Test_C_OpenACC_2.c

/*
**Copyright (C) 2017, Jyong-Jhih Lin.
**All rights reserved.
**Licensed under the GNU Affero General Public License v3.0. See LICENSE file in the project root for full license information.
*/
/*
Test programs for various systems.
Type: C11, OpenACC, standard output, time
compile:
gcc -std=c11 -Wall -Wextra -O0 <file> -lm -static
pgcc -c11 -Minform=inform -O0 -Minfo -ta=host/multicore/tesla,9.0,cc50 <file>
 */


/*
size_t is the unsigned integer type of the result of sizeof , alignof (since C11) and offsetof.

time_t
This is an arithmetic type returned by the functions timer() and mktime() (usually defined as long).

TIME_UTC is int
#define TIME_UTC implementation-defined (since C11)
*/




#include <stdlib.h>
#include <stdio.h>
#include <time.h>
#include <math.h>
#include <openacc.h>


//size of 2D area
#define COLUMNS 1000
#define ROWS 1000
#define dt_upper_bound 0.01
#define iter_max 4000


int main(void)
{
size_t i=0;
size_t j=0;
double time_difference=0.0;
double time_difference2=0.0;
struct timespec start, end;	//C11 only, new method to measure time.
double temp[ROWS+2][COLUMNS+2];
double temp_last[ROWS+2][COLUMNS+2];
double dt;
size_t iter;


timespec_get(&start, 1/*TIME_UTC*/);
//initialize two area variables
dt=100.0;
iter=0;
for(i=0;i<ROWS+2;i++){
     for(j=1;j<COLUMNS+1;j++){
          temp_last[i][j] = 0.0;
          temp[i][j] = 0.0;
     }
}
for(i=0;i<ROWS+2;i++){
     temp_last[i][0] = 0.0;
     temp_last[i][COLUMNS+1] = (100.0/(double)(ROWS+1)) * (double)i;
}
for(j=0;j<COLUMNS+2;j++){
     temp_last[0][j] = 0.0;
     temp_last[ROWS+1][j] = (100.0/(double)(COLUMNS+1)) * (double)j;
}


//do until dt is smaller than dt_upper_bound or iterations are more than iter_max
while(dt>dt_upper_bound && iter<iter_max){

     //average four neighbors
     for(i=1;i<=ROWS;i++){
          for(j=1;j<=COLUMNS;j++){
               temp[i][j] = 0.25 * (temp_last[i-1][j] + temp_last[i+1][j] + temp_last[i][j-1] + temp_last[i][j+1]);
          }
     }

     dt = 0.0;//reset dt

     for(i=1;i<=ROWS;i++){
          for(j=1;j<=COLUMNS;j++){
               dt = fmax( fabs( temp[i][j] - temp_last[i][j] ), dt );

               temp_last[i][j] = temp[i][j];
          }
     }

     /*if(iter % 100 == 0){
          printf("The iter is %li\n", iter);
          printf("The dt is %lf\n", dt);
     }*/
     iter++;

}
printf("The part 1 is done.\n");
timespec_get(&end, 1/*TIME_UTC*/);

//type "time_t" is defined by the OS, not in C standard.
//Therefore, the size of time_t decides how long the calendar time can be.
time_difference = (double)(end.tv_sec - start.tv_sec) + (double)(end.tv_nsec - start.tv_nsec) / 1.0e9;
/*
*
*
*
*/

//initialize two area variables
dt=100.0;
iter=0;

#pragma acc enter data create(temp[0:ROWS+2][0:COLUMNS+2],temp_last[0:ROWS+2][0:COLUMNS+2])
timespec_get(&start, 1/*TIME_UTC*/);
#pragma acc parallel loop collapse(2) present(temp, temp_last)
for(i=1;i<ROWS+1;i++){
     for(j=0;j<COLUMNS+2;j++){
          temp_last[i][j] = 0.0;
          temp[i][j] = 0.0;
     }
}

#pragma acc parallel loop present(temp_last) async
for(i=0;i<ROWS+2;i++){
     temp_last[i][0] = 0.0;
}
#pragma acc parallel loop present(temp_last) async
for(i=0;i<ROWS+2;i++){
     temp_last[i][COLUMNS+1] = (100.0/(double)(ROWS+1)) * (double)i;
}
#pragma acc parallel loop present(temp_last) async
for(j=0;j<COLUMNS+2;j++){
     temp_last[0][j] = 0.0;
}
#pragma acc parallel loop present(temp_last) async
for(j=0;j<COLUMNS+2;j++){
     temp_last[ROWS+1][j] = (100.0/(double)(COLUMNS+1)) * (double)j;
}
#pragma acc wait


//do until dt is smaller than 0.01 or iterations are more than 4000
while(dt>dt_upper_bound && iter<iter_max){

     //average four neighbors
     #pragma acc parallel loop collapse(2) present(temp, temp_last)
     for(i=1;i<=ROWS;i++){
          for(j=1;j<=COLUMNS;j++){
               temp[i][j] = 0.25 * (temp_last[i-1][j] + temp_last[i+1][j] + temp_last[i][j-1] + temp_last[i][j+1]);
          }
     }

     dt = 0.0;//reset dt
     #pragma acc enter data copyin(dt)
     #pragma acc parallel loop collapse(2) present(temp, temp_last, dt) reduction(max:dt)
     for(i=1;i<=ROWS;i++){
          for(j=1;j<=COLUMNS;j++){
               dt = fmax( fabs( temp[i][j] - temp_last[i][j] ), dt );

               temp_last[i][j] = temp[i][j];
          }
     }
     #pragma acc exit data copyout(dt)

     /*if(iter % 100 == 0){
          printf("The iter is %li\n", iter);
          printf("The dt is %lf\n", dt);
     }*/
     iter++;

}
timespec_get(&end, 1/*TIME_UTC*/);
#pragma acc exit data copyout(temp, temp_last)

printf("The part 2 is done.\n");


time_difference2 = (double)(end.tv_sec - start.tv_sec) + (double)(end.tv_nsec - start.tv_nsec) / 1.0e9;

printf("Type: C11, OpenACC, standard output, time\n");
printf("The time difference is %.9lf\n", time_difference);
printf("The time difference2 is %.9lf\n", time_difference2);
printf("The speed-up ratio is %.9lf\n", time_difference/time_difference2);
printf("The iter is %li\n", iter);
printf("The dt is %lf\n", dt);

return 0;
}

/*
TIMESPEC_GET(3)            Library Functions Manual            TIMESPEC_GET(3)

NAME
     timespec_get - get current calendar time

LIBRARY
     Standard C Library (libc, -lc)

SYNOPSIS
     #include <time.h>

     #define TIME_UTC 1

     int
     timespec_get(struct timespec *ts, int base);

DESCRIPTION
     The timespec_get function sets the interval pointed to by ts to hold the
     current calendar time based on the specified time base in base.

     Currently the only supported valid base is TIME_UTC.  It returns time
     elapsed since epoch.

RETURN VALUES
     The timespec_get function returns the passed value of base if successful,
     otherwise 0 on failure.

SEE ALSO
     clock_gettime(2)

STANDARDS
     The timespec_get function conforms to ISO/IEC 9899:2011 ("ISO C11").

HISTORY
     This interface first appeared in NetBSD 8.

AUTHORS
     Kamil Rytarowski <kamil@NetBSD.org>

NetBSD 8.99                     October 4, 2016                    NetBSD 8.99

https://www.daemon-systems.org/man/timespec_get.3.html
*/