[llvm] [AMDGPU] Compiler should synthesize private buffer resource descriptor from flat_scratch_init (PR #79586)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> > We do need a follow up change to prune this from the calling convention lowering. It's going to cause a huge amount of test churn, so it should be done separately
> 
> So, could you please formalize the request a bit and possibly open the Jira ticket? Just to clarify what exactly did you mean to prune: Rsrc23 constants "black magic" or... what?

It's really the same ticket. As it stands, this PR is pointless and just increases the instruction count for no benefit. The point was to prune out the resource descriptor in the kernel user SGPRs, which this isn't doing yet. All of the kernel uses of s[0:3] should be removed and shifted down by 4 SGPRs 

https://github.com/llvm/llvm-project/pull/79586