James Price

> I suspect it's something to do with the VLA usage for etot and friends. FWIW there are no VLA's in the OpenCL kernel. The `NUM_TD_PER_THREAD` value is a preprocessor...

Sorry for the slow response. The reason we haven't switched to the final v4.0 release is because some things changed in the domain decomposition that cause problems for our benchmarking....

Thanks for noticing this. I'll need to think about an appropriate fix - the `cl::Platform::get()` function in question has been around for a long time and simply removing it would...

When using the `Program` constructor to auto-build a program, the header will pass the `-cl-std=CL2.0` flag to the OpenCL compiler by default. This may be causing your crash, if you...

> My currently private repo is [here](https://github.com/MathiasMagnus/OpenCL-KernelLoad). We can't see this repo because it's private. Can you either make it public, or perhaps open as a PR here so we...

One high-level question that comes to mind here is about which languages we want to provide examples for. These examples use C++ with the C++ bindings. Do we also want...

No, I didn't consider different primitives and only tested with a single triangle.

Thanks for reporting this. It certainly seems like LLVM is transforming a program which has no data races into one that does, which is unfortunate. Tracking loads to see if...

Some Adreno GPUs do support the `storageBuffer16BitAccess` capability: https://vulkan.gpuinfo.org/listdevicescoverage.php?platform=android&extensionname=VK_KHR_16bit_storage&extensionfeature=storageBuffer16BitAccess It's only the `uniformAndStorageBuffer16BitAccess` capability that excludes all Adreno GPUs. Implementations could consider polyfillying support for 16-bit loads from uniform buffers...

> i'd like to know of this project is still being maintained. Yes, but it's not as regularly updated as I'd like. I try to keep on top of new...