Merge pull request #642 from egaburov/launch3d

concept of 3d tasking

examples/mandelbrot_tasks/mandelbrot_tasks.cpp

@@ -38,7 +38,8 @@
 #pragma warning (disable: 4305)
 #endif
 
-#include <stdio.h>
+#include <cstdio>
+#include <cstdlib>
 #include <algorithm>
 #include <string.h>
 #include "../timing.h"

examples/mandelbrot_tasks/mandelbrot_tasks.ispc

@@ -57,21 +57,26 @@ task void
 mandelbrot_scanline(uniform float x0, uniform float dx, 
                     uniform float y0, uniform float dy,
                     uniform int width, uniform int height, 
-                    uniform int span,
+                    uniform int xspan, uniform int yspan,
                     uniform int maxIterations, uniform int output[]) {
-    uniform int ystart = taskIndex * span;
-    uniform int yend = min((taskIndex+1) * span, (unsigned int)height);
+    const uniform int xstart = taskIndex0 * xspan;
+    const uniform int xend   = min(xstart  + xspan, width);
 
-    foreach (yi = ystart ... yend, xi = 0 ... width) {
+    const uniform int ystart = taskIndex1 * yspan;
+    const uniform int yend   = min(ystart  + yspan, height);
+    
+
+    foreach (yi = ystart ... yend, xi = xstart ... xend) {
         float x = x0 + xi * dx;
         float y = y0 + yi * dy;
 
         int index = yi * width + xi;
         output[index] = mandel(x, y, maxIterations);
     }
+    
 }
-                               
 
+#if 1
 export void
 mandelbrot_ispc(uniform float x0, uniform float y0, 
                 uniform float x1, uniform float y1,
@@ -79,8 +84,16 @@ mandelbrot_ispc(uniform float x0, uniform float y0,
                 uniform int maxIterations, uniform int output[]) {
     uniform float dx = (x1 - x0) / width;
     uniform float dy = (y1 - y0) / height;
-    uniform int span = 4;
+    const uniform int xspan = max(32, programCount*2);  /* make sure it is big enough to avoid false-sharing */
+    const uniform int yspan = 16; 
 
-    launch[height/span] mandelbrot_scanline(x0, dx, y0, dy, width, height, span,
-                                            maxIterations, output);
+
+#if 1
+    launch [width/xspan, height/yspan]
+#else
+    launch [height/yspan][width/xspan]
+#endif
+      mandelbrot_scanline(x0, dx, y0, dy, width, height, xspan, yspan,
+                          maxIterations, output);
 }
+#endif