Move the S0 indirect send generation before post-RA

To improve the performance with scheduling

Author: bcheng0127

visa/Optimizer.cpp

@@ -992,6 +992,7 @@ int Optimizer::optimization() {
   runPass(PI_reassignBlockIDs);
 
   runPass(PI_FoldAddrImmediate);
+  runPass(PI_s0SubAfterRA);
 
   // FIXME houjenko: Disable local scheduling due to issues when
   // using extra regiser that may corrupt unknown liveout
@@ -1010,7 +1011,6 @@ int Optimizer::optimization() {
   runPass(PI_legalizeType);
 
   runPass(PI_changeMoveType);
-  runPass(PI_s0SubAfterRA);
 
   // No pass after this should expect def-use to be preserved as this pass
   // removes raw movs with identical src/dst physical GRFs.