candle-flash-attn/kernels/flash_fwd_hdim32_bf16_causal_sm80.cu


1
2
3
4
5
6
7
8
9
10

// Copyright (c) 2024, 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, true>(Flash_fwd_params &params, cudaStream_t stream) {
    run_mha_fwd_hdim32<cutlass::bfloat16_t, true>(params, stream);
}