Adapt multithreaded cluster finder to storage cell data - continued (non-functional)

This commit is contained in:
hinger_v 2025-02-06 20:42:36 +01:00
parent 3a1945116a
commit fcbb87b71a
5 changed files with 221 additions and 95 deletions

View File

@ -46,8 +46,10 @@ void printDatatypeSize(hid_t dataset) {
}
/*
/* **********************
* Class member functions
* **********************
*/
// Default constructor

View File

@ -82,7 +82,7 @@ public:
/**
* Read an image into current_image,
* increment Z-offset (frame) and (if rank==4) storage cell
* @returns frame number read,
* @returns frame number read from file,
*/
int ReadImage (uint16_t* image, std::vector<hsize_t>& offset);

View File

@ -46,13 +46,19 @@ std::string getRootString( std::string const& filepath ) {
* \param fext file extension (e.g. "raw")
*/
std::string createFileName( std::string const& dir, std::string const& fprefix="run",
std::string const& fsuffix="", std::string const& fext="raw", int const outfilecounter=-1 ) {
if (outfilecounter >= 0)
return fmt::format("{:s}/{:s}_{:s}_f{:05d}.{:s}", dir, fprefix, fsuffix, outfilecounter, fext);
else if (fsuffix.length()!=0)
return fmt::format("{:s}/{:s}_{:s}.{:s}", dir, fprefix, fsuffix, fext);
else
return fmt::format("{:s}/{:s}.{:s}", dir, fprefix, fext);
std::string const& fsuffix="", std::string const& fext="raw", int const outfilecounter=-1 ) {
std::string return_string;
if (outfilecounter >= 0)
return_string = fmt::format("{:s}/{:s}_{:s}_f{:05d}", dir, fprefix, fsuffix, outfilecounter);
else if (fsuffix.length()!=0)
return_string = fmt::format("{:s}/{:s}_{:s}", dir, fprefix, fsuffix);
else
return_string = fmt::format("{:s}/{:s}", dir, fprefix);
if (fext.length()!=0)
return_string += "." + fext;
return return_string;
}
/* Adjusts number of threads to be a multiple of number of storage cells
@ -208,6 +214,7 @@ int main(int argc, char *argv[]) {
std::cout << std::ctime(&end_time) << std::endl;
int nThreads = nthreads;
int nSC = 1;
// Validate number of threads for number of storage cells (if applicable)
// Determine the dimensions of the dataset from the first datafile
@ -219,7 +226,8 @@ int main(int argc, char *argv[]) {
// Validate number of threads
if( firstfileh5->GetRank() == 4 ) {
auto h5dims = firstfileh5->GetDatasetDimensions();
auto nThreads = adjustThreads(nthreads,h5dims[1]);
nSC = h5dims[1];
nThreads = adjustThreads(nthreads,nSC);
}
firstfileh5->CloseResources();
@ -229,7 +237,7 @@ int main(int argc, char *argv[]) {
}
multiThreadedCountingDetector* mt =
new multiThreadedCountingDetector(filter, nThreads, fifosize);
new multiThreadedCountingDetector(filter, nThreads, fifosize, nSC);
//auto mt = std::make_unique<multiThreadedCountingDetector>(filter.get(), nthreads, fifosize);
mt->setClusterSize(csize, csize);
mt->newDataSet(); //Initialize new dataset for each thread
@ -242,7 +250,7 @@ int main(int argc, char *argv[]) {
int ifr = 0; //frame counter of while loop
int framenumber = 0; //framenumber as read from file (detector)
std::vector<hsize_t> h5offset; //frame counter internal to HDF5File::ReadImage (provided for sanity check/debugging)
std::vector<hsize_t> h5offset; //hyperslab offset internal to HDF5File::ReadImage
hsize_t h5rank;
if (pedfilename.length()>1) {
@ -279,20 +287,31 @@ int main(int argc, char *argv[]) {
}
int storageCell = 0;
if (h5rank == 4)
hsize_t n_storageCells = 1;
if (h5rank == 4) {
storageCell = h5offset[1];
n_storageCells = pedefile->GetDatasetDimensions()[1];
}
mt->pushData(buff, storageCell); //HERE!!!
mt->nextThread();
mt->popFree(buff);
++ifr;
if (ifr % 100 == 0) {
std::cout << " ****" << ifr << " " << framenumber << " " << h5offset[0]
<< std::endl;
} // else
// push buff into fifoData for a thread corresponding to the active storage cell
mt->pushData(buff, storageCell);
// increment (round-robin) the internal thread counter for that storage cell
mt->nextThread(storageCell);
// get a free memory address from fifoFree of the active storage cell for the next read operation
mt->popFree(buff,storageCell);
/* NOTE: the buff that was popped free from the current thread, will be (likely) pushed into
* the fifoData of a different thread in the next iteration of the loop! */
if (ifr >= 1000)
break;
++ifr;
if (ifr % 100 == 0) {
std::cout << " ****" << ifr << " " << framenumber << " " << h5offset[0];
if (n_storageCells>1)
std::cout << " sc " << storageCell;
std::cout << "\n";
} // else
if (ifr >= 1000*n_storageCells)
break;
//framenumber = 0;
}
@ -302,9 +321,9 @@ int main(int argc, char *argv[]) {
}
std::cout << "froot " << froot << std::endl;
auto imgfname = createFileName( outdir, froot, "ped", "tiff");
auto imgfname = createFileName( outdir, froot, "ped", "" );
mt->writePedestal(imgfname.c_str());
imgfname = createFileName( outdir, froot, "rms", "tiff");
imgfname = createFileName( outdir, froot, "rms", "");
mt->writePedestalRMS(imgfname.c_str());
} else

View File

@ -341,9 +341,9 @@ class threadedAnalogDetector {
class multiThreadedAnalogDetector {
public:
multiThreadedAnalogDetector(analogDetector<uint16_t> *d, int n,
int fs = 1000)
: stop(0), nThreads(n), ithread(0) {
multiThreadedAnalogDetector(analogDetector<uint16_t> *d, int num_threads,
int fs = 1000, int num_sc = 1)
: stop(0), nThreads(num_threads), nSC(num_sc) {
dd[0] = d;
if (nThreads == 1)
dd[0]->setId(100);
@ -359,6 +359,18 @@ class multiThreadedAnalogDetector {
dets[i] = new threadedAnalogDetector(dd[i], fs);
}
// Pre-assign threads to storage cells in a round-robin manner
int threads_per_sc = nThreads / nSC;
sc_to_threads.clear();
for (int s = 0; s < nSC; ++s) {
for (int t = 0; t < threads_per_sc; ++t) {
int assigned_thread = s * threads_per_sc + t;
sc_to_threads[s].push_back(assigned_thread);
}
}
// Set all thread counter to zero for each storage cell
thread_counters_by_sc.resize(nSC,0);
image = nullptr;
ff = NULL;
ped = NULL;
@ -522,82 +534,133 @@ class multiThreadedAnalogDetector {
return ret;
}
virtual bool pushData(char *&ptr, int SC_value) {
/*
virtual std::vector<int> getThreadsForSc(int sc) {
return sc_to_threads[sc];
}
*/
virtual bool pushData(char *&ptr, int sc=0) {
//Additional logic implemented to accommodate storage cells
std::unique_lock<std::mutex> lock(map_mutex);
// Get assigned threads for this storage cell
std::vector<int>& assigned_threads = sc_to_threads[SC_value];
auto& assigned_threads = sc_to_threads[sc];
auto& counter = thread_counters_by_sc[sc];
return dets[assigned_thread]->pushData(ptr);
// Distribute workload among threads using round-robin
int selected_thread = assigned_threads[counter % assigned_threads.size()];
return dets[selected_thread]->pushData(ptr);
}
virtual bool popFree(char *&ptr) {
// cout << ithread << endl;
return dets[ithread]->popFree(ptr);
virtual bool popFree(char *&ptr, int sc=0) {
//Additional logic implemented to accommodate storage cells
std::unique_lock<std::mutex> lock(map_mutex);
// Get assigned threads for this storage cell
auto& assigned_threads = sc_to_threads[sc];
auto& counter = thread_counters_by_sc[sc];
// Distribute workload among threads using round-robin
int selected_thread = assigned_threads[counter % assigned_threads.size()];
return dets[selected_thread]->popFree(ptr);
}
virtual int nextThread() {
ithread++;
if (ithread == nThreads)
ithread = 0;
return ithread;
virtual int nextThread(int sc=0) {
//Additional logic implemented to accommodate storage cells
auto& counter = thread_counters_by_sc[sc];
//counter++;
if (++counter == nThreads/nSC)
counter = 0;
return counter;
}
virtual double *getPedestal() {
// Storage cell sensitive
virtual double *getPedestal(int sc = 0) {
int nx, ny;
dets[0]->getDetectorSize(nx, ny);
if (ped)
delete[] ped;
ped = new double[nx * ny];
if (sc_pedestals.count(sc) && sc_pedestals[sc]) {
delete[] sc_pedestals[sc];
sc_pedestals[sc] = nullptr;
}
// allocate memory and initialize all values to zero
sc_pedestals[sc] = new double[nx * ny](); // parentheses initialize elements to zero
//std::fill(sc_pedestals[sc], sc_pedestals[sc] + (nx * ny), 0.0); // explicit zero initialization
double *p0 = new double[nx * ny];
for (int i = 0; i < nThreads; i++) {
// Get the threads assigned to this storage cell
auto const& assigned_threads = sc_to_threads[sc];
int num_threads = assigned_threads.size();
// Only iterate over threads assigned to this storage cell
for ( int thread_id : assigned_threads ) {
// inte=(slsInterpolation*)dets[i]->getInterpolation(nb,emi,ema);
// cout << i << endl;
p0 = dets[i]->getPedestal(p0);
if (p0) {
if (i == 0) {
//p0 = dets[thread_id]->getPedestal(p0);
dets[thread_id]->getPedestal(p0);
if (p0) { /*
if (i == 0) {
// If first thread, initialize ped with first thread's values
for (int ib = 0; ib < nx * ny; ib++) {
ped[ib] = p0[ib] / ((double)nThreads);
// cout << p0[ib] << " ";
}
} else {
for (int ib = 0; ib < nx * ny; ib++) {
ped[ib] += p0[ib] / ((double)nThreads);
// cout << p0[ib] << " ";
}
*/
// For subsequent threads, accumulate pedestal values
// if ( i == 0 ) becomes superfluous if we zero-initialize earlier
for (int ib = 0; ib < nx * ny; ib++) {
sc_pedestals[sc][ib] += p0[ib] / ((double)num_threads);
// cout << p0[ib] << " ";
}
//}
}
}
delete[] p0;
return ped;
return sc_pedestals[sc];
};
virtual double *getPedestalRMS() {
// Storage cell sensitive
virtual double *getPedestalRMS(int sc = 0) {
int nx, ny;
dets[0]->getDetectorSize(nx, ny);
// if (ped) delete [] ped;
double *rms = new double[nx * ny];
if (sc_pedestals_rms.count(sc) && sc_pedestals_rms[sc]) {
delete[] sc_pedestals_rms[sc];
sc_pedestals_rms = nullptr;
}
// allocate memory and initialize all values to zero
sc_pedestals_rms[sc] = new double[nx * ny](); // Zero-initialize
//std::fill(sc_pedestals_rms[sc], sc_pedestals_rms[sc] + (nx * ny), 0.0); // explicit zero initialization
//double *rms = sc_pedestals_rms[sc];
double *p0 = new double[nx * ny];
for (int i = 0; i < nThreads; i++) {
// Get the threads assigned to this storage cell
auto const& assigned_threads = sc_to_threads[sc];
int num_threads = assigned_threads.size();
// Only iterate over threads assigned to this storage cell
for (int thread_id : assigned_threads) {
// inte=(slsInterpolation*)dets[i]->getInterpolation(nb,emi,ema);
// cout << i << endl;
p0 = dets[i]->getPedestalRMS(p0);
if (p0) {
if (i == 0) {
//p0 = dets[thread_id]->getPedestalRMS(p0);
dets[thread_id]->getPedestalRMS(p0);
for (int ib = 0; ib < nx * ny; ib++) {
rms[ib] = p0[ib] * p0[ib] / ((double)nThreads);
if (p0) {
for (int ib = 0; ib < nx * ny; ib++) {
sc_pedestals_rms[sc][ib] += (p0[ib] * p0[ib]) / ((double)num_threads);
// cout << p0[ib] << " ";
}
} else {
for (int ib = 0; ib < nx * ny; ib++) {
rms[ib] += p0[ib] * p0[ib] / ((double)nThreads);
// cout << p0[ib] << " ";
}
}
}
}
delete[] p0;
@ -608,9 +671,10 @@ class multiThreadedAnalogDetector {
/* rms[ib]=0; */
/* } */
return rms;
return sc_pedestals_rms[sc];
};
// Does not differentiate for storage cells!
virtual double *setPedestal(double *h = NULL) {
// int nb=0;
@ -625,44 +689,79 @@ class multiThreadedAnalogDetector {
return NULL;
};
virtual void *writePedestal(const char *imgname) {
// Storage cell sensitive
virtual void *writePedestal(char const* base_imgname) {
int nx, ny;
dets[0]->getDetectorSize(nx, ny);
getPedestal();
float *gm = new float[nx * ny];
if (gm) {
for (int ix = 0; ix < nx * ny; ix++) {
gm[ix] = ped[ix];
}
WriteToTiff(gm, imgname, nx, ny);
delete[] gm;
} else
std::cout << "Could not allocate float image " << std::endl;
//float *gm = new float[nx * ny];
if(ped)
delete[] ped; //Memory cleanup
// Loop over each storage cell
for ( auto const& entry : sc_to_threads ) {
int sc = entry.first;
std::string imgname = std::string(base_imgname);
if (nSC>1)
imgname += "_SC" + std::to_string(sc);
imgname += ".tiff";
getPedestal(sc); // Compute pedestal for this storage cell
std::vector<float> gm(nx * ny);
// Copy pedestal data into the float array
/*
for (int ix = 0; ix < nx * ny; ix++) {
gm[ix] = sc_pedestals[sc][ix];
}
*/
std::copy(sc_pedestals[sc], sc_pedestals[sc] + (nx * ny), gm.data());
WriteToTiff(gm.data(), imgname.c_str(), nx, ny);
// Clean up memory
if(sc_pedestals[sc]) {
delete[] sc_pedestals[sc];
sc_pedestals[sc] = nullptr;
}
}
return NULL;
};
virtual void *writePedestalRMS(const char *imgname) {
// Storage cell sensitive
virtual void *writePedestalRMS(char const* base_imgname) {
int nx, ny;
dets[0]->getDetectorSize(nx, ny);
double *rms = getPedestalRMS();
float *gm = new float[nx * ny];
if (gm) {
//float *gm = new float[nx * ny];
// Loop over each stoarge cell
for ( auto const& entry : sc_to_threads ) {
int sc = entry.first;
std::string imgname = std::string(base_imgname);
if (nSC>1)
imgname += "_SC" + std::to_string(sc);
imgname += ".tiff";
double *rms = getPedestalRMS(sc); // Compute pedestal RMS for this storage cell
std::vector<float> gm(nx * ny);
// Copy rms data into the float array
/*
for (int ix = 0; ix < nx * ny; ix++) {
gm[ix] = rms[ix];
gm[ix] = rms[ix];
}
WriteToTiff(gm, imgname, nx, ny);
delete[] gm;
*/
std::copy(rms, rms + (nx * ny), gm.data());
WriteToTiff(gm.data(), imgname.c_str(), nx, ny);
// Clean up memory
delete[] rms;
} else
std::cout << "Could not allocate float image " << std::endl;
if(sc_pedestals_rms[sc]) {
delete[] sc_pedestals_rms[sc];
sc_pedestals_rms[sc] = nullptr;
}
}
return NULL;
};
@ -713,13 +812,19 @@ class multiThreadedAnalogDetector {
const int nThreads;
threadedAnalogDetector *dets[MAXTHREADS];
analogDetector<uint16_t> *dd[MAXTHREADS];
int ithread;
//int ithread{0}; // Thread index
std::vector<int> thread_counters_by_sc{}; // Round-robin counters for threads for each storage cell
int* image;
int* ff;
double* ped;
//pthread_mutex_t fmutex; //unused
std::unordered_map< int, std::vector<int> > sc_to_threads; // Maps storage cell -> assigned threads
std::mutex map_mutex; // Ensure thread safe access to the map
std::unordered_map< int, std::vector<int> > sc_to_threads; // Maps storage cell -> vector of assigned thread ids
std::mutex map_mutex; // Ensure thread-safe access to the map
int nSC{1}; // Number of storage cells
std::unordered_map<int, double*> sc_pedestals; // Store pedestal arrays per storage cell
std::unordered_map<int, double*> sc_pedestals_rms; // Store pedestal RMS arrays per storage cell
// at the moment, these maps could be avoided, but this implementation is more robust in allowing future changes
};
#endif

View File

@ -19,8 +19,8 @@ using namespace std;
class multiThreadedCountingDetector : public multiThreadedAnalogDetector {
public:
multiThreadedCountingDetector(singlePhotonDetector *d, int n, int fs = 1000)
: multiThreadedAnalogDetector(d, n, fs){};
multiThreadedCountingDetector(singlePhotonDetector *d, int num_threads, int fs = 1000, int num_sc = 1)
: multiThreadedAnalogDetector(d, num_threads, fs, num_sc){};
// virtual
// ~multiThreadedCountingDetector{multiThreadedAnalogDetector::~multiThreadedAnalogDetector();};
virtual double setNSigma(double n) {