[ModuleInliner] Try moving function-simplification before module inliner

[mingmingl]

Turns out the improvement I saw is an artifact of testing flow (in a/b or b/a, test is always worse than base for a reason when I enable new changes in baseline). Thanks for the offline discussions!

Plan to move function-simplication before module inliner and see how it works out.

llvm/lib/Passes/PassBuilderPipelines.cpp

PassBuilder::buildModuleInlinerPipeline(OptimizationLevel Level,
   // The inline deferral logic is used to avoid losing some
   // inlining chance in future. It is helpful in SCC inliner, in which
   // inlining is processed in bottom-up order.
   // While in module inliner, the inlining order is a priority-based order
   // by default. The inline deferral is unnecessary there. So we disable the
   // inline deferral logic in module inliner.
   IP.EnableDeferral = false;
 
-  MPM.addPass(ModuleInlinerPass(IP, UseInlineAdvisor, Phase));
+  MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(
+      PostOrderFunctionAttrsPass(/*SkipNonRecursive*/ true)));
+
+  if (Level == OptimizationLevel::O3)
+    MPM.addPass(
+        createModuleToPostOrderCGSCCPassAdaptor(ArgumentPromotionPass()));
 
   MPM.addPass(createModuleToFunctionPassAdaptor(
       buildFunctionSimplificationPipeline(Level, Phase),
       PTO.EagerlyInvalidateAnalyses));
 
+  MPM.addPass(ModuleInlinerPass(IP, UseInlineAdvisor, Phase));
+
+  MPM.addPass(
+      createModuleToPostOrderCGSCCPassAdaptor(PostOrderFunctionAttrsPass()));
+
   MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(
       CoroSplitPass(Level != OptimizationLevel::O0)));
 
   return MPM;
 }
 
 ModulePassManager
 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level,