// Copyright (c) 2024, Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao. | |
// Splitting the different template instantiations to different files to speed up compilation. | |
// This file is auto-generated. See "generate_kernels.py" | |
template<> | |
void run_mha_bwd_<90, cutlass::bfloat16_t, 192, true>(Flash_bwd_params ¶ms, cudaStream_t stream) { | |
run_mha_bwd_hdim192<90, cutlass::bfloat16_t, true>(params, stream); | |
} | |