Parallelized the CALIC implementation

refs/calic/AdaptiveModel.cc

@@ -1,5 +1,6 @@
 #include "Arithm.h"
 #include <stdio.h>
+#include <string.h>
 
 AdaptiveModel::AdaptiveModel() {
   numb_symb = 0;

refs/calic/Arithm.h

@@ -3,7 +3,7 @@
 
 #include <stdlib.h>
 #include <stdio.h>
-
+#include <omp.h>
 #define MaxSymbols 257
 
 #define CodeValueBits   16

refs/calic/ArithmeticDecoder.cc

@@ -56,15 +56,21 @@ void ArithmeticDecoder::Input_Byte() {
 }
 
 void ArithmeticDecoder::update() {
-  for (;;) {
-    if (high >= Half) {
-      if (low >= Half) {
+
+  for (;;) 
+  {
+    if (high >= Half) 
+    {
+      if (low >= Half) 
+      {
         value -= Half;  
         low -= Half;  
         high -= Half; 
       }
-      else {
-        if ((low >= FirstQtr) && (high < ThirdQtr)) {
+      else 
+      {
+        if ((low >= FirstQtr) && (high < ThirdQtr)) 
+        {
           value -= FirstQtr;  low -= FirstQtr;
           high -= FirstQtr; 
         }

refs/calic/BinaryModeModel.cc

@@ -4,41 +4,7 @@
 void BinaryModeModel::update(int index) {
 //  if (incr == 1)
     AdaptiveModel::update(index); 
-/*
-  else {
-    int i, cum, symb, new_freq;
 
-    if (cum_freq[0] >= MaxFrequency) {
-      for (cum = 0, i = numb_symb; i >= 0; i--) {
-        freq[i] = (freq[i] + 1) >> 1;
-        cum_freq[i] = cum;  
-        cum += freq[i]; 
-      }
-      incr = (incr + 1) / 2;
-    }
-
-    new_freq = freq[index] + incr;
-    symb = index_to_symb[index];
-
-    int tmp_idx, tmp_symb;
-    for (i = index; new_freq >= freq[i-1] && i > 1; i--) {
-      freq[i] = freq[i-1];
-      tmp_symb = index_to_symb[i-1];
-      index_to_symb[i] = tmp_symb;
-      symb_to_index[tmp_symb] = i;
-    }
-
-    freq[i] = new_freq;
-
-    if (i < index) {
-      index_to_symb[i] = symb;
-      symb_to_index[symb] = i;
-    }
-
-    while (i) 
-      cum_freq[--i] += incr;
-  }
-*/
 }
 
 void BinaryModeModel::init(int ns) {
@@ -52,28 +18,50 @@ void ContinuousModeModel::init(int ns) {
 
   int i, cum;
 
-  int freqCount = (MaxFrequency * 2 / 4);
+  int freqCount = (MaxFrequency / 2);
   double *freqPercentage = new double[ns];
 
   double sum = 0.0;
-  for (int i = 0; i < ns; i++) {
-    freqPercentage[i] = pow(15.0, 0.4 * (double)(-i-1));
-    sum += freqPercentage[i];
+  double temp;
+  #pragma omp parallel for private(temp) reduction(+:sum)
+  for (int i = 0; i < ns; i++) 
+  {
+    //prevent false sharing and double indirect addressing overhead using a `temp`orary variable
+    temp = pow(15.0, 0.4 * (double)(-i-1));  
+    freqPercentage[i] = temp;  
+    sum += temp; 
+    //temp avoids data dependency issues instead of freqPercentage
   }
 
-  for (int i = 0; i < ns; i++) {
-    freqPercentage[i] /= sum;
+  #pragma omp parallel for shared(freqPercentage)
+  for (int i = 0; i < ns; i++) 
+  {
+    //Get and store probabiliity
+      freqPercentage[i] /= sum;
   }
 
-  for (i = 1; i <= ns; i++) {
+  for (i = 1; i <= ns; i++) 
+  {
+
+     //Absolute frequency
     freq[i] = (int)((double)freqCount * freqPercentage[i-1]);
     if (freq[i] == 0)
       freq[i] = 1;
   }
-
-  for (cum = 0, i = ns; i >= 0; i--) {
-    cum_freq[i] = cum;
-    cum += freq[i];
+
+  //Cumulative Frequency => loop dependencies => prefix sum
+cum_freq[ns] = 0;
+ int j;
+  #pragma omp parallel for schedule(dynamic)
+  for ( i = ns - 1; i >= 0; i--) 
+  {   
+    cum_freq[i] = 0;
+    //#pragma omp parallel for reduction is an overhead
+    for(j=ns;j>=i;j--)
+    {
+        cum_freq[i] += freq[j];
+    }
+
   }
 
   delete [] freqPercentage;

refs/calic/CommandLine.cc

@@ -32,8 +32,7 @@ CommandLine::~CommandLine() {
 //   Otherwise it returns False.
 //
 //////////////////////////////////////////////////////////////////////
-bool CommandLine::getOption(char *str) {
-
+bool CommandLine::getOption(char *str) {  
   for (int i = 1; i < argc; i++)
     if (argv[i][0] == '-' && (strcmp(argv[i] + 1, str) == 0)) 
       return true;
@@ -90,7 +89,7 @@ bool CommandLine::getParameter(char *str, float *var) {
 bool CommandLine::getParameter(char *str, double *var) {
 
   for (int i = 1; i < argc - 1; i++)
-    if (argv[i][0] == '-' && (strcmp(argv[i] + 1, str) == 0)) {
+    if (argv[i][0] == '-' && (strcmp(argv[i] + 1, str) == 0)) { //false sharing if parallelized
       *var = (double) atof(argv[i+1]);
       return true;
     }
@@ -113,6 +112,7 @@ void CommandLine::setArguements(int _argc, char *_argv[]) {
     argc = _argc;
 
     argv = new char * [argc];
+    #pragma omp  parallel for
     for (int i = 0; i < argc; i++) {
        argv[i] = new char [MAX_STR_LEN];
        strcpy(argv[i], _argv[i]);

refs/calic/CommandLine.h

@@ -15,6 +15,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <omp.h>
 
 #ifndef MAX_STR_LEN
 #define MAX_STR_LEN 64

refs/calic/ImageCodec.h

@@ -5,6 +5,7 @@
 #define __IMAGECODEC_H__
 
 #include <pnm.h>
+#include <string.h>
 #include "Arithm.h"
 #include "Image.h"
 

refs/calic/Makefile

@@ -1,6 +1,6 @@
 ARCH = $(shell uname)
 
-CC = g++ -Wall
+CC = g++ -Wall -fopenmp
 
 PROGRAMS = encode decode 
 
@@ -30,7 +30,7 @@ OBJS = $(ARITHM_OBJS) $(PNM_OBJS) $(MISC_OBJS) $(IMAGECODEC_OBJS)
 
 INCDIR = .
 
-CFLAGS = -O2 -I$(INCDIR) -D$(ARCH)
+CFLAGS = -O2 -I$(INCDIR) -D$(ARCH) -fopenmp
 
 .SUFFIXES: .cc .
 .cc.o:

refs/calic/StatisticalModel.cc

@@ -2,7 +2,10 @@
 
 void StatisticalModel::reset() {
   for (int i = 0; i < size; i++)
-    model[i] = modelCount[i] = 0;
+   {
+    model[i]  = 0;
+    modelCount[i] = 0;
+   } 
 }
 
 int StatisticalModel::write(char *filename) {

refs/calic/encode.cc

@@ -57,9 +57,9 @@ int main(int argc, char *argv[]) {
 
   if (debugLevel > 1) {
     ShortImage * errImg = codec->getErrorImage();
-    int w = errImg->w;;
+    int w = errImg->w;
     int h = errImg->h;
-
+    
     FILE *ofp;
 
     char errorFilename[256];
@@ -71,9 +71,10 @@ int main(int argc, char *argv[]) {
     else {
       PGMImage *img = new PGMImage(w, h);
 
-      for (int y = 0; y < h; y++)
-        for (int x = 0; x < w; x++) {
-          fprintf(ofp, "%d %d   %d\n", x, y, errImg->getPixel(x, y));
+      #pragma omp parallel for
+      for (int y = 0; y < h; y++) 
+        for (int x = 0; x < w; x++) { //overhead in parallelizing
+          fprintf(ofp, "%d %d   %d\n", x, y, errImg->getPixel(x, y)); //thread safe
           img->setPixel(x, y, (errImg->getPixel(x, y) < 0) ? 0 : 255);
         }
 

refs/calic/pnm.cc

@@ -101,7 +101,6 @@ int PPMImage::write(char *filename) {
   }
 
   fprintf(fp, "P6\n%d %d\n%d\n", width, height, maxIntensity);
-
   for (int i = 0; i < height; i++) {
     for (int j = 0; j < width; j++) {
       tmp = pixels[i*width + j] << 8;
@@ -184,6 +183,7 @@ PPMImage *PPMImage::copyArea(int x, int y, int w, int h) {
   register int i, j, imgsize;
 
   imgsize = w * h;
+  #pragma omp parallel for
   for (i = 0; i < imgsize; i++) 
     newimg->pixels[i] = 0;
 

refs/calic/pnm.h

@@ -3,7 +3,7 @@
 
 #include <stdio.h>
 #include <stdlib.h>
-
+#include <omp.h>
 typedef unsigned char byte;
 typedef unsigned short int16;