public/vio
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add fast kernel

Browse Source
This commit is contained in:
gfuse
2021-05-05 19:12:01 +02:00
parent 5687fe5057
commit 5a693259d6
6 changed files with 85 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files
GPU_version/svo/include/svo/cl_class.h
+42 -50
View File
@@ -10,7 +10,6 @@
#include <algorithm>
#include <fstream>
#include <exception>
#define __CL_ENABLE_EXCEPTIONS
#include <CL/opencl.h>
#include <opencv2/opencv.hpp>
@@ -18,45 +17,40 @@ class kernel{
public:
kernel(cl::Program* program,std::string name){_kernel = new cl::Kernel(*program,name.c_str());};
~kernel(){
//for(auto i:_buffers)i.first.first.;
_buffers.clear();
_images.clear();
};
template<typename T>
int32_t write(size_t id/*buffer ID*/,T* buf,cl::CommandQueue* queue,cl::Context* context,size_t buf_size){
try{
_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, sizeof(T) * buf_size),
buf_size),id));
queue->enqueueWriteBuffer(_buffers.back().first.first,CL_TRUE, 0, sizeof(T) * buf_size, buf);
_kernel->setArg(id,_buffers.back().first.first);
return CL_SUCCESS;
}catch (std::exception& err){
std::cerr<<"Error C:Kernel, F:write\n"<<err.what()<<'\n';
return CL_MAP_FAILURE;
cl_int err=queue->enqueueWriteBuffer(_buffers.back().first.first,CL_TRUE, 0, sizeof(T) * buf_size, buf);
if(err!=0){
std::cerr<<"Error C:Kernel, F:write, L:enqueueWriteBuffer\n"<<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 CL_SUCCESS;
};
int32_t write(size_t id/*buffer ID*/,cv::Mat& buf,cl::Context* context){
try{
//if(id==0){
_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_COPY_HOST_PTR,
cl::ImageFormat(CL_R, CL_UNSIGNED_INT8),
buf.size().width,
buf.size().height,
0,
reinterpret_cast<uchar*>(buf.data)),id));
/* }else{
_images.push_back(std::pair<cl::Image2D,size_t>(cl::Image2D(*context, CL_MEM_READ_WRITE,
cl::ImageFormat(CL_RGBA, CL_UNORM_INT8),
buf.size().width,
buf.size().height),id));
}*/
_kernel->setArg(id,_images.back().first);
return CL_SUCCESS;
}catch (std::exception& err){
std::cerr<<"Error C:Kernel, F:write image\n"<<err.what()<<'\n';
return CL_MAP_FAILURE;
};
cl_int err=_kernel->setArg(id,_images.back().first);
if(err!=0){
std::cerr<<"Error C:Kernel, F:write image, L:setArg\n"<<err<<'\n';
return err;
}
return CL_SUCCESS;
};
template<typename T>
@@ -87,6 +81,12 @@ public:
}
};
cl::Kernel* _kernel = NULL;
std::pair<cl::Buffer,size_t> read(size_t id){
for(auto i:_buffers)if(i.second==id)return i.first;
}
cl::Image2D read_img(size_t id){
for(auto i:_images)if(i.second==id)return i.first;
}
private:
std::vector<std::pair<std::pair<cl::Buffer,size_t>,size_t>> _buffers;
std::vector<std::pair<cl::Image2D,size_t>> _images;
@@ -132,7 +132,7 @@ public:
}
}
template<typename T>
int32_t write_buf(size_t id1/*kernal ID*/,size_t id2/*buffer ID*/,T& value){
int32_t write_buf(size_t id1/*kernal ID*/,size_t id2/*buffer ID*/,T&& value){
_kernels.at(id1)._kernel->setArg(id2,value);
};
template<typename T>
@@ -154,37 +154,29 @@ public:
return CL_MAP_FAILURE;
}
}
template<typename T>
int32_t run(size_t id1/*kernal ID*/,std::size_t x=1,std::size_t y=1,std::size_t z=1,size_t id2=0/*last id of the kernel related to the output*/,
size_t size_out=0/*out put buffer size*/,
T* out= nullptr) {
try{
//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: ";
cl::Buffer out_buf(*context, CL_MEM_READ_WRITE, sizeof(T) * size_out);
_kernels.at(id1)._kernel->setArg(id2,out_buf);
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;
if(z>1 && y>1){
std::cerr<<queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x,y,z)/*Global*/, cl::NullRange/*local*/)<<'\n';
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){
queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x,y)/*Global*/, cl::NullRange/*local*/);
//std::cerr<<queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x,y)/*Global*/, cl::NullRange/*local*/)<<'\n';
/*std::cerr<<_kernels.at(id1)._kernel->getArgInfo<CL_KERNEL_ARG_TYPE_NAME>(0)<<'\n';
std::cerr<<_kernels.at(id1)._kernel->getArgInfo<CL_KERNEL_ARG_TYPE_NAME>(1)<<'\n';
std::cerr<<_kernels.at(id1)._kernel->getArgInfo<CL_KERNEL_ARG_TYPE_NAME>(2)<<'\n';
std::cerr<<_kernels.at(id1)._kernel->getArgInfo<CL_KERNEL_ARG_TYPE_NAME>(3)<<'\n';*/
err=queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x,y)/*Global*/, cl::NullRange/*local*/);
}else{
std::cerr<<queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x)/*Global*/, cl::NullRange/*local*/)<<'\n';
err=queue->enqueueNDRangeKernel(*_kernels.at(id1)._kernel, cl::NullRange/*offset*/, cl::NDRange(x)/*Global*/, cl::NullRange/*local*/);
};
queue->enqueueReadBuffer(out_buf, CL_TRUE, 0, sizeof(T) * size_out, out);
//wait for kernel to finish
queue->finish();
return CL_SUCCESS;
} catch (std::exception& err) {
std::cerr<<"RUN kernel Error\n"<<err.what()<<'\n';
return CL_EXEC_STATUS_ERROR_FOR_EVENTS_IN_WAIT_LIST;
}
if(err!=0){
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';
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';
}
return err;
}
template<typename T>
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);
}
void clear_buf();
private:
std::vector<kernel> _kernels;
GPU_version/svo/include/svo/feature_detection.h
+6 -1
View File
@@ -19,6 +19,7 @@
#include <svo/global.h>
#include <svo/frame.h>
#include <svo/cl_class.h>
namespace svo {
@@ -49,7 +50,10 @@ public:
const int cell_size,
const int n_pyr_levels);
virtual ~AbstractDetector() {};
virtual ~AbstractDetector() {
gpu_fast_->clear_buf();
gpu_fast_->make_kernel("fast_gray");
};
virtual void detect(
Frame* frame,
@@ -62,6 +66,7 @@ public:
/// Set grid cells of existing features as occupied
void setExistingFeatures(const Features& fts);
opencl* gpu_fast_= nullptr;
protected: