/**
* Cellular automata sample main function.
* */
int main(int argc, char* argv[]) {
/* Wrappers for OpenCL objects. */
CCLContext* ctx;
CCLDevice* dev;
CCLImage* img1;
CCLImage* img2;
CCLProgram* prg;
CCLKernel* krnl;
CCLEvent* evt1;
CCLEvent* evt2;
/* Other variables. */
CCLEventWaitList ewl = NULL;
/* Profiler object. */
CCLProf* prof;
/* Output images filename. */
char* filename;
/* Selected device, may be given in command line. */
int dev_idx = -1;
/* Error handling object (must be NULL). */
GError* err = NULL;
/* Does selected device support images? */
cl_bool image_ok;
/* Initial sim state. */
cl_uchar4* input_image;
/* Simulation states. */
cl_uchar4** output_images;
/* RNG seed, may be given in command line. */
unsigned int seed;
/* Image file write status. */
int file_write_status;
/* Image format. */
cl_image_format image_format = { CL_RGBA, CL_UNSIGNED_INT8 };
/* Thread data. */
struct thread_data td;
/* Global and local worksizes. */
size_t gws[2];
size_t lws[2];
/* Threads. */
GThread* comm_thread;
GThread* exec_thread;
/* Check arguments. */
if (argc >= 2) {
/* Check if a device was specified in the command line. */
dev_idx = atoi(argv[1]);
}
if (argc >= 3) {
/* Check if a RNG seed was specified. */
seed = atoi(argv[2]);
} else {
seed = (unsigned int) time(NULL);
}
/* Initialize RNG. */
srand(seed);
/* Create random initial state. */
input_image = (cl_uchar4*)
malloc(CA_WIDTH * CA_HEIGHT * sizeof(cl_uchar4));
for (cl_uint i = 0; i < CA_WIDTH * CA_HEIGHT; ++i) {
cl_uchar state = (rand() & 0x3) ? 0xFF : 0x00;
input_image[i] = (cl_uchar4) {{ state, state, state, 0xFF }};
}
/* Allocate space for simulation results. */
output_images = (cl_uchar4**)
malloc((CA_ITERS + 1) * sizeof(cl_uchar4*));
for (cl_uint i = 0; i < CA_ITERS + 1; ++i)
output_images[i] = (cl_uchar4*)
malloc(CA_WIDTH * CA_HEIGHT * sizeof(cl_uchar4));
/* Create context using device selected from menu. */
ctx = ccl_context_new_from_menu_full(&dev_idx, &err);
HANDLE_ERROR(err);
/* Get first device in context. */
dev = ccl_context_get_device(ctx, 0, &err);
HANDLE_ERROR(err);
/* Ask device if it supports images. */
image_ok = ccl_device_get_info_scalar(
dev, CL_DEVICE_IMAGE_SUPPORT, cl_bool, &err);
HANDLE_ERROR(err);
if (!image_ok)
ERROR_MSG_AND_EXIT("Selected device doesn't support images.");
/* Create command queues. */
queue_exec = ccl_queue_new(ctx, dev, CL_QUEUE_PROFILING_ENABLE, &err);
HANDLE_ERROR(err);
queue_comm = ccl_queue_new(ctx, dev, CL_QUEUE_PROFILING_ENABLE, &err);
HANDLE_ERROR(err);
/* Create 2D image for initial state. */
img1 = ccl_image_new(ctx, CL_MEM_READ_WRITE,
&image_format, NULL, &err,
"image_type", (cl_mem_object_type) CL_MEM_OBJECT_IMAGE2D,
"image_width", (size_t) CA_WIDTH,
"image_height", (size_t) CA_HEIGHT,
NULL);
HANDLE_ERROR(err);
/* Create another 2D image for double buffering. */
img2 = ccl_image_new(ctx, CL_MEM_READ_WRITE,
&image_format, NULL, &err,
"image_type", (cl_mem_object_type) CL_MEM_OBJECT_IMAGE2D,
"image_width", (size_t) CA_WIDTH,
"image_height", (size_t) CA_HEIGHT,
NULL);
HANDLE_ERROR(err);
/* Create program from kernel source and compile it. */
prg = ccl_program_new_from_source(ctx, CA_KERNEL, &err);
HANDLE_ERROR(err);
ccl_program_build(prg, NULL, &err);
HANDLE_ERROR(err);
/* Get kernel wrapper. */
krnl = ccl_program_get_kernel(prg, "ca", &err);
HANDLE_ERROR(err);
/* Determine nice local and global worksizes. */
ccl_kernel_suggest_worksizes(krnl, dev, 2, real_ws, gws, lws, &err);
HANDLE_ERROR(err);
printf("\n * Global work-size: (%d, %d)\n", (int) gws[0], (int) gws[1]);
printf(" * Local work-size: (%d, %d)\n", (int) lws[0], (int) lws[1]);
/* Create thread communication queues. */
comm_thread_queue = g_async_queue_new();
exec_thread_queue = g_async_queue_new();
host_thread_queue = g_async_queue_new();
/* Setup thread data. */
td.krnl = krnl;
td.img1 = img1;
td.img2 = img2;
td.gws = gws;
td.lws = lws;
td.output_images = output_images;
/* Create threads. */
exec_thread = g_thread_new("exec_thread", exec_func, &td);
comm_thread = g_thread_new("comm_thread", comm_func, &td);
/* Start profiling. */
prof = ccl_prof_new();
ccl_prof_start(prof);
/* Write initial state. */
ccl_image_enqueue_write(img1, queue_comm, CL_TRUE,
origin, region, 0, 0, input_image, NULL, &err);
HANDLE_ERROR(err);
/* Run CA_ITERS iterations of the CA. */
for (cl_uint i = 0; i < CA_ITERS; ++i) {
/* Send message to comms thread. */
g_async_queue_push(comm_thread_queue, &go_msg);
/* Send message to exec thread. */
g_async_queue_push(exec_thread_queue, &go_msg);
/* Get event wrappers from both threads. */
evt1 = (CCLEvent*) g_async_queue_pop(host_thread_queue);
evt2 = (CCLEvent*) g_async_queue_pop(host_thread_queue);
/* Can't continue until this iteration is over. */
ccl_event_wait_list_add(&ewl, evt1, evt2, NULL);
/* Wait for events. */
ccl_event_wait(&ewl, &err);
HANDLE_ERROR(err);
}
/* Send message to comms thread to read last result. */
g_async_queue_push(comm_thread_queue, &go_msg);
/* Send stop messages to both threads. */
g_async_queue_push(comm_thread_queue, &stop_msg);
g_async_queue_push(exec_thread_queue, &stop_msg);
/* Get event wrapper from comms thread. */
evt1 = (CCLEvent*) g_async_queue_pop(host_thread_queue);
/* Can't continue until final read is over. */
ccl_event_wait_list_add(&ewl, evt1, NULL);
ccl_event_wait(&ewl, &err);
HANDLE_ERROR(err);
/* Make sure both queues are finished. */
ccl_queue_finish(queue_comm, &err);
HANDLE_ERROR(err);
ccl_queue_finish(queue_exec, &err);
HANDLE_ERROR(err);
/* Stop profiling timer and add queues for analysis. */
ccl_prof_stop(prof);
ccl_prof_add_queue(prof, "Comms", queue_comm);
ccl_prof_add_queue(prof, "Exec", queue_exec);
/* Allocate space for base filename. */
filename = (char*) malloc(
(strlen(IMAGE_FILE_PREFIX ".png") + IMAGE_FILE_NUM_DIGITS + 1) * sizeof(char));
/* Write results to image files. */
for (cl_uint i = 0; i < CA_ITERS; ++i) {
/* Determine next filename. */
sprintf(filename, "%s%0" G_STRINGIFY(IMAGE_FILE_NUM_DIGITS) "d.png", IMAGE_FILE_PREFIX, i);
/* Save next image. */
file_write_status = stbi_write_png(filename, CA_WIDTH, CA_HEIGHT, 4,
output_images[i], CA_WIDTH * sizeof(cl_uchar4));
/* Give feedback if unable to save image. */
if (!file_write_status) {
ERROR_MSG_AND_EXIT("Unable to save image in file.");
}
}
/* Process profiling info. */
ccl_prof_calc(prof, &err);
HANDLE_ERROR(err);
/* Print profiling info. */
ccl_prof_print_summary(prof);
/* Save profiling info. */
ccl_prof_export_info_file(prof, "prof.tsv", &err);
HANDLE_ERROR(err);
/* Destroy threads. */
g_thread_join(exec_thread);
g_thread_join(comm_thread);
/* Destroy thread communication queues. */
g_async_queue_unref(comm_thread_queue);
g_async_queue_unref(exec_thread_queue);
g_async_queue_unref(host_thread_queue);
/* Release host buffers. */
free(filename);
free(input_image);
for (cl_uint i = 0; i < CA_ITERS + 1; ++i)
free(output_images[i]);
free(output_images);
/* Release wrappers. */
ccl_image_destroy(img1);
ccl_image_destroy(img2);
ccl_program_destroy(prg);
ccl_queue_destroy(queue_comm);
ccl_queue_destroy(queue_exec);
ccl_context_destroy(ctx);
/* Destroy profiler. */
ccl_prof_destroy(prof);
/* Check all wrappers have been destroyed. */
g_assert(ccl_wrapper_memcheck());
/* Terminate. */
return 0;
}
/**
* Image filter main function.
* */
int main(int argc, char * argv[]) {
/* Wrappers for OpenCL objects. */
CCLContext * ctx;
CCLDevice * dev;
CCLImage * img_in;
CCLImage * img_out;
CCLQueue * queue;
CCLProgram * prg;
CCLKernel * krnl;
CCLSampler * smplr;
/* Device selected specified in the command line. */
int dev_idx = -1;
/* Error handling object (must be initialized to NULL). */
CCLErr * err = NULL;
/* Does selected device support images? */
cl_bool image_ok;
/* Image data in host. */
unsigned char * input_image;
unsigned char * output_image;
/* Image properties. */
int width, height, n_channels;
/* Image file write status. */
int file_write_status;
/* Image parameters. */
cl_image_format image_format = { CL_RGBA, CL_UNSIGNED_INT8 };
/* Origin and region of complete image. */
size_t origin[3] = { 0, 0, 0 };
size_t region[3];
/* Real worksize. */
size_t real_ws[2];
/* Global and local worksizes. */
size_t gws[2];
size_t lws[2];
/* Check arguments. */
if (argc < 2) {
ERROR_MSG_AND_EXIT("Usage: image_filter <image_file> [device_index]");
} else if (argc >= 3) {
/* Check if a device was specified in the command line. */
dev_idx = atoi(argv[2]);
}
/* Load image. */
input_image = stbi_load(argv[1], &width, &height, &n_channels, 4);
if (!input_image) ERROR_MSG_AND_EXIT(stbi_failure_reason());
/* Real work size. */
real_ws[0] = width; real_ws[1] = height;
/* Set image region. */
region[0] = width; region[1] = height; region[2] = 1;
/* Create context using device selected from menu. */
ctx = ccl_context_new_from_menu_full(&dev_idx, &err);
HANDLE_ERROR(err);
/* Get first device in context. */
dev = ccl_context_get_device(ctx, 0, &err);
HANDLE_ERROR(err);
/* Ask device if it supports images. */
image_ok = ccl_device_get_info_scalar(
dev, CL_DEVICE_IMAGE_SUPPORT, cl_bool, &err);
HANDLE_ERROR(err);
if (!image_ok)
ERROR_MSG_AND_EXIT("Selected device doesn't support images.");
/* Create a command queue. */
queue = ccl_queue_new(ctx, dev, 0, &err);
HANDLE_ERROR(err);
/* Create 2D input image using loaded image data. */
img_in = ccl_image_new(ctx, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
&image_format, input_image, &err,
"image_type", (cl_mem_object_type) CL_MEM_OBJECT_IMAGE2D,
"image_width", (size_t) width,
"image_height", (size_t) height,
NULL);
HANDLE_ERROR(err);
/* Create 2D output image. */
img_out = ccl_image_new(ctx, CL_MEM_WRITE_ONLY,
&image_format, NULL, &err,
"image_type", (cl_mem_object_type) CL_MEM_OBJECT_IMAGE2D,
"image_width", (size_t) width,
"image_height", (size_t) height,
NULL);
HANDLE_ERROR(err);
/* Create program from kernel source and compile it. */
prg = ccl_program_new_from_source(ctx, FILTER_KERNEL, &err);
HANDLE_ERROR(err);
ccl_program_build(prg, NULL, &err);
HANDLE_ERROR(err);
/* Get kernel wrapper. */
krnl = ccl_program_get_kernel(prg, "do_filter", &err);
HANDLE_ERROR(err);
/* Determine nice local and global worksizes. */
ccl_kernel_suggest_worksizes(krnl, dev, 2, real_ws, gws, lws, &err);
HANDLE_ERROR(err);
/* Show information to user. */
printf("\n * Image size: %d x %d, %d channels\n",
width, height, n_channels);
printf(" * Global work-size: (%d, %d)\n", (int) gws[0], (int) gws[1]);
printf(" * Local work-size: (%d, %d)\n", (int) lws[0], (int) lws[1]);
/* Create sampler (this could also be created in-kernel). */
smplr = ccl_sampler_new(ctx, CL_FALSE, CL_ADDRESS_CLAMP_TO_EDGE,
CL_FILTER_NEAREST, &err);
HANDLE_ERROR(err);
/* Apply filter. */
ccl_kernel_set_args_and_enqueue_ndrange(
krnl, queue, 2, NULL, gws, lws, NULL, &err,
img_in, img_out, smplr, NULL);
HANDLE_ERROR(err);
/* Allocate space for output image. */
output_image = (unsigned char *)
malloc(width * height * 4 * sizeof(unsigned char));
/* Read image data back to host. */
ccl_image_enqueue_read(img_out, queue, CL_TRUE, origin, region,
0, 0, output_image, NULL, &err);
HANDLE_ERROR(err);
/* Write image to file. */
file_write_status = stbi_write_png(IMAGE_FILE, width, height, 4,
output_image, width * 4);
/* Give feedback. */
if (file_write_status) {
fprintf(stdout, "\nImage saved in file '" IMAGE_FILE "'\n");
} else {
ERROR_MSG_AND_EXIT("Unable to save image in file.");
}
/* Release host images. */
free(output_image);
stbi_image_free(input_image);
/* Release wrappers. */
ccl_image_destroy(img_in);
ccl_image_destroy(img_out);
ccl_sampler_destroy(smplr);
ccl_program_destroy(prg);
ccl_queue_destroy(queue);
ccl_context_destroy(ctx);
/* Check all wrappers have been destroyed. */
assert(ccl_wrapper_memcheck());
/* Terminate. */
return EXIT_SUCCESS;
}