[GlobalISel] Add computeNumSignBits for ASHR #139503
Open
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
The code is taken from SelectionDAG::computeKnownBits.
|
@llvm/pr-subscribers-llvm-globalisel @llvm/pr-subscribers-backend-aarch64 Author: David Green (davemgreen) ChangesFull diff: https://github.com/llvm/llvm-project/pull/139503.diff 4 Files Affected:
diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
index 21990be21bbf7..41e36e1e6640b 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
@@ -864,6 +864,16 @@ unsigned GISelValueTracking::computeNumSignBits(Register R,
return TyBits - 1; // Every always-zero bit is a sign bit.
break;
}
+ case TargetOpcode::G_ASHR: {
+ Register Src1 = MI.getOperand(1).getReg();
+ Register Src2 = MI.getOperand(2).getReg();
+ LLT SrcTy = MRI.getType(Src1);
+ FirstAnswer = computeNumSignBits(Src1, DemandedElts, Depth + 1);
+ if (auto C = getIConstantSplatVal(Src2, MRI))
+ FirstAnswer = std::max<uint64_t>(FirstAnswer + C->getZExtValue(),
+ SrcTy.getScalarSizeInBits());
+ break;
+ }
case TargetOpcode::G_INTRINSIC:
case TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS:
case TargetOpcode::G_INTRINSIC_CONVERGENT:
diff --git a/llvm/test/CodeGen/AArch64/aarch64-smull.ll b/llvm/test/CodeGen/AArch64/aarch64-smull.ll
index 951001c84aed0..591bc65bf3226 100644
--- a/llvm/test/CodeGen/AArch64/aarch64-smull.ll
+++ b/llvm/test/CodeGen/AArch64/aarch64-smull.ll
@@ -2265,33 +2265,12 @@ define <2 x i64> @lsr_const(<2 x i64> %a, <2 x i64> %b) {
}
define <2 x i64> @asr(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-NEON-LABEL: asr:
-; CHECK-NEON: // %bb.0:
-; CHECK-NEON-NEXT: shrn v0.2s, v0.2d, #32
-; CHECK-NEON-NEXT: shrn v1.2s, v1.2d, #32
-; CHECK-NEON-NEXT: smull v0.2d, v0.2s, v1.2s
-; CHECK-NEON-NEXT: ret
-;
-; CHECK-SVE-LABEL: asr:
-; CHECK-SVE: // %bb.0:
-; CHECK-SVE-NEXT: shrn v0.2s, v0.2d, #32
-; CHECK-SVE-NEXT: shrn v1.2s, v1.2d, #32
-; CHECK-SVE-NEXT: smull v0.2d, v0.2s, v1.2s
-; CHECK-SVE-NEXT: ret
-;
-; CHECK-GI-LABEL: asr:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: sshr v0.2d, v0.2d, #32
-; CHECK-GI-NEXT: sshr v1.2d, v1.2d, #32
-; CHECK-GI-NEXT: fmov x8, d0
-; CHECK-GI-NEXT: fmov x9, d1
-; CHECK-GI-NEXT: mov x10, v0.d[1]
-; CHECK-GI-NEXT: mov x11, v1.d[1]
-; CHECK-GI-NEXT: mul x8, x8, x9
-; CHECK-GI-NEXT: mul x9, x10, x11
-; CHECK-GI-NEXT: mov v0.d[0], x8
-; CHECK-GI-NEXT: mov v0.d[1], x9
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: asr:
+; CHECK: // %bb.0:
+; CHECK-NEXT: shrn v0.2s, v0.2d, #32
+; CHECK-NEXT: shrn v1.2s, v1.2d, #32
+; CHECK-NEXT: smull v0.2d, v0.2s, v1.2s
+; CHECK-NEXT: ret
%x = ashr <2 x i64> %a, <i64 32, i64 32>
%y = ashr <2 x i64> %b, <i64 32, i64 32>
%z = mul nsw <2 x i64> %x, %y
@@ -2299,34 +2278,12 @@ define <2 x i64> @asr(<2 x i64> %a, <2 x i64> %b) {
}
define <2 x i64> @asr_const(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-NEON-LABEL: asr_const:
-; CHECK-NEON: // %bb.0:
-; CHECK-NEON-NEXT: movi v1.2s, #31
-; CHECK-NEON-NEXT: shrn v0.2s, v0.2d, #32
-; CHECK-NEON-NEXT: smull v0.2d, v0.2s, v1.2s
-; CHECK-NEON-NEXT: ret
-;
-; CHECK-SVE-LABEL: asr_const:
-; CHECK-SVE: // %bb.0:
-; CHECK-SVE-NEXT: movi v1.2s, #31
-; CHECK-SVE-NEXT: shrn v0.2s, v0.2d, #32
-; CHECK-SVE-NEXT: smull v0.2d, v0.2s, v1.2s
-; CHECK-SVE-NEXT: ret
-;
-; CHECK-GI-LABEL: asr_const:
-; CHECK-GI: // %bb.0:
-; CHECK-GI-NEXT: adrp x8, .LCPI81_0
-; CHECK-GI-NEXT: sshr v0.2d, v0.2d, #32
-; CHECK-GI-NEXT: ldr q1, [x8, :lo12:.LCPI81_0]
-; CHECK-GI-NEXT: fmov x8, d0
-; CHECK-GI-NEXT: fmov x9, d1
-; CHECK-GI-NEXT: mov x10, v0.d[1]
-; CHECK-GI-NEXT: mov x11, v1.d[1]
-; CHECK-GI-NEXT: mul x8, x8, x9
-; CHECK-GI-NEXT: mul x9, x10, x11
-; CHECK-GI-NEXT: mov v0.d[0], x8
-; CHECK-GI-NEXT: mov v0.d[1], x9
-; CHECK-GI-NEXT: ret
+; CHECK-LABEL: asr_const:
+; CHECK: // %bb.0:
+; CHECK-NEXT: movi v1.2s, #31
+; CHECK-NEXT: shrn v0.2s, v0.2d, #32
+; CHECK-NEXT: smull v0.2d, v0.2s, v1.2s
+; CHECK-NEXT: ret
%x = ashr <2 x i64> %a, <i64 32, i64 32>
%z = mul nsw <2 x i64> %x, <i64 31, i64 31>
ret <2 x i64> %z
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-abs-rv64.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-abs-rv64.mir
index 78a2227b84a3a..a7c1c6355bff6 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-abs-rv64.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-abs-rv64.mir
@@ -88,8 +88,7 @@ body: |
; RV64I-NEXT: [[ADD:%[0-9]+]]:_(s64) = G_ADD [[ASSERT_SEXT]], [[ASHR]]
; RV64I-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s64) = G_SEXT_INREG [[ADD]], 32
; RV64I-NEXT: [[XOR:%[0-9]+]]:_(s64) = G_XOR [[SEXT_INREG]], [[ASHR]]
- ; RV64I-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s64) = G_SEXT_INREG [[XOR]], 32
- ; RV64I-NEXT: $x10 = COPY [[SEXT_INREG1]](s64)
+ ; RV64I-NEXT: $x10 = COPY [[XOR]](s64)
; RV64I-NEXT: PseudoRET implicit $x10
;
; RV64ZBB-LABEL: name: abs_i32
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbb.ll b/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbb.ll
index 8549a7c526e45..747dda692529e 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbb.ll
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/rv64zbb.ll
@@ -1053,9 +1053,8 @@ define signext i32 @abs_i32_sext(i32 signext %x) {
; RV64I-LABEL: abs_i32_sext:
; RV64I: # %bb.0:
; RV64I-NEXT: srai a1, a0, 31
-; RV64I-NEXT: add a0, a0, a1
+; RV64I-NEXT: addw a0, a0, a1
; RV64I-NEXT: xor a0, a0, a1
-; RV64I-NEXT: sext.w a0, a0
; RV64I-NEXT: ret
;
; RV64ZBB-LABEL: abs_i32_sext:
|
jayfoad
requested changes
May 12, 2025
| LLT SrcTy = MRI.getType(Src1); | ||
| FirstAnswer = computeNumSignBits(Src1, DemandedElts, Depth + 1); | ||
| if (auto C = getIConstantSplatVal(Src2, MRI)) | ||
| FirstAnswer = std::max<uint64_t>(FirstAnswer + C->getZExtValue(), |
There was a problem hiding this comment.
Suggested change
| FirstAnswer = std::max<uint64_t>(FirstAnswer + C->getZExtValue(), | |
| FirstAnswer = std::min<uint64_t>(FirstAnswer + C->getZExtValue(), |
arsenm
reviewed
May 12, 2025
| Register Src2 = MI.getOperand(2).getReg(); | ||
| LLT SrcTy = MRI.getType(Src1); | ||
| FirstAnswer = computeNumSignBits(Src1, DemandedElts, Depth + 1); | ||
| if (auto C = getIConstantSplatVal(Src2, MRI)) |
There was a problem hiding this comment.
At some point we should make these properly elementwise and not only handle splats
davemgreen
changed the base branch from
users/davemgreen/gh-gi-shuffleknownbits
to
main
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.
No description provided.