[llvm] [AArch64] Use SVE for load-extend-broadcast (PR #221689)

[Jack Styles via llvm-commits]

================
@@ -1022,6 +1022,10 @@ def FeatureLimited64bitVectorMulBandwidth : SubtargetFeature<
     "Has limited 64bit vector multiply bandwidth compared to scalar multiply",
     [], InlineIgnore>;
 
+def FeaturePreferSVEVectors : SubtargetFeature <
----------------
Stylie777 wrote:

I'd be hesitant to do so. This was introduced to stop cores using SVE LDR1 for FP instructions, whereas we are trying to do the opposite here which is to use SVE. I've introduced a new flag as I've only found this to be beneficial on V3/V3AE cores.

https://github.com/llvm/llvm-project/pull/221689