Skip to content

GitLab

Commit e5ebcc41 authored by Artur Wojcik's avatar Artur Wojcik
Browse files

Merge branch 'develop' into uif2-migraphx

parents 57cdd70b abac8b07
Hide whitespace changes
Inline Side-by-side
Showing with 367 additions and 113 deletions
example/27_layernorm/common.hpp → example/27_layernorm2d_fwd/common.hpp
View file @ e5ebcc41
......@@ -10,8 +10,8 @@
#include <getopt.h>
#include "ck/ck.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/impl/device_normalization_fwd_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/device_memory.hpp"
......
example/27_layernorm2d_fwd/layernorm2d_fwd_fp16.cpp 0 → 100644
View file @ e5ebcc41
// 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 @ e5ebcc41
// 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 @ e5ebcc41
......@@ -4,28 +4,19 @@
#pragma once
template <typename DeviceInstance>
int run_groupnorm_example()
int run_layernorm2d_fwd_example()
{
bool time_kernel = false;
ck::index_t M = 1024;
ck::index_t N = 1024;
ck::index_t Stride = N;
ck::index_t M = 1024;
ck::index_t N = 1024;
auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
return HostTensorDescriptor({len}, {stride});
};
auto f_host_tensor_descriptor2d = [](std::size_t row, std::size_t col, std::size_t stride) {
using namespace ck::literals;
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));
Tensor<XDataType> x({M, N});
Tensor<GammaDataType> gamma({N});
Tensor<BetaDataType> beta({N});
Tensor<YDataType> y({M, N});
Tensor<SaveMeanInvStdDataType> save_mean({M});
Tensor<SaveMeanInvStdDataType> save_inv_std({M});
x.GenerateTensorValue(GeneratorTensor_3<XDataType>{0.0, 1.0});
gamma.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{0.0, 1.0});
......@@ -35,6 +26,11 @@ int run_groupnorm_example()
DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
DeviceMem beta_dev(sizeof(BetaDataType) * beta.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());
gamma_dev.ToDevice(gamma.mData.data());
......@@ -47,14 +43,23 @@ int run_groupnorm_example()
{0, 1},
{0, 1},
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},
1e-4,
x_dev.GetDeviceBuffer(),
gamma_dev.GetDeviceBuffer(),
beta_dev.GetDeviceBuffer(),
y_dev.GetDeviceBuffer(),
#ifdef SAVE_MEAN_INV_STD
save_mean_dev.GetDeviceBuffer(),
save_inv_std_dev.GetDeviceBuffer(),
#else
nullptr,
nullptr,
#endif
PassThrough{});
if(!device_instance.IsSupportedArgument(argument_ptr.get()))
......@@ -72,24 +77,45 @@ int run_groupnorm_example()
bool pass = true;
{
Tensor<YDataType> host_y(f_host_tensor_descriptor2d(M, N, Stride));
using ReferenceInstance = ck::tensor_operation::host::ReferenceLayernorm<XDataType,
GammaDataType,
BetaDataType,
YDataType,
ComputeDataType,
PassThrough,
Rank,
NumReduceDim>;
Tensor<YDataType> host_y({M, N});
Tensor<SaveMeanInvStdDataType> host_save_mean({M});
Tensor<SaveMeanInvStdDataType> host_save_inv_std({M});
using ReferenceInstance =
ck::tensor_operation::host::ReferenceLayernorm<XDataType,
GammaDataType,
BetaDataType,
YDataType,
SaveMeanInvStdDataType,
ComputeDataType,
PassThrough,
Rank,
NumReduceDim>;
ReferenceInstance ref;
auto ref_argument =
ref.MakeArgument(x, gamma, beta, host_y, PassThrough{}, {M, N}, {1}, 1e-4);
auto ref_invoker = ref.MakeInvoker();
auto ref_argument = ref.MakeArgument(x,
gamma,
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);
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);
......
example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
View file @ e5ebcc41
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
endif()
add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
example/29_batched_gemm_bias_e_permute/CMakeLists.txt
View file @ e5ebcc41
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
endif()
if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
endif()
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
View file @ e5ebcc41
......@@ -3,44 +3,38 @@ list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
add_custom_target(example_grouped_conv_fwd_multiple_d)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
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_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
endif()
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
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_int8)
endif()
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)
add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
endif() # USE_BITINT_EXTENSION_INT4
set(target 1)
endif()
if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
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)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
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)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
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_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
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)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
endif() # USE_BITINT_EXTENSION_INT4
set(target 1)
endif()
endforeach()
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
endif()
set(target 1)
endif()
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_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
set(target 1)
endif()
endforeach()
example/30_grouped_conv_fwd_multiple_d/README.md
View file @ e5ebcc41
......@@ -4,7 +4,7 @@ arg1: verification (0=no, 1=yes)
arg2: initialization (0=no init, 1=integer value, 2=decimal value)
arg3: time kernel (0=no, 1=yes)
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,
<filter spatial dimensions>, (ie Y, X for 2D)
<input image spatial dimensions>, (ie Hi, Wi for 2D)
......@@ -26,5 +26,5 @@ out: dim 5, lengths {1, 128, 256, 36, 36}, strides {256, 331776, 1, 9216, 256}
launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1}
Warm up 1 time
Start running 10 times...
Perf: 1.55981 ms, 94.0927 TFlops, 213.868 GB/s, DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<256, 128, 256, 16, Default>
Perf: 1.55981 ms, 94.0927 TFlops, 213.868 GB/s, DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<256, 128, 256, 16, Default>
```
example/30_grouped_conv_fwd_multiple_d/common.hpp
View file @ e5ebcc41
......@@ -12,7 +12,7 @@
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/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"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
......
example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
View file @ e5ebcc41
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
#error Should compile this file with ck::int4_t support
#endif
#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
#include "common.hpp"
......@@ -29,3 +27,4 @@ using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
#include "run_grouped_conv_fwd_bias_relu_add_example.inc"
int main(int argc, char* argv[]) { return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv); }
#endif
example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc
View file @ e5ebcc41
......@@ -34,7 +34,7 @@ using ResidualLayout = typename LayoutSettingSelector<NDimSpatial>::ResidualLayo
template <ck::index_t NDimSpatial>
using DeviceConvFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
NDimSpatial,
InputLayout<NDimSpatial>,
WeightLayout<NDimSpatial>,
......
example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc
View file @ e5ebcc41
......@@ -3,7 +3,7 @@
template <ck::index_t NDimSpatial>
using DeviceConvFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<
NDimSpatial,
InputLayout<NDimSpatial>,
WeightLayout<NDimSpatial>,
......
example/31_batched_gemm_gemm/CMakeLists.txt
View file @ e5ebcc41
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list2 gfx908 gfx90a)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
endif()
add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
endif(USE_BITINT_EXTENSION_INT4)
......@@ -20,7 +14,5 @@ foreach(gpu IN LISTS GPU_TARGETS)
endforeach()
if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
endif()
add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
endif()
example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp
View file @ e5ebcc41
......@@ -9,9 +9,7 @@ Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
Gemm1
*/
#ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
#error Should compile this file with ck::int4_t support
#endif
#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
#include <iostream>
#include <numeric>
......@@ -144,3 +142,4 @@ static_assert(sizeof(ck::int4_t) == sizeof(int8_t));
#endif
int main(int argc, char* argv[]) { return run_batched_gemm_gemm_example(argc, argv) ? 0 : 1; }
#endif
example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
View file @ e5ebcc41
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
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_permute_xdl_fp16 batched_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)
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_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
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_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
endif()
add_custom_target(example_gemm_scale_softmax_gemm)
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_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_permute_xdl_fp16)
add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
add_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_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_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_permute_xdl_bf16)
endif()
add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
add_example_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
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_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
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_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
example/35_splitK_gemm/CMakeLists.txt
View file @ e5ebcc41
......@@ -3,26 +3,30 @@ set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_splitK_gemm_xdl)
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bfp16)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
endif()
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
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)
add_example_executable(example_splitK_gemm_xdl_fp16_fp8 splitK_gemm_xdl_fp16_fp8.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16_fp8)
add_example_executable(example_splitK_gemm_xdl_lds_direct_load_fp16 splitK_gemm_xdl_lds_direct_load_fp16.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_lds_direct_load_fp16)
add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16)
add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
endif()
set(target 1)
endif()
endforeach()
example/35_splitK_gemm/run_splitK_gemm_example.inc
View file @ e5ebcc41
......@@ -157,7 +157,7 @@ bool run_splitK_gemm(const ProblemSize& problem_size, const ExecutionConfig& con
if(config.time_kernel)
{
float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel, 1});
std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype =
......
example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
View file @ e5ebcc41
......@@ -42,7 +42,7 @@ using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::KPadding;
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle
// clang-format off
......
example/35_splitK_gemm/splitK_gemm_xdl_fp16_fp8.cpp 0 → 100644
View file @ e5ebcc41
// SPDX-License-Identifier: MIT
// Copyright (c) 2024 Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/literals.hpp"
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using F8 = ck::f8_t;
using F16 = ck::half_t;
using F32 = float;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using ADataType = F16;
using BDataType = F8;
using AccDataType = F32;
using CDataType = F16;
using ALayout = Row;
using BLayout = Col;
using CLayout = Row;
using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle
// clang-format off
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 128, 16, 64, 8, 16, 16, 16, 1, 2, S<1, 8, 8, 2>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 16, 16, true, 1, 1, S<1, 16, 1, 8>, 4, F16, ck::PipelineVersion::v1, ck::LoopScheduler::Default, ADataType, BDataType>;
// clang-format on
#include "run_splitK_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_splitK_gemm_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