Commit 4a4d7d10 authored by Elwasif, Wael's avatar Elwasif, Wael
Browse files

Merge code.ornl.gov:elwasif/dmrgppPluginSc

parents 657bf875 72e864f0

src/BatchedGemm.cpp

+28 −1
Original line number Diff line number Diff line
@@ -3,6 +3,15 @@ 
#include "setup_sparse_batch.h"



#include "BatchedGemm.h"

//#include "timer.h"

#define _POSIX_C_SOURCE 200809L

#include <inttypes.h>

#include <math.h>

#include <stdio.h>



sem_t* BatchedGemm<FpType>:: gpu_access = NULL;       

extern "C" void get_current_time_with_ms (time_t *sec , long *millisec);





BatchedGemm<FpType>::BatchedGemm(int noperator,

                                 int npatches,
@@ -28,6 +37,20 @@ BatchedGemm<FpType>::BatchedGemm(int noperator, 
	xy_patch_start_ = 0;





        if (gpu_access == NULL){

            gpu_access = sem_open("/dmrg_gpu", O_CREAT, 0666, 1); 

            if(gpu_access == SEM_FAILED) { 

                perror("Semaphore init");

            }

        }

        time_t sec;

        long  ms;

        get_current_time_with_ms (&sec , &ms);

        printf(" %"PRIdMAX".%03ld : waiting for GPU\n", (intmax_t)sec, ms);

        sem_wait(gpu_access);

        get_current_time_with_ms (&sec , &ms);

        printf(" %"PRIdMAX".%03ld : acuired GPU\n", (intmax_t)sec, ms);



        if (use_sparse) {

          nC_ =  0;

          gAbatch_ = 0;
@@ -109,7 +132,11 @@ BatchedGemm<FpType>::~BatchedGemm() { 
	               &Bbatch_

                       );

    };



    sem_post(gpu_access);

    time_t sec;

    long  ms;

    get_current_time_with_ms (&sec , &ms);

    printf(" %"PRIdMAX".%03ld : released GPU\n", (intmax_t)sec, ms);



}

src/BatchedGemm.h

+14 −0
Original line number Diff line number Diff line
@@ -4,6 +4,18 @@ 


#include "dmrg_types.h"

#include "estimate_work.h"

#include <sys/types.h>

#include <sys/ipc.h>

#include <sys/shm.h>

#include <semaphore.h>

#include <stdio.h>

#include <string.h>

#include <stdlib.h>

#include <errno.h>

#include <fcntl.h>

#include <sys/stat.h>

#include <sys/wait.h>

#include <unistd.h>



template<typename T>

class BatchedGemm {
@@ -55,6 +67,8 @@ private: 
        int *ld_gAbatch_;

        int *ld_gBbatch_;



        static sem_t* gpu_access; 



	BatchedGemm(const BatchedGemm<FpType>&);



	BatchedGemm& operator=(const BatchedGemm<FpType>&);

src/Makefile

+2 −1
Original line number Diff line number Diff line
@@ -14,7 +14,8 @@ OBJS=\ 
        unsetup_sparse_batch.o \

        apply_Htarget_sparse.o \

        estimate_work.o \

        BatchedGemm.o

	BatchedGemm.o \

        timer.o 



libdmrgppPluginSc.a: $(OBJS)

	ar cr libdmrgppPluginSc.a $(OBJS)

src/apply_Htarget_pvbatch.c

+5 −6
Original line number Diff line number Diff line 
#include "dmrg_vbatch.h"

#include "timer.h"

#ifdef _OPENMP

#include <omp.h>

#endif
@@ -367,18 +368,16 @@ void apply_Htarget_pvbatch( 
#ifdef _OPENMP

   time_2nd_vbatch += omp_get_wtime();

   gflops2 = gflops2/(giga);

   time_t sec;

   long  ms;

   get_current_time_with_ms (&sec , &ms);



   printf("1st vbatch %f gflops (gflops1=%lf,time=%lf)\n", 

          gflops1/time_1st_vbatch,  gflops1, time_1st_vbatch );

   printf("2nd vbatch %f gflops (gflops2=%lf,time=%lf)\n", 

          gflops2/time_2nd_vbatch, gflops2, time_2nd_vbatch );



   printf("overall %f gflops\n", 

          (gflops1+gflops2)/(time_1st_vbatch + time_2nd_vbatch) );



   printf("apply_Htarget_pvbatch:memory BX (%f GBytes)\n",

          (double) nbytes_BX/(giga) );



   printf(" %"PRIdMAX".%03ld : overall %f gflops memory BX (%f GBytes)\n", (intmax_t)s, ms, (gflops1+gflops2)/(time_1st_vbatch + time_2nd_vbatch), (double) nbytes_BX/(giga) );

#endif

src/apply_Htarget_sparse.c

+7 −4
Original line number Diff line number Diff line
@@ -5,6 +5,7 @@ 
#ifdef _OPENMP

#include <omp.h>

#endif

#include "timer.h"



#ifndef MAX

#define MAX(x,y) ( ((x) > (y)) ?(x):(y) )
@@ -489,12 +490,14 @@ void apply_Htarget_sparse( 
   gflops2 = gflops2/giga;



   if (idebug >= 1) {

       time_t sec;

       long  ms;

       get_current_time_with_ms (&sec , &ms);

       printf("1st vbatch %lf gflops/sec (gflops1=%lf,time=%lf)\n", 

          gflops1/time_1st_vbatch,  gflops1, time_1st_vbatch );

       printf("2nd vbatch %lf gflops/sec (gflops2=%lf,time=%lf)\n", 

          gflops2/time_2nd_vbatch, gflops2, time_2nd_vbatch );



   printf("overall %lf gflops/sec\n", 

       printf(" %"PRIdMAX".%03ld : overall %f gflops\n", (intmax_t)sec, ms, 

           (gflops1+gflops2)/(time_1st_vbatch + time_2nd_vbatch) );



   };