candle/candle-flash-attn/kernels/flash_fwd_hdim32_bf16_sm80.cu

11 lines
398 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, 32, false>(Flash_fwd_params &params, cudaStream_t stream) {
run_mha_fwd_hdim32<cutlass::bfloat16_t, false>(params, stream);
}