Reland of "[heap] Switch to 500k pages"

src/objects.h

 // Copyright 2015 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.
 
 #ifndef V8_OBJECTS_H_
 #define V8_OBJECTS_H_
 
 #include <iosfwd>
 #include <memory>
 
@@ skipping 4694 matching lines @@
   inline static FreeSpace* cast(HeapObject* obj);
 
   // Dispatched behavior.
   DECLARE_PRINTER(FreeSpace)
   DECLARE_VERIFIER(FreeSpace)
 
   // Layout description.
   // Size is smi tagged when it is stored.
   static const int kSizeOffset = HeapObject::kHeaderSize;
   static const int kNextOffset = POINTER_SIZE_ALIGN(kSizeOffset + kPointerSize);
+  static const int kSize = kNextOffset + kPointerSize;
 
  private:
   DISALLOW_IMPLICIT_CONSTRUCTORS(FreeSpace);
 };
 
 
 // V has parameters (Type, type, TYPE, C type, element_size)
 #define TYPED_ARRAYS(V) \
   V(Uint8, uint8, UINT8, uint8_t, 1)                                           \
   V(Int8, int8, INT8, int8_t, 1)                                               \
@@ skipping 5734 matching lines @@
   DECLARE_PRINTER(JSArray)
   DECLARE_VERIFIER(JSArray)
 
   // Number of element slots to pre-allocate for an empty array.
   static const int kPreallocatedArrayElements = 4;
 
   // Layout description.
   static const int kLengthOffset = JSObject::kHeaderSize;
   static const int kSize = kLengthOffset + kPointerSize;
 
-  // 600 * KB is the Page::kMaxRegularHeapObjectSize defined in spaces.h which
+  // 400 * KB is the Page::kMaxRegularHeapObjectSize defined in spaces.h which

[Hannes Payer (out of office), 2016/08/10 21:17:35]

Why not 300?

[Michael Lippautz, 2016/08/11 09:48:11]

Crankshaft (and esp hydrogen stubs) don't check for kMaxRegularHeapObjectSize at
all places. The problem is more involved, as the stubs often require objects to
be allocated in new space to avoid write barriers.

Since those stubs are already deprecated and in the process of being move to
turbo fan we just mitigate it by staying at 400 (which will not trigger a case).
This is merely a problem with our verification, as allocating a large object in
new space works like expected (except that it should not go there ;)).

Turbofan already suffers from similar issues: tracking bug: chromium:636391

   // we do not want to include in objects.h
   // Note that Page::kMaxRegularHeapObjectSize has to be in sync with
   // kInitialMaxFastElementArray which is checked in a DCHECK in heap.cc.
   static const int kInitialMaxFastElementArray =
-      (600 * KB - FixedArray::kHeaderSize - kSize - AllocationMemento::kSize) /
+      (400 * KB - FixedArray::kHeaderSize - kSize - AllocationMemento::kSize) /
       kPointerSize;
 
  private:
   DISALLOW_IMPLICIT_CONSTRUCTORS(JSArray);
 };
 
 
 Handle<Object> CacheInitialJSArrayMaps(Handle<Context> native_context,
                                        Handle<Map> initial_map);
 
@@ skipping 588 matching lines @@
     }
     return value;
   }
 };
 
 
 }  // NOLINT, false-positive due to second-order macros.
 }  // NOLINT, false-positive due to second-order macros.
 
 #endif  // V8_OBJECTS_H_