public/feature-extraction-opencl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add banchmark image alignment

Browse Source
This commit is contained in:
gfuse
2021-08-23 15:30:49 +02:00
parent 30918ce750
commit 8df5e42a5b
5 changed files with 90 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CMakeLists.txt
+9 -4
View File
@@ -1,14 +1,19 @@
cmake_minimum_required(VERSION 3.7) cmake_minimum_required(VERSION 3.7)
project(Vio_OpenCL) project(Vio_OpenCL)
SET(CMAKE_BUILD_TYPE Release) # Release, RelWithDebInfo
SET(CMAKE_VERBOSE_MAKEFILE OFF)
SET(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${PROJECT_SOURCE_DIR}/CMakeModules/")
set(CMAKE_CXX_STANDARD 14) set(CMAKE_CXX_STANDARD 14)
find_package (OpenCL REQUIRED) find_package (OpenCL REQUIRED)
find_package(OpenCV REQUIRED) find_package(OpenCV REQUIRED)
FIND_PACKAGE(Eigen REQUIRED)
FIND_PACKAGE(Sophus REQUIRED)
include_directories (include ${OpenCL_INCLUDE_DIRS} ${OpenCV_INCLUDE_DIRS} kernel) include_directories (include ${OpenCL_INCLUDE_DIRS} ${OpenCV_INCLUDE_DIRS} kernel ${Eigen_INCLUDE_DIRS} ${Sophus_INCLUDE_DIRS})
add_executable(${PROJECT_NAME} main.cpp include/cl_class.h src/cl_class.cpp) add_executable(${PROJECT_NAME} main2.cpp include/cl_class.h src/cl_class.cpp src/sparse_img_align_gpu.cpp src/math_utils.cpp src/robust_cost.cpp)
target_compile_features(${PROJECT_NAME} PRIVATE cxx_range_for) target_compile_features(${PROJECT_NAME} PRIVATE cxx_range_for)
target_link_libraries(${PROJECT_NAME} ${OpenCL_LIBRARIES} ${OpenCV_LIBS}) target_link_libraries(${PROJECT_NAME} ${OpenCL_LIBRARIES} ${OpenCV_LIBS} ${Sophus_LIBRARIES})
ADD_DEFINITIONS(-DTEST_DATA_DIR=\"${PROJECT_SOURCE_DIR}/test/data\") ADD_DEFINITIONS(-DTEST_DATA_DIR=\"${PROJECT_SOURCE_DIR}/test/data\")
ADD_DEFINITIONS(-DKERNEL_DIR=\"${PROJECT_SOURCE_DIR}/kernel\")
include/cl_class.h
+40 -28
View File
@@ -4,40 +4,39 @@
#ifndef VIO_OPENCL_CL_CLASS_H #ifndef VIO_OPENCL_CL_CLASS_H
#define VIO_OPENCL_CL_CLASS_H #define VIO_OPENCL_CL_CLASS_H
#include "cl.hpp" #include <cl.hpp>
#include <iostream> #include <iostream>
#include <vector> #include <vector>
#include <algorithm> #include <algorithm>
#include <fstream> #include <fstream>
#include <exception> #include <exception>
#include <CL/opencl.h>
#include <opencv2/opencv.hpp> #include <opencv2/opencv.hpp>
class kernel{ class kernel{
public: public:
kernel(cl::Program* program,std::string name){_kernel = new cl::Kernel(*program,name.c_str());}; kernel(cl::Program* program,std::string name){_kernel = new cl::Kernel(*program,name.c_str());};
~kernel(){ ~kernel(){
for(auto&& i:_images)i.first.setDestructorCallback((void (*)(_cl_mem *, void *))notify, NULL);
for(auto&& i:_buffers)i.first.first.setDestructorCallback((void (*)(_cl_mem *, void *))notify, NULL);
_buffers.clear(); _buffers.clear();
_images.clear(); _images.clear();
}; };
template<typename T> template<typename T>
int32_t write(size_t id/*buffer ID*/,T* buf,cl::CommandQueue* queue,cl::Context* context,size_t buf_size){ int32_t write(size_t id/*buffer ID*/,T* buf,cl::CommandQueue* queue,cl::Context* context,size_t buf_size){
_buffers.push_back(std::pair<std::pair<cl::Buffer,size_t>,size_t>(std::pair<cl::Buffer,size_t>(cl::Buffer(*context, CL_MEM_READ_WRITE, sizeof(T) * buf_size), _buffers.push_back(std::pair<std::pair<cl::Buffer,size_t>,size_t>(std::pair<cl::Buffer,size_t>(cl::Buffer(*context, CL_MEM_READ_WRITE | CL_MEM_USE_HOST_PTR
, sizeof(T) * buf_size,buf),
buf_size),id)); buf_size),id));
cl_int err=queue->enqueueWriteBuffer(_buffers.back().first.first,CL_TRUE, 0, sizeof(T) * buf_size, buf); cl_int err=_kernel->setArg(id,_buffers.back().first.first);
if(err!=0){ if(err!=0){
std::cerr<<"Error C:Kernel, F:write, L:enqueueWriteBuffer\n"<<err<<'\n'; std::cerr<<"Error C:Kernel, F:write, L3:setArg\t"<<id<<"\t"<<err<<'\n';
return err;
}
err=_kernel->setArg(id,_buffers.back().first.first);
if(err!=0){
std::cerr<<"Error C:Kernel, F:write, L3:setArg\n"<<err<<'\n';
return err; return err;
} }
return CL_SUCCESS; return CL_SUCCESS;
}; };
int32_t write(size_t id/*buffer ID*/,cv::Mat& buf,cl::Context* context){ int32_t write(size_t id/*buffer ID*/,cv::Mat& buf,cl::Context* context){
_images.push_back(std::pair<cl::Image2D,size_t>(cl::Image2D(*context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, _images.push_back(std::pair<cl::Image2D,size_t>(cl::Image2D(*context, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR,
cl::ImageFormat(CL_R, CL_UNSIGNED_INT8), cl::ImageFormat(CL_R, CL_UNSIGNED_INT8),
buf.size().width, buf.size().width,
buf.size().height, buf.size().height,
@@ -46,7 +45,7 @@ public:
cl_int err=_kernel->setArg(id,_images.back().first); cl_int err=_kernel->setArg(id,_images.back().first);
if(err!=0){ if(err!=0){
std::cerr<<"Error C:Kernel, F:write image, L:setArg\n"<<err<<'\n'; std::cerr<<"Error C:Kernel, F:write image, L:setArg ID:\t"<<id<<"\t"<<err<<'\n';
return err; return err;
} }
return CL_SUCCESS; return CL_SUCCESS;
@@ -54,25 +53,29 @@ public:
template<typename T> template<typename T>
int32_t reload(size_t id,T* buf,cl::CommandQueue* queue){ int32_t reload(size_t id,T* buf,cl::CommandQueue* queue){
try{ for(auto&& i:_buffers)if(i.second==id){
if(id>_buffers.size())return CL_MAP_FAILURE; T* Map_buf=(T*)queue->enqueueMapBuffer(i.first.first,CL_TRUE,CL_MAP_WRITE,0,sizeof(T) * i.first.second);
for(auto i:_buffers)if(i.second==id) memcpy(Map_buf,buf,sizeof(T) * i.first.second);
queue->enqueueWriteBuffer(i.first.first,CL_TRUE, 0, sizeof(T) * i.first.second, buf); if(queue->enqueueUnmapMemObject(i.first.first,Map_buf)!=CL_SUCCESS)std::cerr<<"Reload buffer on GPU failed"<<'\n';
return CL_SUCCESS;
}catch (std::exception& err){
std::cerr<<"Reload buffer on GPU failed\n"<<err.what()<<'\n';
return CL_MAP_FAILURE;
} }
return CL_SUCCESS;
}; };
int32_t reload(size_t id,cv::Mat& buf,cl::Context* context){ int32_t reload(size_t id,cv::Mat& buf,cl::Context* context){
try{ try{
for(auto i:_images)if(i.second==id) for(auto&& i:_images)if(i.second==id){
i.first=cl::Image2D(*context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, i.first.setDestructorCallback((void (*)(_cl_mem *, void *))notify, NULL);
i.first=cl::Image2D(*context, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR,
cl::ImageFormat(CL_R, CL_UNSIGNED_INT8), cl::ImageFormat(CL_R, CL_UNSIGNED_INT8),
buf.size().width, buf.size().width,
buf.size().height, buf.size().height,
0, 0,
reinterpret_cast<uchar*>(buf.data)); reinterpret_cast<uchar*>(buf.data));
cl_int err=_kernel->setArg(id,i.first);
if(err!=0){
std::cerr<<"Error C:Kernel, F:write image, L:setArg ID:\t"<<id<<"\t"<<err<<'\n';
return err;
}
}
return CL_SUCCESS; return CL_SUCCESS;
}catch (std::exception& err){ }catch (std::exception& err){
std::cerr<<"Reload buffer on GPU failed\n"<<err.what()<<'\n'; std::cerr<<"Reload buffer on GPU failed\n"<<err.what()<<'\n';
@@ -87,6 +90,9 @@ public:
for(auto i:_images)if(i.second==id)return i.first; for(auto i:_images)if(i.second==id)return i.first;
} }
private: private:
static void notify(cl_mem *, void * user_data) {
//std::cerr << "Memory object was deleted." << std::endl;
}
std::vector<std::pair<std::pair<cl::Buffer,size_t>,size_t>> _buffers; std::vector<std::pair<std::pair<cl::Buffer,size_t>,size_t>> _buffers;
std::vector<std::pair<cl::Image2D,size_t>> _images; std::vector<std::pair<cl::Image2D,size_t>> _images;
@@ -126,7 +132,7 @@ public:
_kernels.at(id1).write(id2,buf,context); _kernels.at(id1).write(id2,buf,context);
return CL_SUCCESS; return CL_SUCCESS;
}catch (std::exception& err){ }catch (std::exception& err){
std::cerr<<"Write buffer on GPU failed\n"<<err.what()<<'\n'; std::cerr<<"Write image on GPU failed\n"<<err.what()<<'\n';
return CL_MAP_FAILURE; return CL_MAP_FAILURE;
} }
} }
@@ -155,6 +161,7 @@ public:
} }
int32_t run(size_t id1/*kernal ID*/,std::size_t x=1,std::size_t y=1,std::size_t z=1) { int32_t run(size_t id1/*kernal ID*/,std::size_t x=1,std::size_t y=1,std::size_t z=1) {
cl_int err=0; cl_int err=0;
queue->flush();
if(z>1 && y>1){ if(z>1 && y>1){
err=queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x,y,z)/*Global*/, cl::NullRange/*local*/); err=queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x,y,z)/*Global*/, cl::NullRange/*local*/);
}else if(z<2 && y>1){ }else if(z<2 && y>1){
@@ -162,27 +169,32 @@ public:
}else{ }else{
err=queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x)/*Global*/, cl::NullRange/*local*/); err=queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x)/*Global*/, cl::NullRange/*local*/);
}; };
queue->finish();
if(err!=0){ if(err!=0){
std::cerr<<"kernel: "<<_kernels.at(id1)._kernel->getInfo<CL_KERNEL_FUNCTION_NAME>(); std::cerr<<"kernel: "<<_kernels.at(id1)._kernel->getInfo<CL_KERNEL_FUNCTION_NAME>();
std::cerr<<" ,Args: "<<_kernels.at(id1)._kernel->getInfo<CL_KERNEL_NUM_ARGS>()<<" ,Run out: "<<err<<'\n'; std::cerr<<" ,Args: "<<_kernels.at(id1)._kernel->getInfo<CL_KERNEL_NUM_ARGS>()<<" ,Run out: "<<err<<'\n';
for(int i=0;i<_kernels.at(id1)._kernel->getInfo<CL_KERNEL_NUM_ARGS>();++i) for(int i=0;i<_kernels.at(id1)._kernel->getInfo<CL_KERNEL_NUM_ARGS>();++i)
std::cerr<<"argument type "<<i<<":"<<_kernels.at(id1)._kernel->getArgInfo<CL_KERNEL_ARG_TYPE_NAME>(i)<<'\n'; std::cerr<<"argument type "<<i<<":"<<_kernels.at(id1)._kernel->getArgInfo<CL_KERNEL_ARG_TYPE_NAME>(i)<<'\n';
} }
return err; return queue->finish();
} }
template<typename T> template<typename T>
void read(size_t id1/*kernal ID*/,size_t id2/*buffer ID*/, T* out){ void read(size_t id1/*kernal ID*/,size_t id2/*buffer ID*/, T* out){
queue->enqueueReadBuffer(_kernels.at(id1).read(id2).first, CL_TRUE, 0, sizeof(T) * _kernels.at(id1).read(id2).second, out); queue->enqueueReadBuffer(_kernels.at(id1).read(id2).first, CL_TRUE, 0, sizeof(T) * _kernels.at(id1).read(id2).second, out);
} }
template<typename T>
void read(size_t id1/*kernal ID*/,size_t id2/*buffer ID*/,size_t size/*size*/, T* out){
T* Map_buf=(T*)queue->enqueueMapBuffer(_kernels.at(id1).read(id2).first,CL_TRUE,CL_MAP_READ,0,sizeof(T) * size);
memcpy(out,Map_buf,sizeof(T) * size);
if(queue->enqueueUnmapMemObject(_kernels.at(id1).read(id2).first,Map_buf)!=CL_SUCCESS)std::cerr<<"Read buffer on GPU failed"<<'\n';
}
void clear_buf(); void clear_buf();
private: private:
std::vector<kernel> _kernels; std::vector<kernel> _kernels;
cl::Context* context = NULL; cl::Context* context = nullptr;
cl::Device device; cl::Device* device = nullptr;
cl::Program* program = NULL; cl::Program* program = nullptr;
cl::CommandQueue* queue=NULL; cl::CommandQueue* queue= nullptr;
}; };
#endif //VIO_OPENCL_CL_CLASS_H #endif //VIO_OPENCL_CL_CLASS_H
main.cpp
+1 -1
View File
@@ -72,7 +72,7 @@ int main() {
gpu.run(0,752,480,1); gpu.run(0,752,480,1);
time_accumulator += ((cv::getTickCount() - t) / cv::getTickFrequency()); time_accumulator += ((cv::getTickCount() - t) / cv::getTickFrequency());
} }
printf("OpenCL took %f s (average over %d trials).\n", time_accumulator, 1000 ); printf("OpenCL took %f ms (average over %d trials).\n", time_accumulator/((double)1000)*1000.0, 1000 );
int icorner_[1]={0}; int icorner_[1]={0};
gpu.read(0,2,icorner_); gpu.read(0,2,icorner_);
cl_int2 corners[740*450]; cl_int2 corners[740*450];
readme.md
+14 -9
View File
@@ -1,11 +1,16 @@
sudo apt-get install -y gpg-agent wget sudo apt-get install -y gpg-agent wge
wget -qO - https://repositories.intel.com/graphics/intel-graphics.key | wget -qO - https://repositories.intel.com/graphics/intel-graphics.key | sudo apt-key add -
sudo apt-key add - sudo apt-add-repository 'deb [arch=amd64] https://repositories.intel.com/graphics/ubuntu bionic main'
sudo apt-add-repository \
'deb [arch=amd64] https://repositories.intel.com/graphics/ubuntu bionic main'
sudo apt-get update
sudo apt install ocl-icd-* opencl-headers intel-level-zero-gpu level-zero intel-opencl intel-igc-opencl-devel level-zero-devel
CodeBlocks build for installing apt-add-repository :
cmake .. -G"CodeBlocks - Unix Makefiles" sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys F42ED6FBAB17C654
sudo apt-get update
sudo apt-get install software-properties-common
sudo apt-get update
sudo apt install ocl-icd-* opencl-headers intel-level-zero-gpu level-zero intel-opencl intel-igc-opencl-devel level-zero-devel
CodeBlocks build
cmake .. -G"CodeBlocks - Unix Makefiles"
src/cl_class.cpp
+26 -17
View File
@@ -1,7 +1,7 @@
// //
// Created by root on 4/27/21. // Created by root on 4/27/21.
// //
#include "cl_class.h" #include <cl_class.h>
opencl::opencl() { opencl::opencl() {
std::vector<cl::Platform> all_platforms; std::vector<cl::Platform> all_platforms;
cl::Platform::get(&all_platforms); cl::Platform::get(&all_platforms);
@@ -12,7 +12,7 @@ opencl::opencl() {
//get default device of the default platform //get default device of the default platform
std::vector<cl::Device> all_devices; std::vector<cl::Device> all_devices;
for(auto i:all_platforms){ for(auto i:all_platforms){
std::cout << "Find platform: " << i.getInfo<CL_PLATFORM_NAME>() << "\n"; std::cout << "Find platform number:"<< i.getInfo<CL_PLATFORM_NAME>() << "\n";
i.getDevices(CL_DEVICE_TYPE_GPU, &all_devices); i.getDevices(CL_DEVICE_TYPE_GPU, &all_devices);
} }
@@ -20,29 +20,38 @@ opencl::opencl() {
std::cout << " No devices found. Check OpenCL installation!\n"; std::cout << " No devices found. Check OpenCL installation!\n";
exit(1); exit(1);
} }
device = all_devices[0]; device = new cl::Device(all_devices[0]);
std::cout << "CL_DEVICE_NAME: " << device.getInfo<CL_DEVICE_NAME>() <<'\n' std::cout << "CL_DEVICE_NAME: " << device->getInfo<CL_DEVICE_NAME>() <<'\n'
<< "CL_DEVICE_OPENCL_C_VERSION: " <<device.getInfo<CL_DEVICE_OPENCL_C_VERSION>()<<'\n' << "CL_DEVICE_OPENCL_C_VERSION: " <<device->getInfo<CL_DEVICE_OPENCL_C_VERSION>()<<'\n'
<< "CL_DEVICE_BUILT_IN_KERNELS: " <<device.getInfo<CL_DEVICE_BUILT_IN_KERNELS>()<<'\n' << "CL_DEVICE_BUILT_IN_KERNELS: " <<device->getInfo<CL_DEVICE_BUILT_IN_KERNELS>()<<'\n'
<< "CL_DEVICE_COMPILER_AVAILABLE: " <<device.getInfo<CL_DEVICE_COMPILER_AVAILABLE>()<<'\n' << "CL_DEVICE_COMPILER_AVAILABLE: " <<device->getInfo<CL_DEVICE_COMPILER_AVAILABLE>()<<'\n'
<< "CL_DEVICE_LOCAL_MEM_SIZE: " <<device.getInfo<CL_DEVICE_LOCAL_MEM_SIZE>()<<'\n' << "CL_DEVICE_LOCAL_MEM_SIZE: " <<device->getInfo<CL_DEVICE_LOCAL_MEM_SIZE>()<<'\n'
<< "CL_DEVICE_GLOBAL_MEM_SIZE: " <<device.getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>()<<'\n'; << "CL_DEVICE_GLOBAL_MEM_SIZE: " <<device->getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>()<<'\n';
context=new cl::Context({ *device });
context=new cl::Context({ device });
cl::Program::Sources sources; cl::Program::Sources sources;
read_cl fast("../kernel/fast-gray.cl"); read_cl fast(std::string(KERNEL_DIR)+"/fast-gray.cl");
sources.push_back({ fast.src_str, fast.size }); sources.push_back({ fast.src_str, fast.size });
read_cl compute_residual(std::string(KERNEL_DIR)+"/compute-residual.cl");
sources.push_back({ compute_residual.src_str, compute_residual.size });
program=new cl::Program(*context, sources); program=new cl::Program(*context, sources);
if(program->build({ device },"-DFAST_THRESH=40") !=0) //296.724480,395.357280,317.877600,238.752640,1.016870
std::cout << " Error building: " << program->getBuildInfo<CL_PROGRAM_BUILD_STATUS>(device)<<'\n' std::string options="-DFAST_THRESH=40 -DPATCH_SIZE=2 -DPATCH_HALFSIZE=1 -DF_X="+ std::to_string((float)296.724480) +
<<program->getBuildInfo<CL_PROGRAM_BUILD_LOG>(device) << '\n'; " -DF_Y="+std::to_string((float)395.357280)+
queue=new cl::CommandQueue(*context,device,0,NULL); " -DC_X="+std::to_string((float)317.877600)+
" -DC_Y="+std::to_string((float)238.752640)+
" -DS="+std::to_string((float)1.016870);
if(program->build({ *device },options.c_str()) !=0)
std::cout << " Error building: " << program->getBuildInfo<CL_PROGRAM_BUILD_STATUS>(*device)<<'\n'
<<program->getBuildInfo<CL_PROGRAM_BUILD_LOG>(*device) << '\n';
queue=new cl::CommandQueue(*context,*device,CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE,NULL);
} }
opencl::~opencl() { opencl::~opencl() {
delete context;
delete device;
} }
void opencl::clear_buf() { void opencl::clear_buf() {
queue->finish(); queue->finish();
queue->flush();
_kernels.clear(); _kernels.clear();
delete program;
} }