101,
111,
1711,
- 3000, // broken in osx
+ 3000,
1000,
1100,
2100,
12800,
- 1500, // broken in osx
+ 1500,
12400,
500,
3200,
10410,
10500,
10600,
- 10700, // broken in osx
+ 10700,
9000,
5200,
6800,
device_param->kernel_params_buf32[26] = kernel_loops;
device_param->kernel_params_buf32[27] = kernel_loops;
+ // init some fake words
+
+ if (data.attack_kern == ATTACK_KERN_BF)
+ {
+ run_kernel_mp (KERN_RUN_MP_L, device_param, kernel_power);
+ run_kernel_mp (KERN_RUN_MP_R, device_param, kernel_loops);
+ }
+ else
+ {
+ for (u32 i = 0; i < kernel_power; i++)
+ {
+ device_param->pws_buf[i].pw_len = i & 7;
+ }
+
+ hc_clEnqueueWriteBuffer (data.ocl, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, kernel_power * sizeof (pw_t), device_param->pws_buf, 0, NULL, NULL);
+
+ if (data.attack_exec == ATTACK_EXEC_OUTSIDE_KERNEL)
+ {
+ run_kernel_amp (device_param, kernel_power);
+ }
+ }
+
// caching run
if (data.attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
const double exec_ms_prev = get_avg_exec_time (device_param, repeat);
+ // reset fake words
+
+ memset (device_param->pws_buf, 0, kernel_power * sizeof (pw_t));
+
+ hc_clEnqueueWriteBuffer (data.ocl, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, kernel_power * sizeof (pw_t), device_param->pws_buf, 0, NULL, NULL);
+ hc_clEnqueueWriteBuffer (data.ocl, device_param->command_queue, device_param->d_pws_amp_buf, CL_TRUE, 0, kernel_power * sizeof (pw_t), device_param->pws_buf, 0, NULL, NULL);
+
return exec_ms_prev;
}
{
const double target_ms = TARGET_MS_PROFILE[data.workload_profile - 1];
- u32 kernel_loops_min = device_param->kernel_loops_min;
- u32 kernel_loops_max = device_param->kernel_loops_max;
+ const u32 kernel_accel_min = device_param->kernel_accel_min;
+ const u32 kernel_accel_max = device_param->kernel_accel_max;
- u32 kernel_accel_min = device_param->kernel_accel_min;
- u32 kernel_accel_max = device_param->kernel_accel_max;
+ const u32 kernel_loops_min = device_param->kernel_loops_min;
+ const u32 kernel_loops_max = device_param->kernel_loops_max;
- u32 kernel_loops = kernel_loops_min;
u32 kernel_accel = kernel_accel_min;
+ u32 kernel_loops = kernel_loops_min;
- // init some fake words
-
- const u32 kernel_power_max = device_param->device_processors * device_param->kernel_threads * device_param->kernel_accel_max;
+ // steps
- for (u32 i = 0; i < kernel_power_max; i++)
- {
- device_param->pws_buf[i].pw_len = 8;
- }
+ #define STEPS_CNT 10
- hc_clEnqueueWriteBuffer (data.ocl, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, device_param->size_pws, device_param->pws_buf, 0, NULL, NULL);
- hc_clEnqueueWriteBuffer (data.ocl, device_param->command_queue, device_param->d_pws_amp_buf, CL_TRUE, 0, device_param->size_pws, device_param->pws_buf, 0, NULL, NULL);
+ #define STEPS_ACCEL_CNT (STEPS_CNT + 2)
+ #define STEPS_LOOPS_CNT (STEPS_CNT + 2)
- // steps for loops
+ u32 steps_accel[STEPS_ACCEL_CNT];
+ u32 steps_loops[STEPS_LOOPS_CNT];
- #define STEPS_LOOPS_CNT 15
+ for (int i = 0; i < STEPS_ACCEL_CNT; i++)
+ {
+ steps_accel[i] = 1 << i;
+ }
- u32 steps_loops[STEPS_LOOPS_CNT];
+ for (int i = 0; i < STEPS_LOOPS_CNT; i++)
+ {
+ steps_loops[i] = 1 << i;
+ }
- steps_loops[ 0] = 1;
- steps_loops[ 1] = 2;
- steps_loops[ 2] = 4;
- steps_loops[ 3] = 8;
- steps_loops[ 4] = 16;
- steps_loops[ 5] = 32;
- steps_loops[ 6] = 64;
- steps_loops[ 7] = 100;
- steps_loops[ 8] = 128;
- steps_loops[ 9] = 200;
- steps_loops[10] = 256;
- steps_loops[11] = 500;
- steps_loops[12] = 512;
- steps_loops[13] = 1000;
- steps_loops[14] = 1024;
-
- // steps for accel
-
- #define STEPS_ACCEL_CNT 13
+ steps_accel[STEPS_CNT + 0] = kernel_accel_min;
+ steps_accel[STEPS_CNT + 1] = kernel_accel_max;
- u32 steps_accel[STEPS_ACCEL_CNT];
+ steps_loops[STEPS_CNT + 0] = kernel_loops_min;
+ steps_loops[STEPS_CNT + 1] = kernel_loops_max;
- steps_accel[ 0] = 1;
- steps_accel[ 1] = 2;
- steps_accel[ 2] = 4;
- steps_accel[ 3] = 8;
- steps_accel[ 4] = 16;
- steps_accel[ 5] = 32;
- steps_accel[ 6] = 64;
- steps_accel[ 7] = 128;
- steps_accel[ 8] = 256;
- steps_accel[ 9] = 384;
- steps_accel[10] = 512;
- steps_accel[11] = 768;
- steps_accel[12] = 1024;
+ qsort (steps_accel, STEPS_ACCEL_CNT, sizeof (u32), sort_by_u32);
+ qsort (steps_loops, STEPS_LOOPS_CNT, sizeof (u32), sort_by_u32);
// find out highest kernel-loops that stays below target_ms, we can use it later for multiplication as this is a linear function
u32 kernel_loops_tmp;
- for (kernel_loops_tmp = MIN (kernel_loops_max, 200); kernel_loops_tmp >= kernel_loops_min; kernel_loops_tmp >>= 1)
+ for (kernel_loops_tmp = kernel_loops_max; kernel_loops_tmp > kernel_loops_min; kernel_loops_tmp >>= 1)
{
const double exec_ms = try_run (device_param, kernel_accel_min, kernel_loops_tmp, 1);
if (exec_ms < target_ms) break;
-
- if (kernel_loops_tmp == kernel_loops_min) break;
}
// kernel-accel
- double e_best = 0;
-
- for (int i = 0; i < STEPS_ACCEL_CNT; i++)
+ if (kernel_accel_min < kernel_accel_max)
{
- const u32 kernel_accel_try = steps_accel[i];
+ double e_best = 0;
- if (kernel_accel_try < kernel_accel_min) continue;
- if (kernel_accel_try > kernel_accel_max) break;
+ for (int i = 0; i < STEPS_ACCEL_CNT; i++)
+ {
+ const u32 kernel_accel_try = steps_accel[i];
- const double exec_ms = try_run (device_param, kernel_accel_try, kernel_loops_tmp, 1);
+ if (kernel_accel_try < kernel_accel_min) continue;
+ if (kernel_accel_try > kernel_accel_max) break;
- if (exec_ms > target_ms) break;
+ const double exec_ms = try_run (device_param, kernel_accel_try, kernel_loops_tmp, 1);
- const double e = kernel_accel_try / exec_ms;
+ if (exec_ms > target_ms) break;
- if (e > e_best)
- {
- kernel_accel = kernel_accel_try;
+ const double e = kernel_accel_try / exec_ms;
+
+ if (e > e_best)
+ {
+ kernel_accel = kernel_accel_try;
- e_best = e;
+ e_best = e;
+ }
}
}
// kernel-loops final
- e_best = 0;
-
- for (int i = 0; i < STEPS_LOOPS_CNT; i++)
+ if (kernel_loops_min < kernel_loops_max)
{
- const u32 kernel_loops_try = steps_loops[i];
+ double e_best = 0;
- if (kernel_loops_try < kernel_loops_min) continue;
- if (kernel_loops_try > kernel_loops_max) break;
+ for (int i = 0; i < STEPS_LOOPS_CNT; i++)
+ {
+ const u32 kernel_loops_try = steps_loops[i];
- const double exec_ms = try_run (device_param, kernel_accel, kernel_loops_try, 1);
+ if (kernel_loops_try < kernel_loops_min) continue;
+ if (kernel_loops_try > kernel_loops_max) break;
- if (exec_ms > target_ms) break;
+ const double exec_ms = try_run (device_param, kernel_accel, kernel_loops_try, 1);
- const double e = kernel_loops_try / exec_ms;
+ if (exec_ms > target_ms) break;
- if (e > e_best)
- {
- kernel_loops = kernel_loops_try;
+ const double e = kernel_loops_try / exec_ms;
- e_best = e;
+ if (e > e_best)
+ {
+ kernel_loops = kernel_loops_try;
+
+ e_best = e;
+ }
}
}
const double exec_ms = try_run (device_param, kernel_accel, kernel_loops, 1);
- e_best = exec_ms;
+ u32 kernel_accel_best = kernel_accel;
+ u32 kernel_loops_best = kernel_loops;
+
+ u32 exec_best = exec_ms;
- u32 kernel_accel_try = kernel_accel;
- u32 kernel_loops_try = kernel_loops;
+ // reset
- for (int i = 0; i < 2; i++)
+ if (kernel_accel_min < kernel_accel_max)
{
- kernel_accel_try >>= 1;
- kernel_loops_try <<= 1;
+ u32 kernel_accel_try = kernel_accel;
+ u32 kernel_loops_try = kernel_loops;
- if (kernel_accel_try < kernel_accel_min) break;
- if (kernel_loops_try > kernel_loops_max) break;
+ for (int i = 0; i < 2; i++)
+ {
+ kernel_accel_try >>= 1;
+ kernel_loops_try <<= 1;
+
+ if (kernel_accel_try < kernel_accel_min) break;
+ if (kernel_loops_try > kernel_loops_max) break;
- const double exec_ms = try_run (device_param, kernel_accel_try, kernel_loops_try, 1);
+ const double exec_ms = try_run (device_param, kernel_accel_try, kernel_loops_try, 1);
- if (exec_ms > e_best) break;
+ if (exec_ms < exec_best)
+ {
+ kernel_accel_best = kernel_accel_try;
+ kernel_loops_best = kernel_loops_try;
- kernel_accel = kernel_accel_try;
- kernel_loops = kernel_loops_try;
+ exec_best = exec_ms;
+ }
+ }
- e_best = exec_ms;
+ kernel_accel = kernel_accel_best;
+ kernel_loops = kernel_loops_best;
}
- kernel_accel_try = kernel_accel;
- kernel_loops_try = kernel_loops;
+ // reset
+
- for (int i = 0; i < 2; i++)
+ if (kernel_loops_min < kernel_loops_max)
{
- kernel_accel_try <<= 1;
- kernel_loops_try >>= 1;
+ u32 kernel_accel_try = kernel_accel;
+ u32 kernel_loops_try = kernel_loops;
+
+ for (int i = 0; i < 2; i++)
+ {
+ kernel_accel_try <<= 1;
+ kernel_loops_try >>= 1;
- if (kernel_accel_try > kernel_accel_max) break;
- if (kernel_loops_try < kernel_loops_min) break;
+ if (kernel_accel_try > kernel_accel_max) break;
+ if (kernel_loops_try < kernel_loops_min) break;
- const double exec_ms = try_run (device_param, kernel_accel_try, kernel_loops_try, 1);
+ const double exec_ms = try_run (device_param, kernel_accel_try, kernel_loops_try, 1);
- if (exec_ms > e_best) break;
+ if (exec_ms < exec_best)
+ {
+ kernel_accel_best = kernel_accel_try;
+ kernel_loops_best = kernel_loops_try;
- kernel_accel = kernel_accel_try;
- kernel_loops = kernel_loops_try;
+ exec_best = exec_ms;
+ }
+ }
- e_best = exec_ms;
+ kernel_accel = kernel_accel_best;
+ kernel_loops = kernel_loops_best;
}
// reset timer
memset (device_param->exec_ms, 0, EXEC_CACHE * sizeof (double));
- // reset fake words
-
- for (u32 i = 0; i < kernel_power_max; i++)
- {
- device_param->pws_buf[i].pw_len = 0;
- }
-
- hc_clEnqueueWriteBuffer (data.ocl, device_param->command_queue, device_param->d_pws_buf, CL_TRUE, 0, device_param->size_pws, device_param->pws_buf, 0, NULL, NULL);
- hc_clEnqueueWriteBuffer (data.ocl, device_param->command_queue, device_param->d_pws_amp_buf, CL_TRUE, 0, device_param->size_pws, device_param->pws_buf, 0, NULL, NULL);
-
// store
device_param->kernel_loops = kernel_loops;
if (device_param->skipped) return NULL;
- const bool run_autotune = ((device_param->kernel_accel == 0) && (device_param->kernel_loops == 0));
-
- if (run_autotune)
- {
- autotune (device_param);
- }
+ autotune (device_param);
const uint attack_kern = data.attack_kern;
}
}
- if (run_autotune)
- {
- device_param->kernel_accel = 0;
- device_param->kernel_loops = 0;
- }
+ device_param->kernel_accel = 0;
+ device_param->kernel_loops = 0;
return NULL;
}
if (device_param->skipped) return NULL;
- const bool run_autotune = ((device_param->kernel_accel == 0) && (device_param->kernel_loops == 0));
-
- if (run_autotune)
- {
- autotune (device_param);
- }
+ autotune (device_param);
const uint attack_mode = data.attack_mode;
const uint attack_kern = data.attack_kern;
fclose (fd);
}
- if (run_autotune)
- {
- device_param->kernel_accel = 0;
- device_param->kernel_loops = 0;
- }
+ device_param->kernel_accel = 0;
+ device_param->kernel_loops = 0;
return NULL;
}
return (-1);
}
- if (kernel_accel_chgd == 1 && kernel_loops_chgd == 0)
- {
- log_error ("ERROR: If kernel-accel is specified, kernel-loops need to be specified as well");
-
- return (-1);
- }
-
- if (kernel_loops_chgd == 1 && kernel_accel_chgd == 0)
- {
- log_error ("ERROR: If kernel-loops is specified, kernel-accel need to be specified as well");
-
- return (-1);
- }
-
if (kernel_accel_chgd == 1)
{
if (kernel_accel < 1)
// tuning db
- tuning_db_entry_t *tuningdb_entry = tuning_db_search (tuning_db, device_param->device_name, attack_mode, hash_mode);
+ tuning_db_entry_t *tuningdb_entry = tuning_db_search (tuning_db, device_param, attack_mode, hash_mode);
// device_version
* kernel accel and loops tuning db adjustment
*/
- uint _kernel_accel = kernel_accel;
- uint _kernel_loops = kernel_loops;
+ device_param->kernel_accel_min = 1;
+ device_param->kernel_accel_max = 1024;
- tuning_db_entry_t *tuningdb_entry = tuning_db_search (tuning_db, device_param->device_name, attack_mode, hash_mode);
+ device_param->kernel_loops_min = 1;
+ device_param->kernel_loops_max = 1024;
- if (kernel_accel_chgd == 0)
+ tuning_db_entry_t *tuningdb_entry = tuning_db_search (tuning_db, device_param, attack_mode, hash_mode);
+
+ if (tuningdb_entry)
{
- if (tuningdb_entry)
+ u32 _kernel_accel = tuningdb_entry->kernel_accel;
+ u32 _kernel_loops = tuningdb_entry->kernel_loops;
+
+ if (_kernel_accel)
{
- _kernel_accel = tuningdb_entry->kernel_accel;
+ device_param->kernel_accel_min = _kernel_accel;
+ device_param->kernel_accel_max = _kernel_accel;
}
- }
- if (kernel_loops_chgd == 0)
- {
- if (tuningdb_entry)
+ if (_kernel_loops)
{
- _kernel_loops = tuningdb_entry->kernel_loops;
-
if (workload_profile == 1)
{
_kernel_loops = (_kernel_loops > 8) ? _kernel_loops / 8 : 1;
{
_kernel_loops = (_kernel_loops > 4) ? _kernel_loops / 4 : 1;
}
+
+ device_param->kernel_loops_min = _kernel_loops;
+ device_param->kernel_loops_max = _kernel_loops;
}
}
- device_param->kernel_accel = _kernel_accel;
- device_param->kernel_loops = _kernel_loops;
+ // commandline parameters overwrite tuningdb entries
+
+ if (kernel_accel)
+ {
+ device_param->kernel_accel_min = kernel_accel;
+ device_param->kernel_accel_max = kernel_accel;
+ }
+
+ if (kernel_loops)
+ {
+ device_param->kernel_loops_min = kernel_loops;
+ device_param->kernel_loops_max = kernel_loops;
+ }
+
+ /**
+ * activate device
+ */
devices_active++;
}
#endif // HAVE_ADK
#endif // HAVE_HWMON
- #ifdef OSX
- if (hash_mode == 3000 || hash_mode == 1500 || hash_mode == 10700)
- {
- if (force == 0)
- {
- log_info ("");
- log_info ("Warning: Hash mode %d is not stable with OSX.", hash_mode);
- log_info ("You can use --force to override this but do not post error reports if you do so");
- log_info ("");
-
- continue;
- }
- }
- #endif
-
#ifdef DEBUG
if (benchmark == 1) log_info ("Hashmode: %d", data.hash_mode);
#endif
* some algorithms need a fixed kernel-loops count
*/
- u32 kernel_loops_min = 1;
- u32 kernel_loops_max = 1024;
-
if (hash_mode == 1500)
{
const u32 kernel_loops_fixed = 1024;
- kernel_loops_min = kernel_loops_fixed;
- kernel_loops_max = kernel_loops_fixed;
+ device_param->kernel_loops_min = kernel_loops_fixed;
+ device_param->kernel_loops_max = kernel_loops_fixed;
}
if (hash_mode == 3000)
{
const u32 kernel_loops_fixed = 1024;
- kernel_loops_min = kernel_loops_fixed;
- kernel_loops_max = kernel_loops_fixed;
+ device_param->kernel_loops_min = kernel_loops_fixed;
+ device_param->kernel_loops_max = kernel_loops_fixed;
}
if (hash_mode == 8900)
{
const u32 kernel_loops_fixed = 1;
- kernel_loops_min = kernel_loops_fixed;
- kernel_loops_max = kernel_loops_fixed;
+ device_param->kernel_loops_min = kernel_loops_fixed;
+ device_param->kernel_loops_max = kernel_loops_fixed;
}
if (hash_mode == 9300)
{
const u32 kernel_loops_fixed = 1;
- kernel_loops_min = kernel_loops_fixed;
- kernel_loops_max = kernel_loops_fixed;
+ device_param->kernel_loops_min = kernel_loops_fixed;
+ device_param->kernel_loops_max = kernel_loops_fixed;
}
if (hash_mode == 12500)
{
const u32 kernel_loops_fixed = ROUNDS_RAR3 / 16;
- kernel_loops_min = kernel_loops_fixed;
- kernel_loops_max = kernel_loops_fixed;
+ device_param->kernel_loops_min = kernel_loops_fixed;
+ device_param->kernel_loops_max = kernel_loops_fixed;
}
+ /**
+ * some algorithms have a maximum kernel-loops count
+ */
+
if (attack_exec == ATTACK_EXEC_OUTSIDE_KERNEL)
{
- if (data.salts_buf[0].salt_iter < kernel_loops_max)
+ if (data.salts_buf[0].salt_iter < device_param->kernel_loops_max)
{
- kernel_loops_max = data.salts_buf[0].salt_iter;
+ device_param->kernel_loops_max = data.salts_buf[0].salt_iter;
}
}
- device_param->kernel_loops_min = kernel_loops_min;
- device_param->kernel_loops_max = kernel_loops_max;
-
- // find out if we would request too much memory on memory blocks which are based on kernel_accel
-
- uint size_pws = 4;
- uint size_tmps = 4;
- uint size_hooks = 4;
-
- uint kernel_accel_min = 1;
- uint kernel_accel_max = 1024;
-
/**
* some algorithms need a special kernel-accel
*/
if (hash_mode == 8900)
{
- kernel_accel_max = 64;
+ device_param->kernel_accel_min = 1;
+ device_param->kernel_accel_max = 64;
}
if (hash_mode == 9300)
{
- kernel_accel_max = 64;
+ device_param->kernel_accel_min = 1;
+ device_param->kernel_accel_max = 64;
}
- while (kernel_accel_max)
+ u32 kernel_accel_min = device_param->kernel_accel_min;
+ u32 kernel_accel_max = device_param->kernel_accel_max;
+
+ // find out if we would request too much memory on memory blocks which are based on kernel_accel
+
+ uint size_pws = 4;
+ uint size_tmps = 4;
+ uint size_hooks = 4;
+
+ while (kernel_accel_max >= kernel_accel_min)
{
uint kernel_power_max = device_processors * kernel_threads * kernel_accel_max;
break;
}
+ /*
if (kernel_accel_max == 0)
{
log_error ("Device #%u: Device does not provide enough allocatable device-memory to handle hash-type %u", device_id + 1, data.hash_mode);
return -1;
}
+ */
device_param->kernel_accel_min = kernel_accel_min;
device_param->kernel_accel_max = kernel_accel_max;
+ /*
if (kernel_accel_max < kernel_accel)
{
if (quiet == 0) log_info ("Device #%u: Reduced maximum kernel-accel to %u", device_id + 1, kernel_accel_max);
device_param->kernel_accel = kernel_accel_max;
}
-
- const u32 kernel_accel = device_param->kernel_accel;
+ */
device_param->size_pws = size_pws;
device_param->size_tmps = size_tmps;
// do not confuse kernel_accel_max with kernel_accel here
- const u32 kernel_power = device_processors * kernel_threads * kernel_accel;
+ const u32 kernel_power = device_processors * kernel_threads * kernel_accel_max;
device_param->kernel_threads = kernel_threads;
device_param->kernel_power_user = kernel_power;
projects / hashcat.git / blobdiff