; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+zvfh,+v,+m -target-abi=ilp32d \
; RUN:     -verify-machineinstrs < %s | FileCheck %s
; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+zvfh,+v,+m -target-abi=lp64d \
; RUN:     -verify-machineinstrs < %s | FileCheck %s

declare <2 x half> @llvm.vp.fma.v2f16(<2 x half>, <2 x half>, <2 x half>, <2 x i1>, i32)
declare <2 x half> @llvm.vp.fneg.v2f16(<2 x half>, <2 x i1>, i32)
declare <2 x half> @llvm.vp.merge.v2f16(<2 x i1>, <2 x half>, <2 x half>, i32)
declare <2 x half> @llvm.vp.select.v2f16(<2 x i1>, <2 x half>, <2 x half>, i32)

define <2 x half> @vfmacc_vv_v2f16(<2 x half> %a, <2 x half> %b, <2 x half> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, tu, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %a, <2 x half> %b, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.merge.v2f16(<2 x i1> %m, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

define <2 x half> @vfmacc_vv_v2f16_unmasked(<2 x half> %a, <2 x half> %b, <2 x half> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, tu, ma
; CHECK-NEXT:    vfmacc.vv v10, v8, v9
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %a, <2 x half> %b, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.merge.v2f16(<2 x i1> %allones, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

define <2 x half> @vfmacc_vf_v2f16(<2 x half> %va, half %b, <2 x half> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x half> poison, half %b, i32 0
  %vb = shufflevector <2 x half> %elt.head, <2 x half> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %va, <2 x half> %vb, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.merge.v2f16(<2 x i1> %m, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

define <2 x half> @vfmacc_vf_v2f16_commute(<2 x half> %va, half %b, <2 x half> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f16_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x half> poison, half %b, i32 0
  %vb = shufflevector <2 x half> %elt.head, <2 x half> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %vb, <2 x half> %va, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.merge.v2f16(<2 x i1> %m, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

define <2 x half> @vfmacc_vf_v2f16_unmasked(<2 x half> %va, half %b, <2 x half> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, tu, ma
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x half> poison, half %b, i32 0
  %vb = shufflevector <2 x half> %elt.head, <2 x half> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %va, <2 x half> %vb, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.merge.v2f16(<2 x i1> %allones, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

define <2 x half> @vfmacc_vv_v2f16_ta(<2 x half> %a, <2 x half> %b, <2 x half> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, ta, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %a, <2 x half> %b, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.select.v2f16(<2 x i1> %m, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

define <2 x half> @vfmacc_vf_v2f16_ta(<2 x half> %va, half %b, <2 x half> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x half> poison, half %b, i32 0
  %vb = shufflevector <2 x half> %elt.head, <2 x half> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %va, <2 x half> %vb, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.select.v2f16(<2 x i1> %m, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

define <2 x half> @vfmacc_vf_v2f16_commute_ta(<2 x half> %va, half %b, <2 x half> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f16_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x half> poison, half %b, i32 0
  %vb = shufflevector <2 x half> %elt.head, <2 x half> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x half> @llvm.vp.fma.v2f16(<2 x half> %vb, <2 x half> %va, <2 x half> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x half> @llvm.vp.select.v2f16(<2 x i1> %m, <2 x half> %v, <2 x half> %c, i32 %evl)
  ret <2 x half> %u
}

declare <4 x half> @llvm.vp.fma.v4f16(<4 x half>, <4 x half>, <4 x half>, <4 x i1>, i32)
declare <4 x half> @llvm.vp.fneg.v4f16(<4 x half>, <4 x i1>, i32)
declare <4 x half> @llvm.vp.merge.v4f16(<4 x i1>, <4 x half>, <4 x half>, i32)
declare <4 x half> @llvm.vp.select.v4f16(<4 x i1>, <4 x half>, <4 x half>, i32)

define <4 x half> @vfmacc_vv_v4f16(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, tu, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.merge.v4f16(<4 x i1> %m, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

define <4 x half> @vfmacc_vv_v4f16_unmasked(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, tu, ma
; CHECK-NEXT:    vfmacc.vv v10, v8, v9
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.merge.v4f16(<4 x i1> %allones, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

define <4 x half> @vfmacc_vf_v4f16(<4 x half> %va, half %b, <4 x half> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x half> poison, half %b, i32 0
  %vb = shufflevector <4 x half> %elt.head, <4 x half> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %va, <4 x half> %vb, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.merge.v4f16(<4 x i1> %m, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

define <4 x half> @vfmacc_vf_v4f16_commute(<4 x half> %va, half %b, <4 x half> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f16_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x half> poison, half %b, i32 0
  %vb = shufflevector <4 x half> %elt.head, <4 x half> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %vb, <4 x half> %va, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.merge.v4f16(<4 x i1> %m, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

define <4 x half> @vfmacc_vf_v4f16_unmasked(<4 x half> %va, half %b, <4 x half> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, tu, ma
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x half> poison, half %b, i32 0
  %vb = shufflevector <4 x half> %elt.head, <4 x half> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %va, <4 x half> %vb, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.merge.v4f16(<4 x i1> %allones, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

define <4 x half> @vfmacc_vv_v4f16_ta(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, ta, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.select.v4f16(<4 x i1> %m, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

define <4 x half> @vfmacc_vf_v4f16_ta(<4 x half> %va, half %b, <4 x half> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x half> poison, half %b, i32 0
  %vb = shufflevector <4 x half> %elt.head, <4 x half> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %va, <4 x half> %vb, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.select.v4f16(<4 x i1> %m, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

define <4 x half> @vfmacc_vf_v4f16_commute_ta(<4 x half> %va, half %b, <4 x half> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f16_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, mf2, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x half> poison, half %b, i32 0
  %vb = shufflevector <4 x half> %elt.head, <4 x half> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x half> @llvm.vp.fma.v4f16(<4 x half> %vb, <4 x half> %va, <4 x half> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x half> @llvm.vp.select.v4f16(<4 x i1> %m, <4 x half> %v, <4 x half> %c, i32 %evl)
  ret <4 x half> %u
}

declare <8 x half> @llvm.vp.fma.v8f16(<8 x half>, <8 x half>, <8 x half>, <8 x i1>, i32)
declare <8 x half> @llvm.vp.fneg.v8f16(<8 x half>, <8 x i1>, i32)
declare <8 x half> @llvm.vp.merge.v8f16(<8 x i1>, <8 x half>, <8 x half>, i32)
declare <8 x half> @llvm.vp.select.v8f16(<8 x i1>, <8 x half>, <8 x half>, i32)

define <8 x half> @vfmacc_vv_v8f16(<8 x half> %a, <8 x half> %b, <8 x half> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, tu, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %a, <8 x half> %b, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.merge.v8f16(<8 x i1> %m, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

define <8 x half> @vfmacc_vv_v8f16_unmasked(<8 x half> %a, <8 x half> %b, <8 x half> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, tu, ma
; CHECK-NEXT:    vfmacc.vv v10, v8, v9
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %a, <8 x half> %b, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.merge.v8f16(<8 x i1> %allones, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

define <8 x half> @vfmacc_vf_v8f16(<8 x half> %va, half %b, <8 x half> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x half> poison, half %b, i32 0
  %vb = shufflevector <8 x half> %elt.head, <8 x half> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %va, <8 x half> %vb, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.merge.v8f16(<8 x i1> %m, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

define <8 x half> @vfmacc_vf_v8f16_commute(<8 x half> %va, half %b, <8 x half> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f16_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x half> poison, half %b, i32 0
  %vb = shufflevector <8 x half> %elt.head, <8 x half> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %vb, <8 x half> %va, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.merge.v8f16(<8 x i1> %m, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

define <8 x half> @vfmacc_vf_v8f16_unmasked(<8 x half> %va, half %b, <8 x half> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, tu, ma
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x half> poison, half %b, i32 0
  %vb = shufflevector <8 x half> %elt.head, <8 x half> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %va, <8 x half> %vb, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.merge.v8f16(<8 x i1> %allones, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

define <8 x half> @vfmacc_vv_v8f16_ta(<8 x half> %a, <8 x half> %b, <8 x half> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, ta, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %a, <8 x half> %b, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.select.v8f16(<8 x i1> %m, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

define <8 x half> @vfmacc_vf_v8f16_ta(<8 x half> %va, half %b, <8 x half> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x half> poison, half %b, i32 0
  %vb = shufflevector <8 x half> %elt.head, <8 x half> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %va, <8 x half> %vb, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.select.v8f16(<8 x i1> %m, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

define <8 x half> @vfmacc_vf_v8f16_commute_ta(<8 x half> %va, half %b, <8 x half> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f16_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m1, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x half> poison, half %b, i32 0
  %vb = shufflevector <8 x half> %elt.head, <8 x half> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x half> @llvm.vp.fma.v8f16(<8 x half> %vb, <8 x half> %va, <8 x half> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x half> @llvm.vp.select.v8f16(<8 x i1> %m, <8 x half> %v, <8 x half> %c, i32 %evl)
  ret <8 x half> %u
}

declare <16 x half> @llvm.vp.fma.v16f16(<16 x half>, <16 x half>, <16 x half>, <16 x i1>, i32)
declare <16 x half> @llvm.vp.fneg.v16f16(<16 x half>, <16 x i1>, i32)
declare <16 x half> @llvm.vp.merge.v16f16(<16 x i1>, <16 x half>, <16 x half>, i32)
declare <16 x half> @llvm.vp.select.v16f16(<16 x i1>, <16 x half>, <16 x half>, i32)

define <16 x half> @vfmacc_vv_v16f16(<16 x half> %a, <16 x half> %b, <16 x half> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v16f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, tu, mu
; CHECK-NEXT:    vfmacc.vv v12, v8, v10, v0.t
; CHECK-NEXT:    vmv2r.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %a, <16 x half> %b, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.merge.v16f16(<16 x i1> %m, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

define <16 x half> @vfmacc_vv_v16f16_unmasked(<16 x half> %a, <16 x half> %b, <16 x half> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v16f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, tu, ma
; CHECK-NEXT:    vfmacc.vv v12, v8, v10
; CHECK-NEXT:    vmv2r.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %a, <16 x half> %b, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.merge.v16f16(<16 x i1> %allones, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

define <16 x half> @vfmacc_vf_v16f16(<16 x half> %va, half %b, <16 x half> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, tu, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x half> poison, half %b, i32 0
  %vb = shufflevector <16 x half> %elt.head, <16 x half> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %va, <16 x half> %vb, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.merge.v16f16(<16 x i1> %m, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

define <16 x half> @vfmacc_vf_v16f16_commute(<16 x half> %va, half %b, <16 x half> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f16_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, tu, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x half> poison, half %b, i32 0
  %vb = shufflevector <16 x half> %elt.head, <16 x half> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %vb, <16 x half> %va, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.merge.v16f16(<16 x i1> %m, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

define <16 x half> @vfmacc_vf_v16f16_unmasked(<16 x half> %va, half %b, <16 x half> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, tu, ma
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x half> poison, half %b, i32 0
  %vb = shufflevector <16 x half> %elt.head, <16 x half> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %va, <16 x half> %vb, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.merge.v16f16(<16 x i1> %allones, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

define <16 x half> @vfmacc_vv_v16f16_ta(<16 x half> %a, <16 x half> %b, <16 x half> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v16f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, ta, mu
; CHECK-NEXT:    vfmacc.vv v12, v8, v10, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %a, <16 x half> %b, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.select.v16f16(<16 x i1> %m, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

define <16 x half> @vfmacc_vf_v16f16_ta(<16 x half> %va, half %b, <16 x half> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, ta, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x half> poison, half %b, i32 0
  %vb = shufflevector <16 x half> %elt.head, <16 x half> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %va, <16 x half> %vb, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.select.v16f16(<16 x i1> %m, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

define <16 x half> @vfmacc_vf_v16f16_commute_ta(<16 x half> %va, half %b, <16 x half> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f16_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m2, ta, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x half> poison, half %b, i32 0
  %vb = shufflevector <16 x half> %elt.head, <16 x half> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x half> @llvm.vp.fma.v16f16(<16 x half> %vb, <16 x half> %va, <16 x half> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x half> @llvm.vp.select.v16f16(<16 x i1> %m, <16 x half> %v, <16 x half> %c, i32 %evl)
  ret <16 x half> %u
}

declare <32 x half> @llvm.vp.fma.v32f16(<32 x half>, <32 x half>, <32 x half>, <32 x i1>, i32)
declare <32 x half> @llvm.vp.fneg.v32f16(<32 x half>, <32 x i1>, i32)
declare <32 x half> @llvm.vp.merge.v32f16(<32 x i1>, <32 x half>, <32 x half>, i32)
declare <32 x half> @llvm.vp.select.v32f16(<32 x i1>, <32 x half>, <32 x half>, i32)

define <32 x half> @vfmacc_vv_v32f16(<32 x half> %a, <32 x half> %b, <32 x half> %c, <32 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v32f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, tu, mu
; CHECK-NEXT:    vfmacc.vv v16, v8, v12, v0.t
; CHECK-NEXT:    vmv4r.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %a, <32 x half> %b, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.merge.v32f16(<32 x i1> %m, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

define <32 x half> @vfmacc_vv_v32f16_unmasked(<32 x half> %a, <32 x half> %b, <32 x half> %c, <32 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v32f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, tu, ma
; CHECK-NEXT:    vfmacc.vv v16, v8, v12
; CHECK-NEXT:    vmv4r.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %a, <32 x half> %b, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.merge.v32f16(<32 x i1> %allones, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

define <32 x half> @vfmacc_vf_v32f16(<32 x half> %va, half %b, <32 x half> %c, <32 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v32f16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, tu, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <32 x half> poison, half %b, i32 0
  %vb = shufflevector <32 x half> %elt.head, <32 x half> poison, <32 x i32> zeroinitializer
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %va, <32 x half> %vb, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.merge.v32f16(<32 x i1> %m, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

define <32 x half> @vfmacc_vf_v32f16_commute(<32 x half> %va, half %b, <32 x half> %c, <32 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v32f16_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, tu, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <32 x half> poison, half %b, i32 0
  %vb = shufflevector <32 x half> %elt.head, <32 x half> poison, <32 x i32> zeroinitializer
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %vb, <32 x half> %va, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.merge.v32f16(<32 x i1> %m, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

define <32 x half> @vfmacc_vf_v32f16_unmasked(<32 x half> %va, half %b, <32 x half> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v32f16_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, tu, ma
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <32 x half> poison, half %b, i32 0
  %vb = shufflevector <32 x half> %elt.head, <32 x half> poison, <32 x i32> zeroinitializer
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %va, <32 x half> %vb, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.merge.v32f16(<32 x i1> %allones, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

define <32 x half> @vfmacc_vv_v32f16_ta(<32 x half> %a, <32 x half> %b, <32 x half> %c, <32 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v32f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, ta, mu
; CHECK-NEXT:    vfmacc.vv v16, v8, v12, v0.t
; CHECK-NEXT:    vmv.v.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %a, <32 x half> %b, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.select.v32f16(<32 x i1> %m, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

define <32 x half> @vfmacc_vf_v32f16_ta(<32 x half> %va, half %b, <32 x half> %c, <32 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v32f16_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, ta, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <32 x half> poison, half %b, i32 0
  %vb = shufflevector <32 x half> %elt.head, <32 x half> poison, <32 x i32> zeroinitializer
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %va, <32 x half> %vb, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.select.v32f16(<32 x i1> %m, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

define <32 x half> @vfmacc_vf_v32f16_commute_ta(<32 x half> %va, half %b, <32 x half> %c, <32 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v32f16_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e16, m4, ta, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <32 x half> poison, half %b, i32 0
  %vb = shufflevector <32 x half> %elt.head, <32 x half> poison, <32 x i32> zeroinitializer
  %splat = insertelement <32 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <32 x i1> %splat, <32 x i1> poison, <32 x i32> zeroinitializer
  %v = call <32 x half> @llvm.vp.fma.v32f16(<32 x half> %vb, <32 x half> %va, <32 x half> %c, <32 x i1> %allones, i32 %evl)
  %u = call <32 x half> @llvm.vp.select.v32f16(<32 x i1> %m, <32 x half> %v, <32 x half> %c, i32 %evl)
  ret <32 x half> %u
}

declare <2 x float> @llvm.vp.fma.v2f32(<2 x float>, <2 x float>, <2 x float>, <2 x i1>, i32)
declare <2 x float> @llvm.vp.fneg.v2f32(<2 x float>, <2 x i1>, i32)
declare <2 x float> @llvm.vp.merge.v2f32(<2 x i1>, <2 x float>, <2 x float>, i32)
declare <2 x float> @llvm.vp.select.v2f32(<2 x i1>, <2 x float>, <2 x float>, i32)

define <2 x float> @vfmacc_vv_v2f32(<2 x float> %a, <2 x float> %b, <2 x float> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, tu, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %a, <2 x float> %b, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.merge.v2f32(<2 x i1> %m, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

define <2 x float> @vfmacc_vv_v2f32_unmasked(<2 x float> %a, <2 x float> %b, <2 x float> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, tu, ma
; CHECK-NEXT:    vfmacc.vv v10, v8, v9
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %a, <2 x float> %b, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.merge.v2f32(<2 x i1> %allones, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

define <2 x float> @vfmacc_vf_v2f32(<2 x float> %va, float %b, <2 x float> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x float> poison, float %b, i32 0
  %vb = shufflevector <2 x float> %elt.head, <2 x float> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %va, <2 x float> %vb, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.merge.v2f32(<2 x i1> %m, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

define <2 x float> @vfmacc_vf_v2f32_commute(<2 x float> %va, float %b, <2 x float> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f32_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x float> poison, float %b, i32 0
  %vb = shufflevector <2 x float> %elt.head, <2 x float> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %vb, <2 x float> %va, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.merge.v2f32(<2 x i1> %m, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

define <2 x float> @vfmacc_vf_v2f32_unmasked(<2 x float> %va, float %b, <2 x float> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, tu, ma
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x float> poison, float %b, i32 0
  %vb = shufflevector <2 x float> %elt.head, <2 x float> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %va, <2 x float> %vb, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.merge.v2f32(<2 x i1> %allones, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

define <2 x float> @vfmacc_vv_v2f32_ta(<2 x float> %a, <2 x float> %b, <2 x float> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, ta, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %a, <2 x float> %b, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.select.v2f32(<2 x i1> %m, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

define <2 x float> @vfmacc_vf_v2f32_ta(<2 x float> %va, float %b, <2 x float> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x float> poison, float %b, i32 0
  %vb = shufflevector <2 x float> %elt.head, <2 x float> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %va, <2 x float> %vb, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.select.v2f32(<2 x i1> %m, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

define <2 x float> @vfmacc_vf_v2f32_commute_ta(<2 x float> %va, float %b, <2 x float> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f32_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, mf2, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x float> poison, float %b, i32 0
  %vb = shufflevector <2 x float> %elt.head, <2 x float> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x float> @llvm.vp.fma.v2f32(<2 x float> %vb, <2 x float> %va, <2 x float> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x float> @llvm.vp.select.v2f32(<2 x i1> %m, <2 x float> %v, <2 x float> %c, i32 %evl)
  ret <2 x float> %u
}

declare <4 x float> @llvm.vp.fma.v4f32(<4 x float>, <4 x float>, <4 x float>, <4 x i1>, i32)
declare <4 x float> @llvm.vp.fneg.v4f32(<4 x float>, <4 x i1>, i32)
declare <4 x float> @llvm.vp.merge.v4f32(<4 x i1>, <4 x float>, <4 x float>, i32)
declare <4 x float> @llvm.vp.select.v4f32(<4 x i1>, <4 x float>, <4 x float>, i32)

define <4 x float> @vfmacc_vv_v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.merge.v4f32(<4 x i1> %m, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

define <4 x float> @vfmacc_vv_v4f32_unmasked(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, ma
; CHECK-NEXT:    vfmacc.vv v10, v8, v9
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.merge.v4f32(<4 x i1> %allones, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

define <4 x float> @vfmacc_vf_v4f32(<4 x float> %va, float %b, <4 x float> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x float> poison, float %b, i32 0
  %vb = shufflevector <4 x float> %elt.head, <4 x float> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %va, <4 x float> %vb, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.merge.v4f32(<4 x i1> %m, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

define <4 x float> @vfmacc_vf_v4f32_commute(<4 x float> %va, float %b, <4 x float> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f32_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x float> poison, float %b, i32 0
  %vb = shufflevector <4 x float> %elt.head, <4 x float> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %vb, <4 x float> %va, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.merge.v4f32(<4 x i1> %m, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

define <4 x float> @vfmacc_vf_v4f32_unmasked(<4 x float> %va, float %b, <4 x float> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, tu, ma
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x float> poison, float %b, i32 0
  %vb = shufflevector <4 x float> %elt.head, <4 x float> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %va, <4 x float> %vb, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.merge.v4f32(<4 x i1> %allones, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

define <4 x float> @vfmacc_vv_v4f32_ta(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.select.v4f32(<4 x i1> %m, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

define <4 x float> @vfmacc_vf_v4f32_ta(<4 x float> %va, float %b, <4 x float> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x float> poison, float %b, i32 0
  %vb = shufflevector <4 x float> %elt.head, <4 x float> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %va, <4 x float> %vb, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.select.v4f32(<4 x i1> %m, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

define <4 x float> @vfmacc_vf_v4f32_commute_ta(<4 x float> %va, float %b, <4 x float> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f32_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m1, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x float> poison, float %b, i32 0
  %vb = shufflevector <4 x float> %elt.head, <4 x float> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x float> @llvm.vp.fma.v4f32(<4 x float> %vb, <4 x float> %va, <4 x float> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x float> @llvm.vp.select.v4f32(<4 x i1> %m, <4 x float> %v, <4 x float> %c, i32 %evl)
  ret <4 x float> %u
}

declare <8 x float> @llvm.vp.fma.v8f32(<8 x float>, <8 x float>, <8 x float>, <8 x i1>, i32)
declare <8 x float> @llvm.vp.fneg.v8f32(<8 x float>, <8 x i1>, i32)
declare <8 x float> @llvm.vp.merge.v8f32(<8 x i1>, <8 x float>, <8 x float>, i32)
declare <8 x float> @llvm.vp.select.v8f32(<8 x i1>, <8 x float>, <8 x float>, i32)

define <8 x float> @vfmacc_vv_v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, tu, mu
; CHECK-NEXT:    vfmacc.vv v12, v8, v10, v0.t
; CHECK-NEXT:    vmv2r.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %m, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

define <8 x float> @vfmacc_vv_v8f32_unmasked(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, tu, ma
; CHECK-NEXT:    vfmacc.vv v12, v8, v10
; CHECK-NEXT:    vmv2r.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %allones, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

define <8 x float> @vfmacc_vf_v8f32(<8 x float> %va, float %b, <8 x float> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, tu, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x float> poison, float %b, i32 0
  %vb = shufflevector <8 x float> %elt.head, <8 x float> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %va, <8 x float> %vb, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %m, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

define <8 x float> @vfmacc_vf_v8f32_commute(<8 x float> %va, float %b, <8 x float> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f32_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, tu, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x float> poison, float %b, i32 0
  %vb = shufflevector <8 x float> %elt.head, <8 x float> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %vb, <8 x float> %va, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %m, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

define <8 x float> @vfmacc_vf_v8f32_unmasked(<8 x float> %va, float %b, <8 x float> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, tu, ma
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x float> poison, float %b, i32 0
  %vb = shufflevector <8 x float> %elt.head, <8 x float> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %va, <8 x float> %vb, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.merge.v8f32(<8 x i1> %allones, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

define <8 x float> @vfmacc_vv_v8f32_ta(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, ta, mu
; CHECK-NEXT:    vfmacc.vv v12, v8, v10, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.select.v8f32(<8 x i1> %m, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

define <8 x float> @vfmacc_vf_v8f32_ta(<8 x float> %va, float %b, <8 x float> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, ta, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x float> poison, float %b, i32 0
  %vb = shufflevector <8 x float> %elt.head, <8 x float> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %va, <8 x float> %vb, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.select.v8f32(<8 x i1> %m, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

define <8 x float> @vfmacc_vf_v8f32_commute_ta(<8 x float> %va, float %b, <8 x float> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f32_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m2, ta, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x float> poison, float %b, i32 0
  %vb = shufflevector <8 x float> %elt.head, <8 x float> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x float> @llvm.vp.fma.v8f32(<8 x float> %vb, <8 x float> %va, <8 x float> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x float> @llvm.vp.select.v8f32(<8 x i1> %m, <8 x float> %v, <8 x float> %c, i32 %evl)
  ret <8 x float> %u
}

declare <16 x float> @llvm.vp.fma.v16f32(<16 x float>, <16 x float>, <16 x float>, <16 x i1>, i32)
declare <16 x float> @llvm.vp.fneg.v16f32(<16 x float>, <16 x i1>, i32)
declare <16 x float> @llvm.vp.merge.v16f32(<16 x i1>, <16 x float>, <16 x float>, i32)
declare <16 x float> @llvm.vp.select.v16f32(<16 x i1>, <16 x float>, <16 x float>, i32)

define <16 x float> @vfmacc_vv_v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v16f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, tu, mu
; CHECK-NEXT:    vfmacc.vv v16, v8, v12, v0.t
; CHECK-NEXT:    vmv4r.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.merge.v16f32(<16 x i1> %m, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

define <16 x float> @vfmacc_vv_v16f32_unmasked(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v16f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, tu, ma
; CHECK-NEXT:    vfmacc.vv v16, v8, v12
; CHECK-NEXT:    vmv4r.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.merge.v16f32(<16 x i1> %allones, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

define <16 x float> @vfmacc_vf_v16f32(<16 x float> %va, float %b, <16 x float> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, tu, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x float> poison, float %b, i32 0
  %vb = shufflevector <16 x float> %elt.head, <16 x float> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %va, <16 x float> %vb, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.merge.v16f32(<16 x i1> %m, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

define <16 x float> @vfmacc_vf_v16f32_commute(<16 x float> %va, float %b, <16 x float> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f32_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, tu, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x float> poison, float %b, i32 0
  %vb = shufflevector <16 x float> %elt.head, <16 x float> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %vb, <16 x float> %va, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.merge.v16f32(<16 x i1> %m, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

define <16 x float> @vfmacc_vf_v16f32_unmasked(<16 x float> %va, float %b, <16 x float> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f32_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, tu, ma
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x float> poison, float %b, i32 0
  %vb = shufflevector <16 x float> %elt.head, <16 x float> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %va, <16 x float> %vb, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.merge.v16f32(<16 x i1> %allones, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

define <16 x float> @vfmacc_vv_v16f32_ta(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v16f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, mu
; CHECK-NEXT:    vfmacc.vv v16, v8, v12, v0.t
; CHECK-NEXT:    vmv.v.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.select.v16f32(<16 x i1> %m, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

define <16 x float> @vfmacc_vf_v16f32_ta(<16 x float> %va, float %b, <16 x float> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f32_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x float> poison, float %b, i32 0
  %vb = shufflevector <16 x float> %elt.head, <16 x float> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %va, <16 x float> %vb, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.select.v16f32(<16 x i1> %m, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

define <16 x float> @vfmacc_vf_v16f32_commute_ta(<16 x float> %va, float %b, <16 x float> %c, <16 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v16f32_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <16 x float> poison, float %b, i32 0
  %vb = shufflevector <16 x float> %elt.head, <16 x float> poison, <16 x i32> zeroinitializer
  %splat = insertelement <16 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <16 x i1> %splat, <16 x i1> poison, <16 x i32> zeroinitializer
  %v = call <16 x float> @llvm.vp.fma.v16f32(<16 x float> %vb, <16 x float> %va, <16 x float> %c, <16 x i1> %allones, i32 %evl)
  %u = call <16 x float> @llvm.vp.select.v16f32(<16 x i1> %m, <16 x float> %v, <16 x float> %c, i32 %evl)
  ret <16 x float> %u
}

declare <2 x double> @llvm.vp.fma.v2f64(<2 x double>, <2 x double>, <2 x double>, <2 x i1>, i32)
declare <2 x double> @llvm.vp.fneg.v2f64(<2 x double>, <2 x i1>, i32)
declare <2 x double> @llvm.vp.merge.v2f64(<2 x i1>, <2 x double>, <2 x double>, i32)
declare <2 x double> @llvm.vp.select.v2f64(<2 x i1>, <2 x double>, <2 x double>, i32)

define <2 x double> @vfmacc_vv_v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, tu, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.merge.v2f64(<2 x i1> %m, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

define <2 x double> @vfmacc_vv_v2f64_unmasked(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f64_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, tu, ma
; CHECK-NEXT:    vfmacc.vv v10, v8, v9
; CHECK-NEXT:    vmv1r.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.merge.v2f64(<2 x i1> %allones, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

define <2 x double> @vfmacc_vf_v2f64(<2 x double> %va, double %b, <2 x double> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x double> poison, double %b, i32 0
  %vb = shufflevector <2 x double> %elt.head, <2 x double> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %va, <2 x double> %vb, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.merge.v2f64(<2 x i1> %m, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

define <2 x double> @vfmacc_vf_v2f64_commute(<2 x double> %va, double %b, <2 x double> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f64_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, tu, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x double> poison, double %b, i32 0
  %vb = shufflevector <2 x double> %elt.head, <2 x double> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %vb, <2 x double> %va, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.merge.v2f64(<2 x i1> %m, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

define <2 x double> @vfmacc_vf_v2f64_unmasked(<2 x double> %va, double %b, <2 x double> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f64_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, tu, ma
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8
; CHECK-NEXT:    vmv1r.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x double> poison, double %b, i32 0
  %vb = shufflevector <2 x double> %elt.head, <2 x double> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %va, <2 x double> %vb, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.merge.v2f64(<2 x i1> %allones, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

define <2 x double> @vfmacc_vv_v2f64_ta(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v2f64_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, ta, mu
; CHECK-NEXT:    vfmacc.vv v10, v8, v9, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.select.v2f64(<2 x i1> %m, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

define <2 x double> @vfmacc_vf_v2f64_ta(<2 x double> %va, double %b, <2 x double> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f64_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x double> poison, double %b, i32 0
  %vb = shufflevector <2 x double> %elt.head, <2 x double> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %va, <2 x double> %vb, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.select.v2f64(<2 x i1> %m, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

define <2 x double> @vfmacc_vf_v2f64_commute_ta(<2 x double> %va, double %b, <2 x double> %c, <2 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v2f64_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m1, ta, mu
; CHECK-NEXT:    vfmacc.vf v9, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v9
; CHECK-NEXT:    ret
  %elt.head = insertelement <2 x double> poison, double %b, i32 0
  %vb = shufflevector <2 x double> %elt.head, <2 x double> poison, <2 x i32> zeroinitializer
  %splat = insertelement <2 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <2 x i1> %splat, <2 x i1> poison, <2 x i32> zeroinitializer
  %v = call <2 x double> @llvm.vp.fma.v2f64(<2 x double> %vb, <2 x double> %va, <2 x double> %c, <2 x i1> %allones, i32 %evl)
  %u = call <2 x double> @llvm.vp.select.v2f64(<2 x i1> %m, <2 x double> %v, <2 x double> %c, i32 %evl)
  ret <2 x double> %u
}

declare <4 x double> @llvm.vp.fma.v4f64(<4 x double>, <4 x double>, <4 x double>, <4 x i1>, i32)
declare <4 x double> @llvm.vp.fneg.v4f64(<4 x double>, <4 x i1>, i32)
declare <4 x double> @llvm.vp.merge.v4f64(<4 x i1>, <4 x double>, <4 x double>, i32)
declare <4 x double> @llvm.vp.select.v4f64(<4 x i1>, <4 x double>, <4 x double>, i32)

define <4 x double> @vfmacc_vv_v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, tu, mu
; CHECK-NEXT:    vfmacc.vv v12, v8, v10, v0.t
; CHECK-NEXT:    vmv2r.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.merge.v4f64(<4 x i1> %m, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

define <4 x double> @vfmacc_vv_v4f64_unmasked(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f64_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, tu, ma
; CHECK-NEXT:    vfmacc.vv v12, v8, v10
; CHECK-NEXT:    vmv2r.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.merge.v4f64(<4 x i1> %allones, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

define <4 x double> @vfmacc_vf_v4f64(<4 x double> %va, double %b, <4 x double> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, tu, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x double> poison, double %b, i32 0
  %vb = shufflevector <4 x double> %elt.head, <4 x double> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %va, <4 x double> %vb, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.merge.v4f64(<4 x i1> %m, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

define <4 x double> @vfmacc_vf_v4f64_commute(<4 x double> %va, double %b, <4 x double> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f64_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, tu, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x double> poison, double %b, i32 0
  %vb = shufflevector <4 x double> %elt.head, <4 x double> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %vb, <4 x double> %va, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.merge.v4f64(<4 x i1> %m, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

define <4 x double> @vfmacc_vf_v4f64_unmasked(<4 x double> %va, double %b, <4 x double> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f64_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, tu, ma
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8
; CHECK-NEXT:    vmv2r.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x double> poison, double %b, i32 0
  %vb = shufflevector <4 x double> %elt.head, <4 x double> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %va, <4 x double> %vb, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.merge.v4f64(<4 x i1> %allones, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

define <4 x double> @vfmacc_vv_v4f64_ta(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v4f64_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, ta, mu
; CHECK-NEXT:    vfmacc.vv v12, v8, v10, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.select.v4f64(<4 x i1> %m, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

define <4 x double> @vfmacc_vf_v4f64_ta(<4 x double> %va, double %b, <4 x double> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f64_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, ta, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x double> poison, double %b, i32 0
  %vb = shufflevector <4 x double> %elt.head, <4 x double> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %va, <4 x double> %vb, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.select.v4f64(<4 x i1> %m, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

define <4 x double> @vfmacc_vf_v4f64_commute_ta(<4 x double> %va, double %b, <4 x double> %c, <4 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v4f64_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m2, ta, mu
; CHECK-NEXT:    vfmacc.vf v10, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v10
; CHECK-NEXT:    ret
  %elt.head = insertelement <4 x double> poison, double %b, i32 0
  %vb = shufflevector <4 x double> %elt.head, <4 x double> poison, <4 x i32> zeroinitializer
  %splat = insertelement <4 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <4 x i1> %splat, <4 x i1> poison, <4 x i32> zeroinitializer
  %v = call <4 x double> @llvm.vp.fma.v4f64(<4 x double> %vb, <4 x double> %va, <4 x double> %c, <4 x i1> %allones, i32 %evl)
  %u = call <4 x double> @llvm.vp.select.v4f64(<4 x i1> %m, <4 x double> %v, <4 x double> %c, i32 %evl)
  ret <4 x double> %u
}

declare <8 x double> @llvm.vp.fma.v8f64(<8 x double>, <8 x double>, <8 x double>, <8 x i1>, i32)
declare <8 x double> @llvm.vp.fneg.v8f64(<8 x double>, <8 x i1>, i32)
declare <8 x double> @llvm.vp.merge.v8f64(<8 x i1>, <8 x double>, <8 x double>, i32)
declare <8 x double> @llvm.vp.select.v8f64(<8 x i1>, <8 x double>, <8 x double>, i32)

define <8 x double> @vfmacc_vv_v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, tu, mu
; CHECK-NEXT:    vfmacc.vv v16, v8, v12, v0.t
; CHECK-NEXT:    vmv4r.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.merge.v8f64(<8 x i1> %m, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}

define <8 x double> @vfmacc_vv_v8f64_unmasked(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f64_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, tu, ma
; CHECK-NEXT:    vfmacc.vv v16, v8, v12
; CHECK-NEXT:    vmv4r.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.merge.v8f64(<8 x i1> %allones, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}

define <8 x double> @vfmacc_vf_v8f64(<8 x double> %va, double %b, <8 x double> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, tu, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x double> poison, double %b, i32 0
  %vb = shufflevector <8 x double> %elt.head, <8 x double> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %va, <8 x double> %vb, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.merge.v8f64(<8 x i1> %m, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}

define <8 x double> @vfmacc_vf_v8f64_commute(<8 x double> %va, double %b, <8 x double> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f64_commute:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, tu, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x double> poison, double %b, i32 0
  %vb = shufflevector <8 x double> %elt.head, <8 x double> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %vb, <8 x double> %va, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.merge.v8f64(<8 x i1> %m, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}

define <8 x double> @vfmacc_vf_v8f64_unmasked(<8 x double> %va, double %b, <8 x double> %c, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f64_unmasked:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, tu, ma
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8
; CHECK-NEXT:    vmv4r.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x double> poison, double %b, i32 0
  %vb = shufflevector <8 x double> %elt.head, <8 x double> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %va, <8 x double> %vb, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.merge.v8f64(<8 x i1> %allones, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}

define <8 x double> @vfmacc_vv_v8f64_ta(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vv_v8f64_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, ta, mu
; CHECK-NEXT:    vfmacc.vv v16, v8, v12, v0.t
; CHECK-NEXT:    vmv.v.v v8, v16
; CHECK-NEXT:    ret
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.select.v8f64(<8 x i1> %m, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}

define <8 x double> @vfmacc_vf_v8f64_ta(<8 x double> %va, double %b, <8 x double> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f64_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, ta, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x double> poison, double %b, i32 0
  %vb = shufflevector <8 x double> %elt.head, <8 x double> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %va, <8 x double> %vb, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.select.v8f64(<8 x i1> %m, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}

define <8 x double> @vfmacc_vf_v8f64_commute_ta(<8 x double> %va, double %b, <8 x double> %c, <8 x i1> %m, i32 zeroext %evl) {
; CHECK-LABEL: vfmacc_vf_v8f64_commute_ta:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsetvli zero, a0, e64, m4, ta, mu
; CHECK-NEXT:    vfmacc.vf v12, fa0, v8, v0.t
; CHECK-NEXT:    vmv.v.v v8, v12
; CHECK-NEXT:    ret
  %elt.head = insertelement <8 x double> poison, double %b, i32 0
  %vb = shufflevector <8 x double> %elt.head, <8 x double> poison, <8 x i32> zeroinitializer
  %splat = insertelement <8 x i1> poison, i1 -1, i32 0
  %allones = shufflevector <8 x i1> %splat, <8 x i1> poison, <8 x i32> zeroinitializer
  %v = call <8 x double> @llvm.vp.fma.v8f64(<8 x double> %vb, <8 x double> %va, <8 x double> %c, <8 x i1> %allones, i32 %evl)
  %u = call <8 x double> @llvm.vp.select.v8f64(<8 x i1> %m, <8 x double> %v, <8 x double> %c, i32 %evl)
  ret <8 x double> %u
}