[wasm-gc] Implement dataref

As per latest wasm-gc spec, the data heap type is a subtype of eq and a
supertype of all array and struct types.
The heap type expected for arrays and structs when interacting with JS
changes from eq to data.

Bug: v8:7748
Change-Id: Idd1670b9e47acc95c098559e674c629ea44ca49d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2649044
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72399}

src/compiler/wasm-compiler.cc

@@ -6243,7 +6243,7 @@ class WasmWrapperGraphBuilder : public WasmGraphBuilder {
             representation == wasm::HeapType::kFunc) {
           return node;
         }
-        if (representation == wasm::HeapType::kEq) {
+        if (representation == wasm::HeapType::kData) {
           // TODO(7748): Update this when JS interop is settled.
           return BuildAllocateObjectWrapper(node);
         }
@@ -6387,10 +6387,11 @@ class WasmWrapperGraphBuilder : public WasmGraphBuilder {
           case wasm::HeapType::kFunc:
             BuildCheckValidRefValue(input, js_context, type);
             return input;
-          case wasm::HeapType::kEq:
+          case wasm::HeapType::kData:
             // TODO(7748): Update this when JS interop has settled.
             BuildCheckValidRefValue(input, js_context, type);
             return BuildUnpackObjectWrapper(input, kReturnNull);
+          case wasm::HeapType::kEq:
           case wasm::HeapType::kI31:
             // If this is reached, then IsJSCompatibleSignature() is too
             // permissive.

src/wasm/function-body-decoder-impl.h

@@ -185,10 +185,10 @@ V8_INLINE WasmFeature feature_for_heap_type(HeapType heap_type) {
       return WasmFeature::kFeature_reftypes;
     case HeapType::kEq:
     case HeapType::kI31:
+    case HeapType::kData:
     case HeapType::kAny:
       return WasmFeature::kFeature_gc;
     case HeapType::kBottom:
-    default:
       UNREACHABLE();
   }
 }
@@ -214,6 +214,7 @@ HeapType read_heap_type(Decoder* decoder, const byte* pc,
       case kEqRefCode:
       case kExternRefCode:
       case kI31RefCode:
+      case kDataRefCode:
       case kAnyRefCode: {
         HeapType result = HeapType::from_code(code);
         if (!VALIDATE(enabled.contains(feature_for_heap_type(result)))) {
@@ -279,10 +280,13 @@ ValueType read_value_type(Decoder* decoder, const byte* pc,
     case kEqRefCode:
     case kExternRefCode:
     case kI31RefCode:
+    case kDataRefCode:
     case kAnyRefCode: {
       HeapType heap_type = HeapType::from_code(code);
-      ValueType result = ValueType::Ref(
-          heap_type, code == kI31RefCode ? kNonNullable : kNullable);
+      Nullability nullability = code == kI31RefCode || code == kDataRefCode
+                                    ? kNonNullable
+                                    : kNullable;
+      ValueType result = ValueType::Ref(heap_type, nullability);
       if (!VALIDATE(enabled.contains(feature_for_heap_type(heap_type)))) {
         DecodeError<validate>(
             decoder, pc,

src/wasm/value-type.h

@@ -57,6 +57,7 @@ class HeapType {
     kExtern,                  // shorthand: e
     kEq,                      // shorthand: q
     kI31,                     // shorthand: j
+    kData,                    // shorthand: o
     kAny,                     // shorthand: a
     // This value is used to represent failures in the parsing of heap types and
     // does not correspond to a wasm heap type.
@@ -79,6 +80,8 @@ class HeapType {
         return HeapType(kI31);
       case ValueTypeCode::kAnyRefCode:
         return HeapType(kAny);
+      case ValueTypeCode::kDataRefCode:
+        return HeapType(kData);
       default:
         return HeapType(kBottom);
     }
@@ -129,6 +132,8 @@ class HeapType {
         return std::string("eq");
       case kI31:
         return std::string("i31");
+      case kData:
+        return std::string("data");
       case kAny:
         return std::string("any");
       default:
@@ -150,6 +155,8 @@ class HeapType {
         return mask | kEqRefCode;
       case kI31:
         return mask | kI31RefCode;
+      case kData:
+        return mask | kDataRefCode;
       case kAny:
         return mask | kAnyRefCode;
       default:
@@ -381,8 +388,14 @@ class ValueType {
             return kOptRefCode;
         }
       case kRef:
-        if (heap_representation() == HeapType::kI31) return kI31RefCode;
+        switch (heap_representation()) {
+          case HeapType::kI31:
+            return kI31RefCode;
+          case HeapType::kData:
+            return kDataRefCode;
+          default:
             return kRefCode;
+        }
       case kStmt:
         return kVoidCode;
       case kRtt:
@@ -403,9 +416,11 @@ class ValueType {
   // Returns true iff the heap type is needed to encode this type in the wasm
   // binary format, taking into account available type shorthands.
   constexpr bool encoding_needs_heap_type() const {
-    return (kind() == kRef && heap_representation() != HeapType::kI31) ||
-           (kind() == kOptRef && (!heap_type().is_generic() ||
-                                  heap_representation() == HeapType::kI31));
+    return (kind() == kRef && heap_representation() != HeapType::kI31 &&
+            heap_representation() != HeapType::kData) ||
+           (kind() == kOptRef && (heap_type().is_index() ||
+                                  heap_representation() == HeapType::kI31 ||
+                                  heap_representation() == HeapType::kData));
   }
 
   static constexpr int kLastUsedBit = 30;
@@ -425,19 +440,12 @@ class ValueType {
     std::ostringstream buf;
     switch (kind()) {
       case kRef:
-        if (heap_representation() == HeapType::kI31) {
-          buf << "i31ref";
-        } else {
-          buf << "(ref " << heap_type().name() << ")";
-        }
-        break;
       case kOptRef:
-        if (heap_type().is_generic() &&
-            heap_representation() != HeapType::kI31) {
-          // We use shorthands to be compatible with the 'reftypes' proposal.
-          buf << heap_type().name() << "ref";
+        if (encoding_needs_heap_type()) {
+          buf << "(ref " << (kind() == kOptRef ? "null " : "")
+              << heap_type().name() << ")";
         } else {
-          buf << "(ref null " << heap_type().name() << ")";
+          buf << heap_type().name() << "ref";
         }
         break;
       case kRttWithDepth:
@@ -511,12 +519,14 @@ constexpr ValueType kWasmI8 = ValueType::Primitive(ValueType::kI8);
 constexpr ValueType kWasmI16 = ValueType::Primitive(ValueType::kI16);
 constexpr ValueType kWasmStmt = ValueType::Primitive(ValueType::kStmt);
 constexpr ValueType kWasmBottom = ValueType::Primitive(ValueType::kBottom);
-// Established wasm shorthands:
+// Established reference-type proposal shorthands.
 constexpr ValueType kWasmFuncRef = ValueType::Ref(HeapType::kFunc, kNullable);
 constexpr ValueType kWasmExternRef =
     ValueType::Ref(HeapType::kExtern, kNullable);
 constexpr ValueType kWasmEqRef = ValueType::Ref(HeapType::kEq, kNullable);
 constexpr ValueType kWasmI31Ref = ValueType::Ref(HeapType::kI31, kNonNullable);
+constexpr ValueType kWasmDataRef =
+    ValueType::Ref(HeapType::kData, kNonNullable);
 constexpr ValueType kWasmAnyRef = ValueType::Ref(HeapType::kAny, kNullable);
 
 #define FOREACH_WASMVALUE_CTYPES(V) \

src/wasm/wasm-constants.h

@@ -40,6 +40,7 @@ enum ValueTypeCode : uint8_t {
   kI31RefCode = 0x6a,
   kRttWithDepthCode = 0x69,
   kRttCode = 0x68,
+  kDataRefCode = 0x67,
 };
 // Binary encoding of other types.
 constexpr uint8_t kWasmFunctionTypeCode = 0x60;

src/wasm/wasm-js.cc

@@ -1368,10 +1368,15 @@ void WebAssemblyGlobal(const v8::FunctionCallbackInfo<v8::Value>& args) {
           }
           break;
         }
+        case internal::wasm::HeapType::kBottom:
+          UNREACHABLE();
         case i::wasm::HeapType::kEq:
+        case internal::wasm::HeapType::kI31:
+        case internal::wasm::HeapType::kData:
         default:
           // TODO(7748): Implement these.
           UNIMPLEMENTED();
+          break;
       }
       break;
     }
@@ -1840,6 +1845,10 @@ void WebAssemblyGlobalGetValueCommon(
         case i::wasm::HeapType::kAny:
           return_value.Set(Utils::ToLocal(receiver->GetRef()));
           break;
+        case internal::wasm::HeapType::kBottom:
+          UNREACHABLE();
+        case internal::wasm::HeapType::kI31:
+        case internal::wasm::HeapType::kData:
         case i::wasm::HeapType::kEq:
         default:
           // TODO(7748): Implement these.
@@ -1932,7 +1941,10 @@ void WebAssemblyGlobalSetValue(
           }
           break;
         }
-
+        case internal::wasm::HeapType::kBottom:
+          UNREACHABLE();
+        case internal::wasm::HeapType::kI31:
+        case internal::wasm::HeapType::kData:
         case i::wasm::HeapType::kEq:
         default:
           // TODO(7748): Implement these.

src/wasm/wasm-objects.cc

@@ -452,6 +452,7 @@ void WasmTableObject::Set(Isolate* isolate, Handle<WasmTableObject> table,
       SetFunctionTableEntry(isolate, table, entries, entry_index, entry);
       return;
     case wasm::HeapType::kEq:
+    case wasm::HeapType::kData:
     case wasm::HeapType::kI31:
       // TODO(7748): Implement once we have a story for struct/arrays/i31ref in
       // JS.
@@ -499,6 +500,7 @@ Handle<Object> WasmTableObject::Get(Isolate* isolate,
       break;
     case wasm::HeapType::kEq:
     case wasm::HeapType::kI31:
+    case wasm::HeapType::kData:
     case wasm::HeapType::kAny:
       // TODO(7748): Implement once we have a story for struct/arrays/i31ref in
       // JS.
@@ -2111,7 +2113,7 @@ bool TypecheckJSObject(Isolate* isolate, const WasmModule* module,
         case HeapType::kExtern:
         case HeapType::kAny:
           return true;
-        case HeapType::kEq: {
+        case HeapType::kData: {
           // TODO(7748): Change this when we have a decision on the JS API for
           // structs/arrays.
           Handle<Name> key = isolate->factory()->wasm_wrapped_object_symbol();
@@ -2119,13 +2121,15 @@ bool TypecheckJSObject(Isolate* isolate, const WasmModule* module,
                             LookupIterator::OWN_SKIP_INTERCEPTOR);
           if (it.state() == LookupIterator::DATA) return true;
           *error_message =
-              "eqref object must be null (if nullable) or wrapped with wasm "
-              "object wrapper";
+              "dataref object must be null (if nullable) or wrapped with the "
+              "wasm object wrapper";
           return false;
         }
+        case HeapType::kEq:
         case HeapType::kI31:
           // TODO(7748): Implement when the JS API for i31ref is decided on.
-          *error_message = "Assigning JS objects to i31ref not supported yet.";
+          *error_message =
+              "Assigning JS objects to eqref/i31ref not supported yet.";
           return false;
         default:
           // Tables defined outside a module can't refer to user-defined types.

src/wasm/wasm-opcodes.cc

@@ -41,7 +41,8 @@ bool IsJSCompatibleSignature(const FunctionSig* sig, const WasmModule* module,
   for (auto type : sig->all()) {
     // TODO(7748): Allow structs, arrays, rtts and i31s when their
     //             JS-interaction is decided on.
-    if (type == kWasmS128 || type.is_reference_to(HeapType::kI31) ||
+    if (type == kWasmS128 || type.is_reference_to(HeapType::kEq) ||
+        type.is_reference_to(HeapType::kI31) ||
         (type.has_index() && !module->has_signature(type.ref_index())) ||
         type.is_rtt()) {
       return false;

src/wasm/wasm-subtyping.cc

@@ -326,7 +326,8 @@ V8_NOINLINE V8_EXPORT_PRIVATE bool IsSubtypeOfImpl(
     case HeapType::kAny:
       return super_heap == HeapType::kAny;
     case HeapType::kI31:
-      return super_heap == HeapType::kI31 || super_heap == HeapType::kEq ||
+    case HeapType::kData:
+      return super_heap == sub_heap || super_heap == HeapType::kEq ||
              super_heap == HeapType::kAny;
     case HeapType::kBottom:
       UNREACHABLE();
@@ -342,6 +343,7 @@ V8_NOINLINE V8_EXPORT_PRIVATE bool IsSubtypeOfImpl(
     case HeapType::kFunc:
       return sub_module->has_signature(sub_index);
     case HeapType::kEq:
+    case HeapType::kData:
       return !sub_module->has_signature(sub_index);
     case HeapType::kExtern:
     case HeapType::kI31:

src/wasm/wasm-subtyping.h

@@ -45,10 +45,14 @@ V8_NOINLINE bool EquivalentTypes(ValueType type1, ValueType type2,
 // - ref(ht1) <: ref/optref(ht2) iff ht1 <: ht2.
 // - rtt1 <: rtt2 iff rtt1 ~ rtt2.
 // For heap types, the following subtyping rules hold:
-// - Each generic heap type is a subtype of itself.
-// - All heap types are subtypes of any.
+// - The abstract heap types form the following type hierarchy:
+//           any
+//         /  |  \
+//       eq func  extern
+//      / \
+//   i31   data
+// - All structs and arrays are subtypes of data.
 // - All functions are subtypes of func.
-// - i31, structs and arrays are subtypes of eq.
 // - Struct subtyping: Subtype must have at least as many fields as supertype,
 //   covariance for immutable fields, equivalence for mutable fields.
 // - Array subtyping (mutable only) is the equivalence relation.

test/cctest/wasm/test-gc.cc

@@ -1251,13 +1251,12 @@ WASM_COMPILED_EXEC_TEST(IndirectNullSetManually) {
 }
 
 TEST(JsAccess) {
-  for (ValueType supertype : {kWasmEqRef, kWasmAnyRef}) {
   WasmGCTester tester;
   const byte type_index = tester.DefineStruct({F(wasm::kWasmI32, true)});
   ValueType kRefType = ref(type_index);
-    ValueType kSupertypeToI[] = {kWasmI32, supertype};
+  ValueType kSupertypeToI[] = {kWasmI32, kWasmDataRef};
   FunctionSig sig_t_v(1, 0, &kRefType);
-    FunctionSig sig_super_v(1, 0, &supertype);
+  FunctionSig sig_super_v(1, 0, &kWasmDataRef);
   FunctionSig sig_i_super(1, 1, kSupertypeToI);
 
   tester.DefineExportedFunction(
@@ -1316,7 +1315,6 @@ TEST(JsAccess) {
   CHECK(try_catch.HasCaught());
   try_catch.Reset();
   isolate->clear_pending_exception();
-  }
 }
 
 }  // namespace test_gc

test/common/wasm/wasm-interpreter.cc

@@ -3026,7 +3026,11 @@ class WasmInterpreterInternals {
               encoded_values->set(encoded_index++, *externref);
               break;
             }
+            case HeapType::kBottom:
+              UNREACHABLE();
             case HeapType::kEq:
+            case HeapType::kData:
+            case HeapType::kI31:
             default:
               // TODO(7748): Implement these.
               UNIMPLEMENTED();

test/unittests/wasm/subtyping-unittest.cc

@@ -57,9 +57,9 @@ TEST_F(WasmSubtypingTest, Subtyping) {
 
   ValueType numeric_types[] = {kWasmI32, kWasmI64, kWasmF32, kWasmF64,
                                kWasmS128};
-  ValueType ref_types[] = {kWasmExternRef, kWasmFuncRef, kWasmEqRef,
-                           kWasmI31Ref,    kWasmAnyRef,  optRef(0),
-                           ref(0),         optRef(2),    ref(2)};
+  ValueType ref_types[] = {
+      kWasmExternRef, kWasmFuncRef, kWasmEqRef, kWasmI31Ref, kWasmDataRef,
+      kWasmAnyRef,    optRef(0),    ref(0),     optRef(2),   ref(2)};
 
   // Type judgements across modules should work the same as within one module.
   for (WasmModule* module : {module1, module2}) {
@@ -80,11 +80,16 @@ TEST_F(WasmSubtypingTest, Subtyping) {
     }
 
     for (ValueType ref_type : ref_types) {
-      // Concrete reference types and i31ref are subtypes of eqref,
+      // Concrete reference types, i31ref and dataref are subtypes of eqref,
       // externref/funcref/anyref are not.
       CHECK_EQ(IsSubtypeOf(ref_type, kWasmEqRef, module1, module),
                ref_type != kWasmFuncRef && ref_type != kWasmExternRef &&
                    ref_type != kWasmAnyRef);
+      // Non-nullable struct/array types are subtypes of dataref.
+      CHECK_EQ(
+          IsSubtypeOf(ref_type, kWasmDataRef, module1, module),
+          ref_type == kWasmDataRef ||
+              (ref_type.kind() == ValueType::kRef && ref_type.has_index()));
       // Each reference type is a subtype of itself.
       CHECK(IsSubtypeOf(ref_type, ref_type, module1, module));
       // Each reference type is a subtype of anyref.