[llvm] [AArch64][SME] Make getRegAllocationHints stricter for multi-vector loads (PR #123081)

[Sander de Smalen via llvm-commits]

================
@@ -1109,24 +1110,93 @@ bool AArch64RegisterInfo::getRegAllocationHints(
   // so we add the strided registers as a hint.
   unsigned RegID = MRI.getRegClass(VirtReg)->getID();
   // Look through uses of the register for FORM_TRANSPOSED_REG_TUPLE.
-  if ((RegID == AArch64::ZPR2StridedOrContiguousRegClassID ||
-       RegID == AArch64::ZPR4StridedOrContiguousRegClassID) &&
-      any_of(MRI.use_nodbg_instructions(VirtReg), [](const MachineInstr &Use) {
-        return Use.getOpcode() ==
-                   AArch64::FORM_TRANSPOSED_REG_TUPLE_X2_PSEUDO ||
-               Use.getOpcode() == AArch64::FORM_TRANSPOSED_REG_TUPLE_X4_PSEUDO;
-      })) {
-    const TargetRegisterClass *StridedRC =
-        RegID == AArch64::ZPR2StridedOrContiguousRegClassID
-            ? &AArch64::ZPR2StridedRegClass
-            : &AArch64::ZPR4StridedRegClass;
+  for (const MachineInstr &Use : MRI.use_nodbg_instructions(VirtReg)) {
----------------
sdesmalen-arm wrote:

The code here seems to be looping through uses of VirtReg, but it does that regardless of it's register class or feature flags.

What about returning early from the function if the function has no SME or is not in streaming-mode?
And also moving this loop inside a `if (RegID == .. || RegID == ...)` condition?

https://github.com/llvm/llvm-project/pull/123081