Skip to content

GitLab

Commit 648f1f13 authored by Adam Osewski's avatar Adam Osewski
Browse files

Merge remote-tracking branch 'origin/develop' into aosewski/gemm_tile_loop

parents 4e5190f5 cb538740
Hide whitespace changes
Inline Side-by-side
Showing with 373 additions and 125 deletions
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
View file @ 648f1f13
......@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// MPerBlock=128, NPerBlock=128
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<4, 4>, S<4, 4>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<2, 8>, S<2, 8>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<4, 4>, S<4, 4>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 128, 16, 2, 4, 4, 1, S<2, 8>, S<2, 8>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=64
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 2, 4, 4, 1, S<4, 2>, S<4, 2>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 2, 4, 4, 1, S<2, 4>, S<2, 4>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 2, 4, 4, 1, S<4, 2>, S<8, 1>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 2, 4, 4, 1, S<4, 2>, S<4, 2>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 2, 4, 4, 1, S<2, 4>, S<2, 4>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 2, 4, 4, 1, S<4, 2>, S<8, 1>, S<4, 1, 2, 2>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<2, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=16, NPerBlock=64
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 64, 16, 2, 1, 4, 1, S<2, 4>, S<2, 4>, S<4, 1, 1, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 16, 64, 16, 2, 1, 4, 1, S<2, 4>, S<2, 4>, S<4, 1, 1, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=16
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 16, 16, 2, 4, 1, 1, S<4, 2>, S<4, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<16, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 16, 16, 2, 4, 1, 1, S<4, 2>, S<4, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<16, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=64
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<4, 1>, S<8, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<2, 2>, S<8, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<4, 1>, S<8, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<2, 2>, S<8, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=8
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<4, 1>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<4, 1>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=8
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<2, 1>, S<4, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<1, 2>, S<1, 4>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<2, 1>, S<4, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<1, 2>, S<1, 4>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>
// clang-format on
>;
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
View file @ 648f1f13
......@@ -32,36 +32,36 @@ using device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances = std::tuple<
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// MPerBlock=128, NPerBlock=128
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<8, 2>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<4, 4>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<2, 8>, S<2, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<8, 2>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<4, 4>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<2, 8>, S<2, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=128, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=128
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<2, 4>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<8, 1>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<8, 1>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<2, 4>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<8, 1>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<8, 1>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=32, NPerBlock=32
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 2, 4, 1, S<4, 2>, S<2, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 2, 4, 1, S<4, 2>, S<2, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=16, NPerBlock=16
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 4, 2, 2, 1, S<2, 2>, S<2, 2>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 4, 2, 2, 1, S<4, 1>, S<4, 1>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 4, 2, 2, 1, S<2, 2>, S<2, 2>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 4, 2, 2, 1, S<4, 1>, S<4, 1>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 4, 4>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 4, 1, 2, 1, S<2, 2>, S<8, 2>, S<1, 1, 4, 4>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 4, 1, 2, 1, S<2, 2>, S<8, 2>, S<1, 1, 4, 4>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=8
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 4, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 4, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=8
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 1, 2, 1, S<4, 1>, S<2, 1>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 1, 2, 1, S<1, 4>, S<1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 2, 1, 1, S<2, 1>, S<4, 1>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 2, 1, 1, S<1, 2>, S<1, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 1, 2, 1, S<4, 1>, S<2, 1>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 1, 2, 1, S<1, 4>, S<1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 2, 1, 1, S<2, 1>, S<4, 1>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 4, 2, 1, 1, S<1, 2>, S<1, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<1, 1, 4, 4>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>
// clang-format on
>;
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
View file @ 648f1f13
......@@ -32,36 +32,36 @@ using device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances = std::tuple<
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// MPerBlock=128, NPerBlock=128
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<8, 2>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<4, 4>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<2, 8>, S<2, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<8, 2>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<4, 4>, S<4, 2>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<2, 8>, S<2, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=128, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=128
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<4, 4>, S<4, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<2, 8>, S<2, 8>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<2, 4>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<8, 1>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<8, 1>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<2, 4>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<8, 1>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<8, 1>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=32, NPerBlock=32
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 2, 4, 1, S<4, 2>, S<2, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 2, 4, 1, S<4, 2>, S<2, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<4, 2>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<2, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=16, NPerBlock=16
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<2, 2>, S<2, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<4, 1>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<2, 2>, S<2, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<4, 1>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<2, 2>, S<8, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<2, 2>, S<8, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=8
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=8
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<4, 1>, S<2, 1>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<1, 4>, S<1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<2, 1>, S<4, 1>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<1, 2>, S<1, 4>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<4, 1>, S<2, 1>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<1, 4>, S<1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<2, 1>, S<4, 1>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<1, 2>, S<1, 4>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>
// clang-format on
>;
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
View file @ 648f1f13
......@@ -32,36 +32,36 @@ using device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances = std::tuple<
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// MPerBlock=128, NPerBlock=128
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<8, 2>, S<4, 2>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<4, 4>, S<4, 2>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<2, 8>, S<2, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<8, 2>, S<4, 2>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<4, 4>, S<4, 2>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 128, 128, 16, 4, 4, 8, 1, S<2, 8>, S<2, 4>, S<8, 1, 2, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 8, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=128, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<4, 4>, S<4, 4>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<2, 8>, S<2, 8>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<4, 4>, S<4, 4>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 128, 64, 16, 4, 4, 2, 1, S<2, 8>, S<2, 8>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=128
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<4, 4>, S<4, 4>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<2, 8>, S<2, 8>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<4, 4>, S<4, 4>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 256, 64, 128, 16, 4, 2, 4, 1, S<2, 8>, S<2, 8>, S<8, 1, 1, 4>, S<2, 1, 64, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<4, 2>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<2, 4>, S<2, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<8, 1>, S<4, 2>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<8, 1>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<4, 2>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<2, 4>, S<2, 4>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<8, 1>, S<4, 2>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 64, 8, 4, 4, 4, 1, S<4, 2>, S<8, 1>, S<4, 1, 2, 4>, S<2, 1, 32, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=32, NPerBlock=32
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 2, 4, 1, S<4, 2>, S<2, 2>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<4, 2>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<2, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 2, 4, 1, S<4, 2>, S<2, 2>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<4, 2>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 32, 32, 32, 8, 4, 4, 2, 1, S<2, 2>, S<2, 4>, S<4, 1, 2, 4>, S<2, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<2, 1, 4, 4>, S<4, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=16, NPerBlock=16
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<2, 2>, S<2, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<4, 1>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<2, 2>, S<2, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 16, 16, 16, 16, 2, 2, 2, 1, S<4, 1>, S<4, 1>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 4, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=64
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<2, 2>, S<8, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 8, 64, 32, 2, 1, 2, 1, S<2, 2>, S<8, 2>, S<4, 1, 1, 2>, S<8, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=64, NPerBlock=8
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 64, 64, 8, 32, 2, 2, 1, 1, S<8, 2>, S<2, 2>, S<8, 1, 4, 2>, S<4, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<32, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
// MPerBlock=8, NPerBlock=8
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<4, 1>, S<2, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<1, 4>, S<1, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<2, 1>, S<4, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<1, 2>, S<1, 4>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<4, 1>, S<2, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 1, 2, 1, S<1, 4>, S<1, 2>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<2, 1>, S<4, 1>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>,
DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 8, 8, 8, 4, 2, 2, 1, 1, S<1, 2>, S<1, 4>, S<4, 1, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<4, 1, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<1, 1, 4, 2>, S<4, 1, 2, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>
// clang-format on
>;
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_kn_mn_instance.cpp 0 → 100644
View file @ 648f1f13
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef CK_ENABLE_FP8
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
using F8 = f8_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
using device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances =
std::tuple<
// clang-format off
//#####################| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//#####################| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//#####################| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 64, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 64, 4, 4, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 64, 16, 16, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 64, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 64, 4, 4, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 64, 4, 4, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 64, 16, 16, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 64, 4, 4, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 64, 16, 16, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>
// clang-format on
>;
void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_nk_mn_instance.cpp 0 → 100644
View file @ 648f1f13
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef CK_ENABLE_FP8
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
using F8 = f8_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
using device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances =
std::tuple<
// clang-format off
//#####################| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//#####################| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//#####################| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 64, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 64, 4, 16, 32, 32, 2, 4, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 64, 16, 16, 32, 32, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 64, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 64, 4, 16, 32, 32, 2, 2, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 64, 4, 16, 32, 32, 2, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 64, 16, 16, 32, 32, 2, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 64, 4, 16, 32, 32, 1, 2, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Col, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 64, 16, 16, 32, 32, 1, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>
// clang-format on
>;
void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp 0 → 100644
View file @ 648f1f13
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef CK_ENABLE_FP8
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
using F8 = f8_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances =
std::tuple<
// clang-format off
//#####################| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//#####################| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//#####################| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 64, 16, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 64, 16, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 64, 16, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 64, 16, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 64, 16, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 64, 16, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 16, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 64, 16, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<16,16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 64, 16, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 64, 16, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 64, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Row, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 64, 16, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 64, 1, 4>, 16>
// clang-format on
>;
void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp 0 → 100644
View file @ 648f1f13
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef CK_ENABLE_FP8
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
using F8 = f8_t;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances =
std::tuple<
// clang-format off
//#####################| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//#####################| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//#####################| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//#####################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 64, 16, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 64, 16, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 64, 64, 16, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 64, 16, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 64, 16, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 32, 64, 16, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 32, 128, 64, 16, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 64, 32, 64, 16, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 16>,
DeviceGemm_Xdl_CShuffle< Row, Col, Row, F8, F8, F8, F32, F8, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 64, 32, 64, 64, 16, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 16>
// clang-format on
>;
void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
View file @ 648f1f13
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_instance_library(device_gemm_add_add_fastgelu_instance
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
)
endif()
library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
View file @ 648f1f13
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_instance_library(device_gemm_add_fastgelu_instance
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
)
endif()
library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
View file @ 648f1f13
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_instance_library(device_gemm_add_relu_add_layernorm_instance
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
)
endif()
library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
View file @ 648f1f13
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_instance_library(device_gemm_bilinear_instance
device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
......@@ -9,4 +8,3 @@ add_instance_library(device_gemm_bilinear_instance
device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp
device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp
)
endif()
library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt
View file @ 648f1f13
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_instance_library(device_gemm_fastgelu_instance
device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
)
endif()
library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
View file @ 648f1f13
set(GEMM_MULTIPLY_ADD_INSTANCES)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp)
list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp)
endif()
if((DTYPES MATCHES "fp16" AND DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp)
list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp)
endif()
list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp)
add_instance_library(device_gemm_multiply_add_instance ${GEMM_MULTIPLY_ADD_INSTANCES})
library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
View file @ 648f1f13
set(GEMM_SPLITK_INSTANCES)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp)
endif()
if((DTYPES MATCHES "fp16" AND DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp)
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp)
endif()
list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_xdl_splitk_fp8_f16_f16_mk_kn_mn_instance.cpp
device_gemm_xdl_splitk_fp8_f16_f16_mk_nk_mn_instance.cpp
device_gemm_xdl_splitk_fp8_f16_f16_km_kn_mn_instance.cpp
device_gemm_xdl_splitk_fp8_f16_f16_km_nk_mn_instance.cpp
device_gemm_xdl_splitk_f16_fp8_f16_mk_kn_mn_instance.cpp
device_gemm_xdl_splitk_f16_fp8_f16_mk_nk_mn_instance.cpp
device_gemm_xdl_splitk_f16_fp8_f16_km_kn_mn_instance.cpp
device_gemm_xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp)
add_instance_library(device_gemm_splitk_instance ${GEMM_SPLITK_INSTANCES})
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment