Commit e74c2686 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Disabled std::runtime_error in getRegistersPerMultiprocessor 

All kernels using this will have to be refucktored, because
__launch_bounds__ has different meaning in HIP than in CUDA. See [1] for
details.

[1]: https://rocmdocs.amd.com/en/latest/Programming_Guides/HIP-GUIDE.html#device-side-dynamic-global-memory-allocation
parent d2248a9f

src/TNL/Hip/DeviceInfo.hpp

+2 −1
Original line number Diff line number Diff line
@@ -202,7 +202,8 @@ DeviceInfo:: 
getRegistersPerMultiprocessor( int deviceNum )

{

   // TODO: regsPerMultiprocessor is not part of hipDeviceProp_t yet.

   throw std::runtime_error( "HIP cannot detect number of registers per multiprocessor." );

//   throw std::runtime_error( "HIP cannot detect number of registers per multiprocessor." );

   return 65536;

#ifdef HAVE_HIP

   // results are cached because they are used for configuration of some kernels

   static std::unordered_map< int, int > results;