Log more performance data, plot also now many threads are running

src/encoder.c

@@ -1685,7 +1685,6 @@ static void encoder_state_write_bitstream_main(encoder_state * const main_state)
 
   int i;
 
-
   if (main_state->global->is_radl_frame) {
     // Access Unit Delimiter (AUD)
     if (encoder->aud_enable)
@@ -1726,16 +1725,23 @@ static void encoder_state_write_bitstream_main(encoder_state * const main_state)
     nal_write(stream,
               main_state->global->is_radl_frame ? NAL_IDR_W_RADL : NAL_TRAIL_R, 0, long_start_code);
   }
-  
+  {
+    PERFORMANCE_MEASURE_START();
   for (i = 0; main_state->children[i].encoder_control; ++i) {
     //Append bitstream to main stream
     bitstream_append(&main_state->stream, &main_state->children[i].stream);
     //FIXME: Move this...
     bitstream_clear(&main_state->children[i].stream);
   }
+    PERFORMANCE_MEASURE_END(main_state->encoder_control->threadqueue, "type=write_bitstream_append,frame=%d,type=%c", main_state->global->frame, main_state->type);
+  }
   
-  // Calculate checksum
-  add_checksum(main_state);
+  {
+    PERFORMANCE_MEASURE_START();
+    // Calculate checksum
+    add_checksum(main_state);
+    PERFORMANCE_MEASURE_END(main_state->encoder_control->threadqueue, "type=write_bitstream_checksum,frame=%d,type=%c", main_state->global->frame, main_state->type);
+  }
   
   //FIXME: Why is this needed?
   main_state->tile->cur_pic->poc = main_state->global->poc;
@@ -1817,9 +1823,19 @@ static void encoder_state_write_bitstream(encoder_state * const main_state) {
 
 void encode_one_frame(encoder_state * const main_state)
 {
-  encoder_state_new_frame(main_state);
-  encoder_state_encode(main_state);
-  encoder_state_write_bitstream(main_state);
+  {
+    PERFORMANCE_MEASURE_START();
+    encoder_state_new_frame(main_state);
+    PERFORMANCE_MEASURE_END(main_state->encoder_control->threadqueue, "type=new_frame,frame=%d", main_state->global->frame);
+  }
+  {
+    PERFORMANCE_MEASURE_START();
+    encoder_state_encode(main_state);
+    PERFORMANCE_MEASURE_END(main_state->encoder_control->threadqueue, "type=encode,frame=%d", main_state->global->frame);
+  }
+  {
+    encoder_state_write_bitstream(main_state);
+  }
 }
 
 static void fill_after_frame(unsigned height, unsigned array_width,

tools/plot-threadqueue-log.py

@@ -99,6 +99,94 @@ class LogThread:
   def plot(self, ax, i):
     ax.barh(i, self._stop - self._start, left=self._start, height=0.9, align='center',label="test", color='yellow')
     
+class IntervalThreadCounter:
+  def __init__(self):
+    self.interval_starts = []
+    self.interval_stops = []
+    
+  def add_interval(self, start, stop):
+    self.interval_starts.append(start)
+    self.interval_stops.append(stop)
+    self.interval_starts.sort()
+    self.interval_stops.sort()
+    
+  def get_values_xd(self):
+    #Double the first and the last items
+    xds = sorted([(x,'+') for x in self.interval_starts] + [(x,'-') for x in self.interval_stops])
+    return xds
+    
+  def get_values_x(self):
+    xs = []
+    for x in self.get_values_xd():
+      xs.append(x[0])
+      xs.append(x[0])
+    return xs
+  
+  def get_values_y(self):
+    xds = self.get_values_xd()
+    ys = []
+    counter = 0
+    for xd in xds:
+      ys.append(counter)
+      
+      if xd[1] == '+':
+        counter += 1
+      elif xd[1] == '-':
+        counter -= 1
+      else:
+        assert False
+      
+      ys.append(counter)
+      
+    return ys
+    
+  def clamp(self, v, minval, maxval):
+    if v < minval:
+      return minval
+    if v > maxval:
+      return maxval
+    return v
+    
+  def get_values_uniform_xy(self, kernel_size, steps):
+    kernel_size=float(kernel_size)
+    xchs = self.get_values_x()
+    ychs = self.get_values_y()
+    
+    minval = xchs[0] - kernel_size
+    maxval = xchs[-1] + kernel_size
+    
+    pos = minval
+    
+    xvalues = []
+    yvalues = []
+
+    while pos < maxval:
+      value = 0
+      for i in range(1,len(xchs)-1):
+        if xchs[i] < pos - kernel_size:
+          continue
+          
+        v1 = self.clamp(xchs[i-1], pos - kernel_size/2., pos+kernel_size/2.)
+        v2 = self.clamp(xchs[i], pos - kernel_size/2., pos+kernel_size/2.)
+        
+        diff=v2-v1
+        
+        value += diff*ychs[i]/kernel_size
+        
+        if xchs[i] > pos + kernel_size:
+          break
+          
+      xvalues.append(pos)
+      yvalues.append(value)
+      
+      pos += kernel_size/steps
+      
+    return xvalues, yvalues
+    
+      
+    
+    
+
 class LogParser:
   def _parse_time(self, base, sign, value):
     if sign == '+':
@@ -232,6 +320,19 @@ class LogParser:
 
     yticks = {}
     
+    #first draw usage
+    itc = IntervalThreadCounter()
+    for o in self._objects:
+      if isinstance(o, LogJob) and o._is_thread_job:
+        itc.add_interval(o._start, o._stop)
+    
+    #exact plot
+    ax.plot(itc.get_values_x(), [y+1.5 for y in itc.get_values_y()])
+    vx,vy = itc.get_values_uniform_xy(0.01,10)
+    ax.plot(vx, [y+1.5 for y in vy], 'r')
+    
+    
+
     #first draw threads
     for o in self._objects:
       if isinstance(o, LogJob) and o._is_thread_job: