[llvm] [LAA] Move scalable vector check into `getStrideFromAddRec()` (PR #154013)

[Florian Hahn via llvm-commits]

================
@@ -0,0 +1,32 @@
+; RUN: opt -mtriple=aarch64-none-elf -mattr=+sve2 -O2 -disable-output
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32"
+target triple = "aarch64--linux-gnueabihf"
+
+; This verifies LAA does not attempt to get a fixed element count on a scalable vector.
+; From issue: https://github.com/llvm/llvm-project/issues/153797
+
+define i32 @gradient_fast_par_for_gradient_fast_s0_x_v18_v22(ptr %gradient_fast, i64 %0, ptr %1) {
+entry:
+  br label %"2_for_gradient_fast.s0.x.v20.v23"
+
+"2_for_gradient_fast.s0.x.v20.v23":               ; preds = %"2_for_gradient_fast.s0.x.v20.v23", %entry
----------------
fhahn wrote:

```suggestion
loop:
```

Would be good to clean up the names a bit here, to make it asier to read the IR test

https://github.com/llvm/llvm-project/pull/154013