Emscripten stack simplification (#1870)

This takes advantage of the recent memory simplification in emscripten, where JS static allocation is done at compile time. That means we know the stack's initial location at compile time, and can apply it. This is the binaryen side of that:

* asm2wasm support for asm.js globals with an initial value var X = Y; where Y is not 0 (which is what the stack now is).
* wasm-emscripten-finalize support for a flag --initial-stack-pointer=X, and remove the old code to import the stack's initial location.

22 files changed, 82 insertions, 40 deletions

diff --git a/auto_update_tests.py b/auto_update_tests.py
index 9d156472c..bef6a8908 100755
--- a/auto_update_tests.py
+++ b/auto_update_tests.py
@@ -87,7 +87,7 @@ def update_lld_tests():
       if ext != '.out' and not os.path.exists(out_path):
         continue
       cmd = (WASM_EMSCRIPTEN_FINALIZE +
-             [wast_path, '-S', '--global-base=568'] + ext_args)
+             [wast_path, '-S', '--global-base=568', '--initial-stack-pointer=16384'] + ext_args)
       actual = run_command(cmd)
       with open(out_path, 'w') as o:
         o.write(actual)
diff --git a/scripts/test/lld.py b/scripts/test/lld.py
index 03a0733e6..03c96d6c3 100755
--- a/scripts/test/lld.py
+++ b/scripts/test/lld.py
@@ -38,7 +38,7 @@ def test_wasm_emscripten_finalize():
         continue
 
       cmd = (WASM_EMSCRIPTEN_FINALIZE +
-             [wast_path, '-S', '--global-base=568'] + ext_args)
+             [wast_path, '-S', '--global-base=568', '--initial-stack-pointer=16384'] + ext_args)
       actual = run_command(cmd)
 
       if not os.path.exists(expected_file):
diff --git a/src/asm2wasm.h b/src/asm2wasm.h
index d958c4b6a..c193791f8 100644
--- a/src/asm2wasm.h
+++ b/src/asm2wasm.h
@@ -426,13 +426,16 @@ public:
   std::map<IString, MappedGlobal> mappedGlobals;
 
 private:
-  void allocateGlobal(IString name, Type type) {
+  void allocateGlobal(IString name, Type type, Literal value=Literal()) {
     assert(mappedGlobals.find(name) == mappedGlobals.end());
+    if (value.type == none) {
+      value = Literal::makeZero(type);
+    }
     mappedGlobals.emplace(name, MappedGlobal(type));
     wasm.addGlobal(builder.makeGlobal(
       name,
       type,
-      LiteralUtils::makeZero(type, wasm),
+      builder.makeConst(value),
       Builder::Mutable
     ));
   }
@@ -986,8 +989,7 @@ void Asm2WasmBuilder::processAsm(Ref ast) {
         Ref value = pair[1];
         if (value->isNumber()) {
           // global int
-          assert(value->getNumber() == 0);
-          allocateGlobal(name, Type::i32);
+          allocateGlobal(name, Type::i32, Literal(int32_t(value->getInteger())));
         } else if (value[0] == BINARY) {
           // int import
           assert(value[1] == OR && value[3]->isNumber() && value[3]->getNumber() == 0);
diff --git a/src/tools/wasm-emscripten-finalize.cpp b/src/tools/wasm-emscripten-finalize.cpp
index 3d74138cc..3ea3629c5 100644
--- a/src/tools/wasm-emscripten-finalize.cpp
+++ b/src/tools/wasm-emscripten-finalize.cpp
@@ -36,6 +36,8 @@ using namespace cashew;
 using namespace wasm;
 
 int main(int argc, const char *argv[]) {
+  const uint64_t INVALID_BASE = -1;
+
   std::string infile;
   std::string outfile;
   std::string inputSourceMapFilename;
@@ -46,7 +48,8 @@ int main(int argc, const char *argv[]) {
   bool debugInfo = false;
   bool legalizeJavaScriptFFI = true;
   unsigned numReservedFunctionPointers = 0;
-  uint64_t globalBase;
+  uint64_t globalBase = INVALID_BASE;
+  uint64_t initialStackPointer = INVALID_BASE;
   Options options("wasm-emscripten-finalize",
                   "Performs Emscripten-specific transforms on .wasm files");
   options
@@ -75,11 +78,16 @@ int main(int argc, const char *argv[]) {
                                           const std::string &argument) {
              numReservedFunctionPointers = std::stoi(argument);
            })
-      .add("--global-base", "", "Where lld started to place globals",
+      .add("--global-base", "", "The address at which static globals were placed",
            Options::Arguments::One,
            [&globalBase](Options*, const std::string&argument ) {
              globalBase = std::stoull(argument);
            })
+      .add("--initial-stack-pointer", "", "The initial location of the stack pointer",
+           Options::Arguments::One,
+           [&initialStackPointer](Options*, const std::string&argument ) {
+             initialStackPointer = std::stoull(argument);
+           })
 
       .add("--input-source-map", "-ism", "Consume source map from the specified file",
            Options::Arguments::One,
@@ -141,6 +149,12 @@ int main(int argc, const char *argv[]) {
   uint32_t dataSize = 0;
 
   if (!isSideModule) {
+    if (globalBase == INVALID_BASE) {
+      Fatal() << "globalBase must be set";
+    }
+    if (initialStackPointer == INVALID_BASE) {
+      Fatal() << "initialStackPointer must be set";
+    }
     Export* dataEndExport = wasm.getExport("__data_end");
     if (dataEndExport == nullptr) {
       Fatal() << "__data_end export not found";
@@ -189,7 +203,7 @@ int main(int argc, const char *argv[]) {
   } else {
     generator.generateRuntimeFunctions();
     generator.generateMemoryGrowthFunction();
-    generator.generateStackInitialization();
+    generator.generateStackInitialization(initialStackPointer);
     // emscripten calls this by default for side libraries so we only need
     // to include in as a static ctor for main module case.
     if (wasm.getExportOrNull("__post_instantiate")) {
diff --git a/src/wasm-emscripten.h b/src/wasm-emscripten.h
index 2b626a7c9..d078cafbf 100644
--- a/src/wasm-emscripten.h
+++ b/src/wasm-emscripten.h
@@ -36,7 +36,7 @@ public:
 
   void generateRuntimeFunctions();
   Function* generateMemoryGrowthFunction();
-  void generateStackInitialization();
+  void generateStackInitialization(Address addr);
 
   // Create thunks for use with emscripten Runtime.dynCall. Creates one for each
   // signature in the indirect function table.
diff --git a/src/wasm/wasm-emscripten.cpp b/src/wasm/wasm-emscripten.cpp
index b18fe0c76..88e66ea07 100644
--- a/src/wasm/wasm-emscripten.cpp
+++ b/src/wasm/wasm-emscripten.cpp
@@ -156,19 +156,9 @@ Function* EmscriptenGlueGenerator::generateMemoryGrowthFunction() {
   return growFunction;
 }
 
-void EmscriptenGlueGenerator::generateStackInitialization() {
-  // Replace a global with a constant initial value with an imported
-  // initial value, which emscripten JS will send us.
-  // TODO: with mutable imported globals, we can avoid adding another
-  //       global for the import.
-  Builder builder(wasm);
-  auto* import = builder.makeGlobal(STACK_INIT, i32, nullptr, Builder::Immutable);
-  import->module = ENV;
-  import->base = STACKTOP;
-  wasm.addGlobal(import);
+void EmscriptenGlueGenerator::generateStackInitialization(Address addr) {
   auto* stackPointer = getStackPointerGlobal();
-  assert(stackPointer->init->is<Const>());
-  stackPointer->init = builder.makeGetGlobal(import->name, i32);
+  stackPointer->init->cast<Const>()->value = Literal(int32_t(addr));
 }
 
 static bool hasI64ResultOrParam(FunctionType* ft) {
diff --git a/test/lld/duplicate_imports.wast.out b/test/lld/duplicate_imports.wast.out
index 6d3417b26..58f5846e6 100644
--- a/test/lld/duplicate_imports.wast.out
+++ b/test/lld/duplicate_imports.wast.out
@@ -9,7 +9,6 @@
  (type $legaltype$puts2 (func (param i32 i32) (result i32)))
  (type $legaltype$invoke_ffd (func (param i32 f64 f64) (result f64)))
  (type $legaltype$invoke_ffd2 (func (param i32 f64 f64) (result f64)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "puts" (func $puts1 (param i32) (result i32)))
  (import "env" "puts" (func $legalimport$puts2 (param i32 i32) (result i32)))
  (import "env" "invoke_ffd" (func $legalimport$invoke_ffd (param i32 f64 f64) (result f64)))
@@ -17,7 +16,7 @@
  (memory $0 2)
  (data (i32.const 568) "Hello, world\00")
  (table $0 1 1 funcref)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66128))
  (global $global$2 i32 (i32.const 581))
  (export "memory" (memory $0))
diff --git a/test/lld/em_asm.wast.out b/test/lld/em_asm.wast.out
index 009dedc55..92b1ea76f 100644
--- a/test/lld/em_asm.wast.out
+++ b/test/lld/em_asm.wast.out
@@ -7,14 +7,13 @@
  (type $FUNCSIG$ii (func (param i32) (result i32)))
  (type $FUNCSIG$iiii (func (param i32 i32 i32) (result i32)))
  (type $FUNCSIG$iii (func (param i32 i32) (result i32)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "emscripten_asm_const_i" (func $emscripten_asm_const_i (param i32) (result i32)))
  (import "env" "emscripten_asm_const_iii" (func $emscripten_asm_const_iii (param i32 i32 i32) (result i32)))
  (import "env" "emscripten_asm_const_ii" (func $emscripten_asm_const_ii (param i32 i32) (result i32)))
  (memory $0 2)
  (data (i32.const 568) "{ Module.print(\"Hello world\"); }\00{ return $0 + $1; }\00{ Module.print(\"Got \" + $0); }\00")
  (table $0 1 1 funcref)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66192))
  (global $global$2 i32 (i32.const 652))
  (export "memory" (memory $0))
diff --git a/test/lld/em_asm_table.wast.out b/test/lld/em_asm_table.wast.out
index 35e593dd7..ae3675d40 100644
--- a/test/lld/em_asm_table.wast.out
+++ b/test/lld/em_asm_table.wast.out
@@ -4,12 +4,11 @@
  (type $FUNCSIG$vii (func (param i32 i32)))
  (type $FUNCSIG$iiii (func (param i32 i32 i32) (result i32)))
  (import "env" "memory" (memory $0 8192))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "emscripten_log" (func $fimport$0 (param i32 i32)))
  (import "env" "emscripten_asm_const_iii" (func $emscripten_asm_const_iii (param i32 i32 i32) (result i32)))
  (table $0 159609 funcref)
  (elem (i32.const 1) $fimport$0 $emscripten_asm_const_iii)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 1048))
  (export "__data_end" (global $global$1))
  (export "stackSave" (func $stackSave))
diff --git a/test/lld/hello_world.wast.mem.out b/test/lld/hello_world.wast.mem.out
index 078620bcd..66dafd460 100644
--- a/test/lld/hello_world.wast.mem.out
+++ b/test/lld/hello_world.wast.mem.out
@@ -3,11 +3,10 @@
  (type $1 (func (result i32)))
  (type $2 (func))
  (type $FUNCSIG$ii (func (param i32) (result i32)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "puts" (func $puts (param i32) (result i32)))
  (memory $0 2)
  (table $0 1 1 funcref)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66128))
  (global $global$2 i32 (i32.const 581))
  (export "memory" (memory $0))
diff --git a/test/lld/hello_world.wast.out b/test/lld/hello_world.wast.out
index df89a5756..1b1e199cb 100644
--- a/test/lld/hello_world.wast.out
+++ b/test/lld/hello_world.wast.out
@@ -3,12 +3,11 @@
  (type $1 (func (result i32)))
  (type $2 (func))
  (type $FUNCSIG$ii (func (param i32) (result i32)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "puts" (func $puts (param i32) (result i32)))
  (memory $0 2)
  (data (i32.const 568) "Hello, world\00")
  (table $0 1 1 funcref)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66128))
  (global $global$2 i32 (i32.const 581))
  (export "memory" (memory $0))
diff --git a/test/lld/init.wast.out b/test/lld/init.wast.out
index b702fb373..dfff8d5eb 100644
--- a/test/lld/init.wast.out
+++ b/test/lld/init.wast.out
@@ -1,11 +1,10 @@
 (module
  (type $0 (func))
  (type $1 (func (result i32)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (memory $0 2)
  (data (i32.const 568) "\00\00\00\00\00\00\00\00")
  (table $0 1 1 funcref)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66112))
  (global $global$2 i32 (i32.const 576))
  (export "memory" (memory $0))
diff --git a/test/lld/recursive.wast.out b/test/lld/recursive.wast.out
index 1c1180e1c..e8f32d99c 100644
--- a/test/lld/recursive.wast.out
+++ b/test/lld/recursive.wast.out
@@ -3,12 +3,11 @@
  (type $1 (func (result i32)))
  (type $2 (func))
  (type $FUNCSIG$iii (func (param i32 i32) (result i32)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "printf" (func $printf (param i32 i32) (result i32)))
  (memory $0 2)
  (data (i32.const 568) "%d:%d\n\00Result: %d\n\00")
  (table $0 1 1 funcref)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66128))
  (global $global$2 i32 (i32.const 587))
  (export "memory" (memory $0))
diff --git a/test/lld/reserved_func_ptr.wast.jscall.out b/test/lld/reserved_func_ptr.wast.jscall.out
index 6242c3b76..bc9b677d6 100644
--- a/test/lld/reserved_func_ptr.wast.jscall.out
+++ b/test/lld/reserved_func_ptr.wast.jscall.out
@@ -18,7 +18,6 @@
  (type $FUNCSIG$v (func))
  (type $FUNCSIG$vii (func (param i32 i32)))
  (type $FUNCSIG$viiii (func (param i32 i32 i32 i32)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "_Z4atoiPKc" (func $_Z4atoiPKc (param i32) (result i32)))
  (import "env" "jsCall_ddi" (func $jsCall_ddi (param i32 f64 i32) (result f64)))
  (import "env" "jsCall_fffi" (func $jsCall_fffi (param i32 f32 f32 i32) (result f32)))
@@ -29,7 +28,7 @@
  (memory $0 2)
  (table $0 21 21 funcref)
  (elem (i32.const 1) $_Z18address_taken_funciii $_Z19address_taken_func2iii $jsCall_ddi_0 $jsCall_ddi_1 $jsCall_ddi_2 $jsCall_fffi_0 $jsCall_fffi_1 $jsCall_fffi_2 $jsCall_iii_0 $jsCall_iii_1 $jsCall_iii_2 $jsCall_v_0 $jsCall_v_1 $jsCall_v_2 $jsCall_vi_0 $jsCall_vi_1 $jsCall_vi_2 $jsCall_viii_0 $jsCall_viii_1 $jsCall_viii_2)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66112))
  (global $global$2 i32 (i32.const 568))
  (export "memory" (memory $0))
diff --git a/test/lld/reserved_func_ptr.wast.out b/test/lld/reserved_func_ptr.wast.out
index 393178a00..e0d9ea156 100644
--- a/test/lld/reserved_func_ptr.wast.out
+++ b/test/lld/reserved_func_ptr.wast.out
@@ -8,12 +8,11 @@
  (type $6 (func (param i32) (result i32)))
  (type $FUNCSIG$ii (func (param i32) (result i32)))
  (type $FUNCSIG$viii (func (param i32 i32 i32)))
- (import "env" "STACKTOP" (global $stack$init i32))
  (import "env" "_Z4atoiPKc" (func $_Z4atoiPKc (param i32) (result i32)))
  (memory $0 2)
  (table $0 3 3 funcref)
  (elem (i32.const 1) $_Z18address_taken_funciii $_Z19address_taken_func2iii)
- (global $global$0 (mut i32) (global.get $stack$init))
+ (global $global$0 (mut i32) (i32.const 16384))
  (global $global$1 i32 (i32.const 66112))
  (global $global$2 i32 (i32.const 568))
  (export "memory" (memory $0))
diff --git a/test/unit.asm.js b/test/unit.asm.js
index 05827ce54..b62c7aefb 100644
--- a/test/unit.asm.js
+++ b/test/unit.asm.js
@@ -30,6 +30,8 @@ function asm(global, env, buffer) {
   var HEAPF32 = new global.Float32Array(buffer);
   var HEAPF64 = new global.Float64Array(buffer);
 
+  var nonZero = 1337;
+
   function big_negative() {
     var temp = 0.0;
     temp = +-2147483648;
@@ -785,6 +787,7 @@ function asm(global, env, buffer) {
     emterpretify_assertions_safeHeap();
     call_emscripten_log();
     mod_detectSign(1.0, 2.31, 9.78);
+    nonZero = nonZero + 1 | 0;
   }
 
   function v() {
diff --git a/test/unit.fromasm b/test/unit.fromasm
index b1a24bf5d..c9e20da52 100644
--- a/test/unit.fromasm
+++ b/test/unit.fromasm
@@ -26,6 +26,7 @@
  (global $Int (mut i32) (i32.const 0))
  (global $Double (mut f64) (f64.const 0))
  (global $n (mut i32) (global.get $n$asm2wasm$import))
+ (global $nonZero (mut i32) (i32.const 1337))
  (global $exportedNumber i32 (i32.const 42))
  (export "big_negative" (func $big_negative))
  (export "pick" (func $big_negative))
@@ -1162,6 +1163,12 @@
     )
    )
   )
+  (global.set $nonZero
+   (i32.add
+    (global.get $nonZero)
+    (i32.const 1)
+   )
+  )
  )
  (func $vi (; 55 ;) (; has Stack IR ;) (param $0 i32)
   (nop)
diff --git a/test/unit.fromasm.clamp b/test/unit.fromasm.clamp
index 58cf67a0d..deec92f97 100644
--- a/test/unit.fromasm.clamp
+++ b/test/unit.fromasm.clamp
@@ -24,6 +24,7 @@
  (global $Int (mut i32) (i32.const 0))
  (global $Double (mut f64) (f64.const 0))
  (global $n (mut i32) (global.get $n$asm2wasm$import))
+ (global $nonZero (mut i32) (i32.const 1337))
  (global $exportedNumber i32 (i32.const 42))
  (export "big_negative" (func $big_negative))
  (export "pick" (func $big_negative))
@@ -1238,6 +1239,12 @@
     )
    )
   )
+  (global.set $nonZero
+   (i32.add
+    (global.get $nonZero)
+    (i32.const 1)
+   )
+  )
  )
  (func $vi (; 57 ;) (; has Stack IR ;) (param $0 i32)
   (nop)
diff --git a/test/unit.fromasm.clamp.no-opts b/test/unit.fromasm.clamp.no-opts
index 7a30149ba..95cb543e2 100644
--- a/test/unit.fromasm.clamp.no-opts
+++ b/test/unit.fromasm.clamp.no-opts
@@ -31,6 +31,7 @@
  (global $tempDoublePtr (mut i32) (global.get $tempDoublePtr$asm2wasm$import))
  (global $n (mut i32) (global.get $n$asm2wasm$import))
  (global $STACKTOP (mut i32) (global.get $STACKTOP$asm2wasm$import))
+ (global $nonZero (mut i32) (i32.const 1337))
  (global $exportedNumber i32 (i32.const 42))
  (export "big_negative" (func $big_negative))
  (export "pick" (func $exportMe))
@@ -2215,6 +2216,12 @@
     (f64.const 9.78)
    )
   )
+  (global.set $nonZero
+   (i32.add
+    (global.get $nonZero)
+    (i32.const 1)
+   )
+  )
  )
  (func $v (; 83 ;)
   (nop)
diff --git a/test/unit.fromasm.imprecise b/test/unit.fromasm.imprecise
index 869873ec1..7ff7a4c18 100644
--- a/test/unit.fromasm.imprecise
+++ b/test/unit.fromasm.imprecise
@@ -22,6 +22,7 @@
  (global $Int (mut i32) (i32.const 0))
  (global $Double (mut f64) (f64.const 0))
  (global $n (mut i32) (global.get $n$asm2wasm$import))
+ (global $nonZero (mut i32) (i32.const 1337))
  (global $exportedNumber i32 (i32.const 42))
  (export "big_negative" (func $big_negative))
  (export "pick" (func $big_negative))
@@ -1141,6 +1142,12 @@
     (f64.const 1)
    )
   )
+  (global.set $nonZero
+   (i32.add
+    (global.get $nonZero)
+    (i32.const 1)
+   )
+  )
  )
  (func $vi (; 54 ;) (; has Stack IR ;) (param $0 i32)
   (nop)
diff --git a/test/unit.fromasm.imprecise.no-opts b/test/unit.fromasm.imprecise.no-opts
index 74dd7efee..69d1ff2ff 100644
--- a/test/unit.fromasm.imprecise.no-opts
+++ b/test/unit.fromasm.imprecise.no-opts
@@ -31,6 +31,7 @@
  (global $tempDoublePtr (mut i32) (global.get $tempDoublePtr$asm2wasm$import))
  (global $n (mut i32) (global.get $n$asm2wasm$import))
  (global $STACKTOP (mut i32) (global.get $STACKTOP$asm2wasm$import))
+ (global $nonZero (mut i32) (i32.const 1337))
  (global $exportedNumber i32 (i32.const 42))
  (export "big_negative" (func $big_negative))
  (export "pick" (func $exportMe))
@@ -2100,6 +2101,12 @@
     (f64.const 9.78)
    )
   )
+  (global.set $nonZero
+   (i32.add
+    (global.get $nonZero)
+    (i32.const 1)
+   )
+  )
  )
  (func $v (; 78 ;)
   (nop)
diff --git a/test/unit.fromasm.no-opts b/test/unit.fromasm.no-opts
index 561062f04..fb82b572f 100644
--- a/test/unit.fromasm.no-opts
+++ b/test/unit.fromasm.no-opts
@@ -33,6 +33,7 @@
  (global $tempDoublePtr (mut i32) (global.get $tempDoublePtr$asm2wasm$import))
  (global $n (mut i32) (global.get $n$asm2wasm$import))
  (global $STACKTOP (mut i32) (global.get $STACKTOP$asm2wasm$import))
+ (global $nonZero (mut i32) (i32.const 1337))
  (global $exportedNumber i32 (i32.const 42))
  (export "big_negative" (func $big_negative))
  (export "pick" (func $exportMe))
@@ -2141,6 +2142,12 @@
     (f64.const 9.78)
    )
   )
+  (global.set $nonZero
+   (i32.add
+    (global.get $nonZero)
+    (i32.const 1)
+   )
+  )
  )
  (func $v (; 81 ;)
   (nop)