[llvm] [AMDGPU] Use fake16 load/store with +real-true16 and sram-ecc (PR #161256)

[Joe Nash via llvm-commits]

================
@@ -2599,6 +2599,9 @@ def UseFakeTrue16Insts : True16PredicateClass<"Subtarget->hasTrue16BitInsts() &&
   // FIXME When we default to RealTrue16 instead of Fake, change the line as follows.
   // AssemblerPredicate<(all_of FeatureTrue16BitInsts, (not FeatureRealTrue16Insts))>;
 
+def UseLoadTrue16WithSramECC : True16PredicateClass<"Subtarget->useRealTrue16Insts() && "
----------------
Sisyph wrote:

We could use some more description here and in the commit message on what behavior is achieved. Also, it is confusing that the commit is titled 'Use fake16 load/store...' But the predicate is called 'UseLoadTrue16...' I might just drop Load here, and call it 'UseTrue16WithSramECC'. In a comment you can say the behavior, that it serves to not use d16 loads.

https://github.com/llvm/llvm-project/pull/161256