[X86] Enhance combineFMA to look for FNEG behind an EXTRACT_VECTOR_ELT.

llvm-svn: 336514

3 files changed, 45 insertions, 111 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 6d5923010cc..5c7b43cc086 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -37762,11 +37762,23 @@ static SDValue combineFMA(SDNode *N, SelectionDAG &DAG,
   SDValue B = N->getOperand(1);
   SDValue C = N->getOperand(2);
 
-  auto invertIfNegative = [](SDValue &V) {
+  auto invertIfNegative = [&DAG](SDValue &V) {
     if (SDValue NegVal = isFNEG(V.getNode())) {
       V = NegVal;
       return true;
     }
+    // Look through extract_vector_elts. If it comes from an FNEG, create a
+    // new extract from the FNEG input.
+    if (V.getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
+        isa<ConstantSDNode>(V.getOperand(1)) &&
+        cast<ConstantSDNode>(V.getOperand(1))->getZExtValue() == 0) {
+      if (SDValue NegVal = isFNEG(V.getOperand(0).getNode())) {
+        V = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(V), V.getValueType(),
+                        NegVal, V.getOperand(1));
+        return true;
+      }
+    }
+
     return false;
   };
 
diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
index 1c7f633df38..24f08585a4d 100644
--- a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
+++ b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
@@ -5048,18 +5048,14 @@ define <4 x float> @test_mm_mask_fmsub_round_ss(<4 x float> %__W, i8 zeroext %__
 ; X86-LABEL: test_mm_mask_fmsub_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vfmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask_fmsub_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vfmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <4 x float> %__W, i64 0
@@ -5104,18 +5100,14 @@ define <4 x float> @test_mm_maskz_fmsub_round_ss(i8 zeroext %__U, <4 x float> %_
 ; X86-LABEL: test_mm_maskz_fmsub_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X86-NEXT:    vfmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_maskz_fmsub_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X64-NEXT:    vfmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <4 x float> %__A, i64 0
@@ -5163,21 +5155,15 @@ define <4 x float> @test_mm_mask3_fmsub_round_ss(<4 x float> %__W, <4 x float> %
 ; X86-LABEL: test_mm_mask3_fmsub_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm2, %xmm3
-; X86-NEXT:    vfmadd213ss %xmm3, %xmm0, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; X86-NEXT:    vfmsub231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X86-NEXT:    vmovaps %xmm2, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask3_fmsub_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm3
-; X64-NEXT:    vfmadd213ss %xmm3, %xmm0, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; X64-NEXT:    vfmsub231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X64-NEXT:    vmovaps %xmm2, %xmm0
 ; X64-NEXT:    retq
 entry:
@@ -5224,18 +5210,14 @@ define <4 x float> @test_mm_mask_fnmadd_round_ss(<4 x float> %__W, i8 zeroext %_
 ; X86-LABEL: test_mm_mask_fnmadd_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm1, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vfnmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask_fnmadd_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vfnmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <4 x float> %__W, i64 0
@@ -5280,18 +5262,14 @@ define <4 x float> @test_mm_maskz_fnmadd_round_ss(i8 zeroext %__U, <4 x float> %
 ; X86-LABEL: test_mm_maskz_fnmadd_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm1, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X86-NEXT:    vfnmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_maskz_fnmadd_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X64-NEXT:    vfnmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <4 x float> %__A, i64 0
@@ -5339,19 +5317,15 @@ define <4 x float> @test_mm_mask3_fnmadd_round_ss(<4 x float> %__W, <4 x float>
 ; X86-LABEL: test_mm_mask3_fnmadd_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm1, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
+; X86-NEXT:    vfnmadd231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X86-NEXT:    vmovaps %xmm2, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask3_fnmadd_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
+; X64-NEXT:    vfnmadd231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X64-NEXT:    vmovaps %xmm2, %xmm0
 ; X64-NEXT:    retq
 entry:
@@ -5399,20 +5373,14 @@ define <4 x float> @test_mm_mask_fnmsub_round_ss(<4 x float> %__W, i8 zeroext %_
 ; X86-LABEL: test_mm_mask_fnmsub_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm1, %xmm1
-; X86-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vfnmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask_fnmsub_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
-; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vfnmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <4 x float> %__W, i64 0
@@ -5459,20 +5427,14 @@ define <4 x float> @test_mm_maskz_fnmsub_round_ss(i8 zeroext %__U, <4 x float> %
 ; X86-LABEL: test_mm_maskz_fnmsub_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm1, %xmm1
-; X86-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X86-NEXT:    vfnmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_maskz_fnmsub_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
-; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X64-NEXT:    vfnmsub213ss {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <4 x float> %__A, i64 0
@@ -5522,23 +5484,15 @@ define <4 x float> @test_mm_mask3_fnmsub_round_ss(<4 x float> %__W, <4 x float>
 ; X86-LABEL: test_mm_mask3_fnmsub_round_ss:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X86-NEXT:    vxorps %xmm3, %xmm1, %xmm1
-; X86-NEXT:    vxorps %xmm3, %xmm2, %xmm3
-; X86-NEXT:    vfmadd213ss %xmm3, %xmm0, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; X86-NEXT:    vfnmsub231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X86-NEXT:    vmovaps %xmm2, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask3_fnmsub_round_ss:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
-; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
-; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm3
-; X64-NEXT:    vfmadd213ss %xmm3, %xmm0, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovss %xmm1, %xmm2, %xmm2 {%k1}
+; X64-NEXT:    vfnmsub231ss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X64-NEXT:    vmovaps %xmm2, %xmm0
 ; X64-NEXT:    retq
 entry:
@@ -6083,20 +6037,14 @@ define <2 x double> @test_mm_mask_fnmsub_round_sd(<2 x double> %__W, i8 zeroext
 ; X86-LABEL: test_mm_mask_fnmsub_round_sd:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
-; X86-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
-; X86-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X86-NEXT:    vfnmsub213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask_fnmsub_round_sd:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
-; X64-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
-; X64-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
+; X64-NEXT:    vfnmsub213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <2 x double> %__W, i64 0
@@ -6143,20 +6091,14 @@ define <2 x double> @test_mm_maskz_fnmsub_round_sd(i8 zeroext %__U, <2 x double>
 ; X86-LABEL: test_mm_maskz_fnmsub_round_sd:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
-; X86-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
-; X86-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vfmadd213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X86-NEXT:    vfnmsub213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_maskz_fnmsub_round_sd:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
-; X64-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
-; X64-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vfmadd213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
+; X64-NEXT:    vfnmsub213sd {rn-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
 ; X64-NEXT:    retq
 entry:
   %0 = extractelement <2 x double> %__A, i64 0
@@ -6206,23 +6148,15 @@ define <2 x double> @test_mm_mask3_fnmsub_round_sd(<2 x double> %__W, <2 x doubl
 ; X86-LABEL: test_mm_mask3_fnmsub_round_sd:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
-; X86-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
-; X86-NEXT:    vxorpd %xmm3, %xmm2, %xmm3
-; X86-NEXT:    vfmadd213sd %xmm3, %xmm0, %xmm1
 ; X86-NEXT:    kmovw %eax, %k1
-; X86-NEXT:    vmovsd %xmm1, %xmm2, %xmm2 {%k1}
+; X86-NEXT:    vfnmsub231sd {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X86-NEXT:    vmovapd %xmm2, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_mm_mask3_fnmsub_round_sd:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
-; X64-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
-; X64-NEXT:    vxorpd %xmm3, %xmm2, %xmm3
-; X64-NEXT:    vfmadd213sd %xmm3, %xmm0, %xmm1
 ; X64-NEXT:    kmovw %edi, %k1
-; X64-NEXT:    vmovsd %xmm1, %xmm2, %xmm2 {%k1}
+; X64-NEXT:    vfnmsub231sd {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
 ; X64-NEXT:    vmovapd %xmm2, %xmm0
 ; X64-NEXT:    retq
 entry:
diff --git a/llvm/test/CodeGen/X86/fma4-fneg-combine.ll b/llvm/test/CodeGen/X86/fma4-fneg-combine.ll
index f29908678a7..771162a2c99 100644
--- a/llvm/test/CodeGen/X86/fma4-fneg-combine.ll
+++ b/llvm/test/CodeGen/X86/fma4-fneg-combine.ll
@@ -20,8 +20,7 @@ define <4 x float> @test1(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 define <4 x float> @test2(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 ; CHECK-LABEL: test2:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vxorps {{.*}}(%rip), %xmm2, %xmm2
-; CHECK-NEXT:    vfmaddss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfmsubss %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
   %res = tail call <4 x float> @llvm.x86.fma4.vfmadd.ss(<4 x float> %a, <4 x float> %b, <4 x float> %sub.i)
@@ -31,8 +30,7 @@ define <4 x float> @test2(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 define <4 x float> @test3(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 ; CHECK-LABEL: test3:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vxorps {{.*}}(%rip), %xmm1, %xmm1
-; CHECK-NEXT:    vfmaddss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %b
   %res = tail call <4 x float> @llvm.x86.fma4.vfmadd.ss(<4 x float> %a, <4 x float> %sub.i, <4 x float> %c)
@@ -42,8 +40,7 @@ define <4 x float> @test3(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 define <4 x float> @test4(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 ; CHECK-LABEL: test4:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vxorps {{.*}}(%rip), %xmm0, %xmm0
-; CHECK-NEXT:    vfmaddss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
   %res = tail call <4 x float> @llvm.x86.fma4.vfmadd.ss(<4 x float> %sub.i, <4 x float> %b, <4 x float> %c)
@@ -53,10 +50,7 @@ define <4 x float> @test4(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 define <4 x float> @test5(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 ; CHECK-LABEL: test5:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00]
-; CHECK-NEXT:    vxorps %xmm3, %xmm0, %xmm0
-; CHECK-NEXT:    vxorps %xmm3, %xmm2, %xmm2
-; CHECK-NEXT:    vfmaddss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfnmsubss %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
   %sub.i.2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
@@ -78,8 +72,7 @@ define <2 x double> @test6(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 define <2 x double> @test7(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: test7:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vxorpd {{.*}}(%rip), %xmm2, %xmm2
-; CHECK-NEXT:    vfmaddsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfmsubsd %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
   %res = tail call <2 x double> @llvm.x86.fma4.vfmadd.sd(<2 x double> %a, <2 x double> %b, <2 x double> %sub.i)
@@ -89,8 +82,7 @@ define <2 x double> @test7(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 define <2 x double> @test8(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: test8:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vxorpd {{.*}}(%rip), %xmm1, %xmm1
-; CHECK-NEXT:    vfmaddsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfnmaddsd %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %b
   %res = tail call <2 x double> @llvm.x86.fma4.vfmadd.sd(<2 x double> %a, <2 x double> %sub.i, <2 x double> %c)
@@ -100,8 +92,7 @@ define <2 x double> @test8(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 define <2 x double> @test9(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: test9:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vxorpd {{.*}}(%rip), %xmm0, %xmm0
-; CHECK-NEXT:    vfmaddsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfnmaddsd %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a
   %res = tail call <2 x double> @llvm.x86.fma4.vfmadd.sd(<2 x double> %sub.i, <2 x double> %b, <2 x double> %c)
@@ -111,10 +102,7 @@ define <2 x double> @test9(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 define <2 x double> @test10(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: test10:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
-; CHECK-NEXT:    vxorpd %xmm3, %xmm0, %xmm0
-; CHECK-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
-; CHECK-NEXT:    vfmaddsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vfnmsubsd %xmm2, %xmm1, %xmm0, %xmm0
 ; CHECK-NEXT:    retq
   %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a
   %sub.i.2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c