[llvm] [AArch64] Avoid streaming mode hazards from FP constant stores (PR #114207)

[Benjamin Maxwell via llvm-commits]

================
@@ -489,6 +489,10 @@ class TargetLoweringBase {
     return true;
   }
 
+  /// Returns true if a floating-point load or store can be replaced with an
+  /// equivalent integer load or store without negatively affecting performance.
+  virtual bool canUseIntLoadStoreForFloatValues() const { return true; }
----------------
MacDue wrote:

I'm not sure it can be, the operations are legal (and the codegen is no different from regular AArch64). The issue is the possible performance impact from accessing the same memory region with FP/vector loads/stores and GPR loads/stores in a streaming SVE function.

https://github.com/llvm/llvm-project/pull/114207