[DAGCombiner] allow undef elts in vector fma matching

llvm-svn: 344525

2 files changed, 70 insertions, 39 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 7ec5fac390b..f2779a3475e 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -10794,17 +10794,18 @@ SDValue DAGCombiner::visitFMULForFMADistributiveCombine(SDNode *N) {
   unsigned PreferredFusedOpcode = HasFMAD ? ISD::FMAD : ISD::FMA;
   bool Aggressive = TLI.enableAggressiveFMAFusion(VT);
 
-  // fold (fmul (fadd x, +1.0), y) -> (fma x, y, y)
-  // fold (fmul (fadd x, -1.0), y) -> (fma x, y, (fneg y))
+  // fold (fmul (fadd x0, +1.0), y) -> (fma x0, y, y)
+  // fold (fmul (fadd x0, -1.0), y) -> (fma x0, y, (fneg y))
   auto FuseFADD = [&](SDValue X, SDValue Y, const SDNodeFlags Flags) {
     if (X.getOpcode() == ISD::FADD && (Aggressive || X->hasOneUse())) {
-      auto XC1 = isConstOrConstSplatFP(X.getOperand(1));
-      if (XC1 && XC1->isExactlyValue(+1.0))
-        return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y,
-                           Y, Flags);
-      if (XC1 && XC1->isExactlyValue(-1.0))
-        return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y,
-                           DAG.getNode(ISD::FNEG, SL, VT, Y), Flags);
+      if (auto *C = isConstOrConstSplatFP(X.getOperand(1), true)) {
+        if (C->isExactlyValue(+1.0))
+          return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y,
+                             Y, Flags);
+        if (C->isExactlyValue(-1.0))
+          return DAG.getNode(PreferredFusedOpcode, SL, VT, X.getOperand(0), Y,
+                             DAG.getNode(ISD::FNEG, SL, VT, Y), Flags);
+      }
     }
     return SDValue();
   };
diff --git a/llvm/test/CodeGen/X86/fma_patterns.ll b/llvm/test/CodeGen/X86/fma_patterns.ll
index d0d0dfed352..5395ae46d47 100644
--- a/llvm/test/CodeGen/X86/fma_patterns.ll
+++ b/llvm/test/CodeGen/X86/fma_patterns.ll
@@ -637,23 +637,38 @@ define <4 x float> @test_v4f32_mul_y_add_x_one(<4 x float> %x, <4 x float> %y) {
 }
 
 define <4 x float> @test_v4f32_mul_y_add_x_one_undefs(<4 x float> %x, <4 x float> %y) {
-; FMA-LABEL: test_v4f32_mul_y_add_x_one_undefs:
-; FMA:       # %bb.0:
-; FMA-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
-; FMA-NEXT:    vmulps %xmm0, %xmm1, %xmm0
-; FMA-NEXT:    retq
+; FMA-INFS-LABEL: test_v4f32_mul_y_add_x_one_undefs:
+; FMA-INFS:       # %bb.0:
+; FMA-INFS-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
+; FMA-INFS-NEXT:    vmulps %xmm0, %xmm1, %xmm0
+; FMA-INFS-NEXT:    retq
 ;
-; FMA4-LABEL: test_v4f32_mul_y_add_x_one_undefs:
-; FMA4:       # %bb.0:
-; FMA4-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
-; FMA4-NEXT:    vmulps %xmm0, %xmm1, %xmm0
-; FMA4-NEXT:    retq
+; FMA4-INFS-LABEL: test_v4f32_mul_y_add_x_one_undefs:
+; FMA4-INFS:       # %bb.0:
+; FMA4-INFS-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
+; FMA4-INFS-NEXT:    vmulps %xmm0, %xmm1, %xmm0
+; FMA4-INFS-NEXT:    retq
 ;
-; AVX512-LABEL: test_v4f32_mul_y_add_x_one_undefs:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    vaddps {{.*}}(%rip){1to4}, %xmm0, %xmm0
-; AVX512-NEXT:    vmulps %xmm0, %xmm1, %xmm0
-; AVX512-NEXT:    retq
+; AVX512-INFS-LABEL: test_v4f32_mul_y_add_x_one_undefs:
+; AVX512-INFS:       # %bb.0:
+; AVX512-INFS-NEXT:    vaddps {{.*}}(%rip){1to4}, %xmm0, %xmm0
+; AVX512-INFS-NEXT:    vmulps %xmm0, %xmm1, %xmm0
+; AVX512-INFS-NEXT:    retq
+;
+; FMA-NOINFS-LABEL: test_v4f32_mul_y_add_x_one_undefs:
+; FMA-NOINFS:       # %bb.0:
+; FMA-NOINFS-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm1
+; FMA-NOINFS-NEXT:    retq
+;
+; FMA4-NOINFS-LABEL: test_v4f32_mul_y_add_x_one_undefs:
+; FMA4-NOINFS:       # %bb.0:
+; FMA4-NOINFS-NEXT:    vfmaddps %xmm1, %xmm1, %xmm0, %xmm0
+; FMA4-NOINFS-NEXT:    retq
+;
+; AVX512-NOINFS-LABEL: test_v4f32_mul_y_add_x_one_undefs:
+; AVX512-NOINFS:       # %bb.0:
+; AVX512-NOINFS-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm1
+; AVX512-NOINFS-NEXT:    retq
   %a = fadd <4 x float> %x, <float 1.0, float undef, float 1.0, float undef>
   %m = fmul <4 x float> %y, %a
   ret <4 x float> %m
@@ -736,23 +751,38 @@ define <4 x float> @test_v4f32_mul_y_add_x_negone(<4 x float> %x, <4 x float> %y
 }
 
 define <4 x float> @test_v4f32_mul_y_add_x_negone_undefs(<4 x float> %x, <4 x float> %y) {
-; FMA-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
-; FMA:       # %bb.0:
-; FMA-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
-; FMA-NEXT:    vmulps %xmm0, %xmm1, %xmm0
-; FMA-NEXT:    retq
+; FMA-INFS-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
+; FMA-INFS:       # %bb.0:
+; FMA-INFS-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
+; FMA-INFS-NEXT:    vmulps %xmm0, %xmm1, %xmm0
+; FMA-INFS-NEXT:    retq
 ;
-; FMA4-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
-; FMA4:       # %bb.0:
-; FMA4-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
-; FMA4-NEXT:    vmulps %xmm0, %xmm1, %xmm0
-; FMA4-NEXT:    retq
+; FMA4-INFS-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
+; FMA4-INFS:       # %bb.0:
+; FMA4-INFS-NEXT:    vaddps {{.*}}(%rip), %xmm0, %xmm0
+; FMA4-INFS-NEXT:    vmulps %xmm0, %xmm1, %xmm0
+; FMA4-INFS-NEXT:    retq
 ;
-; AVX512-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    vaddps {{.*}}(%rip){1to4}, %xmm0, %xmm0
-; AVX512-NEXT:    vmulps %xmm0, %xmm1, %xmm0
-; AVX512-NEXT:    retq
+; AVX512-INFS-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
+; AVX512-INFS:       # %bb.0:
+; AVX512-INFS-NEXT:    vaddps {{.*}}(%rip){1to4}, %xmm0, %xmm0
+; AVX512-INFS-NEXT:    vmulps %xmm0, %xmm1, %xmm0
+; AVX512-INFS-NEXT:    retq
+;
+; FMA-NOINFS-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
+; FMA-NOINFS:       # %bb.0:
+; FMA-NOINFS-NEXT:    vfmsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm1
+; FMA-NOINFS-NEXT:    retq
+;
+; FMA4-NOINFS-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
+; FMA4-NOINFS:       # %bb.0:
+; FMA4-NOINFS-NEXT:    vfmsubps %xmm1, %xmm1, %xmm0, %xmm0
+; FMA4-NOINFS-NEXT:    retq
+;
+; AVX512-NOINFS-LABEL: test_v4f32_mul_y_add_x_negone_undefs:
+; AVX512-NOINFS:       # %bb.0:
+; AVX512-NOINFS-NEXT:    vfmsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm1
+; AVX512-NOINFS-NEXT:    retq
   %a = fadd <4 x float> %x, <float undef, float -1.0, float undef, float -1.0>
   %m = fmul <4 x float> %y, %a
   ret <4 x float> %m