cl_int CLFW::get(cl::Program::Sources &sources, std::vector<std::string> &files) {
sources.clear();
cl_int error = 0;
if (files.size() == 0) return CL_INVALID_VALUE;
for (int i = 0; i < files.size(); ++i) {
Print("adding " + files[i] + " to sources.", infoFG, infoBG);
long length = 0;
char *source = 0;
error|=loadFile(files[i].c_str(), &source, &length);
sources.push_back({ source, length });
}
return error;
}
cl_int CLFW::get(cl::Program::Sources &sources) {
sources.clear();
char* text = 0;
long temp;
loadFile("./opencl_sources.txt", &text, &temp);
char* file = strtok(text, "\n\r\n\0");
do {
Print("adding " + std::string(file) + " to sources.", infoFG, infoBG);
long length = 0;
char *source = 0;
loadFile(file, &source, &length);
sources.push_back({ source, length });
file = strtok(NULL, "\n\r\n\0");
} while (file != NULL);
return CL_SUCCESS;
}
int main(int argc, char *argv[])
{
cl_int err = CL_SUCCESS;
cl::Event evt;
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);
if (platforms.size() == 0) {
return false;
}
platform_ = platforms[0];
cl_context_properties properties[] =
{ CL_CONTEXT_PLATFORM, (cl_context_properties)(platforms[0])(), 0};
context_ = cl::Context(CL_DEVICE_TYPE_GPU, properties, NULL, NULL, &err);
CHECK_CL_ERROR(err, "cl::Context");
std::vector<cl::Device> devices = context_.getInfo<CL_CONTEXT_DEVICES>();
if (devices.size() == 0) {
return false;
}
device_ = devices[0];
sources_.push_back(std::make_pair(source_str.c_str(), source_str.size()));
program_ = cl::Program(context_, sources_);
err = program_.build(devices);
if (err != CL_SUCCESS) {
std::string log = program_.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[0]);
std::cout << "program.build() ERROR: " << log.c_str() << std::endl;
return false;
}
kernel_ = cl::Kernel(program_, "hello", &err);
CHECK_CL_ERROR(err, "cl::Kernel");
buf_ = cl::Buffer(context_, CL_MEM_READ_ONLY, 1024, NULL, &err);
queue_ = cl::CommandQueue(context_, device_, 0, &err);
CHECK_CL_ERROR(err, "cl::CommandQueue");
kernel_.setArg(0, buf_);
err = queue_.enqueueNDRangeKernel(kernel_, cl::NullRange, cl::NDRange(10, 10), cl::NullRange, NULL, &evt);
evt.wait();
CHECK_CL_ERROR(err, "queue.enqueueNDRangeKernel()");
return 0;
}
CL_helper(const char *kernelFileName, std::vector<std::string> kernelNames, bool loadBinary, bool writeBinary,
bool usePreProcArgs = false, std::map<std::string, std::string> preProcArgs = std::map<std::string, std::string>()){
cl::Platform::get(&platforms);
if(platforms.size()==0)
throw std::runtime_error("No OpenCL platforms found.\n");
int selectedPlatform=0;
platform=platforms.at(selectedPlatform);
platform.getDevices(CL_DEVICE_TYPE_ALL, &devices);
if(devices.size()==0)
throw std::runtime_error("No opencl devices found.\n");
int selectedDevice=0;
device=devices.at(selectedDevice);
context = cl::Context(devices);
try{
if(loadBinary){
std::string vendor=platform.getInfo<CL_PLATFORM_VENDOR>();
size_t found = vendor.find("NVIDIA");
if(found != std::string::npos){
FILE* fp;
fp = fopen("src/kernels/julia_filter.ptx", "r");
if (!fp) {
std::cerr << "Error loading kernel binary" << std::endl;
std::cerr << "Building kernel from .cl file" << std::endl;
loadBinary = false;
} else {
fseek(fp, 0, SEEK_END);
size_t kernel_sz = ftell(fp);
rewind(fp);
char* kernel_str = (char*)malloc(kernel_sz);
unsigned bytes = fread(kernel_str, 1, kernel_sz, fp);
fclose(fp);
binaries.push_back(std::make_pair((void*)kernel_str,kernel_sz+1));
program = cl::Program(context, devices, binaries);
program.build(devices);
}
} else{
std::cerr << "Vendor not NVIDIA, cannot load .ptx binary" << std::endl;
std::cerr << "Building kernel from .cl file" << std::endl;
loadBinary = false;
}
}
if(!loadBinary){
std::string kernelSource=CL_helper::LoadSource(kernelFileName);
sources.push_back(std::make_pair(kernelSource.c_str(), kernelSource.size()+1));
program = cl::Program(context, sources);
if(usePreProcArgs && !preProcArgs.empty()){
std::string preProcArgsString;
for(auto& arg : preProcArgs) {
preProcArgsString += "-D" + arg.first + "=" + arg.second + " ";
}
program.build(devices, preProcArgsString.c_str());
} else {
//std::string params = "-cl-unsafe-math-optimizations";
program.build(devices);
}
}
}catch (cl::Error er) {
for(unsigned i=0;i<devices.size();i++){
std::cerr <<"Log for device " << devices[i].getInfo<CL_DEVICE_NAME>().c_str()<<std::endl;
std::cerr << program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[i]).c_str() <<std::endl;
}
std::cerr << "ERROR:" << er.what() << " Code " << er.err()<<std::endl;
throw;
}
for(unsigned i=0; i<kernelNames.size();i++){
kernels.push_back( cl::Kernel(program, kernelNames.at(i).c_str()) );
}
queue = cl::CommandQueue(context, device);
if(writeBinary){
size_t bin_sz;
program.getInfo( CL_PROGRAM_BINARY_SIZES, &bin_sz);
unsigned char *bin = (unsigned char *)malloc(bin_sz);
program.getInfo(CL_PROGRAM_BINARIES, &bin);
FILE* fp = fopen("src/kernels/julia_filter.ptx", "wb");
fwrite(bin, sizeof(char), bin_sz, fp);
fclose(fp);
free(bin);
}
}
