Shuffle around some of the optimization passes.

The end goal here is to end up with a good pass ordering that will allow
us to only run one set of these passes, rather than running them
twice. This is a start in that direction.

No real impact measured on compile times as of this change. On
benchmarks I see a mix of regressions and improvements.

-O improvements:
  Calculator           -17.6%     1.21x
  Chars                -54.4%     2.19x
  PolymorphicCalls     -14.7%     1.17x
  SetIsSubsetOf        -14.1%     1.16x
  Sim2DArray           -14.1%     1.16x
  StrToInt             -30.4%     1.44x

-O regressions:
  CaptureProp          +32.9%     0.75x
  DictionarySwap       +36.0%     0.74x
  XorLoop              +39.8%     0.72x

-Ounchecked improvements:
  Chars                -58.0%     2.38x

-Ounchecked regressions:
  CaptureProp          +33.3%     0.75x

-Onone improvements:
  StrToInt             -14.9%     1.18x
  StringWalk           -47.6%     1.91x
  StringWithCString    -17.2%     1.21x
  (many more smaller improvements)

-Onone regressions:
  Calculator           +21.5%     0.82x
  OpenClose            +10.1%     0.91x

Author: rudkx

lib/SILOptimizer/PassManager/Passes.cpp

@@ -172,36 +172,25 @@ void AddHighLevelLoopOptPasses(SILPassManager &PM) {
   PM.addSwiftArrayOpts();
 }
 
+// Perform classic SSA optimizations.
 void AddSSAPasses(SILPassManager &PM, OptimizationLevelKind OpLevel) {
-  AddSimplifyCFGSILCombine(PM);
+  // Promote box allocations to stack allocations.
   PM.addAllocBoxToStack();
+
+  // Propagate copies through stack locations.  Should run after
+  // box-to-stack promotion since it is limited to propagating through
+  // stack locations. Should run before aggregate lowering since that
+  // splits up copy_addr.
   PM.addCopyForwarding();
-  PM.addLowerAggregateInstrs();
-  PM.addSILCombine();
-  PM.addSROA();
-  PM.addMem2Reg();
 
-  // Perform classic SSA optimizations.
-  PM.addGlobalOpt();
-  PM.addLetPropertiesOpt();
-  PM.addPerformanceConstantPropagation();
-  PM.addDCE();
-  PM.addCSE();
-  PM.addSILCombine();
-  PM.addJumpThreadSimplifyCFG();
-  // Jump threading can expose opportunity for silcombine (enum -> is_enum_tag->
-  // cond_br).
-  PM.addSILCombine();
-  // Which can expose opportunity for simplifcfg.
-  PM.addSimplifyCFG();
+  // Split up opaque operations (copy_addr, retain_value, etc.).
+  PM.addLowerAggregateInstrs();
 
-  // Perform retain/release code motion and run the first ARC optimizer.
-  PM.addRedundantLoadElimination();
-  PM.addCSE();
-  PM.addEarlyCodeMotion();
-  PM.addARCSequenceOpts();
+  // Split up operations on stack-allocated aggregates (struct, tuple).
+  PM.addSROA();
 
-  PM.addSILLinker();
+  // Promote stack allocations to values.
+  PM.addMem2Reg();
 
   // Run the devirtualizer, specializer, and inliner. If any of these
   // makes a change we'll end up restarting the function passes on the
@@ -217,12 +206,37 @@ void AddSSAPasses(SILPassManager &PM, OptimizationLevelKind OpLevel) {
       // Does inline semantics-functions (except "availability"), but not
       // global-init functions.
       PM.addPerfInliner();
+      PM.addGlobalOpt();
+      PM.addLetPropertiesOpt();
       break;
     case OptimizationLevelKind::LowLevel:
       // Inlines everything
       PM.addLateInliner();
       break;
   }
+
+  PM.addMem2Reg();
+  AddSimplifyCFGSILCombine(PM);
+
+  PM.addPerformanceConstantPropagation();
+  PM.addDCE();
+  PM.addCSE();
+  PM.addSILCombine();
+  PM.addJumpThreadSimplifyCFG();
+  // Jump threading can expose opportunity for silcombine (enum -> is_enum_tag->
+  // cond_br).
+  PM.addSILCombine();
+  // Which can expose opportunity for simplifcfg.
+  PM.addSimplifyCFG();
+
+  // Perform retain/release code motion and run the first ARC optimizer.
+  PM.addRedundantLoadElimination();
+  PM.addCSE();
+  PM.addEarlyCodeMotion();
+  PM.addARCSequenceOpts();
+
+  PM.addSILLinker();
+
   PM.addSimplifyCFG();
   // Only hoist releases very late.
   if (OpLevel == OptimizationLevelKind::LowLevel)

test/SILOptimizer/devirt_single_module_in_multiple_files.swift

@@ -6,12 +6,12 @@ public func test() {
   e.evaluate(1)
 }
 
-// CHECK-LABEL: sil shared [transparent] [reabstraction_thunk] @_TTSf3cpfr69_TFFC38devirt_single_module_in_multiple_files9EvaluatorcFT_S0_U_FT_Si_n_n___TTRXFo__dSi_XFo_iT__iSi_
+// CHECK-LABEL: sil shared @_TFFC38devirt_single_module_in_multiple_files9Evaluatorc
 // CHECK: %{{.*}} = class_method %{{.*}} : $Problem1, #Problem1.run!1 : (Problem1) -> () -> Int , $@convention(method) (@guaranteed Problem1) -> Int
 // CHECK-NEXT: apply
 // CHECK: return
 
-// CHECK-LABEL: sil shared [transparent] [reabstraction_thunk] @_TTSf3cpfr70_TFFC38devirt_single_module_in_multiple_files9EvaluatorcFT_S0_U0_FT_Si_n_n___TTRXFo__dSi_XFo_iT__iSi_
+// CHECK-LABEL: sil shared @_TFFC38devirt_single_module_in_multiple_files9Evaluatorc
 // CHECK: %{{.*}} = class_method %{{.*}} : $Problem2, #Problem2.run!1 : (Problem2) -> () -> Int , $@convention(method) (@guaranteed Problem2) -> Int
 // CHECK-NEXT: apply
 // CHECK: return