ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-18 12:00:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

conv:tf32:add missed instances (#3081)

Browse Source 
* conv:tf32:add missed instances

[ROCm/composable_kernel commit: 6bbc05e1bd]
This commit is contained in:
yinglu
2025-10-24 16:28:36 +08:00
committed by GitHub
parent e316ba18ed
commit 6a7861bbec
35 changed files with 1331 additions and 120 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_scale_instance.hpp
View File

@@ -159,25 +159,25 @@ using device_grouped_conv_bwd_data_xdl_scale_f32_tf32_instances =
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| _NPerBlock|
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// generic instance
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, 1, 1, 1, S<1, 16, 1, 4>, 1>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, 1, 1, 1, S<1, 16, 1, 4>, 1, make_default_loop_scheduler(), TF32, TF32>,
// instances for small conv.K and conv.C
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, 1, 1, 1, S<1, 32, 1, 4>, 1>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, 1, 1, 1, S<1, 32, 1, 4>, 1, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 8>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8>
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 4>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 4>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 32, 1, 4>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 8>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 1, 1, 1, S<1, 16, 1, 4>, 4, make_default_loop_scheduler(), TF32, TF32>,
DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F32, F32, F32, F32, Tuple<>, F32, PassThrough, PassThrough, Scale, ConvSpec, true, true, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, make_default_loop_scheduler(), TF32, TF32>
// clang-format on
>;

21
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp
View File

@@ -367,6 +367,27 @@ using device_grouped_conv_fwd_xdl_f32_16x16_instances = std::tuple<
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
typename DsDataTypes = Tuple<>,
typename OutElementOp = PassThrough>
using device_grouped_conv_fwd_xdl_f32_tf32_16x16_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| 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|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| 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|
//########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// generic instance
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F32, F32, F32, F32, DsDataTypes, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 64, 64, 32, 8, 8, 16, 16, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 1, TF32, TF32>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F32, F32, F32, F32, DsDataTypes, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 64, 64, 32, 8, 8, 16, 16, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 4>, 2, TF32, TF32>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F32, F32, F32, F32, DsDataTypes, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 64, 64, 32, 8, 8, 16, 16, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, 1, 1, S<1, 32, 1, 4>, 4, TF32, TF32>
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,

3
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
View File

@@ -286,8 +286,7 @@ struct DeviceOperationInstanceFactory<
#endif
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
is_same_v<OutDataType, F32> && is_same_v<ComputeTypeA, F32> &&
is_same_v<ComputeTypeB, F32>)
is_same_v<OutDataType, F32>)
{
static_assert(is_same_v<ComputeTypeA, ComputeTypeB>,
"Error: this operator requires the same compute type");

33
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data_bilinear.hpp
View File

@@ -44,6 +44,22 @@ void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f32_in
PassThrough,
PassThrough,
Bilinear>>>& instances);
void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f32_tf32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
Tuple<NDHWGC>,
NDHWGC,
F32,
F32,
Tuple<F32>,
F32,
PassThrough,
PassThrough,
Bilinear,
TF32,
TF32>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
@@ -121,11 +137,20 @@ struct DeviceOperationInstanceFactory<
#endif
#ifdef CK_ENABLE_FP32
else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
is_same_v<OutDataType, F32> && is_same_v<ComputeTypeA, F32> &&
is_same_v<ComputeTypeB, F32>)
is_same_v<OutDataType, F32>)
{
add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f32_instances(
op_ptrs);
static_assert(is_same_v<ComputeTypeA, ComputeTypeB>,
"ComputeTypeA and ComputeTypeB must be the same");
if constexpr(is_same_v<ComputeTypeA, F32>)
{
add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f32_instances(
op_ptrs);
}
else if constexpr(is_same_v<ComputeTypeA, TF32>)
{
add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f32_tf32_instances(
op_ptrs);
}
}
#endif
#ifdef CK_ENABLE_BF16

33
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data_scale.hpp
View File

@@ -44,6 +44,22 @@ void add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_f32_insta
PassThrough,
PassThrough,
Scale>>>& instances);
void add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_f32_tf32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
Tuple<>,
NDHWGC,
F32,
F32,
Tuple<>,
F32,
PassThrough,
PassThrough,
Scale,
TF32,
TF32>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
@@ -121,11 +137,20 @@ struct DeviceOperationInstanceFactory<
#endif
#ifdef CK_ENABLE_FP32
else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
is_same_v<OutDataType, F32> && is_same_v<ComputeTypeA, F32> &&
is_same_v<ComputeTypeB, F32>)
is_same_v<OutDataType, F32>)
{
add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_f32_instances(
op_ptrs);
static_assert(is_same_v<ComputeTypeA, ComputeTypeB>,
" only support same compute type");
if constexpr(is_same_v<ComputeTypeA, F32>)
{
add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_f32_instances(
op_ptrs);
}
else if constexpr(is_same_v<ComputeTypeA, TF32>)
{
add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_f32_tf32_instances(
op_ptrs);
}
}
#endif
#ifdef CK_ENABLE_BF16

49
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
View File

@@ -200,20 +200,41 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
is_same_v<OutDataType, float> && is_same_v<AComputeType, float> &&
is_same_v<BComputeType, float>)
is_same_v<OutDataType, float>)
{
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_16x16_instances(op_ptrs);
add_device_grouped_conv2d_fwd_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_comp_instances(op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_mem_inter_instances(
op_ptrs);
static_assert(is_same_v<AComputeType, BComputeType>,
"Error: AComputeType and BComputeType should be the same!");
if constexpr(is_same_v<AComputeType, float>)
{
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_16x16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_comp_instances(op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_mem_inter_instances(
op_ptrs);
}
else if constexpr(is_same_v<AComputeType, TF32>)
{
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_instances(op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_mem_inter_instances(
op_ptrs);
}
}
#endif
#ifdef CK_ENABLE_FP16
@@ -434,6 +455,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_xdl_merged_groups_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_comp_instances(

108
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_bnorm_clamp.hpp
View File

@@ -127,23 +127,45 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
#endif
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
is_same_v<OutDataType, float> && is_same_v<AComputeType, float> &&
is_same_v<BComputeType, float>)
is_same_v<OutDataType, float>)
{
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_16x16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_mem_inter_instances(
op_ptrs);
static_assert(is_same_v<AComputeType, BComputeType>,
"A and B compute types should be the same");
if constexpr(is_same_v<AComputeType, float>)
{
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_16x16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_mem_inter_instances(
op_ptrs);
}
else if constexpr(is_same_v<AComputeType, TF32>)
{
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_mem_inter_instances(
op_ptrs);
}
}
#endif
}
@@ -205,23 +227,45 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
#endif
#ifdef CK_ENABLE_FP32
if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
is_same_v<OutDataType, float> && is_same_v<AComputeType, float> &&
is_same_v<BComputeType, float>)
is_same_v<OutDataType, float>)
{
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16x16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_merged_groups_ndhwgc_gkzyxc_ndhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_comp_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_inter_instances(
op_ptrs);
static_assert(is_same_v<AComputeType, BComputeType>,
"A and B compute types should be the same");
if constexpr(is_same_v<AComputeType, float>)
{
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16x16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_merged_groups_ndhwgc_gkzyxc_ndhwgk_f32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_comp_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_inter_instances(
op_ptrs);
}
else if constexpr(is_same_v<AComputeType, TF32>)
{
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_merged_groups_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_comp_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_mem_intra_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_mem_inter_instances(
op_ptrs);
}
}
#endif
}

237
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_bnorm_clamp_xdl.inc
View File

@@ -663,6 +663,125 @@ void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_mem_i
PassThrough,
BiasNormalizeInInferClamp>>>& instances);
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
std::vector<
std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
std::vector<
std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
std::vector<
std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
std::vector<
std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_comp_instances(
std::vector<
std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_mem_intra_instances(
std::vector<
std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_mem_inter_instances(
std::vector<
std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
@@ -768,6 +887,124 @@ void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_me
PassThrough,
BiasNormalizeInInferClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_merged_groups_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_comp_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_mem_intra_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_mem_inter_instances(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32, F32, F32, F32, F32>,
F32,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp,
TF32,
TF32>>>& instances);
#endif
} // namespace instance

4
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_clamp.hpp
View File

@@ -135,6 +135,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
add_device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_comp_instances(
@@ -225,6 +227,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
add_device_grouped_conv3d_fwd_bias_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_xdl_merged_groups_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_comp_instances(

32
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_clamp_xdl.inc
View File

@@ -510,6 +510,22 @@ void add_device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_16x16_in
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
F32,
F32,
Tuple<F32>,
F32,
PassThrough,
PassThrough,
AddClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_bias_clamp_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
@@ -720,6 +736,22 @@ void add_device_grouped_conv3d_fwd_bias_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16x16
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
F32,
F32,
Tuple<F32>,
F32,
PassThrough,
PassThrough,
AddClamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,

4
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_clamp.hpp
View File

@@ -133,6 +133,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
add_device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_comp_instances(
@@ -222,6 +224,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
{
add_device_grouped_conv3d_fwd_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_clamp_xdl_merged_groups_ndhwgc_gkzyxc_ndhwgk_f32_tf32_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_comp_instances(

32
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_clamp_xdl.inc
View File

@@ -510,6 +510,22 @@ void add_device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_16x16_instanc
PassThrough,
Clamp>>>& instances);
void add_device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
F32,
F32,
Tuple<>,
F32,
PassThrough,
PassThrough,
Clamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_clamp_xdl_large_tensor_nhwgc_gkyxc_nhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
@@ -704,6 +720,22 @@ void add_device_grouped_conv3d_fwd_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16x16_inst
PassThrough,
Clamp>>>& instances);
void add_device_grouped_conv3d_fwd_clamp_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
F32,
F32,
Tuple<>,
F32,
PassThrough,
PassThrough,
Clamp,
TF32,
TF32>>>& instances);
void add_device_grouped_conv3d_fwd_clamp_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,

32
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_xdl.inc
View File

@@ -227,6 +227,22 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_instances(
PassThrough,
TF32,
TF32>>>& instances);
void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_16x16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough,
TF32,
TF32>>>& instances);
#endif
#ifdef CK_ENABLE_INT8
@@ -604,6 +620,22 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16x16_instances(
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_tf32_16x16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough,
TF32,
TF32>>>& instances);
#endif
#ifdef CK_ENABLE_INT8