/**
* @internal
*
* @brief Tests creation, getting info from and destruction of
* buffer wrapper objects.
* */
static void create_info_destroy_test() {
/* Test variables. */
CCLContext * ctx = NULL;
CCLBuffer * b = NULL;
CCLErr * err = NULL;
size_t buf_size = sizeof(cl_uint) * CCL_TEST_BUFFER_SIZE;
/* Get the test context with the pre-defined device. */
ctx = ccl_test_context_new(&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);
/* Get some info and check if the return value is as expected. */
cl_mem_object_type mot;
mot = ccl_memobj_get_info_scalar(
b, CL_MEM_TYPE, cl_mem_object_type, &err);
g_assert_no_error(err);
g_assert_cmphex(mot, ==, CL_MEM_OBJECT_BUFFER);
cl_mem_flags flags;
flags = ccl_memobj_get_info_scalar(
b, CL_MEM_FLAGS, cl_mem_flags, &err);
g_assert_no_error(err);
g_assert_cmphex(flags, ==, CL_MEM_READ_WRITE);
size_t mem_size;
mem_size = ccl_memobj_get_info_scalar(b, CL_MEM_SIZE, size_t, &err);
g_assert_no_error(err);
g_assert_cmpuint(mem_size, ==, buf_size);
void * host_ptr = NULL;
host_ptr = ccl_memobj_get_info_scalar(
b, CL_MEM_HOST_PTR, void*, &err);
g_assert((err == NULL) || (err->code == CCL_ERROR_INFO_UNAVAILABLE_OCL));
g_assert_cmphex(GPOINTER_TO_UINT(host_ptr), ==,
GPOINTER_TO_UINT(NULL));
g_clear_error(&err);
cl_context context;
context = ccl_memobj_get_info_scalar(
b, CL_MEM_CONTEXT, cl_context, &err);
g_assert_no_error(err);
g_assert_cmphex(GPOINTER_TO_UINT(context), ==,
GPOINTER_TO_UINT(ccl_context_unwrap(ctx)));
/* Destroy stuff. */
ccl_buffer_destroy(b);
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* @internal
*
* @brief Tests map/unmap operations in buffer objects.
* */
static void map_unmap_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;
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);
/* Map buffer onto host memory. */
h_out = ccl_buffer_enqueue_map(
b, q, CL_TRUE, CL_MAP_READ, 0, buf_size, NULL, 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]);
/* Unmap buffer. */
ccl_memobj_enqueue_unmap(
(CCLMemObj *) b, q, h_out, NULL, &err);
g_assert_no_error(err);
/* 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 buffer wrapper class reference counting.
* */
static void ref_unref_test() {
/* Test variables. */
CCLContext * ctx = NULL;
CCLBuffer * b = NULL;
CCLErr * err = NULL;
size_t buf_size = sizeof(cl_uint) * CCL_TEST_BUFFER_SIZE;
/* Get the test context with the pre-defined device. */
ctx = ccl_test_context_new(&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);
/* Increase buffer reference count. */
ccl_memobj_ref(b);
/* Check that buffer ref count is 2. */
g_assert_cmpuint(2, ==, ccl_wrapper_ref_count((CCLWrapper *) b));
/* Increase buffer reference count again, this time using helper
* macro. */
ccl_buffer_ref(b);
/* Check that buffer ref count is 3. */
g_assert_cmpuint(3, ==, ccl_wrapper_ref_count((CCLWrapper *) b));
/* Unref buffer, twice. */
ccl_buffer_unref(b);
ccl_buffer_unref(b);
/* Check that buffer ref count is 1. */
g_assert_cmpuint(1, ==, ccl_wrapper_ref_count((CCLWrapper *) b));
/* Destroy stuff. */
ccl_buffer_unref(b);
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* @internal
*
* @brief Test memory object destructor callbacks.
* */
static void destructor_callback_test() {
/* Test variables. */
CCLContext * ctx = NULL;
CCLBuffer * b = NULL;
CCLErr * err = NULL;
GTimer * timer = NULL;
cl_bool test_var = CL_FALSE;
/* Get the test context with the pre-defined device. */
ctx = ccl_test_context_new(&err);
g_assert_no_error(err);
/* Create a buffer. */
b = ccl_buffer_new(
ctx, CL_MEM_READ_WRITE, 128 * sizeof(cl_uint), NULL, &err);
/* Add destructor callback. */
ccl_memobj_set_destructor_callback(
(CCLMemObj *) b, destructor_callback, &test_var, &err);
g_assert_no_error(err);
/* Destroy buffer. */
ccl_buffer_destroy(b);
/* Destroy context. */
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
/* Wait some more... */
timer = g_timer_new();
while (g_timer_elapsed(timer, NULL) < 2.0);
g_timer_stop(timer);
g_timer_destroy(timer);
/* Confirm that test_var is CL_TRUE. */
g_assert_cmpuint(test_var, ==, CL_TRUE);
}
/**
* @internal
* Perform sort using device data.
* */
static CCLEvent* clo_sort_gselect_sort_with_device_data(
CloSort* sorter, CCLQueue* cq_exec, CCLQueue* cq_comm,
CCLBuffer* data_in, CCLBuffer* data_out, size_t numel,
size_t lws_max, GError** err) {
/* Make sure err is NULL or it is not set. */
g_return_val_if_fail(err == NULL || *err == NULL, NULL);
/* Make sure cq_exec is not NULL. */
g_return_val_if_fail(cq_exec != NULL, NULL);
/* Worksizes. */
size_t lws, gws;
/* OpenCL object wrappers. */
CCLContext* ctx = NULL;
CCLDevice* dev = NULL;
CCLKernel* krnl = NULL;
CCLEvent* evt = NULL;
/* Event wait list. */
CCLEventWaitList ewl = NULL;
/* Internal error reporting object. */
GError* err_internal = NULL;
/* Flag indicating if sorted data is to be copied back to original
* buffer, simulating an in-place sort. */
cl_bool copy_back = CL_FALSE;
/* If data transfer queue is NULL, use exec queue for data
* transfers. */
if (cq_comm == NULL) cq_comm = cq_exec;
/* Get device where sort will occurr. */
dev = ccl_queue_get_device(cq_exec, &err_internal);
g_if_err_propagate_goto(err, err_internal, error_handler);
/* Get the kernel wrapper. */
krnl = ccl_program_get_kernel(clo_sort_get_program(sorter),
"gselect", &err_internal);
g_if_err_propagate_goto(err, err_internal, error_handler);
/* Determine worksizes. */
gws = numel;
lws = lws_max;
ccl_kernel_suggest_worksizes(
krnl, dev, 1, &gws, NULL, &lws, &err_internal);
g_if_err_propagate_goto(err, err_internal, error_handler);
/* Check if data_out is set. */
if (data_out == NULL) {
/* If not create it and set the copy back flag to TRUE. */
/* Get context. */
ctx = ccl_queue_get_context(cq_comm, &err_internal);
g_if_err_propagate_goto(err, err_internal, error_handler);
/* Set copy-back flag to true. */
copy_back = CL_TRUE;
/* Create output buffer. */
data_out = ccl_buffer_new(ctx, CL_MEM_WRITE_ONLY,
numel * clo_sort_get_element_size(sorter), NULL,
&err_internal);
g_if_err_propagate_goto(err, err_internal, error_handler);
} else {
/* Set copy back flag to FALSE. */
copy_back = CL_FALSE;
}
/* Set kernel arguments. */
cl_ulong numel_l = numel;
ccl_kernel_set_args(
krnl, data_in, data_out, ccl_arg_priv(numel_l, cl_ulong), NULL);
/* Perform global memory selection sort. */
evt = ccl_kernel_enqueue_ndrange(
krnl, cq_exec, 1, NULL, &gws, &lws, NULL, &err_internal);
g_if_err_propagate_goto(err, err_internal, error_handler);
ccl_event_set_name(evt, "gselect_ndrange");
/* If copy-back flag is set, copy sorted data back to original
* buffer. */
if (copy_back) {
ccl_event_wait_list_add(&ewl, evt, NULL);
evt = ccl_buffer_enqueue_copy(data_out, data_in, cq_comm, 0, 0,
numel * clo_sort_get_element_size(sorter), &ewl,
&err_internal);
g_if_err_propagate_goto(err, err_internal, error_handler);
ccl_event_set_name(evt, "gselect_copy");
}
/* If we got here, everything is OK. */
g_assert(err == NULL || *err == NULL);
goto finish;
error_handler:
/* If we got here there was an error, verify that it is so. */
g_assert(err == NULL || *err != NULL);
evt = NULL;
finish:
/* Free data out buffer if copy-back flag is set. */
if ((copy_back) && (data_out != NULL)) ccl_buffer_destroy(data_out);
/* Return event wait list. */
return evt;
}
/**
* Test RNG with GID-based device generated seeds.
* */
static void seed_dev_gid_test() {
/* Test variables. */
CCLContext* ctx = NULL;
CCLDevice* dev = NULL;
CCLQueue* cq = NULL;
CCLProgram* prg = NULL;
CCLKernel* krnl = NULL;
CCLBuffer* seeds_dev = NULL;
CCLBuffer* output_dev = NULL;
GError* err = NULL;
CloRng* rng = NULL;
size_t lws = 0;
size_t ws = CLO_RNG_TEST_NUM_SEEDS;
gchar* src;
/* Get context and device. */
ctx = ccl_context_new_any(&err);
g_assert_no_error(err);
dev = ccl_context_get_device(ctx, 0, &err);
g_assert_no_error(err);
/* Create command queue. */
cq = ccl_queue_new(ctx, dev, 0, &err);
g_assert_no_error(err);
/* Test all RNGs. */
for (cl_uint i = 0; clo_rng_infos[i].name != NULL; ++i) {
/* Create RNG object. */
rng = clo_rng_new(clo_rng_infos[i].name, CLO_RNG_SEED_DEV_GID,
NULL, CLO_RNG_TEST_NUM_SEEDS, CLO_RNG_TEST_INIT_SEED,
CLO_RNG_TEST_HASH, ctx, cq, &err);
g_assert_no_error(err);
/* Get RNG seeds device buffer. */
seeds_dev = clo_rng_get_device_seeds(rng);
/* Get RNG kernels source. */
src = g_strconcat(
clo_rng_get_source(rng), CLO_RNG_TEST_SRC, NULL);
/* Create and build program. */
prg = ccl_program_new_from_source(ctx, src, &err);
g_assert_no_error(err);
ccl_program_build(prg, NULL, &err);
g_assert_no_error(err);
/* Create output buffer. */
output_dev = ccl_buffer_new(ctx, CL_MEM_WRITE_ONLY,
CLO_RNG_TEST_NUM_SEEDS * sizeof(cl_ulong), NULL, &err);
g_assert_no_error(err);
/* Get kernel from program. */
krnl = ccl_program_get_kernel(prg, CLO_RNG_TEST_KERNEL, &err);
g_assert_no_error(err);
/* Get a "nice" local worksize. */
ccl_kernel_suggest_worksizes(
krnl, dev, 1, &ws, NULL, &lws, &err);
g_assert_no_error(err);
/* Execute kernel. */
ccl_kernel_set_args_and_enqueue_ndrange(
krnl, cq, 1, NULL, &ws, &lws, NULL, &err,
seeds_dev, output_dev, NULL);
g_assert_no_error(err);
/* Release this iteration stuff. */
g_free(src);
ccl_buffer_destroy(output_dev);
ccl_program_destroy(prg);
clo_rng_destroy(rng);
}
/* Destroy queue and context. */
ccl_queue_destroy(cq);
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* Test RNG with client generated seeds in device.
* */
static void seed_ext_dev_test() {
/* Test variables. */
CCLContext* ctx = NULL;
CCLDevice* dev = NULL;
CCLQueue* cq = NULL;
CCLProgram* prg = NULL;
CCLKernel* krnl = NULL;
CCLBuffer* seeds_dev = NULL;
CCLBuffer* output_dev = NULL;
GError* err = NULL;
CloRng* rng = NULL;
size_t lws = 0;
size_t ws = CLO_RNG_TEST_NUM_SEEDS;
gchar* src;
cl_uchar* host_seeds;
/* Get context and device. */
ctx = ccl_context_new_any(&err);
g_assert_no_error(err);
dev = ccl_context_get_device(ctx, 0, &err);
g_assert_no_error(err);
/* Create command queue. */
cq = ccl_queue_new(ctx, dev, 0, &err);
g_assert_no_error(err);
/* Test all RNGs. */
for (cl_uint i = 0; clo_rng_infos[i].name != NULL; ++i) {
/* Host seeds must account for the seed size of current RNG. */
size_t seed_size =
clo_rng_infos[i].seed_size * CLO_RNG_TEST_NUM_SEEDS;
host_seeds = g_slice_alloc(seed_size);
/* Initialize host seeds with any value. */
for (cl_uint i = 0; i < seed_size; ++i)
host_seeds[i] = (cl_uchar) (((i + 1) * 3) & 0xFF);
/* Allocate memory for device seeds and copy host seeds. */
seeds_dev = ccl_buffer_new(
ctx, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, seed_size,
host_seeds, &err);
g_assert_no_error(err);
/* Create RNG object. */
rng = clo_rng_new(clo_rng_infos[i].name, CLO_RNG_SEED_EXT_DEV,
seeds_dev, CLO_RNG_TEST_NUM_SEEDS, CLO_RNG_TEST_INIT_SEED,
NULL, ctx, cq, &err);
g_assert_no_error(err);
/* Get RNG kernels source. */
src = g_strconcat(
clo_rng_get_source(rng), CLO_RNG_TEST_SRC, NULL);
/* Create and build program. */
prg = ccl_program_new_from_source(ctx, src, &err);
g_assert_no_error(err);
ccl_program_build(prg, NULL, &err);
g_assert_no_error(err);
/* Create output buffer. */
output_dev = ccl_buffer_new(ctx, CL_MEM_WRITE_ONLY,
CLO_RNG_TEST_NUM_SEEDS * sizeof(cl_ulong), NULL, &err);
g_assert_no_error(err);
/* Get kernel from program. */
krnl = ccl_program_get_kernel(prg, CLO_RNG_TEST_KERNEL, &err);
g_assert_no_error(err);
/* Get a "nice" local worksize. */
ccl_kernel_suggest_worksizes(
krnl, dev, 1, &ws, NULL, &lws, &err);
g_assert_no_error(err);
/* Execute kernel. */
ccl_kernel_set_args_and_enqueue_ndrange(
krnl, cq, 1, NULL, &ws, &lws, NULL, &err,
seeds_dev, output_dev, NULL);
g_assert_no_error(err);
/* Release this iteration stuff. */
g_slice_free1(seed_size, host_seeds);
g_free(src);
ccl_buffer_destroy(seeds_dev);
ccl_buffer_destroy(output_dev);
ccl_program_destroy(prg);
clo_rng_destroy(rng);
}
/* Destroy queue and context. */
ccl_queue_destroy(cq);
ccl_context_destroy(ctx);
/* Confirm that memory allocated by wrappers has been properly
* freed. */
g_assert(ccl_wrapper_memcheck());
}
/**
* @internal
*
* @brief Tests memory object migration.
* */
static void migrate_test() {
/* Test variables. */
CCLPlatforms * ps;
CCLPlatform * p;
CCLContext * ctx = NULL;
CCLDevice * d = NULL;
CCLBuffer * b = NULL;
CCLQueue * q;
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 associated with first device in
* context. */
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);
/* Assign buffer to first device in context (via the command
* queue). */
ccl_memobj_enqueue_migrate((CCLMemObj **) &b, 1, q, 0, NULL, &err);
g_assert_no_error(err);
/* Migrate buffer to host. */
ccl_memobj_enqueue_migrate(
(CCLMemObj **) &b, 1, q, CL_MIGRATE_MEM_OBJECT_HOST, NULL, &err);
g_assert_no_error(err);
/* Wait for queue to finish... */
ccl_queue_finish(q, &err);
g_assert_no_error(err);
/* 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 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 rect buffer operations.
* */
static void rect_read_write_copy_test() {
/* Test variables. */
CCLContext * ctx = NULL;
CCLDevice * d = NULL;
CCLBuffer * b1 = NULL;
CCLBuffer * b2 = NULL;
CCLQueue * cq;
cl_uchar h1[CCL_TEST_BUFFER_SIZE * CCL_TEST_BUFFER_SIZE];
cl_uchar h2[CCL_TEST_BUFFER_SIZE * CCL_TEST_BUFFER_SIZE];
size_t buf_size = sizeof(cl_uchar) * sizeof(cl_uchar)
* CCL_TEST_BUFFER_SIZE * CCL_TEST_BUFFER_SIZE;
CCLErr * err = NULL;
const size_t origin[] = {0, 0, 0};
const size_t region[] = {CCL_TEST_BUFFER_SIZE * sizeof(cl_uchar),
CCL_TEST_BUFFER_SIZE * sizeof(cl_uchar), 1};
/* Create a "2D" host array, put some stuff in it. */
for (cl_uint i = 0; i < CCL_TEST_BUFFER_SIZE; ++i)
for (cl_uint j = 0; j < CCL_TEST_BUFFER_SIZE; ++j)
h1[i * CCL_TEST_BUFFER_SIZE + j] =
(cl_uchar) (g_test_rand_int() % 0xFF);
/* 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. */
cq = ccl_queue_new(ctx, d, 0, &err);
g_assert_no_error(err);
/* Create device buffers. */
b1 = ccl_buffer_new(ctx, CL_MEM_READ_WRITE, buf_size, NULL, &err);
g_assert_no_error(err);
b2 = ccl_buffer_new(ctx, CL_MEM_READ_WRITE, buf_size, NULL, &err);
g_assert_no_error(err);
/* Write "rect" data to first buffer in device. */
ccl_buffer_enqueue_write_rect(
b1, cq, CL_TRUE, origin, origin, region, 0, 0, 0, 0, h1, NULL, &err);
g_assert_no_error(err);
/* Copy "rect" data from first buffer to second buffer. */
ccl_buffer_enqueue_copy_rect(
b1, b2, cq, origin, origin, region, 0, 0, 0, 0, NULL, &err);
g_assert_no_error(err);
/* Read data "rect" back to host from the second buffer. */
ccl_buffer_enqueue_read_rect(
b2, cq, CL_TRUE, origin, origin, region, 0, 0, 0, 0, h2, NULL, &err);
g_assert_no_error(err);
/* Check data is OK doing a flat comparison. */
for (cl_uint i = 0; i < CCL_TEST_BUFFER_SIZE * 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(cq);
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());
}
/**
* Tests basic read/write operations from/to buffer objects.
* */
static void buffer_read_write() {
/* 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;
GError* 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 a context with any device. */
ctx = ccl_context_new_any(&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());
}
/**
* Tests creation, getting info from and destruction of
* profiler objects, and their relationship with context, device and
* queue wrapper objects.
* */
static void create_add_destroy_test() {
/* Test variables. */
CCLErr* err = NULL;
CCLBuffer* buf1 = NULL;
CCLBuffer* buf2 = NULL;
CCLProf* prof = NULL;
CCLContext* ctx = NULL;
CCLDevice* d = NULL;
CCLQueue* cq1 = NULL;
CCLQueue* cq2 = NULL;
CCLEvent* evt = NULL;
CCLEventWaitList ewl = NULL;
size_t buf_size = 8 * sizeof(cl_short);
cl_short hbuf[8] = {1, 2, 3, 4, 5, 6, 7, 8};
cl_ulong duration, eff_duration;
double time_elapsed;
/* Create a new profile object. */
prof = ccl_prof_new();
/* Get a context and a device. */
ctx = ccl_test_context_new(&err);
g_assert_no_error(err);
d = ccl_context_get_device(ctx, 0, &err);
g_assert_no_error(err);
/* Create two command queue wrappers. */
cq1 = ccl_queue_new(ctx, d, CL_QUEUE_PROFILING_ENABLE, &err);
g_assert_no_error(err);
cq2 = ccl_queue_new(ctx, d, CL_QUEUE_PROFILING_ENABLE, &err);
g_assert_no_error(err);
/* Create device buffers. */
buf1 = ccl_buffer_new(ctx, CL_MEM_READ_ONLY, buf_size, NULL, &err);
g_assert_no_error(err);
buf2 = ccl_buffer_new(ctx, CL_MEM_READ_WRITE, buf_size, NULL, &err);
g_assert_no_error(err);
/* Start profile object timer. */
ccl_prof_start(prof);
/* Transfer data to buffer. */
evt = ccl_buffer_enqueue_write(
buf1, cq1, CL_FALSE, 0, buf_size, hbuf, NULL, &err);
g_assert_no_error(err);
/* Transfer data from one buffer to another. */
evt = ccl_buffer_enqueue_copy(buf1, buf2, cq2, 0, 0, buf_size,
ccl_ewl(&ewl, evt, NULL), &err);
g_assert_no_error(err);
/* Wait for copy. */
ccl_event_wait(ccl_ewl(&ewl, evt, NULL), &err);
g_assert_no_error(err);
/* Stop profile object timer. */
ccl_prof_stop(prof);
/* Add both queues to profile object. */
ccl_prof_add_queue(prof, "A Queue", cq1);
ccl_prof_add_queue(prof, "Another Queue", cq2);
/* Process queues. */
ccl_prof_calc(prof, &err);
g_assert_no_error(err);
/* Request some profiling information. */
time_elapsed = ccl_prof_time_elapsed(prof);
duration = ccl_prof_get_duration(prof);
eff_duration = ccl_prof_get_eff_duration(prof);
g_debug("Profiling time elapsed: %lf", time_elapsed);
g_debug("Profiling duration: %d", (cl_int) duration);
g_debug("Profiling eff. duration: %d", (cl_int) eff_duration);
/* Destroy buffers. */
ccl_buffer_destroy(buf1);
ccl_buffer_destroy(buf2);
/* Unref cq1, which should not be destroyed because it is held
* by the profile object. */
ccl_queue_destroy(cq1);
/* Destroy the profile object, which will also destroy cq1. cq2
* will me merely unrefed and must still be explicitly destroyed. */
ccl_prof_destroy(prof);
/* Destroy cq2. */
ccl_queue_destroy(cq2);
/* Destroy the context. */
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;
}