[llvm] [AMDGPU] More accurately account for AVGPR pressure (PR #150711)

[Lucas Ramirez via llvm-commits]

================
@@ -238,14 +288,15 @@ class GCNRPTarget {
 private:
   /// Current register pressure.
   GCNRegPressure RP;
-
   /// Target number of SGPRs.
   unsigned MaxSGPRs;
   /// Target number of ArchVGPRs and AGPRs.
   unsigned MaxVGPRs;
   /// Target number of overall VGPRs for subtargets with unified RFs. Always 0
   /// for subtargets with non-unified RFs.
   unsigned MaxUnifiedVGPRs;
+  /// The maximum number of arch vgprs allowed by the subtarget.
+  unsigned AddressableNumArchVGPRs;
----------------
lucas-rami wrote:

Needs to be initialized in the constructors. Let me know if I can help with this, since I designed this class. 

https://github.com/llvm/llvm-project/pull/150711