11 lines
400 B
Plaintext
11 lines
400 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_fwd_launch_template.h"
|
|
|
|
template<>
|
|
void run_mha_fwd_<cutlass::bfloat16_t, 256, false>(Flash_fwd_params ¶ms, cudaStream_t stream) {
|
|
run_mha_fwd_hdim256<cutlass::bfloat16_t, false>(params, stream);
|
|
}
|