ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-15 18:42:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Extend permute scale support up to 6D (#1168)

Browse Source 
* Extend permute scale support up to 6D

* Fixes

* Fixes

* Update profiler/README.md

Co-authored-by: Lisa <lisajdelaney@gmail.com>

* Update profiler/README.md

Co-authored-by: Lisa <lisajdelaney@gmail.com>

* Update profiler/README.md

Co-authored-by: Lisa <lisajdelaney@gmail.com>

* Update profiler/README.md

Co-authored-by: Lisa <lisajdelaney@gmail.com>

* Update profiler/README.md

Co-authored-by: Lisa <lisajdelaney@gmail.com>

* Update profiler/README.md

Co-authored-by: Lisa <lisajdelaney@gmail.com>

* Update profiler/README.md

Co-authored-by: Lisa <lisajdelaney@gmail.com>

---------

Co-authored-by: Lisa <lisajdelaney@gmail.com>

[ROCm/composable_kernel commit: 66736edb95]
This commit is contained in:
Bartłomiej Kocot
2024-02-20 18:56:54 +01:00
committed by GitHub
parent 4acafee7c2
commit a08052cf6a
15 changed files with 898 additions and 299 deletions
Download Patch File Download Diff File Expand all files Collapse all files

195
library/include/ck/library/tensor_operation_instance/gpu/permute_scale.hpp
View File

@@ -1,5 +1,5 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
@@ -17,7 +17,32 @@ namespace tensor_operation {
namespace device {
namespace instance {
void add_device_permute_scale_f16_instances(
#ifdef CK_ENABLE_FP16
void add_device_permute_scale_1d_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>,
ck::Tuple<F16>,
PassThrough,
element_wise::UnarySquare,
Scale,
1>>>&);
void add_device_permute_scale_2d_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>,
ck::Tuple<F16>,
PassThrough,
element_wise::UnarySquare,
Scale,
2>>>&);
void add_device_permute_scale_3d_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>,
ck::Tuple<F16>,
PassThrough,
element_wise::UnarySquare,
Scale,
3>>>&);
void add_device_permute_scale_4d_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>,
ck::Tuple<F16>,
PassThrough,
@@ -25,7 +50,50 @@ void add_device_permute_scale_f16_instances(
Scale,
4>>>&);
void add_device_permute_scale_f32_instances(
void add_device_permute_scale_5d_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>,
ck::Tuple<F16>,
PassThrough,
element_wise::UnarySquare,
Scale,
5>>>&);
void add_device_permute_scale_6d_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>,
ck::Tuple<F16>,
PassThrough,
element_wise::UnarySquare,
Scale,
6>>>&);
#endif
#ifdef CK_ENABLE_FP32
void add_device_permute_scale_1d_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>,
ck::Tuple<F32>,
PassThrough,
element_wise::UnarySquare,
Scale,
1>>>&);
void add_device_permute_scale_2d_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>,
ck::Tuple<F32>,
PassThrough,
element_wise::UnarySquare,
Scale,
2>>>&);
void add_device_permute_scale_3d_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>,
ck::Tuple<F32>,
PassThrough,
element_wise::UnarySquare,
Scale,
3>>>&);
void add_device_permute_scale_4d_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>,
ck::Tuple<F32>,
PassThrough,
@@ -33,6 +101,23 @@ void add_device_permute_scale_f32_instances(
Scale,
4>>>&);
void add_device_permute_scale_5d_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>,
ck::Tuple<F32>,
PassThrough,
element_wise::UnarySquare,
Scale,
5>>>&);
void add_device_permute_scale_6d_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>,
ck::Tuple<F32>,
PassThrough,
element_wise::UnarySquare,
Scale,
6>>>&);
#endif
template <typename InDataTypeTuple,
typename OutDataTypeTuple,
typename ElementwiseOperation,
@@ -57,15 +142,107 @@ struct DeviceOperationInstanceFactory<
static auto GetInstances()
{
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F32>>)
if constexpr(NumDim == 1)
{
add_device_permute_scale_f32_instances(op_ptrs);
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F32>>)
{
add_device_permute_scale_1d_f32_instances(op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F16>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F16>>)
{
add_device_permute_scale_1d_f16_instances(op_ptrs);
}
#endif
}
else if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F16>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F16>>)
else if constexpr(NumDim == 2)
{
add_device_permute_scale_f16_instances(op_ptrs);
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F32>>)
{
add_device_permute_scale_2d_f32_instances(op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F16>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F16>>)
{
add_device_permute_scale_2d_f16_instances(op_ptrs);
}
#endif
}
else if constexpr(NumDim == 3)
{
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F32>>)
{
add_device_permute_scale_3d_f32_instances(op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F16>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F16>>)
{
add_device_permute_scale_3d_f16_instances(op_ptrs);
}
#endif
}
else if constexpr(NumDim == 4)
{
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F32>>)
{
add_device_permute_scale_4d_f32_instances(op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F16>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F16>>)
{
add_device_permute_scale_4d_f16_instances(op_ptrs);
}
#endif
}
else if constexpr(NumDim == 5)
{
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F32>>)
{
add_device_permute_scale_5d_f32_instances(op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F16>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F16>>)
{
add_device_permute_scale_5d_f16_instances(op_ptrs);
}
#endif
}
else if constexpr(NumDim == 6)
{
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F32>>)
{
add_device_permute_scale_6d_f32_instances(op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F16>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F16>>)
{
add_device_permute_scale_6d_f16_instances(op_ptrs);
}
#endif
}
return op_ptrs;
}

42
library/include/ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp Normal file
View File

@@ -0,0 +1,42 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_scale_impl.hpp"
#include "ck/utility/data_type.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
using Pass = ck::tensor_operation::element_wise::PassThrough;
using UnaryOp = ck::tensor_operation::element_wise::UnarySquare;
using Scale = ck::tensor_operation::element_wise::Scale;
// clang-format off
template <index_t NDims>
using device_permute_scale_f16_instances =
std::tuple <
DeviceElementwiseImpl<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, NDims, 1, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, NDims, 8, ck::Sequence<8>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, NDims, 4, ck::Sequence<4>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, NDims, 2, ck::Sequence<2>, ck::Sequence<1>>
>;
template <index_t NDims>
using device_permute_scale_f32_instances = std::tuple<
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, NDims, 1, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, NDims, 8, ck::Sequence<8>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, NDims, 4, ck::Sequence<4>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, NDims, 2, ck::Sequence<2>, ck::Sequence<1>>
>;
// clang-format on
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck