[DAGCombiner] Generalize (or (and X, c1), c2) -> (and (or X, c2), c1|c2) combine to work on non-splat vectors

The knownbits_mask_or_shuffle_uitofp change is interesting - shuffle combines manage to kick in, removing the AND constant mask load. For targets with fast-variable-shuffle this should reduce further to VPOR+VPSHUFB+VCVTDQ2PS.

llvm-svn: 321279

3 files changed, 20 insertions, 18 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index e8f76a2bb70..5a393581ee7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -4687,16 +4687,16 @@ SDValue DAGCombiner::visitOR(SDNode *N) {
 
   // Canonicalize (or (and X, c1), c2) -> (and (or X, c2), c1|c2)
   // iff (c1 & c2) != 0.
-  if (N1C && N0.getOpcode() == ISD::AND && N0.getNode()->hasOneUse()) {
-    if (ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(N0.getOperand(1))) {
-      if (C1->getAPIntValue().intersects(N1C->getAPIntValue())) {
-        if (SDValue COR =
-                DAG.FoldConstantArithmetic(ISD::OR, SDLoc(N1), VT, N1C, C1))
-          return DAG.getNode(
-              ISD::AND, SDLoc(N), VT,
-              DAG.getNode(ISD::OR, SDLoc(N0), VT, N0.getOperand(0), N1), COR);
-        return SDValue();
-      }
+  auto MatchIntersect = [](ConstantSDNode *LHS, ConstantSDNode *RHS) {
+    return LHS->getAPIntValue().intersects(RHS->getAPIntValue());
+  };
+  if (N0.getOpcode() == ISD::AND && N0.getNode()->hasOneUse() &&
+      matchBinaryPredicate(N0.getOperand(1), N1, MatchIntersect)) {
+    if (SDValue COR = DAG.FoldConstantArithmetic(
+            ISD::OR, SDLoc(N1), VT, N1.getNode(), N0.getOperand(1).getNode())) {
+      SDValue IOR = DAG.getNode(ISD::OR, SDLoc(N0), VT, N0.getOperand(0), N1);
+      AddToWorklist(IOR.getNode());
+      return DAG.getNode(ISD::AND, SDLoc(N), VT, COR, IOR);
     }
   }
 
diff --git a/llvm/test/CodeGen/X86/combine-or.ll b/llvm/test/CodeGen/X86/combine-or.ll
index 5ab2e3d2508..65184c4d278 100644
--- a/llvm/test/CodeGen/X86/combine-or.ll
+++ b/llvm/test/CodeGen/X86/combine-or.ll
@@ -435,8 +435,8 @@ define <4 x i32> @test2f(<4 x i32> %a, <4 x i32> %b) {
 define <2 x i64> @or_and_v2i64(<2 x i64> %a0) {
 ; CHECK-LABEL: or_and_v2i64:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    andps {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    orps {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    andps {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    retq
   %1 = and <2 x i64> %a0, <i64 7, i64 7>
   %2 = or <2 x i64> %1, <i64 3, i64 3>
@@ -446,8 +446,8 @@ define <2 x i64> @or_and_v2i64(<2 x i64> %a0) {
 define <4 x i32> @or_and_v4i32(<4 x i32> %a0) {
 ; CHECK-LABEL: or_and_v4i32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    andps {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    orps {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    andps {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    retq
   %1 = and <4 x i32> %a0, <i32 1, i32 3, i32 5, i32 7>
   %2 = or <4 x i32> %1, <i32 3, i32 2, i32 15, i32 2>
diff --git a/llvm/test/CodeGen/X86/known-bits-vector.ll b/llvm/test/CodeGen/X86/known-bits-vector.ll
index 283d1f93dfb..46a888f3b9b 100644
--- a/llvm/test/CodeGen/X86/known-bits-vector.ll
+++ b/llvm/test/CodeGen/X86/known-bits-vector.ll
@@ -160,17 +160,19 @@ define <4 x float> @knownbits_mask_shuffle_uitofp(<4 x i32> %a0) nounwind {
 define <4 x float> @knownbits_mask_or_shuffle_uitofp(<4 x i32> %a0) nounwind {
 ; X32-LABEL: knownbits_mask_or_shuffle_uitofp:
 ; X32:       # %bb.0:
-; X32-NEXT:    vandps {{\.LCPI.*}}, %xmm0, %xmm0
-; X32-NEXT:    vorps {{\.LCPI.*}}, %xmm0, %xmm0
-; X32-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; X32-NEXT:    vpor {{\.LCPI.*}}, %xmm0, %xmm0
+; X32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X32-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,6,6]
+; X32-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; X32-NEXT:    vcvtdq2ps %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: knownbits_mask_or_shuffle_uitofp:
 ; X64:       # %bb.0:
-; X64-NEXT:    vandps {{.*}}(%rip), %xmm0, %xmm0
-; X64-NEXT:    vorps {{.*}}(%rip), %xmm0, %xmm0
-; X64-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; X64-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
+; X64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X64-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,6,6]
+; X64-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; X64-NEXT:    vcvtdq2ps %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = and <4 x i32> %a0, <i32 -1, i32 -1, i32 255, i32 4085>