nuclear@0: #include <stdio.h>
nuclear@8: #include <string.h>
nuclear@2: #include <math.h>
nuclear@58: #include <limits.h>
nuclear@0: #include <assert.h>
nuclear@47: #include "rt.h"
John@14: #include "ogl.h"
nuclear@0: #include "ocl.h"
nuclear@22: #include "scene.h"
nuclear@32: #include "timer.h"
nuclear@45: #include "common.h"
nuclear@0: 
nuclear@12: // kernel arguments
nuclear@12: enum {
nuclear@12: 	KARG_FRAMEBUFFER,
nuclear@12: 	KARG_RENDER_INFO,
nuclear@12: 	KARG_FACES,
nuclear@12: 	KARG_MATLIB,
nuclear@12: 	KARG_LIGHTS,
nuclear@12: 	KARG_PRIM_RAYS,
nuclear@12: 	KARG_XFORM,
John@14: 	KARG_INVTRANS_XFORM,
nuclear@28: 	KARG_KDTREE,
John@14: 
John@14: 	NUM_KERNEL_ARGS
nuclear@12: };
John@11: 
nuclear@47: static void update_render_info();
nuclear@3: static Ray get_primary_ray(int x, int y, int w, int h, float vfov_deg);
nuclear@43: static float *create_kdimage(const KDNodeGPU *kdtree, int num_nodes, int *xsz_ret, int *ysz_ret);
nuclear@3: 
nuclear@13: static Face *faces;
nuclear@3: static Ray *prim_rays;
nuclear@3: static CLProgram *prog;
nuclear@3: static int global_size;
nuclear@3: 
nuclear@7: 
nuclear@4: static RendInfo rinf;
nuclear@55: static RenderStats rstat;
nuclear@47: static int saved_iter_val;
nuclear@4: 
nuclear@43: static long timing_sample_sum;
nuclear@43: static long num_timing_samples;
nuclear@43: 
nuclear@4: 
nuclear@39: bool init_renderer(int xsz, int ysz, Scene *scn, unsigned int tex)
nuclear@0: {
nuclear@4: 	// render info
nuclear@22: 	rinf.ambient[0] = rinf.ambient[1] = rinf.ambient[2] = 0.0;
nuclear@16: 	rinf.ambient[3] = 0.0;
nuclear@16: 
nuclear@4: 	rinf.xsz = xsz;
nuclear@4: 	rinf.ysz = ysz;
nuclear@13: 	rinf.num_faces = scn->get_num_faces();
nuclear@54: 	rinf.num_lights = scn->get_num_lights();
nuclear@47: 	rinf.max_iter = saved_iter_val = 6;
nuclear@47: 	rinf.cast_shadows = true;
nuclear@4: 
nuclear@3: 	/* calculate primary rays */
nuclear@3: 	prim_rays = new Ray[xsz * ysz];
nuclear@2: 
nuclear@2: 	for(int i=0; i<ysz; i++) {
nuclear@2: 		for(int j=0; j<xsz; j++) {
nuclear@2: 			prim_rays[i * xsz + j] = get_primary_ray(j, i, xsz, ysz, 45.0);
nuclear@2: 		}
nuclear@0: 	}
nuclear@54: 	dbg_set_primary_rays(prim_rays);	// give them to the debug renderer
nuclear@0: 
nuclear@2: 	/* setup opencl */
nuclear@3: 	prog = new CLProgram("render");
nuclear@54: 	if(!prog->load("src/rt.cl")) {
nuclear@8: 		return false;
nuclear@0: 	}
nuclear@0: 
nuclear@24: 	if(!(faces = (Face*)scn->get_face_buffer())) {
nuclear@13: 		fprintf(stderr, "failed to create face buffer\n");
nuclear@13: 		return false;
nuclear@13: 	}
nuclear@13: 
nuclear@28: 	const KDNodeGPU *kdbuf = scn->get_kdtree_buffer();
nuclear@28: 	if(!kdbuf) {
nuclear@28: 		fprintf(stderr, "failed to create kdtree buffer\n");
nuclear@28: 		return false;
nuclear@28: 	}
nuclear@43: 
nuclear@43: 	int kdimg_xsz, kdimg_ysz;
nuclear@43: 	float *kdimg_pixels = create_kdimage(kdbuf, scn->get_num_kdnodes(), &kdimg_xsz, &kdimg_ysz);
nuclear@28: 
nuclear@3: 	/* setup argument buffers */
nuclear@41: #ifdef CLGL_INTEROP
nuclear@39: 	prog->set_arg_texture(KARG_FRAMEBUFFER, ARG_WR, tex);
nuclear@41: #else
nuclear@41: 	prog->set_arg_image(KARG_FRAMEBUFFER, ARG_WR, xsz, ysz);
nuclear@41: #endif
nuclear@12: 	prog->set_arg_buffer(KARG_RENDER_INFO, ARG_RD, sizeof rinf, &rinf);
John@14: 	prog->set_arg_buffer(KARG_FACES, ARG_RD, rinf.num_faces * sizeof(Face), faces);
John@14: 	prog->set_arg_buffer(KARG_MATLIB, ARG_RD, scn->get_num_materials() * sizeof(Material), scn->get_materials());
nuclear@54: 	prog->set_arg_buffer(KARG_LIGHTS, ARG_RD, scn->get_num_lights() * sizeof(Light), scn->get_lights());
nuclear@12: 	prog->set_arg_buffer(KARG_PRIM_RAYS, ARG_RD, xsz * ysz * sizeof *prim_rays, prim_rays);
nuclear@12: 	prog->set_arg_buffer(KARG_XFORM, ARG_RD, 16 * sizeof(float));
nuclear@12: 	prog->set_arg_buffer(KARG_INVTRANS_XFORM, ARG_RD, 16 * sizeof(float));
nuclear@43: 	//prog->set_arg_buffer(KARG_KDTREE, ARG_RD, scn->get_num_kdnodes() * sizeof *kdbuf, kdbuf);
nuclear@43: 	prog->set_arg_image(KARG_KDTREE, ARG_RD, kdimg_xsz, kdimg_ysz, kdimg_pixels);
nuclear@43: 
nuclear@43: 	delete [] kdimg_pixels;
nuclear@43: 
nuclear@12: 
John@14: 	if(prog->get_num_args() < NUM_KERNEL_ARGS) {
John@14: 		return false;
John@14: 	}
John@14: 
nuclear@45: 	const char *opt = "-Isrc -cl-mad-enable -cl-single-precision-constant -cl-fast-relaxed-math";
nuclear@45: 	if(!prog->build(opt)) {
nuclear@16: 		return false;
nuclear@16: 	}
nuclear@16: 
nuclear@54: 	//delete [] prim_rays; now dbg_renderer handles them
nuclear@2: 
nuclear@3: 	global_size = xsz * ysz;
nuclear@54: 
nuclear@54: 
nuclear@54: 	init_dbg_renderer(xsz, ysz, scn, tex);
nuclear@3: 	return true;
nuclear@3: }
nuclear@3: 
nuclear@3: void destroy_renderer()
nuclear@3: {
nuclear@3: 	delete prog;
nuclear@43: 
nuclear@54: 	destroy_dbg_renderer();
nuclear@54: 
nuclear@54: 	if(num_timing_samples) {
nuclear@54: 		printf("rendertime mean: %ld msec\n", timing_sample_sum / num_timing_samples);
nuclear@54: 	}
nuclear@3: }
nuclear@3: 
nuclear@3: bool render()
nuclear@3: {
nuclear@58: 	long tm0 = get_msec();
nuclear@39: 
nuclear@58: 	// initialize render-stats
nuclear@58: 	memset(&rstat, 0, sizeof rstat);
nuclear@58: 	rstat.min_aabb_tests = rstat.min_triangle_tests = INT_MAX;
nuclear@58: 	rstat.max_aabb_tests = rstat.max_triangle_tests = 0;
nuclear@32: 
nuclear@40: #ifdef CLGL_INTEROP
nuclear@39: 	cl_event ev;
nuclear@39: 	CLMemBuffer *texbuf = prog->get_arg_buffer(KARG_FRAMEBUFFER);
nuclear@39: 
nuclear@39: 	if(!acquire_gl_object(texbuf, &ev)) {
nuclear@39: 		return false;
nuclear@39: 	}
nuclear@39: 
nuclear@39: 	// make sure that we will wait for the acquire to finish before running
nuclear@39: 	prog->set_wait_event(ev);
nuclear@40: #endif
nuclear@39: 
nuclear@3: 	if(!prog->run(1, global_size)) {
nuclear@3: 		return false;
nuclear@0: 	}
John@15: 
nuclear@40: #ifdef CLGL_INTEROP
nuclear@39: 	if(!release_gl_object(texbuf, &ev)) {
nuclear@39: 		return false;
nuclear@39: 	}
nuclear@39: 	clWaitForEvents(1, &ev);
nuclear@40: #endif
nuclear@39: 
nuclear@40: #ifndef CLGL_INTEROP
nuclear@40: 	/* if we don't compile in CL/GL interoperability support, we need
nuclear@40: 	 * to copy the output buffer to the OpenGL texture used to displaying
nuclear@40: 	 * the image.
nuclear@40: 	 */
nuclear@13: 	CLMemBuffer *mbuf = prog->get_arg_buffer(KARG_FRAMEBUFFER);
nuclear@12: 	void *fb = map_mem_buffer(mbuf, MAP_RD);
nuclear@13: 	if(!fb) {
nuclear@13: 		fprintf(stderr, "FAILED\n");
nuclear@13: 		return false;
nuclear@13: 	}
nuclear@13: 
nuclear@12: 	glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, rinf.xsz, rinf.ysz, GL_RGBA, GL_FLOAT, fb);
nuclear@2: 	unmap_mem_buffer(mbuf);
nuclear@40: #endif
nuclear@32: 
nuclear@58: 	rstat.render_time = get_msec() - tm0;
nuclear@58: 
nuclear@58: 	timing_sample_sum += rstat.render_time;
nuclear@43: 	num_timing_samples++;
nuclear@43: 
nuclear@3: 	return true;
nuclear@0: }
nuclear@2: 
nuclear@12: 
nuclear@12: void set_xform(float *matrix, float *invtrans)
nuclear@12: {
nuclear@12: 	CLMemBuffer *mbuf_xform = prog->get_arg_buffer(KARG_XFORM);
nuclear@12: 	CLMemBuffer *mbuf_invtrans = prog->get_arg_buffer(KARG_INVTRANS_XFORM);
nuclear@12: 	assert(mbuf_xform && mbuf_invtrans);
nuclear@12: 
nuclear@12: 	float *mem = (float*)map_mem_buffer(mbuf_xform, MAP_WR);
nuclear@12: 	memcpy(mem, matrix, 16 * sizeof *mem);
nuclear@12: 	unmap_mem_buffer(mbuf_xform);
nuclear@12: 
nuclear@12: 	mem = (float*)map_mem_buffer(mbuf_invtrans, MAP_WR);
nuclear@12: 	memcpy(mem, invtrans, 16 * sizeof *mem);
nuclear@12: 	unmap_mem_buffer(mbuf_invtrans);
nuclear@8: }
nuclear@8: 
nuclear@54: 
nuclear@54: const RendInfo *get_render_info()
nuclear@54: {
nuclear@54: 	return &rinf;
nuclear@54: }
nuclear@54: 
nuclear@55: const RenderStats *get_render_stats()
nuclear@55: {
nuclear@55: 	return &rstat;
nuclear@55: }
nuclear@55: 
nuclear@55: void print_render_stats(FILE *fp)
nuclear@55: {
nuclear@55: 	fprintf(fp, "-- render stats --\n");
nuclear@55: 	fprintf(fp, "> timing\n");
nuclear@55: 	fprintf(fp, "   render time (msec): %lu\n", rstat.render_time);
nuclear@55: 	fprintf(fp, "   tex update time (msec): %lu\n", rstat.tex_update_time);
nuclear@55: 	fprintf(fp, "> counters\n");
nuclear@55: 	fprintf(fp, "   AABB tests: %d\n", rstat.aabb_tests);
nuclear@55: 	fprintf(fp, "   AABB tests per ray (min/max/avg): %d/%d/%f\n",
nuclear@55: 			rstat.min_aabb_tests, rstat.max_aabb_tests, rstat.avg_aabb_tests);
nuclear@55: 	fprintf(fp, "   triangle tests: %d\n", rstat.triangle_tests);
nuclear@55: 	fprintf(fp, "   triangle tests per ray (min/max/avg): %d/%d/%f\n",
nuclear@55: 			rstat.min_triangle_tests, rstat.max_triangle_tests, rstat.avg_triangle_tests);
nuclear@55: 	fprintf(fp, "   rays cast: %dp %dr %ds (sum: %d)\n", rstat.prim_rays,
nuclear@55: 			rstat.refl_rays, rstat.shadow_rays, rstat.rays_cast);
nuclear@55: 	fprintf(fp, "   rays per second: %d\n", rstat.rays_per_sec);
nuclear@55: 	fprintf(fp, "   BRDF evaluations: %d\n", rstat.brdf_evals);
nuclear@55: 	fputc('\n', fp);
nuclear@55: }
nuclear@55: 
nuclear@47: void set_render_option(int opt, bool val)
nuclear@47: {
nuclear@47: 	switch(opt) {
nuclear@47: 	case ROPT_ITER:
nuclear@47: 	case ROPT_REFL:
nuclear@47: 		rinf.max_iter = val ? saved_iter_val : 0;
nuclear@47: 		break;
nuclear@47: 
nuclear@47: 	case ROPT_SHAD:
nuclear@47: 		rinf.cast_shadows = val;
nuclear@47: 		break;
nuclear@47: 
nuclear@47: 	default:
nuclear@47: 		return;
nuclear@47: 	}
nuclear@47: 
nuclear@47: 	update_render_info();
nuclear@47: }
nuclear@47: 
nuclear@47: void set_render_option(int opt, int val)
nuclear@47: {
nuclear@47: 	switch(opt) {
nuclear@47: 	case ROPT_ITER:
nuclear@47: 		rinf.max_iter = saved_iter_val = val;
nuclear@47: 		break;
nuclear@47: 
nuclear@47: 	case ROPT_SHAD:
nuclear@47: 		rinf.cast_shadows = val;
nuclear@47: 		break;
nuclear@47: 
nuclear@47: 	case ROPT_REFL:
nuclear@47: 		rinf.max_iter = val ? saved_iter_val : 0;
nuclear@47: 		break;
nuclear@47: 
nuclear@47: 	default:
nuclear@47: 		return;
nuclear@47: 	}
nuclear@47: 
nuclear@47: 	update_render_info();
nuclear@47: }
nuclear@47: 
nuclear@47: void set_render_option(int opt, float val)
nuclear@47: {
nuclear@47: 	set_render_option(opt, (int)val);
nuclear@47: }
nuclear@47: 
nuclear@47: bool get_render_option_bool(int opt)
nuclear@47: {
nuclear@47: 	switch(opt) {
nuclear@47: 	case ROPT_ITER:
nuclear@47: 		return rinf.max_iter;
nuclear@47: 	case ROPT_SHAD:
nuclear@47: 		return rinf.cast_shadows;
nuclear@47: 	case ROPT_REFL:
nuclear@47: 		return rinf.max_iter == saved_iter_val;
nuclear@47: 	default:
nuclear@47: 		break;
nuclear@47: 	}
nuclear@47: 	return false;
nuclear@47: }
nuclear@47: 
nuclear@47: int get_render_option_int(int opt)
nuclear@47: {
nuclear@47: 	switch(opt) {
nuclear@47: 	case ROPT_ITER:
nuclear@47: 		return rinf.max_iter;
nuclear@47: 	case ROPT_SHAD:
nuclear@47: 		return rinf.cast_shadows ? 1 : 0;
nuclear@47: 	case ROPT_REFL:
nuclear@47: 		return rinf.max_iter == saved_iter_val ? 1 : 0;
nuclear@47: 	default:
nuclear@47: 		break;
nuclear@47: 	}
nuclear@47: 	return -1;
nuclear@47: }
nuclear@47: 
nuclear@47: float get_render_option_float(int opt)
nuclear@47: {
nuclear@47: 	return (float)get_render_option_int(opt);
nuclear@47: }
nuclear@47: 
nuclear@47: static void update_render_info()
nuclear@47: {
nuclear@47: 	if(!prog) {
nuclear@47: 		return;
nuclear@47: 	}
nuclear@47: 
nuclear@47: 	CLMemBuffer *mbuf = prog->get_arg_buffer(KARG_RENDER_INFO);
nuclear@47: 	assert(mbuf);
nuclear@47: 
nuclear@47: 	RendInfo *rinf_ptr = (RendInfo*)map_mem_buffer(mbuf, MAP_WR);
nuclear@47: 	*rinf_ptr = rinf;
nuclear@47: 	unmap_mem_buffer(mbuf);
nuclear@47: }
nuclear@47: 
nuclear@3: static Ray get_primary_ray(int x, int y, int w, int h, float vfov_deg)
nuclear@2: {
nuclear@2: 	float vfov = M_PI * vfov_deg / 180.0;
nuclear@2: 	float aspect = (float)w / (float)h;
nuclear@2: 
nuclear@2: 	float ysz = 2.0;
nuclear@2: 	float xsz = aspect * ysz;
nuclear@2: 
nuclear@2: 	float px = ((float)x / (float)w) * xsz - xsz / 2.0;
nuclear@2: 	float py = 1.0 - ((float)y / (float)h) * ysz;
nuclear@2: 	float pz = 1.0 / tan(0.5 * vfov);
nuclear@2: 
nuclear@43: 	float mag = sqrt(px * px + py * py + pz * pz);
nuclear@43: 
nuclear@45: 	px = px * RAY_MAG / mag;
nuclear@45: 	py = py * RAY_MAG / mag;
nuclear@45: 	pz = pz * RAY_MAG / mag;
nuclear@2: 
nuclear@18: 	Ray ray = {{0, 0, 0, 1}, {px, py, -pz, 1}};
nuclear@2: 	return ray;
nuclear@2: }
nuclear@43: 
nuclear@54: #define MIN(a, b)	((a) < (b) ? (a) : (b))
nuclear@54: 
nuclear@43: static float *create_kdimage(const KDNodeGPU *kdtree, int num_nodes, int *xsz_ret, int *ysz_ret)
nuclear@43: {
nuclear@45: 	int ysz = MIN(num_nodes, KDIMG_MAX_HEIGHT);
nuclear@45: 	int columns = (num_nodes - 1) / KDIMG_MAX_HEIGHT + 1;
nuclear@45: 	int xsz = KDIMG_NODE_WIDTH * columns;
nuclear@43: 
nuclear@43: 	printf("creating kdtree image %dx%d (%d nodes)\n", xsz, ysz, num_nodes);
nuclear@43: 
nuclear@43: 	float *img = new float[4 * xsz * ysz];
nuclear@43: 	memset(img, 0, 4 * xsz * ysz * sizeof *img);
nuclear@43: 
nuclear@43: 	for(int i=0; i<num_nodes; i++) {
nuclear@45: 		int x = KDIMG_NODE_WIDTH * (i / KDIMG_MAX_HEIGHT);
nuclear@45: 		int y = i % KDIMG_MAX_HEIGHT;
nuclear@45: 
nuclear@45: 		float *ptr = img + (y * xsz + x) * 4;
nuclear@43: 
nuclear@43: 		*ptr++ = kdtree[i].aabb.min[0];
nuclear@43: 		*ptr++ = kdtree[i].aabb.min[1];
nuclear@43: 		*ptr++ = kdtree[i].aabb.min[2];
nuclear@43: 		*ptr++ = 0.0;
nuclear@43: 
nuclear@43: 		*ptr++ = kdtree[i].aabb.max[0];
nuclear@43: 		*ptr++ = kdtree[i].aabb.max[1];
nuclear@43: 		*ptr++ = kdtree[i].aabb.max[2];
nuclear@43: 		*ptr++ = 0.0;
nuclear@43: 
nuclear@43: 		for(int j=0; j<MAX_NODE_FACES; j++) {
nuclear@43: 			*ptr++ = j < kdtree[i].num_faces ? (float)kdtree[i].face_idx[j] : 0.0f;
nuclear@43: 		}
nuclear@43: 
nuclear@43: 		*ptr++ = (float)kdtree[i].num_faces;
nuclear@43: 		*ptr++ = (float)kdtree[i].left;
nuclear@43: 		*ptr++ = (float)kdtree[i].right;
nuclear@43: 		*ptr++ = 0.0;
nuclear@43: 	}
nuclear@43: 
nuclear@43: 	if(xsz_ret) *xsz_ret = xsz;
nuclear@43: 	if(ysz_ret) *ysz_ret = ysz;
nuclear@43: 	return img;
nuclear@43: }