[llvm] [Instcombine] Combine extractelement from a vector_extract at index 0 (PR #151491)
Kerry McLaughlin via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 31 06:07:31 PDT 2025
https://github.com/kmclaughlin-arm updated https://github.com/llvm/llvm-project/pull/151491
>From 03413139be5f006859c933203922c7f63bfd2a60 Mon Sep 17 00:00:00 2001
From: Kerry McLaughlin <kerry.mclaughlin at arm.com>
Date: Thu, 17 Jul 2025 08:49:24 +0000
Subject: [PATCH 1/3] - Add a test for extractelement from scalable
vector_extract at index 0
---
.../InstCombine/scalable-extract-subvec-elt.ll | 14 ++++++++++++++
1 file changed, 14 insertions(+)
create mode 100644 llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
diff --git a/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll b/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
new file mode 100644
index 0000000000000..d0cfc48420b43
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
@@ -0,0 +1,14 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -passes=instcombine < %s | FileCheck %s
+
+define i1 @scalable_test(<vscale x 4 x i1> %a) {
+; CHECK-LABEL: define i1 @scalable_test(
+; CHECK-SAME: <vscale x 4 x i1> [[A:%.*]]) {
+; CHECK-NEXT: [[SUBVEC:%.*]] = call <vscale x 2 x i1> @llvm.vector.extract.nxv2i1.nxv4i1(<vscale x 4 x i1> [[A]], i64 0)
+; CHECK-NEXT: [[ELT:%.*]] = extractelement <vscale x 2 x i1> [[SUBVEC]], i64 1
+; CHECK-NEXT: ret i1 [[ELT]]
+;
+ %subvec = call <vscale x 2 x i1> @llvm.vector.extract.nxv2i1.nxv4i1.i64(<vscale x 4 x i1> %a, i64 0)
+ %elt = extractelement <vscale x 2 x i1> %subvec, i32 1
+ ret i1 %elt
+}
>From d93bdead6c6d36907c3ed539407736070792f92b Mon Sep 17 00:00:00 2001
From: Kerry McLaughlin <kerry.mclaughlin at arm.com>
Date: Thu, 17 Jul 2025 08:51:47 +0000
Subject: [PATCH 2/3] [Instcombine] Combine extractelement from a
vector_extract at index 0
Extracting any element from a subvector starting at index 0 is
equivalent to extracting from the original vector, i.e.
extract_elt(vector_extract(x, 0), y) -> extract_elt(x, y)
---
llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp | 6 +++++-
.../Transforms/InstCombine/scalable-extract-subvec-elt.ll | 3 +--
2 files changed, 6 insertions(+), 3 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
index 00b877b8a07ef..6f2adba9e3f6b 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
@@ -444,7 +444,11 @@ Instruction *InstCombinerImpl::visitExtractElementInst(ExtractElementInst &EI) {
else
Idx = PoisonValue::get(Ty);
return replaceInstUsesWith(EI, Idx);
- }
+ } else if (IID == Intrinsic::vector_extract)
+ // If II is a subvector starting at index 0, extract from the wider
+ // source vector
+ if (cast<ConstantInt>(II->getArgOperand(1))->getZExtValue() == 0)
+ return ExtractElementInst::Create(II->getArgOperand(0), Index);
}
// InstSimplify should handle cases where the index is invalid.
diff --git a/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll b/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
index d0cfc48420b43..38cbeb0df98bd 100644
--- a/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
+++ b/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
@@ -4,8 +4,7 @@
define i1 @scalable_test(<vscale x 4 x i1> %a) {
; CHECK-LABEL: define i1 @scalable_test(
; CHECK-SAME: <vscale x 4 x i1> [[A:%.*]]) {
-; CHECK-NEXT: [[SUBVEC:%.*]] = call <vscale x 2 x i1> @llvm.vector.extract.nxv2i1.nxv4i1(<vscale x 4 x i1> [[A]], i64 0)
-; CHECK-NEXT: [[ELT:%.*]] = extractelement <vscale x 2 x i1> [[SUBVEC]], i64 1
+; CHECK-NEXT: [[ELT:%.*]] = extractelement <vscale x 4 x i1> [[A]], i64 1
; CHECK-NEXT: ret i1 [[ELT]]
;
%subvec = call <vscale x 2 x i1> @llvm.vector.extract.nxv2i1.nxv4i1.i64(<vscale x 4 x i1> %a, i64 0)
>From d9b7aa9557075c166a34ac7dc9ba4d77c2ac36d4 Mon Sep 17 00:00:00 2001
From: Kerry McLaughlin <kerry.mclaughlin at arm.com>
Date: Thu, 31 Jul 2025 12:26:50 +0000
Subject: [PATCH 3/3] - Add variable index & negative tests
---
.../InstCombine/InstCombineVectorOps.cpp | 15 ++++++-----
.../scalable-extract-subvec-elt.ll | 27 +++++++++++++++++--
2 files changed, 34 insertions(+), 8 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
index 6f2adba9e3f6b..cb188ea9a68ec 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
@@ -419,6 +419,7 @@ Instruction *InstCombinerImpl::visitExtractElementInst(ExtractElementInst &EI) {
// If extracting a specified index from the vector, see if we can recursively
// find a previously computed scalar that was inserted into the vector.
auto *IndexC = dyn_cast<ConstantInt>(Index);
+ auto *II = dyn_cast<IntrinsicInst>(SrcVec);
bool HasKnownValidIndex = false;
if (IndexC) {
// Canonicalize type of constant indices to i64 to simplify CSE
@@ -429,7 +430,7 @@ Instruction *InstCombinerImpl::visitExtractElementInst(ExtractElementInst &EI) {
unsigned NumElts = EC.getKnownMinValue();
HasKnownValidIndex = IndexC->getValue().ult(NumElts);
- if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(SrcVec)) {
+ if (II) {
Intrinsic::ID IID = II->getIntrinsicID();
// Index needs to be lower than the minimum size of the vector, because
// for scalable vector, the vector size is known at run time.
@@ -444,11 +445,7 @@ Instruction *InstCombinerImpl::visitExtractElementInst(ExtractElementInst &EI) {
else
Idx = PoisonValue::get(Ty);
return replaceInstUsesWith(EI, Idx);
- } else if (IID == Intrinsic::vector_extract)
- // If II is a subvector starting at index 0, extract from the wider
- // source vector
- if (cast<ConstantInt>(II->getArgOperand(1))->getZExtValue() == 0)
- return ExtractElementInst::Create(II->getArgOperand(0), Index);
+ }
}
// InstSimplify should handle cases where the index is invalid.
@@ -466,6 +463,12 @@ Instruction *InstCombinerImpl::visitExtractElementInst(ExtractElementInst &EI) {
return ScalarPHI;
}
+ // If SrcVec is a subvector starting at index 0, extract from the
+ // wider source vector
+ if (II && II->getIntrinsicID() == Intrinsic::vector_extract)
+ if (cast<ConstantInt>(II->getArgOperand(1))->isZero())
+ return ExtractElementInst::Create(II->getArgOperand(0), Index);
+
// TODO come up with a n-ary matcher that subsumes both unary and
// binary matchers.
UnaryOperator *UO;
diff --git a/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll b/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
index 38cbeb0df98bd..1e089e1168f66 100644
--- a/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
+++ b/llvm/test/Transforms/InstCombine/scalable-extract-subvec-elt.ll
@@ -1,8 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -passes=instcombine < %s | FileCheck %s
-define i1 @scalable_test(<vscale x 4 x i1> %a) {
-; CHECK-LABEL: define i1 @scalable_test(
+define i1 @extract_const_idx(<vscale x 4 x i1> %a) {
+; CHECK-LABEL: define i1 @extract_const_idx(
; CHECK-SAME: <vscale x 4 x i1> [[A:%.*]]) {
; CHECK-NEXT: [[ELT:%.*]] = extractelement <vscale x 4 x i1> [[A]], i64 1
; CHECK-NEXT: ret i1 [[ELT]]
@@ -11,3 +11,26 @@ define i1 @scalable_test(<vscale x 4 x i1> %a) {
%elt = extractelement <vscale x 2 x i1> %subvec, i32 1
ret i1 %elt
}
+
+define float @extract_variable_idx(<vscale x 4 x float> %a, i32 %idx) {
+; CHECK-LABEL: define float @extract_variable_idx(
+; CHECK-SAME: <vscale x 4 x float> [[A:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[ELT:%.*]] = extractelement <vscale x 4 x float> [[A]], i32 [[IDX]]
+; CHECK-NEXT: ret float [[ELT]]
+;
+ %subvec = call <vscale x 2 x float> @llvm.vector.extract.nxv2f32.nxv4f32.i64(<vscale x 4 x float> %a, i64 0)
+ %elt = extractelement <vscale x 2 x float> %subvec, i32 %idx
+ ret float %elt
+}
+
+define float @negative_test(<vscale x 4 x float> %a) {
+; CHECK-LABEL: define float @negative_test(
+; CHECK-SAME: <vscale x 4 x float> [[A:%.*]]) {
+; CHECK-NEXT: [[SUBVEC:%.*]] = call <vscale x 2 x float> @llvm.vector.extract.nxv2f32.nxv4f32(<vscale x 4 x float> [[A]], i64 2)
+; CHECK-NEXT: [[ELT:%.*]] = extractelement <vscale x 2 x float> [[SUBVEC]], i64 1
+; CHECK-NEXT: ret float [[ELT]]
+;
+ %subvec = call <vscale x 2 x float> @llvm.vector.extract.nxv2f32.nxv4f32.i64(<vscale x 4 x float> %a, i64 2)
+ %elt = extractelement <vscale x 2 x float> %subvec, i32 1
+ ret float %elt
+}
More information about the llvm-commits
mailing list