[X86] LowerVectorAllEqual - handle i64 comparisons on 32-bit targets

llvm · Apr 1, 2023 · 88e5df0 · 88e5df0
1 parent 7de8cd6
commit 88e5df0
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 7 deletions.
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -24366,8 +24366,21 @@ static SDValue LowerVectorAllEqual(const SDLoc &DL, SDValue LHS, SDValue RHS,
   // For sub-128-bit vector, cast to (legal) integer and compare with zero.
   if (VT.getSizeInBits() < 128) {
     EVT IntVT = EVT::getIntegerVT(*DAG.getContext(), VT.getSizeInBits());
-    if (!DAG.getTargetLoweringInfo().isTypeLegal(IntVT))
-      return SDValue();
+    if (!DAG.getTargetLoweringInfo().isTypeLegal(IntVT)) {
+      if (IntVT != MVT::i64)
+        return SDValue();
+      auto SplitLHS = DAG.SplitScalar(DAG.getBitcast(IntVT, MaskBits(LHS)), DL,
+                                      MVT::i32, MVT::i32);
+      auto SplitRHS = DAG.SplitScalar(DAG.getBitcast(IntVT, MaskBits(RHS)), DL,
+                                      MVT::i32, MVT::i32);
+      SDValue Lo =
+          DAG.getNode(ISD::XOR, DL, MVT::i32, SplitLHS.first, SplitRHS.first);
+      SDValue Hi =
+          DAG.getNode(ISD::XOR, DL, MVT::i32, SplitLHS.second, SplitRHS.second);
+      return DAG.getNode(X86ISD::CMP, DL, MVT::i32,
+                         DAG.getNode(ISD::OR, DL, MVT::i32, Lo, Hi),
+                         DAG.getConstant(0, DL, MVT::i32));
+    }
     return DAG.getNode(X86ISD::CMP, DL, MVT::i32,
                        DAG.getBitcast(IntVT, MaskBits(LHS)),
                        DAG.getBitcast(IntVT, MaskBits(RHS)));

diff --git a/llvm/test/CodeGen/X86/pr53419.ll b/llvm/test/CodeGen/X86/pr53419.ll
@@ -149,11 +149,11 @@ define i1 @vector_version_v8i8(ptr align 1 %arg, ptr align 1 %arg1) {
 ; X86:       # %bb.0: # %bb
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; X86-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; X86-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; X86-NEXT:    vpmovmskb %xmm0, %eax
-; X86-NEXT:    xorb $-1, %al
+; X86-NEXT:    movl (%ecx), %edx
+; X86-NEXT:    movl 4(%ecx), %ecx
+; X86-NEXT:    xorl 4(%eax), %ecx
+; X86-NEXT:    xorl (%eax), %edx
+; X86-NEXT:    orl %ecx, %edx
 ; X86-NEXT:    sete %al
 ; X86-NEXT:    retl
 bb: