[LLVM][ConstProp] Enable intrinsic simplifications for vector ConstantInt based operands. #159358
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
…tInt based operands. Simplifcation of vector.reduce intrinsics are prevented by an early bailout for ConstantInt base operands. This PR removes the bailout and updates the tests to show matching output when -use-constant-int-for-*-splat is used. No new simplications are added, which is why I only add scalable vector tests for things like bswap and not the vector reductions.
llvmbot
added
llvm:instcombine
|
@llvm/pr-subscribers-llvm-transforms Author: Paul Walker (paulwalker-arm) ChangesSimplifcation of vector.reduce intrinsics are prevented by an early bailout for ConstantInt base operands. This PR removes the bailout and updates the tests to show matching output when -use-constant-int-for-*-splat is used. No new simplications are added, which is why I only add scalable vector tests for things like bswap and not the vector reductions. Full diff: https://github.com/llvm/llvm-project/pull/159358.diff 6 Files Affected:
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index a3b2e62a1b8ba..e50120d345ebe 100755
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -2165,7 +2165,7 @@ Constant *constantFoldVectorReduce(Intrinsic::ID IID, Constant *Op) {
return PoisonValue::get(VT->getElementType());
// TODO: Handle undef.
- if (!isa<ConstantVector>(Op) && !isa<ConstantDataVector>(Op))
+ if (!isa<ConstantVector, ConstantDataVector, ConstantInt>(Op))
return nullptr;
auto *EltC = dyn_cast<ConstantInt>(Op->getAggregateElement(0U));
@@ -3040,9 +3040,6 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
Val = Val | Val << 1;
return ConstantInt::get(Ty, Val);
}
-
- default:
- return nullptr;
}
}
@@ -3063,9 +3060,8 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
}
// Support ConstantVector in case we have an Undef in the top.
- if (isa<ConstantVector>(Operands[0]) ||
- isa<ConstantDataVector>(Operands[0]) ||
- isa<ConstantAggregateZero>(Operands[0])) {
+ if (isa<ConstantVector, ConstantDataVector, ConstantAggregateZero,
+ ConstantInt>(Operands[0])) {
auto *Op = cast<Constant>(Operands[0]);
switch (IntrinsicID) {
default: break;
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll
index 7b30edbf7792b..71dad41b971b5 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll
@@ -1,6 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -passes=instsimplify -S < %s | FileCheck %s
+; RUN: opt -passes=instsimplify -use-constant-int-for-fixed-length-splat -S < %s | FileCheck %s
; Test that intrinsics wasm call are constant folded
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll b/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll
index 68b45a94af4bc..f68b85ed4db26 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll
@@ -1,5 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -use-constant-int-for-scalable-splat -S | FileCheck %s
declare i31 @llvm.ctpop.i31(i31 %val)
declare i32 @llvm.cttz.i32(i32 %val, i1)
@@ -120,6 +121,22 @@ define <2 x i31> @ctpop_vector() {
ret <2 x i31> %x
}
+define <2 x i31> @ctpop_vector_splat_v2i31() {
+; CHECK-LABEL: @ctpop_vector_splat_v2i31(
+; CHECK-NEXT: ret <2 x i31> splat (i31 1)
+;
+ %x = call <2 x i31> @llvm.ctpop.v2i31(<2 x i31> splat(i31 16))
+ ret <2 x i31> %x
+}
+
+define <vscale x 2 x i31> @ctpop_vector_splat_nxv2i31() {
+; CHECK-LABEL: @ctpop_vector_splat_nxv2i31(
+; CHECK-NEXT: ret <vscale x 2 x i31> splat (i31 1)
+;
+ %x = call <vscale x 2 x i31> @llvm.ctpop.nxv2i31(<vscale x 2 x i31> splat(i31 16))
+ ret <vscale x 2 x i31> %x
+}
+
define <2 x i31> @ctpop_vector_undef() {
; CHECK-LABEL: @ctpop_vector_undef(
; CHECK-NEXT: ret <2 x i31> zeroinitializer
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/bitreverse.ll b/llvm/test/Transforms/InstSimplify/ConstProp/bitreverse.ll
new file mode 100644
index 0000000000000..409141a2c872b
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/bitreverse.ll
@@ -0,0 +1,51 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -use-constant-int-for-scalable-splat -S | FileCheck %s
+
+define i16 @W() {
+; CHECK-LABEL: define i16 @W() {
+; CHECK-NEXT: ret i16 -32768
+;
+ %Z = call i16 @llvm.bitreverse.i16(i16 1)
+ ret i16 %Z
+}
+
+define i32 @X() {
+; CHECK-LABEL: define i32 @X() {
+; CHECK-NEXT: ret i32 -2147483648
+;
+ %Z = call i32 @llvm.bitreverse.i32(i32 1)
+ ret i32 %Z
+}
+
+define i64 @Y() {
+; CHECK-LABEL: define i64 @Y() {
+; CHECK-NEXT: ret i64 -9223372036854775808
+;
+ %Z = call i64 @llvm.bitreverse.i64(i64 1)
+ ret i64 %Z
+}
+
+define i80 @Z() {
+; CHECK-LABEL: define i80 @Z() {
+; CHECK-NEXT: ret i80 23777929115895377691656
+;
+ %Z = call i80 @llvm.bitreverse.i80(i80 76151636403560493650080)
+ ret i80 %Z
+}
+
+define <4 x i32> @bitreverse_splat_v4i32() {
+; CHECK-LABEL: define <4 x i32> @bitreverse_splat_v4i32() {
+; CHECK-NEXT: ret <4 x i32> splat (i32 -2147483648)
+;
+ %Z = call <4 x i32> @llvm.bitreverse.v4i32(<4 x i32> splat(i32 1))
+ ret <4 x i32> %Z
+}
+
+define <vscale x 4 x i32> @bitreverse_splat_nxv4i32() {
+; CHECK-LABEL: define <vscale x 4 x i32> @bitreverse_splat_nxv4i32() {
+; CHECK-NEXT: ret <vscale x 4 x i32> splat (i32 -2147483648)
+;
+ %Z = call <vscale x 4 x i32> @llvm.bitreverse.v4i32(<vscale x 4 x i32> splat(i32 1))
+ ret <vscale x 4 x i32> %Z
+}
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll b/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll
index 42bb73344995b..4db8ced58327a 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll
@@ -2,6 +2,7 @@
; bswap should be constant folded when it is passed a constant argument
; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -use-constant-int-for-scalable-splat -S | FileCheck %s
declare i16 @llvm.bswap.i16(i16)
@@ -42,3 +43,19 @@ define i80 @Z() {
%Z = call i80 @llvm.bswap.i80( i80 76151636403560493650080 )
ret i80 %Z
}
+
+define <4 x i32> @bswap_splat_v4i32() {
+; CHECK-LABEL: define <4 x i32> @bswap_splat_v4i32() {
+; CHECK-NEXT: ret <4 x i32> splat (i32 16777216)
+;
+ %Z = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> splat(i32 1))
+ ret <4 x i32> %Z
+}
+
+define <vscale x 4 x i32> @bswap_splat_nxv4i32() {
+; CHECK-LABEL: define <vscale x 4 x i32> @bswap_splat_nxv4i32() {
+; CHECK-NEXT: ret <vscale x 4 x i32> splat (i32 16777216)
+;
+ %Z = call <vscale x 4 x i32> @llvm.bswap.v4i32(<vscale x 4 x i32> splat(i32 1))
+ ret <vscale x 4 x i32> %Z
+}
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll b/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll
index e994921f62574..9f9e3f9ffc070 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll
@@ -1,5 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -S | FileCheck %s
declare i32 @llvm.vector.reduce.add.v1i32(<1 x i32> %a)
declare i32 @llvm.vector.reduce.add.v8i32(<8 x i32> %a)
|
|
@llvm/pr-subscribers-llvm-analysis Author: Paul Walker (paulwalker-arm) ChangesSimplifcation of vector.reduce intrinsics are prevented by an early bailout for ConstantInt base operands. This PR removes the bailout and updates the tests to show matching output when -use-constant-int-for-*-splat is used. No new simplications are added, which is why I only add scalable vector tests for things like bswap and not the vector reductions. Full diff: https://github.com/llvm/llvm-project/pull/159358.diff 6 Files Affected:
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index a3b2e62a1b8ba..e50120d345ebe 100755
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -2165,7 +2165,7 @@ Constant *constantFoldVectorReduce(Intrinsic::ID IID, Constant *Op) {
return PoisonValue::get(VT->getElementType());
// TODO: Handle undef.
- if (!isa<ConstantVector>(Op) && !isa<ConstantDataVector>(Op))
+ if (!isa<ConstantVector, ConstantDataVector, ConstantInt>(Op))
return nullptr;
auto *EltC = dyn_cast<ConstantInt>(Op->getAggregateElement(0U));
@@ -3040,9 +3040,6 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
Val = Val | Val << 1;
return ConstantInt::get(Ty, Val);
}
-
- default:
- return nullptr;
}
}
@@ -3063,9 +3060,8 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
}
// Support ConstantVector in case we have an Undef in the top.
- if (isa<ConstantVector>(Operands[0]) ||
- isa<ConstantDataVector>(Operands[0]) ||
- isa<ConstantAggregateZero>(Operands[0])) {
+ if (isa<ConstantVector, ConstantDataVector, ConstantAggregateZero,
+ ConstantInt>(Operands[0])) {
auto *Op = cast<Constant>(Operands[0]);
switch (IntrinsicID) {
default: break;
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll
index 7b30edbf7792b..71dad41b971b5 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/WebAssembly/any_all_true.ll
@@ -1,6 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -passes=instsimplify -S < %s | FileCheck %s
+; RUN: opt -passes=instsimplify -use-constant-int-for-fixed-length-splat -S < %s | FileCheck %s
; Test that intrinsics wasm call are constant folded
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll b/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll
index 68b45a94af4bc..f68b85ed4db26 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/bitcount.ll
@@ -1,5 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -use-constant-int-for-scalable-splat -S | FileCheck %s
declare i31 @llvm.ctpop.i31(i31 %val)
declare i32 @llvm.cttz.i32(i32 %val, i1)
@@ -120,6 +121,22 @@ define <2 x i31> @ctpop_vector() {
ret <2 x i31> %x
}
+define <2 x i31> @ctpop_vector_splat_v2i31() {
+; CHECK-LABEL: @ctpop_vector_splat_v2i31(
+; CHECK-NEXT: ret <2 x i31> splat (i31 1)
+;
+ %x = call <2 x i31> @llvm.ctpop.v2i31(<2 x i31> splat(i31 16))
+ ret <2 x i31> %x
+}
+
+define <vscale x 2 x i31> @ctpop_vector_splat_nxv2i31() {
+; CHECK-LABEL: @ctpop_vector_splat_nxv2i31(
+; CHECK-NEXT: ret <vscale x 2 x i31> splat (i31 1)
+;
+ %x = call <vscale x 2 x i31> @llvm.ctpop.nxv2i31(<vscale x 2 x i31> splat(i31 16))
+ ret <vscale x 2 x i31> %x
+}
+
define <2 x i31> @ctpop_vector_undef() {
; CHECK-LABEL: @ctpop_vector_undef(
; CHECK-NEXT: ret <2 x i31> zeroinitializer
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/bitreverse.ll b/llvm/test/Transforms/InstSimplify/ConstProp/bitreverse.ll
new file mode 100644
index 0000000000000..409141a2c872b
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/bitreverse.ll
@@ -0,0 +1,51 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -use-constant-int-for-scalable-splat -S | FileCheck %s
+
+define i16 @W() {
+; CHECK-LABEL: define i16 @W() {
+; CHECK-NEXT: ret i16 -32768
+;
+ %Z = call i16 @llvm.bitreverse.i16(i16 1)
+ ret i16 %Z
+}
+
+define i32 @X() {
+; CHECK-LABEL: define i32 @X() {
+; CHECK-NEXT: ret i32 -2147483648
+;
+ %Z = call i32 @llvm.bitreverse.i32(i32 1)
+ ret i32 %Z
+}
+
+define i64 @Y() {
+; CHECK-LABEL: define i64 @Y() {
+; CHECK-NEXT: ret i64 -9223372036854775808
+;
+ %Z = call i64 @llvm.bitreverse.i64(i64 1)
+ ret i64 %Z
+}
+
+define i80 @Z() {
+; CHECK-LABEL: define i80 @Z() {
+; CHECK-NEXT: ret i80 23777929115895377691656
+;
+ %Z = call i80 @llvm.bitreverse.i80(i80 76151636403560493650080)
+ ret i80 %Z
+}
+
+define <4 x i32> @bitreverse_splat_v4i32() {
+; CHECK-LABEL: define <4 x i32> @bitreverse_splat_v4i32() {
+; CHECK-NEXT: ret <4 x i32> splat (i32 -2147483648)
+;
+ %Z = call <4 x i32> @llvm.bitreverse.v4i32(<4 x i32> splat(i32 1))
+ ret <4 x i32> %Z
+}
+
+define <vscale x 4 x i32> @bitreverse_splat_nxv4i32() {
+; CHECK-LABEL: define <vscale x 4 x i32> @bitreverse_splat_nxv4i32() {
+; CHECK-NEXT: ret <vscale x 4 x i32> splat (i32 -2147483648)
+;
+ %Z = call <vscale x 4 x i32> @llvm.bitreverse.v4i32(<vscale x 4 x i32> splat(i32 1))
+ ret <vscale x 4 x i32> %Z
+}
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll b/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll
index 42bb73344995b..4db8ced58327a 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/bswap.ll
@@ -2,6 +2,7 @@
; bswap should be constant folded when it is passed a constant argument
; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -use-constant-int-for-scalable-splat -S | FileCheck %s
declare i16 @llvm.bswap.i16(i16)
@@ -42,3 +43,19 @@ define i80 @Z() {
%Z = call i80 @llvm.bswap.i80( i80 76151636403560493650080 )
ret i80 %Z
}
+
+define <4 x i32> @bswap_splat_v4i32() {
+; CHECK-LABEL: define <4 x i32> @bswap_splat_v4i32() {
+; CHECK-NEXT: ret <4 x i32> splat (i32 16777216)
+;
+ %Z = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> splat(i32 1))
+ ret <4 x i32> %Z
+}
+
+define <vscale x 4 x i32> @bswap_splat_nxv4i32() {
+; CHECK-LABEL: define <vscale x 4 x i32> @bswap_splat_nxv4i32() {
+; CHECK-NEXT: ret <vscale x 4 x i32> splat (i32 16777216)
+;
+ %Z = call <vscale x 4 x i32> @llvm.bswap.v4i32(<vscale x 4 x i32> splat(i32 1))
+ ret <vscale x 4 x i32> %Z
+}
diff --git a/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll b/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll
index e994921f62574..9f9e3f9ffc070 100644
--- a/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/vecreduce.ll
@@ -1,5 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify -use-constant-int-for-fixed-length-splat -S | FileCheck %s
declare i32 @llvm.vector.reduce.add.v1i32(<1 x i32> %a)
declare i32 @llvm.vector.reduce.add.v8i32(<8 x i32> %a)
|
nikic
reviewed
Sep 17, 2025
|
|
||
| // TODO: Handle undef. | ||
| if (!isa<ConstantVector>(Op) && !isa<ConstantDataVector>(Op)) | ||
| if (!isa<ConstantVector, ConstantDataVector, ConstantInt>(Op)) |
There was a problem hiding this comment.
Can we skip these checks entirely?
There was a problem hiding this comment.
Looking at getAggregateElement there is a "bad" path for scalable vectors, but swapping the following dyn_cast with dyn_cast_or_null should account for that, so yes.
nikic
reviewed
Sep 17, 2025
nikic
reviewed
Sep 21, 2025
| // Check each element individually | ||
| if (!isa<ConstantVector, ConstantDataVector, ConstantAggregateZero, | ||
| ConstantInt>(Op)) | ||
| break; |
There was a problem hiding this comment.
Also drop this check, and replace it with an early exit if getAggregateElement fails?
There was a problem hiding this comment.
I moved the check here because it's not just a case of adding an early exit. The code will then have to track "known true" values, so I figured it was better to keep the code simple?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Simplifcation of vector.reduce intrinsics are prevented by an early bailout for ConstantInt base operands. This PR removes the bailout and updates the tests to show matching output when -use-constant-int-for-*-splat is used.
No new simplications are added, which is why I only add scalable vector tests for things like bswap and not the vector reductions.