[SCEVExp] Add helper to clean up dead instructions after expansion. #157308
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
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-llvm-transforms Author: Florian Hahn (fhahn) ChangesAdd new helper to erase dead instructions inserted during SCEV expansion but not being used due to InstSimplifyFolder simplifications. Together with #157307 this also allows removing some specialized folds, e.g. https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp#L2205 Patch is 27.11 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/157308.diff 13 Files Affected:
diff --git a/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h b/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h
index 39fef921a9590..310118078695c 100644
--- a/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h
+++ b/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h
@@ -433,6 +433,10 @@ class SCEVExpander : public SCEVVisitor<SCEVExpander, Value *> {
LLVM_ABI BasicBlock::iterator
findInsertPointAfter(Instruction *I, Instruction *MustDominate) const;
+ /// Remove inserted instructions that are dead, e.g. due to InstSimplifyFolder
+ /// simplifications. \p Root is assumed to be used and won't be removed.
+ void eraseDeadInstructions(Value *Root);
+
private:
LLVMContext &getContext() const { return SE.getContext(); }
diff --git a/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp b/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
index 7c12dac73f3d1..2bb74f24b0af7 100644
--- a/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
+++ b/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
@@ -26,6 +26,7 @@
#include "llvm/IR/PatternMatch.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/raw_ostream.h"
+#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/LoopUtils.h"
#if LLVM_ENABLE_ABI_BREAKING_CHECKS
@@ -175,6 +176,16 @@ SCEVExpander::findInsertPointAfter(Instruction *I,
return IP;
}
+void SCEVExpander::eraseDeadInstructions(Value *Root) {
+ for (Instruction *I : getAllInsertedInstructions()) {
+ if (I == Root || !isInstructionTriviallyDead(I))
+ continue;
+ InsertedValues.erase(I);
+ InsertedPostIncValues.erase(I);
+ I->eraseFromParent();
+ }
+}
+
BasicBlock::iterator
SCEVExpander::GetOptimalInsertionPointForCastOf(Value *V) const {
// Cast the argument at the beginning of the entry block, after
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 21f1089104587..1f800f6056bb0 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1849,6 +1849,9 @@ class GeneratedRTChecks {
"claimed checks are required");
}
+ SCEVExp.eraseDeadInstructions(SCEVCheckCond);
+ MemCheckExp.eraseDeadInstructions(MemRuntimeCheckCond);
+
if (!MemCheckBlock && !SCEVCheckBlock)
return;
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll
index 0a4ccb0a3b961..884eeac09e1e3 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll
@@ -696,7 +696,6 @@ define void @test_conditional_interleave_group (ptr noalias %src.1, ptr noalias
; DEFAULT-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; DEFAULT-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
; DEFAULT-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; DEFAULT-NEXT: [[TMP1:%.*]] = sub i64 0, [[MUL_RESULT]]
; DEFAULT-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[DST]], i64 [[MUL_RESULT]]
; DEFAULT-NEXT: [[TMP3:%.*]] = icmp ult ptr [[TMP2]], [[DST]]
; DEFAULT-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
@@ -704,7 +703,6 @@ define void @test_conditional_interleave_group (ptr noalias %src.1, ptr noalias
; DEFAULT-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; DEFAULT-NEXT: [[MUL_RESULT2:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
; DEFAULT-NEXT: [[MUL_OVERFLOW3:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
-; DEFAULT-NEXT: [[TMP5:%.*]] = sub i64 0, [[MUL_RESULT2]]
; DEFAULT-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT2]]
; DEFAULT-NEXT: [[TMP7:%.*]] = icmp ult ptr [[TMP6]], [[SCEVGEP]]
; DEFAULT-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW3]]
@@ -712,7 +710,6 @@ define void @test_conditional_interleave_group (ptr noalias %src.1, ptr noalias
; DEFAULT-NEXT: [[MUL5:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; DEFAULT-NEXT: [[MUL_RESULT6:%.*]] = extractvalue { i64, i1 } [[MUL5]], 0
; DEFAULT-NEXT: [[MUL_OVERFLOW7:%.*]] = extractvalue { i64, i1 } [[MUL5]], 1
-; DEFAULT-NEXT: [[TMP9:%.*]] = sub i64 0, [[MUL_RESULT6]]
; DEFAULT-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[SCEVGEP4]], i64 [[MUL_RESULT6]]
; DEFAULT-NEXT: [[TMP11:%.*]] = icmp ult ptr [[TMP10]], [[SCEVGEP4]]
; DEFAULT-NEXT: [[TMP12:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW7]]
@@ -881,7 +878,6 @@ define void @test_conditional_interleave_group (ptr noalias %src.1, ptr noalias
; PRED-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; PRED-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
; PRED-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; PRED-NEXT: [[TMP1:%.*]] = sub i64 0, [[MUL_RESULT]]
; PRED-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[DST]], i64 [[MUL_RESULT]]
; PRED-NEXT: [[TMP3:%.*]] = icmp ult ptr [[TMP2]], [[DST]]
; PRED-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
@@ -889,7 +885,6 @@ define void @test_conditional_interleave_group (ptr noalias %src.1, ptr noalias
; PRED-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; PRED-NEXT: [[MUL_RESULT2:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
; PRED-NEXT: [[MUL_OVERFLOW3:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
-; PRED-NEXT: [[TMP5:%.*]] = sub i64 0, [[MUL_RESULT2]]
; PRED-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT2]]
; PRED-NEXT: [[TMP7:%.*]] = icmp ult ptr [[TMP6]], [[SCEVGEP]]
; PRED-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW3]]
@@ -897,7 +892,6 @@ define void @test_conditional_interleave_group (ptr noalias %src.1, ptr noalias
; PRED-NEXT: [[MUL5:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; PRED-NEXT: [[MUL_RESULT6:%.*]] = extractvalue { i64, i1 } [[MUL5]], 0
; PRED-NEXT: [[MUL_OVERFLOW7:%.*]] = extractvalue { i64, i1 } [[MUL5]], 1
-; PRED-NEXT: [[TMP9:%.*]] = sub i64 0, [[MUL_RESULT6]]
; PRED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[SCEVGEP4]], i64 [[MUL_RESULT6]]
; PRED-NEXT: [[TMP11:%.*]] = icmp ult ptr [[TMP10]], [[SCEVGEP4]]
; PRED-NEXT: [[TMP12:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW7]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/predicated-costs.ll b/llvm/test/Transforms/LoopVectorize/AArch64/predicated-costs.ll
index 019d2ee9886a6..ca6636869660e 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/predicated-costs.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/predicated-costs.ll
@@ -35,7 +35,6 @@ define void @test_predicated_load_cast_hint(ptr %dst.1, ptr %dst.2, ptr %src, i8
; CHECK-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 512, i64 [[TMP12]])
; CHECK-NEXT: [[MUL_RESULT2:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW3:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
-; CHECK-NEXT: [[TMP13:%.*]] = sub i64 0, [[MUL_RESULT2]]
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT2]]
; CHECK-NEXT: [[TMP15:%.*]] = icmp ult ptr [[TMP14]], [[SCEVGEP]]
; CHECK-NEXT: [[TMP16:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW3]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll b/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll
index 2c0562bf145ef..3aad98145e2aa 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll
@@ -153,7 +153,6 @@ define void @test_interleave_store_one_constant(ptr noalias %src, ptr noalias %d
; CHECK-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
-; CHECK-NEXT: [[TMP1:%.*]] = sub i64 0, [[MUL_RESULT]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[DST]], i64 [[MUL_RESULT]]
; CHECK-NEXT: [[TMP3:%.*]] = icmp ult ptr [[TMP2]], [[DST]]
; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
@@ -161,7 +160,6 @@ define void @test_interleave_store_one_constant(ptr noalias %src, ptr noalias %d
; CHECK-NEXT: [[MUL2:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; CHECK-NEXT: [[MUL_RESULT3:%.*]] = extractvalue { i64, i1 } [[MUL2]], 0
; CHECK-NEXT: [[MUL_OVERFLOW4:%.*]] = extractvalue { i64, i1 } [[MUL2]], 1
-; CHECK-NEXT: [[TMP5:%.*]] = sub i64 0, [[MUL_RESULT3]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT3]]
; CHECK-NEXT: [[TMP7:%.*]] = icmp ult ptr [[TMP6]], [[SCEVGEP]]
; CHECK-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW4]]
@@ -287,14 +285,12 @@ define void @single_fmul_used_by_each_member(ptr noalias %A, ptr noalias %B, ptr
; CHECK-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; CHECK-NEXT: [[TMP1:%.*]] = sub i64 0, [[MUL_RESULT]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT]]
; CHECK-NEXT: [[TMP3:%.*]] = icmp ult ptr [[TMP2]], [[SCEVGEP]]
; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
; CHECK-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; CHECK-NEXT: [[MUL_RESULT2:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW3:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
-; CHECK-NEXT: [[TMP5:%.*]] = sub i64 0, [[MUL_RESULT2]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[MUL_RESULT2]]
; CHECK-NEXT: [[TMP7:%.*]] = icmp ult ptr [[TMP6]], [[B]]
; CHECK-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW3]]
@@ -302,14 +298,12 @@ define void @single_fmul_used_by_each_member(ptr noalias %A, ptr noalias %B, ptr
; CHECK-NEXT: [[MUL5:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; CHECK-NEXT: [[MUL_RESULT6:%.*]] = extractvalue { i64, i1 } [[MUL5]], 0
; CHECK-NEXT: [[MUL_OVERFLOW7:%.*]] = extractvalue { i64, i1 } [[MUL5]], 1
-; CHECK-NEXT: [[TMP9:%.*]] = sub i64 0, [[MUL_RESULT6]]
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[SCEVGEP4]], i64 [[MUL_RESULT6]]
; CHECK-NEXT: [[TMP11:%.*]] = icmp ult ptr [[TMP10]], [[SCEVGEP4]]
; CHECK-NEXT: [[TMP12:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW7]]
; CHECK-NEXT: [[MUL8:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]])
; CHECK-NEXT: [[MUL_RESULT9:%.*]] = extractvalue { i64, i1 } [[MUL8]], 0
; CHECK-NEXT: [[MUL_OVERFLOW10:%.*]] = extractvalue { i64, i1 } [[MUL8]], 1
-; CHECK-NEXT: [[TMP13:%.*]] = sub i64 0, [[MUL_RESULT9]]
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[C]], i64 [[MUL_RESULT9]]
; CHECK-NEXT: [[TMP15:%.*]] = icmp ult ptr [[TMP14]], [[C]]
; CHECK-NEXT: [[TMP16:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW10]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll b/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll
index 2a2f82bfe4b3a..391653a2efe34 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll
@@ -853,7 +853,6 @@ define i32 @g(i64 %n) {
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP1]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
; CHECK: vector.scevcheck:
-; CHECK-NEXT: [[TMP28:%.*]] = trunc i64 [[N]] to i32
; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt i64 [[N]], 4294967295
; CHECK-NEXT: br i1 [[TMP2]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
; CHECK: vector.main.loop.iter.check:
diff --git a/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll b/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
index 6210a9aa66d50..9d79b625f99a4 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
@@ -119,7 +119,6 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
; CHECK-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[ARG1]])
; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; CHECK-NEXT: [[TMP1:%.*]] = sub i64 0, [[MUL_RESULT]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT]]
; CHECK-NEXT: [[TMP3:%.*]] = icmp ult ptr [[TMP2]], [[SCEVGEP]]
; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
@@ -273,14 +272,13 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: [[TMP0:%.*]] = lshr i64 [[N]], 3
; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[TMP0]], 1
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP1]], 64
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP1]], 56
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[N]], 3
; CHECK-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; CHECK-NEXT: [[TMP3:%.*]] = sub i64 0, [[MUL_RESULT]]
; CHECK-NEXT: [[TMP32:%.*]] = getelementptr i8, ptr [[A]], i64 [[MUL_RESULT]]
; CHECK-NEXT: [[TMP41:%.*]] = icmp ult ptr [[TMP32]], [[A]]
; CHECK-NEXT: [[TMP44:%.*]] = or i1 [[TMP41]], [[MUL_OVERFLOW]]
@@ -288,7 +286,6 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT2:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW3:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
-; CHECK-NEXT: [[TMP55:%.*]] = sub i64 0, [[MUL_RESULT2]]
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT2]]
; CHECK-NEXT: [[TMP5:%.*]] = icmp ult ptr [[TMP4]], [[SCEVGEP]]
; CHECK-NEXT: [[TMP57:%.*]] = or i1 [[TMP5]], [[MUL_OVERFLOW3]]
@@ -296,7 +293,6 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL2:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT3:%.*]] = extractvalue { i64, i1 } [[MUL2]], 0
; CHECK-NEXT: [[MUL_OVERFLOW4:%.*]] = extractvalue { i64, i1 } [[MUL2]], 1
-; CHECK-NEXT: [[TMP7:%.*]] = sub i64 0, [[MUL_RESULT3]]
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[SCEVGEP1]], i64 [[MUL_RESULT3]]
; CHECK-NEXT: [[TMP9:%.*]] = icmp ult ptr [[TMP8]], [[SCEVGEP1]]
; CHECK-NEXT: [[TMP10:%.*]] = or i1 [[TMP9]], [[MUL_OVERFLOW4]]
@@ -304,7 +300,6 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL6:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT7:%.*]] = extractvalue { i64, i1 } [[MUL6]], 0
; CHECK-NEXT: [[MUL_OVERFLOW8:%.*]] = extractvalue { i64, i1 } [[MUL6]], 1
-; CHECK-NEXT: [[TMP11:%.*]] = sub i64 0, [[MUL_RESULT7]]
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[SCEVGEP5]], i64 [[MUL_RESULT7]]
; CHECK-NEXT: [[TMP13:%.*]] = icmp ult ptr [[TMP12]], [[SCEVGEP5]]
; CHECK-NEXT: [[TMP14:%.*]] = or i1 [[TMP13]], [[MUL_OVERFLOW8]]
@@ -312,7 +307,6 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL10:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT11:%.*]] = extractvalue { i64, i1 } [[MUL10]], 0
; CHECK-NEXT: [[MUL_OVERFLOW12:%.*]] = extractvalue { i64, i1 } [[MUL10]], 1
-; CHECK-NEXT: [[TMP15:%.*]] = sub i64 0, [[MUL_RESULT11]]
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[SCEVGEP9]], i64 [[MUL_RESULT11]]
; CHECK-NEXT: [[TMP17:%.*]] = icmp ult ptr [[TMP16]], [[SCEVGEP9]]
; CHECK-NEXT: [[TMP18:%.*]] = or i1 [[TMP17]], [[MUL_OVERFLOW12]]
@@ -320,7 +314,6 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL14:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT15:%.*]] = extractvalue { i64, i1 } [[MUL14]], 0
; CHECK-NEXT: [[MUL_OVERFLOW16:%.*]] = extractvalue { i64, i1 } [[MUL14]], 1
-; CHECK-NEXT: [[TMP19:%.*]] = sub i64 0, [[MUL_RESULT15]]
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[SCEVGEP13]], i64 [[MUL_RESULT15]]
; CHECK-NEXT: [[TMP21:%.*]] = icmp ult ptr [[TMP20]], [[SCEVGEP13]]
; CHECK-NEXT: [[TMP22:%.*]] = or i1 [[TMP21]], [[MUL_OVERFLOW16]]
@@ -328,7 +321,6 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL18:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT19:%.*]] = extractvalue { i64, i1 } [[MUL18]], 0
; CHECK-NEXT: [[MUL_OVERFLOW20:%.*]] = extractvalue { i64, i1 } [[MUL18]], 1
-; CHECK-NEXT: [[TMP23:%.*]] = sub i64 0, [[MUL_RESULT19]]
; CHECK-NEXT: [[TMP24:%.*]] = getelementptr i8, ptr [[SCEVGEP17]], i64 [[MUL_RESULT19]]
; CHECK-NEXT: [[TMP25:%.*]] = icmp ult ptr [[TMP24]], [[SCEVGEP17]]
; CHECK-NEXT: [[TMP26:%.*]] = or i1 [[TMP25]], [[MUL_OVERFLOW20]]
@@ -336,7 +328,6 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL22:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT23:%.*]] = extractvalue { i64, i1 } [[MUL22]], 0
; CHECK-NEXT: [[MUL_OVERFLOW24:%.*]] = extractvalue { i64, i1 } [[MUL22]], 1
-; CHECK-NEXT: [[TMP27:%.*]] = sub i64 0, [[MUL_RESULT23]]
; CHECK-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[SCEVGEP21]], i64 [[MUL_RESULT23]]
; CHECK-NEXT: [[TMP29:%.*]] = icmp ult ptr [[TMP28]], [[SCEVGEP21]]
; CHECK-NEXT: [[TMP30:%.*]] = or i1 [[TMP29]], [[MUL_OVERFLOW24]]
@@ -344,14 +335,12 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
; CHECK-NEXT: [[MUL29:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT30:%.*]] = extractvalue { i64, i1 } [[MUL29]], 0
; CHECK-NEXT: [[MUL_OVERFLOW31:%.*]] = extractvalue { i64, i1 } [[MUL29]], 1
-; CHECK-NEXT: [[TMP67:%.*]] = sub i64 0, [[MUL_RESULT30]]
; CHECK-NEXT: [[TMP68:%.*]] = getelementptr i8, ptr [[SCEVGEP31]], i64 [[MUL_RESULT30]]
; CHECK-NEXT: [[TMP69:%.*]] = icmp ult ptr [[TMP68]], [[SCEVGEP31]]
; CHECK-NEXT: [[TMP70:%.*]] = or i1 [[TMP69]], [[MUL_OVERFLOW31]]
; CHECK-NEXT: [[MUL25:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 32, i64 [[TMP2]])
; CHECK-NEXT: [[MUL_RESULT26:%.*]] = extractvalue { i64, i1 } [[MUL25]], 0
; CHECK-NEXT: [[MUL_OVERFLOW27:%.*]] = extractvalue { i64, i1 } [[MUL25]], 1
-; CHECK-NEXT: [[TMP31:%.*]] = sub i64 0, [[MUL_RESULT26]]
; CHECK-NEXT: [[TMP71:%.*]] = getelementptr i8, ptr [[B]], i64 [[MUL_RESULT26]]
; CHECK-NEXT: [[TMP33:%.*]] = icmp ult ptr [[TMP71]], [[B]]
; CHECK-NEXT: [[TMP34:%.*]] = or i1 [[TMP33]], [[MUL_OVERFLOW27]]
diff --git a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-requiring-scev-predicates.ll b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-requiring-scev-predicates.ll
index 9bf3c448e2cce..51bbc50723263 100644
--- a/llvm/test/Transforms/LoopVectori...
[truncated]
|
nikic
reviewed
Sep 6, 2025
There was a problem hiding this comment.
Probably makes sense to iterate in reverse(), as a later dead instruction may render an earlier one dead as well.
There was a problem hiding this comment.
I switched to a worklist, as the inserted values are stored in DenseSets to make sure there's no differences based on the iteration order over the sets
Add new helper to erase dead instructions inserted during SCEV expansion but not being used due to InstSimplifyFolder simplifications. Together with llvm#157307 this also allows removing some specialized folds, e.g. https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp#L2205
fhahn
force-pushed
the
scevexp-remove-dead
branch
from
b50a0ae to
728902c
Compare
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Sep 8, 2025
…xpansion. (#157308) Add new helper to erase dead instructions inserted during SCEV expansion but not being used due to InstSimplifyFolder simplifications. Together with llvm/llvm-project#157307 this also allows removing some specialized folds, e.g. https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp#L2205 PR: llvm/llvm-project#157308
fhahn
added a commit
that referenced
this pull request
Sep 8, 2025
…nsion. (#157308)" This reverts commit 528b13d. Triggers MSan errors in some configurations, e.g. https://lab.llvm.org/buildbot/#/builders/169/builds/14799
fhahn
added a commit
that referenced
this pull request
Sep 9, 2025
…ansion. (#157308)" This reverts commit eeb4380. Recommit with with a fix for MSan failure ( https://lab.llvm.org/buildbot/#/builders/169/builds/14799), by adding a set to track deleted values. Using the InsertedInstructions set is not sufficient, as it use asserting value handles as keys, which may dereference the value at construction. Original message: Add new helper to erase dead instructions inserted during SCEV expansion but not being used due to InstSimplifyFolder simplifications. Together with #157307 this also allows removing some specialized folds, e.g. https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp#L2205 PR: #157308
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Sep 9, 2025
…s after expansion. (#157308)" This reverts commit eeb4380. Recommit with with a fix for MSan failure ( https://lab.llvm.org/buildbot/#/builders/169/builds/14799), by adding a set to track deleted values. Using the InsertedInstructions set is not sufficient, as it use asserting value handles as keys, which may dereference the value at construction. Original message: Add new helper to erase dead instructions inserted during SCEV expansion but not being used due to InstSimplifyFolder simplifications. Together with llvm/llvm-project#157307 this also allows removing some specialized folds, e.g. https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp#L2205 PR: llvm/llvm-project#157308
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Add new helper to erase dead instructions inserted during SCEV expansion but not being used due to InstSimplifyFolder simplifications.
Together with #157307 this also allows removing some specialized folds, e.g. https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp#L2205