[llvm] 2d7bab6 - [InstCombine] Always create new GEPs when swapping GEPs
Nikita Popov via llvm-commits
llvm-commits at lists.llvm.org
Mon May 30 07:49:25 PDT 2022
Author: Nikita Popov
Date: 2022-05-30T16:48:52+02:00
New Revision: 2d7bab666f5a71a70da0ed7925fda5ce634d21e5
URL: https://github.com/llvm/llvm-project/commit/2d7bab666f5a71a70da0ed7925fda5ce634d21e5
DIFF: https://github.com/llvm/llvm-project/commit/2d7bab666f5a71a70da0ed7925fda5ce634d21e5.diff
LOG: [InstCombine] Always create new GEPs when swapping GEPs
As the long explanatory comment attests, performing the modification
in place is pretty tricky. Drop this unnecessary complexity and
always create new instructions.
This should be NFC-ish, but can probably cause difference due to
worklist order.
Added:
Modified:
llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
llvm/test/Transforms/InstCombine/gep-combine-loop-invariant.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
index cd1e4f409d9e..ad17f8ab5394 100644
--- a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -1969,43 +1969,15 @@ Instruction *InstCombinerImpl::visitGEPOfGEP(GetElementPtrInst &GEP,
// invariant: this breaks the dependence between GEPs and allows LICM
// to hoist the invariant part out of the loop.
if (L->isLoopInvariant(GO1) && !L->isLoopInvariant(SO1)) {
- // We have to be careful here.
- // We have something like:
- // %src = getelementptr <ty>, <ty>* %base, <ty> %idx
- // %gep = getelementptr <ty>, <ty>* %src, <ty> %idx2
- // If we just swap idx & idx2 then we could inadvertantly
- // change %src from a vector to a scalar, or vice versa.
- // Cases:
- // 1) %base a scalar & idx a scalar & idx2 a vector
- // => Swapping idx & idx2 turns %src into a vector type.
- // 2) %base a scalar & idx a vector & idx2 a scalar
- // => Swapping idx & idx2 turns %src in a scalar type
- // 3) %base, %idx, and %idx2 are scalars
- // => %src & %gep are scalars
- // => swapping idx & idx2 is safe
- // 4) %base a vector
- // => %src is a vector
- // => swapping idx & idx2 is safe.
- auto *SO0 = Src->getOperand(0);
- auto *SO0Ty = SO0->getType();
- if (!isa<VectorType>(GEP.getType()) || // case 3
- isa<VectorType>(SO0Ty)) { // case 4
- Src->setOperand(1, GO1);
- GEP.setOperand(1, SO1);
- return &GEP;
- } else {
- // Case 1 or 2
- // -- have to recreate %src & %gep
- // put NewSrc at same location as %src
- Builder.SetInsertPoint(cast<Instruction>(Src));
- Value *NewSrc =
- Builder.CreateGEP(GEP.getSourceElementType(), SO0, GO1,
- Src->getName(), Src->isInBounds());
- GetElementPtrInst *NewGEP = GetElementPtrInst::Create(
- GEP.getSourceElementType(), NewSrc, {SO1});
- NewGEP->setIsInBounds(GEP.isInBounds());
- return NewGEP;
- }
+ // Put NewSrc at same location as %src.
+ Builder.SetInsertPoint(cast<Instruction>(Src));
+ Value *NewSrc = Builder.CreateGEP(GEP.getSourceElementType(),
+ Src->getPointerOperand(), GO1,
+ Src->getName(), Src->isInBounds());
+ GetElementPtrInst *NewGEP = GetElementPtrInst::Create(
+ GEP.getSourceElementType(), NewSrc, {SO1});
+ NewGEP->setIsInBounds(GEP.isInBounds());
+ return NewGEP;
}
}
}
diff --git a/llvm/test/Transforms/InstCombine/gep-combine-loop-invariant.ll b/llvm/test/Transforms/InstCombine/gep-combine-loop-invariant.ll
index adfe20bb2ad7..5a98df1536ed 100644
--- a/llvm/test/Transforms/InstCombine/gep-combine-loop-invariant.ll
+++ b/llvm/test/Transforms/InstCombine/gep-combine-loop-invariant.ll
@@ -20,9 +20,9 @@ define i32 @foo(i8* nocapture readnone %match, i32 %cur_match, i32 %best_len, i3
; CHECK-NEXT: br label [[IF_THEN:%.*]]
; CHECK: do.body:
; CHECK-NEXT: [[IDX_EXT:%.*]] = zext i32 [[TMP4:%.*]] to i64
-; CHECK-NEXT: [[ADD_PTR:%.*]] = getelementptr inbounds i8, i8* [[WIN]], i64 [[IDX_EXT1]]
-; CHECK-NEXT: [[ADD_PTR2:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR]], i64 -1
-; CHECK-NEXT: [[ADD_PTR3:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR2]], i64 [[IDX_EXT]]
+; CHECK-NEXT: [[ADD_PTR1:%.*]] = getelementptr inbounds i8, i8* [[WIN]], i64 [[IDX_EXT1]]
+; CHECK-NEXT: [[ADD_PTR22:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR1]], i64 -1
+; CHECK-NEXT: [[ADD_PTR3:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR22]], i64 [[IDX_EXT]]
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8* [[ADD_PTR3]] to i32*
; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[TMP3]], [[SCAN_END]]
@@ -164,8 +164,8 @@ define void @PR37005_3(<2 x i8*> %base, i8** %in) {
; CHECK-NEXT: [[PI1:%.*]] = ptrtoint <2 x i8**> [[E4]] to <2 x i64>
; CHECK-NEXT: [[TMP0:%.*]] = lshr <2 x i64> [[PI1]], <i64 14, i64 14>
; CHECK-NEXT: [[SL1:%.*]] = and <2 x i64> [[TMP0]], <i64 1125899906842496, i64 1125899906842496>
-; CHECK-NEXT: [[E5:%.*]] = getelementptr inbounds i8, <2 x i8*> [[BASE:%.*]], i64 80
-; CHECK-NEXT: [[E6:%.*]] = getelementptr inbounds i8, <2 x i8*> [[E5]], <2 x i64> [[SL1]]
+; CHECK-NEXT: [[E51:%.*]] = getelementptr inbounds i8, <2 x i8*> [[BASE:%.*]], i64 80
+; CHECK-NEXT: [[E6:%.*]] = getelementptr inbounds i8, <2 x i8*> [[E51]], <2 x i64> [[SL1]]
; CHECK-NEXT: call void @blackhole(<2 x i8*> [[E6]])
; CHECK-NEXT: br label [[LOOP]]
;
@@ -264,8 +264,8 @@ define void @only_one_inbounds(i8* %ptr, i1 %c, i32 noundef %arg1, i32 noundef %
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[ARG1_EXT:%.*]] = zext i32 [[ARG1:%.*]] to i64
-; CHECK-NEXT: [[PTR2:%.*]] = getelementptr inbounds i8, i8* [[PTR:%.*]], i64 [[ARG2_EXT]]
-; CHECK-NEXT: [[PTR3:%.*]] = getelementptr i8, i8* [[PTR2]], i64 [[ARG1_EXT]]
+; CHECK-NEXT: [[PTR21:%.*]] = getelementptr inbounds i8, i8* [[PTR:%.*]], i64 [[ARG2_EXT]]
+; CHECK-NEXT: [[PTR3:%.*]] = getelementptr i8, i8* [[PTR21]], i64 [[ARG1_EXT]]
; CHECK-NEXT: call void @use(i8* [[PTR3]])
; CHECK-NEXT: br i1 [[C:%.*]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
@@ -292,8 +292,8 @@ define void @both_inbounds_one_neg(i8* %ptr, i1 %c, i32 noundef %arg) {
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[ARG_EXT:%.*]] = zext i32 [[ARG:%.*]] to i64
-; CHECK-NEXT: [[PTR2:%.*]] = getelementptr inbounds i8, i8* [[PTR:%.*]], i64 -1
-; CHECK-NEXT: [[PTR3:%.*]] = getelementptr inbounds i8, i8* [[PTR2]], i64 [[ARG_EXT]]
+; CHECK-NEXT: [[PTR21:%.*]] = getelementptr inbounds i8, i8* [[PTR:%.*]], i64 -1
+; CHECK-NEXT: [[PTR3:%.*]] = getelementptr inbounds i8, i8* [[PTR21]], i64 [[ARG_EXT]]
; CHECK-NEXT: call void @use(i8* nonnull [[PTR3]])
; CHECK-NEXT: br i1 [[C:%.*]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
@@ -319,8 +319,8 @@ define void @both_inbounds_pos(i8* %ptr, i1 %c, i32 noundef %arg) {
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[ARG_EXT:%.*]] = zext i32 [[ARG:%.*]] to i64
-; CHECK-NEXT: [[PTR2:%.*]] = getelementptr inbounds i8, i8* [[PTR:%.*]], i64 1
-; CHECK-NEXT: [[PTR3:%.*]] = getelementptr inbounds i8, i8* [[PTR2]], i64 [[ARG_EXT]]
+; CHECK-NEXT: [[PTR21:%.*]] = getelementptr inbounds i8, i8* [[PTR:%.*]], i64 1
+; CHECK-NEXT: [[PTR3:%.*]] = getelementptr inbounds i8, i8* [[PTR21]], i64 [[ARG_EXT]]
; CHECK-NEXT: call void @use(i8* nonnull [[PTR3]])
; CHECK-NEXT: br i1 [[C:%.*]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
More information about the llvm-commits
mailing list