Skip to content

GitLab

"web/git@developer.sourcefind.cn:orangecat/ollama.git" did not exist on "df67a90e644f0c406b2c03d60741e4c19ef6da34"
Unverified Commit 712e464c authored by Adam Osewski's avatar Adam Osewski Committed by GitHub
Browse files

Post PR183 review fixes. (#224) 



* Suppress additional warnings for googltest.

* Rename file conv_fwd_util to conv_util.

* Update includes and ConvParams member access.

* Formatting.

* Change conv_fwd_util target to conv_util

* Fix compiler errors.

* Fix leftovers.
Co-authored-by: default avatarAdam Osewski <aosewski@amd.com>
Co-authored-by: default avatarChao Liu <chao.liu2@amd.com>
parent f03a1738
Hide whitespace changes
Inline Side-by-side
Showing with 450 additions and 447 deletions
cmake/googletest.cmake
View file @ 712e464c
...@@ -18,6 +18,7 @@ list(APPEND GTEST_CMAKE_CXX_FLAGS ...@@ -18,6 +18,7 @@ list(APPEND GTEST_CMAKE_CXX_FLAGS
-Wno-switch-enum -Wno-switch-enum
-Wno-zero-as-null-pointer-constant -Wno-zero-as-null-pointer-constant
-Wno-unused-member-function -Wno-unused-member-function
-Wno-comma
) )
message(STATUS "Suppressing googltest warnings with flags: ${GTEST_CMAKE_CXX_FLAGS}") message(STATUS "Suppressing googltest warnings with flags: ${GTEST_CMAKE_CXX_FLAGS}")
...@@ -33,4 +34,5 @@ FetchContent_MakeAvailable(googletest) ...@@ -33,4 +34,5 @@ FetchContent_MakeAvailable(googletest)
target_compile_options(gtest PRIVATE ${GTEST_CMAKE_CXX_FLAGS}) target_compile_options(gtest PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
target_compile_options(gtest_main PRIVATE ${GTEST_CMAKE_CXX_FLAGS}) target_compile_options(gtest_main PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
target_compile_options(gmock PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
example/06_conv2d_fwd_bias_relu/CMakeLists.txt
View file @ 712e464c
add_example_executable(example_conv2d_fwd_xdl_bias_relu conv2d_fwd_xdl_bias_relu.cpp) add_example_executable(example_conv2d_fwd_xdl_bias_relu conv2d_fwd_xdl_bias_relu.cpp)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu PRIVATE conv_fwd_util) target_link_libraries(example_conv2d_fwd_xdl_bias_relu PRIVATE conv_util)
example/06_conv2d_fwd_bias_relu/conv2d_fwd_xdl_bias_relu.cpp
View file @ 712e464c
...@@ -7,7 +7,7 @@ ...@@ -7,7 +7,7 @@
#include "check_err.hpp" #include "check_err.hpp"
#include "config.hpp" #include "config.hpp"
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "device.hpp" #include "device.hpp"
#include "device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp" #include "device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp"
#include "device_tensor.hpp" #include "device_tensor.hpp"
...@@ -120,40 +120,40 @@ ck::utils::conv::ConvParams ParseConvParams(int argc, char* argv[]) ...@@ -120,40 +120,40 @@ ck::utils::conv::ConvParams ParseConvParams(int argc, char* argv[])
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
int arg_idx = 4; int arg_idx = 4;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -184,21 +184,21 @@ int main(int argc, char* argv[]) ...@@ -184,21 +184,21 @@ int main(int argc, char* argv[])
params = ParseConvParams(argc, argv); params = ParseConvParams(argc, argv);
} }
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims), input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths), std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths)); std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K), std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims), filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths), std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths)); std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths(); const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K)}; static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims), output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths), std::begin(output_spatial_lengths),
std::end(output_spatial_lengths)); std::end(output_spatial_lengths));
...@@ -211,7 +211,7 @@ int main(int argc, char* argv[]) ...@@ -211,7 +211,7 @@ int main(int argc, char* argv[])
get_output_host_tensor_descriptor(output_dims, num_dim_spatial)); get_output_host_tensor_descriptor(output_dims, num_dim_spatial));
// bias: assume contiguous 1d vector // bias: assume contiguous 1d vector
Tensor<OutDataType> bias( Tensor<OutDataType> bias(
HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K)}))); HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K_)})));
std::cout << "input: " << input.mDesc << std::endl; std::cout << "input: " << input.mDesc << std::endl;
std::cout << "weights: " << weights.mDesc << std::endl; std::cout << "weights: " << weights.mDesc << std::endl;
...@@ -248,16 +248,16 @@ int main(int argc, char* argv[]) ...@@ -248,16 +248,16 @@ int main(int argc, char* argv[])
static_cast<const WeiDataType*>(wei_device_buf.GetDeviceBuffer()), static_cast<const WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()), static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
static_cast<const OutDataType*>(bias_device_buf.GetDeviceBuffer()), static_cast<const OutDataType*>(bias_device_buf.GetDeviceBuffer()),
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths, output_spatial_lengths,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
...@@ -272,15 +272,15 @@ int main(int argc, char* argv[]) ...@@ -272,15 +272,15 @@ int main(int argc, char* argv[])
float ave_time = invoker.Run(argument, nrepeat); float ave_time = invoker.Run(argument, nrepeat);
std::size_t flop = get_flops( std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths); params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = std::size_t num_btype =
get_btype<InDataType, WeiDataType, OutDataType>(params.N, get_btype<InDataType, WeiDataType, OutDataType>(params.N_,
params.C, params.C_,
params.K, params.K_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths) + output_spatial_lengths) +
sizeof(OutDataType) * (params.K); sizeof(OutDataType) * (params.K_);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time; float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time; float gb_per_sec = num_btype / 1.E6 / ave_time;
...@@ -296,10 +296,10 @@ int main(int argc, char* argv[]) ...@@ -296,10 +296,10 @@ int main(int argc, char* argv[])
weights, weights,
host_output, host_output,
bias, bias,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
......
example/07_conv2d_fwd_bias_relu_add/CMakeLists.txt
View file @ 712e464c
add_example_executable(example_conv2d_fwd_xdl_bias_relu_add conv2d_fwd_xdl_bias_relu_add.cpp) add_example_executable(example_conv2d_fwd_xdl_bias_relu_add conv2d_fwd_xdl_bias_relu_add.cpp)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu_add PRIVATE conv_fwd_util) target_link_libraries(example_conv2d_fwd_xdl_bias_relu_add PRIVATE conv_util)
example/07_conv2d_fwd_bias_relu_add/conv2d_fwd_xdl_bias_relu_add.cpp
View file @ 712e464c
...@@ -7,7 +7,7 @@ ...@@ -7,7 +7,7 @@
#include "check_err.hpp" #include "check_err.hpp"
#include "config.hpp" #include "config.hpp"
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "device.hpp" #include "device.hpp"
#include "device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp" #include "device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp"
#include "device_tensor.hpp" #include "device_tensor.hpp"
...@@ -117,40 +117,40 @@ ck::utils::conv::ConvParams ParseConvParams(int argc, char* argv[]) ...@@ -117,40 +117,40 @@ ck::utils::conv::ConvParams ParseConvParams(int argc, char* argv[])
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
int arg_idx = 4; int arg_idx = 4;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -181,21 +181,21 @@ int main(int argc, char* argv[]) ...@@ -181,21 +181,21 @@ int main(int argc, char* argv[])
params = ParseConvParams(argc, argv); params = ParseConvParams(argc, argv);
} }
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims), input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths), std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths)); std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K), std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims), filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths), std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths)); std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths(); const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K)}; static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims), output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths), std::begin(output_spatial_lengths),
std::end(output_spatial_lengths)); std::end(output_spatial_lengths));
...@@ -209,7 +209,7 @@ int main(int argc, char* argv[]) ...@@ -209,7 +209,7 @@ int main(int argc, char* argv[])
// bias: assume contiguous 1d vector // bias: assume contiguous 1d vector
Tensor<OutDataType> bias( Tensor<OutDataType> bias(
HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K)}))); HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K_)})));
// residual: assume same layout as output tensor // residual: assume same layout as output tensor
Tensor<OutDataType> residual(get_output_host_tensor_descriptor(output_dims, num_dim_spatial)); Tensor<OutDataType> residual(get_output_host_tensor_descriptor(output_dims, num_dim_spatial));
...@@ -259,16 +259,16 @@ int main(int argc, char* argv[]) ...@@ -259,16 +259,16 @@ int main(int argc, char* argv[])
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()), static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
static_cast<const OutDataType*>(bias_device_buf.GetDeviceBuffer()), static_cast<const OutDataType*>(bias_device_buf.GetDeviceBuffer()),
static_cast<const OutDataType*>(resi_device_buf.GetDeviceBuffer()), static_cast<const OutDataType*>(resi_device_buf.GetDeviceBuffer()),
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths, output_spatial_lengths,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
in_element_op, in_element_op,
wei_element_op, wei_element_op,
out_element_op); out_element_op);
...@@ -283,17 +283,17 @@ int main(int argc, char* argv[]) ...@@ -283,17 +283,17 @@ int main(int argc, char* argv[])
float ave_time = invoker.Run(argument, nrepeat); float ave_time = invoker.Run(argument, nrepeat);
std::size_t flop = get_flops( std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths); params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = std::size_t num_btype =
get_btype<InDataType, WeiDataType, OutDataType>(params.N, get_btype<InDataType, WeiDataType, OutDataType>(params.N_,
params.C, params.C_,
params.K, params.K_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths) + output_spatial_lengths) +
sizeof(OutDataType) * (params.K) + sizeof(OutDataType) * (params.K_) +
sizeof(OutDataType) * sizeof(OutDataType) *
(params.N * params.K * output_spatial_lengths[0] * output_spatial_lengths[1]); (params.N_ * params.K_ * output_spatial_lengths[0] * output_spatial_lengths[1]);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time; float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time; float gb_per_sec = num_btype / 1.E6 / ave_time;
...@@ -310,10 +310,10 @@ int main(int argc, char* argv[]) ...@@ -310,10 +310,10 @@ int main(int argc, char* argv[])
host_output, host_output,
bias, bias,
residual, residual,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
in_element_op, in_element_op,
wei_element_op, wei_element_op,
out_element_op); out_element_op);
......
example/09_convnd_fwd/CMakeLists.txt
View file @ 712e464c
add_example_executable(example_convnd_fwd_xdl convnd_fwd_xdl.cpp) add_example_executable(example_convnd_fwd_xdl convnd_fwd_xdl.cpp)
target_link_libraries(example_convnd_fwd_xdl PRIVATE conv_fwd_util) target_link_libraries(example_convnd_fwd_xdl PRIVATE conv_util)
add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp) add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
target_link_libraries(example_convnd_fwd_xdl_int8 PRIVATE conv_fwd_util) target_link_libraries(example_convnd_fwd_xdl_int8 PRIVATE conv_util)
add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp) add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
target_link_libraries(example_convnd_fwd_xdl_fp16 PRIVATE conv_fwd_util) target_link_libraries(example_convnd_fwd_xdl_fp16 PRIVATE conv_util)
example/09_convnd_fwd/convnd_fwd_xdl.cpp
View file @ 712e464c
...@@ -5,7 +5,7 @@ ...@@ -5,7 +5,7 @@
#include "check_err.hpp" #include "check_err.hpp"
#include "config.hpp" #include "config.hpp"
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "device.hpp" #include "device.hpp"
#include "device_tensor.hpp" #include "device_tensor.hpp"
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp" #include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
...@@ -134,40 +134,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha ...@@ -134,40 +134,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
int arg_idx = 5; int arg_idx = 5;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -199,21 +199,21 @@ int main(int argc, char* argv[]) ...@@ -199,21 +199,21 @@ int main(int argc, char* argv[])
params = parse_conv_params(num_dim_spatial, argc, argv); params = parse_conv_params(num_dim_spatial, argc, argv);
} }
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims), input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths), std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths)); std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K), std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims), filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths), std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths)); std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths(); const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K)}; static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims), output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths), std::begin(output_spatial_lengths),
std::end(output_spatial_lengths)); std::end(output_spatial_lengths));
...@@ -255,16 +255,16 @@ int main(int argc, char* argv[]) ...@@ -255,16 +255,16 @@ int main(int argc, char* argv[])
conv->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()), conv->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()),
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths, output_spatial_lengths,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
...@@ -279,13 +279,13 @@ int main(int argc, char* argv[]) ...@@ -279,13 +279,13 @@ int main(int argc, char* argv[])
float ave_time = invoker->Run(argument.get(), nrepeat); float ave_time = invoker->Run(argument.get(), nrepeat);
std::size_t flop = get_flops( std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths); params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = std::size_t num_btype =
get_btype<InDataType, WeiDataType, OutDataType>(params.N, get_btype<InDataType, WeiDataType, OutDataType>(params.N_,
params.C, params.C_,
params.K, params.K_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths); output_spatial_lengths);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time; float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
...@@ -301,10 +301,10 @@ int main(int argc, char* argv[]) ...@@ -301,10 +301,10 @@ int main(int argc, char* argv[])
auto ref_argument = ref_conv.MakeArgument(input, auto ref_argument = ref_conv.MakeArgument(input,
weights, weights,
host_output, host_output,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
......
example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
View file @ 712e464c
...@@ -5,7 +5,7 @@ ...@@ -5,7 +5,7 @@
#include "check_err.hpp" #include "check_err.hpp"
#include "config.hpp" #include "config.hpp"
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "device.hpp" #include "device.hpp"
#include "device_tensor.hpp" #include "device_tensor.hpp"
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp" #include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
...@@ -137,40 +137,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha ...@@ -137,40 +137,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
int arg_idx = 5; int arg_idx = 5;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -202,21 +202,21 @@ int main(int argc, char* argv[]) ...@@ -202,21 +202,21 @@ int main(int argc, char* argv[])
params = parse_conv_params(num_dim_spatial, argc, argv); params = parse_conv_params(num_dim_spatial, argc, argv);
} }
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims), input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths), std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths)); std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K), std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims), filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths), std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths)); std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths(); const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K)}; static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims), output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths), std::begin(output_spatial_lengths),
std::end(output_spatial_lengths)); std::end(output_spatial_lengths));
...@@ -256,16 +256,16 @@ int main(int argc, char* argv[]) ...@@ -256,16 +256,16 @@ int main(int argc, char* argv[])
conv->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()), conv->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()),
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths, output_spatial_lengths,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
...@@ -280,13 +280,13 @@ int main(int argc, char* argv[]) ...@@ -280,13 +280,13 @@ int main(int argc, char* argv[])
float ave_time = invoker->Run(argument.get(), nrepeat); float ave_time = invoker->Run(argument.get(), nrepeat);
std::size_t flop = get_flops( std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths); params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = get_btype<InDataType, WeiDataType, OutDataType>( std::size_t num_btype = get_btype<InDataType, WeiDataType, OutDataType>(
params.N, params.N_,
params.C, params.C_,
params.K, params.K_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths); output_spatial_lengths);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time; float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
...@@ -302,10 +302,10 @@ int main(int argc, char* argv[]) ...@@ -302,10 +302,10 @@ int main(int argc, char* argv[])
auto ref_argument = ref_conv.MakeArgument(input, auto ref_argument = ref_conv.MakeArgument(input,
weights, weights,
host_output, host_output,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
......
example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
View file @ 712e464c
...@@ -5,7 +5,7 @@ ...@@ -5,7 +5,7 @@
#include "check_err.hpp" #include "check_err.hpp"
#include "config.hpp" #include "config.hpp"
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "device.hpp" #include "device.hpp"
#include "device_tensor.hpp" #include "device_tensor.hpp"
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp" #include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
...@@ -139,40 +139,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha ...@@ -139,40 +139,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
int arg_idx = 5; int arg_idx = 5;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -204,21 +204,21 @@ int main(int argc, char* argv[]) ...@@ -204,21 +204,21 @@ int main(int argc, char* argv[])
params = parse_conv_params(num_dim_spatial, argc, argv); params = parse_conv_params(num_dim_spatial, argc, argv);
} }
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims), input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths), std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths)); std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K), std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims), filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths), std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths)); std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths(); const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K)}; static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims), output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths), std::begin(output_spatial_lengths),
std::end(output_spatial_lengths)); std::end(output_spatial_lengths));
...@@ -258,16 +258,16 @@ int main(int argc, char* argv[]) ...@@ -258,16 +258,16 @@ int main(int argc, char* argv[])
conv->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()), conv->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()),
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths, output_spatial_lengths,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
...@@ -282,13 +282,13 @@ int main(int argc, char* argv[]) ...@@ -282,13 +282,13 @@ int main(int argc, char* argv[])
float ave_time = invoker->Run(argument.get(), nrepeat); float ave_time = invoker->Run(argument.get(), nrepeat);
std::size_t flop = get_flops( std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths); params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = get_btype<InDataType, WeiDataType, OutDataType>( std::size_t num_btype = get_btype<InDataType, WeiDataType, OutDataType>(
params.N, params.N_,
params.C, params.C_,
params.K, params.K_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths); output_spatial_lengths);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time; float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
...@@ -304,10 +304,10 @@ int main(int argc, char* argv[]) ...@@ -304,10 +304,10 @@ int main(int argc, char* argv[])
auto ref_argument = ref_conv.MakeArgument(input, auto ref_argument = ref_conv.MakeArgument(input,
weights, weights,
host_output, host_output,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
......
example/10_conv2d_bwd_data/CMakeLists.txt
View file @ 712e464c
add_example_executable(example_conv2d_bwd_data_xdl conv2d_bwd_data_xdl.cpp) add_example_executable(example_conv2d_bwd_data_xdl conv2d_bwd_data_xdl.cpp)
target_link_libraries(example_conv2d_bwd_data_xdl PRIVATE conv_fwd_util) target_link_libraries(example_conv2d_bwd_data_xdl PRIVATE conv_util)
example/11_conv2d_bwd_weight/CMakeLists.txt
View file @ 712e464c
add_example_executable(example_conv2d_bwd_weight_xdl conv2d_bwd_weight_xdl.cpp) add_example_executable(example_conv2d_bwd_weight_xdl conv2d_bwd_weight_xdl.cpp)
target_link_libraries(example_conv2d_bwd_weight_xdl PRIVATE conv_fwd_util) target_link_libraries(example_conv2d_bwd_weight_xdl PRIVATE conv_util)
example/17_convnd_bwd_data_xdl/CMakeLists.txt
View file @ 712e464c
add_example_executable(example_convnd_bwd_data_xdl convnd_bwd_data_xdl.cpp) add_example_executable(example_convnd_bwd_data_xdl convnd_bwd_data_xdl.cpp)
target_link_libraries(example_convnd_bwd_data_xdl PRIVATE conv_fwd_util) target_link_libraries(example_convnd_bwd_data_xdl PRIVATE conv_util)
example/17_convnd_bwd_data_xdl/convnd_bwd_data_xdl.cpp
View file @ 712e464c
...@@ -6,7 +6,7 @@ ...@@ -6,7 +6,7 @@
#include <half.hpp> #include <half.hpp>
#include "config.hpp" #include "config.hpp"
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "print.hpp" #include "print.hpp"
#include "device.hpp" #include "device.hpp"
#include "host_tensor.hpp" #include "host_tensor.hpp"
...@@ -105,40 +105,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, char* argv[]) ...@@ -105,40 +105,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, char* argv[])
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
int arg_idx = 5; int arg_idx = 5;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -171,7 +171,7 @@ int main(int argc, char* argv[]) ...@@ -171,7 +171,7 @@ int main(int argc, char* argv[])
int num_dim_spatial = 2; int num_dim_spatial = 2;
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
params.C = 128; params.C_ = 128;
if(argc == 4) if(argc == 4)
{ {
...@@ -202,21 +202,21 @@ int main(int argc, char* argv[]) ...@@ -202,21 +202,21 @@ int main(int argc, char* argv[])
exit(1); exit(1);
} }
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims), input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths), std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths)); std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K), std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C)}; static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims), filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths), std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths)); std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths(); const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N), std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K)}; static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims), output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths), std::begin(output_spatial_lengths),
std::end(output_spatial_lengths)); std::end(output_spatial_lengths));
...@@ -263,16 +263,16 @@ int main(int argc, char* argv[]) ...@@ -263,16 +263,16 @@ int main(int argc, char* argv[])
conv->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()), conv->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()),
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths, output_spatial_lengths,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
...@@ -287,13 +287,13 @@ int main(int argc, char* argv[]) ...@@ -287,13 +287,13 @@ int main(int argc, char* argv[])
float ave_time = invoker->Run(argument.get(), nrepeat); float ave_time = invoker->Run(argument.get(), nrepeat);
std::size_t flop = ck::utils::conv::get_flops( std::size_t flop = ck::utils::conv::get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths); params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = ck::utils::conv::get_btype<InDataType, WeiDataType, OutDataType>( std::size_t num_btype = ck::utils::conv::get_btype<InDataType, WeiDataType, OutDataType>(
params.N, params.N_,
params.C, params.C_,
params.K, params.K_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
output_spatial_lengths); output_spatial_lengths);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time; float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
...@@ -310,10 +310,10 @@ int main(int argc, char* argv[]) ...@@ -310,10 +310,10 @@ int main(int argc, char* argv[])
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi_host_result, auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi_host_result,
wei_k_c_y_x, wei_k_c_y_x,
out_n_k_ho_wo, out_n_k_ho_wo,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{});
......
include/ck/tensor_operation/gpu/device/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp
View file @ 712e464c
...@@ -4,7 +4,7 @@ ...@@ -4,7 +4,7 @@
#include <iostream> #include <iostream>
#include <memory> #include <memory>
#include <sstream> #include <sstream>
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "device.hpp" #include "device.hpp"
#include "device_conv_fwd.hpp" #include "device_conv_fwd.hpp"
#include "common_header.hpp" #include "common_header.hpp"
......
library/include/ck/library/utility/conv_fwd_util.hpp → library/include/ck/library/utility/conv_util.hpp
View file @ 712e464c
...@@ -146,19 +146,19 @@ struct ConvParams ...@@ -146,19 +146,19 @@ struct ConvParams
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);
ck::index_t num_dim_spatial; ck::index_t num_dim_spatial_;
ck::index_t N; ck::index_t N_;
ck::index_t K; ck::index_t K_;
ck::index_t C; ck::index_t C_;
std::vector<ck::index_t> filter_spatial_lengths; std::vector<ck::index_t> filter_spatial_lengths_;
std::vector<ck::index_t> input_spatial_lengths; std::vector<ck::index_t> input_spatial_lengths_;
std::vector<ck::index_t> conv_filter_strides; std::vector<ck::index_t> conv_filter_strides_;
std::vector<ck::index_t> conv_filter_dilations; std::vector<ck::index_t> conv_filter_dilations_;
std::vector<ck::index_t> input_left_pads; std::vector<ck::index_t> input_left_pads_;
std::vector<ck::index_t> input_right_pads; std::vector<ck::index_t> input_right_pads_;
std::vector<ck::index_t> GetOutputSpatialLengths() const; std::vector<ck::index_t> GetOutputSpatialLengths() const;
}; };
...@@ -268,10 +268,10 @@ void run_reference_convolution_forward(const ConvParams& params, ...@@ -268,10 +268,10 @@ void run_reference_convolution_forward(const ConvParams& params,
auto ref_argument = ref_conv.MakeArgument(input, auto ref_argument = ref_conv.MakeArgument(input,
weights, weights,
output, output,
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads, params.input_right_pads_,
PassThrough{}, PassThrough{},
PassThrough{}, PassThrough{},
PassThrough{}); PassThrough{});
...@@ -437,17 +437,17 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType, ...@@ -437,17 +437,17 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType,
virtual InTensorsTuple GetInputTensors() const override virtual InTensorsTuple GetInputTensors() const override
{ {
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params_.N), std::vector<std::size_t> input_dims{static_cast<std::size_t>(params_.N_),
static_cast<std::size_t>(params_.C)}; static_cast<std::size_t>(params_.C_)};
input_dims.insert(std::end(input_dims), input_dims.insert(std::end(input_dims),
std::begin(params_.input_spatial_lengths), std::begin(params_.input_spatial_lengths_),
std::end(params_.input_spatial_lengths)); std::end(params_.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params_.K), std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params_.K_),
static_cast<std::size_t>(params_.C)}; static_cast<std::size_t>(params_.C_)};
filter_dims.insert(std::end(filter_dims), filter_dims.insert(std::end(filter_dims),
std::begin(params_.filter_spatial_lengths), std::begin(params_.filter_spatial_lengths_),
std::end(params_.filter_spatial_lengths)); std::end(params_.filter_spatial_lengths_));
auto input = std::make_unique<Tensor<InDataType>>( auto input = std::make_unique<Tensor<InDataType>>(
get_host_tensor_descriptor(input_dims, InLayout{})); get_host_tensor_descriptor(input_dims, InLayout{}));
...@@ -465,8 +465,8 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType, ...@@ -465,8 +465,8 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType,
virtual TensorPtr<OutDataType> GetOutputTensor() const override virtual TensorPtr<OutDataType> GetOutputTensor() const override
{ {
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params_.N), std::vector<std::size_t> output_dims{static_cast<std::size_t>(params_.N_),
static_cast<std::size_t>(params_.K)}; static_cast<std::size_t>(params_.K_)};
output_dims.insert(std::end(output_dims), output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths_), std::begin(output_spatial_lengths_),
std::end(output_spatial_lengths_)); std::end(output_spatial_lengths_));
...@@ -522,16 +522,16 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType, ...@@ -522,16 +522,16 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType,
static_cast<InDataType*>(in_device_buffers[0]->GetDeviceBuffer()), static_cast<InDataType*>(in_device_buffers[0]->GetDeviceBuffer()),
static_cast<WeiDataType*>(in_device_buffers[1]->GetDeviceBuffer()), static_cast<WeiDataType*>(in_device_buffers[1]->GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buffer->GetDeviceBuffer()), static_cast<OutDataType*>(out_device_buffer->GetDeviceBuffer()),
params_.N, params_.N_,
params_.K, params_.K_,
params_.C, params_.C_,
params_.input_spatial_lengths, params_.input_spatial_lengths_,
params_.filter_spatial_lengths, params_.filter_spatial_lengths_,
output_spatial_lengths_, output_spatial_lengths_,
params_.conv_filter_strides, params_.conv_filter_strides_,
params_.conv_filter_dilations, params_.conv_filter_dilations_,
params_.input_left_pads, params_.input_left_pads_,
params_.input_right_pads, params_.input_right_pads_,
InElementwiseOp{}, InElementwiseOp{},
WeiElementwiseOp{}, WeiElementwiseOp{},
OutElementwiseOp{}); OutElementwiseOp{});
...@@ -539,20 +539,20 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType, ...@@ -539,20 +539,20 @@ class ConvFwdOpInstance : public ck::utils::OpInstance<OutDataType, InDataType,
virtual std::size_t GetFlops() const override virtual std::size_t GetFlops() const override
{ {
return get_flops(params_.N, return get_flops(params_.N_,
params_.C, params_.C_,
params_.K, params_.K_,
params_.filter_spatial_lengths, params_.filter_spatial_lengths_,
output_spatial_lengths_); output_spatial_lengths_);
} }
virtual std::size_t GetBtype() const override virtual std::size_t GetBtype() const override
{ {
return get_btype<InDataType, WeiDataType, OutDataType>(params_.N, return get_btype<InDataType, WeiDataType, OutDataType>(params_.N_,
params_.C, params_.C_,
params_.K, params_.K_,
params_.input_spatial_lengths, params_.input_spatial_lengths_,
params_.filter_spatial_lengths, params_.filter_spatial_lengths_,
output_spatial_lengths_); output_spatial_lengths_);
} }
......
library/src/utility/CMakeLists.txt
View file @ 712e464c
...@@ -8,14 +8,14 @@ include_directories(BEFORE ...@@ -8,14 +8,14 @@ include_directories(BEFORE
${PROJECT_SOURCE_DIR}/library/include/ck/library/utility ${PROJECT_SOURCE_DIR}/library/include/ck/library/utility
) )
set(CONV_FWD_UTIL_SOURCE set(CONV_UTIL_SOURCE
conv_fwd_util.cpp conv_util.cpp
) )
add_library(conv_fwd_util SHARED ${CONV_FWD_UTIL_SOURCE}) add_library(conv_util SHARED ${CONV_UTIL_SOURCE})
target_link_libraries(conv_fwd_util PRIVATE host_tensor) target_link_libraries(conv_util PRIVATE host_tensor)
target_compile_features(conv_fwd_util PUBLIC) target_compile_features(conv_util PUBLIC)
set_target_properties(conv_fwd_util PROPERTIES POSITION_INDEPENDENT_CODE ON) set_target_properties(conv_util PROPERTIES POSITION_INDEPENDENT_CODE ON)
target_include_directories(conv_fwd_util SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>) target_include_directories(conv_util SYSTEM PUBLIC $<BUILD_INTERFACE:${HALF_INCLUDE_DIR}>)
clang_tidy_check(conv_fwd_util) clang_tidy_check(conv_util)
library/src/utility/conv_fwd_util.cpp → library/src/utility/conv_util.cpp
View file @ 712e464c
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
namespace ck { namespace ck {
namespace utils { namespace utils {
...@@ -37,16 +37,16 @@ std::size_t get_flops(ck::index_t N, ...@@ -37,16 +37,16 @@ std::size_t get_flops(ck::index_t N,
} }
ConvParams::ConvParams() ConvParams::ConvParams()
: num_dim_spatial(2), : num_dim_spatial_(2),
N(128), N_(128),
K(256), K_(256),
C(192), C_(192),
filter_spatial_lengths(2, 3), filter_spatial_lengths_(2, 3),
input_spatial_lengths(2, 71), input_spatial_lengths_(2, 71),
conv_filter_strides(2, 2), conv_filter_strides_(2, 2),
conv_filter_dilations(2, 1), conv_filter_dilations_(2, 1),
input_left_pads(2, 1), input_left_pads_(2, 1),
input_right_pads(2, 1) input_right_pads_(2, 1)
{ {
} }
...@@ -60,23 +60,23 @@ ConvParams::ConvParams(ck::index_t n_dim, ...@@ -60,23 +60,23 @@ ConvParams::ConvParams(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(n_dim), : num_dim_spatial_(n_dim),
N(n_batch), N_(n_batch),
K(n_out_channels), K_(n_out_channels),
C(n_in_channels), C_(n_in_channels),
filter_spatial_lengths(filters_len), filter_spatial_lengths_(filters_len),
input_spatial_lengths(input_len), input_spatial_lengths_(input_len),
conv_filter_strides(strides), conv_filter_strides_(strides),
conv_filter_dilations(dilations), conv_filter_dilations_(dilations),
input_left_pads(left_pads), input_left_pads_(left_pads),
input_right_pads(right_pads) input_right_pads_(right_pads)
{ {
if(ck::type_convert<ck::index_t>(filter_spatial_lengths.size()) != num_dim_spatial || if(ck::type_convert<ck::index_t>(filter_spatial_lengths_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(input_spatial_lengths.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(input_spatial_lengths_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(conv_filter_strides.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(conv_filter_strides_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(conv_filter_dilations.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(conv_filter_dilations_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(input_left_pads.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(input_left_pads_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(input_right_pads.size()) != num_dim_spatial) ck::type_convert<ck::index_t>(input_right_pads_.size()) != num_dim_spatial_)
{ {
throw( throw(
std::runtime_error("ConvParams::GetOutputSpatialLengths: " std::runtime_error("ConvParams::GetOutputSpatialLengths: "
...@@ -86,27 +86,28 @@ ConvParams::ConvParams(ck::index_t n_dim, ...@@ -86,27 +86,28 @@ ConvParams::ConvParams(ck::index_t n_dim,
std::vector<ck::index_t> ConvParams::GetOutputSpatialLengths() const std::vector<ck::index_t> ConvParams::GetOutputSpatialLengths() const
{ {
if(ck::type_convert<ck::index_t>(filter_spatial_lengths.size()) != num_dim_spatial || if(ck::type_convert<ck::index_t>(filter_spatial_lengths_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(input_spatial_lengths.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(input_spatial_lengths_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(conv_filter_strides.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(conv_filter_strides_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(conv_filter_dilations.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(conv_filter_dilations_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(input_left_pads.size()) != num_dim_spatial || ck::type_convert<ck::index_t>(input_left_pads_.size()) != num_dim_spatial_ ||
ck::type_convert<ck::index_t>(input_right_pads.size()) != num_dim_spatial) ck::type_convert<ck::index_t>(input_right_pads_.size()) != num_dim_spatial_)
{ {
throw( throw(
std::runtime_error("ConvParams::GetOutputSpatialLengths: " std::runtime_error("ConvParams::GetOutputSpatialLengths: "
"parameter size is different from number of declared dimensions!")); "parameter size is different from number of declared dimensions!"));
} }
std::vector<ck::index_t> out_spatial_len(num_dim_spatial, 0); std::vector<ck::index_t> out_spatial_len(num_dim_spatial_, 0);
for(ck::index_t i = 0; i < num_dim_spatial; ++i) for(ck::index_t i = 0; i < num_dim_spatial_; ++i)
{ {
// 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 idx_eff = (filter_spatial_lengths[i] - 1) * conv_filter_dilations[i] + 1; const ck::index_t idx_eff =
(filter_spatial_lengths_[i] - 1) * conv_filter_dilations_[i] + 1;
out_spatial_len[i] = out_spatial_len[i] =
(input_spatial_lengths[i] + input_left_pads[i] + input_right_pads[i] - idx_eff) / (input_spatial_lengths_[i] + input_left_pads_[i] + input_right_pads_[i] - idx_eff) /
conv_filter_strides[i] + conv_filter_strides_[i] +
1; 1;
} }
return out_spatial_len; return out_spatial_len;
...@@ -116,40 +117,40 @@ ConvParams parse_conv_params(int num_dim_spatial, int arg_idx, char* const argv[ ...@@ -116,40 +117,40 @@ ConvParams parse_conv_params(int num_dim_spatial, int arg_idx, char* const argv[
{ {
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -228,12 +229,12 @@ HostTensorDescriptor get_input_host_tensor_descriptor(const std::vector<std::siz ...@@ -228,12 +229,12 @@ HostTensorDescriptor get_input_host_tensor_descriptor(const std::vector<std::siz
std::ostream& operator<<(std::ostream& os, const ck::utils::conv::ConvParams& p) std::ostream& operator<<(std::ostream& os, const ck::utils::conv::ConvParams& p)
{ {
os << "ConvParams {" os << "ConvParams {"
<< "\nnum_dim_spatial: " << p.num_dim_spatial << "\nN: " << p.N << "\nK: " << p.K << "\nnum_dim_spatial: " << p.num_dim_spatial_ << "\nN: " << p.N_ << "\nK: " << p.K_
<< "\nC: " << p.C << "\nfilter_spatial_lengths: " << p.filter_spatial_lengths << "\nC: " << p.C_ << "\nfilter_spatial_lengths: " << p.filter_spatial_lengths_
<< "\ninput_spatial_lengths: " << p.input_spatial_lengths << "\ninput_spatial_lengths: " << p.input_spatial_lengths_
<< "\nconv_filter_strides: " << p.conv_filter_strides << "\nconv_filter_strides: " << p.conv_filter_strides_
<< "\nconv_filter_dilations: " << p.conv_filter_dilations << "\nconv_filter_dilations: " << p.conv_filter_dilations_
<< "\ninput_left_pads: " << p.input_left_pads << "\ninput_left_pads: " << p.input_left_pads_
<< "\ninput_right_pads: " << p.input_right_pads; << "\ninput_right_pads: " << p.input_right_pads_;
return os; return os;
} }
profiler/CMakeLists.txt
View file @ 712e464c
...@@ -43,7 +43,7 @@ set(PROFILER_SOURCE ...@@ -43,7 +43,7 @@ set(PROFILER_SOURCE
add_executable(ckProfiler ${PROFILER_SOURCE}) add_executable(ckProfiler ${PROFILER_SOURCE})
target_link_libraries(ckProfiler PRIVATE host_tensor) target_link_libraries(ckProfiler PRIVATE host_tensor)
target_link_libraries(ckProfiler PRIVATE conv_fwd_util) target_link_libraries(ckProfiler PRIVATE conv_util)
target_link_libraries(ckProfiler PRIVATE device_gemm_reduce_instance) target_link_libraries(ckProfiler PRIVATE device_gemm_reduce_instance)
target_link_libraries(ckProfiler PRIVATE device_gemm_instance) target_link_libraries(ckProfiler PRIVATE device_gemm_instance)
target_link_libraries(ckProfiler PRIVATE device_gemm_bias2d_instance) target_link_libraries(ckProfiler PRIVATE device_gemm_bias2d_instance)
......
profiler/include/profile_convnd_bwd_data_impl.hpp
View file @ 712e464c
#pragma once #pragma once
#include "config.hpp" #include "config.hpp"
#include "device.hpp" #include "device.hpp"
#include "conv_fwd_util.hpp" #include "conv_util.hpp"
#include "host_tensor.hpp" #include "host_tensor.hpp"
#include "host_tensor_generator.hpp" #include "host_tensor_generator.hpp"
#include "tensor_layout.hpp" #include "tensor_layout.hpp"
......
profiler/src/profile_convnd_bwd_data.cpp
View file @ 712e464c
...@@ -39,40 +39,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, char* argv[], ...@@ -39,40 +39,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, char* argv[],
// (N, K, C) + num_dim_spatial * 6 (filter, input, strides, dilations, pad left, pad right) // (N, K, C) + num_dim_spatial * 6 (filter, input, strides, dilations, pad left, pad right)
ck::utils::conv::ConvParams params; ck::utils::conv::ConvParams params;
params.num_dim_spatial = num_dim_spatial; params.num_dim_spatial_ = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]); params.N_ = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]); params.K_ = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]); params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial); params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_spatial_lengths.resize(num_dim_spatial); params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]); params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_strides.resize(num_dim_spatial); params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]); params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
} }
params.conv_filter_dilations.resize(num_dim_spatial); params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]); params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_left_pads.resize(num_dim_spatial); params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_left_pads[i] = std::stoi(argv[arg_idx++]); params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
} }
params.input_right_pads.resize(num_dim_spatial); params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i) for(int i = 0; i < num_dim_spatial; ++i)
{ {
params.input_right_pads[i] = std::stoi(argv[arg_idx++]); params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
} }
return params; return params;
...@@ -133,16 +133,16 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -133,16 +133,16 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
init_method, init_method,
do_log, do_log,
nrepeat, nrepeat,
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
params.GetOutputSpatialLengths(), params.GetOutputSpatialLengths(),
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads); params.input_right_pads_);
break; break;
case 2: case 2:
...@@ -158,16 +158,16 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -158,16 +158,16 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
init_method, init_method,
do_log, do_log,
nrepeat, nrepeat,
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
params.GetOutputSpatialLengths(), params.GetOutputSpatialLengths(),
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads); params.input_right_pads_);
break; break;
case 3: case 3:
...@@ -183,16 +183,16 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -183,16 +183,16 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
init_method, init_method,
do_log, do_log,
nrepeat, nrepeat,
params.N, params.N_,
params.K, params.K_,
params.C, params.C_,
params.input_spatial_lengths, params.input_spatial_lengths_,
params.filter_spatial_lengths, params.filter_spatial_lengths_,
params.GetOutputSpatialLengths(), params.GetOutputSpatialLengths(),
params.conv_filter_strides, params.conv_filter_strides_,
params.conv_filter_dilations, params.conv_filter_dilations_,
params.input_left_pads, params.input_left_pads_,
params.input_right_pads); params.input_right_pads_);
break; break;
default: break; default: break;
......
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