// 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" | |
namespace FLASH_NAMESPACE { | |
template<> | |
void run_mha_fwd_<cutlass::half_t, 192, true>(Flash_fwd_params ¶ms, cudaStream_t stream) { | |
run_mha_fwd_hdim192<cutlass::half_t, true>(params, stream); | |
} | |
} // namespace FLASH_NAMESPACE |