[llvm] 8fc4fce - [RISCV] Add VL patterns for vector widening floating-point fused multiply-add instructions.
via llvm-commits
llvm-commits at lists.llvm.org
Fri May 20 00:11:39 PDT 2022
Author: jacquesguan
Date: 2022-05-20T06:56:48Z
New Revision: 8fc4fcecb8cedffcbf5fdd564c5a3427274db4e2
URL: https://github.com/llvm/llvm-project/commit/8fc4fcecb8cedffcbf5fdd564c5a3427274db4e2
DIFF: https://github.com/llvm/llvm-project/commit/8fc4fcecb8cedffcbf5fdd564c5a3427274db4e2.diff
LOG: [RISCV] Add VL patterns for vector widening floating-point fused multiply-add instructions.
This patch adds VL patterns for vector widening floating-point fused multiply-add instructions to support fixed length vector type.
Differential Revision: https://reviews.llvm.org/D124505
Added:
llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfwmacc.ll
Modified:
llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
Removed:
################################################################################
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
index 47c3375b6470b..15ad6424b19dc 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td
@@ -900,6 +900,162 @@ multiclass VPatNarrowShiftSplat_WX_WI<SDNode op, string instruction_name> {
}
}
+multiclass VPatWidenFPMulAccVL_VV_VF<string instruction_name> {
+ foreach vtiToWti = AllWidenableFloatVectors in {
+ defvar vti = vtiToWti.Vti;
+ defvar wti = vtiToWti.Wti;
+ def : Pat<(riscv_fma_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs1),
+ (vti.Mask true_mask), VLOpFrag)),
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag)),
+ (wti.Vector wti.RegClass:$rd), (vti.Mask true_mask),
+ VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_VV_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.RegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ def : Pat<(riscv_fma_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector (SplatFPOp vti.ScalarRegClass:$rs1)),
+ (vti.Mask true_mask), VLOpFrag)),
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag)),
+ (wti.Vector wti.RegClass:$rd), (vti.Mask true_mask),
+ VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.ScalarRegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ }
+}
+
+multiclass VPatWidenFPNegMulAccVL_VV_VF<string instruction_name> {
+ foreach vtiToWti = AllWidenableFloatVectors in {
+ defvar vti = vtiToWti.Vti;
+ defvar wti = vtiToWti.Wti;
+ def : Pat<(riscv_fma_vl
+ (riscv_fneg_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs1),
+ (vti.Mask true_mask), VLOpFrag)),
+ (wti.Mask true_mask), VLOpFrag),
+ (riscv_fpextend_vl_oneuse (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fneg_vl wti.RegClass:$rd, (wti.Mask true_mask),
+ VLOpFrag),
+ (vti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_VV_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.RegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ def : Pat<(riscv_fma_vl
+ (riscv_fpextend_vl_oneuse
+ (vti.Vector (SplatFPOp vti.ScalarRegClass:$rs1)),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fneg_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag)),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fneg_vl wti.RegClass:$rd, (wti.Mask true_mask),
+ VLOpFrag),
+ (vti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.ScalarRegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ def : Pat<(riscv_fma_vl
+ (riscv_fneg_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector (SplatFPOp vti.ScalarRegClass:$rs1)),
+ (vti.Mask true_mask), VLOpFrag)),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fpextend_vl_oneuse (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fneg_vl wti.RegClass:$rd, (wti.Mask true_mask),
+ VLOpFrag),
+ (vti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.ScalarRegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ }
+}
+
+multiclass VPatWidenFPMulSacVL_VV_VF<string instruction_name> {
+ foreach vtiToWti = AllWidenableFloatVectors in {
+ defvar vti = vtiToWti.Vti;
+ defvar wti = vtiToWti.Wti;
+ def : Pat<(riscv_fma_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs1),
+ (vti.Mask true_mask), VLOpFrag)),
+ (riscv_fpextend_vl_oneuse (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fneg_vl wti.RegClass:$rd, (vti.Mask true_mask),
+ VLOpFrag),
+ (vti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_VV_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.RegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ def : Pat<(riscv_fma_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector (SplatFPOp vti.ScalarRegClass:$rs1)),
+ (vti.Mask true_mask), VLOpFrag)),
+ (riscv_fpextend_vl_oneuse (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fneg_vl wti.RegClass:$rd, (vti.Mask true_mask),
+ VLOpFrag),
+ (vti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.ScalarRegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ }
+}
+
+multiclass VPatWidenFPNegMulSacVL_VV_VF<string instruction_name> {
+ foreach vtiToWti = AllWidenableFloatVectors in {
+ defvar vti = vtiToWti.Vti;
+ defvar wti = vtiToWti.Wti;
+ def : Pat<(riscv_fma_vl
+ (riscv_fneg_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs1),
+ (vti.Mask true_mask), VLOpFrag)),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fpextend_vl_oneuse (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag),
+ wti.RegClass:$rd, (wti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_VV_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.RegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ def : Pat<(riscv_fma_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector (SplatFPOp vti.ScalarRegClass:$rs1)),
+ (vti.Mask true_mask), VLOpFrag)),
+ (riscv_fneg_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag)),
+ (wti.Mask true_mask), VLOpFrag),
+ wti.RegClass:$rd, (wti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.ScalarRegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ def : Pat<(riscv_fma_vl
+ (riscv_fneg_vl
+ (wti.Vector (riscv_fpextend_vl_oneuse
+ (vti.Vector (SplatFPOp vti.ScalarRegClass:$rs1)),
+ (vti.Mask true_mask), VLOpFrag)),
+ (vti.Mask true_mask), VLOpFrag),
+ (riscv_fpextend_vl_oneuse (vti.Vector vti.RegClass:$rs2),
+ (vti.Mask true_mask), VLOpFrag),
+ wti.RegClass:$rd, (vti.Mask true_mask), VLOpFrag),
+ (!cast<Instruction>(instruction_name#"_V"#vti.ScalarSuffix#"_"#vti.LMul.MX)
+ wti.RegClass:$rd, vti.ScalarRegClass:$rs1, vti.RegClass:$rs2,
+ GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
+ }
+}
+
//===----------------------------------------------------------------------===//
// Patterns.
//===----------------------------------------------------------------------===//
@@ -1408,6 +1564,12 @@ foreach vti = AllFloatVectors in {
(vti.Mask V0), GPR:$vl, vti.Log2SEW, TAIL_AGNOSTIC)>;
}
+// 14.7. Vector Widening Floating-Point Fused Multiply-Add Instructions
+defm : VPatWidenFPMulAccVL_VV_VF<"PseudoVFWMACC">;
+defm : VPatWidenFPNegMulAccVL_VV_VF<"PseudoVFWNMACC">;
+defm : VPatWidenFPMulSacVL_VV_VF<"PseudoVFWMSAC">;
+defm : VPatWidenFPNegMulSacVL_VV_VF<"PseudoVFWNMSAC">;
+
// 14.11. Vector Floating-Point MIN/MAX Instructions
defm : VPatBinaryFPVL_VV_VF<riscv_fminnum_vl, "PseudoVFMIN">;
defm : VPatBinaryFPVL_VV_VF<riscv_fmaxnum_vl, "PseudoVFMAX">;
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfwmacc.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfwmacc.ll
new file mode 100644
index 0000000000000..ce2415220a149
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfwmacc.ll
@@ -0,0 +1,1312 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+experimental-zvfh,+v -target-abi=ilp32d \
+; RUN: -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+experimental-zvfh,+v -target-abi=lp64d \
+; RUN: -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s
+
+declare <1 x float> @llvm.fma.v1f32(<1 x float>, <1 x float>, <1 x float>)
+
+define <1 x float> @vfwmacc_vv_nxv1f32(<1 x float> %va, <1 x half> %vb, <1 x half> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %vc to <1 x float>
+ %vf = call <1 x float> @llvm.fma.v1f32(<1 x float> %vd, <1 x float> %ve, <1 x float> %va)
+ ret <1 x float> %vf
+}
+
+define <1 x float> @vfwmacc_vf_nxv1f32(<1 x float> %va, <1 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x half> poison, half %c, i32 0
+ %splat = shufflevector <1 x half> %head, <1 x half> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %splat to <1 x float>
+ %vf = call <1 x float> @llvm.fma.v1f32(<1 x float> %vd, <1 x float> %ve, <1 x float> %va)
+ ret <1 x float> %vf
+}
+
+define <1 x float> @vfwnmacc_vv_nxv1f32(<1 x float> %va, <1 x half> %vb, <1 x half> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %vc to <1 x float>
+ %vf = fneg <1 x float> %va
+ %vg = fneg <1 x float> %vd
+ %vh = call <1 x float> @llvm.fma.v1f32(<1 x float> %vg, <1 x float> %ve, <1 x float> %vf)
+ ret <1 x float> %vh
+}
+
+define <1 x float> @vfwnmacc_vf_nxv1f32(<1 x float> %va, <1 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x half> poison, half %c, i32 0
+ %splat = shufflevector <1 x half> %head, <1 x half> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %splat to <1 x float>
+ %vf = fneg <1 x float> %va
+ %vg = fneg <1 x float> %vd
+ %vh = call <1 x float> @llvm.fma.v1f32(<1 x float> %vg, <1 x float> %ve, <1 x float> %vf)
+ ret <1 x float> %vh
+}
+
+define <1 x float> @vfwnmacc_fv_nxv1f32(<1 x float> %va, <1 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x half> poison, half %c, i32 0
+ %splat = shufflevector <1 x half> %head, <1 x half> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %splat to <1 x float>
+ %vf = fneg <1 x float> %va
+ %vg = fneg <1 x float> %ve
+ %vh = call <1 x float> @llvm.fma.v1f32(<1 x float> %vd, <1 x float> %vg, <1 x float> %vf)
+ ret <1 x float> %vh
+}
+
+define <1 x float> @vfwmsac_vv_nxv1f32(<1 x float> %va, <1 x half> %vb, <1 x half> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %vc to <1 x float>
+ %vf = fneg <1 x float> %va
+ %vg = call <1 x float> @llvm.fma.v1f32(<1 x float> %vd, <1 x float> %ve, <1 x float> %vf)
+ ret <1 x float> %vg
+}
+
+define <1 x float> @vfwmsac_vf_nxv1f32(<1 x float> %va, <1 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x half> poison, half %c, i32 0
+ %splat = shufflevector <1 x half> %head, <1 x half> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %splat to <1 x float>
+ %vf = fneg <1 x float> %va
+ %vg = call <1 x float> @llvm.fma.v1f32(<1 x float> %vd, <1 x float> %ve, <1 x float> %vf)
+ ret <1 x float> %vg
+}
+
+define <1 x float> @vfwnmsac_vv_nxv1f32(<1 x float> %va, <1 x half> %vb, <1 x half> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %vc to <1 x float>
+ %vf = fneg <1 x float> %vd
+ %vg = call <1 x float> @llvm.fma.v1f32(<1 x float> %vf, <1 x float> %ve, <1 x float> %va)
+ ret <1 x float> %vg
+}
+
+define <1 x float> @vfwnmsac_vf_nxv1f32(<1 x float> %va, <1 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x half> poison, half %c, i32 0
+ %splat = shufflevector <1 x half> %head, <1 x half> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %splat to <1 x float>
+ %vf = fneg <1 x float> %vd
+ %vg = call <1 x float> @llvm.fma.v1f32(<1 x float> %vf, <1 x float> %ve, <1 x float> %va)
+ ret <1 x float> %vg
+}
+
+define <1 x float> @vfwnmsac_fv_nxv1f32(<1 x float> %va, <1 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv1f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x half> poison, half %c, i32 0
+ %splat = shufflevector <1 x half> %head, <1 x half> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x half> %vb to <1 x float>
+ %ve = fpext <1 x half> %splat to <1 x float>
+ %vf = fneg <1 x float> %ve
+ %vg = call <1 x float> @llvm.fma.v1f32(<1 x float> %vd, <1 x float> %vf, <1 x float> %va)
+ ret <1 x float> %vg
+}
+
+declare <2 x float> @llvm.fma.v2f32(<2 x float>, <2 x float>, <2 x float>)
+
+define <2 x float> @vfwmacc_vv_nxv2f32(<2 x float> %va, <2 x half> %vb, <2 x half> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %vc to <2 x float>
+ %vf = call <2 x float> @llvm.fma.v2f32(<2 x float> %vd, <2 x float> %ve, <2 x float> %va)
+ ret <2 x float> %vf
+}
+
+define <2 x float> @vfwmacc_vf_nxv2f32(<2 x float> %va, <2 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x half> poison, half %c, i32 0
+ %splat = shufflevector <2 x half> %head, <2 x half> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %splat to <2 x float>
+ %vf = call <2 x float> @llvm.fma.v2f32(<2 x float> %vd, <2 x float> %ve, <2 x float> %va)
+ ret <2 x float> %vf
+}
+
+define <2 x float> @vfwnmacc_vv_nxv2f32(<2 x float> %va, <2 x half> %vb, <2 x half> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %vc to <2 x float>
+ %vf = fneg <2 x float> %va
+ %vg = fneg <2 x float> %vd
+ %vh = call <2 x float> @llvm.fma.v2f32(<2 x float> %vg, <2 x float> %ve, <2 x float> %vf)
+ ret <2 x float> %vh
+}
+
+define <2 x float> @vfwnmacc_vf_nxv2f32(<2 x float> %va, <2 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x half> poison, half %c, i32 0
+ %splat = shufflevector <2 x half> %head, <2 x half> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %splat to <2 x float>
+ %vf = fneg <2 x float> %va
+ %vg = fneg <2 x float> %vd
+ %vh = call <2 x float> @llvm.fma.v2f32(<2 x float> %vg, <2 x float> %ve, <2 x float> %vf)
+ ret <2 x float> %vh
+}
+
+define <2 x float> @vfwnmacc_fv_nxv2f32(<2 x float> %va, <2 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x half> poison, half %c, i32 0
+ %splat = shufflevector <2 x half> %head, <2 x half> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %splat to <2 x float>
+ %vf = fneg <2 x float> %va
+ %vg = fneg <2 x float> %ve
+ %vh = call <2 x float> @llvm.fma.v2f32(<2 x float> %vd, <2 x float> %vg, <2 x float> %vf)
+ ret <2 x float> %vh
+}
+
+define <2 x float> @vfwmsac_vv_nxv2f32(<2 x float> %va, <2 x half> %vb, <2 x half> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %vc to <2 x float>
+ %vf = fneg <2 x float> %va
+ %vg = call <2 x float> @llvm.fma.v2f32(<2 x float> %vd, <2 x float> %ve, <2 x float> %vf)
+ ret <2 x float> %vg
+}
+
+define <2 x float> @vfwmsac_vf_nxv2f32(<2 x float> %va, <2 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x half> poison, half %c, i32 0
+ %splat = shufflevector <2 x half> %head, <2 x half> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %splat to <2 x float>
+ %vf = fneg <2 x float> %va
+ %vg = call <2 x float> @llvm.fma.v2f32(<2 x float> %vd, <2 x float> %ve, <2 x float> %vf)
+ ret <2 x float> %vg
+}
+
+define <2 x float> @vfwnmsac_vv_nxv2f32(<2 x float> %va, <2 x half> %vb, <2 x half> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %vc to <2 x float>
+ %vf = fneg <2 x float> %vd
+ %vg = call <2 x float> @llvm.fma.v2f32(<2 x float> %vf, <2 x float> %ve, <2 x float> %va)
+ ret <2 x float> %vg
+}
+
+define <2 x float> @vfwnmsac_vf_nxv2f32(<2 x float> %va, <2 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x half> poison, half %c, i32 0
+ %splat = shufflevector <2 x half> %head, <2 x half> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %splat to <2 x float>
+ %vf = fneg <2 x float> %vd
+ %vg = call <2 x float> @llvm.fma.v2f32(<2 x float> %vf, <2 x float> %ve, <2 x float> %va)
+ ret <2 x float> %vg
+}
+
+define <2 x float> @vfwnmsac_fv_nxv2f32(<2 x float> %va, <2 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv2f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e16, mf4, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x half> poison, half %c, i32 0
+ %splat = shufflevector <2 x half> %head, <2 x half> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x half> %vb to <2 x float>
+ %ve = fpext <2 x half> %splat to <2 x float>
+ %vf = fneg <2 x float> %ve
+ %vg = call <2 x float> @llvm.fma.v2f32(<2 x float> %vd, <2 x float> %vf, <2 x float> %va)
+ ret <2 x float> %vg
+}
+
+
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>)
+
+define <4 x float> @vfwmacc_vv_nxv4f32(<4 x float> %va, <4 x half> %vb, <4 x half> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %vc to <4 x float>
+ %vf = call <4 x float> @llvm.fma.v4f32(<4 x float> %vd, <4 x float> %ve, <4 x float> %va)
+ ret <4 x float> %vf
+}
+
+define <4 x float> @vfwmacc_vf_nxv4f32(<4 x float> %va, <4 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <4 x half> poison, half %c, i32 0
+ %splat = shufflevector <4 x half> %head, <4 x half> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %splat to <4 x float>
+ %vf = call <4 x float> @llvm.fma.v4f32(<4 x float> %vd, <4 x float> %ve, <4 x float> %va)
+ ret <4 x float> %vf
+}
+
+define <4 x float> @vfwnmacc_vv_nxv4f32(<4 x float> %va, <4 x half> %vb, <4 x half> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %vc to <4 x float>
+ %vf = fneg <4 x float> %va
+ %vg = fneg <4 x float> %vd
+ %vh = call <4 x float> @llvm.fma.v4f32(<4 x float> %vg, <4 x float> %ve, <4 x float> %vf)
+ ret <4 x float> %vh
+}
+
+define <4 x float> @vfwnmacc_vf_nxv4f32(<4 x float> %va, <4 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <4 x half> poison, half %c, i32 0
+ %splat = shufflevector <4 x half> %head, <4 x half> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %splat to <4 x float>
+ %vf = fneg <4 x float> %va
+ %vg = fneg <4 x float> %vd
+ %vh = call <4 x float> @llvm.fma.v4f32(<4 x float> %vg, <4 x float> %ve, <4 x float> %vf)
+ ret <4 x float> %vh
+}
+
+define <4 x float> @vfwnmacc_fv_nxv4f32(<4 x float> %va, <4 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <4 x half> poison, half %c, i32 0
+ %splat = shufflevector <4 x half> %head, <4 x half> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %splat to <4 x float>
+ %vf = fneg <4 x float> %va
+ %vg = fneg <4 x float> %ve
+ %vh = call <4 x float> @llvm.fma.v4f32(<4 x float> %vd, <4 x float> %vg, <4 x float> %vf)
+ ret <4 x float> %vh
+}
+
+define <4 x float> @vfwmsac_vv_nxv4f32(<4 x float> %va, <4 x half> %vb, <4 x half> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %vc to <4 x float>
+ %vf = fneg <4 x float> %va
+ %vg = call <4 x float> @llvm.fma.v4f32(<4 x float> %vd, <4 x float> %ve, <4 x float> %vf)
+ ret <4 x float> %vg
+}
+
+define <4 x float> @vfwmsac_vf_nxv4f32(<4 x float> %va, <4 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <4 x half> poison, half %c, i32 0
+ %splat = shufflevector <4 x half> %head, <4 x half> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %splat to <4 x float>
+ %vf = fneg <4 x float> %va
+ %vg = call <4 x float> @llvm.fma.v4f32(<4 x float> %vd, <4 x float> %ve, <4 x float> %vf)
+ ret <4 x float> %vg
+}
+
+define <4 x float> @vfwnmsac_vv_nxv4f32(<4 x float> %va, <4 x half> %vb, <4 x half> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %vc to <4 x float>
+ %vf = fneg <4 x float> %vd
+ %vg = call <4 x float> @llvm.fma.v4f32(<4 x float> %vf, <4 x float> %ve, <4 x float> %va)
+ ret <4 x float> %vg
+}
+
+define <4 x float> @vfwnmsac_vf_nxv4f32(<4 x float> %va, <4 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <4 x half> poison, half %c, i32 0
+ %splat = shufflevector <4 x half> %head, <4 x half> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %splat to <4 x float>
+ %vf = fneg <4 x float> %vd
+ %vg = call <4 x float> @llvm.fma.v4f32(<4 x float> %vf, <4 x float> %ve, <4 x float> %va)
+ ret <4 x float> %vg
+}
+
+define <4 x float> @vfwnmsac_fv_nxv4f32(<4 x float> %va, <4 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv4f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <4 x half> poison, half %c, i32 0
+ %splat = shufflevector <4 x half> %head, <4 x half> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x half> %vb to <4 x float>
+ %ve = fpext <4 x half> %splat to <4 x float>
+ %vf = fneg <4 x float> %ve
+ %vg = call <4 x float> @llvm.fma.v4f32(<4 x float> %vd, <4 x float> %vf, <4 x float> %va)
+ ret <4 x float> %vg
+}
+
+declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>)
+
+define <8 x float> @vfwmacc_vv_nxv8f32(<8 x float> %va, <8 x half> %vb, <8 x half> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %vc to <8 x float>
+ %vf = call <8 x float> @llvm.fma.v8f32(<8 x float> %vd, <8 x float> %ve, <8 x float> %va)
+ ret <8 x float> %vf
+}
+
+define <8 x float> @vfwmacc_vf_nxv8f32(<8 x float> %va, <8 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <8 x half> poison, half %c, i32 0
+ %splat = shufflevector <8 x half> %head, <8 x half> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %splat to <8 x float>
+ %vf = call <8 x float> @llvm.fma.v8f32(<8 x float> %vd, <8 x float> %ve, <8 x float> %va)
+ ret <8 x float> %vf
+}
+
+define <8 x float> @vfwnmacc_vv_nxv8f32(<8 x float> %va, <8 x half> %vb, <8 x half> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %vc to <8 x float>
+ %vf = fneg <8 x float> %va
+ %vg = fneg <8 x float> %vd
+ %vh = call <8 x float> @llvm.fma.v8f32(<8 x float> %vg, <8 x float> %ve, <8 x float> %vf)
+ ret <8 x float> %vh
+}
+
+define <8 x float> @vfwnmacc_vf_nxv8f32(<8 x float> %va, <8 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <8 x half> poison, half %c, i32 0
+ %splat = shufflevector <8 x half> %head, <8 x half> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %splat to <8 x float>
+ %vf = fneg <8 x float> %va
+ %vg = fneg <8 x float> %vd
+ %vh = call <8 x float> @llvm.fma.v8f32(<8 x float> %vg, <8 x float> %ve, <8 x float> %vf)
+ ret <8 x float> %vh
+}
+
+define <8 x float> @vfwnmacc_fv_nxv8f32(<8 x float> %va, <8 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <8 x half> poison, half %c, i32 0
+ %splat = shufflevector <8 x half> %head, <8 x half> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %splat to <8 x float>
+ %vf = fneg <8 x float> %va
+ %vg = fneg <8 x float> %ve
+ %vh = call <8 x float> @llvm.fma.v8f32(<8 x float> %vd, <8 x float> %vg, <8 x float> %vf)
+ ret <8 x float> %vh
+}
+
+define <8 x float> @vfwmsac_vv_nxv8f32(<8 x float> %va, <8 x half> %vb, <8 x half> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %vc to <8 x float>
+ %vf = fneg <8 x float> %va
+ %vg = call <8 x float> @llvm.fma.v8f32(<8 x float> %vd, <8 x float> %ve, <8 x float> %vf)
+ ret <8 x float> %vg
+}
+
+define <8 x float> @vfwmsac_vf_nxv8f32(<8 x float> %va, <8 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <8 x half> poison, half %c, i32 0
+ %splat = shufflevector <8 x half> %head, <8 x half> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %splat to <8 x float>
+ %vf = fneg <8 x float> %va
+ %vg = call <8 x float> @llvm.fma.v8f32(<8 x float> %vd, <8 x float> %ve, <8 x float> %vf)
+ ret <8 x float> %vg
+}
+
+define <8 x float> @vfwnmsac_vv_nxv8f32(<8 x float> %va, <8 x half> %vb, <8 x half> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %vc to <8 x float>
+ %vf = fneg <8 x float> %vd
+ %vg = call <8 x float> @llvm.fma.v8f32(<8 x float> %vf, <8 x float> %ve, <8 x float> %va)
+ ret <8 x float> %vg
+}
+
+define <8 x float> @vfwnmsac_vf_nxv8f32(<8 x float> %va, <8 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <8 x half> poison, half %c, i32 0
+ %splat = shufflevector <8 x half> %head, <8 x half> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %splat to <8 x float>
+ %vf = fneg <8 x float> %vd
+ %vg = call <8 x float> @llvm.fma.v8f32(<8 x float> %vf, <8 x float> %ve, <8 x float> %va)
+ ret <8 x float> %vg
+}
+
+define <8 x float> @vfwnmsac_fv_nxv8f32(<8 x float> %va, <8 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv8f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <8 x half> poison, half %c, i32 0
+ %splat = shufflevector <8 x half> %head, <8 x half> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x half> %vb to <8 x float>
+ %ve = fpext <8 x half> %splat to <8 x float>
+ %vf = fneg <8 x float> %ve
+ %vg = call <8 x float> @llvm.fma.v8f32(<8 x float> %vd, <8 x float> %vf, <8 x float> %va)
+ ret <8 x float> %vg
+}
+
+declare <16 x float> @llvm.fma.v16f32(<16 x float>, <16 x float>, <16 x float>)
+
+define <16 x float> @vfwmacc_vv_nxv16f32(<16 x float> %va, <16 x half> %vb, <16 x half> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %vc to <16 x float>
+ %vf = call <16 x float> @llvm.fma.v16f32(<16 x float> %vd, <16 x float> %ve, <16 x float> %va)
+ ret <16 x float> %vf
+}
+
+define <16 x float> @vfwmacc_vf_nxv16f32(<16 x float> %va, <16 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <16 x half> poison, half %c, i32 0
+ %splat = shufflevector <16 x half> %head, <16 x half> poison, <16 x i32> zeroinitializer
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %splat to <16 x float>
+ %vf = call <16 x float> @llvm.fma.v16f32(<16 x float> %vd, <16 x float> %ve, <16 x float> %va)
+ ret <16 x float> %vf
+}
+
+define <16 x float> @vfwnmacc_vv_nxv16f32(<16 x float> %va, <16 x half> %vb, <16 x half> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %vc to <16 x float>
+ %vf = fneg <16 x float> %va
+ %vg = fneg <16 x float> %vd
+ %vh = call <16 x float> @llvm.fma.v16f32(<16 x float> %vg, <16 x float> %ve, <16 x float> %vf)
+ ret <16 x float> %vh
+}
+
+define <16 x float> @vfwnmacc_vf_nxv16f32(<16 x float> %va, <16 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <16 x half> poison, half %c, i32 0
+ %splat = shufflevector <16 x half> %head, <16 x half> poison, <16 x i32> zeroinitializer
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %splat to <16 x float>
+ %vf = fneg <16 x float> %va
+ %vg = fneg <16 x float> %vd
+ %vh = call <16 x float> @llvm.fma.v16f32(<16 x float> %vg, <16 x float> %ve, <16 x float> %vf)
+ ret <16 x float> %vh
+}
+
+define <16 x float> @vfwnmacc_fv_nxv16f32(<16 x float> %va, <16 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <16 x half> poison, half %c, i32 0
+ %splat = shufflevector <16 x half> %head, <16 x half> poison, <16 x i32> zeroinitializer
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %splat to <16 x float>
+ %vf = fneg <16 x float> %va
+ %vg = fneg <16 x float> %ve
+ %vh = call <16 x float> @llvm.fma.v16f32(<16 x float> %vd, <16 x float> %vg, <16 x float> %vf)
+ ret <16 x float> %vh
+}
+
+define <16 x float> @vfwmsac_vv_nxv16f32(<16 x float> %va, <16 x half> %vb, <16 x half> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %vc to <16 x float>
+ %vf = fneg <16 x float> %va
+ %vg = call <16 x float> @llvm.fma.v16f32(<16 x float> %vd, <16 x float> %ve, <16 x float> %vf)
+ ret <16 x float> %vg
+}
+
+define <16 x float> @vfwmsac_vf_nxv16f32(<16 x float> %va, <16 x half> %vb, half %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <16 x half> poison, half %c, i32 0
+ %splat = shufflevector <16 x half> %head, <16 x half> poison, <16 x i32> zeroinitializer
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %splat to <16 x float>
+ %vf = fneg <16 x float> %va
+ %vg = call <16 x float> @llvm.fma.v16f32(<16 x float> %vd, <16 x float> %ve, <16 x float> %vf)
+ ret <16 x float> %vg
+}
+
+define <16 x float> @vfwnmsac_vv_nxv16f32(<16 x float> %va, <16 x half> %vb, <16 x half> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %vc to <16 x float>
+ %vf = fneg <16 x float> %vd
+ %vg = call <16 x float> @llvm.fma.v16f32(<16 x float> %vf, <16 x float> %ve, <16 x float> %va)
+ ret <16 x float> %vg
+}
+
+define <16 x float> @vfwnmsac_vf_nxv16f32(<16 x float> %va, <16 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <16 x half> poison, half %c, i32 0
+ %splat = shufflevector <16 x half> %head, <16 x half> poison, <16 x i32> zeroinitializer
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %splat to <16 x float>
+ %vf = fneg <16 x float> %vd
+ %vg = call <16 x float> @llvm.fma.v16f32(<16 x float> %vf, <16 x float> %ve, <16 x float> %va)
+ ret <16 x float> %vg
+}
+
+define <16 x float> @vfwnmsac_fv_nxv16f32(<16 x float> %va, <16 x half> %vb, half %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv16f32:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <16 x half> poison, half %c, i32 0
+ %splat = shufflevector <16 x half> %head, <16 x half> poison, <16 x i32> zeroinitializer
+ %vd = fpext <16 x half> %vb to <16 x float>
+ %ve = fpext <16 x half> %splat to <16 x float>
+ %vf = fneg <16 x float> %ve
+ %vg = call <16 x float> @llvm.fma.v16f32(<16 x float> %vd, <16 x float> %vf, <16 x float> %va)
+ ret <16 x float> %vg
+}
+
+declare <1 x double> @llvm.fma.v1f64(<1 x double>, <1 x double>, <1 x double>)
+
+define <1 x double> @vfwmacc_vv_nxv1f64(<1 x double> %va, <1 x float> %vb, <1 x float> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %vc to <1 x double>
+ %vf = call <1 x double> @llvm.fma.v1f64(<1 x double> %vd, <1 x double> %ve, <1 x double> %va)
+ ret <1 x double> %vf
+}
+
+define <1 x double> @vfwmacc_vf_nxv1f64(<1 x double> %va, <1 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x float> poison, float %c, i32 0
+ %splat = shufflevector <1 x float> %head, <1 x float> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %splat to <1 x double>
+ %vf = call <1 x double> @llvm.fma.v1f64(<1 x double> %vd, <1 x double> %ve, <1 x double> %va)
+ ret <1 x double> %vf
+}
+
+define <1 x double> @vfwnmacc_vv_nxv1f64(<1 x double> %va, <1 x float> %vb, <1 x float> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %vc to <1 x double>
+ %vf = fneg <1 x double> %va
+ %vg = fneg <1 x double> %vd
+ %vh = call <1 x double> @llvm.fma.v1f64(<1 x double> %vg, <1 x double> %ve, <1 x double> %vf)
+ ret <1 x double> %vh
+}
+
+define <1 x double> @vfwnmacc_vf_nxv1f64(<1 x double> %va, <1 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x float> poison, float %c, i32 0
+ %splat = shufflevector <1 x float> %head, <1 x float> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %splat to <1 x double>
+ %vf = fneg <1 x double> %va
+ %vg = fneg <1 x double> %vd
+ %vh = call <1 x double> @llvm.fma.v1f64(<1 x double> %vg, <1 x double> %ve, <1 x double> %vf)
+ ret <1 x double> %vh
+}
+
+define <1 x double> @vfwnmacc_fv_nxv1f64(<1 x double> %va, <1 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x float> poison, float %c, i32 0
+ %splat = shufflevector <1 x float> %head, <1 x float> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %splat to <1 x double>
+ %vf = fneg <1 x double> %va
+ %vg = fneg <1 x double> %ve
+ %vh = call <1 x double> @llvm.fma.v1f64(<1 x double> %vd, <1 x double> %vg, <1 x double> %vf)
+ ret <1 x double> %vh
+}
+
+define <1 x double> @vfwmsac_vv_nxv1f64(<1 x double> %va, <1 x float> %vb, <1 x float> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %vc to <1 x double>
+ %vf = fneg <1 x double> %va
+ %vg = call <1 x double> @llvm.fma.v1f64(<1 x double> %vd, <1 x double> %ve, <1 x double> %vf)
+ ret <1 x double> %vg
+}
+
+define <1 x double> @vfwmsac_vf_nxv1f64(<1 x double> %va, <1 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x float> poison, float %c, i32 0
+ %splat = shufflevector <1 x float> %head, <1 x float> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %splat to <1 x double>
+ %vf = fneg <1 x double> %va
+ %vg = call <1 x double> @llvm.fma.v1f64(<1 x double> %vd, <1 x double> %ve, <1 x double> %vf)
+ ret <1 x double> %vg
+}
+
+define <1 x double> @vfwnmsac_vv_nxv1f64(<1 x double> %va, <1 x float> %vb, <1 x float> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %vc to <1 x double>
+ %vf = fneg <1 x double> %vd
+ %vg = call <1 x double> @llvm.fma.v1f64(<1 x double> %vf, <1 x double> %ve, <1 x double> %va)
+ ret <1 x double> %vg
+}
+
+define <1 x double> @vfwnmsac_vf_nxv1f64(<1 x double> %va, <1 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x float> poison, float %c, i32 0
+ %splat = shufflevector <1 x float> %head, <1 x float> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %splat to <1 x double>
+ %vf = fneg <1 x double> %vd
+ %vg = call <1 x double> @llvm.fma.v1f64(<1 x double> %vf, <1 x double> %ve, <1 x double> %va)
+ ret <1 x double> %vg
+}
+
+define <1 x double> @vfwnmsac_fv_nxv1f64(<1 x double> %va, <1 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv1f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 1, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <1 x float> poison, float %c, i32 0
+ %splat = shufflevector <1 x float> %head, <1 x float> poison, <1 x i32> zeroinitializer
+ %vd = fpext <1 x float> %vb to <1 x double>
+ %ve = fpext <1 x float> %splat to <1 x double>
+ %vf = fneg <1 x double> %ve
+ %vg = call <1 x double> @llvm.fma.v1f64(<1 x double> %vd, <1 x double> %vf, <1 x double> %va)
+ ret <1 x double> %vg
+}
+
+declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>)
+
+define <2 x double> @vfwmacc_vv_nxv2f64(<2 x double> %va, <2 x float> %vb, <2 x float> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %vc to <2 x double>
+ %vf = call <2 x double> @llvm.fma.v2f64(<2 x double> %vd, <2 x double> %ve, <2 x double> %va)
+ ret <2 x double> %vf
+}
+
+define <2 x double> @vfwmacc_vf_nxv2f64(<2 x double> %va, <2 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x float> poison, float %c, i32 0
+ %splat = shufflevector <2 x float> %head, <2 x float> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %splat to <2 x double>
+ %vf = call <2 x double> @llvm.fma.v2f64(<2 x double> %vd, <2 x double> %ve, <2 x double> %va)
+ ret <2 x double> %vf
+}
+
+define <2 x double> @vfwnmacc_vv_nxv2f64(<2 x double> %va, <2 x float> %vb, <2 x float> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %vc to <2 x double>
+ %vf = fneg <2 x double> %va
+ %vg = fneg <2 x double> %vd
+ %vh = call <2 x double> @llvm.fma.v2f64(<2 x double> %vg, <2 x double> %ve, <2 x double> %vf)
+ ret <2 x double> %vh
+}
+
+define <2 x double> @vfwnmacc_vf_nxv2f64(<2 x double> %va, <2 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x float> poison, float %c, i32 0
+ %splat = shufflevector <2 x float> %head, <2 x float> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %splat to <2 x double>
+ %vf = fneg <2 x double> %va
+ %vg = fneg <2 x double> %vd
+ %vh = call <2 x double> @llvm.fma.v2f64(<2 x double> %vg, <2 x double> %ve, <2 x double> %vf)
+ ret <2 x double> %vh
+}
+
+define <2 x double> @vfwnmacc_fv_nxv2f64(<2 x double> %va, <2 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x float> poison, float %c, i32 0
+ %splat = shufflevector <2 x float> %head, <2 x float> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %splat to <2 x double>
+ %vf = fneg <2 x double> %va
+ %vg = fneg <2 x double> %ve
+ %vh = call <2 x double> @llvm.fma.v2f64(<2 x double> %vd, <2 x double> %vg, <2 x double> %vf)
+ ret <2 x double> %vh
+}
+
+define <2 x double> @vfwmsac_vv_nxv2f64(<2 x double> %va, <2 x float> %vb, <2 x float> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %vc to <2 x double>
+ %vf = fneg <2 x double> %va
+ %vg = call <2 x double> @llvm.fma.v2f64(<2 x double> %vd, <2 x double> %ve, <2 x double> %vf)
+ ret <2 x double> %vg
+}
+
+define <2 x double> @vfwmsac_vf_nxv2f64(<2 x double> %va, <2 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x float> poison, float %c, i32 0
+ %splat = shufflevector <2 x float> %head, <2 x float> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %splat to <2 x double>
+ %vf = fneg <2 x double> %va
+ %vg = call <2 x double> @llvm.fma.v2f64(<2 x double> %vd, <2 x double> %ve, <2 x double> %vf)
+ ret <2 x double> %vg
+}
+
+define <2 x double> @vfwnmsac_vv_nxv2f64(<2 x double> %va, <2 x float> %vb, <2 x float> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v9, v10
+; CHECK-NEXT: ret
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %vc to <2 x double>
+ %vf = fneg <2 x double> %vd
+ %vg = call <2 x double> @llvm.fma.v2f64(<2 x double> %vf, <2 x double> %ve, <2 x double> %va)
+ ret <2 x double> %vg
+}
+
+define <2 x double> @vfwnmsac_vf_nxv2f64(<2 x double> %va, <2 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x float> poison, float %c, i32 0
+ %splat = shufflevector <2 x float> %head, <2 x float> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %splat to <2 x double>
+ %vf = fneg <2 x double> %vd
+ %vg = call <2 x double> @llvm.fma.v2f64(<2 x double> %vf, <2 x double> %ve, <2 x double> %va)
+ ret <2 x double> %vg
+}
+
+define <2 x double> @vfwnmsac_fv_nxv2f64(<2 x double> %va, <2 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv2f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
+; CHECK-NEXT: ret
+ %head = insertelement <2 x float> poison, float %c, i32 0
+ %splat = shufflevector <2 x float> %head, <2 x float> poison, <2 x i32> zeroinitializer
+ %vd = fpext <2 x float> %vb to <2 x double>
+ %ve = fpext <2 x float> %splat to <2 x double>
+ %vf = fneg <2 x double> %ve
+ %vg = call <2 x double> @llvm.fma.v2f64(<2 x double> %vd, <2 x double> %vf, <2 x double> %va)
+ ret <2 x double> %vg
+}
+
+
+declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>)
+
+define <4 x double> @vfwmacc_vv_nxv4f64(<4 x double> %va, <4 x float> %vb, <4 x float> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %vc to <4 x double>
+ %vf = call <4 x double> @llvm.fma.v4f64(<4 x double> %vd, <4 x double> %ve, <4 x double> %va)
+ ret <4 x double> %vf
+}
+
+define <4 x double> @vfwmacc_vf_nxv4f64(<4 x double> %va, <4 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <4 x float> poison, float %c, i32 0
+ %splat = shufflevector <4 x float> %head, <4 x float> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %splat to <4 x double>
+ %vf = call <4 x double> @llvm.fma.v4f64(<4 x double> %vd, <4 x double> %ve, <4 x double> %va)
+ ret <4 x double> %vf
+}
+
+define <4 x double> @vfwnmacc_vv_nxv4f64(<4 x double> %va, <4 x float> %vb, <4 x float> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %vc to <4 x double>
+ %vf = fneg <4 x double> %va
+ %vg = fneg <4 x double> %vd
+ %vh = call <4 x double> @llvm.fma.v4f64(<4 x double> %vg, <4 x double> %ve, <4 x double> %vf)
+ ret <4 x double> %vh
+}
+
+define <4 x double> @vfwnmacc_vf_nxv4f64(<4 x double> %va, <4 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <4 x float> poison, float %c, i32 0
+ %splat = shufflevector <4 x float> %head, <4 x float> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %splat to <4 x double>
+ %vf = fneg <4 x double> %va
+ %vg = fneg <4 x double> %vd
+ %vh = call <4 x double> @llvm.fma.v4f64(<4 x double> %vg, <4 x double> %ve, <4 x double> %vf)
+ ret <4 x double> %vh
+}
+
+define <4 x double> @vfwnmacc_fv_nxv4f64(<4 x double> %va, <4 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <4 x float> poison, float %c, i32 0
+ %splat = shufflevector <4 x float> %head, <4 x float> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %splat to <4 x double>
+ %vf = fneg <4 x double> %va
+ %vg = fneg <4 x double> %ve
+ %vh = call <4 x double> @llvm.fma.v4f64(<4 x double> %vd, <4 x double> %vg, <4 x double> %vf)
+ ret <4 x double> %vh
+}
+
+define <4 x double> @vfwmsac_vv_nxv4f64(<4 x double> %va, <4 x float> %vb, <4 x float> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %vc to <4 x double>
+ %vf = fneg <4 x double> %va
+ %vg = call <4 x double> @llvm.fma.v4f64(<4 x double> %vd, <4 x double> %ve, <4 x double> %vf)
+ ret <4 x double> %vg
+}
+
+define <4 x double> @vfwmsac_vf_nxv4f64(<4 x double> %va, <4 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <4 x float> poison, float %c, i32 0
+ %splat = shufflevector <4 x float> %head, <4 x float> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %splat to <4 x double>
+ %vf = fneg <4 x double> %va
+ %vg = call <4 x double> @llvm.fma.v4f64(<4 x double> %vd, <4 x double> %ve, <4 x double> %vf)
+ ret <4 x double> %vg
+}
+
+define <4 x double> @vfwnmsac_vv_nxv4f64(<4 x double> %va, <4 x float> %vb, <4 x float> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v10, v11
+; CHECK-NEXT: ret
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %vc to <4 x double>
+ %vf = fneg <4 x double> %vd
+ %vg = call <4 x double> @llvm.fma.v4f64(<4 x double> %vf, <4 x double> %ve, <4 x double> %va)
+ ret <4 x double> %vg
+}
+
+define <4 x double> @vfwnmsac_vf_nxv4f64(<4 x double> %va, <4 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <4 x float> poison, float %c, i32 0
+ %splat = shufflevector <4 x float> %head, <4 x float> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %splat to <4 x double>
+ %vf = fneg <4 x double> %vd
+ %vg = call <4 x double> @llvm.fma.v4f64(<4 x double> %vf, <4 x double> %ve, <4 x double> %va)
+ ret <4 x double> %vg
+}
+
+define <4 x double> @vfwnmsac_fv_nxv4f64(<4 x double> %va, <4 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv4f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v10
+; CHECK-NEXT: ret
+ %head = insertelement <4 x float> poison, float %c, i32 0
+ %splat = shufflevector <4 x float> %head, <4 x float> poison, <4 x i32> zeroinitializer
+ %vd = fpext <4 x float> %vb to <4 x double>
+ %ve = fpext <4 x float> %splat to <4 x double>
+ %vf = fneg <4 x double> %ve
+ %vg = call <4 x double> @llvm.fma.v4f64(<4 x double> %vd, <4 x double> %vf, <4 x double> %va)
+ ret <4 x double> %vg
+}
+
+declare <8 x double> @llvm.fma.v8f64(<8 x double>, <8 x double>, <8 x double>)
+
+define <8 x double> @vfwmacc_vv_nxv8f64(<8 x double> %va, <8 x float> %vb, <8 x float> %vc) {
+; CHECK-LABEL: vfwmacc_vv_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwmacc.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %vc to <8 x double>
+ %vf = call <8 x double> @llvm.fma.v8f64(<8 x double> %vd, <8 x double> %ve, <8 x double> %va)
+ ret <8 x double> %vf
+}
+
+define <8 x double> @vfwmacc_vf_nxv8f64(<8 x double> %va, <8 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmacc_vf_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwmacc.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <8 x float> poison, float %c, i32 0
+ %splat = shufflevector <8 x float> %head, <8 x float> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %splat to <8 x double>
+ %vf = call <8 x double> @llvm.fma.v8f64(<8 x double> %vd, <8 x double> %ve, <8 x double> %va)
+ ret <8 x double> %vf
+}
+
+define <8 x double> @vfwnmacc_vv_nxv8f64(<8 x double> %va, <8 x float> %vb, <8 x float> %vc) {
+; CHECK-LABEL: vfwnmacc_vv_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwnmacc.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %vc to <8 x double>
+ %vf = fneg <8 x double> %va
+ %vg = fneg <8 x double> %vd
+ %vh = call <8 x double> @llvm.fma.v8f64(<8 x double> %vg, <8 x double> %ve, <8 x double> %vf)
+ ret <8 x double> %vh
+}
+
+define <8 x double> @vfwnmacc_vf_nxv8f64(<8 x double> %va, <8 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_vf_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <8 x float> poison, float %c, i32 0
+ %splat = shufflevector <8 x float> %head, <8 x float> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %splat to <8 x double>
+ %vf = fneg <8 x double> %va
+ %vg = fneg <8 x double> %vd
+ %vh = call <8 x double> @llvm.fma.v8f64(<8 x double> %vg, <8 x double> %ve, <8 x double> %vf)
+ ret <8 x double> %vh
+}
+
+define <8 x double> @vfwnmacc_fv_nxv8f64(<8 x double> %va, <8 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmacc_fv_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwnmacc.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <8 x float> poison, float %c, i32 0
+ %splat = shufflevector <8 x float> %head, <8 x float> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %splat to <8 x double>
+ %vf = fneg <8 x double> %va
+ %vg = fneg <8 x double> %ve
+ %vh = call <8 x double> @llvm.fma.v8f64(<8 x double> %vd, <8 x double> %vg, <8 x double> %vf)
+ ret <8 x double> %vh
+}
+
+define <8 x double> @vfwmsac_vv_nxv8f64(<8 x double> %va, <8 x float> %vb, <8 x float> %vc) {
+; CHECK-LABEL: vfwmsac_vv_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwmsac.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %vc to <8 x double>
+ %vf = fneg <8 x double> %va
+ %vg = call <8 x double> @llvm.fma.v8f64(<8 x double> %vd, <8 x double> %ve, <8 x double> %vf)
+ ret <8 x double> %vg
+}
+
+define <8 x double> @vfwmsac_vf_nxv8f64(<8 x double> %va, <8 x float> %vb, float %c) {
+; CHECK-LABEL: vfwmsac_vf_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwmsac.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <8 x float> poison, float %c, i32 0
+ %splat = shufflevector <8 x float> %head, <8 x float> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %splat to <8 x double>
+ %vf = fneg <8 x double> %va
+ %vg = call <8 x double> @llvm.fma.v8f64(<8 x double> %vd, <8 x double> %ve, <8 x double> %vf)
+ ret <8 x double> %vg
+}
+
+define <8 x double> @vfwnmsac_vv_nxv8f64(<8 x double> %va, <8 x float> %vb, <8 x float> %vc) {
+; CHECK-LABEL: vfwnmsac_vv_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwnmsac.vv v8, v12, v14
+; CHECK-NEXT: ret
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %vc to <8 x double>
+ %vf = fneg <8 x double> %vd
+ %vg = call <8 x double> @llvm.fma.v8f64(<8 x double> %vf, <8 x double> %ve, <8 x double> %va)
+ ret <8 x double> %vg
+}
+
+define <8 x double> @vfwnmsac_vf_nxv8f64(<8 x double> %va, <8 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_vf_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <8 x float> poison, float %c, i32 0
+ %splat = shufflevector <8 x float> %head, <8 x float> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %splat to <8 x double>
+ %vf = fneg <8 x double> %vd
+ %vg = call <8 x double> @llvm.fma.v8f64(<8 x double> %vf, <8 x double> %ve, <8 x double> %va)
+ ret <8 x double> %vg
+}
+
+define <8 x double> @vfwnmsac_fv_nxv8f64(<8 x double> %va, <8 x float> %vb, float %c) {
+; CHECK-LABEL: vfwnmsac_fv_nxv8f64:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, mu
+; CHECK-NEXT: vfwnmsac.vf v8, fa0, v12
+; CHECK-NEXT: ret
+ %head = insertelement <8 x float> poison, float %c, i32 0
+ %splat = shufflevector <8 x float> %head, <8 x float> poison, <8 x i32> zeroinitializer
+ %vd = fpext <8 x float> %vb to <8 x double>
+ %ve = fpext <8 x float> %splat to <8 x double>
+ %vf = fneg <8 x double> %ve
+ %vg = call <8 x double> @llvm.fma.v8f64(<8 x double> %vd, <8 x double> %vf, <8 x double> %va)
+ ret <8 x double> %vg
+}
More information about the llvm-commits
mailing list