Fall back to 64 threads default (from 256) because modern GPU create too many workitems which then results in too less GPU memory available

include/types.h

@@ -147,7 +147,7 @@ typedef enum amplifier_count
   KERNEL_BFS              = 1024,
   KERNEL_COMBS            = 1024,
   KERNEL_RULES            = 1024,
-  KERNEL_THREADS_MAX      = 256,
+  KERNEL_THREADS_MAX      = 64,
   KERNEL_THREADS_MAX_CPU  = 1
 
 } amplifier_count_t;