; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-- -mcpu=bdver2 | FileCheck %s --check-prefixes=FMA4
; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=FMA3
; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=FMA3

; PR173172

define void @fnma(ptr %_0, ptr %a, ptr %b, ptr %c) {
; FMA4-LABEL: fnma:
; FMA4:       # %bb.0:
; FMA4-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
; FMA4-NEXT:    vmovups (%rdx), %xmm1
; FMA4-NEXT:    vmovups (%rcx), %xmm2
; FMA4-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm3
; FMA4-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
; FMA4-NEXT:    vfnmaddps {{.*#+}} ymm0 = -(ymm0 * ymm3) + ymm1
; FMA4-NEXT:    vmovups %ymm0, (%rdi)
; FMA4-NEXT:    vzeroupper
; FMA4-NEXT:    retq
;
; FMA3-LABEL: fnma:
; FMA3:       # %bb.0:
; FMA3-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
; FMA3-NEXT:    vmovups (%rdx), %xmm1
; FMA3-NEXT:    vmovups (%rcx), %xmm2
; FMA3-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm3
; FMA3-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
; FMA3-NEXT:    vfnmadd231ps {{.*#+}} ymm1 = -(ymm0 * ymm3) + ymm1
; FMA3-NEXT:    vmovups %ymm1, (%rdi)
; FMA3-NEXT:    vzeroupper
; FMA3-NEXT:    retq
  %i = load <4 x float>, ptr %a, align 4
  %i1 = fneg <4 x float> %i
  %i2 = shufflevector <4 x float> %i1, <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
  %i3 = load <4 x float>, ptr %b, align 4
  %i4 = load <4 x float>, ptr %c, align 4
  %i5 = shufflevector <4 x float> %i3, <4 x float> %i4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %i6 = shufflevector <4 x float> %i4, <4 x float> %i3, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %i7 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %i2, <8 x float> %i5, <8 x float> %i6)
  store <8 x float> %i7, ptr %_0, align 4
  ret void
}

define void @fnma2(ptr %_0, ptr %a, ptr %b, ptr %c) {
; FMA4-LABEL: fnma2:
; FMA4:       # %bb.0:
; FMA4-NEXT:    vmovups (%rsi), %xmm0
; FMA4-NEXT:    vmovups (%rdx), %xmm1
; FMA4-NEXT:    vmovups (%rcx), %xmm2
; FMA4-NEXT:    vfnmaddps {{.*#+}} xmm3 = -(xmm0 * xmm2) + xmm1
; FMA4-NEXT:    vfnmaddps {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
; FMA4-NEXT:    vmovups %xmm0, (%rdi)
; FMA4-NEXT:    vmovups %xmm3, 16(%rdi)
; FMA4-NEXT:    retq
;
; FMA3-LABEL: fnma2:
; FMA3:       # %bb.0:
; FMA3-NEXT:    vmovups (%rsi), %xmm0
; FMA3-NEXT:    vmovups (%rdx), %xmm1
; FMA3-NEXT:    vmovups (%rcx), %xmm2
; FMA3-NEXT:    vmovaps %xmm2, %xmm3
; FMA3-NEXT:    vfnmadd213ps {{.*#+}} xmm3 = -(xmm0 * xmm3) + xmm1
; FMA3-NEXT:    vfnmadd213ps {{.*#+}} xmm1 = -(xmm0 * xmm1) + xmm2
; FMA3-NEXT:    vmovups %xmm1, (%rdi)
; FMA3-NEXT:    vmovups %xmm3, 16(%rdi)
; FMA3-NEXT:    retq
  %va = load <4 x float>, ptr %a, align 4
  %vb = load <4 x float>, ptr %b, align 4
  %vc = load <4 x float>, ptr %c, align 4
  %i = fneg <4 x float> %va
  %i1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %i, <4 x float> %vc, <4 x float> %vb)
  %i2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %i, <4 x float> %vb, <4 x float> %vc)
  store <4 x float> %i2, ptr %_0, align 4
  %i3 = getelementptr inbounds nuw i8, ptr %_0, i64 16
  store <4 x float> %i1, ptr %i3, align 4
  ret void
}

define void @fnma3(ptr %_0, ptr %a, ptr %b, ptr %c) {
; FMA4-LABEL: fnma3:
; FMA4:       # %bb.0:
; FMA4-NEXT:    vmovups (%rdx), %xmm0
; FMA4-NEXT:    vmovups (%rcx), %xmm1
; FMA4-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = mem[0,1,0,1]
; FMA4-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm3
; FMA4-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; FMA4-NEXT:    vfnmaddps {{.*#+}} ymm0 = -(ymm2 * ymm3) + ymm0
; FMA4-NEXT:    vmovups %ymm0, (%rdi)
; FMA4-NEXT:    vzeroupper
; FMA4-NEXT:    retq
;
; FMA3-LABEL: fnma3:
; FMA3:       # %bb.0:
; FMA3-NEXT:    vmovups (%rdx), %xmm0
; FMA3-NEXT:    vmovups (%rcx), %xmm1
; FMA3-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = mem[0,1,0,1]
; FMA3-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm3
; FMA3-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; FMA3-NEXT:    vfnmadd231ps {{.*#+}} ymm0 = -(ymm2 * ymm3) + ymm0
; FMA3-NEXT:    vmovups %ymm0, (%rdi)
; FMA3-NEXT:    vzeroupper
; FMA3-NEXT:    retq
  %va = load <4 x float>, ptr %a, align 4
  %vb = load <4 x float>, ptr %b, align 4
  %vc = load <4 x float>, ptr %c, align 4
  %i = fneg <4 x float> %va
  %i1 = shufflevector <4 x float> %i, <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
  %i2 = shufflevector <4 x float> %vb, <4 x float> %vc, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %i3 = shufflevector <4 x float> %vc, <4 x float> %vb, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %i4 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %i1, <8 x float> %i2, <8 x float> %i3)
  store <8 x float> %i4, ptr %_0, align 4
  ret void
}