aobench: define grid size in ao.h

Author: Hugobros3

samples/aobench/ao.c

@@ -58,7 +58,7 @@ FUNCTION static void vnormalize(vec *c)
 FUNCTION static void
 ray_sphere_intersect(Isect *isect, const Ray *ray, const Sphere *sphere)
 {
-    vec rs;
+    vec rs = { 0 };
 
     rs.x = ray->org.x - sphere->center.x;
     rs.y = ray->org.y - sphere->center.y;

samples/aobench/ao.cl

@@ -17,7 +17,7 @@ void debug_printf_i32_i32(const __constant char*, int, int) __asm__("__shady::pr
 
 }
 
-__attribute__((reqd_work_group_size(16, 16, 1)))
+__attribute__((reqd_work_group_size(BLOCK_SIZE, BLOCK_SIZE, 1)))
 kernel void aobench_kernel(global unsigned char* out) {
     int x = get_global_id(0);
     int y = get_global_id(1);

samples/aobench/ao.cu

@@ -10,7 +10,7 @@ __global__ void aobench_kernel(unsigned char* out) {
     long int ptr = (long int) out;
     Ctx ctx = get_init_context();
     init_scene(&ctx);
-    // render_pixel(&ctx, x, y, WIDTH, HEIGHT, NSUBSAMPLES, out);
+    render_pixel(&ctx, x, y, WIDTH, HEIGHT, NSUBSAMPLES, out);
 }
 
 }

samples/aobench/ao.h

@@ -2,6 +2,7 @@
 #define HEIGHT       2048
 #define NSUBSAMPLES  1
 #define NAO_SAMPLES  8
+#define BLOCK_SIZE 16
 
 typedef float Scalar;
 

samples/aobench/ao_main.c

@@ -119,11 +119,11 @@ void render_device(Args* args, unsigned char *img, int w, int h, int nsubsamples
     Program* program = load_program_from_disk(runtime, &args->compiler_config, path);
 
     // run it twice to compile everything and benefit from caches
-    wait_completion(launch_kernel(program, device, "aobench_kernel", WIDTH / 16, HEIGHT / 16, 1, 1, (void*[]) { &buf_addr }));
+    wait_completion(launch_kernel(program, device, "aobench_kernel", WIDTH / BLOCK_SIZE, HEIGHT / BLOCK_SIZE, 1, 1, (void*[]) { &buf_addr }));
     struct timespec ts;
     timespec_get(&ts, TIME_UTC);
     uint64_t tsn = timespec_to_nano(ts);
-    wait_completion(launch_kernel(program, device, "aobench_kernel", WIDTH / 16, HEIGHT / 16, 1, 1, (void*[]) { &buf_addr }));
+    wait_completion(launch_kernel(program, device, "aobench_kernel", WIDTH / BLOCK_SIZE, HEIGHT / BLOCK_SIZE, 1, 1, (void*[]) { &buf_addr }));
     struct timespec tp;
     timespec_get(&tp, TIME_UTC);
     uint64_t tpn = timespec_to_nano(tp);