[llvm] AMDGPU/GFX12: Fix s_barrier_signal_isfirst for single-wave workgroups (PR #143634)

[Nicolai Hähnle via llvm-commits]

nhaehnle wrote:

> The change seems fine; however, the lit test code gen indicates we are not handling the other conditions required to read SCC. Specifically we need to wait on `KMcnt`, and a `S_BARRIER_WAIT`?

Good catch (though it's or, not and; a successful s_barrier_wait implies that the s_barrier_signal_isfirst has also completed). I'm going to raise an issue internally for this.

> Maybe AMDGPUUsage should document that the intrinsic now provides a slightly better guarantee than the underlying instruction?

Done.

https://github.com/llvm/llvm-project/pull/143634