multithreading support

main.cpp

@@ -16,23 +16,67 @@
 #include <Magick++.h>
 #include <string.h>
 #include <sys/time.h>
+#include <inttypes.h>
+#include <semaphore>
+#include <barrier>
 #include <vector>
 #include <unordered_map>
+#include <mutex>
+#include <thread>
+#include <condition_variable>
 
 // which basic nail placement algorithms should be used
 // #define grid
-// #define multicircle
+#define multicircle
-#define circle
+// #define circle
 
-/// the actual weight function to calculate how off we are to the target
+class Main {
-/// \param value current value
+ private:
-/// \param target desired target
+  /// indicator, which threads are busy
-/// \retval distance to target
+  uint64_t m_busyFlag;
-inline int64_t weightFunction(int16_t value, int16_t target) {
-  return (value - target) * (value - target);
-}
 
-/// definition of a point for our dwarn line vector
+  /// number of threads
+  int16_t m_numThreads;
+
+  /// worker threads
+  std::vector<std::thread> m_worker;
+
+  /// result from checkLine thread
+  int64_t* m_lineResult;
+
+  /// next target to process in thread pool
+  int16_t m_nextTarget;
+
+  /// result pool lock
+  std::mutex m_resultLock;
+
+  /// mutex for work notification
+  std::mutex m_workMutex;
+
+  /// work notification condition
+  std::condition_variable m_workNotification;
+
+  /// mutex for finished motification
+  std::mutex m_finishedMutex;
+  /// finished notification condition
+  std::binary_semaphore m_finishedNotification{0};
+
+  /// sync point for worker threads and main
+  std::barrier<void(*)()> m_syncPoint;
+
+  /// last position (number of nail)
+  int16_t m_lastPosition = 0;
+
+  /// tasks left
+  int32_t m_tasksLeft;
+
+  /// running flag for threads
+  bool m_running;
+
+  /// list of nails
+  const std::vector<uint32_t>& m_nails;
+
+  /// definition of a point for our drawn line vector
   struct Point {
     /// x coordinate
     uint16_t x;
@@ -44,9 +88,39 @@ struct Point {
 
   /// typedef for a list of points tha make a line
   typedef std::vector<Point> td_pointsInLine;
+
   /// the line from src to dst, key = (src << 16) + dst
   typedef std::unordered_map<uint32_t, td_pointsInLine> td_linesFromSource;
 
+  /// our line storage
+  td_linesFromSource m_linesFromSource;
+
+  /// internal image width
+  uint16_t m_imgWidth;
+
+  /// current state of image
+  int16_t* m_currentState;
+
+  /// desired target state
+  uint8_t* m_targetState;
+
+  /// penalty duplication factor
+  float m_duplicateFactor;
+
+  /// map of used paths
+  uint16_t *m_usedpaths;
+
+  /// number of nails
+  int16_t m_numberOfNails;
+
+  /// the actual weight function to calculate how off we are to the target
+  /// \param value current value
+  /// \param target desired target
+  /// \retval distance to target
+  inline int64_t weightFunction(int16_t value, int16_t target) const {
+    return (value - target) * (value - target);
+  }
+
   /// swaps two numbers
   /// \param a first number
   /// \param b second number
@@ -65,6 +139,7 @@ inline float fPartOfNumber(float x) {
     }
     return x - (floor(x) + 1);
   }
+
   /// add given point to vector if col is > 0
   /// \param pil pointer to vector
   /// \param x x coordinate
@@ -134,140 +209,153 @@ td_pointsInLine drawAALine(int16_t x0 , int16_t y0 , int16_t x1 , int16_t y1, ui
     return pil;
   }
 
-/// \brief main entry point
+  float checkLine(int16_t target) const {
-/// \param argc number of command line arguments
+    if (m_lastPosition == target) return INT64_MAX;
-/// \param argv command line arguments
+    /// the diff on current lastPosition -> target
-int main(int argc, char* argv[]) {
+    int64_t testDiff = 0;
-  if (argc != 8) {
+    uint16_t src = std::min(m_lastPosition, target);
-    printf("usage: %s <image name> <resolution x> <resolution y> <number of nails> <max number of iterations> <penalty for duplicate path usage> <lineColor>\n", argv[0]);
+    uint16_t dst = std::max(m_lastPosition, target);
-    return 1;
+    td_linesFromSource::const_iterator lttIter = m_linesFromSource.find((src << 16) + dst);
+    // calculate difference to target
+    // for each point
+    td_pointsInLine::const_iterator pilIter = lttIter->second.begin();
+    while (pilIter != lttIter->second.end()) {
+      uint16_t x = (*pilIter).x;
+      uint16_t y = (*pilIter).y;
+      uint8_t sub = (*pilIter).color;
+      uint32_t index = y * m_imgWidth + x;
+      int16_t cur = m_currentState[index];
+      int16_t goal = m_targetState[index];
+      // subtract previous error
+      testDiff -= weightFunction(cur, goal);
+      cur -= sub;
+      // add new error
+      testDiff += weightFunction(cur, goal);
+      ++pilIter;
+    }
+    float duplicatePenalty = (m_duplicateFactor != 1) ?
+      pow(m_duplicateFactor, m_usedpaths[std::min(m_lastPosition, target)* m_numberOfNails + std::max(m_lastPosition, target)])
+      : 1;
+    return testDiff * duplicatePenalty;
   }
-  // copy command line data to easier to use variables
-  const char* imageName = argv[1];
-  uint16_t resolutionX = atoi(argv[2]);
-  uint16_t resolutionY = atoi(argv[3]);
-  uint16_t numberOfNails = atoi(argv[4]);
-  uint16_t maxIter = atoi(argv[5]);
-  float duplicateFactor = atof(argv[6]);
-  uint8_t lineColor = atoi(argv[7]);
 
-  // our line storage
+  void checkLines(int16_t tid) {
-  td_linesFromSource linesFromSource;
+    bool first = true;
+    while (m_running) {
+      m_syncPoint.arrive_and_wait();
+      // wait for notification
+      {
+        std::unique_lock<std::mutex> lock(m_workMutex);
+        m_busyFlag &= ~(1 << tid);
+        if (!m_busyFlag) {
+          if (!first) {
+            m_finishedNotification.release();
+          } else {
+            first = false;
+          }
+        }
+        m_workNotification.wait(lock);
+        m_busyFlag |= (1 << tid);
+      }
+      // did we wake up because of shutdown?
+      if (!m_running) break;
+      int16_t tasksLeft = 0;
+      int16_t target = 0;
+      do {  // as long as there is work
+        {
+          std::unique_lock<std::mutex> lock(m_resultLock);
+          target = m_nextTarget++;
+          tasksLeft = m_tasksLeft--;
+        }
+        if (target < m_numberOfNails) {
+          m_lineResult[target] = checkLine(target);
+        }
+      } while (tasksLeft > 0);
+    }
+  }
 
-  printf("res: %ix%i nails: %i maxIter: %i duplicatePenalty %.1f color: %i\n", resolutionX, resolutionY, numberOfNails, maxIter, duplicateFactor, lineColor);
+ public:
+  /// \brief constructor
+  /// \param nail vector with nail positions
+  /// \param duplicateFactor duplication penalty factor
+  Main(const std::vector<uint32_t>& nails, float duplicateFactor, int16_t numThreads) : m_syncPoint(numThreads + 1, [](){}), m_nails(nails) {
+    m_duplicateFactor = duplicateFactor;
+    m_numberOfNails = nails.size();
+    m_lineResult = reinterpret_cast<int64_t*>(malloc(sizeof(int64_t) * m_numberOfNails));
+    for (uint16_t i = 0; i < m_numberOfNails; ++i) {
+      m_lineResult[i] = INT64_MAX;
+    }
+    // a lookup of used paths to count repeats
+    m_usedpaths = reinterpret_cast<uint16_t*>(malloc(m_numberOfNails * m_numberOfNails * sizeof(uint16_t)));
+    bzero(m_usedpaths, m_numberOfNails * m_numberOfNails * sizeof(uint16_t));
+    m_nextTarget = 0;
+    m_imgWidth = 0;
+    m_running = true;
+    m_targetState = NULL;
+    m_currentState = NULL;
+    m_tasksLeft = 0;
+    m_numThreads = numThreads;
+
+    m_busyFlag = 0;
+    m_worker.reserve(numThreads);
+    for (uint16_t i = 0; i < numThreads; ++i) {
+      m_worker.push_back(std::thread(&Main::checkLines, this, i));
+    }
+    // wait until all threads are there
+    m_syncPoint.arrive_and_wait();
+  }
+
+  /// destructor
+  ~Main() {
+    m_running = false;
+    {
+      std::unique_lock<std::mutex> lock(m_workMutex);
+      m_workNotification.notify_all();
+    }
+    for (uint16_t i = 0; i < m_numThreads; ++i) {
+      m_worker[i].join();
+    }
+    free(m_lineResult);
+  }
+
+  /// \brief main function
+  /// \param resolutionX X resolution of internal image
+  /// \param resolutionY Y resolution of internal image
+  /// \param maxIter maximal number of iterations to run
+  /// \param lineColor line color to use
+  int run(const char* imageName, Magick::Image* img, uint16_t resolutionX, uint16_t resolutionY, int16_t requestedNumberOfNails, uint16_t maxIter, uint8_t lineColor) {
+    printf("res: %ix%i nails: %i maxIter: %i duplicatePenalty %.1f color: %i\n", resolutionX, resolutionY, m_numberOfNails, maxIter, m_duplicateFactor, lineColor);
 
-  /// nail positions (x << 16) + y
-  std::vector<uint32_t> nails;
     // for time measurement
     struct timeval tv1, tv2;
     gettimeofday(&tv1, NULL);
-  // initialize image magick, load image and resize to target coordinates
-  Magick::InitializeMagick(NULL);
-  Magick::Image img(imageName);
-  if (img.depth() != 8) {
-    printf("only 8 bit images supported\n");
-    return 1;
-  }
-  img.sample(Magick::Geometry(resolutionX, resolutionY));
 
-  // fix potential size differences between requested and delivered size
+    for (uint16_t src = 0; src < m_numberOfNails; ++src) {
-  uint16_t realWidth = img.columns();
+      for (uint16_t dst = src + 1; dst < m_numberOfNails; ++dst) {
-  uint16_t realHeight = img.rows();
+        td_pointsInLine pointsInLine = drawAALine(m_nails[src] >> 16, m_nails[src] & 0xFFFF, m_nails[dst] >> 16, m_nails[dst] & 0xFFFF, lineColor);
-
+        m_linesFromSource.insert(std::make_pair((src << 16) + dst, pointsInLine));
-  // position nails
-#ifdef circle
-  for (uint16_t i = 0; i < numberOfNails; ++i) {
-    float x = sin(2.0 * M_PI * i / numberOfNails) * (realWidth-1) / 2.0 + realWidth / 2.0;
-    float y = cos(2.0 * M_PI * i / numberOfNails) * (realHeight-1) / 2.0 + realHeight / 2.0;
-    nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
-  }
-#endif
-#ifdef multicircle
-  uint16_t count = numberOfNails/1.5;
-  for (uint16_t i = 0; i < count; ++i) {
-    float x = sin(2.0 * M_PI * i / count) * (realWidth-1) / 2.0 + realWidth / 2.0;
-    float y = cos(2.0 * M_PI * i / count) * (realHeight-1) / 2.0 + realHeight / 2.0;
-    nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
-  }
-  uint16_t width = realWidth/1.2 -1;
-  uint16_t height = realHeight/1.2 -1;
-  count = numberOfNails/1.5;
-  for (uint16_t i = 0; i < count; ++i) {
-      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + realWidth / 2.0;
-      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + realHeight / 2.0;
-      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
-  }
-  width = realWidth/1.5 -1;
-  height = realHeight/1.5 -1;
-  count = numberOfNails/2;
-  for (uint16_t i = 0; i < count; ++i) {
-      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + realWidth / 2.0;
-      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + realHeight / 2.0;
-      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
-  }
-  width = realWidth/2 -1;
-  height = realHeight/2 -1;
-  count = numberOfNails/3;
-  for (uint16_t i = 0; i < count; ++i) {
-      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + realWidth / 2.0;
-      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + realHeight / 2.0;
-      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
-  }
-  width = realWidth/3 -1;
-  height = realHeight/3 -1;
-  count = numberOfNails/4;
-  for (uint16_t i = 0; i < count; ++i) {
-      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + realWidth / 2.0;
-      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + realHeight / 2.0;
-      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
-  }
-  width = realWidth/5 -1;
-  height = realHeight/5 -1;
-  count = numberOfNails/6;
-  for (uint16_t i = 0; i < count; ++i) {
-      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + realWidth / 2.0;
-      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + realHeight / 2.0;
-      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
-  }
-  nails.push_back((static_cast<uint32_t>(floor(realWidth / 2.0)) << 16) +  static_cast<uint16_t>(floor(realHeight / 2.0)));
-#endif
-#ifdef grid
-  uint8_t sq_pins = sqrt(numberOfNails);
-  float distX = static_cast<float>(realWidth - 1) / (sq_pins - 1);
-  float distY = static_cast<float>(realHeight - 1) / (sq_pins - 1);
-  for (uint16_t y = 0; y < sq_pins; ++y) {
-    for (uint16_t x = 0; x < sq_pins; ++x) {
-      nails.push_back((static_cast<uint32_t>(floor(distX * x)) << 16) +  static_cast<uint16_t>(floor(distY * y)));
-    }
-  }
-#endif
-  // number of nails might have been changed above or loaded (in the future)
-  numberOfNails = nails.size();
-  printf("num %i\n", numberOfNails);
-
-  for (uint16_t src = 0; src < numberOfNails; ++src) {
-    for (uint16_t dst = src + 1; dst < numberOfNails; ++dst) {
-      td_pointsInLine pointsInLine = drawAALine(nails[src] >> 16, nails[src] & 0xFFFF, nails[dst] >> 16, nails[dst] & 0xFFFF, lineColor);
-      linesFromSource.insert(std::make_pair((src << 16) + dst, pointsInLine));
       }
     }
 
-  uint32_t channels = img.channels();
+    uint32_t channels = img->channels();
-  printf("target image %i x %i x %i\n", realWidth, realHeight, channels);
+    m_imgWidth = img->columns();
-  MagickCore::Quantum *pixels = img.getPixels(0, 0, realWidth, realHeight);
+    uint16_t imgHeight = img->rows();
-  uint8_t* targetState = reinterpret_cast<uint8_t*>(malloc(realWidth * realHeight));
+
+    printf("target image %i x %i x %i\n", m_imgWidth, imgHeight, channels);
+    MagickCore::Quantum *pixels = img->getPixels(0, 0, m_imgWidth, imgHeight);
+    m_targetState = reinterpret_cast<uint8_t*>(malloc(m_imgWidth * imgHeight));
 
     if (channels == 1) {  // monochrome image
-    for (uint32_t i = 0; i < realWidth * realHeight; ++i) {
+      for (uint32_t i = 0; i < m_imgWidth * imgHeight; ++i) {
-      targetState[i] = pixels[i] >> 8;
+        m_targetState[i] = pixels[i] >> 8;
       }
     } else if (channels == 2) {  // color + alpha?
-    for (uint32_t i = 0; i < realWidth * realHeight; ++i) {
+      for (uint32_t i = 0; i < m_imgWidth * imgHeight; ++i) {
-      targetState[i] = pixels[i << 1] >> 8;
+        m_targetState[i] = pixels[i << 1] >> 8;
       }
     } else {  // RGB or RGBA
-    for (uint32_t i = 0; i < realWidth * realHeight; ++i) {
+      for (uint32_t i = 0; i < m_imgWidth * imgHeight; ++i) {
-      targetState[i] = ((pixels[i*channels] + pixels[i*channels + 1] + pixels[i*channels + 2]) / 3) >> 8;
+       m_targetState[i] = ((pixels[i*channels] + pixels[i*channels + 1] + pixels[i*channels + 2]) / 3) >> 8;
       }
     }
 #ifdef DEBUGIMG
@@ -281,27 +369,21 @@ int main(int argc, char* argv[]) {
     std::vector<uint16_t> path;
     // add start position
     path.push_back(0);
-  // a lookup of used paths to count repeats
-  uint16_t usedpaths[numberOfNails][numberOfNails];
-  bzero(usedpaths, numberOfNails * numberOfNails * 2);
 
     /// last thread end position
-  int16_t lastPosition = 0;
+    m_lastPosition = 0;
     /// storage for the current state (all previous drawn threads)
-  int16_t* currentState = reinterpret_cast<int16_t*>(malloc(realWidth * realHeight * 2));
+    m_currentState = reinterpret_cast<int16_t*>(malloc(m_imgWidth * imgHeight * 2));
-  /// temp storage to save (current) best version, will be continously updated
-  int16_t* bestState = reinterpret_cast<int16_t*>(malloc(realWidth * realHeight * 2));
 
     // clear states
-  uint32_t widthXheight = realWidth * realHeight;
+    uint32_t widthXheight = m_imgWidth * imgHeight;
     for (uint32_t i = 0; i < widthXheight; ++i) {
-    currentState[i] = 255;
+      m_currentState[i] = 255;
-    bestState[i] = 255;
     }
     // current iteration
     uint32_t iter = 0;
     // list of used nails with their counter
-  uint8_t usedPins[numberOfNails] = {0};
+    uint8_t usedPins[m_numberOfNails] = {0};
     // number of continous jump tries if we got stuck
     uint16_t jumps = 0;
     /// total diff from currentState to targetState
@@ -309,10 +391,10 @@ int main(int argc, char* argv[]) {
 
     // calculate inital difference
     for (uint32_t i = 0; i < widthXheight; ++i) {
-    totalDiff += weightFunction(255, targetState[i]);
+      totalDiff += weightFunction(255, m_targetState[i]);
     }
-  printf("start %li\n", totalDiff);
+    printf("start diff %li\n", totalDiff);
-  while ((iter < maxIter) && jumps*2 < numberOfNails) {
+    while ((iter < maxIter) && jumps*2 < m_numberOfNails) {
       ++iter;
 #ifdef SANITYCHECK
       int64_t sanity = 0;
@@ -326,90 +408,56 @@ int main(int argc, char* argv[]) {
       }
 #endif
       /// current best difference
-    int64_t realBestDiff = INT64_MAX;
+      int64_t bestDiff = INT64_MAX;
-    /// compensated diff includes penality when reusing paths
-    int64_t compensatedBestDiff = INT64_MAX;
       int16_t bestTarget = -1;
-    // printf("source %i\n", lastPosition); fflush(stdout);
+      {
-    for (int16_t target = 0; target < numberOfNails; ++target) {
+        std::unique_lock<std::mutex> lock(m_resultLock);
-      if (target == lastPosition) continue;
+        m_nextTarget = 0;
-      /// the diff on current lastPosition -> target
+        m_tasksLeft = m_numberOfNails;
-      int64_t testDiff = 0;
-      uint16_t src = std::min(lastPosition, target);
-      uint16_t dst = std::max(lastPosition, target);
-      td_linesFromSource::const_iterator lttIter = linesFromSource.find((src << 16) + dst);
-      // calculate difference to target
-      // for each point
-      td_pointsInLine::const_iterator pilIter = lttIter->second.begin();
-      while (pilIter != lttIter->second.end()) {
-        uint16_t x = (*pilIter).x;
-        uint16_t y = (*pilIter).y;
-        uint8_t sub = (*pilIter).color;
-        uint32_t index = y * realWidth + x;
-        int16_t cur = currentState[index];
-        int16_t goal = targetState[index];
-        // subtract previous error
-        testDiff -= weightFunction(cur, goal);
-        cur -= sub;
-        // add new error
-        testDiff += weightFunction(cur, goal);
-        ++pilIter;
       }
-      float duplicatePenalty = (duplicateFactor != 1) ?
+      // notify threads
-          pow(duplicateFactor, usedpaths[std::min(lastPosition, target)][std::max(lastPosition, target)])
+      {
-          : 1;
+        std::unique_lock<std::mutex> lock(m_workMutex);
-      if ((testDiff / duplicatePenalty) < compensatedBestDiff) {
+        m_workNotification.notify_all();
-        // printf("    new best %i - %i(%i,%i) %li d %li\n", lastPosition, target, pos[target]>>16, pos[target]&0xffff, testDiff, totalDiff - testDiff);
+      }
-        compensatedBestDiff = testDiff * duplicatePenalty;
+      m_syncPoint.arrive_and_wait();
-        realBestDiff = testDiff;
+      // wait for threads, results are in m_lineResults
-        // a new best
+      m_finishedNotification.acquire();
-        if (bestTarget != -1) {
+
-          // printf("undo %i:%i\n", lastPosition, bestTarget);
+      // search best result
-          // undo previous best
+      for (int16_t target = 0; target < m_numberOfNails; ++target) {
-          uint16_t tmpSrc = std::min(lastPosition, bestTarget);
+        if (target == m_lastPosition) continue;
-          uint16_t tmpDst = std::max(lastPosition, bestTarget);
+        if (m_lineResult[target] < bestDiff) {
-          td_linesFromSource::const_iterator tmpLfsIter = linesFromSource.find((tmpSrc << 16) + tmpDst);
+          bestTarget = target;
-          td_pointsInLine::const_iterator pilIter = tmpLfsIter->second.begin();
+          bestDiff = m_lineResult[target];
-          while (pilIter != tmpLfsIter->second.end()) {
-            uint16_t x = (*pilIter).x;
-            uint16_t y = (*pilIter).y;
-            uint32_t index = y * realWidth + x;
-            int16_t cur = currentState[index];
-            bestState[index] = cur;
-            ++pilIter;
         }
       }
+      if (bestDiff < 0) {
         // apply current best
+        td_linesFromSource::const_iterator lttIter = m_linesFromSource.find((std::min(m_lastPosition, bestTarget) << 16) + std::max(m_lastPosition, bestTarget));
         td_pointsInLine::const_iterator pilIter = lttIter->second.begin();
         while (pilIter != lttIter->second.end()) {
           uint16_t x = (*pilIter).x;
           uint16_t y = (*pilIter).y;
           int16_t sub = (*pilIter).color;
-          uint32_t index = y * realWidth + x;
+          uint32_t index = y * m_imgWidth + x;
-          int16_t cur = currentState[index];
+          m_currentState[index] -= sub;
-          cur -= sub;
-          bestState[index] = cur;
           ++pilIter;
         }
-        bestTarget = target;
+      } else {
-      }
-    }
-    if (realBestDiff >= 0) {
         // we got worse, jump to random place to continue
-      // printf("j %3i %3i -> %3i(%4i, %4i) bestDiff %8li (%li) iter %i path %i\n", jumps, lastPosition, bestTarget, pos[bestTarget] >> 16, pos[bestTarget] & 0xFFFF, realBestDiff, totalDiff - realBestDiff, iter, usedpaths[std::min(lastPosition, bestTarget)][std::max(lastPosition, bestTarget)]);
+        // printf("j %3i %3i -> %3i(%4i, %4i) bestDiff %8li (%li) iter %i path %i\n", jumps, lastPosition, bestTarget, pos[bestTarget] >> 16, pos[bestTarget] & 0xFFFF, realBestDiff, totalDiff - realBestDiff, iter, m_usedpaths[std::min(m_lastPosition, bestTarget)* m_numberOfNails + std::max(m_lastPosition, bestTarget)]]);
         if (jumps) {  // undo last jump, was not working anyway
-        --usedPins[lastPosition];
+          --usedPins[m_lastPosition];
           path.pop_back();
         }
         // select next target randomly (kind of, intentially producing the same numbers)
-      bestTarget = random() % numberOfNails;
+        bestTarget = random() % m_numberOfNails;
         path.push_back(bestTarget);
-      lastPosition = bestTarget;
+        m_lastPosition = bestTarget;
         ++usedPins[bestTarget];
         ++jumps;
         --iter;
-      // fix bestState, easier to copy over than manually reversing. should only happen a few times anyway
-      memcpy(bestState, currentState, widthXheight * 2);
         continue;
       }
       /// reset jump counter (faster to always set)
@@ -419,33 +467,32 @@ int main(int argc, char* argv[]) {
         break;
       }
       // update path map
-    ++usedpaths[std::min(lastPosition, bestTarget)][std::max(lastPosition, bestTarget)];
+      ++m_usedpaths[std::min(m_lastPosition, bestTarget)* m_numberOfNails + std::max(m_lastPosition, bestTarget)];
       // add new stop
       path.push_back(bestTarget);
       // update used pins
       ++usedPins[bestTarget];
+      // update diff
+      totalDiff += bestDiff;
       // progress report
       if (iter % 100 == 0) {
-      printf("best %4i -> %4i(%4i, %4i) diff %9li (%12li) iter %5i path %i\n", lastPosition, bestTarget, nails[bestTarget] >> 16, nails[bestTarget] & 0xFFFF, compensatedBestDiff, totalDiff + realBestDiff, iter, usedpaths[std::min(lastPosition, bestTarget)][std::max(lastPosition, bestTarget)]);
+        printf("best %4i -> %4i(%4i, %4i) diff %12li iter %5i path %i\n", m_lastPosition, bestTarget, m_nails[bestTarget] >> 16, m_nails[bestTarget] & 0xFFFF,
+            totalDiff, iter, m_usedpaths[std::min(m_lastPosition, bestTarget) * m_numberOfNails + std::max(m_lastPosition, bestTarget)]);
       }
       // set new start position
-    lastPosition = bestTarget;
+      m_lastPosition = bestTarget;
-    // update diff
-    totalDiff += realBestDiff;
       // update current state from best map
-    memcpy(currentState, bestState, widthXheight * 2);
     }
-
     printf("size %li\n", path.size());
     // we are done, create output svg
     FILE* fh = fopen("map.svg", "wb");
-  fprintf(fh, "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<svg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 %i %i\">\n<rect width=\"%i\" height=\"%i\" fill=\"#ffffff\" />\n<g style=\"fill:none;stroke:#000000;stroke-opacity:%.2f;stroke-width:1\">\n", realWidth, realHeight, realWidth, realHeight, lineColor / 255.0);
+    fprintf(fh, "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<svg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 %i %i\">\n<rect width=\"%i\" height=\"%i\" fill=\"#ffffff\" />\n<g style=\"fill:none;stroke:#000000;stroke-opacity:%.2f;stroke-width:1\">\n", m_imgWidth, imgHeight, m_imgWidth, imgHeight, lineColor / 255.0);
     uint32_t counter = 0;
     for (uint16_t i : path) {
       if ((counter & 255) == 0) {
-      fprintf(fh, "<path d=\"M%i %i", nails[i] >> 16, nails[i] & 0xffff);
+        fprintf(fh, "<path d=\"M%i %i", m_nails[i] >> 16, m_nails[i] & 0xffff);
       } else {
-      fprintf(fh, "L%i %i", nails[i] >> 16, nails[i] & 0xffff);
+        fprintf(fh, "L%i %i", m_nails[i] >> 16, m_nails[i] & 0xffff);
       }
       if ((counter & 255) == 255) {
         fprintf(fh, "\" />\n");
@@ -458,14 +505,122 @@ int main(int argc, char* argv[]) {
     gettimeofday(&tv2, NULL);
     float timeNeeded = (tv2.tv_sec - tv1.tv_sec) + (tv2.tv_usec - tv1.tv_usec) / 1000000.0;
     fprintf(fh, "</g>\n");
-  fprintf(fh, "<text x=\"30\" y=\"10\" style=\"font-weight:bold;font-size:60px;font-family:'DejaVu Serif'\"><tspan x=\"30\" y=\"70\">%s %s %i %i</tspan><tspan x=\"70\" y=\"150\"> %i %i %.2f %i</tspan><tspan x=\"30\" y=\"%i\">%i nails, %li paths, %.1f sec</tspan></text>", argv[0], imageName, resolutionX, resolutionY, atoi(argv[4]), maxIter, duplicateFactor, lineColor, realHeight - 30, numberOfNails, path.size(), timeNeeded);
+    fprintf(fh, "<text x=\"30\" y=\"10\" style=\"font-weight:bold;font-size:60px;font-family:'DejaVu Serif'\"><tspan x=\"30\" y=\"70\">%s %i %i</tspan><tspan x=\"70\" y=\"150\"> %i %i %.2f %i</tspan><tspan x=\"30\" y=\"%i\">%i nails, %li paths, %.1f sec</tspan></text>", imageName, resolutionX, resolutionY, requestedNumberOfNails, maxIter, m_duplicateFactor, lineColor, imgHeight - 30, m_numberOfNails, path.size(), timeNeeded);
     fprintf(fh, "</svg>");
     fclose(fh);
     // cleanup
-  free(targetState);
+    free(m_targetState);
-  free(bestState);
+    free(m_currentState);
-  free(currentState);
+    free(m_usedpaths);
     path.clear();
-  linesFromSource.clear();
+    m_linesFromSource.clear();
+    return 0;
+  }
+};
+/// \brief main entry point
+/// \param argc number of command line arguments
+/// \param argv command line arguments
+int main(int argc, char* argv[]) {
+  if (argc != 8) {
+    printf("usage: %s <image name> <resolution x> <resolution y> <number of nails> <max number of iterations> <penalty for duplicate path usage> <lineColor>\n", argv[0]);
+    return 1;
+  }
+  // copy command line data to easier to use variables
+  const char* imageName = argv[1];
+  uint16_t resolutionX = atoi(argv[2]);
+  uint16_t resolutionY = atoi(argv[3]);
+  uint16_t requestedNumberOfNails = atoi(argv[4]);
+  uint16_t maxIter = atoi(argv[5]);
+  float duplicateFactor = atof(argv[6]);
+  uint8_t lineColor = atoi(argv[7]);
+
+  /// nail positions (x << 16) + y
+  std::vector<uint32_t> nails;
+
+  // initialize image magick, load image and resize to target coordinates
+  Magick::InitializeMagick(NULL);
+  Magick::Image img(imageName);
+  if (img.depth() != 8) {
+    printf("only 8 bit images supported\n");
+    return 1;
+  }
+  img.sample(Magick::Geometry(resolutionX, resolutionY));
+
+  // fix potential size differences between requested and delivered size
+  uint16_t imgWidth = img.columns();
+  uint16_t imgHeight = img.rows();
+
+
+  // position nails
+#ifdef circle
+    for (uint16_t i = 0; i < requestedNumberOfNails; ++i) {
+      float x = sin(2.0 * M_PI * i / requestedNumberOfNails) * (imgWidth-1) / 2.0 + imgWidth / 2.0;
+      float y = cos(2.0 * M_PI * i / requestedNumberOfNails) * (imgHeight-1) / 2.0 + imgHeight / 2.0;
+      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
+    }
+#endif
+#ifdef multicircle
+    uint16_t count = requestedNumberOfNails/1.5;
+    for (uint16_t i = 0; i < count; ++i) {
+      float x = sin(2.0 * M_PI * i / count) * (imgWidth-1) / 2.0 + imgWidth / 2.0;
+      float y = cos(2.0 * M_PI * i / count) * (imgHeight-1) / 2.0 + imgHeight / 2.0;
+      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
+    }
+    uint16_t width = imgWidth / 1.2 - 1;
+    uint16_t height = imgHeight / 1.2 - 1;
+    count = requestedNumberOfNails / 1.5;
+    for (uint16_t i = 0; i < count; ++i) {
+      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + imgWidth / 2.0;
+      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + imgHeight / 2.0;
+      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
+    }
+    width = imgWidth / 1.5 - 1;
+    height = imgHeight / 1.5 - 1;
+    count = requestedNumberOfNails / 2;
+    for (uint16_t i = 0; i < count; ++i) {
+      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + imgWidth / 2.0;
+      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + imgHeight / 2.0;
+      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
+    }
+    width = imgWidth / 2 - 1;
+    height = imgHeight / 2 - 1;
+    count = requestedNumberOfNails / 3;
+    for (uint16_t i = 0; i < count; ++i) {
+      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + imgWidth / 2.0;
+      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + imgHeight / 2.0;
+      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
+    }
+    width = imgWidth / 3 - 1;
+    height = imgHeight / 3 - 1;
+    count = requestedNumberOfNails / 4;
+    for (uint16_t i = 0; i < count; ++i) {
+      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + imgWidth / 2.0;
+      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + imgHeight / 2.0;
+      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
+    }
+    width = imgWidth / 5 - 1;
+    height = imgHeight / 5 - 1;
+    count = requestedNumberOfNails / 6;
+    for (uint16_t i = 0; i < count; ++i) {
+      float x = sin(2.0 * M_PI * i / count) * (width-1) / 2.0 + imgWidth / 2.0;
+      float y = cos(2.0 * M_PI * i / count) * (height-1) / 2.0 + imgHeight / 2.0;
+      nails.push_back((static_cast<uint32_t>(floor(x)) << 16) +  static_cast<uint16_t>(floor(y)));
+  }
+  nails.push_back((static_cast<uint32_t>(floor(imgWidth / 2.0)) << 16) +  static_cast<uint16_t>(floor(imgHeight / 2.0)));
+#endif
+#ifdef grid
+  uint8_t sq_pins = sqrt(requestedNumberOfNails);
+  float distX = static_cast<float>(imgWidth - 1) / (sq_pins - 1);
+  float distY = static_cast<float>(imgHeight - 1) / (sq_pins - 1);
+  for (uint16_t y = 0; y < sq_pins; ++y) {
+    for (uint16_t x = 0; x < sq_pins; ++x) {
+      nails.push_back((static_cast<uint32_t>(floor(distX * x)) << 16) +  static_cast<uint16_t>(floor(distY * y)));
+    }
+  }
+#endif
+
+
+  Main m(nails, duplicateFactor, std::min(std::thread::hardware_concurrency(), 64U));
+  m.run(imageName, &img, resolutionX, resolutionY, requestedNumberOfNails, maxIter, lineColor);
   Magick::TerminateMagick();
 }