Wrapped code into functions (80d02718) · Commits · Fabio Roberto Vitello / euroexa_GaiaGsrParSim

aprod.c

+311 KiB (327 KiB)

File changed.

No diff preview for this file type.

cblas.h

+2 −2

Original line number	Original line	Diff line number	Diff line
	@@ -50,7 +50,7 @@ cblas_daxpy(const int N, const double alpha, const double *X,

	void		void
	cblas_dcopy(const long int N, const double *X, const int incX,		cblas_dcopy(const long int N, const double *X, const int incX,
	double *Y, const int incY);		double *Y, const int incY, int ntasks=0);


	double		double
	@@ -58,7 +58,7 @@ cblas_ddot(const int N, const double *X, const int incX,
	const double *Y, const int incY);		const double *Y, const int incY);

	double		double
	cblas_dnrm2(const long int N, const double *X, const int incX);		cblas_dnrm2(const long int N, const double *X, const int incX, int ntasks=0);

	void		void
	cblas_dscal(const long int N, const double alpha, double *X, const int incX);		cblas_dscal(const long int N, const double alpha, double *X, const int incX);

lsqr.c

+1340 −1192

File changed.

Preview size limit exceeded, changes collapsed.

lsqrblas.c

+41 −25

Original line number	Original line	Diff line number	Diff line
	@@ -26,6 +26,10 @@
	*/		*/
	#include <stdio.h>		#include <stdio.h>
	#include <stdlib.h>		#include <stdlib.h>
			#define MIN(a,b) (((a)<(b))?(a):(b))
			#define MAX(a,b) (((a)>(b))?(a):(b))


	/*!		/*!
	\file		\file

	@@ -95,7 +99,7 @@ cblas_daxpy( const int N, const double alpha, const double *X,
	*/		*/
	void		void
	cblas_dcopy( const long int N, const double *X,		cblas_dcopy( const long int N, const double *X,
	const int incX, double *Y, const int incY)		const int incX, double *Y, const int incY, int ntasks)
	{		{
	long int i,ix,iy;		long int i,ix,iy;
	// int ix = OFFSET(N, incX);		// int ix = OFFSET(N, incX);
	@@ -107,17 +111,24 @@ cblas_dcopy( const long int N, const double *X,
	else iy=((N) - 1) * (-(incY));		else iy=((N) - 1) * (-(incY));


	#pragma omp parallel shared(ix, iy)		//#pragma omp parallel shared(ix, iy)
			//#pragma omp task shared(ix, iy)
	{		{


	long ixThread,iyThread;		long ixThread,iyThread;
	long ixThreadOffset=ix, iyThreadOffset=iy;		long ixThreadOffset=ix, iyThreadOffset=iy;
	#pragma omp for		//#pragma omp for
	for (i = 0; i < N; i++) {		for (i = 0; i < N; i++) {
			//printf("Inside task %d\n", i);
	ixThread=ixThreadOffset+incX*i;		ixThread=ixThreadOffset+incX*i;
	iyThread=iyThreadOffset+incY*i;		iyThread=iyThreadOffset+incY*i;
	Y[iyThread] = X[ixThread];		Y[iyThread] = X[ixThread];
	}		}


	}//end pragma		}//end pragma
			//#pragma omp taskwait
	}		}

	/*!		/*!
	@@ -157,7 +168,7 @@ cblas_ddot( const int N, const double *X,
	*/		*/

	double		double
	cblas_dnrm2( const long int N, const double *X, const int incX)		cblas_dnrm2( const long int N, const double *X, const int incX, int ntasks)
	{		{
	double		double
	scale = 0.0,		scale = 0.0,
	@@ -190,9 +201,9 @@ cblas_dnrm2( const long int N, const double *X, const int incX)
	return scale * sqrt(ssq);		return scale * sqrt(ssq);
	}		}

			//FV_ not used ??
	double		double
	cblas_dnrm21( const long int N, const double *X, const int incX)		cblas_dnrm21( const long int N, const double *X, const int incX, int ntasks=0)
	{		{
	double		double
	ssq[100], ssqFinal=0;		ssq[100], ssqFinal=0;
	@@ -204,21 +215,25 @@ cblas_dnrm21( const long int N, const double *X, const int incX)

	if (N <= 0 \|\| incX <= 0) return 0;		if (N <= 0 \|\| incX <= 0) return 0;
	else if (N == 1) return fabs(X[0]);		else if (N == 1) return fabs(X[0]);
	#pragma omp parallel private(tid,nthreads,ix)
			////#pragma omp parallel private(tid,nthreads,ix)
	{		{

			/*
	#ifdef OMP		#ifdef OMP
	tid = omp_get_thread_num();		tid = omp_get_thread_num();
	nthreads = omp_get_num_threads();		nthreads = omp_get_num_threads();
	if(nthreads>100) exit(1);		if(nthreads>100) exit(1);
	#endif		#endif
			*/

			////#pragma omp for
	#pragma omp for
	for (long i = 0; i < N; i++) {		for (long i = 0; i < N; i++) {
	ix=incX*i;		ix=incX*i;
	ssq[tid] += X[ix]*X[ix];		ssq[tid] += X[ix]*X[ix];
	}		}
	} // pragma		}
			// pragma

	for(int j=0;j<100;j++)		for(int j=0;j<100;j++)
	ssqFinal+=ssq[j];		ssqFinal+=ssq[j];
	@@ -226,7 +241,7 @@ for(int j=0;j<100;j++)
	return sqrt(ssqFinal);		return sqrt(ssqFinal);
	}		}

			//FV_ not used ??
	double		double
	cblas_dnrm22( const long int N, const double *X, const int incX)		cblas_dnrm22( const long int N, const double *X, const int incX)
	{		{
	@@ -243,14 +258,14 @@ cblas_dnrm22( const long int N, const double *X, const int incX)

	if (N <= 0 \|\| incX <= 0) return 0;		if (N <= 0 \|\| incX <= 0) return 0;
	else if (N == 1) return fabs(X[0]);		else if (N == 1) return fabs(X[0]);
	#pragma omp parallel private(ix)		///#pragma omp parallel private(ix)
	{		////{
	#ifdef OMP		////#ifdef OMP
	tid = omp_get_thread_num();		//// tid = omp_get_thread_num();
	nthreads = omp_get_num_threads();		//// nthreads = omp_get_num_threads();
	if(nthreads>100) exit(1);		//// if(nthreads>100) exit(1);
	#endif		////#endif
	#pragma omp for		////#pragma omp for
	for (i = 0; i < N; i++) {		for (i = 0; i < N; i++) {
	ix=incX*i;		ix=incX*i;
	const double x = X[ix];		const double x = X[ix];
	@@ -269,7 +284,7 @@ cblas_dnrm22( const long int N, const double *X, const int incX)
	}		}

	resultSqr[tid]=(scale[tid] * sqrt(ssq[tid]))(scale[tid] sqrt(ssq[tid]));		resultSqr[tid]=(scale[tid] * sqrt(ssq[tid]))(scale[tid] sqrt(ssq[tid]));
	}// pragma		////}// pragma

	double resultFinal=0.0;		double resultFinal=0.0;
	for(int j=0;j<100;j++) resultFinal+=resultSqr[j];		for(int j=0;j<100;j++) resultFinal+=resultSqr[j];
	@@ -301,6 +316,7 @@ void cblas_dscal(const long int N, const double alpha, double *X, const int incX
	}		}
	}		}

			//FV_ not used ??
	void		void
	cblas_dscal1(const long int N, const double alpha, double *X, const int incX)		cblas_dscal1(const long int N, const double alpha, double *X, const int incX)
	{		{
	@@ -312,14 +328,14 @@ cblas_dscal1(const long int N, const double alpha, double *X, const int incX)
	if (incX > 0) ix=0;		if (incX > 0) ix=0;
	else ix=((N) - 1) * (-(incX));		else ix=((N) - 1) * (-(incX));

	#pragma omp parallel shared(ix)		////#pragma omp parallel shared(ix)
	{		/////{
	long ixThread;		long ixThread;
	long ixThreadOffset=ix;		long ixThreadOffset=ix;
	#pragma omp for		/////#pragma omp for
	for (i = 0; i < N; i++) {		for (i = 0; i < N; i++) {
	ixThread=ixThreadOffset+incX*i;		ixThread=ixThreadOffset+incX*i;
	X[ixThread] *= alpha;		X[ixThread] *= alpha;
	}		}
	}// end pragma		/////}// end pragma
	}		}

memRequest.c

100755 → 100644

+0 −0

File mode changed from 100755 to 100644.

View file