[llvm] [GlobalISel] Add SVE support for alloca (PR #178976)

[David Green via llvm-commits]

================
@@ -30624,18 +30624,12 @@ bool AArch64TargetLowering::fallBackToDAGISel(const Instruction &Inst) const {
   // Note that if EnableSVEGISel is true, we allow scalable vector types for
   // all instructions, regardless of whether they are actually supported.
   if (!EnableSVEGISel) {
-    if (Inst.getType()->isScalableTy()) {
+    if (Inst.getType()->isScalableTy())
       return true;
-    }
 
     for (unsigned i = 0; i < Inst.getNumOperands(); ++i)
       if (Inst.getOperand(i)->getType()->isScalableTy())
         return true;
-
-    if (const AllocaInst *AI = dyn_cast<AllocaInst>(&Inst)) {
----------------
davemgreen wrote:

It is about what becomes a blocker for enabling GlobalISel by default, and whether the whole of SVE (performance and correctness) is needed, or whether we can enable it for a smaller subset before that. It might become a blocker at any rate, but having the option is still useful.

https://github.com/llvm/llvm-project/pull/178976