void writer_run(const reader_t* reader)
{
context_t context;
context_initialize(&context);
calc_work(&context, reader);
allocate_work(&context);
read(&context, &reader);
context_finalize(&context);
}
int main (int argc, char *argv[])
{
uint64_t skip = 0;
uint64_t left = -1;
if (argc >= 2) skip = atoll (argv[1]);
if (argc >= 3) left = atoll (argv[2]);
printf ("Loading Kernel...\n");
const char *filename = KERNEL_SRC;
struct stat s;
if (stat (filename, &s) == -1)
{
fprintf (stderr, "%s: %s in line %d\n", filename, strerror (errno), __LINE__);
return (-1);
}
FILE *fp = fopen (filename, "rb");
if (fp == NULL)
{
fprintf (stderr, "%s: %s in line %d\n", filename, strerror (errno), __LINE__);
return (-1);
}
char *source_buf = (char *) malloc (s.st_size + 1);
if (!fread (source_buf, sizeof (char), s.st_size, fp))
{
fprintf (stderr, "%s: %s in line %d\n", filename, strerror (errno), __LINE__);
return (-1);
}
source_buf[s.st_size] = 0;
fclose (fp);
const char *sourceBuf[] = { source_buf };
const size_t sourceLen[] = { s.st_size + 1 };
printf ("Initializing OpenCL...\n");
cl_platform_id platform;
cl_uint num_devices = 0;
cl_device_id devices[MAX_PLATFORM];
gc_clGetPlatformIDs (1, &platform, NULL);
gc_clGetDeviceIDs (platform, DEV_TYPE, MAX_PLATFORM, devices, &num_devices);
gpu_ctx_t gpu_ctxs[MAX_GPU];
memset (gpu_ctxs, 0, sizeof (gpu_ctxs));
for (cl_uint device_id = 0; device_id < num_devices; device_id++)
{
cl_device_id device = devices[device_id];
cl_context context = gc_clCreateContext (NULL, 1, &device, NULL, NULL);
cl_program program = gc_clCreateProgramWithSource (context, 1, sourceBuf, sourceLen);
gc_clBuildProgram (program, 1, &device, BUILD_OPTS, NULL, NULL);
cl_kernel kernel = gc_clCreateKernel (program, KERNEL_NAME);
cl_command_queue command_queue = gc_clCreateCommandQueue (context, device, 0);
cl_uint max_compute_units;
gc_clGetDeviceInfo (device, CL_DEVICE_MAX_COMPUTE_UNITS, sizeof (max_compute_units), &max_compute_units, NULL);
char device_name[BUFSIZ];
memset (device_name, 0, sizeof (device_name));
gc_clGetDeviceInfo (device, CL_DEVICE_NAME, sizeof (device_name), &device_name, NULL);
printf ("Found new device #%2d: %s, %u compute units\n", device_id, device_name, max_compute_units);
const int num_threads = GPU_THREADS;
const int num_elements = max_compute_units * num_threads * GPU_ACCEL;
/**
* GPU memory
*/
const size_t size_block = num_elements * sizeof (block_t);
const size_t size_results = num_threads * sizeof (uint32_t);
cl_mem d_block = gc_clCreateBuffer (context, CL_MEM_READ_ONLY, size_block, NULL);
cl_mem d_results = gc_clCreateBuffer (context, CL_MEM_WRITE_ONLY, size_results, NULL);
gc_clSetKernelArg (kernel, 0, sizeof (cl_mem), (void *) &d_block);
gc_clSetKernelArg (kernel, 1, sizeof (cl_mem), (void *) &d_results);
/**
* Host memory
*/
block_t *h_block = (block_t *) malloc (size_block);
uint32_t *h_results = (uint32_t *) malloc (size_results);
memset (h_results, 0xff, size_results);
gc_clEnqueueWriteBuffer (command_queue, d_results, CL_TRUE, 0, size_results, h_results, 0, NULL, NULL);
/**
* Buffers for candidates
*/
uint8_t **plains_buf = (uint8_t **) calloc (num_elements * VECT_SIZE, sizeof (uint8_t *));
for (int i = 0; i < num_elements * VECT_SIZE; i++)
{
/* Agreed, this is not nice. But who cares nowadays? */
plains_buf[i] = (uint8_t *) malloc (MAX_LINELEN);
}
size_t *plains_len = (size_t *) calloc (num_elements * VECT_SIZE, sizeof (size_t));
gpu_ctx_t *gpu_ctx = &gpu_ctxs[device_id];
gpu_ctx->context = context;
gpu_ctx->program = program;
gpu_ctx->kernel = kernel;
gpu_ctx->command_queue = command_queue;
gpu_ctx->max_compute_units = max_compute_units;
gpu_ctx->d_block = d_block;
gpu_ctx->d_results = d_results;
gpu_ctx->h_block = h_block;
gpu_ctx->h_results = h_results;
gpu_ctx->num_threads = num_threads;
gpu_ctx->num_elements = num_elements;
gpu_ctx->plains_buf = plains_buf;
gpu_ctx->plains_len = plains_len;
}
/* static salt */
const uint8_t salt_buf[16] =
{
0x97, 0x48, 0x6C, 0xAA,
0x22, 0x5F, 0xE8, 0x77,
0xC0, 0x35, 0xCC, 0x03,
0x73, 0x23, 0x6D, 0x51
};
const size_t salt_len = sizeof (salt_buf);
/* main loop */
printf ("Initialization done, accepting candidates from stdin...\n\n");
cl_uint cur_device_id = 0;
while (!feof (stdin))
{
/* Get new password candidate from stdin */
uint8_t line_buf[MAX_LINELEN];
int cur_c = 0;
int prev_c = 0;
size_t line_len = 0;
for (size_t i = 0; i < MAX_LINELEN - 100; i++) // - 100 = we need some space for salt and padding
{
cur_c = getchar ();
if (cur_c == EOF) break;
if ((prev_c == '\n') && (cur_c == '\0'))
{
line_len--;
break;
}
line_buf[line_len] = cur_c;
line_len++;
prev_c = cur_c;
}
/* chop \r if it exists for some reason (in case user used a dictionary) */
if (line_len >= 2)
{
if ((prev_c == '\r') && (cur_c == '\0')) line_len -= 2;
}
/* skip empty lines */
if (line_len == 0) continue;
/* The following enables distributed computing / resume work */
if (skip)
{
skip--;
continue;
}
if (left)
{
left--;
}
else
{
break;
}
/* Append constant salt */
memcpy (line_buf + line_len, salt_buf, salt_len);
line_len += salt_len;
/* Generate digest out of it */
uint32_t digest[4];
md5_transform ((uint32_t *) line_buf, (uint32_t) line_len, digest);
/* Next garanteed free GPU */
gpu_ctx_t *gpu_ctx = &gpu_ctxs[cur_device_id];
/* Save original buffer in case it cracks it */
memcpy (gpu_ctx->plains_buf[gpu_ctx->num_cached], line_buf, line_len - salt_len);
gpu_ctx->plains_len[gpu_ctx->num_cached] = line_len - salt_len;
/* Next garanteed free memory element on that GPU */
const uint32_t element_div = gpu_ctx->num_cached / 4;
const uint32_t element_mod = gpu_ctx->num_cached % 4;
/* Copy new digest */
gpu_ctx->h_block[element_div].A[element_mod] = digest[0];
gpu_ctx->h_block[element_div].B[element_mod] = digest[1];
gpu_ctx->h_block[element_div].C[element_mod] = digest[2];
gpu_ctx->h_block[element_div].D[element_mod] = digest[3];
gpu_ctx->num_cached++;
/* If memory elements on that GPU are full, switch to the next GPU */
if ((gpu_ctx->num_cached / VECT_SIZE) < gpu_ctx->num_elements) continue;
cur_device_id++;
/* If there is no more GPU left, run the calculation */
if (cur_device_id < num_devices) continue;
/* Fire! */
calc_work (num_devices, gpu_ctxs);
launch_kernel (num_devices, gpu_ctxs);
/* Collecting data has a blocking effect */
check_results (num_devices, gpu_ctxs);
/* Reset buffer state */
for (cl_uint device_id = 0; device_id < num_devices; device_id++)
{
gpu_ctx_t *gpu_ctx = &gpu_ctxs[device_id];
gpu_ctx->num_cached = 0;
}
cur_device_id = 0;
}
/* Final calculation of leftovers */
calc_work (num_devices, gpu_ctxs);
launch_kernel (num_devices, gpu_ctxs);
check_results (num_devices, gpu_ctxs);
return -1;
}
