[llvm] [AMDGPU] Support TFE D16 format buffer loads on gfx8/10/11/12 (PR #211465)

[Arseniy Obolenskiy via llvm-commits]

================
@@ -6862,6 +6862,17 @@ bool AMDGPULegalizerInfo::legalizeBufferLoad(MachineInstr &MI,
     return true;
   }
 
+  if (IsFormat && !IsTyped && IsD16 && IsTFE && ST.hasGFX90AInsts()) {
----------------
aobolensk wrote:

TargetFeatures will gate on the whole Intrinsic::ID, but we have some non-TFE loads that are supposed to be allowed

https://github.com/llvm/llvm-project/pull/211465