
@@ -68,3 +69,8 @@ define void @test(ptr %ptr, ptr %out) nou...
[truncated]
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
jinhuang1102
requested review from
alanzhao1,
boomanaiden154,
mtrofin and
nikic
jinhuang1102
force-pushed
the
fixup/prof-missing-ExpandLargeDivRem
branch
from
115a17a to
99fcfbd
Compare
alanzhao1
approved these changes
Dec 19, 2025
🪟 Windows x64 Test Results
Failed Tests(click on a test name to see its output) LLVMLLVM.CodeGen/RISCV/idiv_large.llLLVM.CodeGen/X86/div-rem-pair-recomposition-signed.llLLVM.CodeGen/X86/div-rem-pair-recomposition-unsigned.llLLVM.CodeGen/X86/pr38539.llLLVM.Transforms/ExpandIRInsts/X86/srem129.llIf these failures are unrelated to your changes (for example tests are broken or flaky at HEAD), please open an issue at https://github.com/llvm/llvm-project/issues and add the |
🐧 Linux x64 Test Results
Failed Tests(click on a test name to see its output) LLVMLLVM.CodeGen/RISCV/idiv_large.llLLVM.CodeGen/X86/div-rem-pair-recomposition-signed.llLLVM.CodeGen/X86/div-rem-pair-recomposition-unsigned.llLLVM.CodeGen/X86/pr38539.llIf these failures are unrelated to your changes (for example tests are broken or flaky at HEAD), please open an issue at https://github.com/llvm/llvm-project/issues and add the |
jinhuang1102
force-pushed
the
fixup/prof-missing-ExpandLargeDivRem
branch
from
99fcfbd to
5555b70
Compare
Contributor
Author
|
Since the |
jinhuang1102
force-pushed
the
fixup/prof-missing-ExpandLargeDivRem
branch
from
5555b70 to
c8b3d08
Compare
jinhuang1102
force-pushed
the
fixup/prof-missing-ExpandLargeDivRem
branch
from
c8b3d08 to
ae8b12f
Compare
jinhuang1102
force-pushed
the
fixup/prof-missing-ExpandLargeDivRem
branch
from
ae8b12f to
bcc91d2
Compare
Contributor
boomanaiden154
left a comment
boomanaiden154
left a comment
There was a problem hiding this comment.
There are still premerge test failures that need to be fixed.
| Value *Tmp2 = Builder.CreateSub(MSB, SR); | ||
| Value *Q = Builder.CreateShl(Dividend, Tmp2); | ||
| // We assume that in the common case, the dividend's magnitude is larger than | ||
| // the divisor's magnitude such that the loop conuter (SR) is non-zero. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR enhance the profile metadata in the ExpandIRinsts pass to ensure accurate profiling information is included for better performance analysis.
The following chagnes were make:
createUnlikelyBranchWeightsto select instructionRet0andconditional branches
ConBrSpecialCasesandConBrBB1.Builder.CreateSelectwithBuilder.CreateSelectWithUnknownProfilefor
RetValandEarlyRetwhen ProfcheckDisableMetadataFixes in not set.setExplicitlyUnknownBranchWeightsIfProfiledfor theConBrDoWhileconditional branch.