Enable StackIR optimizations in release mode and for generating WAT

In WebAssembly/binaryen#6568, StackIR was changed from a set of passes
to a global option. This meant that the `true` argument passed into
_BinaryenModuleAllocateAndWriteStackIR (to optimize the StackIR) no
longer had an effect. This commit restores that and also runs the
optimizations under the same {optimize,shrink}Levels as wasm-opt.

Author: CountBleck

cli/index.js

@@ -945,9 +945,10 @@ export async function main(argv, options) {
       try {
         // use superset text format when extension is `.wast`.
         // Otherwise use official stack IR format (wat).
+        binaryen.setOptimizeStackIR(true);
         out = opts.textFile?.endsWith(".wast")
           ? binaryenModule.emitText()
-          : binaryenModule.emitStackIR(true);
+          : binaryenModule.emitStackIR();
       } catch (e) {
         crash("emitText", e);
       }

src/module.ts

@@ -2503,6 +2503,22 @@ export class Module {
     binaryen._BinaryenSetFastMath(on);
   }
 
+  getGenerateStackIR(): bool {
+    return binaryen._BinaryenGetGenerateStackIR();
+  }
+
+  setGenerateStackIR(on: bool): void {
+    binaryen._BinaryenSetGenerateStackIR(on);
+  }
+
+  getOptimizeStackIR(): bool {
+    return binaryen._BinaryenGetOptimizeStackIR();
+  }
+
+  setOptimizeStackIR(on: bool): void {
+    binaryen._BinaryenSetOptimizeStackIR(on);
+  }
+
   getPassArgument(key: string): string | null {
     let cStr = this.allocStringCached(key);
     let ptr = binaryen._BinaryenGetPassArgument(cStr);
@@ -2594,6 +2610,11 @@ export class Module {
     this.setFastMath(true);
     this.clearPassArguments();
 
+    // OptimizationOptions#parse in src/tools/optimization-options.h
+    const stackIR = optimizeLevel >= 2 || shrinkLevel >= 1;
+    this.setGenerateStackIR(stackIR);
+    this.setOptimizeStackIR(stackIR);
+
     // Tweak inlining limits based on optimization levels
     if (optimizeLevel >= 2 && shrinkLevel == 0) {
       this.setAlwaysInlineMaxSize(12);

tests/compiler/NonNullable.debug.wat

@@ -105,40 +105,38 @@
     end
    end
   end
-  block $while-break|1
-   loop $while-continue|1
-    local.get $len
-    local.tee $7
-    i32.const 1
-    i32.sub
-    local.set $len
-    local.get $7
+  loop $while-continue|1
+   local.get $len
+   local.tee $7
+   i32.const 1
+   i32.sub
+   local.set $len
+   local.get $7
+   if
+    local.get $ptr1
+    i32.load16_u
+    local.set $a
+    local.get $ptr2
+    i32.load16_u
+    local.set $b
+    local.get $a
+    local.get $b
+    i32.ne
     if
-     local.get $ptr1
-     i32.load16_u
-     local.set $a
-     local.get $ptr2
-     i32.load16_u
-     local.set $b
      local.get $a
      local.get $b
-     i32.ne
-     if
-      local.get $a
-      local.get $b
-      i32.sub
-      return
-     end
-     local.get $ptr1
-     i32.const 2
-     i32.add
-     local.set $ptr1
-     local.get $ptr2
-     i32.const 2
-     i32.add
-     local.set $ptr2
-     br $while-continue|1
+     i32.sub
+     return
     end
+    local.get $ptr1
+    i32.const 2
+    i32.add
+    local.set $ptr1
+    local.get $ptr2
+    i32.const 2
+    i32.add
+    local.set $ptr2
+    br $while-continue|1
    end
   end
   i32.const 0

tests/compiler/NonNullable.release.wat

@@ -330,12 +330,10 @@
   i32.const 0
   call $~lib/string/String.__eq
   i32.eqz
-  local.set $0
   global.get $~lib/memory/__stack_pointer
   i32.const 8
   i32.add
   global.set $~lib/memory/__stack_pointer
-  local.get $0
  )
  (func $NonNullable/assertNonNull<~lib/string/String>
   global.get $~lib/memory/__stack_pointer

tests/compiler/assert-nonnull.release.wat

@@ -174,7 +174,6 @@
    i32.store
    local.get $0
    i32.load
-   local.set $0
    global.get $~lib/memory/__stack_pointer
    i32.const 8
    i32.add
@@ -183,7 +182,6 @@
    i32.const 4
    i32.add
    global.set $~lib/memory/__stack_pointer
-   local.get $0
    return
   end
   i32.const 34256
@@ -453,7 +451,6 @@
     i32.eqz
     br_if $folding-inner1
     global.get $~lib/memory/__stack_pointer
-    local.set $1
     global.get $~lib/memory/__stack_pointer
     local.get $0
     i32.store offset=4
@@ -468,7 +465,6 @@
     global.get $~lib/memory/__stack_pointer
     local.get $0
     i32.store
-    local.get $1
     local.get $0
     i32.load
     local.tee $0
@@ -535,7 +531,6 @@
     i32.eqz
     br_if $folding-inner1
     global.get $~lib/memory/__stack_pointer
-    local.set $1
     global.get $~lib/memory/__stack_pointer
     local.get $0
     i32.store offset=4
@@ -550,7 +545,6 @@
     global.get $~lib/memory/__stack_pointer
     local.get $0
     i32.store
-    local.get $1
     local.get $0
     i32.load
     local.tee $0
@@ -623,7 +617,6 @@
    local.get $0
    i32.load
    call_indirect (type $1)
-   drop
    unreachable
   end
   i32.const 34256
@@ -662,7 +655,6 @@
    local.get $0
    i32.store
    global.get $~lib/memory/__stack_pointer
-   local.set $1
    local.get $0
    i32.eqz
    if
@@ -673,13 +665,11 @@
     call $~lib/builtins/abort
     unreachable
    end
-   local.get $1
    local.get $0
    i32.store offset=4
    local.get $0
    i32.load
    call_indirect (type $1)
-   drop
    unreachable
   end
   i32.const 34256
@@ -723,7 +713,6 @@
     local.get $0
     i32.load
     call_indirect (type $1)
-    drop
     unreachable
    end
    i32.const 34256
@@ -785,7 +774,6 @@
    local.get $0
    i32.load
    call_indirect (type $1)
-   drop
    unreachable
   end
   i32.const 34256
@@ -837,7 +825,6 @@
     local.get $0
     i32.load
     call_indirect (type $1)
-    drop
     unreachable
    end
    i32.const 34256