James Price

I notice you're passing `NULL` for the work-group size in your `clEnqueueNDRangeKernel` calls, which might lead to pocl using a different work-group size than the NVIDIA driver (which could have...

I don't have the scripts. I probably just ran the benchmarks manually, but it's been so long I can't remember any details on how to do that. The graph is...

I'd also prefer erroring at shader module creation time. For `const_expression` indexes, I don't see a reason why we shouldn't do this for all arrays/matrices/vectors, not just just those using...

Agreed, but this is something that will require external contributors who have experience with these package managers to do (as I personally have neither the experience nor the time right...

Are you using CMake version older than 3.0 by any chance? Looks like `VERSION_LESS` may have not been introduced until 3.0.2, which would explain the weird error message. I guess...

Weird, same formula seems to work fine for me. Those linker errors look bizarre, haven't seen anything like that before...

> However, the latest release of Oclgrind does not work with LLVM 7. Could we have a new release soon, please? Sure. I'll aim to release shortly after the LLVM...

> On the other hand, Oclgrind is too well-behaved, never leaking memory even if leaks were supposed to exist. I'm surprised by this - there isn't anything the Oclgrind code...

Ah, can you try putting `oclgrind` _before_ `valgrind`? With that way round, I get: ``` [james@headnode leak-test]$ oclgrind valgrind --leak-check=full ./a.out ==11577== Memcheck, a memory error detector ==11577== Copyright (C)...

> it seems like the versions are not matching anymore: `Xcode_12.4` is now `Xcode_13.2.1` on the CI. Right, it looks like the images don't provide a version of Xcode that...