ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 14:59:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK][Examples] Extending support for rdna3/4 part 3:

Browse Source 
-example_gemm_xdl_int8
-example_gemm_xdl_fp8
-example_gemm_xdl_fp8_bf8
-example_gemm_xdl_fp16_fp8
-example_gemm_add_add_fastgelu_xdl_int8
-example_grouped_gemm_xdl_int8
-example_grouped_conv_bwd_weight_xdl_bf16
-example_cgemm_xdl_fp32
-example_cgemm_xdl_int8

fixing cmdlines for:
-example_22_cgemm
-example_24_batched_gemm
-example_batched_gemm_xdl_fp16int4_b_scale_v3

Signed-off-by: Michal Kulikowski <Michal.Kulikowski@amd.com>
This commit is contained in:
Michal Kulikowski
2025-10-03 15:30:40 +02:00
committed by Michał Kulikowski
parent 7259b9c4db
commit 2444c44895
13 changed files with 138 additions and 94 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
example/22_cgemm/cgemm_xdl_fp32.cpp
View File

@@ -1,5 +1,5 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
@@ -48,10 +48,10 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_
16, // index_t KPerBlock
4, // index_t AK1
4, // index_t BK1
32, // index_t MPerXDL
32, // index_t NPerXDL
4, // index_t MXdlPerWave
2, // index_t NXdlPerWave
16, // index_t MPerXDL
16, // index_t NPerXDL
8, // index_t MXdlPerWave
4, // index_t NXdlPerWave
S<4, 64, 1>, // typename ABlockTransferThreadClusterLengths_AK0_M_AK1
S<1, 0, 2>, // typename ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // typename ABlockTransferSrcAccessOrder
@@ -69,11 +69,16 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_
1, // index_t CShuffleMXdlPerWavePerShuffle
1, // index_t CShuffleNXdlPerWavePerShuffle
S<1, 16, 1, 16>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
4>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock
2>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock
// clang-format on
int main(int argc, char* argv[])
{
if(ck::is_gfx11_supported() || ck::is_gfx12_supported())
{
return 0;
}
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
@@ -87,25 +92,25 @@ int main(int argc, char* argv[])
ck::index_t StrideB = 4096;
ck::index_t StrideC = 4096;
if(argc == 4)
if(argc == 1)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
// use default case
}
else if(argc == 10)
else if(argc == 4 || argc == 10)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
if(argc == 10)
{
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
K = std::stoi(argv[6]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
K = std::stoi(argv[6]);
StrideA = std::stoi(argv[7]);
StrideB = std::stoi(argv[8]);
StrideC = std::stoi(argv[9]);
StrideA = std::stoi(argv[7]);
StrideB = std::stoi(argv[8]);
StrideC = std::stoi(argv[9]);
}
}
else
{
@@ -114,7 +119,7 @@ int main(int argc, char* argv[])
<< "arg3: run kernel # of times (>1)\n"
<< "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n"
<< std::endl;
exit(0);
exit(1);
}
return !run_cgemm_xdl<ADataType,

40
example/22_cgemm/cgemm_xdl_int8.cpp
View File

@@ -1,5 +1,5 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
@@ -48,10 +48,10 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_
64, // index_t KPerBlock
16, // index_t AK1
16, // index_t BK1
32, // index_t MPerXDL
32, // index_t NPerXDL
4, // index_t MXdlPerWave
2, // index_t NXdlPerWave
16, // index_t MPerXDL
16, // index_t NPerXDL
8, // index_t MXdlPerWave
4, // index_t NXdlPerWave
S<4, 64, 1>, // typename ABlockTransferThreadClusterLengths_AK0_M_AK1
S<1, 0, 2>, // typename ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // typename ABlockTransferSrcAccessOrder
@@ -68,8 +68,8 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_
1, // index_t BBlockLdsExtraN
1, // index_t CShuffleMXdlPerWavePerShuffle
1, // index_t CShuffleNXdlPerWavePerShuffle
S<1, 64, 1, 4>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
16>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock
S<1, 32, 1, 8>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
4>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock
// clang-format on
int main(int argc, char* argv[])
@@ -87,25 +87,25 @@ int main(int argc, char* argv[])
ck::index_t StrideB = 4096;
ck::index_t StrideC = 4096;
if(argc == 4)
if(argc == 1)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
// use default case
}
else if(argc == 10)
else if(argc == 4 || argc == 10)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
if(argc == 10)
{
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
K = std::stoi(argv[6]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
K = std::stoi(argv[6]);
StrideA = std::stoi(argv[7]);
StrideB = std::stoi(argv[8]);
StrideC = std::stoi(argv[9]);
StrideA = std::stoi(argv[7]);
StrideB = std::stoi(argv[8]);
StrideC = std::stoi(argv[9]);
}
}
else
{
@@ -114,7 +114,7 @@ int main(int argc, char* argv[])
<< "arg3: run kernel # of times (>1)\n"
<< "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n"
<< std::endl;
exit(0);
exit(1);
}
return !run_cgemm_xdl<ADataType,