[clang] [llvm] [Clang][inlineasm] Add special support for "rm" output constraints (PR #92040)

[Bill Wendling via cfe-commits]

================
@@ -985,6 +985,9 @@ void TargetPassConfig::addISelPrepare() {
   if (getOptLevel() != CodeGenOptLevel::None)
     addPass(createObjCARCContractPass());
 
+  if (getOptLevel() == CodeGenOptLevel::None)
----------------
bwendling wrote:

This is how the fast register allocator is selected (see around line 1173). I'm *not* interested in the value of `OptimizeRegAlloc` (the `-regalloc` flag) because that's hidden for a reason. :-) But remember that this isn't an "optimization" in the normal sense of the word---that the middle-end does its thing. As far as I know, the `OptNone` attribute doesn't change the activities of the back-end at that level (i.e., changing the register allocator on the fly).

https://github.com/llvm/llvm-project/pull/92040