This fixes a problem in the upstream code.  There is an extra argument in
the call to the x_buffer.CopyToAsync function that must be removed. Otherwise
the build fails when CUDA is enabled.
--- CLBlast/src/routines/level3/xtrsm.cpp	2022-05-21 08:44:13.815816361 -0600
+++ CLBlast_fixed/src/routines/level3/xtrsm.cpp	2022-05-21 08:28:06.756355738 -0600
@@ -246,7 +246,7 @@
   }
 
   // Retrieves the results
-  x_buffer.CopyToAsync(queue_, b_size, b_buffer, event_);
+  x_buffer.CopyToAsync(queue_, b_size, b_buffer);
 }
 
 // =================================================================================================