-
Notifications
You must be signed in to change notification settings - Fork 12k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[IndVarSimplify] Fix poison-safety when reusing instructions #80458
Conversation
IndVars may replace an instruction with one of its operands, if they have the same SCEV expression. However, such a replacement may be more poisonous. First, check whether the operand being poison implies that the instruction is also poison, in which case the replacement is always safe. If this fails, check whether SCEV can determine that reusing the instruction is safe, using the same check as SCEVExpander. Fixes llvm#79861.
@llvm/pr-subscribers-llvm-transforms Author: Nikita Popov (nikic) ChangesIndVars may replace an instruction with one of its operands, if they have the same SCEV expression. However, such a replacement may be more poisonous. First, check whether the operand being poison implies that the instruction is also poison, in which case the replacement is always safe. If this fails, check whether SCEV can determine that reusing the instruction is safe, using the same check as SCEVExpander. Fixes #79861. Full diff: https://github.com/llvm/llvm-project/pull/80458.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp b/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
index 0ed3324a27b6c..1b142f14d8113 100644
--- a/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
@@ -16,6 +16,7 @@
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instructions.h"
@@ -713,8 +714,11 @@ bool SimplifyIndvar::replaceFloatIVWithIntegerIV(Instruction *UseInst) {
bool SimplifyIndvar::eliminateIdentitySCEV(Instruction *UseInst,
Instruction *IVOperand) {
if (!SE->isSCEVable(UseInst->getType()) ||
- (UseInst->getType() != IVOperand->getType()) ||
- (SE->getSCEV(UseInst) != SE->getSCEV(IVOperand)))
+ UseInst->getType() != IVOperand->getType())
+ return false;
+
+ const SCEV *UseSCEV = SE->getSCEV(UseInst);
+ if (UseSCEV != SE->getSCEV(IVOperand))
return false;
// getSCEV(X) == getSCEV(Y) does not guarantee that X and Y are related in the
@@ -742,6 +746,16 @@ bool SimplifyIndvar::eliminateIdentitySCEV(Instruction *UseInst,
if (!LI->replacementPreservesLCSSAForm(UseInst, IVOperand))
return false;
+ // Make sure the operand is not more poisonous than the instruction.
+ if (!impliesPoison(IVOperand, UseInst)) {
+ SmallVector<Instruction *> DropPoisonGeneratingInsts;
+ if (!SE->canReuseInstruction(UseSCEV, IVOperand, DropPoisonGeneratingInsts))
+ return false;
+
+ for (Instruction *I : DropPoisonGeneratingInsts)
+ I->dropPoisonGeneratingFlagsAndMetadata();
+ }
+
LLVM_DEBUG(dbgs() << "INDVARS: Eliminated identity: " << *UseInst << '\n');
SE->forgetValue(UseInst);
diff --git a/llvm/test/Transforms/IndVarSimplify/pr55925.ll b/llvm/test/Transforms/IndVarSimplify/pr55925.ll
index 420fc209949d4..312a8295ccdc9 100644
--- a/llvm/test/Transforms/IndVarSimplify/pr55925.ll
+++ b/llvm/test/Transforms/IndVarSimplify/pr55925.ll
@@ -18,9 +18,9 @@ define void @test(ptr %p) personality ptr undef {
; CHECK-NEXT: [[RES:%.*]] = invoke i32 @foo(i32 returned [[TMP0]])
; CHECK-NEXT: to label [[LOOP_LATCH]] unwind label [[EXIT:%.*]]
; CHECK: loop.latch:
-; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
; CHECK-NEXT: [[TMP2:%.*]] = call i32 @foo(i32 [[TMP1]])
+; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
; CHECK-NEXT: br label [[LOOP]]
; CHECK: exit:
; CHECK-NEXT: [[LP:%.*]] = landingpad { ptr, i32 }
@@ -64,8 +64,8 @@ define void @test_critedge(i1 %c, ptr %p) personality ptr undef {
; CHECK-NEXT: br label [[LOOP_LATCH]]
; CHECK: loop.latch:
; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ [[TMP1]], [[LOOP_INVOKE]] ], [ 0, [[LOOP_OTHER]] ]
-; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
; CHECK-NEXT: [[TMP2:%.*]] = call i32 @foo(i32 [[PHI]])
+; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
; CHECK-NEXT: br label [[LOOP]]
; CHECK: exit:
; CHECK-NEXT: [[LP:%.*]] = landingpad { ptr, i32 }
diff --git a/llvm/test/Transforms/IndVarSimplify/pr79861.ll b/llvm/test/Transforms/IndVarSimplify/pr79861.ll
index 7e267d04c94cc..6625094496139 100644
--- a/llvm/test/Transforms/IndVarSimplify/pr79861.ll
+++ b/llvm/test/Transforms/IndVarSimplify/pr79861.ll
@@ -12,7 +12,9 @@ define void @or_disjoint() {
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 2, [[ENTRY:%.*]] ], [ [[IV_DEC:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[OR:%.*]] = or disjoint i64 [[IV]], 1
-; CHECK-NEXT: call void @use(i64 [[OR]])
+; CHECK-NEXT: [[ADD:%.*]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT: [[SEL:%.*]] = select i1 false, i64 [[OR]], i64 [[ADD]]
+; CHECK-NEXT: call void @use(i64 [[SEL]])
; CHECK-NEXT: [[IV_DEC]] = add nsw i64 [[IV]], -1
; CHECK-NEXT: [[EXIT_COND:%.*]] = icmp eq i64 [[IV_DEC]], 0
; CHECK-NEXT: br i1 [[EXIT_COND]], label [[EXIT:%.*]], label [[LOOP]]
@@ -44,7 +46,7 @@ define void @add_nowrap_flags(i64 %n) {
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_INC:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[ADD1:%.*]] = add nuw nsw i64 [[IV]], 123
+; CHECK-NEXT: [[ADD1:%.*]] = add i64 [[IV]], 123
; CHECK-NEXT: call void @use(i64 [[ADD1]])
; CHECK-NEXT: [[IV_INC]] = add i64 [[IV]], 1
; CHECK-NEXT: [[EXIT_COND:%.*]] = icmp eq i64 [[IV_INC]], [[N]]
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
I'm mildly concerned about compile time impact here, but since this is a correctness fix, I'm fine landing it and revisiting if the compile time is problematic in practice.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, thanks!
) IndVars may replace an instruction with one of its operands, if they have the same SCEV expression. However, such a replacement may be more poisonous. First, check whether the operand being poison implies that the instruction is also poison, in which case the replacement is always safe. If this fails, check whether SCEV can determine that reusing the instruction is safe, using the same check as SCEVExpander. Fixes llvm#79861.
) IndVars may replace an instruction with one of its operands, if they have the same SCEV expression. However, such a replacement may be more poisonous. First, check whether the operand being poison implies that the instruction is also poison, in which case the replacement is always safe. If this fails, check whether SCEV can determine that reusing the instruction is safe, using the same check as SCEVExpander. Fixes llvm#79861. (cherry picked from commit 7d2b6f0)
) IndVars may replace an instruction with one of its operands, if they have the same SCEV expression. However, such a replacement may be more poisonous. First, check whether the operand being poison implies that the instruction is also poison, in which case the replacement is always safe. If this fails, check whether SCEV can determine that reusing the instruction is safe, using the same check as SCEVExpander. Fixes llvm#79861. (cherry picked from commit 7d2b6f0)
) IndVars may replace an instruction with one of its operands, if they have the same SCEV expression. However, such a replacement may be more poisonous. First, check whether the operand being poison implies that the instruction is also poison, in which case the replacement is always safe. If this fails, check whether SCEV can determine that reusing the instruction is safe, using the same check as SCEVExpander. Fixes llvm#79861. (cherry picked from commit 7d2b6f0)
IndVars may replace an instruction with one of its operands, if they have the same SCEV expression. However, such a replacement may be more poisonous.
First, check whether the operand being poison implies that the instruction is also poison, in which case the replacement is always safe. If this fails, check whether SCEV can determine that reusing the instruction is safe, using the same check as SCEVExpander.
Fixes #79861.