diff --git a/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll b/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll
index 4f1affb94e61..af0529fe41ba 100644
--- a/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll
+++ b/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll
@@ -4282,6 +4282,28 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_mask_fmsub_pd_unary_fneg(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X86-LABEL: test_mm_mask_fmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsub132pd {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132pd {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %neg.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_mm_mask3_fmadd_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm_mask3_fmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -4329,6 +4351,30 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_mask3_fnmadd_pd_unary_fneg(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm_mask3_fnmadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__A
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %neg.i, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_mm_maskz_fmadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
 ; X86-LABEL: test_mm_maskz_fmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -4372,6 +4418,28 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_maskz_fmsub_pd_unary_fneg(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X86-LABEL: test_mm_maskz_fmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %neg.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_mm_maskz_fnmadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
 ; X86-LABEL: test_mm_maskz_fnmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -4394,6 +4462,28 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_maskz_fnmadd_pd_unary_fneg(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X86-LABEL: test_mm_maskz_fnmadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__A
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %neg.i, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_mm_maskz_fnmsub_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
 ; X86-LABEL: test_mm_maskz_fnmsub_pd:
 ; X86:       # %bb.0: # %entry
@@ -4417,6 +4507,29 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_maskz_fnmsub_pd_unary_fneg(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X86-LABEL: test_mm_maskz_fnmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__A
+  %neg1.i = fneg <2 x double> %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %neg.i, <2 x double> %__B, <2 x double> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
 define <4 x double> @test_mm256_mask_fmadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
 ; X86-LABEL: test_mm256_mask_fmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -4460,6 +4573,28 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_mask_fmsub_pd_unary_fneg(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X86-LABEL: test_mm256_mask_fmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %neg.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
 define <4 x double> @test_mm256_mask3_fmadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm256_mask3_fmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -4507,6 +4642,30 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_mask3_fnmadd_pd_unary_fneg(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm256_mask3_fnmadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) + ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fnmadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__A
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %neg.i, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
 define <4 x double> @test_mm256_maskz_fmadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
 ; X86-LABEL: test_mm256_maskz_fmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -4550,6 +4709,28 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_maskz_fmsub_pd_unary_fneg(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X86-LABEL: test_mm256_maskz_fmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %neg.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
 define <4 x double> @test_mm256_maskz_fnmadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
 ; X86-LABEL: test_mm256_maskz_fnmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -4572,6 +4753,28 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_maskz_fnmadd_pd_unary_fneg(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X86-LABEL: test_mm256_maskz_fnmadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fnmadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__A
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %neg.i, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
 define <4 x double> @test_mm256_maskz_fnmsub_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
 ; X86-LABEL: test_mm256_maskz_fnmsub_pd:
 ; X86:       # %bb.0: # %entry
@@ -4595,6 +4798,29 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_maskz_fnmsub_pd_unary_fneg(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X86-LABEL: test_mm256_maskz_fnmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fnmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__A
+  %neg1.i = fneg <4 x double> %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %neg.i, <4 x double> %__B, <4 x double> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
 define <4 x float> @test_mm_mask_fmadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
 ; X86-LABEL: test_mm_mask_fmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -4638,6 +4864,28 @@ entry:
   ret <4 x float> %2
 }
 
+define <4 x float> @test_mm_mask_fmsub_ps_unary_fneg(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X86-LABEL: test_mm_mask_fmsub_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsub132ps {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132ps {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %neg.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
 define <4 x float> @test_mm_mask3_fmadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm_mask3_fmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -4685,6 +4933,30 @@ entry:
   ret <4 x float> %2
 }
 
+define <4 x float> @test_mm_mask3_fnmadd_ps_unary_fneg(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm_mask3_fnmadd_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd231ps {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X86-NEXT:    vmovaps %xmm2, %xmm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231ps {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__A
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %neg.i, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
 define <4 x float> @test_mm_maskz_fmadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
 ; X86-LABEL: test_mm_maskz_fmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -4728,6 +5000,28 @@ entry:
   ret <4 x float> %2
 }
 
+define <4 x float> @test_mm_maskz_fmsub_ps_unary_fneg(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X86-LABEL: test_mm_maskz_fmsub_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %neg.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
 define <4 x float> @test_mm_maskz_fnmadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
 ; X86-LABEL: test_mm_maskz_fnmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -4750,6 +5044,28 @@ entry:
   ret <4 x float> %2
 }
 
+define <4 x float> @test_mm_maskz_fnmadd_ps_unary_fneg(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X86-LABEL: test_mm_maskz_fnmadd_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__A
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %neg.i, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
 define <4 x float> @test_mm_maskz_fnmsub_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
 ; X86-LABEL: test_mm_maskz_fnmsub_ps:
 ; X86:       # %bb.0: # %entry
@@ -4773,6 +5089,29 @@ entry:
   ret <4 x float> %2
 }
 
+define <4 x float> @test_mm_maskz_fnmsub_ps_unary_fneg(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X86-LABEL: test_mm_maskz_fnmsub_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__A
+  %neg1.i = fneg <4 x float> %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %neg.i, <4 x float> %__B, <4 x float> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
 define <8 x float> @test_mm256_mask_fmadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
 ; X86-LABEL: test_mm256_mask_fmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -4814,6 +5153,27 @@ entry:
   ret <8 x float> %2
 }
 
+define <8 x float> @test_mm256_mask_fmsub_ps_unary_fneg(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X86-LABEL: test_mm256_mask_fmsub_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmsub_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <8 x float> %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %neg.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
 define <8 x float> @test_mm256_mask3_fmadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm256_mask3_fmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -5547,6 +5907,32 @@ entry:
   ret <2 x double> %4
 }
 
+define <2 x double> @test_mm_mask3_fmsubadd_pd_unary_fneg(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm_mask3_fmsubadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsubadd231pd {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsubadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231pd {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %neg.i) #9
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract.i, <2 x double> %2, <2 x double> %__C
+  ret <2 x double> %4
+}
+
 define <4 x double> @test_mm256_mask3_fmsubadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm256_mask3_fmsubadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -5573,6 +5959,32 @@ entry:
   ret <4 x double> %4
 }
 
+define <4 x double> @test_mm256_mask3_fmsubadd_pd_unary_fneg(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm256_mask3_fmsubadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsubadd231pd {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmsubadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231pd {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %neg.i) #9
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x double> %2, <4 x double> %__C
+  ret <4 x double> %4
+}
+
 define <4 x float> @test_mm_mask3_fmsubadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm_mask3_fmsubadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -5599,6 +6011,32 @@ entry:
   ret <4 x float> %4
 }
 
+define <4 x float> @test_mm_mask3_fmsubadd_ps_unary_fneg(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm_mask3_fmsubadd_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsubadd231ps {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X86-NEXT:    vmovaps %xmm2, %xmm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsubadd_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231ps {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %neg.i) #9
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x float> %2, <4 x float> %__C
+  ret <4 x float> %4
+}
+
 define <8 x float> @test_mm256_mask3_fmsubadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm256_mask3_fmsubadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -5624,6 +6062,31 @@ entry:
   ret <8 x float> %4
 }
 
+define <8 x float> @test_mm256_mask3_fmsubadd_ps_unary_fneg(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm256_mask3_fmsubadd_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfmsubadd231ps {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X86-NEXT:    vmovaps %ymm2, %ymm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmsubadd_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231ps {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <8 x float> %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %neg.i) #9
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> %__C
+  ret <8 x float> %4
+}
+
 define <2 x double> @test_mm_mask_fnmadd_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
 ; X86-LABEL: test_mm_mask_fnmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -5646,6 +6109,28 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_mask_fnmadd_pd_unary_fneg(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X86-LABEL: test_mm_mask_fnmadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__B
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %neg.i, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
 define <4 x double> @test_mm256_mask_fnmadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
 ; X86-LABEL: test_mm256_mask_fnmadd_pd:
 ; X86:       # %bb.0: # %entry
@@ -5668,6 +6153,28 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_mask_fnmadd_pd_unary_fneg(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X86-LABEL: test_mm256_mask_fnmadd_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fnmadd_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__B
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %neg.i, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
 define <4 x float> @test_mm_mask_fnmadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
 ; X86-LABEL: test_mm_mask_fnmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -5690,6 +6197,28 @@ entry:
   ret <4 x float> %2
 }
 
+define <4 x float> @test_mm_mask_fnmadd_ps_unary_fneg(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X86-LABEL: test_mm_mask_fnmadd_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__B
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %neg.i, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
 define <8 x float> @test_mm256_mask_fnmadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
 ; X86-LABEL: test_mm256_mask_fnmadd_ps:
 ; X86:       # %bb.0: # %entry
@@ -5711,6 +6240,27 @@ entry:
   ret <8 x float> %2
 }
 
+define <8 x float> @test_mm256_mask_fnmadd_ps_unary_fneg(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X86-LABEL: test_mm256_mask_fnmadd_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fnmadd_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <8 x float> %__B
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %neg.i, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
 define <2 x double> @test_mm_mask_fnmsub_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
 ; X86-LABEL: test_mm_mask_fnmsub_pd:
 ; X86:       # %bb.0: # %entry
@@ -5734,6 +6284,29 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_mask_fnmsub_pd_unary_fneg(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X86-LABEL: test_mm_mask_fnmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__B
+  %neg1.i = fneg <2 x double> %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %neg.i, <2 x double> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_mm_mask3_fnmsub_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm_mask3_fnmsub_pd:
 ; X86:       # %bb.0: # %entry
@@ -5759,6 +6332,31 @@ entry:
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mm_mask3_fnmsub_pd_unary_fneg(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm_mask3_fnmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <2 x double> %__B
+  %neg1.i = fneg <2 x double> %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %neg.i, <2 x double> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
 define <4 x double> @test_mm256_mask_fnmsub_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
 ; X86-LABEL: test_mm256_mask_fnmsub_pd:
 ; X86:       # %bb.0: # %entry
@@ -5782,6 +6380,29 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_mask_fnmsub_pd_unary_fneg(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X86-LABEL: test_mm256_mask_fnmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) - ymm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fnmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__B
+  %neg1.i = fneg <4 x double> %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %neg.i, <4 x double> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
 define <4 x double> @test_mm256_mask3_fnmsub_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm256_mask3_fnmsub_pd:
 ; X86:       # %bb.0: # %entry
@@ -5807,6 +6428,31 @@ entry:
   ret <4 x double> %2
 }
 
+define <4 x double> @test_mm256_mask3_fnmsub_pd_unary_fneg(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm256_mask3_fnmsub_pd_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) - ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fnmsub_pd_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x double> %__B
+  %neg1.i = fneg <4 x double> %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %neg.i, <4 x double> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
 define <4 x float> @test_mm_mask_fnmsub_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
 ; X86-LABEL: test_mm_mask_fnmsub_ps:
 ; X86:       # %bb.0: # %entry
@@ -5830,6 +6476,29 @@ entry:
   ret <4 x float> %2
 }
 
+define <4 x float> @test_mm_mask_fnmsub_ps_unary_fneg(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X86-LABEL: test_mm_mask_fnmsub_ps_unary_fneg:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    kmovw %eax, %k1
+; X86-NEXT:    vfnmsub132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_ps_unary_fneg:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %neg.i = fneg <4 x float> %__B
+  %neg1.i = fneg <4 x float> %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %neg.i, <4 x float> %neg1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
 define <4 x float> @test_mm_mask3_fnmsub_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
 ; X86-LABEL: test_mm_mask3_fnmsub_ps:
 ; X86:       # %bb.0: # %entry
diff --git a/test/CodeGen/X86/combine-fabs.ll b/test/CodeGen/X86/combine-fabs.ll
index b779c589cf9d..67b695ebd018 100644
--- a/test/CodeGen/X86/combine-fabs.ll
+++ b/test/CodeGen/X86/combine-fabs.ll
@@ -85,6 +85,22 @@ define float @combine_fabs_fneg(float %a) {
   ret float %2
 }
 
+define float @combine_fabs_unary_fneg(float %a) {
+; SSE-LABEL: combine_fabs_unary_fneg:
+; SSE:       # %bb.0:
+; SSE-NEXT:    andps {{.*}}(%rip), %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_fabs_unary_fneg:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vbroadcastss {{.*}}(%rip), %xmm1
+; AVX-NEXT:    vandps %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %1 = fneg float %a
+  %2 = call float @llvm.fabs.f32(float %1)
+  ret float %2
+}
+
 define <4 x float> @combine_vec_fabs_fneg(<4 x float> %a) {
 ; SSE-LABEL: combine_vec_fabs_fneg:
 ; SSE:       # %bb.0:
@@ -101,6 +117,22 @@ define <4 x float> @combine_vec_fabs_fneg(<4 x float> %a) {
   ret <4 x float> %2
 }
 
+define <4 x float> @combine_vec_fabs_unary_fneg(<4 x float> %a) {
+; SSE-LABEL: combine_vec_fabs_unary_fneg:
+; SSE:       # %bb.0:
+; SSE-NEXT:    andps {{.*}}(%rip), %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_vec_fabs_unary_fneg:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vbroadcastss {{.*}}(%rip), %xmm1
+; AVX-NEXT:    vandps %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %1 = fneg <4 x float> %a
+  %2 = call <4 x float> @llvm.fabs.v4f32(<4 x float> %1)
+  ret <4 x float> %2
+}
+
 ; fabs(fcopysign(x, y)) -> fabs(x)
 define float @combine_fabs_fcopysign(float %a, float %b) {
 ; SSE-LABEL: combine_fabs_fcopysign: