[llvm] [AMDGPU][MC] Disallow op_sel in v_dot4 and v_dot8 with 4- or 8-bit packed data (PR #100485)

[Jun Wang via llvm-commits]

jwanggit86 wrote:

> > > I believe this is only true for gfx940, and I think the older targets did support op_sel with these. I see a note that not supporting op_sel is a change, but don't see documentation for what it did before
> > 
> > 
> > @arsenm Is one of the following the note you referred to above?
> 
> No, I'm reading the ISA documentation, not the implementation
> 
> > Even these two places are inconsistent. One says "do not use on gfx940+", another seems to apply only to gfx940 and gfx11.
> 
> This looks like it wasn't cleanly implemented to track which targets support it and which don't. I wouldn't trust whatever is implemented here, this requires double checking the hardware docs
> 
> > Second, are you suggesting op_sel should be disallowed for gfx940+, but left alone for older targets?
> 
> It should be left alone on any target where it's supported. We shouldn't just drop it everywhere

@arsenm I checked the documentation. For gfx940 it is explicitly stated that op_sel is not allowed. However, it is vague (i.e., not explicitly said) for other targets. So, should we just disallow it for gfx940 for now?

https://github.com/llvm/llvm-project/pull/100485