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
127 lines
7.1 KiB
MLIR
127 lines
7.1 KiB
MLIR
// RUN: mlir-opt %s -convert-amdgpu-to-rocdl=chipset=gfx942 | FileCheck %s
|
|
|
|
// CHECK-LABEL: func @ext_scalar
|
|
// CHECK: [[V:%.+]] = builtin.unrealized_conversion_cast %{{.+}} : f8E5M2FNUZ to i8
|
|
// CHECK-DAG: [[UNDEF:%.+]] = llvm.mlir.undef : vector<4xi8>
|
|
// CHECK-DAG: [[C0_1:%.+]] = llvm.mlir.constant(0 : i32) : i32
|
|
// CHECK: [[VEC:%.+]] = llvm.insertelement [[V]], [[UNDEF]]{{\[}}[[C0_1]] : i32] : vector<4xi8>
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[VEC]] : vector<4xi8> to i32
|
|
// CHECK: [[EXT:%.+]] = rocdl.cvt.f32.bf8 [[CAST]][0] : f32
|
|
// CHECK: return [[EXT]] : f32
|
|
func.func @ext_scalar(%v: f8E5M2FNUZ) -> f32 {
|
|
%ret = amdgpu.ext_packed_fp8 %v[0] : f8E5M2FNUZ to f32
|
|
func.return %ret : f32
|
|
}
|
|
|
|
// CHECK-LABEL: func @ext_short_vec
|
|
// CHECK: [[V:%.+]] = builtin.unrealized_conversion_cast %{{.+}} : vector<2xf8E4M3FNUZ> to vector<2xi8>
|
|
// CHECK-DAG: [[UNDEF:%.+]] = llvm.mlir.undef : vector<4xi8>
|
|
// CHECK-DAG: [[C0:%.+]] = llvm.mlir.constant(0 : i32) : i32
|
|
// CHECK: [[ELEM_0:%.+]] = llvm.extractelement [[V]]{{\[}}[[C0]] : i32] : vector<2xi8>
|
|
// CHECK: [[VEC_0:%.+]] = llvm.insertelement [[ELEM_0]], [[UNDEF]]{{\[}}[[C0]] : i32] : vector<4xi8>
|
|
// CHECK: [[C1_1:%.+]] = llvm.mlir.constant(1 : i32) : i32
|
|
// CHECK: [[ELEM_1:%.+]] = llvm.extractelement [[V]]{{\[}}[[C1_1]] : i32] : vector<2xi8>
|
|
// CHECK: [[VEC_1:%.+]] = llvm.insertelement [[ELEM_1]], [[VEC_0]]{{\[}}[[C1_1]] : i32] : vector<4xi8>
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[VEC_1]] : vector<4xi8> to i32
|
|
// CHECK: [[EXT:%.+]] = rocdl.cvt.f32.fp8 [[CAST]][1] : f32
|
|
// CHECK: return [[EXT]] : f32
|
|
func.func @ext_short_vec(%v: vector<2xf8E4M3FNUZ>) -> f32 {
|
|
%ret = amdgpu.ext_packed_fp8 %v[1] : vector<2xf8E4M3FNUZ> to f32
|
|
func.return %ret : f32
|
|
}
|
|
|
|
// CHECK-LABEL: func @ext_full_vec
|
|
// CHECK: [[V:%.+]] = builtin.unrealized_conversion_cast %{{.+}} : vector<4xf8E4M3FNUZ> to vector<4xi8>
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[V]] : vector<4xi8> to i32
|
|
// CHECK: [[EXT:%.+]] = rocdl.cvt.f32.fp8 [[CAST]][3] : f32
|
|
// CHECK: return [[EXT]] : f32
|
|
func.func @ext_full_vec(%v: vector<4xf8E4M3FNUZ>) -> f32 {
|
|
%ret = amdgpu.ext_packed_fp8 %v[3] : vector<4xf8E4M3FNUZ> to f32
|
|
func.return %ret : f32
|
|
}
|
|
|
|
// CHECK-LABEL: func @ext_packed_2xfp8
|
|
// CHECK: [[V:%.+]] = builtin.unrealized_conversion_cast %{{.+}} : vector<2xf8E4M3FNUZ> to vector<2xi8>
|
|
// CHECK-DAG: [[UNDEF:%.+]] = llvm.mlir.undef : vector<4xi8>
|
|
// CHECK-DAG: [[C0:%.+]] = llvm.mlir.constant(0 : i32) : i32
|
|
// CHECK: [[ELEM_0:%.+]] = llvm.extractelement [[V]]{{\[}}[[C0]] : i32] : vector<2xi8>
|
|
// CHECK: [[VEC_0:%.+]] = llvm.insertelement [[ELEM_0]], [[UNDEF]]{{\[}}[[C0]] : i32] : vector<4xi8>
|
|
// CHECK: [[C1_1:%.+]] = llvm.mlir.constant(1 : i32) : i32
|
|
// CHECK: [[ELEM_1:%.+]] = llvm.extractelement [[V]]{{\[}}[[C1_1]] : i32] : vector<2xi8>
|
|
// CHECK: [[VEC_1:%.+]] = llvm.insertelement [[ELEM_1]], [[VEC_0]]{{\[}}[[C1_1]] : i32] : vector<4xi8>
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[VEC_1]] : vector<4xi8> to i32
|
|
// CHECK: [[EXT:%.+]] = rocdl.cvt.pk.f32.fp8 [[CAST]][false] : vector<2xf32>
|
|
// CHECK: return [[EXT]]
|
|
func.func @ext_packed_2xfp8(%v: vector<2xf8E4M3FNUZ>) -> vector<2xf32> {
|
|
%ret = amdgpu.ext_packed_fp8 %v[0] : vector<2xf8E4M3FNUZ> to vector<2xf32>
|
|
func.return %ret : vector<2xf32>
|
|
}
|
|
|
|
// CHECK-LABEL: func @ext_packed_4xfp8(
|
|
// CHECK: [[V:%.+]] = builtin.unrealized_conversion_cast %{{.+}} : vector<4xf8E4M3FNUZ> to vector<4xi8>
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[V]] : vector<4xi8> to i32
|
|
// CHECK: [[EXT:%.+]] = rocdl.cvt.pk.f32.fp8 [[CAST]][true] : vector<2xf32>
|
|
// CHECK: return [[EXT]] : vector<2xf32>
|
|
func.func @ext_packed_4xfp8(%v: vector<4xf8E4M3FNUZ>) -> vector<2xf32> {
|
|
%ret = amdgpu.ext_packed_fp8 %v[1] : vector<4xf8E4M3FNUZ> to vector<2xf32>
|
|
func.return %ret : vector<2xf32>
|
|
}
|
|
|
|
// CHECK-LABEL: func @packed_trunc
|
|
// CHECK-SAME: ([[V:%.+]]: f32)
|
|
// CHECK: [[V2:%.+]] = llvm.mlir.undef : f32
|
|
// CHECK: [[EXISTING:%.+]] = llvm.mlir.undef : i32
|
|
// CHECK: [[PACKED:%.+]] = rocdl.cvt.pk.fp8.f32 [[V]], [[V2]] -> [[EXISTING]][false] : i32
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[PACKED]] : i32 to vector<4xi8>
|
|
// CHECK: builtin.unrealized_conversion_cast [[CAST]] : vector<4xi8> to vector<4xf8E4M3FNUZ>
|
|
func.func @packed_trunc(%v: f32) -> vector<4xf8E4M3FNUZ> {
|
|
%ret = amdgpu.packed_trunc_2xfp8 %v, undef into undef[word 0] : f32 to vector<4xf8E4M3FNUZ>
|
|
func.return %ret : vector<4xf8E4M3FNUZ>
|
|
}
|
|
|
|
// CHECK-LABEL: func @packed_truncx2
|
|
// CHECK-SAME: ([[V:%.+]]: f32, [[W:%.+]]: f32)
|
|
// CHECK: [[EXISTING:%.+]] = llvm.mlir.undef : i32
|
|
// CHECK: [[PACKED:%.+]] = rocdl.cvt.pk.fp8.f32 [[V]], [[W]] -> [[EXISTING]][false] : i32
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[PACKED]] : i32 to vector<4xi8>
|
|
// CHECK: builtin.unrealized_conversion_cast [[CAST]] : vector<4xi8> to vector<4xf8E4M3FNUZ>
|
|
func.func @packed_truncx2(%v: f32, %w: f32) -> vector<4xf8E4M3FNUZ> {
|
|
%ret = amdgpu.packed_trunc_2xfp8 %v, %w into undef[word 0] : f32 to vector<4xf8E4M3FNUZ>
|
|
func.return %ret : vector<4xf8E4M3FNUZ>
|
|
}
|
|
|
|
// CHECK-LABEL: func @packed_truncx2_into
|
|
// CHECK-SAME: ([[V:%.+]]: f32, [[W:%.+]]: f32, [[EXISTING:%.+]]: vector<4xf8E5M2FNUZ>)
|
|
// CHECK: [[EXISTING_BYTES:%.+]] = builtin.unrealized_conversion_cast [[EXISTING]] : vector<4xf8E5M2FNUZ> to vector<4xi8>
|
|
// CHECK: [[EXISTING_INT:%.+]] = llvm.bitcast [[EXISTING_BYTES]] : vector<4xi8> to i32
|
|
// CHECK: [[PACKED:%.+]] = rocdl.cvt.pk.bf8.f32 [[V]], [[W]] -> [[EXISTING_INT]][true] : i32
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[PACKED]] : i32 to vector<4xi8>
|
|
// CHECK: builtin.unrealized_conversion_cast [[CAST]] : vector<4xi8> to vector<4xf8E5M2FNUZ>
|
|
func.func @packed_truncx2_into(%v: f32, %w: f32, %existing: vector<4xf8E5M2FNUZ>) -> vector<4xf8E5M2FNUZ> {
|
|
%ret = amdgpu.packed_trunc_2xfp8 %v, %w into %existing[word 1] : f32 to vector<4xf8E5M2FNUZ> into vector<4xf8E5M2FNUZ>
|
|
func.return %ret : vector<4xf8E5M2FNUZ>
|
|
}
|
|
|
|
// CHECK-LABEL: func @packed_stoch_round
|
|
// CHECK-SAME: ([[V:%.+]]: f32, [[S:%.+]]: i32)
|
|
// CHECK: [[EXISTING:%.+]] = llvm.mlir.undef : i32
|
|
// CHECK: [[PACKED:%.+]] = rocdl.cvt.sr.fp8.f32 [[V]], [[S]] -> [[EXISTING]][0] : i32
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[PACKED]] : i32 to vector<4xi8>
|
|
// CHECK: builtin.unrealized_conversion_cast [[CAST]] : vector<4xi8> to vector<4xf8E4M3FNUZ>
|
|
func.func @packed_stoch_round(%v: f32, %s: i32) -> vector<4xf8E4M3FNUZ> {
|
|
%ret = amdgpu.packed_stoch_round_fp8 %v + %s into undef[0] : f32 to vector<4xf8E4M3FNUZ>
|
|
func.return %ret : vector<4xf8E4M3FNUZ>
|
|
}
|
|
|
|
// CHECK-LABEL: func @packed_stoch_round_into
|
|
// CHECK-SAME: ([[V:%.+]]: f32, [[S:%.+]]: i32, [[EXISTING:%.+]]: vector<4xf8E5M2FNUZ>)
|
|
// CHECK: [[EXISTING_BYTES:%.+]] = builtin.unrealized_conversion_cast [[EXISTING]] : vector<4xf8E5M2FNUZ> to vector<4xi8>
|
|
// CHECK: [[EXISTING_INT:%.+]] = llvm.bitcast [[EXISTING_BYTES]] : vector<4xi8> to i32
|
|
// CHECK: [[PACKED:%.+]] = rocdl.cvt.sr.bf8.f32 [[V]], [[S]] -> [[EXISTING_INT]][1] : i32
|
|
// CHECK: [[CAST:%.+]] = llvm.bitcast [[PACKED]] : i32 to vector<4xi8>
|
|
// CHECK: builtin.unrealized_conversion_cast [[CAST]] : vector<4xi8> to vector<4xf8E5M2FNUZ>
|
|
func.func @packed_stoch_round_into(%v: f32, %s: i32, %existing: vector<4xf8E5M2FNUZ>) -> vector<4xf8E5M2FNUZ> {
|
|
%ret = amdgpu.packed_stoch_round_fp8 %v + %s into %existing[1] : f32 to vector<4xf8E5M2FNUZ> into vector<4xf8E5M2FNUZ>
|
|
func.return %ret : vector<4xf8E5M2FNUZ>
|
|
}
|