Merge pull request #244 from WebAssembly/interpret-binary

wasm-binary method

3 files changed, 131 insertions, 55 deletions

diff --git a/src/js/wasm.js-post.js b/src/js/wasm.js-post.js
index e3ab2bb4f..12e9e5315 100644
--- a/src/js/wasm.js-post.js
+++ b/src/js/wasm.js-post.js
@@ -17,8 +17,9 @@
 function integrateWasmJS(Module) {
   // wasm.js has several methods for creating the compiled code module here:
   //  * 'native-wasm' : use native WebAssembly support in the browser
-  //  * 'wasm-s-parser': load s-expression code from a .wast and create wasm
-  //  * 'asm2wasm': load asm.js code and translate to wasm
+  //  * 'wasm-s-parser': load s-expression code from a .wast and interpret
+  //  * 'wasm-binary': load binary wasm and interpret
+  //  * 'asm2wasm': load asm.js code, translate to wasm, and interpret
   //  * 'just-asm': no wasm, just load the asm.js code and use that (good for testing)
   // The method can be set at compile time (BINARYEN_METHOD), or runtime by setting Module['wasmJSMethod'].
   // The method can be a comma-separated list, in which case, we will try the
@@ -29,8 +30,8 @@ function integrateWasmJS(Module) {
 
   var method = Module['wasmJSMethod'] || {{{ wasmJSMethod }}} || 'native-wasm,wasm-s-parser'; // by default, try native and then .wast
 
-  var wasmCodeFile = Module['wasmCodeFile'] || {{{ wasmCodeFile }}};
-
+  var wasmTextFile = Module['wasmTextFile'] || {{{ wasmTextFile }}};
+  var wasmBinaryFile = Module['wasmBinaryFile'] || {{{ wasmBinaryFile }}};
   var asmjsCodeFile = Module['asmjsCodeFile'] || {{{ asmjsCodeFile }}};
 
   // utilities
@@ -109,8 +110,8 @@ function integrateWasmJS(Module) {
   // be present in the wasm output of asm2wasm, so we store it in a side file. If we load asm2wasm
   // output, either generated ahead of time or on the client, we need to apply those mapped
   // globals after loading the module.
-  function applyMappedGlobals() {
-    var mappedGlobals = JSON.parse(Module['read'](wasmCodeFile + '.mappedGlobals'));
+  function applyMappedGlobals(globalsFileBase) {
+    var mappedGlobals = JSON.parse(Module['read'](globalsFileBase + '.mappedGlobals'));
     for (var name in mappedGlobals) {
       var global = mappedGlobals[name];
       if (!global.import) continue; // non-imports are initialized to zero in the typed array anyhow, so nothing to do here
@@ -136,6 +137,18 @@ function integrateWasmJS(Module) {
     return ret;
   }
 
+  function getBinary() {
+    var binary;
+    if (ENVIRONMENT_IS_WEB || ENVIRONMENT_IS_WORKER) {
+      binary = Module['wasmBinary'];
+      assert(binary, "on the web, we need the wasm binary to be preloaded and set on Module['wasmBinary']. emcc.py will do that for you when generating HTML (but not JS)");
+      binary = new Uint8Array(binary);
+    } else {
+      binary = Module['readBinary'](wasmBinaryFile);
+    }
+    return binary;
+  }
+
   // do-method functions
 
   function doJustAsm() {
@@ -161,16 +174,7 @@ function integrateWasmJS(Module) {
       global = fixImports(global);
       env = fixImports(env);
 
-      // Load the wasm module
-      var binary;
-      if (ENVIRONMENT_IS_WEB || ENVIRONMENT_IS_WORKER) {
-        binary = Module['wasmBinary'];
-        assert(binary, "on the web, we need the wasm binary to be preloaded and set on Module['wasmBinary']. emcc.py will do that for you when generating HTML (but not JS)");
-        binary = new Uint8Array(binary);
-      } else {
-        binary = Module['readBinary'](wasmCodeFile);
-      }
-      // Create an instance of the module using native support in the JS engine.
+      // Load the wasm module and create an instance of using native support in the JS engine.
       info['global'] = {
         'NaN': NaN,
         'Infinity': Infinity
@@ -178,10 +182,10 @@ function integrateWasmJS(Module) {
       info['global.Math'] = global.Math;
       info['env'] = env;
       var instance;
-      instance = Wasm.instantiateModule(binary, info);
+      instance = Wasm.instantiateModule(getBinary(), info);
       mergeMemory(instance.exports.memory);
 
-      applyMappedGlobals();
+      applyMappedGlobals(wasmBinaryFile);
 
       return instance.exports;
     };
@@ -226,13 +230,27 @@ function integrateWasmJS(Module) {
       wasmJS['providedTotalMemory'] = Module['buffer'].byteLength;
 
       // Prepare to generate wasm, using either asm2wasm or wasm-s-parser
-      var code = Module['read'](method == 'asm2wasm' ? asmjsCodeFile : wasmCodeFile);
-      var temp = wasmJS['_malloc'](code.length + 1);
-      wasmJS['writeAsciiToMemory'](code, temp);
+      var code;
+      if (method === 'wasm-binary') {
+        code = getBinary();
+      } else {
+        code = Module['read'](method == 'asm2wasm' ? asmjsCodeFile : wasmTextFile);
+      }
+      var temp;
       if (method == 'asm2wasm') {
+        temp = wasmJS['_malloc'](code.length + 1);
+        wasmJS['writeAsciiToMemory'](code, temp);
         wasmJS['_load_asm2wasm'](temp);
-      } else {
+      } else if (method === 'wasm-s-parser') {
+        temp = wasmJS['_malloc'](code.length + 1);
+        wasmJS['writeAsciiToMemory'](code, temp);
         wasmJS['_load_s_expr2wasm'](temp);
+      } else if (method === 'wasm-binary') {
+        temp = wasmJS['_malloc'](code.length);
+        wasmJS['HEAPU8'].set(code, temp);
+        wasmJS['_load_binary2wasm'](temp, code.length);
+      } else {
+        throw 'what? ' + method;
       }
       wasmJS['_free'](temp);
 
@@ -244,7 +262,9 @@ function integrateWasmJS(Module) {
       }
 
       if (method == 'wasm-s-parser') {
-        applyMappedGlobals();
+        applyMappedGlobals(wasmTextFile);
+      } else if (method == 'wasm-binary') {
+        applyMappedGlobals(wasmBinaryFile);
       }
 
       return wasmJS['asmExports'];
@@ -263,7 +283,7 @@ function integrateWasmJS(Module) {
       if (doNativeWasm()) return;
     } else if (curr === 'just-asm') {
       if (doJustAsm()) return;
-    } else if (curr === 'asm2wasm' || curr === 'wasm-s-parser') {
+    } else if (curr === 'asm2wasm' || curr === 'wasm-s-parser' || curr === 'wasm-binary') {
       if (doWasmPolyfill(curr)) return;
     } else {
       throw 'bad method: ' + curr;
diff --git a/src/wasm-binary.h b/src/wasm-binary.h
index 715c06287..081924ba0 100644
--- a/src/wasm-binary.h
+++ b/src/wasm-binary.h
@@ -522,7 +522,7 @@ public:
         o << int8_t(BinaryConsts::EndMarker);
         assert(depth == 0);
         size_t size = o.size() - start;
-        assert(size <= std::numeric_limits<uint16_t>::max());
+        assert(size <= std::numeric_limits<uint32_t>::max());
         if (debug) std::cerr << "body size: " << size << ", writing at " << sizePos << ", next starts at " << o.size() << std::endl;
         o.writeAt(sizePos, uint32_t(size)); // XXX int32, diverge from v8 format, to get more code to compile
       } else {
@@ -650,7 +650,7 @@ public:
     o << int8_t(BinaryConsts::EndMarker);
   }
 
-  int getBreakIndex(Name name) { // -1 if not found
+  int32_t getBreakIndex(Name name) { // -1 if not found
     for (int i = breakStack.size() - 1; i >= 0; i--) {
       if (breakStack[i] == name) {
         return breakStack.size() - 1 - i;
@@ -669,15 +669,15 @@ public:
     }
     if (curr->condition) recurse(curr->condition);
     o << int8_t(curr->condition ? BinaryConsts::BrIf : BinaryConsts::Br)
-      << int8_t(getBreakIndex(curr->name));
+      << int32_t(getBreakIndex(curr->name));
   }
   void visitSwitch(Switch *curr) {
     if (debug) std::cerr << "zz node: Switch" << std::endl;
     o << int8_t(BinaryConsts::TableSwitch) << int16_t(curr->targets.size() + 1) << int8_t(curr->value != nullptr);
     for (auto target : curr->targets) {
-      o << (int16_t)getBreakIndex(target);
+      o << (int32_t)getBreakIndex(target);
     }
-    o << (int16_t)getBreakIndex(curr->default_);
+    o << (int32_t)getBreakIndex(curr->default_);
     recurse(curr->condition);
     o << int8_t(BinaryConsts::EndMarker);
     if (curr->value) {
@@ -1079,6 +1079,12 @@ public:
     assert(x == y);
   }
 
+  void ungetInt8() {
+    assert(pos > 0);
+    if (debug) std::cerr << "ungetInt8 (at " << pos << ")" << std::endl;
+    pos--;
+  }
+
   void readStart() {
     if (debug) std::cerr << "== readStart" << std::endl;
     wasm.start = getString();
@@ -1335,19 +1341,44 @@ public:
 
   void visitBlock(Block *curr) {
     if (debug) std::cerr << "zz node: Block" << std::endl;
-    curr->name = getNextLabel();
-    breakStack.push_back(curr->name);
-    size_t start = expressionStack.size(); // everything after this, that is left when we see the marker, is ours
-    processExpressions();
-    size_t end = expressionStack.size();
-    assert(end >= start);
-    for (size_t i = start; i < end; i++) {
-      if (debug) std::cerr << "  " << size_t(expressionStack[i]) << "\n zz Block element " << curr->list.size() << std::endl;
-      curr->list.push_back(expressionStack[i]);
+    // special-case Block and de-recurse nested blocks in their first position, as that is
+    // a common pattern that can be very highly nested.
+    std::vector<Block*> stack;
+    while (1) {
+      curr->name = getNextLabel();
+      breakStack.push_back(curr->name);
+      stack.push_back(curr);
+      if (getInt8() == BinaryConsts::Block) {
+        // a recursion
+        curr = allocator.alloc<Block>();
+        continue;
+      } else {
+        // end of recursion
+        ungetInt8();
+        break;
+      }
+    }
+    Block* last = nullptr;
+    while (stack.size() > 0) {
+      curr = stack.back();
+      stack.pop_back();
+      size_t start = expressionStack.size(); // everything after this, that is left when we see the marker, is ours
+      if (last) {
+        // the previous block is our first-position element
+        expressionStack.push_back(last);
+      }
+      last = curr;
+      processExpressions();
+      size_t end = expressionStack.size();
+      assert(end >= start);
+      for (size_t i = start; i < end; i++) {
+        if (debug) std::cerr << "  " << size_t(expressionStack[i]) << "\n zz Block element " << curr->list.size() << std::endl;
+        curr->list.push_back(expressionStack[i]);
+      }
+      expressionStack.resize(start);
+      curr->finalize();
+      breakStack.pop_back();
     }
-    expressionStack.resize(start);
-    curr->finalize();
-    breakStack.pop_back();
   }
   void visitIf(If *curr, uint8_t code) {
     if (debug) std::cerr << "zz node: If" << std::endl;
@@ -1374,14 +1405,14 @@ public:
     curr->finalize();
   }
 
-  Name getBreakName(int offset) {
+  Name getBreakName(int32_t offset) {
     assert(breakStack.size() - 1 - offset < breakStack.size());
     return breakStack[breakStack.size() - 1 - offset];
   }
 
   void visitBreak(Break *curr, uint8_t code) {
     if (debug) std::cerr << "zz node: Break" << std::endl;
-    curr->name = getBreakName(getInt8());
+    curr->name = getBreakName(getInt32());
     if (code == BinaryConsts::BrIf) curr->condition = popExpression();
     curr->value = popExpression();
   }
@@ -1390,9 +1421,9 @@ public:
     auto numTargets = getInt16();
     auto hasValue = getInt8();
     for (auto i = 0; i < numTargets - 1; i++) {
-      curr->targets.push_back(getBreakName(getInt16()));
+      curr->targets.push_back(getBreakName(getInt32()));
     }
-    curr->default_ = getBreakName(getInt16());
+    curr->default_ = getBreakName(getInt32());
     processExpressions();
     curr->condition = popExpression();
     if (hasValue) {
@@ -1624,6 +1655,7 @@ public:
     if (debug) std::cerr << "zz node: Binary" << std::endl;
     curr->right = popExpression();
     curr->left = popExpression();
+    curr->finalize();
     return true;
     #undef TYPED_CODE
     #undef INT_TYPED_CODE
@@ -1656,6 +1688,7 @@ public:
       default: return false;
     }
     if (debug) std::cerr << "zz node: Host" << std::endl;
+    curr->finalize();
     return true;
   }
   void visitNop(Nop *curr) {
diff --git a/src/wasm-js.cpp b/src/wasm-js.cpp
index 000092da8..3083b3b9e 100644
--- a/src/wasm-js.cpp
+++ b/src/wasm-js.cpp
@@ -26,6 +26,7 @@
 #include "asm2wasm.h"
 #include "wasm-interpreter.h"
 #include "wasm-s-parser.h"
+#include "wasm-binary.h"
 #include "wasm-printing.h"
 
 using namespace cashew;
@@ -100,8 +101,22 @@ extern "C" void EMSCRIPTEN_KEEPALIVE load_asm2wasm(char *input) {
   }
 }
 
+void finalizeModule() {
+  uint32_t providedMemory = EM_ASM_INT_V({
+    return Module['providedTotalMemory']; // we receive the size of memory from emscripten
+  });
+  if (providedMemory & ~Memory::kPageMask) {
+    std::cerr << "Error: provided memory is not a multiple of the 64k wasm page size\n";
+    exit(EXIT_FAILURE);
+  }
+  module->memory.initial = providedMemory / Memory::kPageSize;
+  module->memory.max = (module->exportsMap.find(GROW_WASM_MEMORY) != module->exportsMap.end()) ? -1 : module->memory.initial;
+
+  // global mapping is done in js in post.js
+}
+
 // loads wasm code in s-expression format
-extern "C" void EMSCRIPTEN_KEEPALIVE load_s_expr2wasm(char *input, char *mappedGlobals) {
+extern "C" void EMSCRIPTEN_KEEPALIVE load_s_expr2wasm(char *input) {
   prepare2wasm();
 
   if (wasmJSDebug) std::cerr << "wasm-s-expression parsing...\n";
@@ -119,17 +134,25 @@ extern "C" void EMSCRIPTEN_KEEPALIVE load_s_expr2wasm(char *input, char *mappedG
     abort();
   });
 
-  uint32_t providedMemory = EM_ASM_INT_V({
-    return Module['providedTotalMemory']; // we receive the size of memory from emscripten
-  });
-  if (providedMemory & ~Memory::kPageMask) {
-    std::cerr << "Error: provided memory is not a multiple of the 64k wasm page size\n";
-    exit(EXIT_FAILURE);
+  finalizeModule();
+}
+
+// loads wasm code in binary format
+extern "C" void EMSCRIPTEN_KEEPALIVE load_binary2wasm(char *raw, int32_t size) {
+  prepare2wasm();
+
+  if (wasmJSDebug) std::cerr << "wasm-binary parsing...\n";
+
+  module = new AllocatingModule();
+  std::vector<char> input;
+  input.resize(size);
+  for (int32_t i = 0; i < size; i++) {
+    input[i] = raw[i];
   }
-  module->memory.initial = providedMemory / Memory::kPageSize;
-  module->memory.max = (module->exportsMap.find(GROW_WASM_MEMORY) != module->exportsMap.end()) ? -1 : module->memory.initial;
+  WasmBinaryBuilder parser(*module, input, debug);
+  parser.read();
 
-  // global mapping is done in js in post.js
+  finalizeModule();
 }
 
 // instantiates the loaded wasm (which might be from asm2wasm, or