ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] Multiple-D GEMM example (#2219)

Browse Source 
* Multiple d, initial commit

* Check Ds Layout

* Readme and clang format

* Update branch & conflicts

* Multiple D - fix clang-formatter

* Rename elemetwise_op

* Fix CI

* Code review part1

* Remove printf

* Remove unnecessary comment

* Add new tests with Col layout

* Review part 2

* Added support for Multiple D GEMM

* Update comment

* Remove maybe_unused

* Clang-format

* Review part 3

* Add comment to function

* Add comment to function: another

* Take number of params for a refrence function

* Remove additional d param for 0 tensor

* Change name of function

* Fix CI fails
This commit is contained in:
Mateusz Ozga
2025-06-13 19:39:11 +02:00
committed by GitHub
parent 3a0cb27966
commit bd96ac9742
34 changed files with 2267 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

101
example/ck_tile/03_gemm/run_gemm_example.inc
View File

@@ -146,11 +146,14 @@ void permute_vectors_i4x4_b(Tensor& tensor)
template <typename ADataType,
typename BDataType,
typename DsDataType,
typename AccDataType,
typename CDataType,
typename ALayout,
typename BLayout,
typename CLayout>
typename DsLayout,
typename CLayout,
typename CDEElementWise = ck_tile::element_wise::PassThrough>
float invoke_gemm(ck_tile::DeviceMem& a_m_k_dev_buf,
ck_tile::DeviceMem& b_k_n_dev_buf,
ck_tile::DeviceMem& c_m_n_dev_buf,
@@ -165,41 +168,48 @@ float invoke_gemm(ck_tile::DeviceMem& a_m_k_dev_buf,
int n_repeat,
bool persistent)
{
ck_tile::GemmHostArgs args;
args.a_ptr = a_m_k_dev_buf.GetDeviceBuffer();
args.b_ptr = b_k_n_dev_buf.GetDeviceBuffer();
args.c_ptr = c_m_n_dev_buf.GetDeviceBuffer();
args.k_batch = kbatch;
args.M = M;
args.N = N;
args.K = K;
args.stride_A = stride_A;
args.stride_B = stride_B;
args.stride_C = stride_C;
ck_tile::GemmHostArgs</*NumDTensor = 0*/> args = {a_m_k_dev_buf.GetDeviceBuffer(),
b_k_n_dev_buf.GetDeviceBuffer(),
{},
c_m_n_dev_buf.GetDeviceBuffer(),
kbatch,
M,
N,
K,
stride_A,
stride_B,
{},
stride_C};
float ave_time;
if(persistent)
{
ave_time = gemm_calc<ADataType,
BDataType,
AccDataType,
CDataType,
ALayout,
BLayout,
CLayout,
true>(
ave_time = gemm<ADataType,
BDataType,
DsDataType,
AccDataType,
CDataType,
ALayout,
BLayout,
DsLayout,
CLayout,
true,
CDEElementWise>(
args, ck_tile::stream_config{nullptr, true, 1, n_warmup, n_repeat, true, true, 50});
}
else
{
ave_time = gemm_calc<ADataType,
BDataType,
AccDataType,
CDataType,
ALayout,
BLayout,
CLayout,
false>(
ave_time = gemm<ADataType,
BDataType,
DsDataType,
AccDataType,
CDataType,
ALayout,
BLayout,
DsLayout,
CLayout,
false,
CDEElementWise>(
args, ck_tile::stream_config{nullptr, true, 1, n_warmup, n_repeat, true, true, 50});
}
@@ -328,20 +338,27 @@ int run_gemm_example_with_layouts(int argc,
c_m_n_dev_buf.SetZero();
c_m_n_dev_result.SetZero();
invoke_gemm<ADataType, BDataType, AccDataType, CDataType, ALayout, BLayout, CLayout>(
a_m_k_dev_buf,
b_k_n_dev_buf,
c_m_n_dev_buf,
M,
N,
K,
stride_A,
stride_B,
stride_C,
kbatch,
n_warmup,
n_repeat,
persistent);
invoke_gemm<ADataType,
BDataType,
ck_tile::tuple<>,
AccDataType,
CDataType,
ALayout,
BLayout,
ck_tile::tuple<>,
CLayout>(a_m_k_dev_buf,
b_k_n_dev_buf,
c_m_n_dev_buf,
M,
N,
K,
stride_A,
stride_B,
stride_C,
kbatch,
n_warmup,
n_repeat,
persistent);
c_m_n_dev_buf.FromDevice(c_m_n_dev_result.data());
bool pass = true;