fixed helpers and added ao_bench example

examples/aobench/Makefile

@@ -2,7 +2,7 @@
 EXAMPLE=ao
 CPP_SRC=ao.cpp ao_serial.cpp
 ISPC_SRC=ao.ispc
-ISPC_IA_TARGETS=sse2-i32x4,sse4-i32x4,avx1-i32x8,avx2-i32x8
+ISPC_IA_TARGETS=avx1-i32x8
 ISPC_ARM_TARGETS=neon
 
 include ../common.mk

examples_ptx/aobench/ao.cpp

@@ -137,8 +137,8 @@ int main(int argc, char **argv)
     //
 #ifndef _CUDA_
     double minTimeISPC = 1e30;
-    memset((void *)fimg, 0, sizeof(float) * width * height * 3);
     for (unsigned int i = 0; i < test_iterations[0]; i++) {
+        ispc_memset((void *)fimg, 0, sizeof(float) * width * height * 3);
         assert(NSUBSAMPLES == 2);
         reset_and_start_timer();
         ispc::ao_ispc(width, height, NSUBSAMPLES, fimg);
@@ -158,8 +158,8 @@ int main(int argc, char **argv)
     // minimum time for any of them.
     //
     double minTimeISPCTasks = 1e30;
-    memset((void *)fimg, 0, sizeof(float) * width * height * 3);
     for (unsigned int i = 0; i < test_iterations[1]; i++) {
+        ispc_memset((void *)fimg, 0, sizeof(float) * width * height * 3);
         assert(NSUBSAMPLES == 2);
 
         reset_and_start_timer();
@@ -179,8 +179,8 @@ int main(int argc, char **argv)
     // minimum time.
     //
     double minTimeSerial = 1e30;
-    memset((void *)fimg, 0, sizeof(float) * width * height * 3);
     for (unsigned int i = 0; i < test_iterations[2]; i++) {
+        ispc_memset((void *)fimg, 0, sizeof(float) * width * height * 3);
         reset_and_start_timer();
         ao_serial(width, height, NSUBSAMPLES, fimg);
         double t = get_elapsed_msec();
@@ -199,6 +199,9 @@ int main(int argc, char **argv)
            minTimeSerial / minTimeISPCTasks);
 #endif
     savePPM("ao-serial.ppm", width, height); 
+
+    ispc_free(img);
+    ispc_free(fimg);
         
     return 0;
 }

examples_ptx/aobench/ao.ispc

@@ -259,7 +259,7 @@ export void ao_ispc(uniform int w, uniform int h, uniform int nsubsamples,
     ao_scanlines(0, h, w, h, nsubsamples, image);
 }
 
-
+#if 0
 static void task ao_task(uniform int width, uniform int height, 
                          uniform int nsubsamples, uniform float image[]) {
     ao_scanlines(taskIndex, taskIndex+1, width, height, nsubsamples, image);
@@ -270,3 +270,95 @@ export void ao_ispc_tasks(uniform int w, uniform int h, uniform int nsubsamples,
                           uniform float image[]) {
     launch[h] ao_task(w, h, nsubsamples, image);
 }
+#else
+
+static  inline void ao_tile(
+    uniform int x0, uniform int x1,
+    uniform int y0, uniform int y1, 
+    uniform int w, uniform int h,
+    uniform int nsubsamples, 
+    uniform float image[]) 
+{
+  uniform Plane plane = { { 0.0f, -0.5f, 0.0f }, { 0.f, 1.f, 0.f } };
+  uniform Sphere spheres[3] = {
+    { { -2.0f, 0.0f, -3.5f }, 0.5f },
+    { { -0.5f, 0.0f, -3.0f }, 0.5f },
+    { { 1.0f, 0.0f, -2.2f }, 0.5f } };
+  RNGState rngstate;
+
+  seed_rng(&rngstate, programIndex + (y0 << (programIndex & 15)));
+  float invSamples = 1.f / nsubsamples;
+  foreach_tiled (y = y0 ... y1, x = x0 ... x1)
+  {
+    const int offset = 3 * (y * w + x);
+    float res = 0.0f;
+
+    for (uniform int u = 0; u < nsubsamples; u++)
+      for (uniform int v = 0; v < nsubsamples; v++)
+      {
+        float du = (float)u * invSamples, dv = (float)v * invSamples;
+
+        // Figure out x,y pixel in NDC
+        float px =  (x + du - (w / 2.0f)) / (w / 2.0f);
+        float py = -(y + dv - (h / 2.0f)) / (h / 2.0f);
+        float ret = 0.f;
+        Ray ray;
+        Isect isect;
+
+        ray.org = 0.f;
+
+        // Poor man's perspective projection
+        ray.dir.x = px;
+        ray.dir.y = py;
+        ray.dir.z = -1.0;
+        vnormalize(ray.dir);
+
+        isect.t   = 1.0e+17;
+        isect.hit = 0;
+
+        for (uniform int snum = 0; snum < 3; ++snum)
+          ray_sphere_intersect(isect, ray, spheres[snum]);
+        ray_plane_intersect(isect, ray, plane);
+
+        // Note use of 'coherent' if statement; the set of rays we
+        // trace will often all hit or all miss the scene
+        if (isect.hit) {
+          ret = ambient_occlusion(isect, plane, spheres, rngstate);
+          ret *= invSamples * invSamples;
+          res += ret;
+        }
+      }
+
+      image[offset  ] = res;
+      image[offset+1] = res;
+      image[offset+2] = res;
+  }
+}
+
+#define TILEX 64
+#define TILEY 4
+
+void task ao_task(uniform int width, uniform int height, 
+    uniform int nsubsamples, uniform float image[]) 
+{
+  if (taskIndex0 >= taskCount0) return;
+  if (taskIndex1 >= taskCount1) return;
+
+  const uniform int x0 = taskIndex0 * TILEX;
+  const uniform int x1 = min(x0 + TILEX, width);
+
+  const uniform int y0 = taskIndex1 * TILEY;
+  const uniform int y1 = min(y0 + TILEY, height);
+  ao_tile(x0,x1,y0,y1, width, height, nsubsamples, image);
+}
+
+
+export void ao_ispc_tasks(uniform int w, uniform int h, uniform int nsubsamples, 
+    uniform float image[]) 
+{
+  const uniform int ntilex = (w+TILEX-1)/TILEX;
+  const uniform int ntiley = (h+TILEY-1)/TILEY;
+  launch[ntilex,ntiley] ao_task(w, h, nsubsamples, image);
+  sync;
+}
+#endif

examples_ptx/ispc_malloc.h

@@ -3,6 +3,7 @@
 #ifdef _CUDA_
 extern void ispc_malloc(void **ptr, const size_t size);
 extern void ispc_free(void *ptr);
+extern void ispc_memset(void *ptr, int value, size_t size);
 #else
 #include <cstdlib>
 static inline void ispc_malloc(void **ptr, const size_t size)
@@ -13,5 +14,9 @@ static inline void ispc_free(void *ptr)
 {
   free(ptr);
 }
+static inline void ispc_memset(void *ptr, int value, size_t size)
+{
+  memset(ptr, value, size);
+}
 
 #endif

examples_ptx/nvcc_helpers.cu

@@ -12,5 +12,9 @@ void ispc_free(void *ptr)
 {
   cudaFree(ptr);
 }
+void ispc_memset(void *ptr, int value, size_t size)
+{
+  cudaMemset(ptr, value, size);
+}