1234567891011 |
- // Copyright (c) 2023, Tri Dao.
- // Splitting the different head dimensions to different files to speed up
- // compilation. This file is auto-generated. See "generate_kernels.py"
- #include "flash_fwd_launch_template.h"
- template <>
- void run_mha_fwd_<cutlass::half_t, 224, false>(Flash_fwd_params& params,
- cudaStream_t stream) {
- run_mha_fwd_hdim224<cutlass::half_t, false>(params, stream);
- }
|