[llvm] [AMDGPU] Add IR LiveReg type-based optimization (PR #66838)

[Jeffrey Byrnes via llvm-commits]

================
@@ -369,9 +450,269 @@ bool AMDGPUCodeGenPrepareImpl::run(Function &F) {
       }
     }
   }
+
+  // GlobalISel should directly use the values, and do not need to emit
+  // CopyTo/CopyFrom Regs across blocks
+  if (UsesGlobalISel)
----------------
jrbyrnes wrote:

I thought this was just a consequence of local ISel -- local ISel needs to use CopyTo/FromRegs to propagate values across blocks, and, in this process, legalizes the values. 

However, it appears that GlobalISel is doing something similar.

https://github.com/llvm/llvm-project/pull/66838