derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
2696 lines
125 KiB
LLVM
2696 lines
125 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+zvfh,+zfbfmin,+zvfbfmin,+v -target-abi=ilp32d \
|
|
; RUN: -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,ZVFBFMIN,ZVFH
|
|
; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+zvfh,+zfbfmin,+zvfbfmin,+v -target-abi=lp64d \
|
|
; RUN: -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,ZVFBFMIN,ZVFH
|
|
; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+zvfhmin,+zfbfmin,+zvfbfmin,+v -target-abi=ilp32d \
|
|
; RUN: -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,ZVFHMIN,ZVFBFMIN
|
|
; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+zvfhmin,+zfbfmin,+zvfbfmin,+v -target-abi=lp64d \
|
|
; RUN: -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,ZVFHMIN,ZVFBFMIN
|
|
; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+zvfhmin,+experimental-zvfbfa,+v \
|
|
; RUN: -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,ZVFHMIN,ZVFBFA
|
|
; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+zvfhmin,+experimental-zvfbfa,+v \
|
|
; RUN: -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,ZVFHMIN,ZVFBFA
|
|
|
|
define <vscale x 1 x float> @vfwmaccbf16_vv_nxv1f32(<vscale x 1 x float> %a, <vscale x 1 x bfloat> %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vv_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %b.ext, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwmaccbf16_vf_nxv1f32(<vscale x 1 x float> %a, bfloat %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vf_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 1 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 1 x bfloat> %b.head, <vscale x 1 x bfloat> poison, <vscale x 1 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b.splat to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %b.ext, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmaccbf16_vv_nxv2f32(<vscale x 2 x float> %a, <vscale x 2 x bfloat> %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vv_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %b.ext, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmaccbf16_vf_nxv2f32(<vscale x 2 x float> %a, bfloat %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vf_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 2 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 2 x bfloat> %b.head, <vscale x 2 x bfloat> poison, <vscale x 2 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b.splat to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %b.ext, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmaccbf16_vv_nxv4f32(<vscale x 4 x float> %a, <vscale x 4 x bfloat> %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v14, v11
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v12, v14
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vv_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vv v8, v10, v11
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %b.ext, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmaccbf16_vf_nxv4f32(<vscale x 4 x float> %a, bfloat %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vf v8, fa5, v12
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vf_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vf v8, fa0, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 4 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 4 x bfloat> %b.head, <vscale x 4 x bfloat> poison, <vscale x 4 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b.splat to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %b.ext, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmaccbf16_vv_nxv8f32(<vscale x 8 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v20, v14
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v16, v20
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vv_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vv v8, v12, v14
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %b.ext, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmaccbf16_vf_nxv8f32(<vscale x 8 x float> %a, bfloat %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vf v8, fa5, v16
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vf_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vf v8, fa0, v12
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 8 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 8 x bfloat> %b.head, <vscale x 8 x bfloat> poison, <vscale x 8 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b.splat to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %b.ext, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmaccbf16_vv_nxv16f32(<vscale x 16 x float> %a, <vscale x 16 x bfloat> %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v0, v20
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v24, v0
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vv_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vv v8, v16, v20
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %b.ext, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmaccbf16_vf_nxv16f32(<vscale x 16 x float> %a, bfloat %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vf v8, fa5, v24
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmaccbf16_vf_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmacc.vf v8, fa0, v16
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 16 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 16 x bfloat> %b.head, <vscale x 16 x bfloat> poison, <vscale x 16 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b.splat to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %b.ext, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwmsacbf16_vv_nxv1f32(<vscale x 1 x float> %a, <vscale x 1 x bfloat> %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vv_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vv_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%neg.a = fneg <vscale x 1 x float> %a
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %b.ext, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %neg.a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwmsacbf16_vf_nxv1f32(<vscale x 1 x float> %a, bfloat %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vf_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vf_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 1 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 1 x bfloat> %b.head, <vscale x 1 x bfloat> poison, <vscale x 1 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b.splat to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%neg.a = fneg <vscale x 1 x float> %a
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %b.ext, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %neg.a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmsacbf16_vv_nxv2f32(<vscale x 2 x float> %a, <vscale x 2 x bfloat> %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vv_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vv_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%neg.a = fneg <vscale x 2 x float> %a
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %b.ext, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %neg.a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmsacbf16_vf_nxv2f32(<vscale x 2 x float> %a, bfloat %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vf_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vf_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 2 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 2 x bfloat> %b.head, <vscale x 2 x bfloat> poison, <vscale x 2 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b.splat to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%neg.a = fneg <vscale x 2 x float> %a
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %b.ext, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %neg.a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmsacbf16_vv_nxv4f32(<vscale x 4 x float> %a, <vscale x 4 x bfloat> %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vv_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v14, v11
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vv v8, v12, v14
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vv_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vv v8, v10, v11
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%neg.a = fneg <vscale x 4 x float> %a
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %b.ext, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %neg.a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmsacbf16_vf_nxv4f32(<vscale x 4 x float> %a, bfloat %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vf_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vf v8, fa5, v12
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vf_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vf v8, fa0, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 4 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 4 x bfloat> %b.head, <vscale x 4 x bfloat> poison, <vscale x 4 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b.splat to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%neg.a = fneg <vscale x 4 x float> %a
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %b.ext, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %neg.a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmsacbf16_vv_nxv8f32(<vscale x 8 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vv_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v20, v14
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vv v8, v16, v20
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vv_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vv v8, v12, v14
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%neg.a = fneg <vscale x 8 x float> %a
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %b.ext, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %neg.a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmsacbf16_vf_nxv8f32(<vscale x 8 x float> %a, bfloat %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vf_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vf v8, fa5, v16
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vf_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vf v8, fa0, v12
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 8 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 8 x bfloat> %b.head, <vscale x 8 x bfloat> poison, <vscale x 8 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b.splat to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%neg.a = fneg <vscale x 8 x float> %a
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %b.ext, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %neg.a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmsacbf16_vv_nxv16f32(<vscale x 16 x float> %a, <vscale x 16 x bfloat> %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vv_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v0, v20
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vv v8, v24, v0
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vv_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vv v8, v16, v20
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%neg.a = fneg <vscale x 16 x float> %a
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %b.ext, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %neg.a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmsacbf16_vf_nxv16f32(<vscale x 16 x float> %a, bfloat %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwmsacbf16_vf_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmsac.vf v8, fa5, v24
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwmsacbf16_vf_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwmsac.vf v8, fa0, v16
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 16 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 16 x bfloat> %b.head, <vscale x 16 x bfloat> poison, <vscale x 16 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b.splat to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%neg.a = fneg <vscale x 16 x float> %a
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %b.ext, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %neg.a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmaccbf16_vv_nxv1f32(<vscale x 1 x float> %a, <vscale x 1 x bfloat> %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vv_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vv_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%neg.a = fneg <vscale x 1 x float> %a
|
|
%neg.b = fneg <vscale x 1 x float> %b.ext
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %neg.b, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %neg.a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmaccbf16_vf_nxv1f32(<vscale x 1 x float> %a, bfloat %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vf_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vf_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 1 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 1 x bfloat> %b.head, <vscale x 1 x bfloat> poison, <vscale x 1 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b.splat to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%neg.a = fneg <vscale x 1 x float> %a
|
|
%neg.b = fneg <vscale x 1 x float> %b.ext
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %neg.b, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %neg.a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmaccbf16_vv_nxv2f32(<vscale x 2 x float> %a, <vscale x 2 x bfloat> %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vv_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vv_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%neg.a = fneg <vscale x 2 x float> %a
|
|
%neg.b = fneg <vscale x 2 x float> %b.ext
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %neg.b, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %neg.a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmaccbf16_vf_nxv2f32(<vscale x 2 x float> %a, bfloat %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vf_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vf_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 2 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 2 x bfloat> %b.head, <vscale x 2 x bfloat> poison, <vscale x 2 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b.splat to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%neg.a = fneg <vscale x 2 x float> %a
|
|
%neg.b = fneg <vscale x 2 x float> %b.ext
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %neg.b, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %neg.a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmaccbf16_vv_nxv4f32(<vscale x 4 x float> %a, <vscale x 4 x bfloat> %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vv_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v14, v11
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vv v8, v12, v14
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vv_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vv v8, v10, v11
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%neg.a = fneg <vscale x 4 x float> %a
|
|
%neg.b = fneg <vscale x 4 x float> %b.ext
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %neg.b, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %neg.a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmaccbf16_vf_nxv4f32(<vscale x 4 x float> %a, bfloat %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vf_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vf v8, fa5, v12
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vf_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vf v8, fa0, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 4 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 4 x bfloat> %b.head, <vscale x 4 x bfloat> poison, <vscale x 4 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b.splat to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%neg.a = fneg <vscale x 4 x float> %a
|
|
%neg.b = fneg <vscale x 4 x float> %b.ext
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %neg.b, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %neg.a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmaccbf16_vv_nxv8f32(<vscale x 8 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vv_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v20, v14
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vv v8, v16, v20
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vv_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vv v8, v12, v14
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%neg.a = fneg <vscale x 8 x float> %a
|
|
%neg.b = fneg <vscale x 8 x float> %b.ext
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %neg.b, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %neg.a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmaccbf16_vf_nxv8f32(<vscale x 8 x float> %a, bfloat %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vf_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vf v8, fa5, v16
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vf_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vf v8, fa0, v12
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 8 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 8 x bfloat> %b.head, <vscale x 8 x bfloat> poison, <vscale x 8 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b.splat to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%neg.a = fneg <vscale x 8 x float> %a
|
|
%neg.b = fneg <vscale x 8 x float> %b.ext
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %neg.b, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %neg.a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmaccbf16_vv_nxv16f32(<vscale x 16 x float> %a, <vscale x 16 x bfloat> %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vv_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v0, v20
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vv v8, v24, v0
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vv_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vv v8, v16, v20
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%neg.a = fneg <vscale x 16 x float> %a
|
|
%neg.b = fneg <vscale x 16 x float> %b.ext
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %neg.b, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %neg.a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmaccbf16_vf_nxv16f32(<vscale x 16 x float> %a, bfloat %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmaccbf16_vf_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmacc.vf v8, fa5, v24
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmaccbf16_vf_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmacc.vf v8, fa0, v16
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 16 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 16 x bfloat> %b.head, <vscale x 16 x bfloat> poison, <vscale x 16 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b.splat to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%neg.a = fneg <vscale x 16 x float> %a
|
|
%neg.b = fneg <vscale x 16 x float> %b.ext
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %neg.b, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %neg.a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmsacbf16_vv_nxv1f32(<vscale x 1 x float> %a, <vscale x 1 x bfloat> %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vv_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vv_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%neg.b = fneg <vscale x 1 x float> %b.ext
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %neg.b, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmsacbf16_vf_nxv1f32(<vscale x 1 x float> %a, bfloat %b, <vscale x 1 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vf_nxv1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vf_nxv1f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 1 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 1 x bfloat> %b.head, <vscale x 1 x bfloat> poison, <vscale x 1 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 1 x bfloat> %b.splat to <vscale x 1 x float>
|
|
%c.ext = fpext <vscale x 1 x bfloat> %c to <vscale x 1 x float>
|
|
%neg.b = fneg <vscale x 1 x float> %b.ext
|
|
%res = call <vscale x 1 x float> @llvm.fma.nxv1f32(<vscale x 1 x float> %neg.b, <vscale x 1 x float> %c.ext, <vscale x 1 x float> %a)
|
|
ret <vscale x 1 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmsacbf16_vv_nxv2f32(<vscale x 2 x float> %a, <vscale x 2 x bfloat> %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vv_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v9
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vv v8, v11, v9
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vv_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vv v8, v9, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%neg.b = fneg <vscale x 2 x float> %b.ext
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %neg.b, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmsacbf16_vf_nxv2f32(<vscale x 2 x float> %a, bfloat %b, <vscale x 2 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vf_nxv2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vf_nxv2f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, mf2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 2 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 2 x bfloat> %b.head, <vscale x 2 x bfloat> poison, <vscale x 2 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 2 x bfloat> %b.splat to <vscale x 2 x float>
|
|
%c.ext = fpext <vscale x 2 x bfloat> %c to <vscale x 2 x float>
|
|
%neg.b = fneg <vscale x 2 x float> %b.ext
|
|
%res = call <vscale x 2 x float> @llvm.fma.nxv2f32(<vscale x 2 x float> %neg.b, <vscale x 2 x float> %c.ext, <vscale x 2 x float> %a)
|
|
ret <vscale x 2 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmsacbf16_vv_nxv4f32(<vscale x 4 x float> %a, <vscale x 4 x bfloat> %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vv_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v14, v11
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vv v8, v12, v14
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vv_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vv v8, v10, v11
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%neg.b = fneg <vscale x 4 x float> %b.ext
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %neg.b, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmsacbf16_vf_nxv4f32(<vscale x 4 x float> %a, bfloat %b, <vscale x 4 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vf_nxv4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vf v8, fa5, v12
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vf_nxv4f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m1, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vf v8, fa0, v10
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 4 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 4 x bfloat> %b.head, <vscale x 4 x bfloat> poison, <vscale x 4 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 4 x bfloat> %b.splat to <vscale x 4 x float>
|
|
%c.ext = fpext <vscale x 4 x bfloat> %c to <vscale x 4 x float>
|
|
%neg.b = fneg <vscale x 4 x float> %b.ext
|
|
%res = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> %neg.b, <vscale x 4 x float> %c.ext, <vscale x 4 x float> %a)
|
|
ret <vscale x 4 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmsacbf16_vv_nxv8f32(<vscale x 8 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vv_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v20, v14
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vv v8, v16, v20
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vv_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vv v8, v12, v14
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%neg.b = fneg <vscale x 8 x float> %b.ext
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %neg.b, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmsacbf16_vf_nxv8f32(<vscale x 8 x float> %a, bfloat %b, <vscale x 8 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vf_nxv8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v12
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vf v8, fa5, v16
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vf_nxv8f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m2, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vf v8, fa0, v12
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 8 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 8 x bfloat> %b.head, <vscale x 8 x bfloat> poison, <vscale x 8 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 8 x bfloat> %b.splat to <vscale x 8 x float>
|
|
%c.ext = fpext <vscale x 8 x bfloat> %c to <vscale x 8 x float>
|
|
%neg.b = fneg <vscale x 8 x float> %b.ext
|
|
%res = call <vscale x 8 x float> @llvm.fma.nxv8f32(<vscale x 8 x float> %neg.b, <vscale x 8 x float> %c.ext, <vscale x 8 x float> %a)
|
|
ret <vscale x 8 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmsacbf16_vv_nxv16f32(<vscale x 16 x float> %a, <vscale x 16 x bfloat> %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vv_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v0, v20
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vv v8, v24, v0
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vv_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vv v8, v16, v20
|
|
; ZVFBFA-NEXT: ret
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%neg.b = fneg <vscale x 16 x float> %b.ext
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %neg.b, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmsacbf16_vf_nxv16f32(<vscale x 16 x float> %a, bfloat %b, <vscale x 16 x bfloat> %c) {
|
|
; ZVFBFMIN-LABEL: vfwnmsacbf16_vf_nxv16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fcvt.s.bf16 fa5, fa0
|
|
; ZVFBFMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v16
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfnmsac.vf v8, fa5, v24
|
|
; ZVFBFMIN-NEXT: ret
|
|
;
|
|
; ZVFBFA-LABEL: vfwnmsacbf16_vf_nxv16f32:
|
|
; ZVFBFA: # %bb.0:
|
|
; ZVFBFA-NEXT: vsetvli a0, zero, e16alt, m4, ta, ma
|
|
; ZVFBFA-NEXT: vfwnmsac.vf v8, fa0, v16
|
|
; ZVFBFA-NEXT: ret
|
|
%b.head = insertelement <vscale x 16 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <vscale x 16 x bfloat> %b.head, <vscale x 16 x bfloat> poison, <vscale x 16 x i32> zeroinitializer
|
|
%b.ext = fpext <vscale x 16 x bfloat> %b.splat to <vscale x 16 x float>
|
|
%c.ext = fpext <vscale x 16 x bfloat> %c to <vscale x 16 x float>
|
|
%neg.b = fneg <vscale x 16 x float> %b.ext
|
|
%res = call <vscale x 16 x float> @llvm.fma.nxv16f32(<vscale x 16 x float> %neg.b, <vscale x 16 x float> %c.ext, <vscale x 16 x float> %a)
|
|
ret <vscale x 16 x float> %res
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwmacc_vv_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, <vscale x 1 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmacc_vv_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vv_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %vc to <vscale x 1 x float>
|
|
%vf = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vd, <vscale x 1 x float> %ve, <vscale x 1 x float> %va)
|
|
ret <vscale x 1 x float> %vf
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwmacc_vf_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmacc_vf_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vf_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 1 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x half> %head, <vscale x 1 x half> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %splat to <vscale x 1 x float>
|
|
%vf = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vd, <vscale x 1 x float> %ve, <vscale x 1 x float> %va)
|
|
ret <vscale x 1 x float> %vf
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmacc_vv_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, <vscale x 1 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmacc_vv_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vv_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %vc to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %va
|
|
%vg = fneg <vscale x 1 x float> %vd
|
|
%vh = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vg, <vscale x 1 x float> %ve, <vscale x 1 x float> %vf)
|
|
ret <vscale x 1 x float> %vh
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmacc_vf_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_vf_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vf_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 1 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x half> %head, <vscale x 1 x half> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %splat to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %va
|
|
%vg = fneg <vscale x 1 x float> %vd
|
|
%vh = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vg, <vscale x 1 x float> %ve, <vscale x 1 x float> %vf)
|
|
ret <vscale x 1 x float> %vh
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmacc_fv_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_fv_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_fv_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 1 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x half> %head, <vscale x 1 x half> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %splat to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %va
|
|
%vg = fneg <vscale x 1 x float> %ve
|
|
%vh = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vd, <vscale x 1 x float> %vg, <vscale x 1 x float> %vf)
|
|
ret <vscale x 1 x float> %vh
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwmsac_vv_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, <vscale x 1 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmsac_vv_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vv_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %vc to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %va
|
|
%vg = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vd, <vscale x 1 x float> %ve, <vscale x 1 x float> %vf)
|
|
ret <vscale x 1 x float> %vg
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwmsac_vf_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmsac_vf_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vf_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 1 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x half> %head, <vscale x 1 x half> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %splat to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %va
|
|
%vg = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vd, <vscale x 1 x float> %ve, <vscale x 1 x float> %vf)
|
|
ret <vscale x 1 x float> %vg
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmsac_vv_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, <vscale x 1 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmsac_vv_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vv_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %vc to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %vd
|
|
%vg = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vf, <vscale x 1 x float> %ve, <vscale x 1 x float> %va)
|
|
ret <vscale x 1 x float> %vg
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmsac_vf_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_vf_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vf_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 1 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x half> %head, <vscale x 1 x half> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %splat to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %vd
|
|
%vg = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vf, <vscale x 1 x float> %ve, <vscale x 1 x float> %va)
|
|
ret <vscale x 1 x float> %vg
|
|
}
|
|
|
|
define <vscale x 1 x float> @vfwnmsac_fv_nxv1f32(<vscale x 1 x float> %va, <vscale x 1 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_fv_nxv1f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_fv_nxv1f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 1 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x half> %head, <vscale x 1 x half> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x half> %vb to <vscale x 1 x float>
|
|
%ve = fpext <vscale x 1 x half> %splat to <vscale x 1 x float>
|
|
%vf = fneg <vscale x 1 x float> %ve
|
|
%vg = call <vscale x 1 x float> @llvm.fma.v1f32(<vscale x 1 x float> %vd, <vscale x 1 x float> %vf, <vscale x 1 x float> %va)
|
|
ret <vscale x 1 x float> %vg
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmacc_vv_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, <vscale x 2 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmacc_vv_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vv_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %vc to <vscale x 2 x float>
|
|
%vf = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vd, <vscale x 2 x float> %ve, <vscale x 2 x float> %va)
|
|
ret <vscale x 2 x float> %vf
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmacc_vf_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmacc_vf_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vf_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 2 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x half> %head, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %splat to <vscale x 2 x float>
|
|
%vf = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vd, <vscale x 2 x float> %ve, <vscale x 2 x float> %va)
|
|
ret <vscale x 2 x float> %vf
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmacc_vv_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, <vscale x 2 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmacc_vv_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vv_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %vc to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %va
|
|
%vg = fneg <vscale x 2 x float> %vd
|
|
%vh = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vg, <vscale x 2 x float> %ve, <vscale x 2 x float> %vf)
|
|
ret <vscale x 2 x float> %vh
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmacc_vf_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_vf_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vf_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 2 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x half> %head, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %splat to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %va
|
|
%vg = fneg <vscale x 2 x float> %vd
|
|
%vh = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vg, <vscale x 2 x float> %ve, <vscale x 2 x float> %vf)
|
|
ret <vscale x 2 x float> %vh
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmacc_fv_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_fv_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_fv_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 2 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x half> %head, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %splat to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %va
|
|
%vg = fneg <vscale x 2 x float> %ve
|
|
%vh = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vd, <vscale x 2 x float> %vg, <vscale x 2 x float> %vf)
|
|
ret <vscale x 2 x float> %vh
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmsac_vv_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, <vscale x 2 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmsac_vv_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vv_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %vc to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %va
|
|
%vg = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vd, <vscale x 2 x float> %ve, <vscale x 2 x float> %vf)
|
|
ret <vscale x 2 x float> %vg
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwmsac_vf_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmsac_vf_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vf_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 2 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x half> %head, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %splat to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %va
|
|
%vg = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vd, <vscale x 2 x float> %ve, <vscale x 2 x float> %vf)
|
|
ret <vscale x 2 x float> %vg
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmsac_vv_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, <vscale x 2 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmsac_vv_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vv v8, v9, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vv_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v9, v10
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vv v8, v11, v9
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %vc to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %vd
|
|
%vg = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vf, <vscale x 2 x float> %ve, <vscale x 2 x float> %va)
|
|
ret <vscale x 2 x float> %vg
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmsac_vf_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_vf_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vf_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 2 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x half> %head, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %splat to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %vd
|
|
%vg = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vf, <vscale x 2 x float> %ve, <vscale x 2 x float> %va)
|
|
ret <vscale x 2 x float> %vg
|
|
}
|
|
|
|
define <vscale x 2 x float> @vfwnmsac_fv_nxv2f32(<vscale x 2 x float> %va, <vscale x 2 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_fv_nxv2f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_fv_nxv2f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, mf2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v10, v9
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v10
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 2 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x half> %head, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x half> %vb to <vscale x 2 x float>
|
|
%ve = fpext <vscale x 2 x half> %splat to <vscale x 2 x float>
|
|
%vf = fneg <vscale x 2 x float> %ve
|
|
%vg = call <vscale x 2 x float> @llvm.fma.v2f32(<vscale x 2 x float> %vd, <vscale x 2 x float> %vf, <vscale x 2 x float> %va)
|
|
ret <vscale x 2 x float> %vg
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmacc_vv_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, <vscale x 4 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmacc_vv_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vv v8, v10, v11
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vv_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v14, v11
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vv v8, v12, v14
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %vc to <vscale x 4 x float>
|
|
%vf = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vd, <vscale x 4 x float> %ve, <vscale x 4 x float> %va)
|
|
ret <vscale x 4 x float> %vf
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmacc_vf_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmacc_vf_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vf v8, fa0, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vf_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vf v8, fa5, v12
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 4 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x half> %head, <vscale x 4 x half> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %splat to <vscale x 4 x float>
|
|
%vf = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vd, <vscale x 4 x float> %ve, <vscale x 4 x float> %va)
|
|
ret <vscale x 4 x float> %vf
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmacc_vv_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, <vscale x 4 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmacc_vv_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vv v8, v10, v11
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vv_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v14, v11
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vv v8, v12, v14
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %vc to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %va
|
|
%vg = fneg <vscale x 4 x float> %vd
|
|
%vh = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vg, <vscale x 4 x float> %ve, <vscale x 4 x float> %vf)
|
|
ret <vscale x 4 x float> %vh
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmacc_vf_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_vf_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vf_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v12
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 4 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x half> %head, <vscale x 4 x half> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %splat to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %va
|
|
%vg = fneg <vscale x 4 x float> %vd
|
|
%vh = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vg, <vscale x 4 x float> %ve, <vscale x 4 x float> %vf)
|
|
ret <vscale x 4 x float> %vh
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmacc_fv_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_fv_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_fv_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v12
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 4 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x half> %head, <vscale x 4 x half> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %splat to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %va
|
|
%vg = fneg <vscale x 4 x float> %ve
|
|
%vh = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vd, <vscale x 4 x float> %vg, <vscale x 4 x float> %vf)
|
|
ret <vscale x 4 x float> %vh
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmsac_vv_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, <vscale x 4 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmsac_vv_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vv v8, v10, v11
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vv_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v14, v11
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vv v8, v12, v14
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %vc to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %va
|
|
%vg = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vd, <vscale x 4 x float> %ve, <vscale x 4 x float> %vf)
|
|
ret <vscale x 4 x float> %vg
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwmsac_vf_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmsac_vf_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vf v8, fa0, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vf_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vf v8, fa5, v12
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 4 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x half> %head, <vscale x 4 x half> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %splat to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %va
|
|
%vg = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vd, <vscale x 4 x float> %ve, <vscale x 4 x float> %vf)
|
|
ret <vscale x 4 x float> %vg
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmsac_vv_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, <vscale x 4 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmsac_vv_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vv v8, v10, v11
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vv_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v14, v11
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vv v8, v12, v14
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %vc to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %vd
|
|
%vg = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vf, <vscale x 4 x float> %ve, <vscale x 4 x float> %va)
|
|
ret <vscale x 4 x float> %vg
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmsac_vf_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_vf_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vf_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v12
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 4 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x half> %head, <vscale x 4 x half> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %splat to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %vd
|
|
%vg = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vf, <vscale x 4 x float> %ve, <vscale x 4 x float> %va)
|
|
ret <vscale x 4 x float> %vg
|
|
}
|
|
|
|
define <vscale x 4 x float> @vfwnmsac_fv_nxv4f32(<vscale x 4 x float> %va, <vscale x 4 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_fv_nxv4f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v10
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_fv_nxv4f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m1, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v12, v10
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v12
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 4 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x half> %head, <vscale x 4 x half> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x half> %vb to <vscale x 4 x float>
|
|
%ve = fpext <vscale x 4 x half> %splat to <vscale x 4 x float>
|
|
%vf = fneg <vscale x 4 x float> %ve
|
|
%vg = call <vscale x 4 x float> @llvm.fma.v4f32(<vscale x 4 x float> %vd, <vscale x 4 x float> %vf, <vscale x 4 x float> %va)
|
|
ret <vscale x 4 x float> %vg
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmacc_vv_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, <vscale x 8 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmacc_vv_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vv v8, v12, v14
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vv_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v20, v14
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vv v8, v16, v20
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %vc to <vscale x 8 x float>
|
|
%vf = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vd, <vscale x 8 x float> %ve, <vscale x 8 x float> %va)
|
|
ret <vscale x 8 x float> %vf
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmacc_vf_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmacc_vf_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vf v8, fa0, v12
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vf_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vf v8, fa5, v16
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 8 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x half> %head, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %splat to <vscale x 8 x float>
|
|
%vf = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vd, <vscale x 8 x float> %ve, <vscale x 8 x float> %va)
|
|
ret <vscale x 8 x float> %vf
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmacc_vv_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, <vscale x 8 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmacc_vv_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vv v8, v12, v14
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vv_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v20, v14
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vv v8, v16, v20
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %vc to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %va
|
|
%vg = fneg <vscale x 8 x float> %vd
|
|
%vh = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vg, <vscale x 8 x float> %ve, <vscale x 8 x float> %vf)
|
|
ret <vscale x 8 x float> %vh
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmacc_vf_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_vf_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v12
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vf_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v16
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 8 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x half> %head, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %splat to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %va
|
|
%vg = fneg <vscale x 8 x float> %vd
|
|
%vh = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vg, <vscale x 8 x float> %ve, <vscale x 8 x float> %vf)
|
|
ret <vscale x 8 x float> %vh
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmacc_fv_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_fv_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v12
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_fv_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v16
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 8 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x half> %head, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %splat to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %va
|
|
%vg = fneg <vscale x 8 x float> %ve
|
|
%vh = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vd, <vscale x 8 x float> %vg, <vscale x 8 x float> %vf)
|
|
ret <vscale x 8 x float> %vh
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmsac_vv_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, <vscale x 8 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmsac_vv_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vv v8, v12, v14
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vv_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v20, v14
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vv v8, v16, v20
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %vc to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %va
|
|
%vg = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vd, <vscale x 8 x float> %ve, <vscale x 8 x float> %vf)
|
|
ret <vscale x 8 x float> %vg
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwmsac_vf_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmsac_vf_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vf v8, fa0, v12
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vf_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vf v8, fa5, v16
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 8 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x half> %head, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %splat to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %va
|
|
%vg = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vd, <vscale x 8 x float> %ve, <vscale x 8 x float> %vf)
|
|
ret <vscale x 8 x float> %vg
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmsac_vv_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, <vscale x 8 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmsac_vv_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vv v8, v12, v14
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vv_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v20, v14
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vv v8, v16, v20
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %vc to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %vd
|
|
%vg = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vf, <vscale x 8 x float> %ve, <vscale x 8 x float> %va)
|
|
ret <vscale x 8 x float> %vg
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmsac_vf_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_vf_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v12
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vf_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v16
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 8 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x half> %head, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %splat to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %vd
|
|
%vg = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vf, <vscale x 8 x float> %ve, <vscale x 8 x float> %va)
|
|
ret <vscale x 8 x float> %vg
|
|
}
|
|
|
|
define <vscale x 8 x float> @vfwnmsac_fv_nxv8f32(<vscale x 8 x float> %va, <vscale x 8 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_fv_nxv8f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v12
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_fv_nxv8f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m2, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v16, v12
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v16
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 8 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x half> %head, <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x half> %vb to <vscale x 8 x float>
|
|
%ve = fpext <vscale x 8 x half> %splat to <vscale x 8 x float>
|
|
%vf = fneg <vscale x 8 x float> %ve
|
|
%vg = call <vscale x 8 x float> @llvm.fma.v8f32(<vscale x 8 x float> %vd, <vscale x 8 x float> %vf, <vscale x 8 x float> %va)
|
|
ret <vscale x 8 x float> %vg
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmacc_vv_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, <vscale x 16 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmacc_vv_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vv v8, v16, v20
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vv_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v0, v20
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vv v8, v24, v0
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %vc to <vscale x 16 x float>
|
|
%vf = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vd, <vscale x 16 x float> %ve, <vscale x 16 x float> %va)
|
|
ret <vscale x 16 x float> %vf
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmacc_vf_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmacc_vf_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwmacc.vf v8, fa0, v16
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmacc_vf_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfmacc.vf v8, fa5, v24
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 16 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 16 x half> %head, <vscale x 16 x half> poison, <vscale x 16 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %splat to <vscale x 16 x float>
|
|
%vf = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vd, <vscale x 16 x float> %ve, <vscale x 16 x float> %va)
|
|
ret <vscale x 16 x float> %vf
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmacc_vv_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, <vscale x 16 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmacc_vv_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vv v8, v16, v20
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vv_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v0, v20
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vv v8, v24, v0
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %vc to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %va
|
|
%vg = fneg <vscale x 16 x float> %vd
|
|
%vh = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vg, <vscale x 16 x float> %ve, <vscale x 16 x float> %vf)
|
|
ret <vscale x 16 x float> %vh
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmacc_vf_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_vf_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v16
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_vf_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v24
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 16 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 16 x half> %head, <vscale x 16 x half> poison, <vscale x 16 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %splat to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %va
|
|
%vg = fneg <vscale x 16 x float> %vd
|
|
%vh = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vg, <vscale x 16 x float> %ve, <vscale x 16 x float> %vf)
|
|
ret <vscale x 16 x float> %vh
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmacc_fv_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmacc_fv_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwnmacc.vf v8, fa0, v16
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmacc_fv_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmacc.vf v8, fa5, v24
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 16 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 16 x half> %head, <vscale x 16 x half> poison, <vscale x 16 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %splat to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %va
|
|
%vg = fneg <vscale x 16 x float> %ve
|
|
%vh = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vd, <vscale x 16 x float> %vg, <vscale x 16 x float> %vf)
|
|
ret <vscale x 16 x float> %vh
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmsac_vv_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, <vscale x 16 x half> %vc) {
|
|
; ZVFH-LABEL: vfwmsac_vv_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vv v8, v16, v20
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vv_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v0, v20
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vv v8, v24, v0
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %vc to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %va
|
|
%vg = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vd, <vscale x 16 x float> %ve, <vscale x 16 x float> %vf)
|
|
ret <vscale x 16 x float> %vg
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwmsac_vf_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwmsac_vf_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwmsac.vf v8, fa0, v16
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwmsac_vf_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfmsac.vf v8, fa5, v24
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 16 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 16 x half> %head, <vscale x 16 x half> poison, <vscale x 16 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %splat to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %va
|
|
%vg = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vd, <vscale x 16 x float> %ve, <vscale x 16 x float> %vf)
|
|
ret <vscale x 16 x float> %vg
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmsac_vv_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, <vscale x 16 x half> %vc) {
|
|
; ZVFH-LABEL: vfwnmsac_vv_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vv v8, v16, v20
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vv_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v0, v20
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vv v8, v24, v0
|
|
; ZVFHMIN-NEXT: ret
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %vc to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %vd
|
|
%vg = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vf, <vscale x 16 x float> %ve, <vscale x 16 x float> %va)
|
|
ret <vscale x 16 x float> %vg
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmsac_vf_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_vf_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v16
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_vf_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v24
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 16 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 16 x half> %head, <vscale x 16 x half> poison, <vscale x 16 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %splat to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %vd
|
|
%vg = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vf, <vscale x 16 x float> %ve, <vscale x 16 x float> %va)
|
|
ret <vscale x 16 x float> %vg
|
|
}
|
|
|
|
define <vscale x 16 x float> @vfwnmsac_fv_nxv16f32(<vscale x 16 x float> %va, <vscale x 16 x half> %vb, half %c) {
|
|
; ZVFH-LABEL: vfwnmsac_fv_nxv16f32:
|
|
; ZVFH: # %bb.0:
|
|
; ZVFH-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFH-NEXT: vfwnmsac.vf v8, fa0, v16
|
|
; ZVFH-NEXT: ret
|
|
;
|
|
; ZVFHMIN-LABEL: vfwnmsac_fv_nxv16f32:
|
|
; ZVFHMIN: # %bb.0:
|
|
; ZVFHMIN-NEXT: vsetvli a0, zero, e16, m4, ta, ma
|
|
; ZVFHMIN-NEXT: vfwcvt.f.f.v v24, v16
|
|
; ZVFHMIN-NEXT: fcvt.s.h fa5, fa0
|
|
; ZVFHMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFHMIN-NEXT: vfnmsac.vf v8, fa5, v24
|
|
; ZVFHMIN-NEXT: ret
|
|
%head = insertelement <vscale x 16 x half> poison, half %c, i32 0
|
|
%splat = shufflevector <vscale x 16 x half> %head, <vscale x 16 x half> poison, <vscale x 16 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 16 x half> %vb to <vscale x 16 x float>
|
|
%ve = fpext <vscale x 16 x half> %splat to <vscale x 16 x float>
|
|
%vf = fneg <vscale x 16 x float> %ve
|
|
%vg = call <vscale x 16 x float> @llvm.fma.v16f32(<vscale x 16 x float> %vd, <vscale x 16 x float> %vf, <vscale x 16 x float> %va)
|
|
ret <vscale x 16 x float> %vg
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwmacc_vv_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, <vscale x 1 x float> %vc) {
|
|
; CHECK-LABEL: vfwmacc_vv_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vv v8, v9, v10
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %vc to <vscale x 1 x double>
|
|
%vf = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vd, <vscale x 1 x double> %ve, <vscale x 1 x double> %va)
|
|
ret <vscale x 1 x double> %vf
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwmacc_vf_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmacc_vf_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vf v8, fa0, v9
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 1 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x float> %head, <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %splat to <vscale x 1 x double>
|
|
%vf = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vd, <vscale x 1 x double> %ve, <vscale x 1 x double> %va)
|
|
ret <vscale x 1 x double> %vf
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwnmacc_vv_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, <vscale x 1 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmacc_vv_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vv v8, v9, v10
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %vc to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %va
|
|
%vg = fneg <vscale x 1 x double> %vd
|
|
%vh = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vg, <vscale x 1 x double> %ve, <vscale x 1 x double> %vf)
|
|
ret <vscale x 1 x double> %vh
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwnmacc_vf_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_vf_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 1 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x float> %head, <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %splat to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %va
|
|
%vg = fneg <vscale x 1 x double> %vd
|
|
%vh = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vg, <vscale x 1 x double> %ve, <vscale x 1 x double> %vf)
|
|
ret <vscale x 1 x double> %vh
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwnmacc_fv_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_fv_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v9
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 1 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x float> %head, <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %splat to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %va
|
|
%vg = fneg <vscale x 1 x double> %ve
|
|
%vh = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vd, <vscale x 1 x double> %vg, <vscale x 1 x double> %vf)
|
|
ret <vscale x 1 x double> %vh
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwmsac_vv_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, <vscale x 1 x float> %vc) {
|
|
; CHECK-LABEL: vfwmsac_vv_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vv v8, v9, v10
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %vc to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %va
|
|
%vg = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vd, <vscale x 1 x double> %ve, <vscale x 1 x double> %vf)
|
|
ret <vscale x 1 x double> %vg
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwmsac_vf_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmsac_vf_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vf v8, fa0, v9
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 1 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x float> %head, <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %splat to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %va
|
|
%vg = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vd, <vscale x 1 x double> %ve, <vscale x 1 x double> %vf)
|
|
ret <vscale x 1 x double> %vg
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwnmsac_vv_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, <vscale x 1 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmsac_vv_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vv v8, v9, v10
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %vc to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %vd
|
|
%vg = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vf, <vscale x 1 x double> %ve, <vscale x 1 x double> %va)
|
|
ret <vscale x 1 x double> %vg
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwnmsac_vf_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_vf_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 1 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x float> %head, <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %splat to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %vd
|
|
%vg = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vf, <vscale x 1 x double> %ve, <vscale x 1 x double> %va)
|
|
ret <vscale x 1 x double> %vg
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwnmsac_fv_nxv1f64(<vscale x 1 x double> %va, <vscale x 1 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_fv_nxv1f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v9
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 1 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 1 x float> %head, <vscale x 1 x float> poison, <vscale x 1 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 1 x float> %vb to <vscale x 1 x double>
|
|
%ve = fpext <vscale x 1 x float> %splat to <vscale x 1 x double>
|
|
%vf = fneg <vscale x 1 x double> %ve
|
|
%vg = call <vscale x 1 x double> @llvm.fma.v1f64(<vscale x 1 x double> %vd, <vscale x 1 x double> %vf, <vscale x 1 x double> %va)
|
|
ret <vscale x 1 x double> %vg
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwmacc_vv_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, <vscale x 2 x float> %vc) {
|
|
; CHECK-LABEL: vfwmacc_vv_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vv v8, v10, v11
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %vc to <vscale x 2 x double>
|
|
%vf = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vd, <vscale x 2 x double> %ve, <vscale x 2 x double> %va)
|
|
ret <vscale x 2 x double> %vf
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwmacc_vf_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmacc_vf_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vf v8, fa0, v10
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 2 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x float> %head, <vscale x 2 x float> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %splat to <vscale x 2 x double>
|
|
%vf = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vd, <vscale x 2 x double> %ve, <vscale x 2 x double> %va)
|
|
ret <vscale x 2 x double> %vf
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwnmacc_vv_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, <vscale x 2 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmacc_vv_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vv v8, v10, v11
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %vc to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %va
|
|
%vg = fneg <vscale x 2 x double> %vd
|
|
%vh = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vg, <vscale x 2 x double> %ve, <vscale x 2 x double> %vf)
|
|
ret <vscale x 2 x double> %vh
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwnmacc_vf_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_vf_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v10
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 2 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x float> %head, <vscale x 2 x float> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %splat to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %va
|
|
%vg = fneg <vscale x 2 x double> %vd
|
|
%vh = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vg, <vscale x 2 x double> %ve, <vscale x 2 x double> %vf)
|
|
ret <vscale x 2 x double> %vh
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwnmacc_fv_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_fv_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v10
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 2 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x float> %head, <vscale x 2 x float> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %splat to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %va
|
|
%vg = fneg <vscale x 2 x double> %ve
|
|
%vh = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vd, <vscale x 2 x double> %vg, <vscale x 2 x double> %vf)
|
|
ret <vscale x 2 x double> %vh
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwmsac_vv_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, <vscale x 2 x float> %vc) {
|
|
; CHECK-LABEL: vfwmsac_vv_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vv v8, v10, v11
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %vc to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %va
|
|
%vg = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vd, <vscale x 2 x double> %ve, <vscale x 2 x double> %vf)
|
|
ret <vscale x 2 x double> %vg
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwmsac_vf_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmsac_vf_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vf v8, fa0, v10
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 2 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x float> %head, <vscale x 2 x float> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %splat to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %va
|
|
%vg = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vd, <vscale x 2 x double> %ve, <vscale x 2 x double> %vf)
|
|
ret <vscale x 2 x double> %vg
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwnmsac_vv_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, <vscale x 2 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmsac_vv_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vv v8, v10, v11
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %vc to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %vd
|
|
%vg = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vf, <vscale x 2 x double> %ve, <vscale x 2 x double> %va)
|
|
ret <vscale x 2 x double> %vg
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwnmsac_vf_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_vf_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v10
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 2 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x float> %head, <vscale x 2 x float> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %splat to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %vd
|
|
%vg = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vf, <vscale x 2 x double> %ve, <vscale x 2 x double> %va)
|
|
ret <vscale x 2 x double> %vg
|
|
}
|
|
|
|
define <vscale x 2 x double> @vfwnmsac_fv_nxv2f64(<vscale x 2 x double> %va, <vscale x 2 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_fv_nxv2f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m1, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v10
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 2 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 2 x float> %head, <vscale x 2 x float> poison, <vscale x 2 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 2 x float> %vb to <vscale x 2 x double>
|
|
%ve = fpext <vscale x 2 x float> %splat to <vscale x 2 x double>
|
|
%vf = fneg <vscale x 2 x double> %ve
|
|
%vg = call <vscale x 2 x double> @llvm.fma.v2f64(<vscale x 2 x double> %vd, <vscale x 2 x double> %vf, <vscale x 2 x double> %va)
|
|
ret <vscale x 2 x double> %vg
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwmacc_vv_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, <vscale x 4 x float> %vc) {
|
|
; CHECK-LABEL: vfwmacc_vv_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vv v8, v12, v14
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %vc to <vscale x 4 x double>
|
|
%vf = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vd, <vscale x 4 x double> %ve, <vscale x 4 x double> %va)
|
|
ret <vscale x 4 x double> %vf
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwmacc_vf_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmacc_vf_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vf v8, fa0, v12
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 4 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x float> %head, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %splat to <vscale x 4 x double>
|
|
%vf = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vd, <vscale x 4 x double> %ve, <vscale x 4 x double> %va)
|
|
ret <vscale x 4 x double> %vf
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwnmacc_vv_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, <vscale x 4 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmacc_vv_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vv v8, v12, v14
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %vc to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %va
|
|
%vg = fneg <vscale x 4 x double> %vd
|
|
%vh = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vg, <vscale x 4 x double> %ve, <vscale x 4 x double> %vf)
|
|
ret <vscale x 4 x double> %vh
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwnmacc_vf_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_vf_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v12
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 4 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x float> %head, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %splat to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %va
|
|
%vg = fneg <vscale x 4 x double> %vd
|
|
%vh = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vg, <vscale x 4 x double> %ve, <vscale x 4 x double> %vf)
|
|
ret <vscale x 4 x double> %vh
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwnmacc_fv_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_fv_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v12
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 4 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x float> %head, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %splat to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %va
|
|
%vg = fneg <vscale x 4 x double> %ve
|
|
%vh = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vd, <vscale x 4 x double> %vg, <vscale x 4 x double> %vf)
|
|
ret <vscale x 4 x double> %vh
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwmsac_vv_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, <vscale x 4 x float> %vc) {
|
|
; CHECK-LABEL: vfwmsac_vv_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vv v8, v12, v14
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %vc to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %va
|
|
%vg = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vd, <vscale x 4 x double> %ve, <vscale x 4 x double> %vf)
|
|
ret <vscale x 4 x double> %vg
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwmsac_vf_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmsac_vf_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vf v8, fa0, v12
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 4 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x float> %head, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %splat to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %va
|
|
%vg = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vd, <vscale x 4 x double> %ve, <vscale x 4 x double> %vf)
|
|
ret <vscale x 4 x double> %vg
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwnmsac_vv_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, <vscale x 4 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmsac_vv_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vv v8, v12, v14
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %vc to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %vd
|
|
%vg = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vf, <vscale x 4 x double> %ve, <vscale x 4 x double> %va)
|
|
ret <vscale x 4 x double> %vg
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwnmsac_vf_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_vf_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v12
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 4 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x float> %head, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %splat to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %vd
|
|
%vg = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vf, <vscale x 4 x double> %ve, <vscale x 4 x double> %va)
|
|
ret <vscale x 4 x double> %vg
|
|
}
|
|
|
|
define <vscale x 4 x double> @vfwnmsac_fv_nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_fv_nxv4f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v12
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 4 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 4 x float> %head, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 4 x float> %vb to <vscale x 4 x double>
|
|
%ve = fpext <vscale x 4 x float> %splat to <vscale x 4 x double>
|
|
%vf = fneg <vscale x 4 x double> %ve
|
|
%vg = call <vscale x 4 x double> @llvm.fma.v4f64(<vscale x 4 x double> %vd, <vscale x 4 x double> %vf, <vscale x 4 x double> %va)
|
|
ret <vscale x 4 x double> %vg
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwmacc_vv_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, <vscale x 8 x float> %vc) {
|
|
; CHECK-LABEL: vfwmacc_vv_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vv v8, v16, v20
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %vc to <vscale x 8 x double>
|
|
%vf = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vd, <vscale x 8 x double> %ve, <vscale x 8 x double> %va)
|
|
ret <vscale x 8 x double> %vf
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwmacc_vf_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmacc_vf_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vf v8, fa0, v16
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 8 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x float> %head, <vscale x 8 x float> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %splat to <vscale x 8 x double>
|
|
%vf = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vd, <vscale x 8 x double> %ve, <vscale x 8 x double> %va)
|
|
ret <vscale x 8 x double> %vf
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwnmacc_vv_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, <vscale x 8 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmacc_vv_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vv v8, v16, v20
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %vc to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %va
|
|
%vg = fneg <vscale x 8 x double> %vd
|
|
%vh = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vg, <vscale x 8 x double> %ve, <vscale x 8 x double> %vf)
|
|
ret <vscale x 8 x double> %vh
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwnmacc_vf_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_vf_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v16
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 8 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x float> %head, <vscale x 8 x float> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %splat to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %va
|
|
%vg = fneg <vscale x 8 x double> %vd
|
|
%vh = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vg, <vscale x 8 x double> %ve, <vscale x 8 x double> %vf)
|
|
ret <vscale x 8 x double> %vh
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwnmacc_fv_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmacc_fv_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwnmacc.vf v8, fa0, v16
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 8 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x float> %head, <vscale x 8 x float> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %splat to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %va
|
|
%vg = fneg <vscale x 8 x double> %ve
|
|
%vh = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vd, <vscale x 8 x double> %vg, <vscale x 8 x double> %vf)
|
|
ret <vscale x 8 x double> %vh
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwmsac_vv_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, <vscale x 8 x float> %vc) {
|
|
; CHECK-LABEL: vfwmsac_vv_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vv v8, v16, v20
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %vc to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %va
|
|
%vg = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vd, <vscale x 8 x double> %ve, <vscale x 8 x double> %vf)
|
|
ret <vscale x 8 x double> %vg
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwmsac_vf_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwmsac_vf_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vf v8, fa0, v16
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 8 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x float> %head, <vscale x 8 x float> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %splat to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %va
|
|
%vg = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vd, <vscale x 8 x double> %ve, <vscale x 8 x double> %vf)
|
|
ret <vscale x 8 x double> %vg
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwnmsac_vv_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, <vscale x 8 x float> %vc) {
|
|
; CHECK-LABEL: vfwnmsac_vv_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vv v8, v16, v20
|
|
; CHECK-NEXT: ret
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %vc to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %vd
|
|
%vg = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vf, <vscale x 8 x double> %ve, <vscale x 8 x double> %va)
|
|
ret <vscale x 8 x double> %vg
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwnmsac_vf_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_vf_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v16
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 8 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x float> %head, <vscale x 8 x float> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %splat to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %vd
|
|
%vg = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vf, <vscale x 8 x double> %ve, <vscale x 8 x double> %va)
|
|
ret <vscale x 8 x double> %vg
|
|
}
|
|
|
|
define <vscale x 8 x double> @vfwnmsac_fv_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x float> %vb, float %c) {
|
|
; CHECK-LABEL: vfwnmsac_fv_nxv8f64:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, m4, ta, ma
|
|
; CHECK-NEXT: vfwnmsac.vf v8, fa0, v16
|
|
; CHECK-NEXT: ret
|
|
%head = insertelement <vscale x 8 x float> poison, float %c, i32 0
|
|
%splat = shufflevector <vscale x 8 x float> %head, <vscale x 8 x float> poison, <vscale x 8 x i32> zeroinitializer
|
|
%vd = fpext <vscale x 8 x float> %vb to <vscale x 8 x double>
|
|
%ve = fpext <vscale x 8 x float> %splat to <vscale x 8 x double>
|
|
%vf = fneg <vscale x 8 x double> %ve
|
|
%vg = call <vscale x 8 x double> @llvm.fma.v8f64(<vscale x 8 x double> %vd, <vscale x 8 x double> %vf, <vscale x 8 x double> %va)
|
|
ret <vscale x 8 x double> %vg
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwma_vv_nxv1f64_same_op(<vscale x 1 x float> %va, <vscale x 1 x double> %vb) {
|
|
; CHECK-LABEL: vfwma_vv_nxv1f64_same_op:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwmacc.vv v9, v8, v8
|
|
; CHECK-NEXT: vmv1r.v v8, v9
|
|
; CHECK-NEXT: ret
|
|
%vc = fpext <vscale x 1 x float> %va to <vscale x 1 x double>
|
|
%vd = call <vscale x 1 x double> @llvm.fma(<vscale x 1 x double> %vc, <vscale x 1 x double> %vc, <vscale x 1 x double> %vb)
|
|
ret <vscale x 1 x double> %vd
|
|
}
|
|
|
|
define <vscale x 1 x double> @vfwmsac_vv_nxv1f64_same_op(<vscale x 1 x float> %va, <vscale x 1 x double> %vb) {
|
|
; CHECK-LABEL: vfwmsac_vv_nxv1f64_same_op:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vsetvli a0, zero, e32, mf2, ta, ma
|
|
; CHECK-NEXT: vfwmsac.vv v9, v8, v8
|
|
; CHECK-NEXT: vmv1r.v v8, v9
|
|
; CHECK-NEXT: ret
|
|
%vc = fpext <vscale x 1 x float> %va to <vscale x 1 x double>
|
|
%vd = fneg <vscale x 1 x double> %vb
|
|
%ve = call <vscale x 1 x double> @llvm.fma(<vscale x 1 x double> %vc, <vscale x 1 x double> %vc, <vscale x 1 x double> %vd)
|
|
ret <vscale x 1 x double> %ve
|
|
}
|