[llvm-branch-commits] [llvm] AMDGPU/GlobalISel: Import D16 load patterns and add combines for them (PR #153178)

[Pierre van Houtryve via llvm-branch-commits]

================
@@ -392,6 +401,102 @@ void AMDGPURegBankCombinerImpl::applyCanonicalizeZextShiftAmt(
   MI.eraseFromParent();
 }
 
+bool AMDGPURegBankCombinerImpl::matchD16Load(MachineInstr &MI,
+                                             D16MatchInfo &MatchInfo) const {
+  if (!STI.d16PreservesUnusedBits())
----------------
Pierre-vh wrote:

We do have a way, it's the field called `Predicates` in the combine rule.

https://github.com/llvm/llvm-project/pull/153178