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

Add gemm universal bf16 instances (#1484)

Browse Source 
* revert ckprofiler change

* temp save

* Add test and test pass

* test pass

* Fix bug inside rotating buffer when tensor is not packed

* bug fix

* clang format

---------

Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
This commit is contained in:
Haocong WANG
2024-09-05 11:58:54 +08:00
committed by GitHub
parent 52410b49c7
commit 5b10dae6a4
35 changed files with 1048 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3.hpp
View File

@@ -221,7 +221,7 @@ struct GridwiseGemm_xdl_cshuffle_v3
make_tuple(Sequence<3>{}, Sequence<0, 1, 2>{}));
}
__device__ static auto MakeAGridDescriptor_AK0_M_AK1(
__host__ __device__ static auto MakeAGridDescriptor_AK0_M_AK1(
index_t M, index_t MPad, index_t K, index_t KPad, index_t StrideA, index_t AK0)
{
const auto a_grid_desc_mraw_kraw = [&]() {
@@ -303,7 +303,7 @@ struct GridwiseGemm_xdl_cshuffle_v3
}
}
__device__ static auto MakeBGridDescriptor_BK0_N_BK1(
__host__ __device__ static auto MakeBGridDescriptor_BK0_N_BK1(
index_t K, index_t KPad, index_t N, index_t NPad, index_t StrideB, index_t BK0)
{
const auto b_grid_desc_nraw_kraw = [&]() {
@@ -576,12 +576,12 @@ struct GridwiseGemm_xdl_cshuffle_v3
}
else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
{
a_k_split_offset = blockIdx.z * karg.KRead * karg.M;
a_k_split_offset = blockIdx.z * karg.KRead * karg.StrideA;
}
if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
{
b_k_split_offset = blockIdx.z * karg.KRead * karg.N;
b_k_split_offset = blockIdx.z * karg.KRead * karg.StrideB;
}
else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
{

8
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_d.hpp
View File

@@ -255,7 +255,7 @@ struct GridwiseGemmMultiD_xdl_cshuffle_v3
make_tuple(Sequence<3>{}, Sequence<0, 1, 2>{}));
}
__device__ static auto MakeAGridDescriptor_AK0_M_AK1(
__host__ __device__ static auto MakeAGridDescriptor_AK0_M_AK1(
index_t M, index_t MPad, index_t K, index_t KPad, index_t StrideA, index_t AK0)
{
const auto a_grid_desc_mraw_kraw = [&]() {
@@ -337,7 +337,7 @@ struct GridwiseGemmMultiD_xdl_cshuffle_v3
}
}
__device__ static auto MakeBGridDescriptor_BK0_N_BK1(
__host__ __device__ static auto MakeBGridDescriptor_BK0_N_BK1(
index_t K, index_t KPad, index_t N, index_t NPad, index_t StrideB, index_t BK0)
{
const auto b_grid_desc_nraw_kraw = [&]() {
@@ -647,12 +647,12 @@ struct GridwiseGemmMultiD_xdl_cshuffle_v3
}
else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
{
a_k_split_offset = blockIdx.z * karg.KRead * karg.M;
a_k_split_offset = blockIdx.z * karg.KRead * karg.StrideA;
}
if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
{
b_k_split_offset = blockIdx.z * karg.KRead * karg.N;
b_k_split_offset = blockIdx.z * karg.KRead * karg.StrideB;
}
else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
{