5018ac6ac5
* base version * restructure pipelines, add special fp8 epilogue * add variants * add fp8 causal and modify dynamic tile scheduler * better causal schedule * maintain two schedules for non causal and causal * removing macros * fix regression * clean up unneeded methods and variants * fix mistake with NumProducerThreads * base version * restructure pipelines, add special fp8 epilogue * add variants * add fp8 causal and modify dynamic tile scheduler * better causal schedule * maintain two schedules for non causal and causal * removing macros * fix regression * clean up unneeded methods and variants * fix mistake with NumProducerThreads * use seqlen traits * add fp8 .cu files and benchmark script * fix merge issue * fix merge issue * fix merge issue * remove duplicate code * fix regression with varseqlen * move varseqlen init in constexpr * fix test script * more constexpr on varseqlen and add max offset * add back test cases
10 lines
334 B
Plaintext
10 lines
334 B
Plaintext
// Copyright (c) 2024, Tri Dao.
|
|
// Splitting the different head dimensions to different files to speed up compilation.
|
|
|
|
#include "flash_fwd_launch_template.h"
|
|
|
|
template<>
|
|
void run_mha_fwd_<cutlass::float_e4m3_t, 64>(Flash_fwd_params ¶ms, cudaStream_t stream) {
|
|
run_mha_fwd_hdim64_fp8<cutlass::float_e4m3_t>(params, stream);
|
|
}
|