Skip to content

Implement device grouped gemm fixed nk multi abd for rdna4 #3619

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
zsotakal wants to merge 12 commits into
base: develop
Choose a base branch
from
+3,568 −77
Open

Implement device grouped gemm fixed nk multi abd for rdna4 #3619

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
b637af1
device struct implementation
Dec 19, 2025
4b32353
added xdl grouped multi abd fixed nk testing
Jan 9, 2026
ab6faa7
wmma implementation fixed
Jan 19, 2026
cf0ebfb
avoid unnecessary device mem allocation and code cleanups
Jan 19, 2026
97db9d6
cleanup instances definitions
Jan 19, 2026
9a10230
wmma examples added
Jan 20, 2026
06dc334
code cleanups
Jan 20, 2026
96d5a96
fix clang format
illsilin Jan 20, 2026
06bac70
typo and compilation fixes related to reference gemm
Jan 21, 2026
15b26a3
fix compilation error due to std::remove_cvref_t
Jan 21, 2026
a84d37b
added missing hip_check_error includes
Jan 22, 2026
55c6b7a
correction to example instances
Jan 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions client_example/31_grouped_gemm_bf16Aint8B/grouped_gemm_bias_fastgelu_xdl_bf16_i8.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,8 @@

#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_multi_abd_fixed_nk.hpp"

#include "ck/host_utility/hip_check_error.hpp"

using ::ck::hip_check_error;

template <ck::index_t... Is>
Expand Down
8 changes: 8 additions & 0 deletions example/59_grouped_gemm_multi_ABD/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,3 +8,11 @@ add_example_dependencies(example_grouped_gemm_xdl_multi_abd example_grouped_gemm

add_example_executable(example_grouped_gemm_multi_abd_xdl_fixed_nk_bias_bf16_i8 grouped_gemm_multi_abd_xdl_fixed_nk_bias_bf16_i8.cpp)
add_example_dependencies(example_grouped_gemm_xdl_multi_abd example_grouped_gemm_multi_abd_xdl_fixed_nk_bias_bf16_i8)

add_custom_target(example_grouped_gemm_wmma_multi_abd)

add_example_executable(example_grouped_gemm_multi_abd_wmma_fixed_nk_bias_fp16 grouped_gemm_multi_abd_wmma_fixed_nk_bias_fp16.cpp)
add_example_dependencies(example_grouped_gemm_wmma_multi_abd example_grouped_gemm_multi_abd_wmma_fixed_nk_bias_fp16)

add_example_executable(example_grouped_gemm_multi_abd_wmma_fixed_nk_bias_bf16_i8 grouped_gemm_multi_abd_wmma_fixed_nk_bias_bf16_i8.cpp)
add_example_dependencies(example_grouped_gemm_wmma_multi_abd example_grouped_gemm_multi_abd_wmma_fixed_nk_bias_bf16_i8)
409 changes: 409 additions & 0 deletions example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_wmma_fixed_nk_bias_bf16_i8.cpp
View file
Open in desktop
Copy link
Contributor

@ErwinTerpstra ErwinTerpstra Jan 22, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As I've placed some comments on the fp16 example, it would be good to place the common code for these examples in a shared file (similar to how other operations usually have a run_grouped_gemm_xxxx_example.inc)

Large diffs are not rendered by default.

403 changes: 403 additions & 0 deletions example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_wmma_fixed_nk_bias_fp16.cpp
View file
Open in desktop

Large diffs are not rendered by default.

4 changes: 3 additions & 1 deletion example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_xdl_fixed_nk_bias_bf16_i8.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,8 @@
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"

#include "ck/host_utility/hip_check_error.hpp"

using ::ck::DeviceMem;
using ::ck::hip_check_error;
using ::ck::HostTensorDescriptor;
Expand Down Expand Up @@ -398,7 +400,7 @@ int main(int argc, char* argv[])
{
printf("arg1: verification (0=no, 1=yes)\n");
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
printf("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg3: time kernel (0=no, 1=yes)\n");
printf("arg4: k_batch (>0)\n");
exit(0);
}
Expand Down
8 changes: 5 additions & 3 deletions example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_xdl_fixed_nk_bias_fp16.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,8 @@
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"

#include "ck/host_utility/hip_check_error.hpp"

using ::ck::DeviceMem;
using ::ck::hip_check_error;
using ::ck::HostTensorDescriptor;
Expand Down Expand Up @@ -47,9 +49,9 @@ using B0DataType = F16;
using BsDataType = ck::Tuple<B0DataType>;
using AccDataType = F32;
using CShuffleDataType = F32;
using D0DataType = F32;
using D0DataType = F16;
using DsDataType = ck::Tuple<D0DataType>;
using EDataType = F32;
using EDataType = F16;

using A0Layout = Row;
using A1Layout = Row;
Expand Down Expand Up @@ -394,7 +396,7 @@ int main(int argc, char* argv[])
{
printf("arg1: verification (0=no, 1=yes)\n");
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
printf("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg3: time kernel (0=no, 1=yes)\n");
printf("arg4: k_batch (>0)\n");
exit(0);
}
Expand Down
Loading