Jason Burmark
Jason Burmark
Support separable compilation / relocatable device code for hip as well as cuda. Without the -fgpu-rdc flag I get an `unsupported indirect call to function` error at link time.
It looks like when slic is enabled log messages will always be created. https://github.com/LLNL/Umpire/blob/6bf231bdbbc797df70d60027ddb714ac2ef7c0a1/src/umpire/util/Macros.hpp#L55
# Support more hip atomics Refactor hip atomic implementation to support more types natively and de-duplicate implementation. - This PR is a refactoring - It does the following: - Refactors...
Our openmp implementation for non-inplace scans are slow because they do a sequential copy and an inplace parallel scan.
When configuring cmake without a host config the CMAKE_CUDA_FLAGS_* variables are overwritten by the RAJA SetupCompilers.cmake script.
The test-plugin-workgroup-Hip test intermittently fails. The error output indicates that the active platform during capture being none instead of hip. Expected equality of these values: loop_data[i].capture_platform_active Which is: 4-byte object...
Some of our gpu API is not stream safe due to the use of memory pools. This includes reducers, scan operations, and sort operations. The problem arises when memory is...
I just ran into an issue where I accidentally used a library built with RAJA configured without Cuda and linked it to a code built with RAJA configured with Cuda....
# Add Allocator Interface that exposes more options and state to users. Refactor our existing internal allocators to use an Allocator interface. Add ability to set allocators with user defined...
With hip clang we get a warning from two lines (849, 850) of AsyncCommPeer.C here https://github.com/LLNL/SAMRAI/blob/master/source/SAMRAI/tbox/AsyncCommPeer.C#L849. The warning: warning: cast to 'void *' from smaller integer type 'SAMRAI_MPI::Request' (aka 'int')...