Dmitry Babokin

> > could you check the change from #3074 on your side? > > This patch isn't in 1.25.1 Could manually apply the changes from #3074 to see if they...

This issue is good one to someone who'd like to start hacking ISPC and is looking for reasonably easy, but important problem.

Bitcast is definitely useful primitive. I would lean towards ``bit_cast()`` syntax (a-la C++20), though it would be a precedent for using ```` in ISPC. I'm not sure that there's much...

Thanks for the report, it appeared to be an interesting problem. The intention was not to have an alignment, but LLVM toolchain has introduced an alignment that it thinks is...

I'll have a look at it.

At this point I see that this is a bug triggered by LLVM optimizations. The most likely it's either the bug in LLVM optimizations or in LLVM IR that we...

The problem remains in the ISPC based on LLVM 6.0. Also, the switch --opt=disable-all-on-optimizations looks like is a workaround for this problem.

That's very reasonable request, we should definitely do that. Though, implementation is not that straightforward, as it seems from the first glance. For scalar code, the most performance friendly control...

@DeepakRajendrakumaran could you have a look? Seems easy to fix and would be good to have it fixed before the release.

The problem is that `lEmitVectorTypedefs()` is not designed to emit varying short vector definitions: https://github.com/ispc/ispc/blob/f390931d5dc5f2c62efaa42441c117430d414bc3/src/module.cpp#L1546 Seems that proper fix would need to: - fix `lEmitVectorTypedefs()` to emit proper definition of...