FunMiles

@lhog , you want to create a PR with your fixes for this?

If I may, a little detour: reading the linked code reminded me of another issue I've left out. Clang complains about a few switch/case where not all cases are handled....

I am my own employer ;) I am going to have a bit more free time starting later this week. I'll create a few PRs. I already have two. I...

I am short of time to enter into development to implement this. If I end up with more free time, I may try to tackle it for UNIX type sockets...

The problem remains under 14.0.3. Changing the one line `auto n = next;` to `thread_local auto n = next;` appears to be a workaround. But without somebody in the know...

This problem exist in the original LLVM and is fixed by https://github.com/llvm/llvm-project/commit/c4672454743e942f148a1aff1e809dae73e464f6 I will close this issue when the patch is brought into this fork.

AMD radeon pro 5500m. Yes it does support 32-invocation subgroups as reported: ``` SubgroupSizeControlProperties: maxComputeWorkgroupSubgroups = 32 maxSubgroupSize = 64 minSubgroupSize = 32 ``` However don't forget the GLSL code...

PS: I also made sure that there's no change in the subgroup size somewhere else. The query of the device does also report the current setting as being 64: ```...

The answer to your first question is that my customer has been working with their workflow for 30 years and making them change this to something else is a very...