Robust/src/Benchmarks/Prefetch/MatrixMultiply/MatrixMultiply4.java

   1 public class MatrixMultiply extends Thread{
   2         MMul mmul;
   3         public int x0, y0, x1, y1;
   4
   5         public MatrixMultiply(MMul mmul, int x0, int y0, int x1, int y1) {
   6                 this.mmul = mmul;
   7                 this.x0 = x0;
   8                 this.y0 = y0;
   9                 this.x1 = x1;
  10                 this.y1 = y1;
  11         }
  12
  13         public void run() {
  14                 double localresults[][];
  15
  16                 atomic {
  17                     //compute the results
  18                     localresults=new double[1+x1-x0][1+y1-y0];
  19
  20                     //Use b transpose for cache performance
  21                     for(int i = x0; i<= x1; i++){
  22                         double a[]=mmul.a[i];
  23                         int M=mmul.M;
  24                         for (int j = y0; j <= y1; j++) {
  25                             double innerProduct=0;
  26                             double b[] = mmul.btranspose[j];
  27                             for(int k = 0; k < M; k++) {
  28                                 innerProduct += a[k] *b[k];
  29                             }
  30                             localresults[i-x0][j-y0]=innerProduct;
  31                         }
  32                     }
  33                 }
  34
  35                 atomic {
  36                     //write the results
  37                     for(int i=x0;i<=x1;i++) {
  38                         double c[]=mmul.c[i];
  39                         for(int j=y0;j<=y1;j++) {
  40                             c[j]=localresults[i-x0][j-y0];
  41                         }
  42                     }
  43                 }
  44         }
  45
  46         public static void main(String[] args) {
  47       int NUM_THREADS = 4;
  48       int[] mid = new int[NUM_THREADS];
  49       mid[0] = (128<<24)|(195<<16)|(175<<8)|80;
  50       mid[1] = (128<<24)|(195<<16)|(175<<8)|73;
  51       mid[2] = (128<<24)|(195<<16)|(175<<8)|78;
  52       mid[3] = (128<<24)|(195<<16)|(175<<8)|79;
  53                 //int mid1 = (128<<24)|(195<<16)|(175<<8)|69;
  54                 //int mid2 = (128<<24)|(195<<16)|(175<<8)|73;
  55                 int p, q, r;
  56                 MatrixMultiply[] mm;
  57                 MatrixMultiply tmp;
  58                 MMul matrix;
  59
  60                 atomic {
  61                         matrix = global new MMul(600, 600, 600);
  62                         matrix.setValues();
  63                         matrix.transpose();
  64                 }
  65
  66                 atomic{
  67                         mm = global new MatrixMultiply[NUM_THREADS];
  68                 }
  69
  70                 atomic {
  71                         mm[0] = global new MatrixMultiply(matrix,0,0,300,300);
  72                         mm[1] = global new MatrixMultiply(matrix,0,301,300,599);
  73                         mm[2] = global new MatrixMultiply(matrix,301,0,599,300);
  74                         mm[3] = global new MatrixMultiply(matrix,301,301,599,599);
  75                 }
  76
  77                 atomic {
  78                         p = matrix.L;
  79                         q = matrix.M;
  80                         r = matrix.N;
  81                 }
  82
  83                 // print out the matrices to be multiplied
  84                 System.printString("\n");
  85                 System.printString("MatrixMultiply: L=");
  86                 System.printInt(p);
  87                 System.printString("\t");
  88                 System.printString("M=");
  89                 System.printInt(q);
  90                 System.printString("\t");
  91                 System.printString("N=");
  92                 System.printInt(r);
  93                 System.printString("\n");
  94
  95                 // start a thread to compute each c[l,n]
  96                 for (int i = 0; i < NUM_THREADS; i++) {
  97                         atomic {
  98                                 tmp = mm[i];
  99                         }
 100                         tmp.start(mid[i]);
 101                 }
 102
 103                 // wait for them to finish
 104                 for (int i = 0; i < NUM_THREADS; i++) {
 105                         atomic {
 106                                 tmp = mm[i];
 107                         }
 108                         tmp.join();
 109                 }
 110
 111                 // print out the result of the matrix multiply
 112                 System.printString("Starting\n");
 113                 System.printString("Matrix Product c =\n");
 114                 double val;
 115                 atomic {
 116                         for (int i = 0; i < p; i++) {
 117                                 double c[]=matrix.c[i];
 118                                 for (int j = 0; j < r; j++) {
 119                                         val = c[j];
 120                                 }
 121                         }
 122                 }
 123                 System.printString("Finished\n");
 124         }
 125 }
 126
 127 public class MMul{
 128
 129         public int L, M, N;
 130         public double[][] a;
 131         public double[][] b;
 132         public double[][] c;
 133         public double[][] btranspose;
 134
 135         public MMul(int L, int M, int N) {
 136                 this.L = L;
 137                 this.M = M;
 138                 this.N = N;
 139                 a = global new double[L][M];
 140                 b = global new double[M][N];
 141                 c = global new double[L][N];
 142                 btranspose = global new double[N][M];
 143         }
 144
 145         public void setValues() {
 146                 for(int i = 0; i < L; i++) {
 147             double ai[] = a[i];
 148                         for(int j = 0; j < M; j++) {
 149                                 ai[j] = j+1;
 150                         }
 151                 }
 152
 153                 for(int i = 0; i < M; i++) {
 154             double bi[] = b[i];
 155                         for(int j = 0; j < N; j++) {
 156                                 bi[j] = j+1;
 157                         }
 158                 }
 159
 160                 for(int i = 0; i < L; i++) {
 161             double ci[] = c[i];
 162                         for(int j = 0; j < N; j++) {
 163                                 ci[j] = 0;
 164                         }
 165                 }
 166                 for(int i = 0; i < N; i++) {
 167             double btransposei[] = btranspose[i];
 168                         for(int j = 0; j < M; j++) {
 169                                 btransposei[j] = 0;
 170                         }
 171                 }
 172         }
 173
 174         public void transpose() {
 175                 for(int row = 0; row < M; row++) {
 176             double brow[] = b[row];
 177                         for(int col = 0; col < N; col++) {
 178                                 btranspose[col][row] = brow[col];
 179                         }
 180                 }
 181         }
 182 }