ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-24 06:44:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

WMMA grouped conv fwd large tensor extra flavors (#3582)

Browse Source 
* Additional flavors for WMMA conv fwd large tensor

- added F16/BF16 clamp operation
- added F16/BF16 bias_clamp operation
- small modification to the device code to accomodate extra tensors

* changed strategy to handle GemmArgs array

* Adding generic instance

* Added generic instance to clamp and bias_clamp ops
This commit is contained in:
Wojciech Laskowski
2026-01-23 12:19:51 +01:00
committed by GitHub
parent 7b3db1a878
commit 81ee19bd2c
27 changed files with 1007 additions and 171 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_large_tensor_instance.hpp
View File

@@ -29,12 +29,32 @@ using S = ck::Sequence<Is...>;
using namespace ck::tensor_layout::convolution;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using AddClamp = ck::tensor_operation::element_wise::AddClamp;
using Clamp = ck::tensor_operation::element_wise::Clamp;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
template <index_t NDSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
typename DsDataType = Empty_Tuple,
typename CDEElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_large_tensor_f16_generic_instances = std::tuple<
// clang-format off
//########################################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MWmma| NWmma| 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| Size| Block| Block| Block| | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN|MWmmaPerWave|NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector|
//########################################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1>
// clang-format on
>;
template <index_t NDSpatial,
typename ALayout,
typename BLayout,
@@ -45,9 +65,10 @@ template <index_t NDSpatial,
typename CDEElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_large_tensor_f16_instances = std::tuple<
// clang-format off
//########################################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MRepeat| NRepeat| 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| Size| Block| Block| Block| | WMMA| WMMA| | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _MBlock_MWaveMPerXdl| ScalarPerVector|
//########################################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
//########################################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MWmma| NWmma| 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| Size| Block| Block| Block| | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN|MWmmaPerWave|NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector|
//########################################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 16, 16, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 1>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 64, 64, 64, 8, 16, 16, 2, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 1>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 128, 128, 32, 8, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 1>,
@@ -56,6 +77,24 @@ using device_grouped_conv_fwd_wmma_large_tensor_f16_instances = std::tuple<
// clang-format on
>;
template <index_t NDSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
typename DsDataType = Empty_Tuple,
typename CDEElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_large_tensor_bf16_generic_instances = std::tuple<
// clang-format off
//########################################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MWmma| NWmma| 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| Size| Block| Block| Block| | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN|MWmmaPerWave|NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector|
//########################################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataType, BF16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 64, 64, 64, 32, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1>
// clang-format on
>;
template <index_t NDSpatial,
typename ALayout,
typename BLayout,
@@ -66,9 +105,10 @@ template <index_t NDSpatial,
typename CDEElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_large_tensor_bf16_instances = std::tuple<
// clang-format off
//########################################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MRepeat| NRepeat| 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| Size| Block| Block| Block| | WMMA| WMMA| | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _MBlock_MWaveMPerXdl| ScalarPerVector|
//########################################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
//########################################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MWmma| NWmma| 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| Size| Block| Block| Block| | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN|MWmmaPerWave|NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector|
//########################################################| | | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataType, BF16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 16, 16, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 1>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataType, BF16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 64, 64, 64, 8, 16, 16, 2, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 1>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataType, BF16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 128, 128, 32, 8, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 1>,

24
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_clamp.hpp
View File

@@ -293,8 +293,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part4(
op_ptrs);
// add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_generic_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
@@ -310,8 +312,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part4(
op_ptrs);
// add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_generic_instances(
op_ptrs);
}
#endif
}
@@ -334,8 +338,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part4(
op_ptrs);
// add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
// op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_generic_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
@@ -351,8 +357,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part4(
op_ptrs);
// add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
// op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_generic_instances(
op_ptrs);
}
#endif
}

164
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_clamp_wmma_cshufflev3.inc
View File

@@ -63,20 +63,33 @@ void add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_
PassThrough,
AddClamp>>>& instances);
// void
// add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Tuple<NHWGK>,
// NHWGK,
// BF16,
// BF16,
// Tuple<BF16>,
// BF16,
// PassThrough,
// PassThrough,
// AddClamp>>>& instances);
void add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
BF16,
BF16,
Tuple<BF16>,
BF16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
BF16,
BF16,
Tuple<BF16>,
BF16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part1(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
@@ -131,20 +144,33 @@ void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhw
PassThrough,
AddClamp>>>& instances);
// void
// add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Tuple<NDHWGK>,
// NDHWGK,
// BF16,
// BF16,
// Tuple<BF16>,
// BF16,
// PassThrough,
// PassThrough,
// AddClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
BF16,
BF16,
Tuple<BF16>,
BF16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
BF16,
BF16,
Tuple<BF16>,
BF16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
#endif
@@ -203,20 +229,33 @@ void add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_
PassThrough,
AddClamp>>>& instances);
// void
// add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Tuple<NHWGK>,
// NHWGK,
// F16,
// F16,
// Tuple<F16>,
// F16,
// PassThrough,
// PassThrough,
// AddClamp>>>& instances);
void add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
F16,
F16,
Tuple<F16>,
F16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv2d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
F16,
F16,
Tuple<F16>,
F16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part1(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
@@ -271,20 +310,33 @@ void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhw
PassThrough,
AddClamp>>>& instances);
// void
// add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Tuple<NDHWGK>,
// NDHWGK,
// F16,
// F16,
// Tuple<F16>,
// F16,
// PassThrough,
// PassThrough,
// AddClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
F16,
F16,
Tuple<F16>,
F16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK>,
NDHWGK,
F16,
F16,
Tuple<F16>,
F16,
PassThrough,
PassThrough,
AddClamp>>>& instances);
#endif

24
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_clamp.hpp
View File

@@ -290,8 +290,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part4(
op_ptrs);
// add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_generic_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
@@ -307,8 +309,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part4(
op_ptrs);
// add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_generic_instances(
op_ptrs);
}
#endif
}
@@ -331,8 +335,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part4(
op_ptrs);
// add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
// op_ptrs);
add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_generic_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
@@ -348,8 +354,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part4(
op_ptrs);
// add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
// op_ptrs);
add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
op_ptrs);
add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_generic_instances(
op_ptrs);
}
#endif
}

164
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_clamp_wmma_cshufflev3.inc
View File

@@ -63,20 +63,33 @@ void add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_
PassThrough,
Clamp>>>& instances);
// void
// add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Tuple<>,
// NHWGK,
// BF16,
// BF16,
// Tuple<>,
// BF16,
// PassThrough,
// PassThrough,
// Clamp>>>& instances);
void add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
BF16,
BF16,
Tuple<>,
BF16,
PassThrough,
PassThrough,
Clamp>>>& instances);
void add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
BF16,
BF16,
Tuple<>,
BF16,
PassThrough,
PassThrough,
Clamp>>>& instances);
void add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part1(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
@@ -131,20 +144,33 @@ void add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf
PassThrough,
Clamp>>>& instances);
// void
// add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Tuple<>,
// NDHWGK,
// BF16,
// BF16,
// Tuple<>,
// BF16,
// PassThrough,
// PassThrough,
// Clamp>>>& instances);
void add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
BF16,
BF16,
Tuple<>,
BF16,
PassThrough,
PassThrough,
Clamp>>>& instances);
void add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
BF16,
BF16,
Tuple<>,
BF16,
PassThrough,
PassThrough,
Clamp>>>& instances);
#endif
@@ -256,35 +282,61 @@ void add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f1
PassThrough,
Clamp>>>& instances);
// void
// add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Tuple<>,
// NHWGK,
// F16,
// F16,
// Tuple<>,
// F16,
// PassThrough,
// PassThrough,
// Clamp>>>& instances);
void add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
F16,
F16,
Tuple<>,
F16,
PassThrough,
PassThrough,
Clamp>>>& instances);
// void
// add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Tuple<>,
// NDHWGK,
// F16,
// F16,
// Tuple<>,
// F16,
// PassThrough,
// PassThrough,
// Clamp>>>& instances);
void add_device_grouped_conv2d_fwd_clamp_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
F16,
F16,
Tuple<>,
F16,
PassThrough,
PassThrough,
Clamp>>>& instances);
void add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
F16,
F16,
Tuple<>,
F16,
PassThrough,
PassThrough,
Clamp>>>& instances);
void add_device_grouped_conv3d_fwd_clamp_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_generic_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
F16,
F16,
Tuple<>,
F16,
PassThrough,
PassThrough,
Clamp>>>& instances);
#endif