[llvm] [AArch64][InstCombine] Bail from combining SRAD on +/-1 divisor (PR #109274)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Matthew Devereau (MDevereau)

<details>
<summary>Changes</summary>

This fixes a crash when svdiv's third parameter is svdup_s64(1)

---
Full diff: https://github.com/llvm/llvm-project/pull/109274.diff


2 Files Affected:

- (modified) llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp (+3) 
- (modified) llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-sdiv.ll (+24) 


``````````diff
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index 649ba1ac8749f5..686503a794d7be 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -1972,7 +1972,10 @@ static std::optional<Instruction *> instCombineSVESDIV(InstCombiner &IC,
   ConstantInt *SplatConstantInt = dyn_cast_or_null<ConstantInt>(SplatValue);
   if (!SplatConstantInt)
     return std::nullopt;
+
   APInt Divisor = SplatConstantInt->getValue();
+  if (Divisor.abs().getZExtValue() == 1)
+    return std::nullopt;
 
   if (Divisor.isPowerOf2()) {
     Constant *DivisorLog2 = ConstantInt::get(Int32Ty, Divisor.logBase2());
diff --git a/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-sdiv.ll b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-sdiv.ll
index 68d871355e2a23..dc134ef109fe75 100644
--- a/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-sdiv.ll
+++ b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-sdiv.ll
@@ -68,6 +68,30 @@ define <vscale x 4 x i32> @sdiv_i32_not_zero(<vscale x 4 x i32> %a, <vscale x 4
   ret <vscale x 4 x i32> %out
 }
 
+; Don't instcombine to SRAD when the divisor is (+/-)1
+define <vscale x 2 x i64> @divide_by_1(<vscale x 16 x i1> %p, <vscale x 2 x i64> %a) #0 {
+; CHECK-LABEL: @divide_by_1(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[P:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.sdiv.nxv2i64(<vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[A:%.*]], <vscale x 2 x i64> shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 1, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer))
+; CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
+;
+  %1 = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.x.nxv2i64(i64 1)
+  %2 = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %p)
+  %3 = call <vscale x 2 x i64> @llvm.aarch64.sve.sdiv.nxv2i64(<vscale x 2 x i1> %2, <vscale x 2 x i64> %a, <vscale x 2 x i64> %1)
+  ret <vscale x 2 x i64> %3
+}
+
+define <vscale x 2 x i64> @divide_by_m1(<vscale x 16 x i1> %p, <vscale x 2 x i64> %a) #0 {
+; CHECK-LABEL: @divide_by_m1(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[P:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.sdiv.nxv2i64(<vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[A:%.*]], <vscale x 2 x i64> shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 -1, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer))
+; CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
+;
+  %1 = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.x.nxv2i64(i64 -1)
+  %2 = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %p)
+  %3 = call <vscale x 2 x i64> @llvm.aarch64.sve.sdiv.nxv2i64(<vscale x 2 x i1> %2, <vscale x 2 x i64> %a, <vscale x 2 x i64> %1)
+  ret <vscale x 2 x i64> %3
+}
 
 declare <vscale x 4 x i32> @llvm.aarch64.sve.sdiv.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 4 x i32>)
 declare <vscale x 2 x i64> @llvm.aarch64.sve.sdiv.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, <vscale x 2 x i64>)

``````````

</details>


https://github.com/llvm/llvm-project/pull/109274