Skip to content

GitLab

"include/vscode:/vscode.git/clone" did not exist on "35b59ac6cb021a23d9e1da93ff0696eb82e84a51"
Unverified Commit a93d07c7 authored by Illia Silin's avatar Illia Silin Committed by GitHub
Browse files

Merge branch 'develop' into ck_codegen_build

parents 9d9ad510 afbf6350
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 452 additions and 162 deletions
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/large_tensor/device_grouped_conv3d_fwd_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp 0 → 100644
View file @ a93d07c7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_large_tensor_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_large_tensor_f16_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/large_tensor/device_grouped_conv3d_fwd_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp 0 → 100644
View file @ a93d07c7
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_large_tensor_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_large_tensor_ndhwgc_gkzyxc_ndhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_large_tensor_f32_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/mha/CMakeLists.txt 0 → 100644
View file @ a93d07c7
set(FMHA_CPP_FOLDER ${CMAKE_CURRENT_BINARY_DIR})
set(FMHA_SRC_FOLDER ${CMAKE_SOURCE_DIR}/example/ck_tile/01_fmha/)
set(CK_TILE_SRC_FOLDER ${CMAKE_SOURCE_DIR}/include/ck_tile/)
# python stuff
find_package(PythonInterp 3 REQUIRED)
rocm_install(DIRECTORY ${CK_TILE_SRC_FOLDER} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/ck_tile)
rocm_install(FILES
"${FMHA_SRC_FOLDER}/fmha_fwd.hpp"
"${FMHA_SRC_FOLDER}/bias.hpp"
"${FMHA_SRC_FOLDER}/mask.hpp"
DESTINATION include/ck_tile/ops
)
# header for building lib
file(COPY ${FMHA_SRC_FOLDER}/fmha_fwd.hpp DESTINATION ${FMHA_CPP_FOLDER})
file(COPY ${FMHA_SRC_FOLDER}/bias.hpp DESTINATION ${FMHA_CPP_FOLDER})
file(COPY ${FMHA_SRC_FOLDER}/mask.hpp DESTINATION ${FMHA_CPP_FOLDER})
# generate a list of kernels, but not actually emit files at config stage
execute_process(
COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_SOURCE_DIR}/example/ck_tile/01_fmha/generate.py
--list_blobs ${FMHA_CPP_FOLDER}/blob_list.txt
)
file(STRINGS ${FMHA_CPP_FOLDER}/blob_list.txt FMHA_FWD_GEN_BLOBS)
# actually generate the cpp files
add_custom_command(
OUTPUT ${FMHA_FWD_GEN_BLOBS}
COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_SOURCE_DIR}/example/ck_tile/01_fmha/generate.py
--output_dir ${FMHA_CPP_FOLDER}
COMMENT "Generating mha kernel (cpp) files now ..."
VERBATIM
)
# This is done to remove path info and just
# have filename. Since, it was cauing the cmake
# to throw "File name too long"
set(device_files)
foreach(filepath IN LISTS FMHA_FWD_GEN_BLOBS)
get_filename_component(filename ${filepath} NAME)
# Append the filename to the device_files list
list(APPEND device_files ${filename})
endforeach()
add_custom_target(generate_cpp_files DEPENDS ${FMHA_FWD_GEN_BLOBS})
add_instance_library(device_mha_instance ${device_files})
if (TARGET device_mha_instance)
add_dependencies(device_mha_instance generate_cpp_files)
endif()
library/src/utility/convolution_parameter.cpp
View file @ a93d07c7
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/host_utility/io.hpp" #include "ck/host_utility/io.hpp"
...@@ -20,6 +20,63 @@ ConvParam::ConvParam(ck::index_t n_dim, ...@@ -20,6 +20,63 @@ ConvParam::ConvParam(ck::index_t n_dim,
const std::vector<ck::index_t>& dilations, const std::vector<ck::index_t>& dilations,
const std::vector<ck::index_t>& left_pads, const std::vector<ck::index_t>& left_pads,
const std::vector<ck::index_t>& right_pads) const std::vector<ck::index_t>& right_pads)
: num_dim_spatial_(static_cast<ck::long_index_t>(n_dim)),
G_(static_cast<ck::long_index_t>(group_count)),
N_(static_cast<ck::long_index_t>(n_batch)),
K_(static_cast<ck::long_index_t>(n_out_channels)),
C_(static_cast<ck::long_index_t>(n_in_channels)),
filter_spatial_lengths_(num_dim_spatial_),
input_spatial_lengths_(num_dim_spatial_),
output_spatial_lengths_(num_dim_spatial_),
conv_filter_strides_(num_dim_spatial_),
conv_filter_dilations_(num_dim_spatial_),
input_left_pads_(num_dim_spatial_),
input_right_pads_(num_dim_spatial_)
{
if(static_cast<ck::index_t>(filter_spatial_lengths_.size()) != num_dim_spatial_ ||
static_cast<ck::index_t>(input_spatial_lengths_.size()) != num_dim_spatial_ ||
static_cast<ck::index_t>(conv_filter_strides_.size()) != num_dim_spatial_ ||
static_cast<ck::index_t>(conv_filter_dilations_.size()) != num_dim_spatial_ ||
static_cast<ck::index_t>(input_left_pads_.size()) != num_dim_spatial_ ||
static_cast<ck::index_t>(input_right_pads_.size()) != num_dim_spatial_)
{
throw(
std::runtime_error("ConvParam::ConvParam: "
"parameter size is different from number of declared dimensions!"));
}
for(ck::index_t i = 0; i < num_dim_spatial_; ++i)
{
filter_spatial_lengths_[i] = static_cast<ck::long_index_t>(filters_len[i]);
input_spatial_lengths_[i] = static_cast<ck::long_index_t>(input_len[i]);
conv_filter_strides_[i] = static_cast<ck::long_index_t>(strides[i]);
conv_filter_dilations_[i] = static_cast<ck::long_index_t>(dilations[i]);
input_left_pads_[i] = static_cast<ck::long_index_t>(left_pads[i]);
input_right_pads_[i] = static_cast<ck::long_index_t>(right_pads[i]);
// XEff = (X - 1) * conv_dilation_w + 1;
// Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
const ck::long_index_t x_eff =
(filter_spatial_lengths_[i] - 1) * conv_filter_dilations_[i] + 1;
output_spatial_lengths_[i] =
(input_spatial_lengths_[i] + input_left_pads_[i] + input_right_pads_[i] - x_eff) /
conv_filter_strides_[i] +
1;
}
}
ConvParam::ConvParam(ck::long_index_t n_dim,
ck::long_index_t group_count,
ck::long_index_t n_batch,
ck::long_index_t n_out_channels,
ck::long_index_t n_in_channels,
const std::vector<ck::long_index_t>& filters_len,
const std::vector<ck::long_index_t>& input_len,
const std::vector<ck::long_index_t>& strides,
const std::vector<ck::long_index_t>& dilations,
const std::vector<ck::long_index_t>& left_pads,
const std::vector<ck::long_index_t>& right_pads)
: num_dim_spatial_(n_dim), : num_dim_spatial_(n_dim),
G_(group_count), G_(group_count),
N_(n_batch), N_(n_batch),
...@@ -49,7 +106,8 @@ ConvParam::ConvParam(ck::index_t n_dim, ...@@ -49,7 +106,8 @@ ConvParam::ConvParam(ck::index_t n_dim,
{ {
// XEff = (X - 1) * conv_dilation_w + 1; // XEff = (X - 1) * conv_dilation_w + 1;
// Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1; // Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
const ck::index_t x_eff = (filter_spatial_lengths_[i] - 1) * conv_filter_dilations_[i] + 1; const ck::long_index_t x_eff =
(filter_spatial_lengths_[i] - 1) * conv_filter_dilations_[i] + 1;
output_spatial_lengths_[i] = output_spatial_lengths_[i] =
(input_spatial_lengths_[i] + input_left_pads_[i] + input_right_pads_[i] - x_eff) / (input_spatial_lengths_[i] + input_left_pads_[i] + input_right_pads_[i] - x_eff) /
...@@ -63,7 +121,7 @@ ConvParam::ConvParam() ...@@ -63,7 +121,7 @@ ConvParam::ConvParam()
{ {
} }
std::vector<ck::index_t> ConvParam::GetOutputSpatialLengths() const std::vector<ck::long_index_t> ConvParam::GetOutputSpatialLengths() const
{ {
return output_spatial_lengths_; return output_spatial_lengths_;
} }
...@@ -97,46 +155,46 @@ std::string get_conv_param_parser_helper_msg() ...@@ -97,46 +155,46 @@ std::string get_conv_param_parser_helper_msg()
ck::utils::conv::ConvParam parse_conv_param(int num_dim_spatial, int arg_idx, char* const argv[]) ck::utils::conv::ConvParam parse_conv_param(int num_dim_spatial, int arg_idx, char* const argv[])
{ {
const ck::index_t G = std::stoi(argv[arg_idx++]); const ck::long_index_t G = std::stol(argv[arg_idx++]);
const ck::index_t N = std::stoi(argv[arg_idx++]); const ck::long_index_t N = std::stol(argv[arg_idx++]);
const ck::index_t K = std::stoi(argv[arg_idx++]); const ck::long_index_t K = std::stol(argv[arg_idx++]);
const ck::index_t C = std::stoi(argv[arg_idx++]); const ck::long_index_t C = std::stol(argv[arg_idx++]);
std::vector<ck::index_t> filter_spatial_lengths(num_dim_spatial); std::vector<ck::long_index_t> filter_spatial_lengths(num_dim_spatial);
std::vector<ck::index_t> input_spatial_lengths(num_dim_spatial); std::vector<ck::long_index_t> input_spatial_lengths(num_dim_spatial);
std::vector<ck::index_t> conv_filter_strides(num_dim_spatial); std::vector<ck::long_index_t> conv_filter_strides(num_dim_spatial);
std::vector<ck::index_t> conv_filter_dilations(num_dim_spatial); std::vector<ck::long_index_t> conv_filter_dilations(num_dim_spatial);
std::vector<ck::index_t> input_left_pads(num_dim_spatial); std::vector<ck::long_index_t> input_left_pads(num_dim_spatial);
std::vector<ck::index_t> input_right_pads(num_dim_spatial); std::vector<ck::long_index_t> input_right_pads(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); filter_spatial_lengths[i] = std::stol(argv[arg_idx++]);
} }
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); input_spatial_lengths[i] = std::stol(argv[arg_idx++]);
} }
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
conv_filter_strides[i] = std::stoi(argv[arg_idx++]); conv_filter_strides[i] = std::stol(argv[arg_idx++]);
} }
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); conv_filter_dilations[i] = std::stol(argv[arg_idx++]);
} }
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
input_left_pads[i] = std::stoi(argv[arg_idx++]); input_left_pads[i] = std::stol(argv[arg_idx++]);
} }
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
input_right_pads[i] = std::stoi(argv[arg_idx++]); input_right_pads[i] = std::stol(argv[arg_idx++]);
} }
return ck::utils::conv::ConvParam{num_dim_spatial, return ck::utils::conv::ConvParam{num_dim_spatial,
......
profiler/include/profiler/profile_conv_bwd_data_impl.hpp
View file @ a93d07c7
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
...@@ -82,6 +82,29 @@ bool profile_conv_bwd_data_impl(int do_verification, ...@@ -82,6 +82,29 @@ bool profile_conv_bwd_data_impl(int do_verification,
Tensor<WeiDataType> weight(wei_g_k_c_xs_desc); Tensor<WeiDataType> weight(wei_g_k_c_xs_desc);
Tensor<OutDataType> output(out_g_n_k_wos_desc); Tensor<OutDataType> output(out_g_n_k_wos_desc);
std::vector<ck::index_t> input_spatial_lengths_i32(NDimSpatial);
std::vector<ck::index_t> filter_spatial_lengths_i32(NDimSpatial);
std::vector<ck::index_t> output_spatial_lengths_i32(NDimSpatial);
std::vector<ck::index_t> conv_filter_strides_i32(NDimSpatial);
std::vector<ck::index_t> conv_filter_dilations_i32(NDimSpatial);
std::vector<ck::index_t> input_left_pads_i32(NDimSpatial);
std::vector<ck::index_t> input_right_pads_i32(NDimSpatial);
for(ck::index_t d = 0; d < NDimSpatial; d++)
{
input_spatial_lengths_i32[d] =
static_cast<ck::index_t>(conv_param.input_spatial_lengths_[d]);
filter_spatial_lengths_i32[d] =
static_cast<ck::index_t>(conv_param.filter_spatial_lengths_[d]);
output_spatial_lengths_i32[d] =
static_cast<ck::index_t>(conv_param.GetOutputSpatialLengths()[d]);
conv_filter_strides_i32[d] = static_cast<ck::index_t>(conv_param.conv_filter_strides_[d]);
conv_filter_dilations_i32[d] =
static_cast<ck::index_t>(conv_param.conv_filter_dilations_[d]);
input_left_pads_i32[d] = static_cast<ck::index_t>(conv_param.input_left_pads_[d]);
input_right_pads_i32[d] = static_cast<ck::index_t>(conv_param.input_right_pads_[d]);
}
std::cout << "input: " << input_host_result.mDesc << std::endl; std::cout << "input: " << input_host_result.mDesc << std::endl;
std::cout << "weight: " << weight.mDesc << std::endl; std::cout << "weight: " << weight.mDesc << std::endl;
std::cout << "output: " << output.mDesc << std::endl; std::cout << "output: " << output.mDesc << std::endl;
...@@ -161,16 +184,16 @@ bool profile_conv_bwd_data_impl(int do_verification, ...@@ -161,16 +184,16 @@ bool profile_conv_bwd_data_impl(int do_verification,
op_ptr->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()), op_ptr->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()), static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()), static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
conv_param.N_, static_cast<ck::index_t>(conv_param.N_),
conv_param.K_, static_cast<ck::index_t>(conv_param.K_),
conv_param.C_, static_cast<ck::index_t>(conv_param.C_),
conv_param.input_spatial_lengths_, input_spatial_lengths_i32,
conv_param.filter_spatial_lengths_, filter_spatial_lengths_i32,
conv_param.output_spatial_lengths_, output_spatial_lengths_i32,
conv_param.conv_filter_strides_, conv_filter_strides_i32,
conv_param.conv_filter_dilations_, conv_filter_dilations_i32,
conv_param.input_left_pads_, input_left_pads_i32,
conv_param.input_right_pads_, input_right_pads_i32,
in_element_op, in_element_op,
wei_element_op, wei_element_op,
out_element_op); out_element_op);
......
profiler/include/profiler/profile_conv_fwd_impl.hpp
View file @ a93d07c7
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
...@@ -60,6 +60,29 @@ bool profile_conv_fwd_impl(int do_verification, ...@@ -60,6 +60,29 @@ bool profile_conv_fwd_impl(int do_verification,
Tensor<OutDataType> host_output(out_g_n_k_wos_desc); Tensor<OutDataType> host_output(out_g_n_k_wos_desc);
Tensor<OutDataType> device_output(out_g_n_k_wos_desc); Tensor<OutDataType> device_output(out_g_n_k_wos_desc);
std::vector<ck::index_t> input_spatial_lengths_i32(NDimSpatial);
std::vector<ck::index_t> filter_spatial_lengths_i32(NDimSpatial);
std::vector<ck::index_t> output_spatial_lengths_i32(NDimSpatial);
std::vector<ck::index_t> conv_filter_strides_i32(NDimSpatial);
std::vector<ck::index_t> conv_filter_dilations_i32(NDimSpatial);
std::vector<ck::index_t> input_left_pads_i32(NDimSpatial);
std::vector<ck::index_t> input_right_pads_i32(NDimSpatial);
for(ck::index_t d = 0; d < NDimSpatial; d++)
{
input_spatial_lengths_i32[d] =
static_cast<ck::index_t>(conv_param.input_spatial_lengths_[d]);
filter_spatial_lengths_i32[d] =
static_cast<ck::index_t>(conv_param.filter_spatial_lengths_[d]);
output_spatial_lengths_i32[d] =
static_cast<ck::index_t>(conv_param.GetOutputSpatialLengths()[d]);
conv_filter_strides_i32[d] = static_cast<ck::index_t>(conv_param.conv_filter_strides_[d]);
conv_filter_dilations_i32[d] =
static_cast<ck::index_t>(conv_param.conv_filter_dilations_[d]);
input_left_pads_i32[d] = static_cast<ck::index_t>(conv_param.input_left_pads_[d]);
input_right_pads_i32[d] = static_cast<ck::index_t>(conv_param.input_right_pads_[d]);
}
std::cout << "input: " << input.mDesc << std::endl; std::cout << "input: " << input.mDesc << std::endl;
std::cout << "weight: " << weight.mDesc << std::endl; std::cout << "weight: " << weight.mDesc << std::endl;
std::cout << "output: " << host_output.mDesc << std::endl; std::cout << "output: " << host_output.mDesc << std::endl;
...@@ -143,16 +166,16 @@ bool profile_conv_fwd_impl(int do_verification, ...@@ -143,16 +166,16 @@ bool profile_conv_fwd_impl(int do_verification,
op_ptr->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()), op_ptr->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()), static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()), static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
conv_param.N_, static_cast<ck::index_t>(conv_param.N_),
conv_param.K_, static_cast<ck::index_t>(conv_param.K_),
conv_param.C_, static_cast<ck::index_t>(conv_param.C_),
conv_param.input_spatial_lengths_, input_spatial_lengths_i32,
conv_param.filter_spatial_lengths_, filter_spatial_lengths_i32,
conv_param.GetOutputSpatialLengths(), output_spatial_lengths_i32,
conv_param.conv_filter_strides_, conv_filter_strides_i32,
conv_param.conv_filter_dilations_, conv_filter_dilations_i32,
conv_param.input_left_pads_, input_left_pads_i32,
conv_param.input_right_pads_, input_right_pads_i32,
in_element_op, in_element_op,
wei_element_op, wei_element_op,
out_element_op); out_element_op);
......
profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
View file @ a93d07c7
...@@ -33,7 +33,8 @@ template <ck::index_t NDimSpatial, ...@@ -33,7 +33,8 @@ template <ck::index_t NDimSpatial,
typename WeiDataType, typename WeiDataType,
typename OutDataType, typename OutDataType,
typename AComputeType = InDataType, typename AComputeType = InDataType,
typename BComputeType = AComputeType> typename BComputeType = AComputeType,
typename IndexType = ck::index_t>
bool profile_grouped_conv_fwd_impl(int do_verification, bool profile_grouped_conv_fwd_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
...@@ -57,16 +58,16 @@ bool profile_grouped_conv_fwd_impl(int do_verification, ...@@ -57,16 +58,16 @@ bool profile_grouped_conv_fwd_impl(int do_verification,
const auto out_g_n_k_wos_desc = const auto out_g_n_k_wos_desc =
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param); ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{}; std::array<IndexType, NDimSpatial + 3> a_g_n_c_wis_lengths{};
std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{}; std::array<IndexType, NDimSpatial + 3> a_g_n_c_wis_strides{};
std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{}; std::array<IndexType, NDimSpatial + 3> b_g_k_c_xs_lengths{};
std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{}; std::array<IndexType, NDimSpatial + 3> b_g_k_c_xs_strides{};
std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{}; std::array<IndexType, NDimSpatial + 3> e_g_n_k_wos_lengths{};
std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{}; std::array<IndexType, NDimSpatial + 3> e_g_n_k_wos_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_strides{}; std::array<IndexType, NDimSpatial> conv_filter_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{}; std::array<IndexType, NDimSpatial> conv_filter_dilations{};
std::array<ck::index_t, NDimSpatial> input_left_pads{}; std::array<IndexType, NDimSpatial> input_left_pads{};
std::array<ck::index_t, NDimSpatial> input_right_pads{}; std::array<IndexType, NDimSpatial> input_right_pads{};
auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); }; auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
......
profiler/src/CMakeLists.txt
View file @ a93d07c7
...@@ -82,6 +82,11 @@ set(PROFILER_EXECUTABLE ckProfiler) ...@@ -82,6 +82,11 @@ set(PROFILER_EXECUTABLE ckProfiler)
add_executable(${PROFILER_EXECUTABLE} ${PROFILER_SOURCES}) add_executable(${PROFILER_EXECUTABLE} ${PROFILER_SOURCES})
target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors) target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors)
# flags to compress the library
if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600241132)
message("Adding --offload-compress flag for ${PROFILER_EXECUTABLE}")
target_compile_options(${PROFILER_EXECUTABLE} PRIVATE --offload-compress)
endif()
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility getopt::getopt) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility getopt::getopt)
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance) target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
......
profiler/src/profile_grouped_conv_fwd.cpp
View file @ a93d07c7
This diff is collapsed.
test/conv_util/conv_util.cpp
View file @ a93d07c7
This diff is collapsed.
test/smfmac_op/smfmac_op_xdl.cpp
View file @ a93d07c7
This diff is collapsed.
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