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

test/cctest/heap/test-heap.cc

 // Copyright 2012 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
@@ skipping 2210 matching lines @@
   Address top = *top_addr;
   // Now force the remaining allocation onto the free list.
   CcTest::heap()->old_space()->EmptyAllocationInfo();
   return top;
 }
 
 
 // Test the case where allocation must be done from the free list, so filler
 // may precede or follow the object.
 TEST(TestAlignedOverAllocation) {
+  Heap* heap = CcTest::heap();
+  // Test checks for fillers before and behind objects and requires a fresh
+  // page and empty free list.
+  heap::AbandonCurrentlyFreeMemory(heap->old_space());
+  // Allocate a dummy object to properly set up the linear allocation info.
+  AllocationResult dummy =
+      heap->old_space()->AllocateRawUnaligned(kPointerSize);
+  CHECK(!dummy.IsRetry());
+  heap->CreateFillerObjectAt(
+      HeapObject::cast(dummy.ToObjectChecked())->address(), kPointerSize,
+      ClearRecordedSlots::kNo);
+
   // Double misalignment is 4 on 32-bit platforms, 0 on 64-bit ones.
   const intptr_t double_misalignment = kDoubleSize - kPointerSize;
   Address start;
   HeapObject* obj;
   HeapObject* filler1;
   HeapObject* filler2;
   if (double_misalignment) {
     start = AlignOldSpace(kDoubleAligned, 0);
     obj = OldSpaceAllocateAligned(kPointerSize, kDoubleAligned);
     // The object is aligned, and a filler object is created after.
@@ skipping 121 matching lines @@
   intptr_t available = new_space->Capacity() - new_space->Size();
   intptr_t number_of_fillers = (available / FixedArray::SizeFor(32)) - 1;
   for (intptr_t i = 0; i < number_of_fillers; i++) {
     CHECK(heap->InNewSpace(*factory->NewFixedArray(32, NOT_TENURED)));
   }
 }
 
 
 TEST(GrowAndShrinkNewSpace) {
   CcTest::InitializeVM();
+  // Avoid shrinking new space in GC epilogue. This can happen if allocation
+  // throughput samples have been taken while executing the benchmark.
+  i::FLAG_predictable = true;

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

Why?
And to avoid shrinking, just set the min_semi_space_size.

[Michael Lippautz, 2016/08/11 09:48:12]

This test checks shrinking. With predictable I merely disabled the shrinking
during the GC. This is already a flake on ToT.

   Heap* heap = CcTest::heap();
   NewSpace* new_space = heap->new_space();
 
   if (heap->MaxSemiSpaceSize() == heap->InitialSemiSpaceSize()) {
     return;
   }
 
   // Explicitly growing should double the space capacity.
   intptr_t old_capacity, new_capacity;
   old_capacity = new_space->TotalCapacity();
@@ skipping 1222 matching lines @@
   // memory is distributed. Since this is non-deterministic because of
   // concurrent sweeping, we disable it for this test.
   i::FLAG_parallel_compaction = false;
   // Concurrent sweeping adds non determinism, depending on when memory is
   // available for further reuse.
   i::FLAG_concurrent_sweeping = false;
   // Fast evacuation of pages may result in a different page count in old space.
   i::FLAG_page_promotion = false;
   CcTest::InitializeVM();
   Isolate* isolate = CcTest::i_isolate();
+  // If there's snapshot available, we don't know whether 20 small arrays will
+  // fit on the initial pages.
+  if (!isolate->snapshot_available()) return;
   Factory* factory = isolate->factory();
   Heap* heap = isolate->heap();
+
   v8::HandleScope scope(CcTest::isolate());
   static const int number_of_test_pages = 20;
 
   // Prepare many pages with low live-bytes count.
   PagedSpace* old_space = heap->old_space();
   const int initial_page_count = old_space->CountTotalPages();
   const int overall_page_count = number_of_test_pages + initial_page_count;
   for (int i = 0; i < number_of_test_pages; i++) {
     AlwaysAllocateScope always_allocate(isolate);
     heap::SimulateFullSpace(old_space);
     factory->NewFixedArray(1, TENURED);
   }
   CHECK_EQ(overall_page_count, old_space->CountTotalPages());
 
   // Triggering one GC will cause a lot of garbage to be discovered but
   // even spread across all allocated pages.
   heap->CollectAllGarbage(Heap::kFinalizeIncrementalMarkingMask,
                           "triggered for preparation");
   CHECK_GE(overall_page_count, old_space->CountTotalPages());
 
   // Triggering subsequent GCs should cause at least half of the pages
   // to be released to the OS after at most two cycles.
   heap->CollectAllGarbage(Heap::kFinalizeIncrementalMarkingMask,
                           "triggered by test 1");
   CHECK_GE(overall_page_count, old_space->CountTotalPages());
   heap->CollectAllGarbage(Heap::kFinalizeIncrementalMarkingMask,
                           "triggered by test 2");
   CHECK_GE(overall_page_count, old_space->CountTotalPages() * 2);
 
-  // Triggering a last-resort GC should cause all pages to be released to the
-  // OS so that other processes can seize the memory.  If we get a failure here
-  // where there are 2 pages left instead of 1, then we should increase the
-  // size of the first page a little in SizeOfFirstPage in spaces.cc.  The
-  // first page should be small in order to reduce memory used when the VM
-  // boots, but if the 20 small arrays don't fit on the first page then that's
-  // an indication that it is too small.
   heap->CollectAllAvailableGarbage("triggered really hard");
+  // Triggering a last-resort GC should release all additional pages.
   CHECK_EQ(initial_page_count, old_space->CountTotalPages());
 }
 
 static int forced_gc_counter = 0;
 
 void MockUseCounterCallback(v8::Isolate* isolate,
                             v8::Isolate::UseCounterFeature feature) {
   isolate->GetCurrentContext();
   if (feature == v8::Isolate::kForcedGC) {
     forced_gc_counter++;
@@ skipping 3422 matching lines @@
       chunk, chunk->area_end() - kPointerSize, chunk->area_end());
   slots[chunk->area_end() - kPointerSize] = false;
   RememberedSet<OLD_TO_NEW>::Iterate(chunk, [&slots](Address addr) {
     CHECK(slots[addr]);
     return KEEP_SLOT;
   });
 }
 
 }  // namespace internal
 }  // namespace v8