Skip to content

GitLab

Commit 0c823497 authored by muozturk's avatar muozturk
Browse files

merge

parents 334cfe1c 68f2b5e7
Hide whitespace changes
Inline Side-by-side
Showing with 458 additions and 253 deletions
example/27_layernorm/layernorm_fp16.cpp deleted 100644 → 0
View file @ 334cfe1c
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8>; // OutScalarPerVector
#include "run_layernorm_example.inc"
int main() { return run_groupnorm_example<DeviceInstance>(); }
example/27_layernorm2d_fwd/CMakeLists.txt 0 → 100644
View file @ 0c823497
add_example_executable(example_layernorm2d_fwd_fp16 layernorm2d_fwd_fp16.cpp)
add_example_executable(example_layernorm2d_fwd_splitk_fp16 layernorm2d_fwd_splitk_fp16.cpp)
example/27_layernorm/common.hpp → example/27_layernorm2d_fwd/common.hpp
View file @ 0c823497
...@@ -10,8 +10,8 @@ ...@@ -10,8 +10,8 @@
#include <getopt.h> #include <getopt.h>
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_normalization_fwd_splitk_impl.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
......
example/27_layernorm2d_fwd/layernorm2d_fwd_fp16.cpp 0 → 100644
View file @ 0c823497
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
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 PassThrough = ck::tensor_operation::element_wise::PassThrough;
#define SAVE_MEAN_INV_STD
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationFwdImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_layernorm_example.inc"
int main() { return run_layernorm2d_fwd_example<DeviceInstance>(); }
example/27_layernorm2d_fwd/layernorm2d_fwd_splitk_fp16.cpp 0 → 100644
View file @ 0c823497
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
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 PassThrough = ck::tensor_operation::element_wise::PassThrough;
#define SAVE_MEAN_INV_STD
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance = ck::tensor_operation::device::DeviceNormalizationFwdSplitKImpl<
XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // XScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8, // YScalarPerVector
1>; // SaveMeanInvStdScalarPerVector
#include "run_layernorm_example.inc"
int main() { return run_layernorm2d_fwd_example<DeviceInstance>(); }
example/27_layernorm/run_layernorm_example.inc → example/27_layernorm2d_fwd/run_layernorm_example.inc
View file @ 0c823497
...@@ -4,28 +4,19 @@ ...@@ -4,28 +4,19 @@
#pragma once #pragma once
template <typename DeviceInstance> template <typename DeviceInstance>
int run_groupnorm_example() int run_layernorm2d_fwd_example()
{ {
bool time_kernel = false; bool time_kernel = false;
ck::index_t M = 1024; ck::index_t M = 1024;
ck::index_t N = 1024; ck::index_t N = 1024;
ck::index_t Stride = N;
auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) { Tensor<XDataType> x({M, N});
return HostTensorDescriptor({len}, {stride}); Tensor<GammaDataType> gamma({N});
}; Tensor<BetaDataType> beta({N});
Tensor<YDataType> y({M, N});
auto f_host_tensor_descriptor2d = [](std::size_t row, std::size_t col, std::size_t stride) { Tensor<SaveMeanInvStdDataType> save_mean({M});
using namespace ck::literals; Tensor<SaveMeanInvStdDataType> save_inv_std({M});
return HostTensorDescriptor({row, col}, {stride, 1_uz});
};
Tensor<XDataType> x(f_host_tensor_descriptor2d(M, N, Stride));
Tensor<GammaDataType> gamma(f_host_tensor_descriptor1d(N, 1));
Tensor<BetaDataType> beta(f_host_tensor_descriptor1d(N, 1));
Tensor<YDataType> y(f_host_tensor_descriptor2d(M, N, Stride));
x.GenerateTensorValue(GeneratorTensor_3<XDataType>{0.0, 1.0}); x.GenerateTensorValue(GeneratorTensor_3<XDataType>{0.0, 1.0});
gamma.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{0.0, 1.0}); gamma.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{0.0, 1.0});
...@@ -35,6 +26,11 @@ int run_groupnorm_example() ...@@ -35,6 +26,11 @@ int run_groupnorm_example()
DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize()); DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize()); DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize());
DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize()); DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
#ifdef SAVE_MEAN_INV_STD
DeviceMem save_mean_dev(sizeof(SaveMeanInvStdDataType) * save_mean.mDesc.GetElementSpaceSize());
DeviceMem save_inv_std_dev(sizeof(SaveMeanInvStdDataType) *
save_inv_std.mDesc.GetElementSpaceSize());
#endif
x_dev.ToDevice(x.mData.data()); x_dev.ToDevice(x.mData.data());
gamma_dev.ToDevice(gamma.mData.data()); gamma_dev.ToDevice(gamma.mData.data());
...@@ -47,14 +43,23 @@ int run_groupnorm_example() ...@@ -47,14 +43,23 @@ int run_groupnorm_example()
{0, 1}, {0, 1},
{0, 1}, {0, 1},
std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()}, std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
save_mean.mDesc.GetStrides().end()},
std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
save_mean.mDesc.GetStrides().end()},
{1}, {1},
1e-4, 1e-4,
x_dev.GetDeviceBuffer(), x_dev.GetDeviceBuffer(),
gamma_dev.GetDeviceBuffer(), gamma_dev.GetDeviceBuffer(),
beta_dev.GetDeviceBuffer(), beta_dev.GetDeviceBuffer(),
y_dev.GetDeviceBuffer(), y_dev.GetDeviceBuffer(),
#ifdef SAVE_MEAN_INV_STD
save_mean_dev.GetDeviceBuffer(),
save_inv_std_dev.GetDeviceBuffer(),
#else
nullptr, nullptr,
nullptr, nullptr,
#endif
PassThrough{}); PassThrough{});
if(!device_instance.IsSupportedArgument(argument_ptr.get())) if(!device_instance.IsSupportedArgument(argument_ptr.get()))
...@@ -72,24 +77,45 @@ int run_groupnorm_example() ...@@ -72,24 +77,45 @@ int run_groupnorm_example()
bool pass = true; bool pass = true;
{ {
Tensor<YDataType> host_y(f_host_tensor_descriptor2d(M, N, Stride)); Tensor<YDataType> host_y({M, N});
using ReferenceInstance = ck::tensor_operation::host::ReferenceLayernorm<XDataType, Tensor<SaveMeanInvStdDataType> host_save_mean({M});
GammaDataType, Tensor<SaveMeanInvStdDataType> host_save_inv_std({M});
BetaDataType,
YDataType, using ReferenceInstance =
ComputeDataType, ck::tensor_operation::host::ReferenceLayernorm<XDataType,
PassThrough, GammaDataType,
Rank, BetaDataType,
NumReduceDim>; YDataType,
SaveMeanInvStdDataType,
ComputeDataType,
PassThrough,
Rank,
NumReduceDim>;
ReferenceInstance ref; ReferenceInstance ref;
auto ref_argument = auto ref_argument = ref.MakeArgument(x,
ref.MakeArgument(x, gamma, beta, host_y, PassThrough{}, {M, N}, {1}, 1e-4); gamma,
auto ref_invoker = ref.MakeInvoker(); beta,
host_y,
host_save_mean,
host_save_inv_std,
PassThrough{},
{M, N},
{1},
1e-4);
auto ref_invoker = ref.MakeInvoker();
ref_invoker.Run(ref_argument); ref_invoker.Run(ref_argument);
y_dev.FromDevice(y.mData.data()); y_dev.FromDevice(y.mData.data());
pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3); pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results (y)", 1e-3, 1e-3);
#ifdef SAVE_MEAN_INV_STD
save_mean_dev.FromDevice(save_mean.mData.data());
save_inv_std_dev.FromDevice(save_inv_std.mData.data());
pass &= ck::utils::check_err(
save_mean, host_save_mean, "Error: Incorrect results (mean)", 1e-3, 1e-3);
pass &= ck::utils::check_err(
save_inv_std, host_save_inv_std, "Error: Incorrect results (inv_std)", 1e-3, 1e-3);
#endif
} }
return (pass ? 0 : 1); return (pass ? 0 : 1);
......
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
View file @ 0c823497
...@@ -3,44 +3,38 @@ list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102) ...@@ -3,44 +3,38 @@ list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0) if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
add_custom_target(example_grouped_conv_fwd_multiple_d) add_custom_target(example_grouped_conv_fwd_multiple_d)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
if(result EQUAL 0) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
endif()
add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp) add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
endif() add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
if(result EQUAL 0)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
endif() add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
if(result EQUAL 0) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
endif()
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp) if(USE_BITINT_EXTENSION_INT4)
if(result EQUAL 0) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
endif() endif() # USE_BITINT_EXTENSION_INT4
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp) set(target 1)
if(result EQUAL 0) endif()
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
endif()
endif() # USE_BITINT_EXTENSION_INT4
set(target 1)
endif()
endforeach() endforeach()
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list2 AND target EQUAL 0) if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
example/30_grouped_conv_fwd_multiple_d/README.md
View file @ 0c823497
...@@ -4,7 +4,7 @@ arg1: verification (0=no, 1=yes) ...@@ -4,7 +4,7 @@ arg1: verification (0=no, 1=yes)
arg2: initialization (0=no init, 1=integer value, 2=decimal value) arg2: initialization (0=no init, 1=integer value, 2=decimal value)
arg3: time kernel (0=no, 1=yes) arg3: time kernel (0=no, 1=yes)
Following arguments (depending on number of spatial dims): Following arguments (depending on number of spatial dims):
Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d) Number of spatial dimensions (1=Conv1D, 2=Conv2D, 3=Conv3D)
G, N, K, C, G, N, K, C,
<filter spatial dimensions>, (ie Y, X for 2D) <filter spatial dimensions>, (ie Y, X for 2D)
<input image spatial dimensions>, (ie Hi, Wi for 2D) <input image spatial dimensions>, (ie Hi, Wi for 2D)
......
example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
View file @ 0c823497
add_custom_target(example_gemm_scale_softmax_gemm) add_custom_target(example_gemm_scale_softmax_gemm)
add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp) add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
endif()
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp) add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
endif()
add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp) add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
endif()
add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp) add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
endif()
add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp) add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
endif()
add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp) add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
if(result EQUAL 0) add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
endif()
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp) add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
if(result EQUAL 0) add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
endif()
example/35_splitK_gemm/CMakeLists.txt
View file @ 0c823497
...@@ -4,28 +4,23 @@ foreach(gpu IN LISTS GPU_TARGETS) ...@@ -4,28 +4,23 @@ foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0) if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_splitK_gemm_xdl) add_custom_target(example_splitK_gemm_xdl)
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp) add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
if(result EQUAL 0) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
endif() add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
if(result EQUAL 0)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16) add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp)
endif() add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16)
add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp)
if(result EQUAL 0) add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
endif()
add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
endif()
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp) add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
if(result EQUAL 0) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
endif()
endif() endif()
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
View file @ 0c823497
...@@ -2,27 +2,26 @@ list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942) ...@@ -2,27 +2,26 @@ list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102) list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
set(target 0) set(target 0)
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0) if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
add_custom_target(example_grouped_conv_bwd_data) add_custom_target(example_grouped_conv_bwd_data)
add_example_executable(example_grouped_conv_bwd_data_xdl_fp16 grouped_conv_bwd_data_xdl_fp16.cpp)
if(result EQUAL 0) add_example_executable(example_grouped_conv_bwd_data_xdl_fp16 grouped_conv_bwd_data_xdl_fp16.cpp)
add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_xdl_fp16) add_example_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_xdl_fp16)
endif()
add_example_executable(example_grouped_conv_bwd_data_bias_relu_xdl_fp16 grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp) add_example_executable(example_grouped_conv_bwd_data_bias_relu_xdl_fp16 grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_xdl_fp16)
add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_xdl_fp16)
endif() set(target 1)
set(target 1) endif()
endif()
endforeach() endforeach()
foreach(gpu IN LISTS GPU_TARGETS) foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0) if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
add_custom_target(example_grouped_conv_bwd_data) add_custom_target(example_grouped_conv_bwd_data)
add_example_executable(example_grouped_conv_bwd_data_wmma_fp16 grouped_conv_bwd_data_wmma_fp16.cpp)
if(result EQUAL 0) add_example_executable(example_grouped_conv_bwd_data_wmma_fp16 grouped_conv_bwd_data_wmma_fp16.cpp)
add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_wmma_fp16) add_example_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_wmma_fp16)
endif()
set(target 1) set(target 1)
endif() endif()
endforeach() endforeach()
example/39_permute/CMakeLists.txt
View file @ 0c823497
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)
if(result EQUAL 0) add_example_dependencies(example_permute example_permute_1xHxW_fp16)
add_dependencies(example_permute example_permute_1xHxW_fp16)
endif()
add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp) add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_permute example_permute_NxHxW_fp16)
add_dependencies(example_permute example_permute_NxHxW_fp16)
endif()
add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp) add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
if(result EQUAL 0) add_example_dependencies(example_permute example_permute_HxWx4_fp16)
add_dependencies(example_permute example_permute_HxWx4_fp16)
endif()
example/42_groupnorm/CMakeLists.txt deleted 100644 → 0
View file @ 334cfe1c
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)
example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp deleted 100644 → 0
View file @ 334cfe1c
// 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 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::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_swish_fp16.cpp deleted 100644 → 0
View file @ 334cfe1c
// 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_fwd/CMakeLists.txt 0 → 100644
View file @ 0c823497
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/common.hpp → example/42_groupnorm_fwd/common.hpp
View file @ 0c823497
...@@ -11,8 +11,8 @@ ...@@ -11,8 +11,8 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/utility/reduction_enums.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_fwd_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_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/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/fill.hpp" #include "ck/library/utility/fill.hpp"
......
example/42_groupnorm_fwd/groupnorm_fwd_sigmoid_mul_fp16.cpp 0 → 100644
View file @ 0c823497
// 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 @ 0c823497
// 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); }
example/42_groupnorm_fwd/groupnorm_fwd_swish_fp16.cpp 0 → 100644
View file @ 0c823497
// 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::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); }
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