From d1dceb110682ff31e1a4dc1767eb3d533e4fd21c Mon Sep 17 00:00:00 2001
From: jihoonl <jihoonl>
Date: Wed, 7 Oct 2009 19:12:18 +0000
Subject: [PATCH] dsm.recovery matrixmultiply final

---
 .../MatrixMultiply/dsm/MatrixMultiplyN.java   | 295 +++++++++++-------
 .../Prefetch/MatrixMultiply/dsm/makefile      |  18 +-
 2 files changed, 190 insertions(+), 123 deletions(-)

diff --git a/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/MatrixMultiplyN.java b/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/MatrixMultiplyN.java
index ebb6d207..f379faa8 100644
--- a/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/MatrixMultiplyN.java
+++ b/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/MatrixMultiplyN.java
@@ -1,117 +1,183 @@
-public class MatrixMultiply extends Thread{
-    MMul mmul;
-    public int x0, y0, x1, y1;
-    public MatrixMultiply(MMul mmul, int x0, int x1, int y0, int y1) {
-	this.mmul = mmul;
-	this.x0 = x0;
-	this.y0 = y0;
-	this.x1 = x1;
-	this.y1 = y1;
-    //System.printString("x0 = " +x0+" x1= "+x1+" y0= "+y0+" y1= "+y1+"\n");
-    }
-    
-    public void run() {
-	atomic {
-	    double la[][]=mmul.a;
-	    double lc[][]=mmul.c;
-	    double lb[][]=mmul.btranspose;
-	    int M=mmul.M;
-        //Use btranspose for cache performance
-	    for(int i = x0; i< x1; i++){
-		double a[]=la[i];
-		double c[]=lc[i];
-		for (int j = y0; j < y1; j++) {
-		    double innerProduct=0;
-		    double b[] = lb[j];
-		    for(int k = 0; k < M; k++) {
-			innerProduct += a[k] *b[k];
-		    }
-		    c[j]=innerProduct;
-		}
-	    }
-	}
-    }
-    
-    public static void main(String[] args) {
-	int NUM_THREADS = 4;
-	int SIZE=600;
-	if (args.length>0) {
-	    NUM_THREADS=Integer.parseInt(args[0]);
-	    if (args.length>1)
-		SIZE=Integer.parseInt(args[1]);
-	}
-	
-	int[] mid = new int[8];
-	mid[0] = (128<<24)|(195<<16)|(136<<8)|162; //dc-1.calit2
-	mid[1] = (128<<24)|(195<<16)|(136<<8)|163; //dc-2.calit2
-	mid[2] = (128<<24)|(195<<16)|(136<<8)|164; //dc-3.calit2
-	mid[3] = (128<<24)|(195<<16)|(136<<8)|165; //dc-4.calit2
-	mid[4] = (128<<24)|(195<<16)|(136<<8)|166; //dc-5.calit2
-	mid[5] = (128<<24)|(195<<16)|(136<<8)|167; //dc-6.calit2
-	mid[6] = (128<<24)|(195<<16)|(136<<8)|168; //dc-7.calit2
-	mid[7] = (128<<24)|(195<<16)|(136<<8)|169; //dc-8.calit2
- 
-	int p, q, r;
-	MatrixMultiply[] mm;
-	MatrixMultiply tmp;
-	MMul matrix;
+public class MatrixMultiply extends Task {
+	MMul mmul;
+	int SIZE;
+	int increment;
 	
-	atomic {
-	    matrix = global new MMul(SIZE, SIZE, SIZE);
-	    matrix.setValues();
-	    matrix.transpose();
-	    mm = global new MatrixMultiply[NUM_THREADS];
-	    int increment=SIZE/NUM_THREADS;
-	    int base=0;
-	    for(int i=0;i<NUM_THREADS;i++) {
-		if ((i+1)==NUM_THREADS)
-		    mm[i]=global new MatrixMultiply(matrix,base, SIZE, 0, SIZE);
-		else
-		    mm[i]=global new MatrixMultiply(matrix,base, base+increment, 0, SIZE);
-		base+=increment;
-	    }
-	    p = matrix.L;
-	    q = matrix.M;
-	    r = matrix.N;
+	public MatrixMultiply(MMul mmul, int num_threads, int size) {
+		this.mmul = mmul;
+
+/*    if ((size % num_threads) == 0) {
+      NUM_TASKS = num_threads*num_threads;
+    }
+    else {
+      NUM_TASKS = (num_threads+1)*(num_threads+1);
+    }*/
+
+		SIZE = size;
+    increment = 80;
+
+    init();
 	}
-	
-	// print out the matrices to be multiplied
-	System.printString("\n");
-	System.printString("MatrixMultiply: L=");
-	System.printInt(p);
-	System.printString("\t");
-	System.printString("M=");
-	System.printInt(q);
-	System.printString("\t");
-	System.printString("N=");
-	System.printInt(r);
-	System.printString("\n");
-	
-	// start a thread to compute each c[l,n]
-	for (int i = 0; i < NUM_THREADS; i++) {
-	    atomic {
-		tmp = mm[i];
-	    }
-	    tmp.start(mid[i]);
+
+	public void init() {
+		todoList = global new Queue();
+		doneList = global new Queue();
+
+		fillTodoList();
 	}
 
-	
-	// wait for them to finish
-	for (int i = 0; i < NUM_THREADS; i++) {
-	    atomic {
-		tmp = mm[i];
-	    }
-	    tmp.join();
+  // fill up the Work Pool
+	public void fillTodoList() {
+    Segment seg;
+    int i;
+
+    for(i = 0; i < SIZE; i +=increment) {
+
+      if(i+increment > SIZE) {
+        seg = global new Segment(i,SIZE);
+      }
+      else {
+        seg = global new Segment(i, i + increment);
+//        System.out.println("Seg = " + i + " - " + (i+increment));
+      }
+			todoList.push(seg);
+    }
+//    System.out.println("TodoSIZE = " + todoList.size());
 	}
-	
-	// print out the result of the matrix multiply
 
-	System.printString("Finished\n");
+	public void execute() {
+    double la[][];
+    double lc[][];
+    double lb[][];
+    double rowA[];
+    double colB[];
+    Segment seg;
+		
+    double innerproduct;
+    int i,j;
+    int x0;
+    int x1;
+		int size;
+
+    // get matrix 
+    atomic {
+			seg = (Segment)myWork;
+			x0 = seg.x0;  // x start row
+			x1 = seg.x1;  // x end row
+      la = mmul.a;          //  first mat
+      lb = mmul.btranspose; // second mat
+//      lc = mmul.c;          // destination mat
+			size = SIZE;
     }
+
+		lc = new double[size][size];
+		System.out.println("Seg x0 = " + x0 + " - x1 = " + x1);
+		
+		for(i = x0; i < x1 ; i++) {
+			System.printString("i = " + i + "\n");
+		  atomic {
+        rowA = la[i];   // grab first mat's row
+
+				for(j = 0; j < size ; j++) {
+          colB = lb[j]; // grab second mat's col
+
+					innerproduct = computeProduct(rowA,colB, size); // computes the value
+
+          lc[i][j] = innerproduct;  // store in dest mat
+				} // end of for j
+			} 
+		}	// end for i 
+//		}
+		System.out.println("Finished comutation");
+
+		atomic {
+			for (i = x0; i < x1; i++) {
+				for (j = 0; j < size; j++) {
+					mmul.c[i][j] = lc[i][j];
+				}
+			}
+		}
+  }
+
+  public double computeProduct(double[] rowA,double[] colB, int size)
+  {
+    int i;
+    double sum = 0;
+
+    for(i = 0 ;i < size; i++) {
+//      atomic {
+        sum += rowA[i] * colB[i];
+//      }
+    }
+
+    return sum;
+  }
+
+	public void done(Object work) {
+		atomic {
+			doneList.push(work);
+		}
+	}
+
+  public static void main(String[] args) {
+		int NUM_THREADS = 4;
+		int SIZE = 1600;
+    int i,j;
+		Work[] works;
+		MMul matrix;
+		MatrixMultiply mm;
+    Segment[] currentWorkList;
+
+		if (args.length > 0) {
+			NUM_THREADS = Integer.parseInt(args[0]);
+		}
+
+		int[] mid = new int[NUM_THREADS];
+//		mid[0] = (128<<24)|(195<<16)|(180<<8)|21; //dc1
+//		mid[1] = (128<<24)|(195<<16)|(180<<8)|24; //dc2
+//		mid[2] = (128<<24)|(195<<16)|(180<<8)|26; //dc3
+    mid[0] = (128<<24)|(195<<16)|(136<<8)|162; //dc4
+		mid[1] = (128<<24)|(195<<16)|(136<<8)|163; //dc5
+		mid[2] = (128<<24)|(195<<16)|(136<<8)|164; //dc6
+		mid[3] = (128<<24)|(195<<16)|(136<<8)|165; //dc6
+		mid[4] = (128<<24)|(195<<16)|(136<<8)|166; //dc6
+		mid[5] = (128<<24)|(195<<16)|(136<<8)|167; //dc6
+
+		atomic {
+			matrix = global new MMul(SIZE, SIZE, SIZE);
+			matrix.setValues();
+			matrix.transpose();
+			mm = global new MatrixMultiply(matrix, NUM_THREADS, SIZE);
+
+			works = global new Work[NUM_THREADS];
+      currentWorkList = global new Segment[NUM_THREADS];
+
+			for(i = 0; i < NUM_THREADS; i++) {
+				works[i] = global new Work(mm, NUM_THREADS, i,currentWorkList);
+			}
+		}
+    System.out.println("Finished to createObjects");
+
+		Work tmp;
+		for (i = 0; i < NUM_THREADS; i++) {
+			atomic {
+				tmp = works[i];
+			}
+			Thread.myStart(tmp,mid[i]);
+		}
+
+		for (i = 0; i < NUM_THREADS; i++) {
+			atomic {
+				tmp = works[i];
+			}
+			tmp.join();
+		}
+    
+    System.printString("Finished\n");
+	}
 }
 
 public class MMul{
-
 	public int L, M, N;
 	public double[][] a;
 	public double[][] b;
@@ -130,27 +196,27 @@ public class MMul{
 
 	public void setValues() {
 		for(int i = 0; i < L; i++) {
-            double ai[] = a[i];
+			double ai[] = a[i];
 			for(int j = 0; j < M; j++) {
 				ai[j] = j+1;
 			}
 		}
 
 		for(int i = 0; i < M; i++) {
-            double bi[] = b[i];
+			double bi[] = b[i];
 			for(int j = 0; j < N; j++) {
 				bi[j] = j+1;
 			}
 		}
 
 		for(int i = 0; i < L; i++) {
-            double ci[] = c[i];
+			double ci[] = c[i];
 			for(int j = 0; j < N; j++) {
 				ci[j] = 0;
 			}
 		}
 		for(int i = 0; i < N; i++) {
-            double btransposei[] = btranspose[i];
+			double btransposei[] = btranspose[i];
 			for(int j = 0; j < M; j++) {
 				btransposei[j] = 0;
 			}
@@ -159,10 +225,21 @@ public class MMul{
 
 	public void transpose() {
 		for(int row = 0; row < M; row++) {
-            double brow[] = b[row];
+			double brow[] = b[row];
 			for(int col = 0; col < N; col++) {
 				btranspose[col][row] = brow[col];
 			}
 		}
 	}
 }
+
+public class Segment {
+	int x0;
+	int x1;
+
+	Segment (int x0, int x1) {
+		this.x0 = x0;
+		this.x1 = x1;
+	}
+}
+
diff --git a/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/makefile b/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/makefile
index f2162cd9..fb7c4ba7 100644
--- a/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/makefile
+++ b/Robust/src/Benchmarks/Prefetch/MatrixMultiply/dsm/makefile
@@ -1,21 +1,11 @@
 MAINCLASS=MatrixMultiply
-SRC1=${MAINCLASS}N.java
-SRC2=${MAINCLASS}Nrun.java
-SRC3=${MAINCLASS}D.java \
-	../../../../ClassLibrary/JavaDSM/Barrier.java
-SRC4=${MAINCLASS}D2.java \
-	../../../../ClassLibrary/JavaDSM/Barrier.java
-SRC5=${MAINCLASS}D3.java \
-	../../../../ClassLibrary/JavaDSM/Barrier.java
-FLAGS=-dsm -dsmcaching -optimize -prefetch -excprefetch MatrixMultiply.main -excprefetch MMul.setValues -excprefetch MMul.transpose -mainclass ${MAINCLASS} -trueprob 0.98
+SRC1=${MAINCLASS}N.java 	 
+FLAGS=-dsm -dsmcaching -prefetch -optimize -excprefetch MatrixMultiply.main -excprefetch MMul.setValues -excprefetch MMul.transpose -mainclass ${MAINCLASS} -trueprob 0.98
 FLAGS1=-dsm -optimize -dsmcaching -mainclass ${MAINCLASS}
 FLAGS2=-dsm -optimize -mainclass ${MAINCLASS}
+FLAGS3= -dsm -32bit -nooptimize -recovery -debug -mainclass ${MAINCLASS}
 default:
-	../../../../buildscript ${FLAGS2} -o ${MAINCLASS}NPNC ${SRC3}
-#	../../../../buildscript ${FLAGS1} -o ${MAINCLASS}NPC ${SRC1}
-	../../../../buildscript ${FLAGS} -o ${MAINCLASS}N ${SRC3}
-#	../../../../buildscript ${FLAGS2} -o ${MAINCLASS}NPNC200 ${SRC2}
-#	../../../../buildscript ${FLAGS} -o ${MAINCLASS}N200 ${SRC2}
+	../../../../buildscript ${FLAGS3} -o ${MAINCLASS} ${SRC1} 
 
 clean:
 	rm -rf tmpbuilddirectory
-- 
2.34.1