ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-20 12:59:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Layernorm welford (#346)

Browse Source 
* Add threadwise and blockwise welford

* Rename gridwise op, prepare to add welford version

* implement welford and integrate welford into layernorm

* Take care of tail loop

* Fix buf when ThreadSliceK > 1

* Fix bug of merging of two empty set

* Rename clip to clamp

* 1. Fix type of count
2. Remove useless static_assert

* Do not inherit Reduction::Argument

* [What] replace __syncthreads() with block_sync_lds()
[Why] __syncthreads might wait both lgkmcnt(0) and vmcnt(0)

* Add y stride

* Rename.
DeviceLayernorm -> DeviceLayernormImpl
DeviceNormalization2 -> DeviceLayernorm

* Move literal ""_uz & ""_zu into namespace 'literals'

* Move namespace 'literals' as 'ck::literals'

Co-authored-by: Po-Yen, Chen <PoYen.Chen@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>

[ROCm/composable_kernel commit: 0bd6b842b9]
This commit is contained in:
rocking5566
2022-08-13 22:43:18 +08:00
committed by GitHub
parent 37f61cbba5
commit 8ff56b19d5
13 changed files with 1097 additions and 476 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
example/27_layernorm/layernorm_blockwise.cpp
View File

@@ -9,7 +9,7 @@
#include "ck/ck.hpp"
#include "ck/utility/reduction_enums.hpp"
#include "ck/tensor_operation/gpu/device/device_layernorm.hpp"
#include "ck/tensor_operation/gpu/device/device_layernorm_impl.hpp"
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/check_err.hpp"
@@ -29,24 +29,24 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance = ck::tensor_operation::device::DeviceLayernorm<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // SrcVecDim (0=M, 1=K)
8, // SrcScalarPerVector
8, // GammaScalarPerVector
8, // BetaScalarPerVector
8>; // OutScalarPerVector
using DeviceInstance = ck::tensor_operation::device::DeviceLayernormImpl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // SrcVecDim (0=M, 1=K)
8, // SrcScalarPerVector
8, // GammaScalarPerVector
8, // BetaScalarPerVector
8>; // OutScalarPerVector
int main()
{
@@ -90,6 +90,7 @@ int main()
std::vector<ck::index_t>{x.mDesc.GetStrides().begin(), x.mDesc.GetStrides().end()},
std::vector<ck::index_t>{gamma.mDesc.GetStrides().begin(), gamma.mDesc.GetStrides().end()},
std::vector<ck::index_t>{beta.mDesc.GetStrides().begin(), beta.mDesc.GetStrides().end()},
std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
{1},
1e-4,
x_dev.GetDeviceBuffer(),