[llvm] [X86] Add missing vNbf16 handling in X86CallingConv.td file (PR #127102)

[Phoebe Wang via llvm-commits]

phoebewang wrote:

> This is puzzling to me. Obviously, if FastISel fails then compiler fallbacks to something else to lower these calls -- but since the X86callingConv table _doesn't_ have entries for vNbf16 how does this other thing manage not to crash? It has to use some other mechanism, one which doesn't use the table.

The DAGISel uses in this way: https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/X86/X86ISelLoweringCall.cpp#L126-L129
I think it can be removed with this change.

> Nonetheless, this PR fixes the crash, though I didn't create a test for it, since I am unsure yet how it should look like. I would like to learn how the working non-FastISel mechanism works; I tried looking for it, but didn't yet manage to find anything

We don't have enough coverage for FastISel, but I think it's ok as long as we didn't touch FastISel code, because it will fallback to DAGISel anyway. The ABI handling may have difference, and that's why it failed. So I think we may create a single test file for bf16 ABI and test both FastISel and DAGISel, or maybe even GISel. It may look like:

```
RUN: llc < %s -mtriple=x86_64 | FileCheck %s --check-prefixes=CHECK,SSE2,SSE2-DISEL
RUN: llc < %s -mtriple=x86_64 -fast-isel | FileCheck %s --check-prefixes=CHECK,SSE2,SSE2-FISEL
RUN: llc < %s -mtriple=x86_64 -mattr=avx512bf16,avx512vl | FileCheck %s --check-prefixes=CHECK,AVX,AVX-DISEL
RUN: llc < %s -mtriple=x86_64 -fast-isel -mattr=avx512bf16,avx512vl | FileCheck %s --check-prefixes=CHECK,AVX,AVX-FISEL

define void @caller128(ptr %x) {
  %y = load <16 x bfloat>, ptr %x
  call void @callee128(<16 x bfloat> %y)
  ret void
}

declare void @callee128(<16 x bfloat>)
...
```

The issue can be reproduced with `-fast-isel -mattr=avx512bf16,avx512vl`

https://github.com/llvm/llvm-project/pull/127102