[llvm] [AArch64] Use correct regclass for spills of ZPR2/ZPR4 (PR #148806)

[Sander de Smalen via llvm-commits]

================
@@ -2625,16 +2625,22 @@ let Predicates = [HasSVE_or_SME] in {
   // These get expanded to individual LDR_ZXI/STR_ZXI instructions in
   // AArch64ExpandPseudoInsts.
   let mayLoad = 1, hasSideEffects = 0 in {
-    def LDR_ZZXI   : Pseudo<(outs   ZZ_b_strided_and_contiguous:$Zd), (ins GPR64sp:$sp, simm4s1:$offset),[]>, Sched<[]>;
+    def LDR_ZZXI_STRIDED_CONTIGUOUS   : Pseudo<(outs   ZZ_b_strided_and_contiguous:$Zd), (ins GPR64sp:$sp, simm4s1:$offset),[]>, Sched<[]>;
----------------
sdesmalen-arm wrote:

Yeah, we've not been very consistent with the naming, I'll fix that up in a follow-up NFC patch. My vote is also for "strided or contiguous" in that case.

https://github.com/llvm/llvm-project/pull/148806