12.1.4.4 : Le main_mask.cpp avec un masque
On pourrait se dire que le calcul du modulo prend du temps, et qu'il serait plus rapide avec un masque.
On commence avec les includes standards :
1 2 3 4 5 6 7 8 |
#include <iostream> #include <vector> #include <numeric> #include <execution> #include <algorithm> #include "micro_benchmark.h" |
On déclare notre kernel :
1 2 3 4 5 6 7 8 9 10 11 |
///Compute the calibration /** @param[out] vecCalibSignal : vector of calibrated signal * @param tabADC : pointer to the table of ADC values * @param tabPed : pointer to the table of pedestal * @param tabGain : pointer to the table of gain * @param vecIdx : vector of index to be used to round robin over the pixel * @param nbPixel : number of pixels in the camera */ void compute_calibration(std::vector<float> & vecCalibSignal, const float * tabADC, const float * tabPed, const float* tabGain, const std::vector<int> & vecIdx, int nbPixel) { |
On calcule un masque pour remplacer notre modulo, mais cela ne foncitonne que si le nombre de pixel est une puissance de 2 :
1 |
int maskModulo = nbPixel - 1; //Ok only if nbPixel is a power of two |
La méthode d'exécution sera définie par la macro EXECUTION_POLICY comme ceci nous pourrons compiler la même source avec plusieurs méthodes :
1 2 |
std::transform(EXECUTION_POLICY, std::begin(vecIdx), std::end(vecIdx), std::begin(vecCalibSignal), [=](int i){ |
Il faut passer notre tableau d'indices pour récupérer les bonnes valeurs de tabPed et tabGain :
1 2 3 4 |
unsigned int idxPedGain = i & maskModulo; //Prevent read of vecGain and vecPedestal outside of the bound return (tabADC[i] - tabPed[idxPedGain])*tabGain[idxPedGain]; }); } |
La fonction qui évaluera notre kernel :
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 |
///Get the number of nanoseconds per elements of the Calibration /** @param nbPixel : number of pixels of the tables */ void evaluateCalibration(size_t nbPixel){ //Let's define size of data : size_t nbEvent(NB_EVENT), nbSlice(2lu); size_t nbElement(nbEvent*nbSlice*nbPixel); std::vector<float> vecGain(nbPixel), vecPedestal(nbPixel); std::fill(vecGain.begin(), vecGain.end(), 0.02f); std::fill(vecPedestal.begin(), vecPedestal.end(), 40.0f); std::vector<float> vecADCSignal(nbElement), vecCalibSignal(nbElement); std::fill(vecADCSignal.begin(), vecADCSignal.end(), 42.0f); std::vector<int> vecIdx(nbElement); //Init vector of index for the computation std::iota(vecIdx.begin(), vecIdx.end(), 0); //Hope some day views will work to avoid allocation of index vector //We have to create pointer to be able to catch them by copy without losing any time float * tabADC = vecADCSignal.data(), *tabGain = vecGain.data(), *tabPed = vecPedestal.data(); size_t fullNbElement(nbElement); micro_benchmarkAutoNsPrint("evaluateCalibration", fullNbElement, compute_calibration, vecCalibSignal, tabADC, tabPed, tabGain, vecIdx, nbPixel); } |
Enfin, nous appellons la fonction d'évaluation de MicroBenchmark :
1 2 3 |
int main(int argc, char** argv){ return micro_benchmarkParseArg(argc, argv, evaluateCalibration); } |
Le fichier main_mask.cpp complet :
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 |
/*************************************** Auteur : Pierre Aubert Mail : pierre.aubert@lapp.in2p3.fr Licence : CeCILL-C ****************************************/ #include <iostream> #include <vector> #include <numeric> #include <execution> #include <algorithm> #include "micro_benchmark.h" ///Compute the calibration /** @param[out] vecCalibSignal : vector of calibrated signal * @param tabADC : pointer to the table of ADC values * @param tabPed : pointer to the table of pedestal * @param tabGain : pointer to the table of gain * @param vecIdx : vector of index to be used to round robin over the pixel * @param nbPixel : number of pixels in the camera */ void compute_calibration(std::vector<float> & vecCalibSignal, const float * tabADC, const float * tabPed, const float* tabGain, const std::vector<int> & vecIdx, int nbPixel) { int maskModulo = nbPixel - 1; //Ok only if nbPixel is a power of two std::transform(EXECUTION_POLICY, std::begin(vecIdx), std::end(vecIdx), std::begin(vecCalibSignal), [=](int i){ unsigned int idxPedGain = i & maskModulo; //Prevent read of vecGain and vecPedestal outside of the bound return (tabADC[i] - tabPed[idxPedGain])*tabGain[idxPedGain]; }); } ///Get the number of nanoseconds per elements of the Calibration /** @param nbPixel : number of pixels of the tables */ void evaluateCalibration(size_t nbPixel){ //Let's define size of data : size_t nbEvent(NB_EVENT), nbSlice(2lu); size_t nbElement(nbEvent*nbSlice*nbPixel); std::vector<float> vecGain(nbPixel), vecPedestal(nbPixel); std::fill(vecGain.begin(), vecGain.end(), 0.02f); std::fill(vecPedestal.begin(), vecPedestal.end(), 40.0f); std::vector<float> vecADCSignal(nbElement), vecCalibSignal(nbElement); std::fill(vecADCSignal.begin(), vecADCSignal.end(), 42.0f); std::vector<int> vecIdx(nbElement); //Init vector of index for the computation std::iota(vecIdx.begin(), vecIdx.end(), 0); //Hope some day views will work to avoid allocation of index vector //We have to create pointer to be able to catch them by copy without losing any time float * tabADC = vecADCSignal.data(), *tabGain = vecGain.data(), *tabPed = vecPedestal.data(); size_t fullNbElement(nbElement); micro_benchmarkAutoNsPrint("evaluateCalibration", fullNbElement, compute_calibration, vecCalibSignal, tabADC, tabPed, tabGain, vecIdx, nbPixel); } int main(int argc, char** argv){ return micro_benchmarkParseArg(argc, argv, evaluateCalibration); } |