[WIP] Add support for continuous integration on MacOS

[proski]

This is a temporary "demo" PR to test CI improvements for #294. It also shows how prpll fails on MacOS.

[tdulcet]

To use cmake on Windows, you probably would need to install the mingw-w64-x86_64-cmake package for it to be able to find the OpenCL library, as every MSYS2 environment uses different directories. However, we should probably test building PRPLL with both make and cmake to prevent breakages in the Makefile.

In #275, I updated the workflow to install the PoCL OpenCL implementation on Linux. However, this currently only supports OpenCL 1.2, so it is not able to perform any actual primality tests, as PRPLL requires OpenCL 2.0. This can be seen from the error it outputs:

20240913 07:18:39 82589933 Build option -cl-std specified OpenCL C version 2.0,but device cpu doesn't support that OpenCL C version.

20240913 07:18:39 82589933 Compiling 'transpose.cl' error COMPILE_PROGRAM_FAILURE (-15) (args -cl-finite-math-only -cl-std=CL2.0  -DEXP=82589933u -DWIDTH=1024u -DSMALL_HEIGHT=256u -DMIDDLE=9u -DCARRY_LEN=8u -DNW=4u -DNH=4u -DWEIGHT_STEP=0.41118935041770088 -DIWEIGHT_STEP=-0.29137787235709517 -DFFT_VARIANT=0u -DTAILT=U2(-7.5298160855459062e-05,-0.012271538285719925) )

Also see #279 (comment). You would probably need to use a different CPU based OpenCL implementation for now if you want to add a "real test" on Linux or Windows.

[proski]

To use cmake on Windows, you probably would need to install the mingw-w64-x86_64-cmake package for it to be able to find the OpenCL library, as every MSYS2 environment uses different directories. However, we should probably test building PRPLL with both make and cmake to prevent breakages in the Makefile.

I just wanted to see if it would work.

In #275, I updated the workflow to install the PoCL OpenCL implementation on Linux. However, this currently only supports OpenCL 1.2, so it is not able to perform any actual primality tests, as PRPLL requires OpenCL 2.0.

I didn't realize that! My Mac (with the latest MacOS Sonoma) reports and OpenCL 1.1 device with OpenCL 1.2 framework, and the MacOS used by the CI reports the same.

I'm not going to spend time fixing MacOS errors in gpuowl for my setup. Perhaps a separate program for Metal or Vulkan is needed.

Also see #279 (comment). You would probably need to use a different CPU based OpenCL implementation for now if you want to add a "real test" on Linux or Windows.

I agree. Video hardware on CI machines can change at any time.

[proski]

Closing this draft PR, it's not needed anymore.