[llvm] [AMDGPU] Register allocation anti-hints to reduce MFMA hazard NOPs (PR #156943)

[Matt Arsenault via llvm-commits]

================
@@ -499,6 +499,45 @@ bool TargetRegisterInfo::getRegAllocationHints(
   return false;
 }
 
+void TargetRegisterInfo::applyRegAllocationAntiHints(
+    Register VirtReg, ArrayRef<MCPhysReg> &Order,
+    SmallVectorImpl<MCPhysReg> &OrderStorage,
+    SmallVector<MCPhysReg, 16> &AntiHints, const MachineFunction &MF,
+    const VirtRegMap *VRM, const LiveRegMatrix *Matrix) const {
+
+  if (AntiHints.empty() || !VRM)
+    return;
+
+  auto isAntiHinted = [&](MCPhysReg Reg) {
+    return std::any_of(
+        AntiHints.begin(), AntiHints.end(),
----------------
arsenm wrote:

```suggestion
    return llvm::any_of(
        AntiHints,
```

https://github.com/llvm/llvm-project/pull/156943