[llvm] [TableGen][AMDGPU][AsmParser] Catch and fix ambiguous instructions (PR #211093)

[Matt Arsenault via llvm-commits]

================
@@ -2911,6 +2913,102 @@ def isWave32Strict : Predicate<"Subtarget->isWave32()">,
 def isWave64Strict : Predicate<"Subtarget->isWave64()">,
   AssemblerPredicate <(all_of FeatureWavefrontSize64)>;
 
+def : MutuallyExclusiveAssemblerPredicates<isGFX8GFX9NotGFX90A,
----------------
arsenm wrote:

We probably probably do have some exclusivity problems even when strictly feature based, though it's mostly configurable cases like wavefrontsize. It would probably be better to do incremental predicate / feature cleanups. 

https://github.com/llvm/llvm-project/pull/211093