From 049b3154af7d47d1c2d265f20eca183d0b435089 Mon Sep 17 00:00:00 2001
From: adash <adash>
Date: Mon, 9 Mar 2009 20:30:56 +0000
Subject: [PATCH] Manual prefetcing for 2DConv

---
 .../2DConv/ConvolutionSizeC.java              | 459 ++++++++++++++++++
 .../Prefetch/ManualPrefetch/2DConv/Makefile   |  10 +
 .../Prefetch/ManualPrefetch/2DConv/README.txt |   8 +
 3 files changed, 477 insertions(+)
 create mode 100644 Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/ConvolutionSizeC.java
 create mode 100644 Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/Makefile
 create mode 100644 Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/README.txt

diff --git a/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/ConvolutionSizeC.java b/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/ConvolutionSizeC.java
new file mode 100644
index 00000000..12784155
--- /dev/null
+++ b/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/ConvolutionSizeC.java
@@ -0,0 +1,459 @@
+public class Convolution extends Thread {
+  Image img;
+  int x0,x1,y0,y1;
+
+  public Convolution(Image img, int x0, int x1, int y0, int y1) {
+    this.img = img;
+    this.x0 = x0;
+    this.x1 = x1;
+    this.y0 = y0;
+    this.y1 = y1;
+  }
+
+  public void run() {
+
+    //
+    //Prefetch this.img.inputImage[] the first 32 objects
+    // Generates compiler error ???
+    Object o = this;
+    short[] offsets = new short[6];
+    offsets[0] = getoffset{Convolution, img};
+    offsets[1] = (short) 0;
+    offsets[2] = getoffset{Image, inputImage};
+    offsets[3] = (short) 0;
+    offsets[4] = (short) x0;
+    offsets[5] = (short) 32;
+    System.rangePrefetch(o, offsets);
+
+    //Prefetch this.img.outputImage[] the first 32 objects 
+    // Generates compiler error ???
+    Object o1 = this;
+    short[] offsets1 = new short[6];
+    offsets1[0] = getoffset{Convolution, img};
+    offsets1[1] = (short) 0;
+    offsets1[2] = getoffset{Image, outputImage};
+    offsets1[3] = (short) 0;
+    offsets1[4] = (short) x0;
+    offsets1[5] = (short) 32;
+    System.rangePrefetch(o1, offsets1);
+
+    int kernelHeight=15;
+    int kernelWidth=15;
+
+    double[][] kernel = new double[kernelHeight][kernelWidth];
+    initKernel15(kernel);
+
+    atomic {
+      double tempinput[][] = img.inputImage;
+      double tempout[][] = img.outputImage;
+
+      double tinput1[] = tempinput[x0];
+      double tinput2[] = tempinput[x0+1];
+      double tinput3[] = tempinput[x0+2];
+      double tinput4[] = tempinput[x0+3];
+      double tinput5[] = tempinput[x0+4];
+      double tinput6[] = tempinput[x0+5];
+      double tinput7[] = tempinput[x0+6];
+      double tinput8[] = tempinput[x0+7];
+      double tinput9[] = tempinput[x0+8];
+      double tinput10[] = tempinput[x0+9];
+      double tinput11[] = tempinput[x0+10];
+      double tinput12[] = tempinput[x0+11];
+      double tinput13[] = tempinput[x0+12];
+      double tinput14[] = tempinput[x0+13];
+      double tinput0[] = tinput1;
+
+      int l=x0+14;
+      for(int i=x0;i<x1;i++,l++){
+        if(i&31 == 0) {  //prefetch every 16th iteration
+          //Prefetch this.img.inputImage[] 
+          Object o2 = img;
+          short[] offsets2 = new short[4];
+          offsets2[0] = getoffset{Image, inputImage};
+          offsets2[1] = (short) 0;
+          offsets2[2] = (short) (i+32);
+          offsets2[3] = (short) 32;
+          System.rangePrefetch(o2, offsets2);
+
+          //Prefetch this.img.outputImage[] the first 32 objects 
+          Object o3 = img;
+          short[] offsets3 = new short[4];
+          offsets3[0] = getoffset{Image, outputImage};
+          offsets3[1] = (short) 0;
+          offsets3[2] = (short) (i+32);
+          offsets3[3] = (short) 32;
+          System.rangePrefetch(o3, offsets3);
+        }
+
+        double tout[] = tempout[i];
+        tinput0 = tinput1; tinput1=tinput2; tinput2=tinput3; tinput3=tinput4; tinput4=tinput5;
+        tinput5 = tinput6; tinput6=tinput7; tinput7=tinput8; tinput8=tinput9; tinput9=tinput10; 
+        tinput10 = tinput11; tinput11=tinput12; tinput12=tinput13; tinput13=tinput14; tinput14=tempinput[l];
+        for(int j=y0;j<y1;++j){
+          double s=0;
+          for(int b=0;b<kernelHeight;++b) {
+            s+=(tinput0[j+b] * kernel[0][b] + tinput1[j+b] * kernel[1][b] + tinput2[j+b]*kernel[2][b] +
+                tinput3[j+b]*kernel[3][b] + tinput4[j+b]*kernel[4][b] + tinput5[j+b]*kernel[5][b]+ 
+                tinput6[j+b]*kernel[6][b] + tinput7[j+b]*kernel[7][b] + tinput8[j+b]*kernel[8][b]+
+                tinput9[j+b]*kernel[9][b] + tinput10[j+b]*kernel[10][b] + tinput11[j+b]*kernel[11][b]+
+                tinput12[j+b]*kernel[12][b]+ tinput13[j+b]*kernel[13][b] + tinput14[j+b]*kernel[14][b]);
+          }
+          tout[j]=s;
+        }
+      }
+    }
+  }
+
+  public static void main(String[] args) {
+    int SIZE = 256;
+    int NUM_THREADS = 1;
+    int kernelHeight=15, kernelWidth=15;
+
+    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; //dw-10
+    mid[1] = (128<<24)|(195<<16)|(136<<8)|163; //dw-11
+    mid[2] = (128<<24)|(195<<16)|(136<<8)|164; //dw-12
+    mid[3] = (128<<24)|(195<<16)|(136<<8)|165; //dw-13
+    mid[4] = (128<<24)|(195<<16)|(136<<8)|166; //dw-14
+    mid[5] = (128<<24)|(195<<16)|(136<<8)|167; //dw-15
+    mid[6] = (128<<24)|(195<<16)|(136<<8)|168; //dw-16
+    mid[7] = (128<<24)|(195<<16)|(136<<8)|169; //dw-17
+
+    Image img;
+    Convolution[] conv;
+    Convolution tmp;
+
+    atomic {
+      img = global new Image(SIZE,SIZE,kernelHeight,kernelWidth);
+      img.setValues();
+      conv = global new Convolution[NUM_THREADS];
+      int increment=SIZE/NUM_THREADS;
+      int base = 0;
+      for(int i = 0; i<NUM_THREADS; i++) {
+        if((i+1)==NUM_THREADS)
+          conv[i] = global new Convolution(img, base, SIZE, 0, SIZE);
+        else 
+          conv[i] = global new Convolution(img, base, base+increment, 0, SIZE);
+        base+=increment;
+      }
+    }
+ 
+    /*
+    atomic{
+      System.printString("img.outputImage[10][20] = " +(int) img.outputImage[10][20] + "\n");
+      System.printString("img.outputImage[256][890] = " +(int) img.outputImage[256][890] + "\n");
+    }
+    */
+
+	System.printString("Convolution: Size=");
+    System.printInt(SIZE);
+	System.printString("\n");
+
+    for(int i = 0; i <NUM_THREADS; i++) {
+      atomic {
+        tmp = conv[i];
+      }
+      tmp.start(mid[i]);
+    }
+
+    for(int i = 0; i < NUM_THREADS; i++) {
+      atomic {
+        tmp = conv[i];
+      }
+      tmp.join();
+    }
+
+    System.printString("2DConv Done!\n");
+
+    /*
+    atomic{
+      System.printString("img.outputImage[10][20] = " +(int) img.outputImage[10][20] + "\n");
+      System.printString("img.outputImage[256][890] = " +(int) img.outputImage[256][890] + "\n");
+    }
+    */
+  }
+
+  //define 15X15 Gaussian kernel
+  public static void initKernel15(double[][] kernel) {
+    kernel[0][0] = 1/256.0;
+    kernel[0][1] = 4/256.0;
+    kernel[0][2] = 6/256.0;
+    kernel[0][3] = 8/256.0;
+    kernel[0][4] = 10/256.0;
+    kernel[0][5] = 12/256.0;
+    kernel[0][6] = 14/256.0;
+    kernel[0][7] = 16/256.0;
+    kernel[0][8] = 14/256.0;
+    kernel[0][9] = 12/256.0;
+    kernel[0][10] = 10/256.0;
+    kernel[0][11] = 8/256.0;
+    kernel[0][12] = 6/256.0;
+    kernel[0][13] = 4/256.0;
+    kernel[0][14] = 1/256.0;
+
+    kernel[1][0] = 4/256.0;
+    kernel[1][1] = 16/256.0;
+    kernel[1][2] = 24/256.0;
+    kernel[1][3] = 32/256.0;
+    kernel[1][4] = 40/256.0;
+    kernel[1][5] = 48/256.0;
+    kernel[1][6] = 56/256.0;
+    kernel[1][7] = 64/256.0;
+    kernel[1][8] = 56/256.0;
+    kernel[1][9] = 48/256.0;
+    kernel[1][10] = 40/256.0;
+    kernel[1][11] = 32/256.0;
+    kernel[1][12] = 24/256.0;
+    kernel[1][13] = 16/256.0;
+    kernel[1][14] = 4/256.0;
+
+    kernel[2][0] = 6/256.0;
+    kernel[2][1] = 24/256.0;
+    kernel[2][2] = 36/256.0;
+    kernel[2][3] = 48/256.0;
+    kernel[2][4] = 60/256.0;
+    kernel[2][5] = 72/256.0;
+    kernel[2][6] = 84/256.0;
+    kernel[2][7] = 96/256.0;
+    kernel[2][8] = 84/256.0;
+    kernel[2][9] = 72/256.0;
+    kernel[2][10] = 60/256.0;
+    kernel[2][11] = 48/256.0;
+    kernel[2][12] = 36/256.0;
+    kernel[2][13] = 24/256.0;
+    kernel[2][14] = 6/256.0;
+
+    kernel[3][0] = 8/256.0;
+    kernel[3][1] = 32/256.0;
+    kernel[3][2] = 48/256.0;
+    kernel[3][3] = 64/256.0;
+    kernel[3][4] = 80/256.0;
+    kernel[3][5] = 96/256.0;
+    kernel[3][6] = 112/256.0;
+    kernel[3][7] = 128/256.0;
+    kernel[3][8] = 112/256.0;
+    kernel[3][9] = 96/256.0;
+    kernel[3][10] = 80/256.0;
+    kernel[3][11] = 64/256.0;
+    kernel[3][12] = 48/256.0;
+    kernel[3][13] = 32/256.0;
+    kernel[3][14] = 8/256.0;
+
+
+    kernel[4][0] = 10/256.0;
+    kernel[4][1] = 40/256.0;
+    kernel[4][2] = 60/256.0;
+    kernel[4][3] = 80/256.0;
+    kernel[4][4] = 100/256.0;
+    kernel[4][5] = 120/256.0;
+    kernel[4][6] = 140/256.0;
+    kernel[4][7] = 160/256.0;
+    kernel[4][8] = 140/256.0;
+    kernel[4][9] = 120/256.0;
+    kernel[4][10] = 100/256.0;
+    kernel[4][11] = 80/256.0;
+    kernel[4][12] = 60/256.0;
+    kernel[4][13] = 40/256.0;
+    kernel[4][14] = 10/256.0;
+
+    kernel[5][0] = 12/256.0;
+    kernel[5][1] = 48/256.0;
+    kernel[5][2] = 72/256.0;
+    kernel[5][3] = 96/256.0;
+    kernel[5][4] = 120/256.0;
+    kernel[5][5] = 144/256.0;
+    kernel[5][6] = 168/256.0;
+    kernel[5][7] = 192/256.0;
+    kernel[5][8] = 168/256.0;
+    kernel[5][9] = 144/256.0;
+    kernel[5][10] = 120/256.0;
+    kernel[5][11] = 96/256.0;
+    kernel[5][12] = 72/256.0;
+    kernel[5][13] = 48/256.0;
+    kernel[5][14] = 12/256.0;
+
+    kernel[6][0] = 14/256.0;
+    kernel[6][1] = 56/256.0;
+    kernel[6][2] = 84/256.0;
+    kernel[6][3] = 112/256.0;
+    kernel[6][4] = 140/256.0;
+    kernel[6][5] = 168/256.0;
+    kernel[6][6] = 196/256.0;
+    kernel[6][7] = 224/256.0;
+    kernel[6][8] = 196/256.0;
+    kernel[6][9] = 168/256.0;
+    kernel[6][10] = 140/256.0;
+    kernel[6][11] = 112/256.0;
+    kernel[6][12] = 84/256.0;
+    kernel[6][13] = 56/256.0;
+    kernel[6][14] = 14/256.0;
+
+    kernel[7][0] = 16/256.0;
+    kernel[7][1] = 64/256.0;
+    kernel[7][2] = 96/256.0;
+    kernel[7][3] = 128/256.0;
+    kernel[7][4] = 160/256.0;
+    kernel[7][5] = 192/256.0;
+    kernel[7][6] = 224/256.0;
+    kernel[7][7] = 256/256.0;
+    kernel[7][8] = 224/256.0;
+    kernel[7][9] = 192/256.0;
+    kernel[7][10] = 160/256.0;
+    kernel[7][11] = 128/256.0;
+    kernel[7][12] = 96/256.0;
+    kernel[7][13] = 64/256.0;
+    kernel[7][14] = 16/256.0;
+
+    kernel[8][0] = 14/256.0;
+    kernel[8][1] = 56/256.0;
+    kernel[8][2] = 84/256.0;
+    kernel[8][3] = 112/256.0;
+    kernel[8][4] = 140/256.0;
+    kernel[8][5] = 168/256.0;
+    kernel[8][6] = 196/256.0;
+    kernel[8][7] = 224/256.0;
+    kernel[8][8] = 196/256.0;
+    kernel[8][9] = 168/256.0;
+    kernel[8][10] = 140/256.0;
+    kernel[8][11] = 112/256.0;
+    kernel[8][12] = 84/256.0;
+    kernel[8][13] = 56/256.0;
+    kernel[8][14] = 14/256.0;
+
+    kernel[9][0] = 12/256.0;
+    kernel[9][1] = 48/256.0;
+    kernel[9][2] = 72/256.0;
+    kernel[9][3] = 96/256.0;
+    kernel[9][4] = 120/256.0;
+    kernel[9][5] = 144/256.0;
+    kernel[9][6] = 168/256.0;
+    kernel[9][7] = 192/256.0;
+    kernel[9][8] = 168/256.0;
+    kernel[9][9] = 144/256.0;
+    kernel[9][10] = 120/256.0;
+    kernel[9][11] = 96/256.0;
+    kernel[9][12] = 72/256.0;
+    kernel[9][13] = 48/256.0;
+    kernel[9][14] = 12/256.0;
+
+    kernel[10][0] = 10/256.0;
+    kernel[10][1] = 40/256.0;
+    kernel[10][2] = 60/256.0;
+    kernel[10][3] = 80/256.0;
+    kernel[10][4] = 100/256.0;
+    kernel[10][5] = 120/256.0;
+    kernel[10][6] = 140/256.0;
+    kernel[10][7] = 160/256.0;
+    kernel[10][8] = 140/256.0;
+    kernel[10][9] = 120/256.0;
+    kernel[10][10] = 100/256.0;
+    kernel[10][11] = 80/256.0;
+    kernel[10][12] = 60/256.0;
+    kernel[10][13] = 40/256.0;
+    kernel[10][14] = 10/256.0;
+
+    kernel[11][0] = 8/256.0;
+    kernel[11][1] = 32/256.0;
+    kernel[11][2] = 48/256.0;
+    kernel[11][3] = 64/256.0;
+    kernel[11][4] = 80/256.0;
+    kernel[11][5] = 96/256.0;
+    kernel[11][6] = 112/256.0;
+    kernel[11][7] = 128/256.0;
+    kernel[11][8] = 112/256.0;
+    kernel[11][9] = 96/256.0;
+    kernel[11][10] = 80/256.0;
+    kernel[11][11] = 64/256.0;
+    kernel[11][12] = 48/256.0;
+    kernel[11][13] = 32/256.0;
+    kernel[11][14] = 8/256.0;
+
+    kernel[12][0] = 6/256.0;
+    kernel[12][1] = 24/256.0;
+    kernel[12][2] = 36/256.0;
+    kernel[12][3] = 48/256.0;
+    kernel[12][4] = 60/256.0;
+    kernel[12][5] = 72/256.0;
+    kernel[12][6] = 84/256.0;
+    kernel[12][7] = 96/256.0;
+    kernel[12][8] = 84/256.0;
+    kernel[12][9] = 72/256.0;
+    kernel[12][10] = 60/256.0;
+    kernel[12][11] = 48/256.0;
+    kernel[12][12] = 36/256.0;
+    kernel[12][13] = 24/256.0;
+    kernel[12][14] = 6/256.0;
+
+    kernel[13][0] = 4/256.0;
+    kernel[13][1] = 16/256.0;
+    kernel[13][2] = 24/256.0;
+    kernel[13][3] = 32/256.0;
+    kernel[13][4] = 40/256.0;
+    kernel[13][5] = 48/256.0;
+    kernel[13][6] = 56/256.0;
+    kernel[13][7] = 64/256.0;
+    kernel[13][8] = 56/256.0;
+    kernel[13][9] = 48/256.0;
+    kernel[13][10] = 40/256.0;
+    kernel[13][11] = 32/256.0;
+    kernel[13][12] = 24/256.0;
+    kernel[13][13] = 16/256.0;
+    kernel[13][14] = 4/256.0;
+
+    kernel[14][0] = 1/256.0;
+    kernel[14][1] = 4/256.0;
+    kernel[14][2] = 6/256.0;
+    kernel[14][3] = 8/256.0;
+    kernel[14][4] = 10/256.0;
+    kernel[14][5] = 12/256.0;
+    kernel[14][6] = 14/256.0;
+    kernel[14][7] = 16/256.0;
+    kernel[14][8] = 14/256.0;
+    kernel[14][9] = 12/256.0;
+    kernel[14][10] = 10/256.0;
+    kernel[14][11] = 8/256.0;
+    kernel[14][12] = 6/256.0;
+    kernel[14][13] = 4/256.0;
+    kernel[14][14] = 1/256.0;
+  }
+}
+
+public class Image {
+  int width, height;
+  int kernelWidth, kernelHeight;
+  double[][] inputImage;
+  double[][] outputImage;
+
+  public Image(int width, int height, int kernelWidth, int kernelHeight) {
+    this.width = width;
+    this.height = height;
+    this.kernelWidth = kernelWidth;
+    this.kernelHeight = kernelHeight;
+    inputImage = global new double[height+kernelHeight-1][width+kernelWidth-1];
+    outputImage = global new double[height][width];
+  }
+
+  /* Create a valid image */
+  public void setValues() {
+    for (int i = 0; i < (height+kernelHeight - 1); i++) {
+      double ainput[] = inputImage[i];
+      for(int j = 0; j < (width+kernelWidth - 1); j++) {
+        ainput[j] = 256-j;
+      }
+    }
+
+    for (int i = 0; i < height; i++){
+      double aout[] = outputImage[i];
+      for(int j = 0; j < width; j++) {
+        aout[j] = 0;
+      }
+    }
+  }
+}
diff --git a/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/Makefile b/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/Makefile
new file mode 100644
index 00000000..afe85a34
--- /dev/null
+++ b/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/Makefile
@@ -0,0 +1,10 @@
+MAINCLASS=Convolution
+SRC2=${MAINCLASS}SizeC.java
+FLAGS =-dsm -dsmcaching -rangeprefetch -nooptimize -debug -trueprob 0.95 -mainclass ${MAINCLASS}
+
+default:
+	../../../../buildscript ${FLAGS} ${SRC2} -o ${MAINCLASS}RangeN
+
+clean:
+	rm -rf tmpbuilddirectory
+	rm *.bin
diff --git a/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/README.txt b/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/README.txt
new file mode 100644
index 00000000..298e9a0b
--- /dev/null
+++ b/Robust/src/Benchmarks/Prefetch/ManualPrefetch/2DConv/README.txt
@@ -0,0 +1,8 @@
+Choosing where to insert manual prefetches
+===========================================
+
+Choice A:
+=========
+1. manual prefetch object this.img.inputImage[] the first 20 objects outside the computation
+   and every 15th iteration prefetch the next N=30 objects
+2. 
-- 
2.34.1