feat: Improve pass pipeline when flattening code (#1368)

Author: MaxGraey

src/module.ts

@@ -1512,6 +1512,7 @@ export class Module {
         passes.push("flatten");
         passes.push("simplify-locals-notee-nostructure");
         passes.push("vacuum");
+
         passes.push("code-folding");
         passes.push("flatten");
         passes.push("local-cse");
@@ -1529,8 +1530,10 @@ export class Module {
       //   passes.push("post-assemblyscript");
       // }
       passes.push("optimize-instructions");
-      passes.push("inlining");
-      passes.push("dce");
+      if (optimizeLevel >= 3 || shrinkLevel >= 1) {
+        passes.push("dce");
+        passes.push("inlining");
+      }
       passes.push("remove-unused-brs");
       passes.push("remove-unused-names");
       passes.push("inlining-optimizing");
@@ -1549,11 +1552,13 @@ export class Module {
       }
       passes.push("simplify-locals-nostructure");
       passes.push("vacuum");
+
       passes.push("reorder-locals");
       passes.push("remove-unused-brs");
       passes.push("coalesce-locals");
       passes.push("simplify-locals");
       passes.push("vacuum");
+
       passes.push("reorder-locals");
       passes.push("coalesce-locals");
       passes.push("reorder-locals");
@@ -1576,8 +1581,8 @@ export class Module {
       }
       if (optimizeLevel >= 2 || shrinkLevel >= 1) {
         passes.push("rse");
+        passes.push("vacuum");
       }
-      passes.push("vacuum");
 
       // --- PassRunner::addDefaultGlobalOptimizationPostPasses ---
 
@@ -1623,22 +1628,30 @@ export class Module {
           // very expensive, so O3 only
           passes.push("simplify-globals");
           passes.push("vacuum");
+
+          passes.push("precompute-propagate");
+          passes.push("vacuum");
+
           // replace indirect with direct calls again and inline
           passes.push("inlining-optimizing");
           passes.push("directize");
           passes.push("dae-optimizing");
-          passes.push("precompute-propagate");
-          passes.push("vacuum");
+
           passes.push("merge-locals");
           passes.push("coalesce-locals");
-          passes.push("simplify-locals-nostructure");
+          passes.push("simplify-locals");
           passes.push("vacuum");
+
           passes.push("inlining-optimizing");
           passes.push("precompute-propagate");
+          passes.push("vacuum");
+
+          passes.push("coalesce-locals");
         }
         passes.push("remove-unused-brs");
         passes.push("remove-unused-names");
         passes.push("vacuum");
+
         passes.push("optimize-instructions");
         passes.push("simplify-globals-optimizing");
       }

tests/compiler/binary.optimized.wat

@@ -114,17 +114,17 @@
      local.get $1
      i32.const 8388608
      i32.ge_u
-     if
+     if (result i32)
       local.get $1
       i32.const 8388608
       i32.eq
       br_if $folding-inner0
       local.get $1
       i32.const 8388608
       i32.sub
-      local.set $1
+     else
+      local.get $1
      end
-     local.get $1
      i32.const 1
      i32.shl
      local.set $1
@@ -261,17 +261,17 @@
      local.get $1
      i64.const 4503599627370496
      i64.ge_u
-     if
+     if (result i64)
       local.get $1
       i64.const 4503599627370496
       i64.eq
       br_if $folding-inner0
       local.get $1
       i64.const 4503599627370496
       i64.sub
-      local.set $1
+     else
+      local.get $1
      end
-     local.get $1
      i64.const 1
      i64.shl
      local.set $1

tests/compiler/class-overloading.optimized.wat

@@ -98,12 +98,11 @@
    local.set $0
   end
   local.get $0
-  i32.eqz
   if (result i32)
+   local.get $0
+  else
    i32.const 3
    call $~lib/rt/stub/__alloc
-  else
-   local.get $0
   end
  )
  (func $~lib/string/String#get:length (param $0 i32) (result i32)