Optimize a few modes for hashcat_tuning.hctab for budget NV cards

[hashcat.git] / src / oclHashcat.c

diff --git a/src/oclHashcat.c b/src/oclHashcat.c
index d24c6cf..ea3c9c3 100644 (file)
--- a/src/oclHashcat.c
+++ b/src/oclHashcat.c
@@ -2859,7 +2859,19 @@ static void autotune (hc_device_param_t *device_param)
 
   #define MAX_RETRIES 1
 
-  double exec_ms_final = 0;
+  if ((kernel_loops_min == kernel_loops_max) || (kernel_accel_min == kernel_accel_max))
+  {
+    // we do this in case the user specified a fixed -u and -n on the commandline
+    // so we have a cached kernel for benchmark
+
+    try_run (device_param, kernel_accel, kernel_loops);
+    try_run (device_param, kernel_accel, kernel_loops);
+    try_run (device_param, kernel_accel, kernel_loops);
+    try_run (device_param, kernel_accel, kernel_loops);
+    try_run (device_param, kernel_accel, kernel_loops);
+  }
+
+  double exec_ms_final = try_run (device_param, kernel_accel, kernel_loops);
 
   // first find out highest kernel-loops that stays below target_ms
 
@@ -2909,10 +2921,10 @@ static void autotune (hc_device_param_t *device_param)
 
   // sometimes we're in a bad situation that the algorithm is so slow that we can not
   // create enough kernel_accel to do both, keep the gpu busy and stay below target_ms.
-  // however, we need to have a minimum kernel_accel of 64.
+  // however, we need to have a minimum kernel_accel and kernel_loops of 32.
   // luckily, at this level of workload, it became a linear function
 
-  while (kernel_accel < 64)
+  while (kernel_accel < 32 && kernel_loops >= 32)
   {
     const u32 kernel_accel_try = kernel_accel * 2;
     const u32 kernel_loops_try = kernel_loops / 2;
@@ -13643,6 +13655,7 @@ int main (int argc, char **argv)
        * some algorithms have a maximum kernel-loops count
        */
 
+      /*
       if (attack_exec == ATTACK_EXEC_OUTSIDE_KERNEL)
       {
         if (data.salts_buf[0].salt_iter < device_param->kernel_loops_max)
@@ -13650,6 +13663,7 @@ int main (int argc, char **argv)
           device_param->kernel_loops_max = data.salts_buf[0].salt_iter;
         }
       }
+      */
 
       /**
        * some algorithms need a special kernel-accel