flash_fwd_hdim32_fp16_sm80.cu 435 B

1234567891011
  1. // Copyright (c) 2023, Tri Dao.
  2. // Splitting the different head dimensions to different files to speed up
  3. // compilation. This file is auto-generated. See "generate_kernels.py"
  4. #include "flash_fwd_launch_template.h"
  5. template <>
  6. void run_mha_fwd_<cutlass::half_t, 32, false>(Flash_fwd_params& params,
  7. cudaStream_t stream) {
  8. run_mha_fwd_hdim32<cutlass::half_t, false>(params, stream);
  9. }