/**
* @internal
*
* @brief Tests buffer fill.
* */
static void fill_test() {
/* Test variables. */
CCLPlatforms * ps;
CCLPlatform * p;
CCLContext * ctx = NULL;
CCLDevice * d = NULL;
CCLBuffer * b = NULL;
CCLQueue * q;
cl_char8 h[CCL_TEST_BUFFER_SIZE];
cl_char8 pattern = {{ 1, -1, 5, 4, -12, 3, 7, -20 }};
size_t buf_size = sizeof(cl_char8) * CCL_TEST_BUFFER_SIZE;
CCLErr * err = NULL;
/* Get a context which supports OpenCL 1.2, if possible. */
ps = ccl_platforms_new(&err);
g_assert_no_error(err);
for (guint i = 0; i < ccl_platforms_count(ps); ++i) {
p = ccl_platforms_get(ps, i);
cl_uint ocl_ver = ccl_platform_get_opencl_version(p, &err);
if (ocl_ver >= 120) {
ctx = ccl_context_new_from_devices(
ccl_platform_get_num_devices(p, NULL),
ccl_platform_get_all_devices(p, NULL),
&err);
g_assert_no_error(err);
break;
}
}
/* If not possible to find a 1.2 or better context, finish this
* test. */
if (ctx == NULL) {
g_test_message("'%s' test not performed because no platform " \
"with OpenCL 1.2 support was found", CCL_STRD);
ccl_platforms_destroy(ps);
return;
}
/* Get first device in context. */
d = ccl_context_get_device(ctx, 0, &err);
g_assert_no_error(err);
/* Create a command queue. */
q = ccl_queue_new(ctx, d, 0, &err);
g_assert_no_error(err);
/* Create regular buffer. */
b = ccl_buffer_new(ctx, CL_MEM_READ_WRITE, buf_size, NULL, &err);
g_assert_no_error(err);
/* Fill buffer with pattern. */
ccl_buffer_enqueue_fill(
b, q, &pattern, sizeof(cl_char8), 0, buf_size, NULL, &err);
g_assert_no_error(err);
/* Read data back to host. */
ccl_buffer_enqueue_read(b, q, CL_TRUE, 0, buf_size, h, NULL, &err);
g_assert_no_error(err);
/* Check data is OK. */
for (guint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
for (guint j = 0; j < 8; ++j)
g_assert_cmpuint(h[i].s[j], ==, pattern.s[j]);
/* Free stuff. */
ccl_buffer_destroy(b);
ccl_queue_destroy(q);
ccl_context_destroy(ctx);
ccl_platforms_destroy(ps);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* @internal
*
* @brief Tests the ccl_buffer_new_from_region() function.
* */
static void create_from_region_test() {
/* Test variables. */
CCLContext * ctx = NULL;
CCLDevice * dev = NULL;
CCLQueue * cq = NULL;
CCLBuffer * buf = NULL;
CCLBuffer * subbuf = NULL;
CCLEvent * evt = NULL;
CCLEventWaitList ewl = NULL;
CCLErr * err = NULL;
cl_ulong * hbuf;
cl_ulong * hsubbuf;
cl_uint min_align;
size_t siz_buf;
size_t siz_subbuf;
/* Get the test context with the pre-defined device. */
ctx = ccl_test_context_new(&err);
g_assert_no_error(err);
/* Get first device in context. */
dev = ccl_context_get_device(ctx, 0, &err);
g_assert_no_error(err);
/* Get minimum alignment for sub-buffer in bits. */
min_align = ccl_device_get_info_scalar(
dev, CL_DEVICE_MEM_BASE_ADDR_ALIGN, cl_uint, &err);
g_assert_no_error(err);
/* Determine buffer and sub-buffer sizes (divide by 64 because its
* the number of bits in cl_ulong). */
siz_subbuf = sizeof(cl_ulong) * min_align / 64;
siz_buf = 4 * siz_subbuf;
/* Allocate memory for host buffer and host sub-buffer. */
hbuf = g_slice_alloc(siz_buf);
hsubbuf = g_slice_alloc(siz_subbuf);
/* Initialize initial host buffer. */
for (cl_uint i = 0; i < siz_buf / sizeof(cl_ulong); ++i)
hbuf[i] = g_test_rand_int();
/* Create a command queue. */
cq = ccl_queue_new(ctx, dev, 0, &err);
g_assert_no_error(err);
/* Create a regular buffer, put some data in it. */
buf = ccl_buffer_new(
ctx, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, siz_buf, hbuf, &err);
g_assert_no_error(err);
/* Create sub-buffer from indexes 16 to 31 (16 positions) of
* original buffer. */
subbuf = ccl_buffer_new_from_region(
buf, 0, siz_subbuf, siz_subbuf, &err);
g_assert_no_error(err);
/* Get data in sub-buffer to a new host buffer. */
evt = ccl_buffer_enqueue_read(
subbuf, cq, CL_FALSE, 0, siz_subbuf, hsubbuf, NULL, &err);
g_assert_no_error(err);
/* Wait for read to be complete. */
ccl_event_wait(ccl_ewl(&ewl, evt, NULL), &err);
g_assert_no_error(err);
/* Check that expected values were successfully read. */
for (cl_uint i = 0; i < siz_subbuf / sizeof(cl_ulong); ++i)
g_assert_cmpuint(
hsubbuf[i], ==, hbuf[i + siz_subbuf / sizeof(cl_ulong)]);
/* Destroy stuff. */
ccl_buffer_destroy(buf);
ccl_buffer_destroy(subbuf);
ccl_queue_destroy(cq);
ccl_context_destroy(ctx);
g_slice_free1(siz_buf, hbuf);
g_slice_free1(siz_subbuf, hsubbuf);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* @internal
*
* @brief Tests basic read/write operations from/to buffer objects.
* */
static void read_write_test() {
/* Test variables. */
CCLContext * ctx = NULL;
CCLDevice * d = NULL;
CCLBuffer * b = NULL;
CCLQueue * q;
cl_uint h_in[CCL_TEST_BUFFER_SIZE];
cl_uint h_out[CCL_TEST_BUFFER_SIZE];
size_t buf_size = sizeof(cl_uint) * CCL_TEST_BUFFER_SIZE;
CCLErr * err = NULL;
/* Create a host array, put some stuff in it. */
for (guint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
h_in[i] = g_test_rand_int();
/* Get the test context with the pre-defined device. */
ctx = ccl_test_context_new(&err);
g_assert_no_error(err);
/* Get first device in context. */
d = ccl_context_get_device(ctx, 0, &err);
g_assert_no_error(err);
/* Create a command queue. */
q = ccl_queue_new(ctx, d, 0, &err);
g_assert_no_error(err);
/* Create regular buffer and write data from the host buffer. */
b = ccl_buffer_new(
ctx, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, buf_size, h_in, &err);
g_assert_no_error(err);
/* Read data back to host. */
ccl_buffer_enqueue_read(
b, q, CL_TRUE, 0, buf_size, (void *) h_out, NULL, &err);
g_assert_no_error(err);
/* Check data is OK. */
for (guint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
g_assert_cmpuint(h_in[i], ==, h_out[i]);
/* Set some other data in host array. */
for (guint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
h_in[i] = g_test_rand_int();
/* Write it explicitly to buffer. */
ccl_buffer_enqueue_write(
b, q, CL_TRUE, 0, buf_size, (void *) h_in, NULL, &err);
g_assert_no_error(err);
/* Read new data to host. */
ccl_buffer_enqueue_read(
b, q, CL_TRUE, 0, buf_size, (void *) h_out, NULL, &err);
g_assert_no_error(err);
/* Check data is OK. */
for (guint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
g_assert_cmpuint(h_in[i], ==, h_out[i]);
/* Free stuff. */
ccl_buffer_destroy(b);
ccl_queue_destroy(q);
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* @internal
*
* @brief Tests copy operations from one buffer to another.
* */
static void copy_test() {
/* Test variables. */
CCLContext * ctx = NULL;
CCLDevice * d = NULL;
CCLBuffer * b1 = NULL;
CCLBuffer * b2 = NULL;
CCLQueue * q;
cl_long h1[CCL_TEST_BUFFER_SIZE];
cl_long h2[CCL_TEST_BUFFER_SIZE];
size_t buf_size = sizeof(cl_long) * CCL_TEST_BUFFER_SIZE;
CCLErr * err = NULL;
/* Create a host array, put some stuff in it. */
for (guint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
h1[i] = g_test_rand_int();
/* Get the test context with the pre-defined device. */
ctx = ccl_test_context_new(&err);
g_assert_no_error(err);
/* Get first device in context. */
d = ccl_context_get_device(ctx, 0, &err);
g_assert_no_error(err);
/* Create a command queue. */
q = ccl_queue_new(ctx, d, 0, &err);
g_assert_no_error(err);
/* Create regular buffer and write data from the host buffer. */
b1 = ccl_buffer_new(
ctx, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, buf_size, h1, &err);
g_assert_no_error(err);
/* Create another buffer, double the size. */
b2 = ccl_buffer_new(ctx, CL_MEM_READ_WRITE, 2 * buf_size, NULL, &err);
g_assert_no_error(err);
/* Copy data from first buffer to second buffer, using an offset on
* the second buffer. */
ccl_buffer_enqueue_copy(
b1, b2, q, 0, buf_size / 2, buf_size, NULL, &err);
g_assert_no_error(err);
/* Read data back to host from the second buffer. */
ccl_buffer_enqueue_read(
b2, q, CL_TRUE, buf_size / 2, buf_size, h2, NULL, &err);
g_assert_no_error(err);
/* Check data is OK. */
for (guint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
g_assert_cmpuint(h1[i], ==, h2[i]);
/* Free stuff. */
ccl_buffer_destroy(b1);
ccl_buffer_destroy(b2);
ccl_queue_destroy(q);
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* Canonical example main function.
* */
int main(int argc, char** argv) {
/* Number of elements in buffer. */
size_t buf_n = DEF_BUF_N;
/* Device selected specified in the command line. */
int dev_idx = -1;
/* Program return value. */
int ret_val;
/* Check if a device was specified in the command line. */
if (argc >= 2) {
dev_idx = atoi(argv[1]);
}
/* Check if a new buffer size was specified in the command line. */
if (argc >= 3) {
buf_n = atoi(argv[2]);
}
/* Wrappers. */
CCLContext* ctx = NULL;
CCLProgram* prg = NULL;
CCLDevice* dev = NULL;
CCLQueue* queue = NULL;
CCLKernel* krnl = NULL;
CCLBuffer* a_dev;
CCLBuffer* b_dev;
CCLBuffer* c_dev;
CCLEvent* evt_write1;
CCLEvent* evt_write2;
CCLEvent* evt_exec;
CCLEventWaitList ewl = NULL;
/* Profiler. */
CCLProf* prof;
/* Global and local worksizes. */
size_t gws = 0;
size_t lws = 0;
/* Host buffers. */
cl_uint* a_host = NULL;
cl_uint* b_host = NULL;
cl_uint* c_host = NULL;
cl_uint d_host;
/* Error reporting object. */
CCLErr* err = NULL;
/* Check results flag. */
cl_bool check_result;
/* Create a context with device selected from menu. */
ctx = ccl_context_new_from_menu_full(&dev_idx, &err);
HANDLE_ERROR(err);
/* Get the selected device. */
dev = ccl_context_get_device(ctx, 0, &err);
HANDLE_ERROR(err);
/* Create a new program from kernel source. */
prg = ccl_program_new_from_source(ctx, KERNEL_SRC, &err);
HANDLE_ERROR(err);
/* Build program. */
ccl_program_build(prg, NULL, &err);
HANDLE_ERROR(err);
/* Create a command queue. */
queue = ccl_queue_new(ctx, dev, CL_QUEUE_PROFILING_ENABLE, &err);
HANDLE_ERROR(err);
/* Get kernel object. */
krnl = ccl_program_get_kernel(prg, KERNEL_NAME, &err);
HANDLE_ERROR(err);
/* Get worksizes. */
lws = ccl_kernel_suggest_worksizes(krnl, dev, 1, &buf_n, &gws, &lws, &err);
HANDLE_ERROR(err);
/* Show worksizes. */
printf("\n");
printf(" * Global worksize: %d\n", (int) gws);
printf(" * Local worksize : %d\n", (int) lws);
/* Initialize host buffers. */
a_host = (cl_uint*) malloc(sizeof(cl_uint) * buf_n);
b_host = (cl_uint*) malloc(sizeof(cl_uint) * buf_n);
c_host = (cl_uint*) malloc(sizeof(cl_uint) * buf_n);
/* Fill host buffers. */
for (cl_uint i = 0; i < buf_n; ++i) {
a_host[i] = i;
b_host[i] = buf_n - i;
}
d_host = buf_n / 4;
/* Create device buffers. */
a_dev = ccl_buffer_new(ctx, CL_MEM_READ_ONLY,
buf_n * sizeof(cl_uint), NULL, &err);
HANDLE_ERROR(err);
b_dev = ccl_buffer_new(ctx, CL_MEM_READ_ONLY,
buf_n * sizeof(cl_uint), NULL, &err);
HANDLE_ERROR(err);
c_dev = ccl_buffer_new(ctx, CL_MEM_WRITE_ONLY,
buf_n * sizeof(cl_uint), NULL, &err);
HANDLE_ERROR(err);
/* Copy host data to device buffers without waiting for transfer
* to terminate before continuing host program. */
evt_write1 = ccl_buffer_enqueue_write(a_dev, queue, CL_FALSE, 0,
buf_n * sizeof(cl_uint), a_host, NULL, &err);
HANDLE_ERROR(err);
evt_write2 = ccl_buffer_enqueue_write(b_dev, queue, CL_FALSE, 0,
buf_n * sizeof(cl_uint), b_host, NULL, &err);
HANDLE_ERROR(err);
/* Initialize event wait list and add the two transfer events. */
ccl_event_wait_list_add(&ewl, evt_write1, evt_write2, NULL);
/* Execute program kernel, waiting for the two transfer events
* to terminate (this will empty the event wait list). */
evt_exec = ccl_program_enqueue_kernel(prg, KERNEL_NAME, queue, 1,
NULL, &gws, &lws, &ewl, &err,
/* Kernel arguments. */
a_dev, b_dev, c_dev,
ccl_arg_priv(d_host, cl_uint), ccl_arg_priv(buf_n, cl_uint),
NULL);
HANDLE_ERROR(err);
/* Add the kernel termination event to the wait list. */
ccl_event_wait_list_add(&ewl, evt_exec, NULL);
/* Sync. queue for events in wait list (just the execute event in
* this case) to terminate before going forward... */
ccl_enqueue_barrier(queue, &ewl, &err);
HANDLE_ERROR(err);
/* Read back results from host waiting for transfer to terminate
* before continuing host program. */
ccl_buffer_enqueue_read(c_dev, queue, CL_TRUE, 0,
buf_n * sizeof(cl_uint), c_host, NULL, &err);
HANDLE_ERROR(err);
/* Check results are as expected (not available with OpenCL stub). */
check_result = CL_TRUE;
for (cl_uint i = 0; i < buf_n; ++i) {
if(c_host[i] != a_host[i] + b_host[i] + d_host) {
check_result = CL_FALSE;
break;
}
}
if (check_result) {
fprintf(stdout, " * Kernel execution produced the expected results.\n");
ret_val = EXIT_SUCCESS;
} else {
fprintf(stderr,
" * Kernel execution failed to produce the expected results.\n");
ret_val = EXIT_FAILURE;
}
/* Perform profiling. */
prof = ccl_prof_new();
ccl_prof_add_queue(prof, "queue1", queue);
ccl_prof_calc(prof, &err);
HANDLE_ERROR(err);
/* Show profiling info. */
ccl_prof_print_summary(prof);
/* Export profiling info. */
ccl_prof_export_info_file(prof, "out.tsv", &err);
HANDLE_ERROR(err);
/* Destroy profiler object. */
ccl_prof_destroy(prof);
/* Destroy host buffers. */
free(a_host);
free(b_host);
free(c_host);
/* Destroy wrappers. */
ccl_buffer_destroy(a_dev);
ccl_buffer_destroy(b_dev);
ccl_buffer_destroy(c_dev);
ccl_queue_destroy(queue);
ccl_program_destroy(prg);
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly freed. */
assert(ccl_wrapper_memcheck());
/* Bye. */
return ret_val;
}
