ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-30 19:57:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add instances for scale and bilinear based on the bf16 NHWGC GKYXC tuning. Keep generic instances for support.

Browse Source
This commit is contained in:
kiefer
2025-12-05 13:37:51 +00:00
parent c2fab9d677
commit 6380f9ab81
2 changed files with 56 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

67
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_bilinear_instance.hpp
View File

@@ -45,27 +45,21 @@ using device_grouped_conv_bwd_weight_wmma_c_shuffle_f16_bilinear_instances = std
//#################################################| Spatial| | | | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| MBlock_MPerBlock| _NPerBlock| Scheduler| Version |
//#################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | |
// generic instance
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, true, 1, 1, S<1, 16, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, true, 1, 1, S<1, 16, 1, 4>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// // instance for small conv.K
// // for fp16 conv.K and conv.C must be divisible by 2
// // since half_t atomic_add require scalar_per_x_vector % 2 == 0
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 1, true, 1, 1, S<1, 32, 1, 4>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 256, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 256, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 128, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 64, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 64, 128, 32, 8, 16, 16, 2, 4, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 64, 32, 8, 16, 16, 2, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 64, 128, 32, 8, 16, 16, 2, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 32, 128, 32, 8, 16, 16, 1, 4, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 16, 16, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, 1, 1, 1, S<1, 16, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, 1, 1, 1, S<1, 16, 1, 4>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// for fp16 conv.K and conv.C must be divisible by 2
// since half_t atomic_add require scalar_per_x_vector % 2 == 0
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 16, 16, 2, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, 1, 1, S<1, 8, 1, 8>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 64, 8, 16, 16, 4, 2, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 256, 64, 8, 16, 16, 8, 2, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 48, 64, 128, 8, 16, 16, 3, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 96, 128, 64, 8, 16, 16, 6, 2, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 64, 64, 128, 8, 16, 16, 4, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 96, 128, 128, 8, 16, 16, 6, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>, // Presumably doesn't produce correct results for f16
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, F16, F16, F16, F32, Tuple<F16>, PassThrough, Bilinear, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1> // Presumably doesn't produce correct results for f16
// clang-format on
>;
@@ -81,24 +75,19 @@ using device_grouped_conv_bwd_weight_wmma_c_shuffle_bf16_bilinear_instances = st
//#################################################| Spatial| | | | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| MBlock_MPerBlock| _NPerBlock| Scheduler| Version |
//#################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | |
// generic instance
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, true, 1, 1, S<1, 16, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// // instance for small conv.K
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 1, true, 1, 1, S<1, 32, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 256, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 256, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 128, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 64, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 64, 128, 32, 8, 16, 16, 2, 4, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 64, 32, 8, 16, 16, 2, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 64, 128, 32, 8, 16, 16, 2, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 32, 128, 32, 8, 16, 16, 1, 4, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 16, 16, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, 1, 1, 1, S<1, 16, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// other instances
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 16, 16, 2, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, 1, 1, S<1, 8, 1, 8>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 64, 64, 64, 64, 8, 16, 16, 4, 2, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 128, 256, 64, 8, 16, 16, 8, 2, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 48, 64, 128, 8, 16, 16, 3, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 96, 128, 64, 8, 16, 16, 6, 2, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 128, 64, 64, 128, 8, 16, 16, 4, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 256, 96, 128, 128, 8, 16, 16, 6, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Tuple<BLayout>, BF16, F32, BF16, F32, Tuple<F32>, PassThrough, Bilinear, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// clang-format on
>;

67
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_scale_instance.hpp
View File

@@ -45,27 +45,21 @@ using device_grouped_conv_bwd_weight_wmma_c_shuffle_f16_scale_instances = std::t
//#################################################| Spatial| | | | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| MBlock_MPerBlock| _NPerBlock| Scheduler| Version |
//#################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | |
// generic instance
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, true, 1, 1, S<1, 16, 1, 4>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// // instance for small conv.K
// // for fp16 conv.K and conv.C must be divisible by 2
// // since half_t atomic_add require scalar_per_x_vector % 2 == 0
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 1, true, 1, 1, S<1, 32, 1, 4>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 256, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 256, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 128, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 64, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 64, 128, 32, 8, 16, 16, 2, 4, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 64, 32, 8, 16, 16, 2, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 64, 128, 32, 8, 16, 16, 2, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 32, 128, 32, 8, 16, 16, 1, 4, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 16, 16, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// clang-format on
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 4, 1, 1, 1, S<1, 16, 1, 4>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// for fp16 conv.K and conv.C must be divisible by 2
// since half_t atomic_add require scalar_per_x_vector % 2 == 0
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 16, 16, 2, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, 1, 1, S<1, 8, 1, 8>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 64, 8, 16, 16, 4, 2, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 256, 64, 8, 16, 16, 8, 2, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 48, 64, 128, 8, 16, 16, 3, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 96, 128, 64, 8, 16, 16, 6, 2, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 64, 64, 128, 8, 16, 16, 4, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 96, 128, 128, 8, 16, 16, 6, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>, // Presumably doesn't produce correct results for fp16
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, F16, F16, F16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1> // Presumably doesn't produce correct results for fp16
// clang-format on
>;
template <ck::index_t NDimSpatial,
@@ -80,24 +74,19 @@ using device_grouped_conv_bwd_weight_wmma_c_shuffle_bf16_scale_instances = std::
//#################################################| Spatial| | | | | | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| MBlock_MPerBlock| _NPerBlock| Scheduler| Version |
//#################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| | | |
// generic instance
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, true, 1, 1, S<1, 16, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// // instance for small conv.K
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 1, true, 1, 1, S<1, 32, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 256, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 256, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 4, 4, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 128, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 64, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 64, 128, 32, 8, 16, 16, 2, 4, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 64, 32, 8, 16, 16, 2, 2, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 64, 128, 32, 8, 16, 16, 2, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 32, 32, 8, 16, 16, 4, 1, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 32, 128, 32, 8, 16, 16, 1, 4, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 1, true, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 32, 32, 8, 16, 16, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 32, 64, 32, 8, 16, 16, 2, 2, S<4, 4, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 2, true, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 1, 4, 1, 1, 1, S<1, 16, 1, 4>, 1, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
// other instances
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 32, 32, 32, 8, 16, 16, 2, 1, S<4, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, S<4, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 2, 2, 0, 1, 1, S<1, 8, 1, 8>, 2, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 128, 128, 32, 8, 16, 16, 8, 2, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<4, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 64, 64, 64, 64, 8, 16, 16, 4, 2, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<8, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 128, 256, 64, 8, 16, 16, 8, 2, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 4, 8, 1, S<8, 32, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 48, 64, 128, 8, 16, 16, 3, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 96, 128, 64, 8, 16, 16, 6, 2, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<8, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 128, 64, 64, 128, 8, 16, 16, 4, 1, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, S<16, 8, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 256, 96, 128, 128, 8, 16, 16, 6, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<16, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>,
DeviceGroupedConvBwdWeightMultipleD_Wmma_CShuffleV3< NDimSpatial, ALayout, BLayout, ELayout, Empty_Tuple, BF16, F32, BF16, F32, Empty_Tuple, PassThrough, Scale, PassThrough, ConvSpec, 96, 96, 96, 48, 8, 16, 16, 6, 2, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 1, S<6, 16, 1>, S<2, 0, 1>, S<1, 0, 2>, 1, 6, 8, 0, 1, 1, S<1, 16, 1, 6>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// clang-format on
>;