kernel_t<COI>* kernel_t<COI>::buildFromBinary(const std::string &filename,
const std::string &functionName_){
OCCA_EXTRACT_DATA(COI, Kernel);
functionName = functionName_;
std::string libPath, soname;
getFilePrefixAndName(filename, libPath, soname);
for(int i = 0; i < soname.size(); ++i){
if(soname[i] == '.'){
soname = soname.substr(0, i);
break;
}
}
COILIBRARY outLibrary;
OCCA_COI_CHECK("Kernel: Loading Kernel To Chief",
COIProcessLoadLibraryFromFile(data_.chiefID,
filename.c_str(),
soname.c_str(),
NULL,
&outLibrary));
const char *c_functionName = functionName.c_str();
OCCA_COI_CHECK("Kernel: Getting Handle",
COIProcessGetFunctionHandles(data_.chiefID,
1,
&c_functionName,
&(data_.kernel)));
return this;
}
void init(const char* cfg)
{
/* get number of Xeon Phi devices */
uint32_t engines = 0;
COIEngineGetCount( COI_ISA_MIC, &engines );
if (engines == 0) throw std::runtime_error("No Xeon Phi device found.");
/* get engine handle */
COIRESULT result;
result = COIEngineGetHandle( COI_ISA_MIC, 0, &engine );
if (result != COI_SUCCESS)
throw std::runtime_error("Failed to load engine number " + std::stringOf(0) + ": " + COIResultGetName(result));
/* print info of engine */
COI_ENGINE_INFO info;
result = COIEngineGetInfo(engine,sizeof(info),&info);
std::cout << "Found Xeon Phi device with " << info.NumCores << " cores and " << (info.PhysicalMemory/1024/1024) << "MB memory" << std::endl;
/* create process */
const std::string executable = std::string(tutorialName)+"_xeonphi_device";
result = COIProcessCreateFromFile
(engine,
executable.c_str(), // The local path to the sink side binary to launch.
0, NULL, // argc and argv for the sink process.
false, NULL, // Environment variables to set for the sink process.
true, NULL, // Enable the proxy but don't specify a proxy root path.
0, // The amount of memory to reserve for COIBuffers.
NULL, // Path to search for dependencies
&process // The resulting process handle.
);
if (result != COI_SUCCESS)
throw std::runtime_error("Failed to create process " + std::string(executable) +": " + COIResultGetName(result));
/* create pipeline */
COI_CPU_MASK cpuMask;
result = COIPipelineClearCPUMask(&cpuMask);
if (result != COI_SUCCESS)
throw std::runtime_error(std::string("COIPipelineClearCPUMask failed: ") + COIResultGetName(result));
result = COIPipelineSetCPUMask(process,info.NumCores-1,0,&cpuMask);
result = COIPipelineSetCPUMask(process,info.NumCores-1,1,&cpuMask);
result = COIPipelineSetCPUMask(process,info.NumCores-1,2,&cpuMask);
result = COIPipelineSetCPUMask(process,info.NumCores-1,3,&cpuMask);
if (result != COI_SUCCESS)
throw std::runtime_error(std::string("COIPipelineSetCPUMask failed: ") + COIResultGetName(result));
result = COIPipelineCreate(process,cpuMask,0,&pipeline);
if (result != COI_SUCCESS)
throw std::runtime_error(std::string("COIPipelineCreate failed: ") + COIResultGetName(result));
/* get run functions */
const char *fctNameArray[8] = { "run_init", "run_key_pressed", "run_create_mesh", "run_create_hairset", "run_create_scene", "run_pick", "run_render", "run_cleanup" };
result = COIProcessGetFunctionHandles (process, 8, fctNameArray, &runInit);
if (result != COI_SUCCESS)
throw std::runtime_error("COIProcessGetFunctionHandles failed: "+std::string(COIResultGetName(result)));
/* run init runfunction */
InitData parms;
strncpy(parms.cfg,cfg,sizeof(parms.cfg));
result = COIPipelineRunFunction (pipeline, runInit, 0, NULL, NULL, 0, NULL, &parms, sizeof(parms), NULL, 0, NULL);
if (result != COI_SUCCESS)
throw std::runtime_error("COIPipelineRunFunction failed: "+std::string(COIResultGetName(result)));
}
kernel_t<COI>* kernel_t<COI>::buildFromSource(const std::string &filename,
const std::string &functionName_,
const kernelInfo &info_){
functionName = functionName_;
kernelInfo info = info_;
info.addDefine("OCCA_USING_CPU", 1);
info.addDefine("OCCA_USING_COI", 1);
info.addOCCAKeywords(occaCOIDefines);
std::stringstream salt;
salt << "COI"
<< info.salt()
<< dev->dHandle->compilerEnvScript
<< dev->dHandle->compiler
<< dev->dHandle->compilerFlags
<< functionName;
std::string cachedBinary = getCachedName(filename, salt.str());
std::string libPath, soname;
getFilePrefixAndName(cachedBinary, libPath, soname);
std::string libName = "lib" + soname + ".so";
cachedBinary = libPath + libName;
struct stat buffer;
bool fileExists = (stat(cachedBinary.c_str(), &buffer) == 0);
if(fileExists){
std::cout << "Found cached binary of [" << filename << "] in [" << cachedBinary << "]\n";
return buildFromBinary(cachedBinary, functionName);
}
if(!haveFile(cachedBinary)){
waitForFile(cachedBinary);
return buildFromBinary(cachedBinary, functionName);
}
std::string iCachedBinary = createIntermediateSource(filename,
cachedBinary,
info);
std::stringstream command;
if(dev->dHandle->compilerEnvScript.size())
command << dev->dHandle->compilerEnvScript << " && ";
command << dev->dHandle->compiler
#if (OCCA_OS == LINUX_OS) || (OCCA_OS == OSX_OS)
<< " -x c++ -w -nodefaultlibs -fPIC -shared"
#else
<< " /TP /LD /D MC_CL_EXE"
#endif
<< ' ' << dev->dHandle->compilerFlags
<< ' ' << info.flags
<< ' ' << iCachedBinary
#if (OCCA_OS == LINUX_OS) || (OCCA_OS == OSX_OS)
<< " -o " << cachedBinary
#else
<< " /link /OUT:" << cachedBinary
#endif
<< std::endl;
const std::string &sCommand = command.str();
std::cout << "Compiling [" << functionName << "]\n" << sCommand << "\n";
const int compileError = system(sCommand.c_str());
if(compileError){
releaseFile(cachedBinary);
throw 1;
}
OCCA_EXTRACT_DATA(COI, Kernel);
COILIBRARY outLibrary;
const COIRESULT loadingLibraryResult = COIProcessLoadLibraryFromFile(data_.chiefID,
cachedBinary.c_str(),
soname.c_str(),
NULL,
&outLibrary);
if(errorCode != COI_SUCCESS)
releaseFile(cachedBinary);
OCCA_COI_CHECK("Kernel: Loading Kernel To Chief", loadingLibraryResult);
const char *c_functionName = functionName.c_str();
const COIRESULT getFunctionHandleResult = COIProcessGetFunctionHandles(data_.chiefID,
1,
&c_functionName,
&(data_.kernel));
if(errorCode != COI_SUCCESS)
releaseFile(cachedBinary);
OCCA_COI_CHECK("Kernel: Getting Handle", getFunctionHandleResult);
releaseFile(cachedBinary);
return this;
}
void device_t<COI>::setup(const int device, const int memoryAllocated){
data = new COIDeviceData_t;
OCCA_EXTRACT_DATA(COI, Device);
uint32_t deviceCount;
OCCA_COI_CHECK("Device: Get Count",
COIEngineGetCount(COI_ISA_MIC, &deviceCount));
OCCA_CHECK(device < deviceCount);
OCCA_COI_CHECK("Device: Get Handle",
COIEngineGetHandle(COI_ISA_MIC, device, &data_.deviceID) );
std::stringstream salt;
salt << "COI"
<< occaCOIMain;
std::string cachedBinary = getCachedName("occaCOIMain", salt.str());
struct stat buffer;
bool fileExists = (stat(cachedBinary.c_str(), &buffer) == 0);
if(fileExists)
std::cout << "Found cached binary of [occaCOIMain] in [" << cachedBinary << "]\n";
else{
//---[ Write File ]-----------------
std::string prefix, name;
getFilePrefixAndName(cachedBinary, prefix, name);
const std::string iCachedBinary = prefix + "i_" + name;
if(haveFile(cachedBinary)){
std::cout << "Making [" << iCachedBinary << "]\n";
std::ofstream fs;
fs.open(iCachedBinary.c_str());
fs << occaCOIMain;
fs.close();
std::stringstream command;
command << dev->dHandle->compiler
<< " -o " << cachedBinary
<< " -x c++"
<< ' ' << dev->dHandle->compilerFlags
<< ' ' << iCachedBinary;
const std::string &sCommand = command.str();
std::cout << "Compiling [" << functionName << "]\n" << sCommand << "\n\n";
system(sCommand.c_str());
releaseFile(cachedBinary);
}
else
waitForFile(cachedBinary);
}
// [-] Tentative
std::string SINK_LD_LIBRARY_PATH;
char *c_SINK_LD_LIBRARY_PATH = getenv("SINK_LD_LIBRARY_PATH");
if(c_SINK_LD_LIBRARY_PATH != NULL)
SINK_LD_LIBRARY_PATH = std::string(c_SINK_LD_LIBRARY_PATH);
OCCA_COI_CHECK("Device: Initializing",
COIProcessCreateFromFile(data_.deviceID,
cachedBinary.c_str(),
0 , NULL,
true, NULL,
true, NULL,
memoryAllocated ? memoryAllocated : (4 << 30), // 4 GB
SINK_LD_LIBRARY_PATH.c_str(),
&(data_.chiefID)) );
const char *kernelNames[] = {"occaKernelWith1Argument" , "occaKernelWith2Arguments" , "occaKernelWith3Arguments" ,
"occaKernelWith4Arguments" , "occaKernelWith5Arguments" , "occaKernelWith6Arguments" ,
"occaKernelWith7Arguments" , "occaKernelWith8Arguments" , "occaKernelWith9Arguments" ,
"occaKernelWith10Arguments", "occaKernelWith11Arguments", "occaKernelWith12Arguments",
"occaKernelWith13Arguments", "occaKernelWith14Arguments", "occaKernelWith15Arguments",
"occaKernelWith16Arguments", "occaKernelWith17Arguments", "occaKernelWith18Arguments",
"occaKernelWith19Arguments", "occaKernelWith20Arguments", "occaKernelWith21Arguments",
"occaKernelWith22Arguments", "occaKernelWith23Arguments", "occaKernelWith24Arguments",
"occaKernelWith25Arguments"};
// [-] More hard-coding, if you know what I mean
OCCA_COI_CHECK("Device: Getting Kernel Wrappers",
COIProcessGetFunctionHandles(data_.chiefID,
25,
kernelNames,
data_.kernelWrapper));
}
