Improve pass pipeline (#1227)

Author: MaxGraey

Diff for: cli/asc.js

@@ -694,48 +694,47 @@ exports.main = function main(argv, options, callback) {
 
       // PassRunner::addDefaultGlobalOptimizationPrePasses
       add("duplicate-function-elimination");
+      add("remove-unused-module-elements"); // differs
 
       // PassRunner::addDefaultFunctionOptimizationPasses
       if (optimizeLevel >= 3 || shrinkLevel >= 1) {
         add("ssa-nomerge");
       }
       if (optimizeLevel >= 3) {
-        add("simplify-locals-nostructure"); // differs
+        add("flatten"); // differs
+        add("simplify-locals-notee-nostructure"); // differs
         add("vacuum"); // differs
-        add("reorder-locals"); // differs
+        add("code-folding"); // differs
         add("flatten");
         add("local-cse");
+        add("reorder-locals"); // differs
       }
       if (optimizeLevel >= 2 || shrinkLevel >= 1) { // differs
         add("rse");
+        add("vacuum");
       }
       if (hasARC) { // differs
         if (optimizeLevel < 3) {
           add("flatten");
         }
         add("post-assemblyscript");
       }
+      add("optimize-instructions"); // differs
+      add("inlining"); // differs
       add("dce");
       add("remove-unused-brs");
       add("remove-unused-names");
-      // add("optimize-instructions"); // differs move 2 lines above
+      add("inlining-optimizing"); // differs
       if (optimizeLevel >= 2 || shrinkLevel >= 1) {
         add("pick-load-signs");
         add("simplify-globals-optimizing"); // differs
       }
-      add("optimize-instructions"); // differs
       if (optimizeLevel >= 3 || shrinkLevel >= 2) {
         add("precompute-propagate");
       } else {
         add("precompute");
       }
-      if (module.getLowMemoryUnused()) {
-        if (optimizeLevel >= 3 || shrinkLevel >= 1) {
-          add("optimize-added-constants-propagate");
-        } else {
-          add("optimize-added-constants");
-        }
-      }
+      add("vacuum"); // differs
       // this will be done later (1)
       // if (optimizeLevel >= 2 || shrinkLevel >= 2) {
       //   add("code-pushing");
@@ -776,25 +775,36 @@ exports.main = function main(argv, options, callback) {
       // } else {
       //   add("precompute");
       // }
-      add("optimize-instructions");
+      if (optimizeLevel >= 3) {
+        add("optimize-instructions");
+      }
       if (optimizeLevel >= 2 || shrinkLevel >= 1) {
         add("rse");
       }
       add("vacuum");
       // PassRunner::addDefaultGlobalOptimizationPostPasses
       if (optimizeLevel >= 2 || shrinkLevel >= 1) {
+        add("simplify-globals-optimizing"); // differs
         add("dae-optimizing");
       }
       if (optimizeLevel >= 2 || shrinkLevel >= 2) {
         add("inlining-optimizing");
       }
+      if (module.getLowMemoryUnused()) {
+        if (optimizeLevel >= 3 || shrinkLevel >= 1) {
+          add("optimize-added-constants-propagate");
+        } else {
+          add("optimize-added-constants");
+        }
+      }
       // "duplicate-function-elimination" will better done later
       // add("duplicate-function-elimination");
       add("duplicate-import-elimination");
       if (optimizeLevel >= 2 || shrinkLevel >= 2) {
         add("simplify-globals-optimizing");
       } else {
         add("simplify-globals");
+        add("vacuum"); // differs
       }
       // moved from (2)
       // it works better after globals optimizations like simplify-globals, inlining-optimizing and etc
@@ -814,28 +824,30 @@ exports.main = function main(argv, options, callback) {
       }
       if (optimizeLevel >= 2 || shrinkLevel >= 1) { // differs
         add("rse");
-        // rearrange / reduce switch cases again
-        add("remove-unused-brs");
-        add("vacuum");
-
         // move some code after early return which potentially could reduce computations
         // do this after CFG cleanup (originally it was done before)
         // moved from (1)
         add("code-pushing");
         if (optimizeLevel >= 3) {
           // this quite expensive so do this only for highest opt level
           add("simplify-globals");
+          add("vacuum");
           // replace indirect calls with direct and inline if possible again.
+          add("inlining-optimizing");
           add("directize");
           add("dae-optimizing");
           add("precompute-propagate");
-          add("coalesce-locals");
+          add("vacuum");
           add("merge-locals");
+          add("coalesce-locals");
           add("simplify-locals-nostructure");
           add("vacuum");
           add("inlining-optimizing");
           add("precompute-propagate");
         }
+        add("remove-unused-brs");
+        add("remove-unused-names");
+        add("vacuum");
         add("optimize-instructions");
         add("simplify-globals-optimizing");
       }

Diff for: tests/compiler/assert-nonnull.optimized.wat

@@ -213,8 +213,8 @@
   (local $1 i32)
   local.get $0
   call_indirect (type $none_=>_i32)
-  local.tee $0
-  local.get $0
+  local.tee $1
+  local.get $1
   i32.eqz
   if
    i32.const 0
@@ -235,8 +235,8 @@
   local.get $0
   i32.load offset=4
   call_indirect (type $none_=>_i32)
-  local.tee $0
-  local.get $0
+  local.tee $1
+  local.get $1
   i32.eqz
   if
    i32.const 0

Diff for: tests/compiler/builtins.optimized.wat

@@ -111,16 +111,16 @@
    local.set $2
    local.get $3
    if
-    local.get $1
+    local.get $0
     i32.load16_u
     local.tee $3
-    local.get $0
+    local.get $1
     i32.load16_u
     local.tee $4
     i32.ne
     if
-     local.get $4
      local.get $3
+     local.get $4
      i32.sub
      return
     end

Diff for: tests/compiler/comma.optimized.wat

@@ -76,21 +76,10 @@
   global.set $comma/b
   i32.const 2
   global.set $comma/a
-  global.get $comma/a
+  i32.const 2
   global.set $comma/b
-  global.get $comma/b
-  global.set $comma/a
-  global.get $comma/a
   i32.const 2
-  i32.ne
-  if
-   i32.const 0
-   i32.const 1040
-   i32.const 18
-   i32.const 1
-   call $~lib/builtins/abort
-   unreachable
-  end
+  global.set $comma/a
   global.get $comma/b
   i32.const 2
   i32.ne