ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 05:01:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Layernorm4d (#1022)

Browse Source 
* Rename folder

* Add layernorm 4d fwd example

* Rename original layernorm example

* Add layernorm 4d f16  test

* Add layernorm4d_fwd client example

* Support layernorm4D in ckProfiler

* Rename groupnorm to groupnorm fwd in example

* Rename layernorm and group fwd in test

* Rename normalization to normalization_fwd (instances)

* Add fwd to DeviceNormalization

* Rename external api header

* Rename folder, because we can also add bwd in this folder

* Add fwd in layernorm and groupnorm (profiler

* Fix compile error

---------

Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
This commit is contained in:
rocking
2023-11-09 08:34:51 +08:00
committed by GitHub
parent ce52621123
commit a3d9a2cd42
59 changed files with 1271 additions and 675 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
include/ck/tensor_operation/gpu/device/device_normalization.hpp → include/ck/tensor_operation/gpu/device/device_normalization_fwd.hpp
View File

@@ -19,7 +19,7 @@ template <typename XDataType,
typename YElementwiseOperation,
index_t Rank,
index_t NumReduceDim>
struct DeviceNormalization : public BaseOperator
struct DeviceNormalizationFwd : public BaseOperator
{
virtual std::unique_ptr<BaseArgument>
MakeArgumentPointer(const std::vector<index_t> lengths,
@@ -50,14 +50,14 @@ template <typename XDataType,
typename YElementwiseOperation,
index_t Rank,
index_t NumReduceDim>
using DeviceNormalizationPtr = std::unique_ptr<DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
YDataType,
SaveMeanInvStdDataType,
YElementwiseOperation,
Rank,
NumReduceDim>>;
using DeviceNormalizationFwdPtr = std::unique_ptr<DeviceNormalizationFwd<XDataType,
GammaDataType,
BetaDataType,
YDataType,
SaveMeanInvStdDataType,
YElementwiseOperation,
Rank,
NumReduceDim>>;
} // namespace device
} // namespace tensor_operation

20
include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp → include/ck/tensor_operation/gpu/device/impl/device_normalization_fwd_impl.hpp
View File

@@ -7,7 +7,7 @@
#include <sstream>
#include "ck/utility/reduction_operator.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization_fwd.hpp"
#include "ck/tensor_operation/gpu/device/device_reduce.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp"
@@ -46,14 +46,14 @@ template <typename XDataType,
index_t YDstVectorSize,
index_t SaveMeanInvStdDstVectorSize,
bool UseWelford = true>
struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
YDataType,
SaveMeanInvStdDataType,
YElementwiseOperation,
Rank,
NumReduceDim>
struct DeviceNormalizationFwdImpl : public DeviceNormalizationFwd<XDataType,
GammaDataType,
BetaDataType,
YDataType,
SaveMeanInvStdDataType,
YElementwiseOperation,
Rank,
NumReduceDim>
{
static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize);
static_assert(
@@ -461,7 +461,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
auto str = std::stringstream();
// clang-format off
str << "DeviceNormalizationImpl<" << BlockSize << ",";
str << "DeviceNormalizationFwdImpl<" << BlockSize << ",";
str << "Cluster_MK_" << MThreadClusterSize << "_" << KThreadClusterSize << ",";
str << "Slice_MK_" << MThreadSliceSize << "_" << KThreadSliceSize << ",";
str << "XYSrcVectorDim_" << XYSrcVectorDim << ",";

20
include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp → include/ck/tensor_operation/gpu/device/impl/device_normalization_fwd_splitk_impl.hpp
View File

@@ -8,7 +8,7 @@
#include "ck/utility/reduction_operator.hpp"
#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization_fwd.hpp"
#include "ck/tensor_operation/gpu/device/device_reduce.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp"
@@ -134,14 +134,14 @@ template <typename XDataType,
index_t BetaSrcVectorSize,
index_t YDstVectorSize,
index_t SaveMeanInvStdDstVectorSize>
struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
YDataType,
SaveMeanInvStdDataType,
YElementwiseOperation,
Rank,
NumReduceDim>
struct DeviceNormalizationFwdSplitKImpl : public DeviceNormalizationFwd<XDataType,
GammaDataType,
BetaDataType,
YDataType,
SaveMeanInvStdDataType,
YElementwiseOperation,
Rank,
NumReduceDim>
{
using WorkspaceMeanVarDataType = SaveMeanInvStdDataType;
@@ -732,7 +732,7 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
auto str = std::stringstream();
// clang-format off
str << "DeviceNormalizationSplitKImpl<" << BlockSize << ",";
str << "DeviceNormalizationFwdSplitKImpl<" << BlockSize << ",";
str << "Cluster_MK_" << MThreadClusterSize << "_" << KThreadClusterSize << ",";
str << "Slice_MK_" << MThreadSliceSize << "_" << KThreadSliceSize << ",";
str << "XYSrcVectorDim_" << XYVectorDim << ",";