remove some unnecessary memory allocation, replace by persistently allocated buffers

bioem.cpp

@@ -362,25 +362,11 @@ int bioem::compareRefMaps(int iProjectionOut, int iConv, const myfloat_t* conv_m
 {
 	if (FFTAlgo)
 	{
-		#pragma omp parallel
+		#pragma omp parallel for
+		for (int iRefMap = startMap; iRefMap < RefMap.ntotRefMap; iRefMap ++)
 		{
-			mycomplex_t *localCCT;
-			myfloat_t *lCC;
-			localCCT = (mycomplex_t *) myfftw_malloc(sizeof(mycomplex_t) * param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D);
-			lCC = (myfloat_t *) myfftw_malloc(sizeof(myfloat_t) * param.param_device.NumberPixels * param.param_device.NumberPixels);
-
-			const int num_threads = omp_get_num_threads();
-			const int thread_id = omp_get_thread_num();
-			const int mapsPerThread = (RefMap.ntotRefMap - startMap + num_threads - 1) / num_threads;
-			const int iStart = startMap + thread_id * mapsPerThread;
-			const int iEnd = min(RefMap.ntotRefMap, startMap + (thread_id + 1) * mapsPerThread);
-
-			for (int iRefMap = iStart; iRefMap < iEnd; iRefMap ++)
-			{
-				calculateCCFFT(iRefMap, iProjectionOut, iConv, sumC, sumsquareC, localmultFFT, localCCT, lCC);
-			}
-			myfftw_free(localCCT);
-			myfftw_free(lCC);
+			const int num = omp_get_thread_num();
+			calculateCCFFT(iRefMap, iProjectionOut, iConv, sumC, sumsquareC, localmultFFT, localCCT[num], lCC[num]);
 		}
 	}
 	else
@@ -420,7 +406,7 @@ int bioem::createProjection(int iMap, mycomplex_t* mapFFT)
 	myfloat_t alpha, gam, beta;
 	myfloat_t* localproj;
 
-	localproj = (myfloat_t *) myfftw_malloc(sizeof(myfloat_t) * param.param_device.NumberPixels * param.param_device.NumberPixels);
+	localproj = lCC[omp_get_thread_num()];
 	memset(localproj, 0, param.param_device.NumberPixels * param.param_device.NumberPixels * sizeof(*localproj));
 
 	alpha = param.angles[iMap].pos[0];
@@ -492,8 +478,6 @@ int bioem::createProjection(int iMap, mycomplex_t* mapFFT)
 	// ********** Omp Critical is necessary with FFTW*******
 	myfftw_execute_dft_r2c(param.fft_plan_r2c_forward, localproj, mapFFT);
 
-	myfftw_free(localproj);
-
 	return(0);
 }
 
@@ -505,8 +489,7 @@ int bioem::createConvolutedProjectionMap(int iMap, int iConv, mycomplex_t* lproj
 	// *************** and Backtransforming it to real Space ********************************
 	// **************************************************************************************
 
-	mycomplex_t* tmp;
-	tmp = (mycomplex_t*) myfftw_malloc(sizeof(mycomplex_t) * param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D);
+	mycomplex_t* tmp = localCCT[omp_get_thread_num()];
 
 	// **** Multiplying FFTmap with corresponding kernel ****
 	const mycomplex_t* refCTF = &param.refCTF[iConv * param.FFTMapSize];
@@ -538,9 +521,6 @@ int bioem::createConvolutedProjectionMap(int iMap, int iConv, mycomplex_t* lproj
 	sumC = sumC / norm2;
 	sumsquareC = sumsquareC / norm4;
 
-	// **** Freeing fftw_complex created (dont know if omp critical is necessary) ****
-	myfftw_free(tmp);
-
 	return(0);
 }
 
@@ -570,10 +550,32 @@ int bioem::deviceInit()
 
 int bioem::deviceStartRun()
 {
+	if (FFTAlgo)
+	{
+		const int count = omp_get_max_threads();
+		localCCT = new mycomplex_t*[count];
+		lCC = new myfloat_t*[count];
+		for (int i = 0;i < count;i++)
+		{
+			localCCT[i] = (mycomplex_t *) myfftw_malloc(sizeof(mycomplex_t) * param.param_device.NumberPixels * param.param_device.NumberFFTPixels1D);
+			lCC[i] = (myfloat_t *) myfftw_malloc(sizeof(myfloat_t) * param.param_device.NumberPixels * param.param_device.NumberPixels);
+		}
+	}
 	return(0);
 }
 
 int bioem::deviceFinishRun()
 {
+	if (FFTAlgo)
+	{
+		const int count = omp_get_max_threads();
+		for (int i = 0;i < count;i++)
+		{
+			myfftw_free(localCCT[i]);
+			myfftw_free(lCC[i]);
+		}
+		delete[] localCCT;
+		delete[] lCC;
+	}
 	return(0);
 }

include/bioem.h

@@ -47,6 +47,9 @@ protected:
 	int nProjectionMapsTotal;	//Maps in total
 
 	int FFTAlgo;
+
+	mycomplex_t** localCCT;
+	myfloat_t** lCC;
 };
 
 #endif