Skip to content

GitLab

Commit 7ff4d613 authored by Mateusz Ozga's avatar Mateusz Ozga
Browse files

Rollback tests, removed gnhwc instances

parent 5002a39c
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 177 additions and 733 deletions
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -17,6 +17,7 @@ using OutElementOp = PassThrough; ...@@ -17,6 +17,7 @@ using OutElementOp = PassThrough;
template <ck::index_t NDimSpatial> template <ck::index_t NDimSpatial>
using DeviceConvBwdWeightInstance = using DeviceConvBwdWeightInstance =
// clang-format on
ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle<
NDimSpatial, NDimSpatial,
ck::tuple_element_t<NDimSpatial - 1, ck::tuple_element_t<NDimSpatial - 1,
...@@ -39,33 +40,34 @@ using DeviceConvBwdWeightInstance = ...@@ -39,33 +40,34 @@ using DeviceConvBwdWeightInstance =
WeiElementOp, // WeiElementwiseOperation WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation OutElementOp, // OutElementwiseOperation
ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
64, // BlockSize 256, // BlockSize
16, // MPerBlock 128, // MPerBlock
16, // NPerBlock 128, // NPerBlock
32, // K0PerBlock 32, // K0PerBlock
8, // K1 8, // K1
16, // MPerXdl 32, // MPerXdl
16, // NPerXdl 32, // NPerXdl
1, // MXdlPerWave 2, // MXdlPerWave
1, // NXdlPerWave 2, // NXdlPerWave
S<1, 4, 16, 4>, // ABlockTransferThreadClusterLengths_K0_M_K1 S<1, 4, 16, 4>, // ABlockTransferThreadClusterLengths_K0_M_K1
S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder
1, // ABlockTransferSrcVectorDim 2, // ABlockTransferSrcVectorDim
1, // ABlockTransferSrcScalarPerVector 1, // ABlockTransferSrcScalarPerVector
4, // ABlockTransferDstScalarPerVector_K1 2, // ABlockTransferDstScalarPerVector_K1
false, // ABlockLdsAddExtraM true, // ABlockLdsAddExtraM
S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1 S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder
1, // BBlockTransferSrcVectorDim 2, // BBlockTransferSrcVectorDim
1, // BBlockTransferSrcScalarPerVector 1, // BBlockTransferSrcScalarPerVector
4, // BBlockTransferDstScalarPerVector_K1 2, // BBlockTransferDstScalarPerVector_K1
false, // BBlockLdsAddExtraN true, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle 1, // CShuffleMXdlPerWavePerShuffle
1, // CShuffleNXdlPerWavePerShuffle 1, // CShuffleNXdlPerWavePerShuffle
S<1, 8, 1, 8>, // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock S<1, 32, 1, 4>, // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
1>; // CBlockTransferScalarPerVector_NWaveNPerXdl 128 / (sizeof(WeiDataType) * CHAR_BIT)>; // CBlockTransferScalarPerVector_NWaveNPerXdl
// clang-format off
template <ck::index_t NDimSpatial> template <ck::index_t NDimSpatial>
using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial, using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
......
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -51,14 +51,14 @@ using DeviceConvBwdWeightInstance = ...@@ -51,14 +51,14 @@ using DeviceConvBwdWeightInstance =
S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim 2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector 1, // ABlockTransferSrcScalarPerVector
2, // ABlockTransferDstScalarPerVector_K1 2, // ABlockTransferDstScalarPerVector_K1
false, // ABlockLdsAddExtraM false, // ABlockLdsAddExtraM
S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1 S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim 2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector 1, // BBlockTransferSrcScalarPerVector
2, // BBlockTransferDstScalarPerVector_K1 2, // BBlockTransferDstScalarPerVector_K1
false, // BBlockLdsAddExtraN false, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle 1, // CShuffleMXdlPerWavePerShuffle
......
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
...@@ -41,32 +41,32 @@ using DeviceConvBwdWeightInstance = ...@@ -41,32 +41,32 @@ using DeviceConvBwdWeightInstance =
WeiElementOp, // WeiElementwiseOperation WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation OutElementOp, // OutElementwiseOperation
ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
64, // BlockSize 256, // BlockSize
16, // MPerBlock 128, // MPerBlock
16, // NPerBlock 128, // NPerBlock
32, // K0PerBlock 32, // K0PerBlock
8, // K1 8, // K1
16, // MPerXdl 32, // MPerXdl
16, // NPerXdl 32, // NPerXdl
1, // MXdlPerWave 2, // MXdlPerWave
1, // NXdlPerWave 2, // NXdlPerWave
S<1, 4, 16, 4>, // ABlockTransferThreadClusterLengths_K0_M_K1 S<1, 4, 16, 4>, // ABlockTransferThreadClusterLengths_K0_M_K1
S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder S<0, 3, 1, 2>, // ABlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder S<0, 2, 1, 3>, // ABlockTransferSrcAccessOrder
1, // ABlockTransferSrcVectorDim 2, // ABlockTransferSrcVectorDim
1, // ABlockTransferSrcScalarPerVector 1, // ABlockTransferSrcScalarPerVector
4, // ABlockTransferDstScalarPerVector_K1 1, // ABlockTransferDstScalarPerVector_K1
false, // ABlockLdsAddExtraM true, // ABlockLdsAddExtraM
S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1 S<1, 4, 16, 4>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder S<0, 3, 1, 2>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder S<0, 2, 1, 3>, // BBlockTransferSrcAccessOrder
1, // BBlockTransferSrcVectorDim 2, // BBlockTransferSrcVectorDim
1, // BBlockTransferSrcScalarPerVector 1, // BBlockTransferSrcScalarPerVector
4, // BBlockTransferDstScalarPerVector_K1 1, // BBlockTransferDstScalarPerVector_K1
false, // BBlockLdsAddExtraN true, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle 1, // CShuffleMXdlPerWavePerShuffle
1, // CShuffleNXdlPerWavePerShuffle 1, // CShuffleNXdlPerWavePerShuffle
S<1, 8, 1, 8>, // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock S<1, 32, 1, 4>, // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
2, // CBlockTransferScalarPerVector_NWaveNPerXdl 2, // CBlockTransferScalarPerVector_NWaveNPerXdl
ck::BlockGemmPipelineScheduler::Intrawave, // BlkGemmPipeSched ck::BlockGemmPipelineScheduler::Intrawave, // BlkGemmPipeSched
ck::BlockGemmPipelineVersion::v1, // BlkGemmPipelineVer ck::BlockGemmPipelineVersion::v1, // BlkGemmPipelineVer
......
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
...@@ -16,7 +16,6 @@ ...@@ -16,7 +16,6 @@
#include "ck/tensor_operation/operator_transform/transform_conv_bwd_weight_to_gemm.hpp" #include "ck/tensor_operation/operator_transform/transform_conv_bwd_weight_to_gemm.hpp"
#include "ck/tensor_operation/operator_transform/transform_conv_bwd_weight_to_gemm_v2.hpp" #include "ck/tensor_operation/operator_transform/transform_conv_bwd_weight_to_gemm_v2.hpp"
#include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp" #include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_bwd_weight_v3.hpp" #include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_bwd_weight_v3.hpp"
#include <ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp> #include <ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp>
...@@ -221,12 +220,12 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle ...@@ -221,12 +220,12 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
// TODO make A/B datatype different // TODO make A/B datatype different
using ABDataType = InDataType; using ABDataType = InDataType;
static inline constexpr auto I0 = Number<0>{}; static inline auto I0 = Number<0>{};
static inline constexpr auto I1 = Number<1>{}; static inline auto I1 = Number<1>{};
static inline constexpr auto I2 = Number<2>{}; static inline auto I2 = Number<2>{};
static inline constexpr auto I3 = Number<3>{}; static inline auto I3 = Number<3>{};
static inline constexpr auto I4 = Number<4>{}; static inline auto I4 = Number<4>{};
static inline constexpr auto I5 = Number<5>{}; static inline auto I5 = Number<5>{};
static constexpr GemmSpecialization GemmSpec = GemmSpecialization::Default; static constexpr GemmSpecialization GemmSpec = GemmSpecialization::Default;
static constexpr auto K1Number = Number<K1>{}; static constexpr auto K1Number = Number<K1>{};
...@@ -1235,21 +1234,10 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle ...@@ -1235,21 +1234,10 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
} }
} }
if(!(arg.Conv_C_ % BBlockTransferSrcScalarPerVector == 0 &&
arg.Conv_K_ % ABlockTransferSrcScalarPerVector == 0))
{
if(!(arg.Conv_K_ == 1 && arg.compute_ptr_offset_of_batch_.BatchStrideA_ == 1))
{
return false;
}
if(!(arg.Conv_C_ == 1 && arg.compute_ptr_offset_of_batch_.BatchStrideB_ == 1))
{
return false;
}
}
// vector load A/B matrix from global memory // vector load A/B matrix from global memory
if(!(ABlockTransferSrcVectorDim == 1 && BBlockTransferSrcVectorDim == 1)) if(!(ABlockTransferSrcVectorDim == 2 && BBlockTransferSrcVectorDim == 2 &&
arg.Conv_K_ % ABlockTransferSrcScalarPerVector == 0 &&
arg.Conv_C_ % BBlockTransferSrcScalarPerVector == 0))
{ {
return false; return false;
} }
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
...@@ -329,13 +329,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -329,13 +329,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> && is_same_v<OutDataType, float> && is_same_v<ComputeTypeA, float> &&
is_same_v<ComputeTypeB, float>) is_same_v<ComputeTypeB, float>)
{ {
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev2_instances( add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instances(
op_ptrs); op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev5_instances( add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instances(
op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_instances(
op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev5_instances(
op_ptrs); op_ptrs);
} }
#endif #endif
...@@ -344,13 +340,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -344,13 +340,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> && is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> &&
is_same_v<ComputeTypeB, half_t>) is_same_v<ComputeTypeB, half_t>)
{ {
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev2_instances( add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instances(
op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev5_instances(
op_ptrs); op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev2_instances( add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instances(
op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev2_instances(
op_ptrs); op_ptrs);
} }
#endif #endif
...@@ -360,13 +352,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe ...@@ -360,13 +352,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
is_same_v<ComputeTypeA, ck::bhalf_t> && is_same_v<ComputeTypeA, ck::bhalf_t> &&
is_same_v<ComputeTypeB, ck::bhalf_t>) is_same_v<ComputeTypeB, ck::bhalf_t>)
{ {
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev2_instances( add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev1_instances(
op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev5_instances(
op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev2_instances(
op_ptrs); op_ptrs);
add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev5_instances( add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev1_instances(
op_ptrs); op_ptrs);
} }
#endif #endif
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight_xdl.inc
View file @ 7ff4d613
...@@ -149,7 +149,7 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_pad0_pipev5_ins ...@@ -149,7 +149,7 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_pad0_pipev5_ins
#endif #endif
// conv2d backward weight // conv2d backward weight
#ifdef CK_ENABLE_BF16 #ifdef CK_ENABLE_BF16
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev2_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -160,29 +160,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_de ...@@ -160,29 +160,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_de
PassThrough, PassThrough,
PassThrough, PassThrough,
PassThrough>>>& instances); PassThrough>>>& instances);
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev5_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev2_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev5_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -241,29 +219,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev5_ ...@@ -241,29 +219,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev5_
PassThrough>>>& instances); PassThrough>>>& instances);
#endif #endif
#ifdef CK_ENABLE_FP32 #ifdef CK_ENABLE_FP32
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev5_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev2_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances);
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -274,7 +230,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_ ...@@ -274,7 +230,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_
PassThrough, PassThrough,
PassThrough, PassThrough,
PassThrough>>>& instances); PassThrough>>>& instances);
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev5_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
View file @ 7ff4d613
# ONLY XDL_AND_DL_KERNELS # ONLY XDL_AND_DL_KERNELS
set(GROUPED_CONV2D_BWD_WEIGHT set(GROUPED_CONV2D_BWD_WEIGHT
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev2_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev1_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev5_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev1_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev2_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev5_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev2_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev5_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev2_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev5_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev2_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev5_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev5_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev2_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev2_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev5_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev5_instance.cpp
xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_default_pipev2_instance.cpp xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_default_pipev2_instance.cpp
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev2_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev1_instance.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp" #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
...@@ -10,7 +10,7 @@ namespace device { ...@@ -10,7 +10,7 @@ namespace device {
namespace instance { namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k] // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev2_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -31,7 +31,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_de ...@@ -31,7 +31,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_de
GNHWK, GNHWK,
ConvBwdWeightDefault, ConvBwdWeightDefault,
BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v2>{}); BlockGemmPipelineVersion::v1>{});
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev5_instance.cpp deleted 100644 → 0
View file @ 5002a39c
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_default_pipev5_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_f32_bf16_instances<
2,
GNHWC,
GKYXC,
GNHWK,
ConvBwdWeightDefault,
BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v5>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev2_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev1_instance.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp" #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
...@@ -10,7 +10,7 @@ namespace device { ...@@ -10,7 +10,7 @@ namespace device {
namespace instance { namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k] // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev2_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -31,7 +31,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pa ...@@ -31,7 +31,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pa
GNHWK, GNHWK,
ConvBwdWeightFilter1x1Stride1Pad0, ConvBwdWeightFilter1x1Stride1Pad0,
BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v2>{}); BlockGemmPipelineVersion::v1>{});
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev5_instance.cpp deleted 100644 → 0
View file @ 5002a39c
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_pad0_pipev5_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_f32_bf16_instances<
2,
GNHWC,
GKYXC,
GNHWK,
ConvBwdWeightFilter1x1Stride1Pad0,
BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v5>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev2_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instance.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp" #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
...@@ -10,7 +10,7 @@ namespace device { ...@@ -10,7 +10,7 @@ namespace device {
namespace instance { namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k] // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev2_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipe ...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipe
GNHWK, GNHWK,
ConvBwdWeightDefault, ConvBwdWeightDefault,
BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v2>{}); BlockGemmPipelineVersion::v1>{});
} }
} // namespace instance } // namespace instance
} // namespace device } // namespace device
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev5_instance.cpp deleted 100644 → 0
View file @ 5002a39c
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev5_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
2,
GNHWC,
GKYXC,
GNHWK,
ConvBwdWeightDefault,
BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v5>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev2_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instance.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp" #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
...@@ -10,7 +10,7 @@ namespace device { ...@@ -10,7 +10,7 @@ namespace device {
namespace instance { namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k] // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev2_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev2_ ...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev2_
GNHWK, GNHWK,
ConvBwdWeightFilter1x1Stride1Pad0, ConvBwdWeightFilter1x1Stride1Pad0,
BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v2>{}); BlockGemmPipelineVersion::v1>{});
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev5_instance.cpp deleted 100644 → 0
View file @ 5002a39c
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev5_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
2,
GNHWC,
GKYXC,
GNHWK,
ConvBwdWeightFilter1x1Stride1Pad0,
BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v5>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev2_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instance.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp" #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
...@@ -10,7 +10,7 @@ namespace device { ...@@ -10,7 +10,7 @@ namespace device {
namespace instance { namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k] // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev2_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipe ...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipe
GNHWK, GNHWK,
ConvBwdWeightDefault, ConvBwdWeightDefault,
BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v2>{}); BlockGemmPipelineVersion::v1>{});
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev5_instance.cpp deleted 100644 → 0
View file @ 5002a39c
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev5_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
2,
GNHWC,
GKYXC,
GNHWK,
ConvBwdWeightDefault,
BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v5>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instance.cpp
View file @ 7ff4d613
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp" #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp" #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
...@@ -10,7 +10,7 @@ namespace device { ...@@ -10,7 +10,7 @@ namespace device {
namespace instance { namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k] // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_instances( void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2, std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC, GNHWC,
GKYXC, GKYXC,
...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_ ...@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev2_
GNHWK, GNHWK,
ConvBwdWeightFilter1x1Stride1Pad0, ConvBwdWeightFilter1x1Stride1Pad0,
BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v2>{}); BlockGemmPipelineVersion::v1>{});
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev5_instance.cpp deleted 100644 → 0
View file @ 5002a39c
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev5_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
GNHWC,
GKYXC,
GNHWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
2,
GNHWC,
GKYXC,
GNHWK,
ConvBwdWeightFilter1x1Stride1Pad0,
BlockGemmPipelineScheduler::Intrawave,
BlockGemmPipelineVersion::v5>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment