[wasm] Prototype I32AtomicWake implementation

Bug=v8:8075
R=adamk@chromium.org

Change-Id: Ib7b18448d59e1f54f664e24908fbd3a8b4607a9f
Reviewed-on: https://chromium-review.googlesource.com/c/1332153
Commit-Queue: Aseem Garg <aseemgarg@chromium.org>
Reviewed-by: Ben Smith <binji@chromium.org>
Cr-Commit-Position: refs/heads/master@{#57623}

src/builtins/builtins-definitions.h

@@ -1231,6 +1231,7 @@ namespace internal {
   /* Wasm */                                                                   \
   ASM(WasmCompileLazy)                                                         \
   TFC(WasmAllocateHeapNumber, AllocateHeapNumber, 1)                           \
+  TFC(WasmAtomicWake, WasmAtomicWake, 1)                                       \
   TFC(WasmCallJavaScript, CallTrampoline, 1)                                   \
   TFC(WasmMemoryGrow, WasmMemoryGrow, 1)                                       \
   TFC(WasmRecordWrite, RecordWrite, 1)                                         \
@@ -1529,6 +1530,7 @@ namespace internal {
 #define WASM_RUNTIME_STUB_LIST(V, VTRAP) \
   FOREACH_WASM_TRAPREASON(VTRAP)         \
   V(WasmAllocateHeapNumber)              \
+  V(WasmAtomicWake)                      \
   V(WasmCallJavaScript)                  \
   V(WasmMemoryGrow)                      \
   V(WasmRecordWrite)                     \

src/builtins/builtins-wasm-gen.cc

@@ -98,6 +98,31 @@ TF_BUILTIN(WasmThrow, WasmBuiltinsAssembler) {
   TailCallRuntimeWithCEntry(Runtime::kThrow, centry, context, exception);
 }
 
+TF_BUILTIN(WasmAtomicWake, WasmBuiltinsAssembler) {
+  TNode<Uint32T> address =
+      UncheckedCast<Uint32T>(Parameter(Descriptor::kAddress));
+  TNode<Uint32T> count = UncheckedCast<Uint32T>(Parameter(Descriptor::kCount));
+
+  TNode<Object> instance = LoadInstanceFromFrame();
+  TNode<Code> centry = LoadCEntryFromInstance(instance);
+
+  TNode<Code> target = LoadBuiltinFromFrame(Builtins::kAllocateHeapNumber);
+
+  // TODO(aseemgarg): Use SMIs if possible for address and count
+  TNode<HeapNumber> address_heap = UncheckedCast<HeapNumber>(
+      CallStub(AllocateHeapNumberDescriptor(), target, NoContextConstant()));
+  StoreHeapNumberValue(address_heap, ChangeUint32ToFloat64(address));
+
+  TNode<HeapNumber> count_heap = UncheckedCast<HeapNumber>(
+      CallStub(AllocateHeapNumberDescriptor(), target, NoContextConstant()));
+  StoreHeapNumberValue(count_heap, ChangeUint32ToFloat64(count));
+
+  TNode<Smi> result_smi = UncheckedCast<Smi>(CallRuntimeWithCEntry(
+      Runtime::kWasmAtomicWake, centry, NoContextConstant(), instance,
+      address_heap, count_heap));
+  ReturnRaw(SmiToInt32(result_smi));
+}
+
 TF_BUILTIN(WasmMemoryGrow, WasmBuiltinsAssembler) {
   TNode<Int32T> num_pages =
       UncheckedCast<Int32T>(Parameter(Descriptor::kNumPages));

src/compiler/wasm-compiler.cc

@@ -4065,6 +4065,27 @@ Node* WasmGraphBuilder::AtomicOp(wasm::WasmOpcode opcode, Node* const* inputs,
   }
     ATOMIC_STORE_LIST(BUILD_ATOMIC_STORE_OP)
 #undef BUILD_ATOMIC_STORE_OP
+    case wasm::kExprAtomicWake: {
+      Node* index = CheckBoundsAndAlignment(
+          wasm::ValueTypes::MemSize(MachineType::Uint32()), inputs[0], offset,
+          position);
+      // Now that we've bounds-checked, compute the effective address.
+      Node* address = graph()->NewNode(mcgraph()->machine()->Int32Add(),
+                                       Uint32Constant(offset), index);
+      WasmAtomicWakeDescriptor interface_descriptor;
+      auto call_descriptor = Linkage::GetStubCallDescriptor(
+          mcgraph()->zone(), interface_descriptor,
+          interface_descriptor.GetStackParameterCount(),
+          CallDescriptor::kNoFlags, Operator::kNoProperties,
+          StubCallMode::kCallWasmRuntimeStub);
+      Node* call_target = mcgraph()->RelocatableIntPtrConstant(
+          wasm::WasmCode::kWasmAtomicWake, RelocInfo::WASM_STUB_CALL);
+      node = graph()->NewNode(mcgraph()->common()->Call(call_descriptor),
+                              call_target, address, inputs[1], Effect(),
+                              Control());
+      break;
+    }
+
     default:
       FATAL_UNSUPPORTED_OPCODE(opcode);
   }

src/interface-descriptors.cc

@@ -356,6 +356,11 @@ void WasmThrowDescriptor::InitializePlatformSpecific(
   DefaultInitializePlatformSpecific(data, kParameterCount);
 }
 
+void WasmAtomicWakeDescriptor::InitializePlatformSpecific(
+    CallInterfaceDescriptorData* data) {
+  DefaultInitializePlatformSpecific(data, kParameterCount);
+}
+
 void CloneObjectWithVectorDescriptor::InitializePlatformSpecific(
     CallInterfaceDescriptorData* data) {
   DefaultInitializePlatformSpecific(data, kParameterCount);

src/interface-descriptors.h

@@ -76,6 +76,7 @@ namespace internal {
   V(RunMicrotasks)                    \
   V(WasmMemoryGrow)                   \
   V(WasmThrow)                        \
+  V(WasmAtomicWake)                   \
   V(CloneObjectWithVector)            \
   BUILTIN_LIST_TFS(V)
 
@@ -1104,6 +1105,15 @@ class WasmThrowDescriptor final : public CallInterfaceDescriptor {
   DECLARE_DESCRIPTOR(WasmThrowDescriptor, CallInterfaceDescriptor)
 };
 
+class WasmAtomicWakeDescriptor final : public CallInterfaceDescriptor {
+ public:
+  DEFINE_PARAMETERS_NO_CONTEXT(kAddress, kCount)
+  DEFINE_RESULT_AND_PARAMETER_TYPES(MachineType::Uint32(),  // result 1
+                                    MachineType::Uint32(),  // kAddress
+                                    MachineType::Uint32())  // kCount
+  DECLARE_DESCRIPTOR(WasmAtomicWakeDescriptor, CallInterfaceDescriptor)
+};
+
 class CloneObjectWithVectorDescriptor final : public CallInterfaceDescriptor {
  public:
   DEFINE_PARAMETERS(kSource, kFlags, kSlot, kVector)

src/runtime/runtime-wasm.cc

@@ -26,7 +26,7 @@ namespace internal {
 
 namespace {
 
-Context* GetNativeContextFromWasmInstanceOnStackTop(Isolate* isolate) {
+WasmInstanceObject* GetWasmInstanceOnStackTop(Isolate* isolate) {
   StackFrameIterator it(isolate, isolate->thread_local_top());
   // On top: C entry stub.
   DCHECK_EQ(StackFrame::EXIT, it.frame()->type());
@@ -34,7 +34,11 @@ Context* GetNativeContextFromWasmInstanceOnStackTop(Isolate* isolate) {
   // Next: the wasm compiled frame.
   DCHECK(it.frame()->is_wasm_compiled());
   WasmCompiledFrame* frame = WasmCompiledFrame::cast(it.frame());
-  return frame->wasm_instance()->native_context();
+  return frame->wasm_instance();
+}
+
+Context* GetNativeContextFromWasmInstanceOnStackTop(Isolate* isolate) {
+  return GetWasmInstanceOnStackTop(isolate)->native_context();
 }
 
 class ClearThreadInWasmScope {
@@ -253,5 +257,25 @@ RUNTIME_FUNCTION(Runtime_WasmCompileLazy) {
   return reinterpret_cast<Object*>(entrypoint);
 }
 
+RUNTIME_FUNCTION(Runtime_WasmAtomicWake) {
+  HandleScope scope(isolate);
+  DCHECK_EQ(3, args.length());
+  CONVERT_ARG_HANDLE_CHECKED(WasmInstanceObject, instance, 0);
+  CONVERT_NUMBER_CHECKED(uint32_t, address, Uint32, args[1]);
+  CONVERT_NUMBER_CHECKED(uint32_t, count, Uint32, args[2]);
+
+  DCHECK(instance->has_memory_object());
+  Handle<JSArrayBuffer> array_buffer(instance->memory_object()->array_buffer(),
+                                     isolate);
+
+  // Validation should have failed if the memory was not shared.
+  DCHECK(array_buffer->is_shared());
+
+  // Should have trapped if address was OOB
+  DCHECK_LT(address, array_buffer->byte_length());
+
+  return FutexEmulation::Wake(array_buffer, address, count);
+}
+
 }  // namespace internal
 }  // namespace v8

src/runtime/runtime.h

@@ -531,6 +531,7 @@ namespace internal {
 #define FOR_EACH_INTRINSIC_WASM(F, I) \
   F(ThrowWasmError, 1, 1)             \
   F(ThrowWasmStackOverflow, 0, 1)     \
+  F(WasmAtomicWake, 3, 1)             \
   F(WasmExceptionGetValues, 1, 1)     \
   F(WasmExceptionGetTag, 1, 1)        \
   F(WasmMemoryGrow, 2, 1)             \

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

@@ -61,6 +61,7 @@ struct WasmException;
                 (message)))
 
 #define ATOMIC_OP_LIST(V)                \
+  V(AtomicWake, Uint32)                  \
   V(I32AtomicLoad, Uint32)               \
   V(I64AtomicLoad, Uint64)               \
   V(I32AtomicLoad8U, Uint8)              \

src/wasm/wasm-opcodes.cc

@@ -263,6 +263,7 @@ const char* WasmOpcodes::OpcodeName(WasmOpcode opcode) {
     CASE_S1x16_OP(AllTrue, "all_true")
 
     // Atomic operations.
+    CASE_OP(AtomicWake, "atomic_wake")
     CASE_UNSIGNED_ALL_OP(AtomicLoad, "atomic_load")
     CASE_UNSIGNED_ALL_OP(AtomicStore, "atomic_store")
     CASE_UNSIGNED_ALL_OP(AtomicAdd, "atomic_add")

src/wasm/wasm-opcodes.h

@@ -418,6 +418,7 @@ bool IsJSCompatibleSignature(const FunctionSig* sig);
   V(TableCopy, 0xfc0e, v_iii)
 
 #define FOREACH_ATOMIC_OPCODE(V)                \
+  V(AtomicWake, 0xfe00, i_ii)                   \
   V(I32AtomicLoad, 0xfe10, i_i)                 \
   V(I64AtomicLoad, 0xfe11, l_i)                 \
   V(I32AtomicLoad8U, 0xfe12, i_i)               \

test/mjsunit/wasm/futex.js

@@ -0,0 +1,141 @@
+// Copyright 2018 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// Flags: --allow-natives-syntax --harmony-sharedarraybuffer
+// Flags: --experimental-wasm-threads
+
+'use strict';
+
+load("test/mjsunit/wasm/wasm-constants.js");
+load("test/mjsunit/wasm/wasm-module-builder.js");
+
+function WasmAtomicWakeFunction(memory, offset, index, num) {
+  let builder = new WasmModuleBuilder();
+  builder.addImportedMemory("m", "memory", 0, 20, "shared");
+  builder.addFunction("main", kSig_i_ii)
+    .addBody([
+      kExprGetLocal, 0,
+      kExprGetLocal, 1,
+      kAtomicPrefix,
+      kExprAtomicWake, /* alignment */ 0, offset])
+    .exportAs("main");
+
+  // Instantiate module, get function exports
+  let module = new WebAssembly.Module(builder.toBuffer());
+  let instance = new WebAssembly.Instance(module, {m: {memory}});
+  return instance.exports.main(index, num);
+}
+
+(function TestInvalidIndex() {
+  let memory = new WebAssembly.Memory({initial: 1, maximum: 1, shared: true});
+
+  // Valid indexes are 0-65535 (1 page).
+  [-2, 65536, 0xffffffff].forEach(function(invalidIndex) {
+    assertThrows(function() {
+      WasmAtomicWakeFunction(memory, 0, invalidIndex, -1);
+    }, Error);
+    assertThrows(function() {
+      WasmAtomicWakeFunction(memory, invalidIndex, 0, -1);
+    }, Error);
+    assertThrows(function() {
+      WasmAtomicWakeFunction(memory, invalidIndex/2, invalidIndex/2, -1);
+    }, Error);
+  });
+})();
+
+(function TestWakeCounts() {
+  let memory = new WebAssembly.Memory({initial: 1, maximum: 1, shared: true});
+
+  [-1, 0, 4, 100, 0xffffffff].forEach(function(count) {
+    WasmAtomicWakeFunction(memory, 0, 0, count);
+  });
+})();
+
+//// WORKER ONLY TESTS
+
+if (this.Worker) {
+  let wasm_wake_adapter = (memory, offset, index, count) => {
+    return WasmAtomicWakeFunction(memory, offset, index, count);
+  };
+
+  // Wait adapter string that can be passed as a parameter to TestWaitWake to generate
+  // custom worker script
+  let js_wait_adapter = `(memory, offset, index, val) => {
+    let i32a = new Int32Array(memory.buffer, offset);
+    let res = Atomics.wait(i32a, index/4, val);
+    if (res == "ok") return 0;
+    if (res == "not-equal") return 1;
+    return 2;
+  }`
+
+  let TestWaitWake = function(wait_adapter, wake_adapter, num_workers, num_workers_wake) {
+    let memory = new WebAssembly.Memory({initial: 1, maximum: 1, shared: true});
+    let i32a = new Int32Array(memory.buffer);
+    // SAB values:
+    // memory[id*4], where id in range [0, num_workers]:
+    //   0 => Worker |id| is still waiting on the futex
+    //   1 => Worker |id| is not waiting on futex, but has not be reaped by the
+    //        main thread.
+    //   2 => Worker |id| has been reaped.
+    //
+    // memory[num_workers*4]:
+    //   always 0. Each worker is waiting on this index.
+
+    let workerScript =
+      `onmessage = function(msg) {
+         let id = msg.id;
+         let memory = msg.memory;
+         let i32a = new Int32Array(memory.buffer);
+         let wait_adapter = eval(msg.wait_adapter);
+         let result = wait_adapter(memory, 0, 4*${num_workers}, 0);
+         // Set i32a[id] to 1 to notify the main thread which workers were
+         // woken up.
+         Atomics.store(i32a, id, 1);
+         postMessage(result);
+       };`;
+
+    let workers = [];
+    for (let id = 0; id < num_workers; id++) {
+      workers[id] = new Worker(workerScript, {type: 'string'});
+      workers[id].postMessage({id, memory, wait_adapter});
+    }
+
+    // Spin until all workers are waiting on the futex.
+    while (%AtomicsNumWaitersForTesting(i32a, num_workers) != num_workers) {}
+
+    if (num_workers_wake < num_workers) {
+      assertEquals(num_workers_wake, wake_adapter(memory, 0, 4*num_workers, num_workers_wake));
+    } else {
+      assertEquals(num_workers, wake_adapter(memory, 0, 4*num_workers, num_workers_wake));
+      num_workers_wake = num_workers;
+    }
+
+    let wokenCount = 0;
+    while (wokenCount < num_workers_wake) {
+      for (let id = 0; id < num_workers; id++) {
+        // Look for workers that have not yet been reaped. Set i32a[id] to 2
+        // when they've been processed so we don't look at them again.
+        if (Atomics.compareExchange(i32a, id, 1, 2) == 1) {
+          assertEquals(0, workers[id].getMessage());
+          workers[id].terminate();
+          wokenCount++;
+        }
+      }
+    }
+
+    assertEquals(num_workers - num_workers_wake,
+                 %AtomicsNumWaitersForTesting(i32a, num_workers));
+
+    // Finally wake and kill all workers.
+    wake_adapter(memory, 0, 4*num_workers, num_workers_wake)
+    for (let id = 0; id < num_workers; id++) {
+      workers[id].terminate();
+    }
+  };
+
+  TestWaitWake(js_wait_adapter, wasm_wake_adapter, 1, 1);
+  TestWaitWake(js_wait_adapter, wasm_wake_adapter, 4, 4);
+  TestWaitWake(js_wait_adapter, wasm_wake_adapter, 4, 3);
+  TestWaitWake(js_wait_adapter, wasm_wake_adapter, 3, 4);
+}

test/mjsunit/wasm/wasm-constants.js

@@ -354,6 +354,7 @@ let kExprF64ReinterpretI64 = 0xbf;
 // Prefix opcodes
 let kAtomicPrefix = 0xfe;
 
+let kExprAtomicWake = 0x00;
 let kExprI32AtomicLoad = 0x10;
 let kExprI32AtomicLoad8U = 0x12;
 let kExprI32AtomicLoad16U = 0x13;