[llvm] [AMDGPU][True16][CodeGen] Support AND/OR/XOR and LDEXP True16 format (PR #102620)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Mon Aug 12 09:50:01 PDT 2024


================
@@ -236,5 +243,38 @@ bool GCNPreRAOptimizations::runOnMachineFunction(MachineFunction &MF) {
     Changed |= processReg(Reg);
   }
 
+  if (!ST.useRealTrue16Insts())
+    return Changed;
+
+  // Add RA hints to improve True16 COPY elimination.
----------------
arsenm wrote:

If you have to hint every single instance, then it's probably something that should be handled by better coalescing or RA directly 

https://github.com/llvm/llvm-project/pull/102620


More information about the llvm-commits mailing list