[llvm] [AMDGPU][GFX12.5] Add support for emitting memory operations with nv bit set (PR #179413)

[Matt Arsenault via llvm-commits]

================
@@ -899,6 +905,23 @@ SIMemOpAccess::getLDSDMAInfo(const MachineBasicBlock::iterator &MI) const {
   return constructFromMIWithMMO(MI);
 }
 
+/// \returns true if \p MI has one or more MMO, and all of them are fit for
+/// being marked as non-volatile. This means that either they are accessing the
+/// constant address space, are accessing a known invariant memory location, or
+/// that they are marked with the non-volatile metadata/MMO flag.
+static bool isNonVolatileMemoryAccess(const MachineInstr &MI) {
+  static constexpr unsigned NVFlags =
+      (MOThreadPrivate | MachineMemOperand::MOInvariant);
+  if (MI.getNumMemOperands() == 0)
+    return false;
+  return all_of(MI.memoperands(), [&](const MachineMemOperand *MMO) {
+    unsigned AS = MMO->getAddrSpace();
+    return AS == AMDGPUAS::CONSTANT_ADDRESS ||
+           AS == AMDGPUAS::CONSTANT_ADDRESS_32BIT ||
----------------
arsenm wrote:

Don't spread the special case of these address spaces around, these MMOs should get tagged as invariant anyway 

https://github.com/llvm/llvm-project/pull/179413