[llvm] [AArch64] Improve scheduling latency into Bundles (PR #86310)

[David Sherwood via llvm-commits]

================
@@ -235,7 +235,9 @@ class TargetSubtargetInfo : public MCSubtargetInfo {
   // and UseOpIdx are the indices of the operands in Def and Use, respectively.
   // Otherwise, either may be -1.
   virtual void adjustSchedDependency(SUnit *Def, int DefOpIdx, SUnit *Use,
-                                     int UseOpIdx, SDep &Dep) const {}
+                                     int UseOpIdx, SDep &Dep,
+                                     const TargetSchedModel *SchedModel) const {
----------------
david-arm wrote:

I noticed that the AMDGPU overrides this and gets access to the schedmodel via the subtarget, so essentially in the AMDGPU version it's possibly now a bit confusing. Does it need an assert that the SchedModel function argument matches the scheduling model it already uses? Alternatively, do we even need to pass in the SchedModel - can the AArch64Subtarget also just store the scheduling model?

https://github.com/llvm/llvm-project/pull/86310