Commit d8e2a6a0 authored by Giovanni Lacopo's avatar Giovanni Lacopo
Browse files

Better scalability of w-correction with OpenMP

parent d7aacba1

phase_correction.cu

+16 −15
Original line number Original line Diff line number Diff line
@@ -150,19 +150,20 @@ void phase_correction(double* gridss, double* image_real, double* image_imag, in 
#else

#else





#ifndef ACCOMP

#ifndef ACCOMP

	

#ifdef _OPENMP

#ifdef _OPENMP

	omp_set_num_threads(num_threads);

	omp_set_num_threads(num_threads);

#endif

#endif

	

	

        #pragma omp parallel for collapse(3) private(wterm)

       #pragma omp parallel for collapse(2) private(wterm) 

	for (int iw=0; iw<num_w_planes; iw++)

	for (int iw=0; iw<num_w_planes; iw++)

	{

	{

	    for (int iv=0; iv<yaxis; iv++)

	    for (int iv=0; iv<yaxis; iv++)

            for (int iu=0; iu<xaxis; iu++)

            for (int iu=0; iu<xaxis; iu++)

            {

            {





		long index = 2*(iu+iv*xaxis+xaxis*yaxis*iw);

		unsigned int index = 2*(iu+iv*xaxis+xaxis*yaxis*iw);

		long img_index = iu+iv*xaxis;

		unsigned int img_index = iu+iv*xaxis;

		wterm = wmin + iw*dw;

		wterm = wmin + iw*dw;

#ifdef PHASE_ON

#ifdef PHASE_ON

                if (num_w_planes > 1)

                if (num_w_planes > 1)
@@ -218,11 +219,11 @@ void phase_correction(double* gridss, double* image_real, double* image_imag, in 
#else

#else

       #if !defined(__clang__)

       #if !defined(__clang__)





       #pragma omp target teams distribute parallel for collapse(3) simd private(wterm) map(to:gridss[0:2*num_w_planes*xaxis*yaxis]) map(from:image_real[0:xaxis*yaxis]) map(from:image_imag[0:xaxis*yaxis]) device(rank % omp_get_num_devices())

       #pragma omp target teams distribute parallel for collapse(2) simd private(wterm) map(to:gridss[0:2*num_w_planes*xaxis*yaxis]) map(from:image_real[0:xaxis*yaxis]) map(from:image_imag[0:xaxis*yaxis]) device(rank % omp_get_num_devices())





       #else

       #else





       #pragma omp target teams distribute parallel for collapse(3) private(wterm) map(to:gridss[0:2*num_w_planes*xaxis*yaxis]) map(from:image_real[0:xaxis*yaxis]) map(from:image_imag[0:xaxis*yaxis]) device(rank % omp_get_num_devices())

       #pragma omp target teams distribute parallel for collapse(2) private(wterm) map(to:gridss[0:2*num_w_planes*xaxis*yaxis]) map(from:image_real[0:xaxis*yaxis]) map(from:image_imag[0:xaxis*yaxis]) device(rank % omp_get_num_devices())

       #endif

       #endif

	

	

	for (int iw=0; iw<num_w_planes; iw++)

	for (int iw=0; iw<num_w_planes; iw++)