[WebAssembly] Move CFG-changing passes before RegStackify

[aheejin]

Summary

FixIrreducibleControlFlow and LateEHPrepare both possibly modify CFG and
create new registers. There seems to be no reason these passes go after
register-related optimization passes (PrepareForLiveIntervals,
OptimizeLiveIntervals, StoreResults, RegStackify, and RegColoring), and
this also possibly create new optimization opportunities. I think we
should put all current and future optimization passes before RegStackify
(and related passes) unless there's a reason not to.

[aheejin]

I'm not sure if I can create an LLVM regression test for this. All current test cases pass.

[kripken]

From the point of view of FixIrreducibleControlFlow this sounds good.

llvm/trunk/lib/Target/WebAssembly/WebAssemblyTargetMachine.cpp

void WebAssemblyPassConfig::addPreEmitPass() {
   // colored, and numbered with the rest of the registers.
   addPass(createWebAssemblyReplacePhysRegs());
 
   // Rewrite pseudo call_indirect instructions as real instructions.
   // This needs to run before register stackification, because we change the
   // order of the arguments.
   addPass(createWebAssemblyCallIndirectFixup());
 
+  // Eliminate multiple-entry loops.
+  addPass(createWebAssemblyFixIrreducibleControlFlow());
+
+  // Do various transformations for exception handling.
+  addPass(createWebAssemblyLateEHPrepare());
+
   if (getOptLevel() != CodeGenOpt::None) {
     // LiveIntervals isn't commonly run this late. Re-establish preconditions.
     addPass(createWebAssemblyPrepareForLiveIntervals());
 
     // Depend on LiveIntervals and perform some optimizations on it.
     addPass(createWebAssemblyOptimizeLiveIntervals());
 
     // Prepare store instructions for register stackifying.
     addPass(createWebAssemblyStoreResults());
 
     // Mark registers as representing wasm's value stack. This is a key
     // code-compression technique in WebAssembly. We run this pass (and
     // StoreResults above) very late, so that it sees as much code as possible,
     // including code emitted by PEI and expanded by late tail duplication.
     addPass(createWebAssemblyRegStackify());
 
     // Run the register coloring pass to reduce the total number of registers.
     // This runs after stackification so that it doesn't consider registers
     // that become stackified.
     addPass(createWebAssemblyRegColoring());
   }
 
-  // Eliminate multiple-entry loops. Do this before inserting explicit get_local
-  // and set_local operators because we create a new variable that we want
-  // converted into a local.
-  addPass(createWebAssemblyFixIrreducibleControlFlow());
-
   // Insert explicit get_local and set_local operators.
   addPass(createWebAssemblyExplicitLocals());
 
-  // Do various transformations for exception handling
-  addPass(createWebAssemblyLateEHPrepare());
-
   // Sort the blocks of the CFG into topological order, a prerequisite for
   // BLOCK and LOOP markers.
   addPass(createWebAssemblyCFGSort());
 
   // Insert BLOCK and LOOP markers.
   addPass(createWebAssemblyCFGStackify());
 
   // Lower br_unless into br_if.