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
348 lines
14 KiB
LLVM
348 lines
14 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc < %s -mtriple=riscv32 -mattr=+v,+zvfh,+zvfbfwma -verify-machineinstrs | FileCheck %s --check-prefix=ZVFBFWMA
|
|
; RUN: llc < %s -mtriple=riscv64 -mattr=+v,+zvfh,+zvfbfwma -verify-machineinstrs | FileCheck %s --check-prefix=ZVFBFWMA
|
|
; RUN: llc < %s -mtriple=riscv32 -mattr=+v,+zvfh,+zvfbfmin -verify-machineinstrs | FileCheck %s --check-prefix=ZVFBFMIN
|
|
; RUN: llc < %s -mtriple=riscv64 -mattr=+v,+zvfh,+zvfbfmin -verify-machineinstrs | FileCheck %s --check-prefix=ZVFBFMIN
|
|
|
|
define <1 x float> @vfwmaccbf16_vv_v1f32(<1 x float> %a, <1 x bfloat> %b, <1 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vv_v1f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 1, e16, mf4, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v9, v10
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_v1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 1, 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
|
|
%b.ext = fpext <1 x bfloat> %b to <1 x float>
|
|
%c.ext = fpext <1 x bfloat> %c to <1 x float>
|
|
%res = call <1 x float> @llvm.fma.v1f32(<1 x float> %b.ext, <1 x float> %c.ext, <1 x float> %a)
|
|
ret <1 x float> %res
|
|
}
|
|
|
|
define <1 x float> @vfwmaccbf16_vf_v1f32(<1 x float> %a, bfloat %b, <1 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vf_v1f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: fmv.x.h a0, fa0
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 1, e16, mf4, ta, ma
|
|
; ZVFBFWMA-NEXT: vmv.s.x v10, a0
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v10, v9
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_v1f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fmv.x.w a0, fa0
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 1, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vmv.s.x v10, a0
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v10
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v11, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
%b.head = insertelement <1 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <1 x bfloat> %b.head, <1 x bfloat> poison, <1 x i32> zeroinitializer
|
|
%b.ext = fpext <1 x bfloat> %b.splat to <1 x float>
|
|
%c.ext = fpext <1 x bfloat> %c to <1 x float>
|
|
%res = call <1 x float> @llvm.fma.v1f32(<1 x float> %b.ext, <1 x float> %c.ext, <1 x float> %a)
|
|
ret <1 x float> %res
|
|
}
|
|
|
|
define <2 x float> @vfwmaccbf16_vv_v2f32(<2 x float> %a, <2 x bfloat> %b, <2 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vv_v2f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 2, e16, mf4, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v9, v10
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_v2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 2, 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
|
|
%b.ext = fpext <2 x bfloat> %b to <2 x float>
|
|
%c.ext = fpext <2 x bfloat> %c to <2 x float>
|
|
%res = call <2 x float> @llvm.fma.v2f32(<2 x float> %b.ext, <2 x float> %c.ext, <2 x float> %a)
|
|
ret <2 x float> %res
|
|
}
|
|
|
|
define <2 x float> @vfwmaccbf16_vf_v2f32(<2 x float> %a, bfloat %b, <2 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vf_v2f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: fmv.x.h a0, fa0
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 2, e16, mf4, ta, ma
|
|
; ZVFBFWMA-NEXT: vmv.v.x v10, a0
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v10, v9
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_v2f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fmv.x.w a0, fa0
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 2, e16, mf4, ta, ma
|
|
; ZVFBFMIN-NEXT: vmv.v.x v10, a0
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v10
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v11, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
%b.head = insertelement <2 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <2 x bfloat> %b.head, <2 x bfloat> poison, <2 x i32> zeroinitializer
|
|
%b.ext = fpext <2 x bfloat> %b.splat to <2 x float>
|
|
%c.ext = fpext <2 x bfloat> %c to <2 x float>
|
|
%res = call <2 x float> @llvm.fma.v2f32(<2 x float> %b.ext, <2 x float> %c.ext, <2 x float> %a)
|
|
ret <2 x float> %res
|
|
}
|
|
|
|
define <4 x float> @vfwmaccbf16_vv_v4f32(<4 x float> %a, <4 x bfloat> %b, <4 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vv_v4f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v9, v10
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_v4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 4, 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
|
|
%b.ext = fpext <4 x bfloat> %b to <4 x float>
|
|
%c.ext = fpext <4 x bfloat> %c to <4 x float>
|
|
%res = call <4 x float> @llvm.fma.v4f32(<4 x float> %b.ext, <4 x float> %c.ext, <4 x float> %a)
|
|
ret <4 x float> %res
|
|
}
|
|
|
|
define <4 x float> @vfwmaccbf16_vf_v4f32(<4 x float> %a, bfloat %b, <4 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vf_v4f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: fmv.x.h a0, fa0
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
|
|
; ZVFBFWMA-NEXT: vmv.v.x v10, a0
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v10, v9
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_v4f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fmv.x.w a0, fa0
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vmv.v.x v10, a0
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v11, v10
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v11, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
%b.head = insertelement <4 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <4 x bfloat> %b.head, <4 x bfloat> poison, <4 x i32> zeroinitializer
|
|
%b.ext = fpext <4 x bfloat> %b.splat to <4 x float>
|
|
%c.ext = fpext <4 x bfloat> %c to <4 x float>
|
|
%res = call <4 x float> @llvm.fma.v4f32(<4 x float> %b.ext, <4 x float> %c.ext, <4 x float> %a)
|
|
ret <4 x float> %res
|
|
}
|
|
|
|
define <8 x float> @vfwmaccbf16_vv_v8f32(<8 x float> %a, <8 x bfloat> %b, <8 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vv_v8f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 8, e16, m1, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v10, v11
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_v8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 8, 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
|
|
%b.ext = fpext <8 x bfloat> %b to <8 x float>
|
|
%c.ext = fpext <8 x bfloat> %c to <8 x float>
|
|
%res = call <8 x float> @llvm.fma.v8f32(<8 x float> %b.ext, <8 x float> %c.ext, <8 x float> %a)
|
|
ret <8 x float> %res
|
|
}
|
|
|
|
define <8 x float> @vfwmaccbf16_vf_v8f32(<8 x float> %a, bfloat %b, <8 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vf_v8f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: fmv.x.h a0, fa0
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 8, e16, m1, ta, ma
|
|
; ZVFBFWMA-NEXT: vmv.v.x v11, a0
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v11, v10
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_v8f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fmv.x.w a0, fa0
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 8, e16, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vmv.v.x v11, a0
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v12, v11
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v14, v10
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v12, v14
|
|
; ZVFBFMIN-NEXT: ret
|
|
%b.head = insertelement <8 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <8 x bfloat> %b.head, <8 x bfloat> poison, <8 x i32> zeroinitializer
|
|
%b.ext = fpext <8 x bfloat> %b.splat to <8 x float>
|
|
%c.ext = fpext <8 x bfloat> %c to <8 x float>
|
|
%res = call <8 x float> @llvm.fma.v8f32(<8 x float> %b.ext, <8 x float> %c.ext, <8 x float> %a)
|
|
ret <8 x float> %res
|
|
}
|
|
|
|
define <16 x float> @vfwmaccbf16_vv_v16f32(<16 x float> %a, <16 x bfloat> %b, <16 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vv_v16f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 16, e16, m2, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v12, v14
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vv_v16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 16, 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
|
|
%b.ext = fpext <16 x bfloat> %b to <16 x float>
|
|
%c.ext = fpext <16 x bfloat> %c to <16 x float>
|
|
%res = call <16 x float> @llvm.fma.v16f32(<16 x float> %b.ext, <16 x float> %c.ext, <16 x float> %a)
|
|
ret <16 x float> %res
|
|
}
|
|
|
|
define <16 x float> @vfwmaccbf16_vf_v16f32(<16 x float> %a, bfloat %b, <16 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vf_v16f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: fmv.x.h a0, fa0
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 16, e16, m2, ta, ma
|
|
; ZVFBFWMA-NEXT: vmv.v.x v14, a0
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v14, v12
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_v16f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fmv.x.w a0, fa0
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 16, e16, m2, ta, ma
|
|
; ZVFBFMIN-NEXT: vmv.v.x v14, a0
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v16, v14
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v20, v12
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v16, v20
|
|
; ZVFBFMIN-NEXT: ret
|
|
%b.head = insertelement <16 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <16 x bfloat> %b.head, <16 x bfloat> poison, <16 x i32> zeroinitializer
|
|
%b.ext = fpext <16 x bfloat> %b.splat to <16 x float>
|
|
%c.ext = fpext <16 x bfloat> %c to <16 x float>
|
|
%res = call <16 x float> @llvm.fma.v16f32(<16 x float> %b.ext, <16 x float> %c.ext, <16 x float> %a)
|
|
ret <16 x float> %res
|
|
}
|
|
|
|
define <32 x float> @vfwmaccbf32_vv_v32f32(<32 x float> %a, <32 x bfloat> %b, <32 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf32_vv_v32f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: li a0, 32
|
|
; ZVFBFWMA-NEXT: vsetvli zero, a0, e16, m4, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v16, v20
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf32_vv_v32f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: li a0, 32
|
|
; ZVFBFMIN-NEXT: vsetvli zero, a0, 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
|
|
%b.ext = fpext <32 x bfloat> %b to <32 x float>
|
|
%c.ext = fpext <32 x bfloat> %c to <32 x float>
|
|
%res = call <32 x float> @llvm.fma.v32f32(<32 x float> %b.ext, <32 x float> %c.ext, <32 x float> %a)
|
|
ret <32 x float> %res
|
|
}
|
|
|
|
define <32 x float> @vfwmaccbf32_vf_v32f32(<32 x float> %a, bfloat %b, <32 x bfloat> %c) {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf32_vf_v32f32:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: fmv.x.h a0, fa0
|
|
; ZVFBFWMA-NEXT: li a1, 32
|
|
; ZVFBFWMA-NEXT: vsetvli zero, a1, e16, m4, ta, ma
|
|
; ZVFBFWMA-NEXT: vmv.v.x v20, a0
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vv v8, v20, v16
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf32_vf_v32f32:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fmv.x.w a0, fa0
|
|
; ZVFBFMIN-NEXT: li a1, 32
|
|
; ZVFBFMIN-NEXT: vsetvli zero, a1, e16, m4, ta, ma
|
|
; ZVFBFMIN-NEXT: vmv.v.x v20, a0
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v24, v20
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v0, v16
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m8, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vv v8, v24, v0
|
|
; ZVFBFMIN-NEXT: ret
|
|
%b.head = insertelement <32 x bfloat> poison, bfloat %b, i32 0
|
|
%b.splat = shufflevector <32 x bfloat> %b.head, <32 x bfloat> poison, <32 x i32> zeroinitializer
|
|
%b.ext = fpext <32 x bfloat> %b.splat to <32 x float>
|
|
%c.ext = fpext <32 x bfloat> %c to <32 x float>
|
|
%res = call <32 x float> @llvm.fma.v32f32(<32 x float> %b.ext, <32 x float> %c.ext, <32 x float> %a)
|
|
ret <32 x float> %res
|
|
}
|
|
|
|
define <4 x float> @vfwmaccbf16_vf_v4f32_scalar_extend(<4 x float> %rd, bfloat %a, <4 x bfloat> %b) local_unnamed_addr #0 {
|
|
; ZVFBFWMA-LABEL: vfwmaccbf16_vf_v4f32_scalar_extend:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwmaccbf16.vf v8, fa0, v9
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: vfwmaccbf16_vf_v4f32_scalar_extend:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: fmv.x.w a0, fa0
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvtbf16.f.f.v v10, v9
|
|
; ZVFBFMIN-NEXT: slli a0, a0, 16
|
|
; ZVFBFMIN-NEXT: fmv.w.x fa5, a0
|
|
; ZVFBFMIN-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFMIN-NEXT: vfmacc.vf v8, fa5, v10
|
|
; ZVFBFMIN-NEXT: ret
|
|
%b_ext = fpext <4 x bfloat> %b to <4 x float>
|
|
%a_extend = fpext bfloat %a to float
|
|
%a_insert = insertelement <4 x float> poison, float %a_extend, i64 0
|
|
%a_shuffle = shufflevector <4 x float> %a_insert, <4 x float> poison, <4 x i32> zeroinitializer
|
|
%fma = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a_shuffle, <4 x float> %b_ext, <4 x float> %rd)
|
|
ret <4 x float> %fma
|
|
}
|
|
|
|
; Negative test with a mix of bfloat and half fpext.
|
|
define <4 x float> @mix(<4 x float> %rd, <4 x half> %a, <4 x bfloat> %b) {
|
|
; ZVFBFWMA-LABEL: mix:
|
|
; ZVFBFWMA: # %bb.0:
|
|
; ZVFBFWMA-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
|
|
; ZVFBFWMA-NEXT: vfwcvt.f.f.v v11, v9
|
|
; ZVFBFWMA-NEXT: vfwcvtbf16.f.f.v v9, v10
|
|
; ZVFBFWMA-NEXT: vsetvli zero, zero, e32, m1, ta, ma
|
|
; ZVFBFWMA-NEXT: vfmacc.vv v8, v11, v9
|
|
; ZVFBFWMA-NEXT: ret
|
|
;
|
|
; ZVFBFMIN-LABEL: mix:
|
|
; ZVFBFMIN: # %bb.0:
|
|
; ZVFBFMIN-NEXT: vsetivli zero, 4, e16, mf2, ta, ma
|
|
; ZVFBFMIN-NEXT: vfwcvt.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
|
|
%a_ext = fpext <4 x half> %a to <4 x float>
|
|
%b_ext = fpext <4 x bfloat> %b to <4 x float>
|
|
%fma = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a_ext, <4 x float> %b_ext, <4 x float> %rd)
|
|
ret <4 x float> %fma
|
|
}
|