ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-29 19:28:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

enable 128x256 tile for blockscale, for agpr control experiment.

Browse Source
This commit is contained in:
aska-0096
2025-06-13 03:44:47 +00:00
parent 5f1ad09b61
commit d0757cccca
3 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle_v3_blockscale_bpreshuffle.hpp
View File

@@ -234,7 +234,7 @@ struct DeviceGemmMultiD_BlockScale_Xdl_CShuffle_V3_BPreshuffle
};
// unconditional 2 to remove agpr usage
constexpr index_t minimum_occupancy = 2;
constexpr index_t minimum_occupancy = MPerBlock * NPerBlock * KPerBlock > (128*128*128/sizeof(ADataType))? 1: 2;
if(has_main_k_block_loop)
{