Skip to content

GitLab

".github/vscode:/vscode.git/clone" did not exist on "ee638329f13a86cad94124fc354a0c33f02f190c"
Commit ef326c73 authored by Alan Turner's avatar Alan Turner
Browse files

Merge remote-tracking branch 'origin/develop' into migraphx-update

parents b7775add e4dfe4d8
Hide whitespace changes
Inline Side-by-side
Showing with 250 additions and 348 deletions
example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_xdl_fp16.cpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
#include "common.hpp"
using OutDataType = FP16;
using WeiDataType = FP16;
using AccDataType = FP32;
using CShuffleDataType = FP16;
using DsDataType = ck::Tuple<>;
using InDataType = FP16;
using OutLayout = ck::tensor_layout::convolution::GNHWK;
using WeiLayout = ck::tensor_layout::convolution::GKYXC;
using DsLayout = ck::Tuple<>;
using InLayout = ck::tensor_layout::convolution::GNHWC;
using OutElementOp = PassThrough;
using WeiElementOp = PassThrough;
using InElementOp = PassThrough;
// clang-format off
using DeviceConvInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
// ######| NDimSpatial| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl| CDEBlockTransfer| CDEBlockTransfer|
// ######| | | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| PrefetchStage| Size| Block| Block| Block| | | XDL| XDL| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| PerWave| PerWave| _MBlock_MPerBlock| ScalarPerVector|
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| _NPerBlock|
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< NDimSpatial, OutLayout, WeiLayout, DsLayout, InLayout, OutDataType, WeiDataType, AccDataType, CShuffleDataType, DsDataType, InDataType, OutElementOp, WeiElementOp, InElementOp, ConvBwdDataDefault, true, true, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>;
// clang-format on
#include "run_grouped_conv_bwd_data_example.inc"
int main(int argc, char* argv[]) { return run_grouped_conv_bwd_data_example(argc, argv); }
example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_xdl_fp16_comp_bf8_fp8.cpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
#include "common.hpp"
using OutDataType = FP16;
using WeiDataType = FP16;
using AccDataType = FP32;
using CShuffleDataType = FP16;
using DsDataType = ck::Tuple<>;
using InDataType = FP16;
using AComputeType = BF8;
using BComputeType = FP8;
using OutLayout = ck::tensor_layout::convolution::GNHWK;
using WeiLayout = ck::tensor_layout::convolution::GKYXC;
using DsLayout = ck::Tuple<>;
using InLayout = ck::tensor_layout::convolution::GNHWC;
using OutElementOp = PassThrough;
using WeiElementOp = PassThrough;
using InElementOp = PassThrough;
static constexpr auto LoopSched = ck::make_default_loop_scheduler();
// clang-format off
using DeviceConvInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
// ######| NDimSpatial| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl| CDEBlockTransfer| CDEBlockTransfer| Loop| ACompute| BCompute|
// ######| | | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| PrefetchStage| Size| Block| Block| Block| | | XDL| XDL| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| PerWave| PerWave| _MBlock_MPerBlock| ScalarPerVector| Scheduler| Type| Type|
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| _NPerBlock| | | |
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< NDimSpatial, OutLayout, WeiLayout, DsLayout, InLayout, OutDataType, WeiDataType, AccDataType, CShuffleDataType, DsDataType, InDataType, OutElementOp, WeiElementOp, InElementOp, ConvBwdDataDefault, true, true, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8, LoopSched, AComputeType, BComputeType>;
// clang-format on
#include "run_grouped_conv_bwd_data_example.inc"
int main(int argc, char* argv[]) { return run_grouped_conv_bwd_data_example(argc, argv); }
example/39_permute/CMakeLists.txt
View file @ ef326c73
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES) add_custom_target(example_permute)
add_custom_target(example_permute)
add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp) add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp) add_example_dependencies(example_permute example_permute_1xHxW_fp16)
add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
add_dependencies(example_permute example_permute_1xHxW_fp16) add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
add_dependencies(example_permute example_permute_NxHxW_fp16) add_example_dependencies(example_permute example_permute_NxHxW_fp16)
add_dependencies(example_permute example_permute_HxWx4_fp16)
endif() add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
add_example_dependencies(example_permute example_permute_HxWx4_fp16)
example/40_conv2d_fwd_quantization/CMakeLists.txt
View file @ ef326c73
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
set(target 0) add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
foreach(gpu IN LISTS GPU_TARGETS) add_example_executable(example_conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
set(target 1)
endif()
endforeach()
if(DL_KERNELS) # Conv perlayer quantization
# Conv perlayer quantization add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp) # Conv perchannel quantization
# Conv perchannel quantization add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp) # Conv + bias + relu perlayer quantization
# Conv + bias + relu perlayer quantization add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp) # Conv + bias + relu perchannel quantization
# Conv + bias + relu perchannel quantization add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp) # Conv + bias + tanh perlayer quantization
# Conv + bias + tanh perlayer quantization add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp) # Conv + bias + tanh perchannel quantization
# Conv + bias + tanh perchannel quantization add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
endif()
endif()
\ No newline at end of file
example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
View file @ ef326c73
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
using InDataType = int8_t; using InDataType = int8_t;
using WeiDataType = int8_t; using WeiDataType = int8_t;
...@@ -33,7 +33,7 @@ template <ck::index_t NDimSpatial, ...@@ -33,7 +33,7 @@ template <ck::index_t NDimSpatial,
typename RequantScaleLayout, typename RequantScaleLayout,
typename OutLayout> typename OutLayout>
using DeviceGroupedConvNDFwdInstance = using DeviceGroupedConvNDFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle< ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
NDimSpatial, NDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
......
example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
View file @ ef326c73
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
using InDataType = int8_t; using InDataType = int8_t;
using WeiDataType = int8_t; using WeiDataType = int8_t;
...@@ -31,7 +31,7 @@ template <ck::index_t NDimSpatial, ...@@ -31,7 +31,7 @@ template <ck::index_t NDimSpatial,
typename BiasLayout, typename BiasLayout,
typename OutLayout> typename OutLayout>
using DeviceGroupedConvNDFwdInstance = using DeviceGroupedConvNDFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle< ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
NDimSpatial, NDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
......
example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp
View file @ ef326c73
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
using InDataType = int8_t; using InDataType = int8_t;
using WeiDataType = int8_t; using WeiDataType = int8_t;
...@@ -31,7 +31,7 @@ template <ck::index_t NDimSpatial, ...@@ -31,7 +31,7 @@ template <ck::index_t NDimSpatial,
typename RequantScaleLayout, typename RequantScaleLayout,
typename OutLayout> typename OutLayout>
using DeviceGroupedConvNDFwdInstance = using DeviceGroupedConvNDFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle< ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
NDimSpatial, NDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
......
example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
View file @ ef326c73
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp" #include "common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp"
using InDataType = int8_t; using InDataType = int8_t;
using WeiDataType = int8_t; using WeiDataType = int8_t;
...@@ -26,7 +26,7 @@ static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecializatio ...@@ -26,7 +26,7 @@ static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecializatio
template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout> template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
using DeviceGroupedConvNDFwdInstance = using DeviceGroupedConvNDFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle< ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
NDimSpatial, NDimSpatial,
InLayout, InLayout,
WeiLayout, WeiLayout,
......
example/41_grouped_conv_conv_fwd/CMakeLists.txt
View file @ ef326c73
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942) add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
list(APPEND gpu_list2 gfx908 gfx90a) add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
set(target 0) add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
foreach(gpu IN LISTS GPU_TARGETS) if(USE_BITINT_EXTENSION_INT4)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0) add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES) endif(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
endif()
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
endif(USE_BITINT_EXTENSION_INT4)
set(target 1)
endif()
endforeach()
if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1") if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES) add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
endif()
endif() endif()
example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp
View file @ ef326c73
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
#error Should compile this file with ck::int4_t support
#endif
#include <cstdlib> #include <cstdlib>
#include <iostream> #include <iostream>
...@@ -120,3 +118,4 @@ static_assert(sizeof(ck::int4_t) == sizeof(int8_t)); ...@@ -120,3 +118,4 @@ static_assert(sizeof(ck::int4_t) == sizeof(int8_t));
#endif #endif
int main(int argc, char* argv[]) { return run_grouped_conv_conv_fwd_example(argc, argv) ? 0 : 1; } int main(int argc, char* argv[]) { return run_grouped_conv_conv_fwd_example(argc, argv) ? 0 : 1; }
#endif
example/42_groupnorm/CMakeLists.txt deleted 100644 → 0
View file @ b7775add
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
endif()
example/42_groupnorm/common.hpp deleted 100644 → 0
View file @ b7775add
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <getopt.h>
#include "ck/ck.hpp"
#include "ck/utility/reduction_enums.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/fill.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_common_util.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp"
example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp deleted 100644 → 0
View file @ b7775add
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
struct YElementOp
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(ck::is_same<T, float>::value || ck::is_same<T, double>::value ||
ck::is_same<T, ck::half_t>::value,
"Data type is not supported by this operation!");
T a;
ck::tensor_operation::element_wise::Sigmoid{}(a, x);
y = x * a;
};
};
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
YElementOp,
Rank,
NumReduceDim,
1024, // BlockSize
1, // ClusterM
1024, // ClusterK
1, // SliceM
32, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2>; // OutScalarPerVector
#include "run_groupnorm_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
example/42_groupnorm/groupnorm_splitk_fp16.cpp deleted 100644 → 0
View file @ b7775add
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
using YElementOp = ck::tensor_operation::element_wise::Swish;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
YElementOp,
Rank,
NumReduceDim,
256, // BlockSize
1, // ClusterM
256, // ClusterK
1, // SliceM
16, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2>; // OutScalarPerVector
#include "run_groupnorm_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
example/42_groupnorm/groupnorm_swish_fp16.cpp deleted 100644 → 0
View file @ b7775add
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
using YElementOp = ck::tensor_operation::element_wise::Swish;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
YElementOp,
Rank,
NumReduceDim,
1024, // BlockSize
1, // ClusterM
1024, // ClusterK
1, // SliceM
32, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2>; // OutScalarPerVector
#include "run_groupnorm_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
example/42_groupnorm/run_groupnorm_example.inc deleted 100644 → 0
View file @ b7775add
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
int run_groupnorm_example(int argc, char* argv[])
{
ck::index_t N = 32;
ck::index_t H = 16;
ck::index_t W = 16;
ck::index_t G = 64;
ck::index_t C = 128;
if(argc == 1)
{
// use default case
}
else if(argc == 6)
{
N = std::stoi(argv[1]);
H = std::stoi(argv[2]);
W = std::stoi(argv[3]);
G = std::stoi(argv[4]);
C = std::stoi(argv[5]);
}
else
{
std::cerr << "arg1 to 5: N, H, W, G, C" << std::endl;
return 1;
}
Tensor<XDataType> x({N, H, W, G, C});
Tensor<YDataType> y({N, H, W, G, C});
Tensor<GammaDataType> gamma({G, C});
Tensor<BetaDataType> beta({G, C});
ck::utils::FillUniformDistribution<XDataType>{0.f, 1.f}(x);
ck::utils::FillUniformDistribution<GammaDataType>{0.f, 1.f}(gamma);
ck::utils::FillUniformDistribution<BetaDataType>{0.f, 1.f}(beta);
DeviceMem x_dev(sizeof(XDataType) * x.mDesc.GetElementSpaceSize());
DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize());
DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
x_dev.ToDevice(x.mData.data());
gamma_dev.ToDevice(gamma.mData.data());
beta_dev.ToDevice(beta.mData.data());
const auto y_element_op = YElementOp{};
auto device_instance = DeviceInstance{};
auto argument_ptr = device_instance.MakeArgumentPointer(
{N, H, W, G, C},
std::vector<ck::index_t>{x.mDesc.GetStrides().begin(), x.mDesc.GetStrides().end()},
{0, 0, 0, C, 1},
{0, 0, 0, C, 1},
std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
{1, 2, 4}, // reduction dimension: [H, W, C]
1e-6,
x_dev.GetDeviceBuffer(),
gamma_dev.GetDeviceBuffer(),
beta_dev.GetDeviceBuffer(),
y_dev.GetDeviceBuffer(),
nullptr,
nullptr,
y_element_op);
if(!device_instance.IsSupportedArgument(argument_ptr.get()))
{
std::cout << "The runtime parameters are not supported" << std::endl;
return 1;
};
size_t workspace_sz = device_instance.GetWorkSpaceSize(argument_ptr.get());
DeviceMem workspace_dev(workspace_sz);
device_instance.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
auto invoker_ptr = device_instance.MakeInvokerPointer();
float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true, true});
std::size_t num_btype = sizeof(XDataType) * N * H * W * G * C +
sizeof(YDataType) * N * H * W * G * C + sizeof(GammaDataType) * G * C +
sizeof(BetaDataType) * G * C;
float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s, "
<< device_instance.GetTypeString() << std::endl;
bool pass = true;
{
Tensor<YDataType> host_y({N, H, W, G, C});
using ReferenceInstance = ck::tensor_operation::host::ReferenceGroupnorm<XDataType,
GammaDataType,
BetaDataType,
YDataType,
ComputeDataType,
YElementOp>;
ReferenceInstance ref;
auto ref_argument =
ref.MakeArgument(x, gamma, beta, host_y, y_element_op, {N, H, W, G, C}, 1e-6);
auto ref_invoker = ref.MakeInvoker();
ref_invoker.Run(ref_argument);
y_dev.FromDevice(y.mData.data());
pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3);
}
return (pass ? 0 : 1);
}
example/42_groupnorm_fwd/CMakeLists.txt 0 → 100644
View file @ ef326c73
add_example_executable(example_groupnorm_fwd_sigmoid_mul_fp16 groupnorm_fwd_sigmoid_mul_fp16.cpp)
add_example_executable(example_groupnorm_fwd_splitk_fp16 groupnorm_fwd_splitk_fp16.cpp)
add_example_executable(example_groupnorm_fwd_swish_fp16 groupnorm_fwd_swish_fp16.cpp)
example/42_groupnorm_fwd/common.hpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <getopt.h>
#include "ck/ck.hpp"
#include "ck/utility/reduction_enums.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_splitk_impl.hpp"
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/fill.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_common_util.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp"
example/42_groupnorm_fwd/groupnorm_fwd_sigmoid_mul_fp16.cpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
#define SAVE_MEAN_INV_STD
struct YElementOp
{
template <typename Y, typename X>
__host__ __device__ void operator()(Y& y, const X& x) const
{
static_assert(ck::is_same<X, float>::value || ck::is_same<X, double>::value ||
ck::is_same<X, ck::half_t>::value,
"Data type is not supported by this operation!");
static_assert(ck::is_same<Y, float>::value || ck::is_same<Y, double>::value ||
ck::is_same<Y, ck::half_t>::value,
"Data type is not supported by this operation!");
X a;
ck::tensor_operation::element_wise::Sigmoid{}(a, x);
y = ck::type_convert<Y>(x * a);
};
};
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationFwdImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
YElementOp,
Rank,
NumReduceDim,
1024, // BlockSize
1, // ClusterM
1024, // ClusterK
1, // SliceM
32, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_groupnorm_fwd_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_fwd_example(argc, argv); }
example/42_groupnorm_fwd/groupnorm_fwd_splitk_fp16.cpp 0 → 100644
View file @ ef326c73
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using ComputeDataType = float;
using YElementOp = ck::tensor_operation::element_wise::Swish;
#define SAVE_MEAN_INV_STD
using DeviceInstance = ck::tensor_operation::device::DeviceNormalizationFwdSplitKImpl<
XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
YElementOp,
Rank,
NumReduceDim,
256, // BlockSize
1, // ClusterM
256, // ClusterK
1, // SliceM
16, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_groupnorm_fwd_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_fwd_example(argc, argv); }
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