int main(int argc, const char * argv[]) {
static mach_timebase_info_data_t sTimebaseInfo;
int seq = 0;
int cpu = 0;
// insert code here...
int* data = (int*) malloc(sizeof(cl_int)*RANGE);
for (int i = 0; i < RANGE; i ++) {
data[i] = START + i;
//printf("%d",data[i]);
}
int* out = (int*) malloc(sizeof(cl_int)*RANGE);
printf("Hello, World!\n");
if (argc > 1){
if (strncmp(argv[1], "seq", 3) == 0) {
seq = TRUE;
} else if (strncmp(argv[1], "cpu", 3) == 0) {
cpu = 1;
}
}
uint64_t start = mach_absolute_time();
if (seq)
euler_totient(data, out);
else {
dispatch_queue_t queue;
if (cpu)
queue = gcl_create_dispatch_queue(CL_DEVICE_TYPE_CPU, NULL);
else
queue = gcl_create_dispatch_queue(CL_DEVICE_TYPE_GPU, NULL);
void *mem_in = gcl_malloc(sizeof(cl_int)*RANGE, data, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR);
void *mem_out = gcl_malloc(sizeof(cl_int)*RANGE, NULL, CL_MEM_WRITE_ONLY);
dispatch_sync(queue, ^{
size_t wgs;
gcl_get_kernel_block_workgroup_info(euler_totient_kernel,
CL_KERNEL_WORK_GROUP_SIZE,
sizeof(wgs), &wgs, NULL);
cl_ndrange range = {
1,
{0, 0, 0},
{RANGE, 0, 0},
{wgs, 0, 0}
};
euler_totient_kernel(&range,(cl_int*)mem_in, (cl_int*)mem_out);
gcl_memcpy(out, mem_out, sizeof(cl_float) * RANGE);
});
}
int main (int argc, const char * argv[]) {
int i, j;
char name[128];
float *a, *b, *c;
void *gpu_a, *gpu_b, *gpu_c;
int ARRAY_SIZE = 128;
dispatch_queue_t queue = gcl_create_dispatch_queue(CL_DEVICE_TYPE_GPU, NULL);
if (queue == NULL) {
queue = gcl_create_dispatch_queue(CL_DEVICE_TYPE_CPU, NULL);
fprintf(stdout, "Created a dispatch queue failed\n", name);
}
cl_device_id gpu = gcl_get_device_id_with_dispatch_queue(queue);
clGetDeviceInfo(gpu, CL_DEVICE_NAME, 128, name, NULL);
fprintf(stdout, "Created a dispatch queue using the %s\n", name);
if(argc >= 2)
ARRAY_SIZE = atoi(argv[1]);
a = (float *)malloc(ARRAY_SIZE*ARRAY_SIZE*sizeof(cl_float));
if(a == NULL)
fprintf(stderr,"allocating array a failed\n");
for(i = 0; i < ARRAY_SIZE*ARRAY_SIZE; i++)
a[i] = (cl_float)rand();
b = (float *)malloc(ARRAY_SIZE*ARRAY_SIZE*sizeof(cl_float));
if(b == NULL)
fprintf(stderr,"allocating array b failed\n");
for(i = 0; i < ARRAY_SIZE*ARRAY_SIZE; i++)
b[i] = (cl_float)rand();
c = (float *)malloc(ARRAY_SIZE*ARRAY_SIZE*sizeof(cl_float));
if(c == NULL)
fprintf(stderr,"allocating array c failed\n");
gpu_a = gcl_malloc(sizeof(cl_float) * ARRAY_SIZE*ARRAY_SIZE, a,
CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR);
gpu_b = gcl_malloc(sizeof(cl_float) * ARRAY_SIZE*ARRAY_SIZE, b,
CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR);
gpu_c = gcl_malloc(sizeof(cl_float) * ARRAY_SIZE*ARRAY_SIZE, NULL,
CL_MEM_WRITE_ONLY);
dispatch_sync(queue, ^{
cl_ndrange range = { 2, {0, 0, 0},{ARRAY_SIZE, ARRAY_SIZE, 0},{16, 16, 0}};
matrix_mul_kernel(&range,(cl_float*)gpu_a,(cl_float*)gpu_b,(cl_float*)gpu_c,(cl_int)ARRAY_SIZE);
gcl_memcpy(c, gpu_c, sizeof(cl_float) * ARRAY_SIZE*ARRAY_SIZE);
});
device_mem OpenCL::malloc(size_t size, void* host_ptr, mem_flags perm) {
#ifdef __APPLE__
void* buff;
buff = gcl_malloc(size, host_ptr, perm);
#else
cl_mem buff;
SAFE_REF(buff = clCreateBuffer(context, perm, size, host_ptr, &err));
#endif
/* FIXME: Should this point to the cl_mem object, or should it just be the
* cl_mem object itself (which internally is the same as _cl_mem*) ?
*/
// device_mem
return {(uintptr_t)buff, sizeof(cl_mem)};
}
int main(int argc, const char * argv[])
{
int i;
char deviceName[128];
if (init() != true)
{
return 1;
}
SDL_Delay(2000);
// try to get the dispatch queue for the GPU
dispatch_queue_t queue = gcl_create_dispatch_queue(CL_DEVICE_TYPE_GPU, NULL);
// in the event that the system does not have an OpenCL GPU, we can use the CPU instead
if (queue == NULL)
{
queue = gcl_create_dispatch_queue(CL_DEVICE_TYPE_CPU, NULL);
}
// let's print some data on the device we're usng! かわいいです！
cl_device_id gpu = gcl_get_device_id_with_dispatch_queue(queue);
clGetDeviceInfo(gpu, CL_DEVICE_NAME, 128, deviceName, NULL);
fprintf(stdout, "Created a dispatch queue using the %s\n", deviceName);
// let's hardcode some handy test data that's easy to understand
float* test_in = (float*)malloc(sizeof(cl_float) * NUM_VALUES);
for (i = 0; i < NUM_VALUES; i++)
{
test_in[i] = (cl_float)i;
}
// Once the computation using the CL is done, we'll need space in RAM for the output
float* test_out = (float*)malloc(sizeof(cl_float) * NUM_VALUES);
// Now we're going to allocate the buffers again in the OpenCL device's memory space
// CL_MEM_COPY_HOST_PTR will copy the values of test_in to mem_in
void* mem_in = gcl_malloc(sizeof(cl_float) * NUM_VALUES, test_in, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR);
void* mem_out = gcl_malloc(sizeof(cl_float) * NUM_VALUES, NULL, CL_MEM_WRITE_ONLY);
// DISPATCH THE KERNEL PROGRAM
dispatch_sync(queue, ^{
//workgroup size, I think
size_t wgs;
//information on sizing of dimensions
gcl_get_kernel_block_workgroup_info(square_kernel, CL_KERNEL_WORK_GROUP_SIZE, sizeof(wgs), &wgs, NULL);
cl_ndrange range = {
1,
{0, 0, 0},
{NUM_VALUES, 0, 0},
{wgs, 0, 0}
};
// call the kernel
square_kernel(&range, (cl_float*)mem_in, (cl_float*)mem_out);
//copy the output into memory
gcl_memcpy(test_out, mem_out, sizeof(cl_float) * NUM_VALUES);
});
