[X86] Support matching VPTERNLOG when the root node is X86ISD::ANDNP.

diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index 5842489..cf0b700 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -3991,6 +3991,12 @@ bool X86DAGToDAGISel::tryVPTERNLOG(SDNode *N) {
 
   switch (N->getOpcode()) {
   default: llvm_unreachable("Unexpected opcode!");
+  case X86ISD::ANDNP:
+    if (A == N0)
+      Imm &= ~TernlogMagicA;
+    else
+      Imm = ~(Imm) & TernlogMagicA;
+    break;
   case ISD::AND: Imm &= TernlogMagicA; break;
   case ISD::OR:  Imm |= TernlogMagicA; break;
   case ISD::XOR: Imm ^= TernlogMagicA; break;
@@ -4592,6 +4598,11 @@ void X86DAGToDAGISel::Select(SDNode *Node) {
       return;
     break;
 
+  case X86ISD::ANDNP:
+    if (tryVPTERNLOG(Node))
+      return;
+    break;
+
   case ISD::AND:
     if (NVT.isVector() && NVT.getVectorElementType() == MVT::i1) {
       // Try to form a masked VPTESTM. Operands can be in either order.

diff --git a/llvm/test/CodeGen/X86/avx512vl-logic.ll b/llvm/test/CodeGen/X86/avx512vl-logic.ll
index 9bd0b05..48ab5bf 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512vl-logic.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-logic.ll
@@ -1301,8 +1301,7 @@ define <4 x i64> @ternlog_masky_xor_and_mask_ymm(<4 x i64> %x, <4 x i64> %y, <4
 define <4 x i32> @ternlog_andn_or(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z) {
 ; CHECK-LABEL: ternlog_andn_or:
 ; CHECK:       ## %bb.0:
-; CHECK-NEXT:    vorps %xmm2, %xmm1, %xmm1
-; CHECK-NEXT:    vandnps %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vpternlogd $14, %xmm2, %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %a = xor <4 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1>
   %b = or <4 x i32> %y, %z
@@ -1313,8 +1312,7 @@ define <4 x i32> @ternlog_andn_or(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z) {
 define <4 x i32> @ternlog_andn_or_2(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z) {
 ; CHECK-LABEL: ternlog_andn_or_2:
 ; CHECK:       ## %bb.0:
-; CHECK-NEXT:    vorps %xmm2, %xmm1, %xmm1
-; CHECK-NEXT:    vandnps %xmm0, %xmm1, %xmm0
+; CHECK-NEXT:    vpternlogd $16, %xmm2, %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %a = or <4 x i32> %y, %z
   %b = xor <4 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1>