#include <stdio.h>
#include <sys/time.h>

#define ROWSA 800
#define COLSA 800
#define ROWSB 800
#define COLSB 800
#define MAXROWS 800
#define MAXCOLS 800

void setMatrix(int a[MAXROWS][MAXCOLS], int rows, int cols)
{
    int i, j;
    
    for (i=0; i<rows; i++)
    {
        for(j=0; j < cols; j++)
	      a[i][j] = rand() % 10;
    }
}


void printMatrix(int a[MAXROWS][MAXCOLS], int rows, int cols)
{
    int i, j;
    
    for (i=0; i<rows; i++)
    {
        for(j=0; j < cols; j++)
	    printf("%5d", a[i][j]);
	printf("\n");
    }
}

void multMatrix(int a[MAXROWS][MAXCOLS], 
                int b[MAXROWS][MAXCOLS], 
                int c[MAXROWS][MAXCOLS],
                  int rowsA, int colsB, int colsA)
{
    int i, j, k;
    int sum = 0;
    
    for (i=0; i<rowsA; i++)
    {
     for(j=0; j < colsB; j++) 
	{
	   sum = 0;
	   for (k=0; k < colsA; k++)
	       sum = a[i][k]*b[k][j] + sum; 
 	   c[i][j] = sum;
	}
    }
}

int main(int argc, char** argv)
{
     int a[MAXROWS][MAXCOLS];
     int b[MAXROWS][MAXCOLS];
     int c[MAXROWS][MAXCOLS];
     int rowsA, rowsB, rowsC;
     int colsA, colsB, colsC;
     int i;
    struct timeval tv;
    struct timezone tz;
    long starttimeSec, starttimeMicro;
    long endtimeSec, endtimeMicro;
    long totalSec, totalMicro;   

     
     rowsC = ROWSA;
     colsC = ROWSB;
     
     setMatrix(a, ROWSA, COLSA);
     printf("Matrix A set, %d rows, %d cols\n", ROWSA, COLSA);
     setMatrix(b, ROWSB, COLSB);
     printf("Matrix B set, %d rows, %d cols\n", ROWSB, COLSB);
     
     gettimeofday(&tv, &tz);
     starttimeSec = tv.tv_sec;
     starttimeMicro = tv.tv_usec;
 
     if (COLSA == ROWSB)
         multMatrix(a, b, c, ROWSA, COLSB, COLSA);
     else
        setMatrix(c, rowsC, colsC);     

     gettimeofday(&tv, &tz);
     endtimeSec = tv.tv_sec;
     endtimeMicro = tv.tv_usec;
     if (endtimeSec > starttimeSec)
     {
        totalSec = endtimeSec-starttimeSec;
	totalMicro = endtimeMicro + (1000000 - starttimeMicro);
     }
     else {
        totalSec = 0;
	totalMicro = endtimeMicro - starttimeMicro;
     }	
 
     printf("Matrix C set, %d rows, %d cols\n", rowsC, colsC);
     printf("Time = %d sec, %d microsec\n", totalSec, totalMicro);
/*     
     printMatrix(a, rowsA, colsA); 
     printf("\n");
     printMatrix(b, rowsB, colsB);
     printf("\n");
     printMatrix(c, rowsC, colsC);    
*/     
     return 0;
}

/*
Cray: cc -h report=vf -o matmult matmult.c -O3

CC-6287 cc: VECTOR File = matmult.c, Line = 15
  A loop was not vectorized because it contains a call to function "rand" on
          line 18.


CC-6287 cc: VECTOR File = matmult.c, Line = 17
  A loop was not vectorized because it contains a call to function "rand" on
          line 18.


CC-6287 cc: VECTOR File = matmult.c, Line = 27
  A loop was not vectorized because it contains a call to function "printf" on
          line 30.


CC-6287 cc: VECTOR File = matmult.c, Line = 29
  A loop was not vectorized because it contains a call to function "printf" on
          line 30.


CC-6294 cc: VECTOR File = matmult.c, Line = 43
  A loop was not vectorized because a better candidate was found at line 45.



CC-6204 cc: VECTOR File = matmult.c, Line = 45
  A loop was vectorized.


CC-6204 cc: VECTOR File = matmult.c, Line = 48
  A loop was vectorized.


CC-6287 cc: VECTOR File = matmult.c, Line = 15
  A loop was not vectorized because it contains a call to function "rand" on
          line 18.


CC-6287 cc: VECTOR File = matmult.c, Line = 17
  A loop was not vectorized because it contains a call to function "rand" on
          line 18.

sn3313% matmult
Matrix A set, 500 rows, 500 cols
Matrix B set, 500 rows, 500 cols
Matrix C set, 500 rows, 500 cols
Time = 0 sec, 48 microsec
sn3313% 

Linux Transformer
./matmult
Matrix A set, 500 rows, 500 cols
Matrix B set, 500 rows, 500 cols
Matrix C set, 500 rows, 500 cols
Time = 12 sec, 857357 microsec

Linux Element
rlatimer@helium:~/MPI/Matrix$ ./matmult
Matrix A set, 500 rows, 500 cols
Matrix B set, 500 rows, 500 cols
Matrix C set, 500 rows, 500 cols
Time = 4 sec, 572827 microsec

sn3313% matmult
Matrix A set, 800 rows, 800 cols
Matrix B set, 800 rows, 800 cols
Matrix C set, 800 rows, 800 cols
Time = 0 sec, 47 microsec

Transformer
./matmult
Matrix A set, 800 rows, 800 cols
Matrix B set, 800 rows, 800 cols
Matrix C set, 800 rows, 800 cols
Time = 83 sec, 1854435 microsec

Element
./matmult
Matrix A set, 800 rows, 800 cols
Matrix B set, 800 rows, 800 cols
Matrix C set, 800 rows, 800 cols
Time = 20 sec, 1578604 microsec




*/