René Widera

CI: refactor and extend atomic tests

5

- [ ] merge after #2582 (first commit contains the fix for atomicDec) clang CUDA is exposing `atomic*_block()` function signatures even if these can not be used by the selected...

test: extend ALPAKA_CHECK to show better error messages

1

To know where a test fails the line and file can be very useful. before: ``` 1: ALPAKA_CHECK failed because '!(equals(operandOrig, ret))' 1: ALPAKA_CHECK failed because '!(equals(operandOrig, ret))' ``` after:...

The unit test for `atomicDec` is not testing the flip around based on the values. https://github.com/alpaka-group/alpaka/blob/0b96515b1c8a24d0c3a2904081bc301311503770/include/alpaka/atomic/Op.hpp#L131 Some for `atomicInc` https://github.com/alpaka-group/alpaka/blob/0b96515b1c8a24d0c3a2904081bc301311503770/include/alpaka/atomic/Op.hpp#L115 We simply test if an value is incremented or decremented:...

compile time issues with nvcc and GCC 10.3

15

The CI is showing compile time issues with nvcc and GCC 10.3 ``` /usr/include/c++/10/chrono: In substitution of 'template template using __is_harmonic = std::__bool_constant [with _Period2 = _Period2; _Rep = _Rep;...

CUDA/HIP device is not freed when alpaka object is destroyed

1

I realized https://github.com/alpaka-group/alpaka/pull/1707#discussion_r867812402 that in cases the alpaka device is destroyed the device is not correctly freed. The class https://github.com/alpaka-group/alpaka/blob/b074b0df68a96321dc73261ab2b9d3d41180f18c/include/alpaka/dev/DevUniformCudaHipRt.hpp#L62 should call `reset()` which is calling `cudaDeviceReset()/hipDeviceReset()` and guarantees that...

CMake option for relocatable device code

12

In #1289 `ALPAKA_CUDA_NVCC_SEPARABLE_COMPILATION` was removed because it was only a renaming of `CUDA_SEPARABLE_COMPILATION`. Today we found out that this feature is supported for clang too, so we can provide this...

remove duplication in compiler flag variables

4

Alpaka appends options to `CUDA_NVCC_FLAGS`, `CMAKE_CXX_FLAGS`, ... In the case where the user is using a third party library which depends e.g. on CUDA and is also appending `CUDA_NVCC_FLAGS` it...

alpaka API call overhead can be high

2

Alpaka is calling with the HIP/CUDA backend for each memcopy, kernel call, ... `hip/cudaSetDavice` and `hip/cudaGetLastError`. Those calls can increase the latency for API calls a lot. In cases where...

high overhead `getDevByIdx`

2

`getDevByIdx(id)` is the equivalent to `cudaSetDevice` but should not be used like that. The overhead of this function in alpaka is very very high. - This method is always asking...

https://github.com/alpaka-group/alpaka/pull/1451 is avoiding that alpaka include files will be showing warning in the user code. We discussed the current "workaround" in the developer meeting and @j-stephan came up with the...