Nicholas Wilson

> There's a reserved PS4 version. OK, we should probably still set `version(FreeBSD)` though, since it is a FreeBSD. > The target CPU should probably default to btver2 How do...

Of course, but this is still very much WIP. I'll be waiting to get the green light from Manu and Pursche and then polish this up.

Looking into this

Very strange, works for me ``` $ninja -j5 [0/1] Re-running CMake... -- Found host D compiler /Users/nicholaswilson/d/ldc2-1.26.0-osx-arm64/bin/ldmd2, with default flags '' -- Host D compiler ID: LDMD -- Host D...

I'm fine if you want to mark that test as `XFAIL` for now. > there's ugly stuff based on hardcoded LLVM type names / trying to derive special types from...

Ahh, I guess not.

> I know! That why I'd love to move to the proper backend in LLVM 15, but that needs new pass infrastructure and support for opaque pointers. Scratch that, looks...

Nice. Is there a way we can separate the intrinsics out into related clusters, like dcompute has now? e.g. send `"llvm.nvvm.read.*"` to `index.d` `"llvm.nvvm.[barrier|membar]*"` to `sync.d`. I think it would...

The generated file will have to be marked `@compute(CompileFor.deviceOnly)`, with the corresponding import of `ldc.dcompute`.

That works too.