[ValueTracking] ComputeNumSignBitsImpl - add basic handling of BITCAST nodes #127218
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
vortex73
changed the title
llvmbot
added
llvm:instcombine
|
@llvm/pr-subscribers-llvm-transforms Author: Narayan (vortex73) ChangesWhen a wider scalar/vector type containing all sign bits is bitcast to a narrower vector type, we can deduce that the resulting narrow elements will also be all sign bits. This matches existing behavior in SelectionDAG and helps optimize cases involving SSE intrinsics where sign-extended values are bitcast between different vector types. The current implementation fails to recognize that an arithmetic right shift is redundant when applied to elements that are already known to be all sign bits. This PR improves ComputeNumSignBitsImpl to track this information through bitcasts, enabling the optimization of such cases. Closes #87624 Full diff: https://github.com/llvm/llvm-project/pull/127218.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 8a9ad55366ee7..b0b963c14f8a5 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -3922,6 +3922,35 @@ static unsigned ComputeNumSignBitsImpl(const Value *V,
if (auto *U = dyn_cast<Operator>(V)) {
switch (Operator::getOpcode(V)) {
default: break;
+ case Instruction::BitCast: {
+ Value *Src = U->getOperand(0);
+ Type *SrcTy = Src->getType();
+ Type *SrcScalarTy = SrcTy->getScalarType();
+
+ if (!SrcScalarTy->isIntegerTy() && !SrcScalarTy->isFloatingPointTy())
+ break;
+
+ unsigned SrcBits = SrcTy->getScalarSizeInBits();
+
+ if ((SrcBits % TyBits) != 0)
+ break;
+
+ if (auto *DstVTy = dyn_cast<FixedVectorType>(Ty)) {
+ unsigned Scale = SrcBits / TyBits;
+
+ APInt SrcDemandedElts =
+ APInt::getSplat(DstVTy->getNumElements() / Scale, APInt(1, 1));
+
+ Tmp = ComputeNumSignBits(Src, SrcDemandedElts, Depth + 1, Q);
+ if (Tmp == SrcBits)
+ return TyBits;
+ } else {
+ Tmp = ComputeNumSignBits(Src, APInt(1, 1), Depth + 1, Q);
+ if (Tmp == SrcBits)
+ return TyBits;
+ }
+ break;
+ }
case Instruction::SExt:
Tmp = TyBits - U->getOperand(0)->getType()->getScalarSizeInBits();
return ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q) +
diff --git a/llvm/test/Transforms/InstCombine/compute-sign-bits-bitcast.ll b/llvm/test/Transforms/InstCombine/compute-sign-bits-bitcast.ll
new file mode 100644
index 0000000000000..35ba53687c646
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/compute-sign-bits-bitcast.ll
@@ -0,0 +1,34 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=instcombine -S < %s | FileCheck %s
+
+define i32 @test_compute_sign_bits() {
+; CHECK-LABEL: define i32 @test_compute_sign_bits() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: ret i32 -1
+;
+entry:
+ %a = add i8 -1, 0
+ %b = bitcast i8 %a to <4 x i2>
+ %c = ashr <4 x i2> %b, <i2 1, i2 1, i2 1, i2 1>
+ %d = bitcast <4 x i2> %c to i8
+ %e = sext i8 %d to i32
+ ret i32 %e
+}
+
+; Test with sign extension to ensure proper sign bit tracking
+define <4 x i2> @test_sext_bitcast(<1 x i8> %a0, <1 x i8> %a1) {
+; CHECK-LABEL: define <4 x i2> @test_sext_bitcast(
+; CHECK-SAME: <1 x i8> [[A0:%.*]], <1 x i8> [[A1:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <1 x i8> [[A0]], [[A1]]
+; CHECK-NEXT: [[EXT:%.*]] = sext <1 x i1> [[CMP]] to <1 x i8>
+; CHECK-NEXT: [[SUB:%.*]] = bitcast <1 x i8> [[EXT]] to <4 x i2>
+; CHECK-NEXT: ret <4 x i2> [[SUB]]
+;
+entry:
+ %cmp = icmp sgt <1 x i8> %a0, %a1
+ %ext = sext <1 x i1> %cmp to <1 x i8>
+ %sub = bitcast <1 x i8> %ext to <4 x i2>
+ %result = ashr <4 x i2> %sub, <i2 1, i2 1, i2 1, i2 1>
+ ret <4 x i2> %result
+}
|
|
@llvm/pr-subscribers-llvm-analysis Author: Narayan (vortex73) ChangesWhen a wider scalar/vector type containing all sign bits is bitcast to a narrower vector type, we can deduce that the resulting narrow elements will also be all sign bits. This matches existing behavior in SelectionDAG and helps optimize cases involving SSE intrinsics where sign-extended values are bitcast between different vector types. The current implementation fails to recognize that an arithmetic right shift is redundant when applied to elements that are already known to be all sign bits. This PR improves ComputeNumSignBitsImpl to track this information through bitcasts, enabling the optimization of such cases. Closes #87624 Full diff: https://github.com/llvm/llvm-project/pull/127218.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 8a9ad55366ee7..b0b963c14f8a5 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -3922,6 +3922,35 @@ static unsigned ComputeNumSignBitsImpl(const Value *V,
if (auto *U = dyn_cast<Operator>(V)) {
switch (Operator::getOpcode(V)) {
default: break;
+ case Instruction::BitCast: {
+ Value *Src = U->getOperand(0);
+ Type *SrcTy = Src->getType();
+ Type *SrcScalarTy = SrcTy->getScalarType();
+
+ if (!SrcScalarTy->isIntegerTy() && !SrcScalarTy->isFloatingPointTy())
+ break;
+
+ unsigned SrcBits = SrcTy->getScalarSizeInBits();
+
+ if ((SrcBits % TyBits) != 0)
+ break;
+
+ if (auto *DstVTy = dyn_cast<FixedVectorType>(Ty)) {
+ unsigned Scale = SrcBits / TyBits;
+
+ APInt SrcDemandedElts =
+ APInt::getSplat(DstVTy->getNumElements() / Scale, APInt(1, 1));
+
+ Tmp = ComputeNumSignBits(Src, SrcDemandedElts, Depth + 1, Q);
+ if (Tmp == SrcBits)
+ return TyBits;
+ } else {
+ Tmp = ComputeNumSignBits(Src, APInt(1, 1), Depth + 1, Q);
+ if (Tmp == SrcBits)
+ return TyBits;
+ }
+ break;
+ }
case Instruction::SExt:
Tmp = TyBits - U->getOperand(0)->getType()->getScalarSizeInBits();
return ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q) +
diff --git a/llvm/test/Transforms/InstCombine/compute-sign-bits-bitcast.ll b/llvm/test/Transforms/InstCombine/compute-sign-bits-bitcast.ll
new file mode 100644
index 0000000000000..35ba53687c646
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/compute-sign-bits-bitcast.ll
@@ -0,0 +1,34 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=instcombine -S < %s | FileCheck %s
+
+define i32 @test_compute_sign_bits() {
+; CHECK-LABEL: define i32 @test_compute_sign_bits() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: ret i32 -1
+;
+entry:
+ %a = add i8 -1, 0
+ %b = bitcast i8 %a to <4 x i2>
+ %c = ashr <4 x i2> %b, <i2 1, i2 1, i2 1, i2 1>
+ %d = bitcast <4 x i2> %c to i8
+ %e = sext i8 %d to i32
+ ret i32 %e
+}
+
+; Test with sign extension to ensure proper sign bit tracking
+define <4 x i2> @test_sext_bitcast(<1 x i8> %a0, <1 x i8> %a1) {
+; CHECK-LABEL: define <4 x i2> @test_sext_bitcast(
+; CHECK-SAME: <1 x i8> [[A0:%.*]], <1 x i8> [[A1:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <1 x i8> [[A0]], [[A1]]
+; CHECK-NEXT: [[EXT:%.*]] = sext <1 x i1> [[CMP]] to <1 x i8>
+; CHECK-NEXT: [[SUB:%.*]] = bitcast <1 x i8> [[EXT]] to <4 x i2>
+; CHECK-NEXT: ret <4 x i2> [[SUB]]
+;
+entry:
+ %cmp = icmp sgt <1 x i8> %a0, %a1
+ %ext = sext <1 x i1> %cmp to <1 x i8>
+ %sub = bitcast <1 x i8> %ext to <4 x i2>
+ %result = ashr <4 x i2> %sub, <i2 1, i2 1, i2 1, i2 1>
+ ret <4 x i2> %result
+}
|
dtcxzyw
reviewed
Feb 24, 2025
RKSimon
reviewed
Feb 24, 2025
|
@RKSimon Please have a look now. Since the original issue was only concerned with * → vector casts, I've not pushed my implementation for casts to scalars. If is a valid addition I could spend some time on that. |
RKSimon
reviewed
Mar 4, 2025
RKSimon
reviewed
Mar 5, 2025
RKSimon
reviewed
Mar 5, 2025
RKSimon
reviewed
Mar 5, 2025
jph-13
pushed a commit
to jph-13/llvm-project
that referenced
this pull request
Mar 21, 2025
…T nodes (llvm#127218) When a wider scalar/vector type containing all sign bits is bitcast to a narrower vector type, we can deduce that the resulting narrow elements will also be all sign bits. This matches existing behavior in SelectionDAG and helps optimize cases involving SSE intrinsics where sign-extended values are bitcast between different vector types. The current implementation fails to recognize that an arithmetic right shift is redundant when applied to elements that are already known to be all sign bits. This PR improves ComputeNumSignBitsImpl to track this information through bitcasts, enabling the optimization of such cases. ``` %ext = sext <1 x i1> %cmp to <1 x i8> %sub = bitcast <1 x i8> %ext to <4 x i2> %sra = ashr <4 x i2> %sub, <i2 1, i2 1, i2 1, i2 1> ; Can be simplified to just: %sub = bitcast <1 x i8> %ext to <4 x i2> ``` Closes llvm#87624
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.
When a wider scalar/vector type containing all sign bits is bitcast to a narrower vector type, we can deduce that the resulting narrow elements will also be all sign bits. This matches existing behavior in SelectionDAG and helps optimize cases involving SSE intrinsics where sign-extended values are bitcast between different vector types.
The current implementation fails to recognize that an arithmetic right shift is redundant when applied to elements that are already known to be all sign bits. This PR improves ComputeNumSignBitsImpl to track this information through bitcasts, enabling the optimization of such cases.
Closes #87624