// Copyright (c) 2023, Tri Dao.

// Splitting the different head dimensions to different files to speed up compilation.

#include "flash_bwd_launch_template.h"

// template<>
// void run_mha_bwd_<cutlass::half_t, 32>(Flash_bwd_params &params, cudaStream_t stream, const bool configure) {
//     using elem_type = cutlass::half_t;
//     run_flash_bwd<Flash_bwd_kernel_traits<32, 128, 128, 8, 4, 4, 4, false, false, elem_type>>(params, stream, configure);
// }

template<>
void run_mha_bwd_<cutlass::half_t, 32>(Flash_bwd_params &params, cudaStream_t stream, const bool configure) {
    run_mha_bwd_hdim32<cutlass::half_t>(params, stream, configure);
}