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;
}
/* 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;
}
/* Can be used to select a 'good' default lws size */
size_t autotune_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(gpu_id) /
(local_memory_size);
else
max_available = get_device_max_lws(gpu_id);
if (max_available > get_kernel_max_lws(gpu_id, crypt_kernel))
return get_kernel_max_lws(gpu_id, crypt_kernel);
return max_available;
}
static void init(struct fmt_main *self)
{
#ifdef CL_VERSION_1_0
char *temp;
cl_ulong maxsize;
global_work_size = 0;
opencl_init("$JOHN/rar_kernel.cl", ocl_gpu_id, platform_id);
// create kernel to execute
crypt_kernel = clCreateKernel(program[ocl_gpu_id], "SetCryptKeys", &ret_code);
HANDLE_CLERROR(ret_code, "Error creating kernel. Double-check kernel name?");
/* We mimic the lengths of cRARk for comparisons */
if (get_device_type(ocl_gpu_id) == CL_DEVICE_TYPE_GPU) {
#ifndef DEBUG
self->params.benchmark_comment = " (6 characters)";
#endif
self->params.tests = gpu_tests;
#if defined(DEBUG) && !defined(ALWAYS_OPENCL)
fprintf(stderr, "Note: will use CPU for some self-tests, and Single mode.\n");
#endif
}
if ((temp = cfg_get_param(SECTION_OPTIONS, SUBSECTION_OPENCL, LWS_CONFIG)))
local_work_size = atoi(temp);
if ((temp = cfg_get_param(SECTION_OPTIONS, SUBSECTION_OPENCL, GWS_CONFIG)))
global_work_size = atoi(temp);
if ((temp = getenv("LWS")))
local_work_size = atoi(temp);
if ((temp = getenv("GWS")))
global_work_size = atoi(temp);
/* Note: we ask for this kernel's max size, not the device's! */
HANDLE_CLERROR(clGetKernelWorkGroupInfo(crypt_kernel, devices[ocl_gpu_id], CL_KERNEL_WORK_GROUP_SIZE, sizeof(maxsize), &maxsize, NULL), "Query max work group size");
#ifdef DEBUG
fprintf(stderr, "Max allowed local work size %d\n", (int)maxsize);
#endif
if (!local_work_size) {
if (get_device_type(ocl_gpu_id) == CL_DEVICE_TYPE_CPU) {
if (get_platform_vendor_id(platform_id) == INTEL)
local_work_size = 8;
else
local_work_size = 1;
} else {
local_work_size = 64;
}
}
if (local_work_size > maxsize) {
fprintf(stderr, "LWS %d is too large for this GPU. Max allowed is %d, using that.\n", (int)local_work_size, (int)maxsize);
local_work_size = maxsize;
}
if (!global_work_size)
find_best_gws(temp == NULL ? 0 : 1);
if (global_work_size < local_work_size)
global_work_size = local_work_size;
fprintf(stderr, "Local worksize (LWS) %d, Global worksize (GWS) %d\n", (int)local_work_size, (int)global_work_size);
create_clobj(global_work_size);
#ifdef DEBUG
{
cl_ulong loc_mem_size;
HANDLE_CLERROR(clGetKernelWorkGroupInfo(crypt_kernel, devices[ocl_gpu_id], CL_KERNEL_LOCAL_MEM_SIZE, sizeof(loc_mem_size), &loc_mem_size, NULL), "Query local memory usage");
fprintf(stderr, "Kernel using %lu bytes of local memory out of %lu available\n", loc_mem_size, get_local_memory_size(ocl_gpu_id));
}
#endif
atexit(release_clobj);
*mkpc = VF * global_work_size;
#endif /* OpenCL */
#if defined (_OPENMP)
omp_t = omp_get_max_threads();
self->params.min_keys_per_crypt *= omp_t;
#ifndef CL_VERSION_1_0 /* OpenCL gets to decide */
*mkpc = omp_t * OMP_SCALE * MAX_KEYS_PER_CRYPT;
#endif
init_locks();
#endif /* _OPENMP */
if (options.utf8)
self->params.plaintext_length = PLAINTEXT_LENGTH * 3;
unpack_data = mem_calloc_tiny(sizeof(unpack_data_t) * omp_t, MEM_ALIGN_WORD);
cracked = mem_calloc_tiny(sizeof(*cracked) * *mkpc, MEM_ALIGN_WORD);
#ifndef CL_VERSION_1_0
saved_key = mem_calloc_tiny(UNICODE_LENGTH * *mkpc, MEM_ALIGN_NONE);
saved_len = mem_calloc_tiny(sizeof(*saved_len) * *mkpc, MEM_ALIGN_WORD);
saved_salt = mem_calloc_tiny(8, MEM_ALIGN_NONE);
aes_key = mem_calloc_tiny(16 * *mkpc, MEM_ALIGN_NONE);
aes_iv = mem_calloc_tiny(16 * *mkpc, MEM_ALIGN_NONE);
#endif
/* OpenSSL init */
init_aesni();
SSL_load_error_strings();
SSL_library_init();
OpenSSL_add_all_algorithms();
#ifndef __APPLE__
atexit(openssl_cleanup);
#endif
/* CRC-32 table init, do it before we start multithreading */
{
CRC32_t crc;
CRC32_Init(&crc);
}
}
static void init(struct fmt_main *self)
{
char *temp;
cl_ulong maxsize, maxsize2;
char build_opts[64];
global_work_size = 0;
snprintf(build_opts, sizeof(build_opts),
"-DHASH_LOOPS=%u -DUNICODE_LENGTH=%u %s",
HASH_LOOPS,
UNICODE_LENGTH,
(options.flags & FLG_VECTORIZE) ? "-DVECTORIZE" :
(options.flags & FLG_SCALAR) ? "-DSCALAR" : "");
opencl_init_opt("$JOHN/office2007_kernel.cl", ocl_gpu_id, platform_id, build_opts);
// create kernel to execute
GenerateSHA1pwhash = clCreateKernel(program[ocl_gpu_id], "GenerateSHA1pwhash", &ret_code);
HANDLE_CLERROR(ret_code, "Error creating kernel. Double-check kernel name?");
crypt_kernel = clCreateKernel(program[ocl_gpu_id], "HashLoop", &ret_code);
HANDLE_CLERROR(ret_code, "Error creating kernel. Double-check kernel name?");
Generate2007key = clCreateKernel(program[ocl_gpu_id], "Generate2007key", &ret_code);
HANDLE_CLERROR(ret_code, "Error creating kernel. Double-check kernel name?");
if (options.flags & FLG_VECTORIZE) {
/* Run vectorized code */
VF = 4;
self->params.algorithm_name = "OpenCL 4x";
}
if ((temp = cfg_get_param(SECTION_OPTIONS, SUBSECTION_OPENCL, LWS_CONFIG)))
local_work_size = atoi(temp);
if ((temp = cfg_get_param(SECTION_OPTIONS, SUBSECTION_OPENCL, GWS_CONFIG)))
global_work_size = atoi(temp);
if ((temp = getenv("LWS")))
local_work_size = atoi(temp);
if ((temp = getenv("GWS")))
global_work_size = atoi(temp);
/* Note: we ask for the kernels' max sizes, not the device's! */
HANDLE_CLERROR(clGetKernelWorkGroupInfo(GenerateSHA1pwhash, devices[ocl_gpu_id], CL_KERNEL_WORK_GROUP_SIZE, sizeof(maxsize), &maxsize, NULL), "Query max work group size");
HANDLE_CLERROR(clGetKernelWorkGroupInfo(crypt_kernel, devices[ocl_gpu_id], CL_KERNEL_WORK_GROUP_SIZE, sizeof(maxsize2), &maxsize2, NULL), "Query max work group size");
if (maxsize2 < maxsize) maxsize = maxsize2;
HANDLE_CLERROR(clGetKernelWorkGroupInfo(Generate2007key, devices[ocl_gpu_id], CL_KERNEL_WORK_GROUP_SIZE, sizeof(maxsize2), &maxsize2, NULL), "Query max work group size");
if (maxsize2 < maxsize) maxsize = maxsize2;
#if 0
/* Our use of local memory sets a limit for LWS */
maxsize2 = get_local_memory_size(ocl_gpu_id) / (24 * VF);
while (maxsize > maxsize2)
maxsize >>= 1;
#endif
/* maxsize is the lowest figure from the three different kernels */
if (!local_work_size) {
if (getenv("LWS")) {
/* LWS was explicitly set to 0 */
int temp = global_work_size;
local_work_size = maxsize;
global_work_size = global_work_size ? global_work_size : 4 * maxsize;
create_clobj(global_work_size, self);
opencl_find_best_workgroup_limit(self, maxsize);
release_clobj();
global_work_size = temp;
} else {
if (cpu(device_info[ocl_gpu_id])) {
if (get_platform_vendor_id(platform_id) == DEV_INTEL)
local_work_size = MIN(maxsize, 8);
else
local_work_size = 1;
} else
local_work_size = MIN(maxsize, 64);
}
}
if (local_work_size > maxsize) {
fprintf(stderr, "LWS %d is too large for this GPU. Max allowed is %d, using that.\n", (int)local_work_size, (int)maxsize);
local_work_size = maxsize;
}
if (!global_work_size)
find_best_gws(getenv("GWS") == NULL ? 0 : 1, self);
if (global_work_size < local_work_size)
global_work_size = local_work_size;
fprintf(stderr, "Local worksize (LWS) %d, Global worksize (GWS) %d\n", (int)local_work_size, (int)global_work_size);
create_clobj(global_work_size, self);
atexit(release_clobj);
if (options.utf8)
self->params.plaintext_length = MIN(125, 3 * PLAINTEXT_LENGTH);
}
