sbalint98

[hipSYCL] Add hipSYCL GitHub CI

5

# Description This PR will add a github CI to test compilation and the CPU target for the BLAS domain with hipSYCL # Checklist ## New features - [ ]...

[BLAS, cuBLAS backend, hipSYCL] Add official hipsycl cuBLAS support

6

As discussed under PR #144 hipSYCL rocBLAS support comes with automatic support for cuBLAS as well. This PR adds this information to the README. Furthermore, this PR differentiates the FindcuBLAS.cmake...

Atomic performance issue when using the buffer-accessor model

14

**Bug summary** When targetting the AMD backend hipSYCL generates inefficient atomic instructions when the buffer-accessor interface is used. In case of the USM interface, the performance is as expected. **To...

This PR fixes the packaging workflow and updates the outdated documentation. The primary problem for the failing packages was that the link and compilation lines were not updated, and therefore...

Added cmake integration to examples, and tests

2

This PR makes the example and test projects use the hipSYCL CMake integration. It is checked in the default installation directory of the hipSYCL package in case CMake fails to...

Since issue #149 has been fixed, we should upgrade the cuda installation to Cuda 10.1.

deviceProp.name is an empty string for MI50 with rocm 4.5

5

In the case of using rocm 4.5 the `deviceProp.name` string does not contain the expected device name. The following code can reproduce the issue: ``` #include #include int main(){ hipDeviceProp_t...