[llvm] [X86] Fixed truncated masked stores (PR #179491)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-x86

Author: None (JaydeepChauhan14)

<details>
<summary>Changes</summary>

Fixes #<!-- -->179489

---
Full diff: https://github.com/llvm/llvm-project/pull/179491.diff


2 Files Affected:

- (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+3-3) 
- (added) llvm/test/CodeGen/X86/pr179489.ll (+25) 


``````````diff
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 1837c8bbedf0e..9c78d049dcbfe 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -54094,9 +54094,6 @@ static SDValue combineMaskedStore(SDNode *N, SelectionDAG &DAG,
   if (Mst->isCompressingStore())
     return SDValue();
 
-  if (SDValue ScalarStore = reduceMaskedStoreToScalarStore(Mst, DAG, Subtarget))
-    return ScalarStore;
-
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   SDLoc DL(N);
 
@@ -54134,6 +54131,9 @@ static SDValue combineMaskedStore(SDNode *N, SelectionDAG &DAG,
     return SDValue();
   }
 
+  if (SDValue ScalarStore = reduceMaskedStoreToScalarStore(Mst, DAG, Subtarget))
+    return ScalarStore;
+
   // Attempt to convert a (vXi1 bitcast(iX Mask)) mask before it might get split
   // by legalization.
   if (SDValue NewMask =
diff --git a/llvm/test/CodeGen/X86/pr179489.ll b/llvm/test/CodeGen/X86/pr179489.ll
new file mode 100644
index 0000000000000..f06d19d7760e2
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr179489.ll
@@ -0,0 +1,25 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx10.2 | FileCheck %s --check-prefix=CHECK
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: write)
+declare void @llvm.masked.store.v8i8.p1(<8 x i8>, ptr addrspace(1) captures(none), <8 x i1>) #0
+
+define void @foo(<8 x i16> %arg, ptr addrspace(1) %add.ptr, i64 %dim0) {
+; CHECK-LABEL: foo:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vbroadcastss {{.*#+}} xmm1 = [257,257,257,257,257,257,257,257]
+; CHECK-NEXT:    vmovsh {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; CHECK-NEXT:    vpsrlw $8, %xmm0, %xmm0
+; CHECK-NEXT:    movb $1, %al
+; CHECK-NEXT:    kmovd %eax, %k1
+; CHECK-NEXT:    vpmovwb %xmm0, (%rdi) {%k1}
+; CHECK-NEXT:    retq
+entry:
+  %i3 = shufflevector <8 x i16> %arg, <8 x i16> <i16 poison, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257>, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+  %i4 = lshr <8 x i16> %i3, splat (i16 8)
+  %i5 = trunc <8 x i16> %i4 to <8 x i8>
+  call void @llvm.masked.store.v8i8.p1(<8 x i8> %i5, ptr addrspace(1) %add.ptr, <8 x i1> <i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>)
+  ret void
+}
+
+attributes #0 = { nocallback nofree nosync nounwind willreturn memory(argmem: write) }

``````````

</details>


https://github.com/llvm/llvm-project/pull/179491