static void init(struct fmt_main *self)
{
cl_ulong maxsize;
size_t selected_gws;
opencl_init_opt("$JOHN/kernels/pwsafe_kernel.cl", ocl_gpu_id, NULL);
init_kernel = clCreateKernel(program[ocl_gpu_id], KERNEL_INIT_NAME, &ret_code);
HANDLE_CLERROR(ret_code, "Error while creating init kernel");
crypt_kernel = clCreateKernel(program[ocl_gpu_id], KERNEL_RUN_NAME, &ret_code);
HANDLE_CLERROR(ret_code, "Error while creating crypt kernel");
finish_kernel = clCreateKernel(program[ocl_gpu_id], KERNEL_FINISH_NAME, &ret_code);
HANDLE_CLERROR(ret_code, "Error while creating finish kernel");
local_work_size = cpu(device_info[ocl_gpu_id]) ? 1 : 64;
global_work_size = 0;
opencl_get_user_preferences(CONFIG_NAME);
//Initialize openCL tuning (library) for this format.
opencl_init_auto_setup(STEP, ROUNDS_DEFAULT/8, 8, split_events,
warn, &multi_profilingEvent[3], self, create_clobj, release_clobj,
sizeof(pwsafe_pass), 0);
self->methods.crypt_all = crypt_all_benchmark;
selected_gws = global_work_size;
/* Note: we ask for the kernels' max sizes, not the device's! */
maxsize = get_current_work_group_size(ocl_gpu_id, init_kernel);
maxsize = MIN(get_current_work_group_size(ocl_gpu_id, crypt_kernel),
maxsize);
maxsize = MIN(get_current_work_group_size(ocl_gpu_id, finish_kernel),
maxsize);
while (local_work_size > maxsize)
local_work_size >>= 1;
self->params.max_keys_per_crypt = (global_work_size ? global_work_size: MAX_KEYS_PER_CRYPT);
if (!local_work_size) {
create_clobj(self->params.max_keys_per_crypt, self);
find_best_lws(self, ocl_gpu_id);
release_clobj();
}
global_work_size = selected_gws;
if (global_work_size)
create_clobj(global_work_size, self);
else
//user chose to die of boredom
find_best_gws(self, ocl_gpu_id);
self->params.min_keys_per_crypt = local_work_size;
self->params.max_keys_per_crypt = global_work_size;
self->methods.crypt_all = crypt_all;
if (options.verbosity > 2)
fprintf(stderr, "Local worksize (LWS) %d, Global worksize (GWS) %d\n", (int)local_work_size, (int)global_work_size);
}
static size_t get_task_max_work_group_size(){
size_t max_available;
max_available = get_max_work_group_size(ocl_gpu_id);
if (max_available > get_current_work_group_size(ocl_gpu_id, crypt_kernel))
return get_current_work_group_size(ocl_gpu_id, crypt_kernel);
return max_available;
}
/* Can be used to select a 'good' default lws size */
size_t common_get_task_max_work_group_size(int use_local_memory,
int local_memory_size, cl_kernel crypt_kernel) {
size_t max_available;
if (use_local_memory)
max_available = get_local_memory_size(ocl_gpu_id) /
(local_memory_size);
else
max_available = get_max_work_group_size(ocl_gpu_id);
if (max_available > get_current_work_group_size(ocl_gpu_id, crypt_kernel))
return get_current_work_group_size(ocl_gpu_id, crypt_kernel);
return max_available;
}
static size_t get_task_max_work_group_size(){
size_t max_available;
if (use_local(source_in_use))
max_available = get_local_memory_size(ocl_gpu_id) /
(sizeof(sha512_password) + sizeof(sha512_ctx) +
sizeof(sha512_buffers)) - 1;
else if (gpu(source_in_use))
max_available = get_local_memory_size(ocl_gpu_id) /
sizeof(sha512_password);
else
max_available = get_max_work_group_size(ocl_gpu_id);
if (max_available > get_current_work_group_size(ocl_gpu_id, crypt_kernel))
return get_current_work_group_size(ocl_gpu_id, crypt_kernel);
return max_available;
}
static size_t get_task_max_size(){
size_t max_available;
max_available = get_max_compute_units(ocl_gpu_id);
if (cpu(device_info[ocl_gpu_id]))
return max_available * KEYS_PER_CORE_CPU;
else
return max_available * get_current_work_group_size(ocl_gpu_id, crypt_kernel) * 2;
}
/* Can be used to select a 'good' default gws size */
size_t common_get_task_max_size(int multiplier, int keys_per_core_cpu,
int keys_per_core_gpu, cl_kernel crypt_kernel) {
size_t max_available;
max_available = get_max_compute_units(ocl_gpu_id);
if (cpu(device_info[ocl_gpu_id]))
return max_available * keys_per_core_cpu;
else
return max_available * multiplier * keys_per_core_gpu *
get_current_work_group_size(ocl_gpu_id, crypt_kernel);
}
/* --
This function could be used to calculated the best num
for the workgroup
Work-items that make up a work-group (also referred to
as the size of the work-group)
-- */
static void find_best_lws(struct fmt_main * self, int sequential_id) {
//Call the default function.
cl_kernel tKernel = init_kernel;
size_t largest = 0;
size_t temp = get_current_work_group_size(ocl_gpu_id, init_kernel);
largest = temp;
temp = get_current_work_group_size(ocl_gpu_id, crypt_kernel);
if(temp > largest)
{
largest = temp;
tKernel = crypt_kernel;
}
temp = get_current_work_group_size(ocl_gpu_id, finish_kernel);
if(temp > largest)
{
largest = temp;
tKernel = finish_kernel;
}
common_find_best_lws(
largest,
sequential_id, tKernel
);
}
