[llvm] [X86] X86ISelDAGToDAG - don't let ADD/SUB(X,1) -> SUB/ADD(X,-1) constant fold (PR #168726)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

<details>
<summary>Changes</summary>

This late into lowering we don't have a good way to handle constant build_vector lowering

Fixes #<!-- -->168594

---
Full diff: https://github.com/llvm/llvm-project/pull/168726.diff


2 Files Affected:

- (modified) llvm/lib/Target/X86/X86ISelDAGToDAG.cpp (+2-1) 
- (added) llvm/test/CodeGen/X86/pr168594.ll (+22) 


``````````diff
diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index 6c16fcfb282e8..2b5a63aa66926 100644
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -1004,7 +1004,8 @@ void X86DAGToDAGISel::PreprocessISelDAG() {
     if ((N->getOpcode() == ISD::ADD || N->getOpcode() == ISD::SUB) &&
         N->getSimpleValueType(0).isVector() && !mayPreventLoadFold()) {
       APInt SplatVal;
-      if (X86::isConstantSplat(N->getOperand(1), SplatVal) &&
+      if (!ISD::isBuildVectorOfConstantSDNodes(N->getOperand(0).getNode()) &&
+          X86::isConstantSplat(N->getOperand(1), SplatVal) &&
           SplatVal.isOne()) {
         SDLoc DL(N);
 
diff --git a/llvm/test/CodeGen/X86/pr168594.ll b/llvm/test/CodeGen/X86/pr168594.ll
new file mode 100644
index 0000000000000..76bb13223d49c
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr168594.ll
@@ -0,0 +1,22 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64    | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX
+; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX
+
+define <8 x i16> @PR168594() {
+; SSE-LABEL: PR168594:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm0, %xmm0
+; SSE-NEXT:    psubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: PR168594:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; AVX-NEXT:    vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %call = call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> splat (i16 1), <8 x i16> zeroinitializer)
+  %sub = sub <8 x i16> zeroinitializer, %call
+  ret <8 x i16> %sub
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/168726