[GlobalISel] Add known bits and sign-bits handling for G_SPLAT_VECTOR#140204
Merged
Conversation
Member
|
@llvm/pr-subscribers-llvm-globalisel Author: David Green (davemgreen) ChangesThis forwards the KnownBits / SignBits from the scalar operand through to the vector elements. The value is implicitly truncated if it is larger than the vector element size. The change in GISelValueTracking::computeNumSignBits allows scalable sign-bits using a single element demanded elts, and brings it in line with what is done in ValueTracking and SDAG (and GISelValueTracking::getKnownBits). That was my main reason for adding this opcode, to prevent scalable vector asserts. Full diff: https://github.com/llvm/llvm-project/pull/140204.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
index 589936b6c260f..fd42b85251398 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
@@ -222,6 +222,14 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
}
break;
}
+ case TargetOpcode::G_SPLAT_VECTOR: {
+ computeKnownBitsImpl(MI.getOperand(1).getReg(), Known, APInt(1, 1),
+ Depth + 1);
+ // Implicitly truncate the bits to match the official semantics of
+ // G_SPLAT_VECTOR.
+ Known = Known.trunc(BitWidth);
+ break;
+ }
case TargetOpcode::COPY:
case TargetOpcode::G_PHI:
case TargetOpcode::PHI: {
@@ -854,6 +862,15 @@ unsigned GISelValueTracking::computeNumSignBits(Register R,
}
break;
}
+ case TargetOpcode::G_SPLAT_VECTOR: {
+ // Check if the sign bits of source go down as far as the truncated value.
+ Register Src = MI.getOperand(1).getReg();
+ unsigned NumSrcSignBits = computeNumSignBits(Src, APInt(1, 1), Depth + 1);
+ unsigned NumSrcBits = MRI.getType(Src).getSizeInBits();
+ if (NumSrcSignBits > (NumSrcBits - TyBits))
+ return NumSrcSignBits - (NumSrcBits - TyBits);
+ break;
+ }
case TargetOpcode::G_INTRINSIC:
case TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS:
case TargetOpcode::G_INTRINSIC_CONVERGENT:
@@ -889,7 +906,7 @@ unsigned GISelValueTracking::computeNumSignBits(Register R,
unsigned GISelValueTracking::computeNumSignBits(Register R, unsigned Depth) {
LLT Ty = MRI.getType(R);
APInt DemandedElts =
- Ty.isVector() ? APInt::getAllOnes(Ty.getNumElements()) : APInt(1, 1);
+ Ty.isFixedVector() ? APInt::getAllOnes(Ty.getNumElements()) : APInt(1, 1);
return computeNumSignBits(R, DemandedElts, Depth);
}
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-sve-splat.mir b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-sve-splat.mir
new file mode 100644
index 0000000000000..27c11ee1641e9
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-sve-splat.mir
@@ -0,0 +1,35 @@
+# NOTE: Assertions have been autogenerated by utils/update_givaluetracking_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple aarch64 -mattr=+sve -passes="print<gisel-value-tracking>" %s -o - 2>&1 | FileCheck %s
+
+---
+name: Scalable
+body: |
+ bb.1:
+ ; CHECK-LABEL: name: @Scalable
+ ; CHECK-NEXT: %0:_ KnownBits:0000000000001010 SignBits:12
+ ; CHECK-NEXT: %1:_ KnownBits:0000000000001010 SignBits:12
+ %0:_(s16) = G_CONSTANT i16 10
+ %1:_(<vscale x 8 x s16>) = G_SPLAT_VECTOR %0(s16)
+...
+---
+name: Scalable_trunc
+body: |
+ bb.1:
+ ; CHECK-LABEL: name: @Scalable_trunc
+ ; CHECK-NEXT: %0:_ KnownBits:00000000000000000000000000001010 SignBits:28
+ ; CHECK-NEXT: %1:_ KnownBits:0000000000001010 SignBits:12
+ %0:_(s32) = G_CONSTANT i32 10
+ %1:_(<vscale x 8 x s16>) = G_SPLAT_VECTOR %0(s32)
+...
+---
+name: Scalable_signbits
+body: |
+ bb.1:
+ ; CHECK-LABEL: name: @Scalable_signbits
+ ; CHECK-NEXT: %0:_ KnownBits:???????????????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:???????????????????????????????? SignBits:17
+ ; CHECK-NEXT: %2:_ KnownBits:???????????????????????????????? SignBits:17
+ %0:_(s16) = COPY $h0
+ %1:_(s32) = G_SEXT %0(s16)
+ %2:_(<vscale x 4 x s32>) = G_SPLAT_VECTOR %1(s32)
+...
|
Member
|
@llvm/pr-subscribers-backend-aarch64 Author: David Green (davemgreen) ChangesThis forwards the KnownBits / SignBits from the scalar operand through to the vector elements. The value is implicitly truncated if it is larger than the vector element size. The change in GISelValueTracking::computeNumSignBits allows scalable sign-bits using a single element demanded elts, and brings it in line with what is done in ValueTracking and SDAG (and GISelValueTracking::getKnownBits). That was my main reason for adding this opcode, to prevent scalable vector asserts. Full diff: https://github.com/llvm/llvm-project/pull/140204.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
index 589936b6c260f..fd42b85251398 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
@@ -222,6 +222,14 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
}
break;
}
+ case TargetOpcode::G_SPLAT_VECTOR: {
+ computeKnownBitsImpl(MI.getOperand(1).getReg(), Known, APInt(1, 1),
+ Depth + 1);
+ // Implicitly truncate the bits to match the official semantics of
+ // G_SPLAT_VECTOR.
+ Known = Known.trunc(BitWidth);
+ break;
+ }
case TargetOpcode::COPY:
case TargetOpcode::G_PHI:
case TargetOpcode::PHI: {
@@ -854,6 +862,15 @@ unsigned GISelValueTracking::computeNumSignBits(Register R,
}
break;
}
+ case TargetOpcode::G_SPLAT_VECTOR: {
+ // Check if the sign bits of source go down as far as the truncated value.
+ Register Src = MI.getOperand(1).getReg();
+ unsigned NumSrcSignBits = computeNumSignBits(Src, APInt(1, 1), Depth + 1);
+ unsigned NumSrcBits = MRI.getType(Src).getSizeInBits();
+ if (NumSrcSignBits > (NumSrcBits - TyBits))
+ return NumSrcSignBits - (NumSrcBits - TyBits);
+ break;
+ }
case TargetOpcode::G_INTRINSIC:
case TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS:
case TargetOpcode::G_INTRINSIC_CONVERGENT:
@@ -889,7 +906,7 @@ unsigned GISelValueTracking::computeNumSignBits(Register R,
unsigned GISelValueTracking::computeNumSignBits(Register R, unsigned Depth) {
LLT Ty = MRI.getType(R);
APInt DemandedElts =
- Ty.isVector() ? APInt::getAllOnes(Ty.getNumElements()) : APInt(1, 1);
+ Ty.isFixedVector() ? APInt::getAllOnes(Ty.getNumElements()) : APInt(1, 1);
return computeNumSignBits(R, DemandedElts, Depth);
}
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-sve-splat.mir b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-sve-splat.mir
new file mode 100644
index 0000000000000..27c11ee1641e9
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-sve-splat.mir
@@ -0,0 +1,35 @@
+# NOTE: Assertions have been autogenerated by utils/update_givaluetracking_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple aarch64 -mattr=+sve -passes="print<gisel-value-tracking>" %s -o - 2>&1 | FileCheck %s
+
+---
+name: Scalable
+body: |
+ bb.1:
+ ; CHECK-LABEL: name: @Scalable
+ ; CHECK-NEXT: %0:_ KnownBits:0000000000001010 SignBits:12
+ ; CHECK-NEXT: %1:_ KnownBits:0000000000001010 SignBits:12
+ %0:_(s16) = G_CONSTANT i16 10
+ %1:_(<vscale x 8 x s16>) = G_SPLAT_VECTOR %0(s16)
+...
+---
+name: Scalable_trunc
+body: |
+ bb.1:
+ ; CHECK-LABEL: name: @Scalable_trunc
+ ; CHECK-NEXT: %0:_ KnownBits:00000000000000000000000000001010 SignBits:28
+ ; CHECK-NEXT: %1:_ KnownBits:0000000000001010 SignBits:12
+ %0:_(s32) = G_CONSTANT i32 10
+ %1:_(<vscale x 8 x s16>) = G_SPLAT_VECTOR %0(s32)
+...
+---
+name: Scalable_signbits
+body: |
+ bb.1:
+ ; CHECK-LABEL: name: @Scalable_signbits
+ ; CHECK-NEXT: %0:_ KnownBits:???????????????? SignBits:1
+ ; CHECK-NEXT: %1:_ KnownBits:???????????????????????????????? SignBits:17
+ ; CHECK-NEXT: %2:_ KnownBits:???????????????????????????????? SignBits:17
+ %0:_(s16) = COPY $h0
+ %1:_(s32) = G_SEXT %0(s16)
+ %2:_(<vscale x 4 x s32>) = G_SPLAT_VECTOR %1(s32)
+...
|
arsenm
approved these changes
May 16, 2025
This forwards the KnownBits / SignBits from the scalar operand through to the vector elements. The value is implicitly truncated if it is larger than the vector element size. The change in GISelValueTracking::computeNumSignBits allows scalable sign-bits using a single element demanded elts, and brings it in line with what is done in ValueTracking and SDAG (and GISelValueTracking::getKnownBits). That was my main reason for adding this opcode, to prevent scalable vector asserts.
d2af7d1 to
0a22c65
Compare
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
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.
This forwards the KnownBits / SignBits from the scalar operand through to the vector elements. The value is implicitly truncated if it is larger than the vector element size. The change in GISelValueTracking::computeNumSignBits allows scalable sign-bits using a single element demanded elts, and brings it in line with what is done in ValueTracking and SDAG (and GISelValueTracking::getKnownBits). That was my main reason for adding this opcode, to prevent scalable vector asserts.