ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-25 07:14:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

V5: experiment with multi-warp

Browse Source
This commit is contained in:
Damien Lejeune
2026-02-12 14:39:20 +00:00
parent 0d7a341d27
commit 11d1c40655
4 changed files with 37 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
example/ck_tile/42_mhc/mhc_v5_bf16_benchmark.cpp
View File

@@ -13,6 +13,7 @@
#include "ck_tile/host/kernel_launch.hpp"
#include "ck_tile/host/reference/reference_mhc.hpp"
#include "ck_tile/host/check_err.hpp"
#include "ck_tile/ops/mhc/pipeline/mhc_problem_v5_4warp.hpp"
// Parse command-line arguments for MHC benchmark
auto create_args(int argc, char* argv[])
@@ -95,6 +96,8 @@ bool run_mhc_benchmark_impl(const ck_tile::ArgParser& arg_parser)
d_phi_mem.ToDevice(h_phi.data());
d_output_mem.ToDevice(h_output.data());
// Reverted to adaptive 2-warp (4-warp produced incorrect results)
// using Problem = ck_tile::MHCProblemV5_4Warp<XDataType, ComputeDataType, YDataType>;
using Problem = ck_tile::MHCProblemV5<XDataType, ComputeDataType, YDataType, MTile>;
// V5 kernel - split-K implementation with adaptive problem