flash-attention/csrc/flash_attn/src/flash_bwd_hdim256_fp16_sm80.cu

11 lines
411 B
Plaintext

// 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_bwd_launch_template.h"
template<>
void run_mha_bwd_<cutlass::half_t, 256>(Flash_bwd_params &params, cudaStream_t stream, const bool configure) {
run_mha_bwd_hdim256<cutlass::half_t>(params, stream, configure);
}