[llvm-branch-commits] [llvm] [AMDGPU] Adds AGPR pressure during candidate init in GCN scheduler. (PR #185733)

[Dhruva Chakrabarti via llvm-branch-commits]

dhruvachak wrote:

On the main branch, there was this method
```
unsigned getMaxNumAGPRs(const Function &F) const { return getMaxNumVGPRs(F); }
```
Additionally, the previous version of this patch added new getMaxNumAGPRs method that started from getTotalNumVGPRs. That could return an incorrect number of AGPRs, larger than what is addressable. All of that logic has been removed in this patch. Now, getMaxNumAGPRs goes through getMaxNumVectorRegs.

https://github.com/llvm/llvm-project/pull/185733