// Copyright (c) 2024, Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, Tri Dao. // Splitting the different template instantiations to different files to speed up compilation. // This file is auto-generated. See "generate_kernels.py" #include "flash_fwd_launch_template.h" #ifndef FLASHATTENTION_DISABLE_HDIM192 template void run_mha_fwd_<90, cutlass::bfloat16_t, 192, true, false, false, true>(Flash_fwd_params ¶ms, cudaStream_t stream); #endif