ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-20 21:09:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '7795e73b47a34a25b48a14f3e4e0e6d681fcbde5' into develop

Browse Source
This commit is contained in:
assistant-librarian[bot]
2025-12-19 05:14:28 +00:00
parent 9073821d40
commit 0779609ddc
13 changed files with 1788 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

83
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_large_tensor_instance.hpp Normal file
View File

@@ -0,0 +1,83 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#pragma once
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle_v3_large_tensor.hpp"
#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using BF16 = ck::bhalf_t;
using F16 = ck::half_t;
using F32 = float;
using I8 = int8_t;
using I32 = int32_t;
using Empty_Tuple = ck::Tuple<>;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using namespace ck::tensor_layout::convolution;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
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_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| | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 64, 64, 8, 16, 16, 2, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 1>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 96, 64, 8, 16, 16, 2, 3, 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, 64, 1, 4>, 1>
// 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_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| | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
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>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataType, BF16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 128, 64, 64, 64, 8, 16, 16, 2, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 1>,
DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_V3_Large_Tensor<NDSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataType, BF16, PassThrough, PassThrough, CDEElementOp, ConvSpec, GemmMNKPadding, 256, 128, 96, 64, 8, 16, 16, 2, 3, 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, 64, 1, 4>, 1>
// clang-format on
>;
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

17
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
View File

@@ -31,6 +31,7 @@
#include "grouped_convolution_forward_wmma_cshufflev3.inc"
#include "grouped_convolution_forward_mem_inter_wmma_cshufflev3.inc"
#include "grouped_convolution_forward_mem_intra_wmma_cshufflev3.inc"
#include "grouped_convolution_forward_wmma_large_tensor.inc"
#endif
namespace ck {
@@ -794,8 +795,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part4(
op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_BF16
@@ -813,8 +814,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part4(
op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
op_ptrs);
}
#endif
}
@@ -837,8 +838,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv3d_fwd_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part4(
op_ptrs);
// add_device_grouped_conv3d_fwd_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
// op_ptrs);
add_device_grouped_conv3d_fwd_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_BF16
@@ -856,8 +857,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv3d_fwd_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part4(
op_ptrs);
// add_device_grouped_conv3d_fwd_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
// op_ptrs);
add_device_grouped_conv3d_fwd_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
op_ptrs);
}
#endif
}

78
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_wmma_large_tensor.inc Normal file
View File

@@ -0,0 +1,78 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#pragma once
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
#ifdef CK_ENABLE_FP16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_large_tensor_nhwgc_gkyxc_nhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_FP16
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck