8251330: Reorder CDS archived heap to speed up relocation

Reviewed-by: iklam, ccheung

Author: Matias Saavedra Silva

src/hotspot/share/cds/archiveHeapLoader.cpp

@@ -164,18 +164,19 @@ void ArchiveHeapLoader::patch_compressed_embedded_pointers(BitMapView bm,
 
   // Optimization: if dumptime shift is the same as runtime shift, we can perform a
   // quick conversion from "dumptime narrowOop" -> "runtime narrowOop".
+  narrowOop* patching_start = (narrowOop*)region.start() + FileMapInfo::current_info()->heap_oopmap_start_pos();
   if (_narrow_oop_shift == CompressedOops::shift()) {
     uint32_t quick_delta = (uint32_t)rt_encoded_bottom - (uint32_t)dt_encoded_bottom;
     log_info(cds)("CDS heap data relocation quick delta = 0x%x", quick_delta);
     if (quick_delta == 0) {
       log_info(cds)("CDS heap data relocation unnecessary, quick_delta = 0");
     } else {
-      PatchCompressedEmbeddedPointersQuick patcher((narrowOop*)region.start(), quick_delta);
+      PatchCompressedEmbeddedPointersQuick patcher(patching_start, quick_delta);
       bm.iterate(&patcher);
     }
   } else {
     log_info(cds)("CDS heap data quick relocation not possible");
-    PatchCompressedEmbeddedPointers patcher((narrowOop*)region.start());
+    PatchCompressedEmbeddedPointers patcher(patching_start);
     bm.iterate(&patcher);
   }
 }
@@ -186,17 +187,10 @@ void ArchiveHeapLoader::patch_embedded_pointers(FileMapInfo* info,
                                                 MemRegion region, address oopmap,
                                                 size_t oopmap_size_in_bits) {
   BitMapView bm((BitMap::bm_word_t*)oopmap, oopmap_size_in_bits);
-
-#ifndef PRODUCT
-  ResourceMark rm;
-  ResourceBitMap checkBm = HeapShared::calculate_oopmap(region);
-  assert(bm.is_same(checkBm), "sanity");
-#endif
-
   if (UseCompressedOops) {
     patch_compressed_embedded_pointers(bm, info, region);
   } else {
-    PatchUncompressedEmbeddedPointers patcher((oop*)region.start());
+    PatchUncompressedEmbeddedPointers patcher((oop*)region.start() + FileMapInfo::current_info()->heap_oopmap_start_pos());
     bm.iterate(&patcher);
   }
 }
@@ -316,7 +310,7 @@ bool ArchiveHeapLoader::load_heap_region_impl(FileMapInfo* mapinfo, LoadedArchiv
   uintptr_t oopmap = bitmap_base + r->oopmap_offset();
   BitMapView bm((BitMap::bm_word_t*)oopmap, r->oopmap_size_in_bits());
 
-  PatchLoadedRegionPointers patcher((narrowOop*)load_address, loaded_region);
+  PatchLoadedRegionPointers patcher((narrowOop*)load_address + FileMapInfo::current_info()->heap_oopmap_start_pos(), loaded_region);
   bm.iterate(&patcher);
   return true;
 }
@@ -449,7 +443,7 @@ void ArchiveHeapLoader::patch_native_pointers() {
   if (r->mapped_base() != nullptr && r->has_ptrmap()) {
     log_info(cds, heap)("Patching native pointers in heap region");
     BitMapView bm = r->ptrmap_view();
-    PatchNativePointers patcher((Metadata**)r->mapped_base());
+    PatchNativePointers patcher((Metadata**)r->mapped_base() + FileMapInfo::current_info()->heap_ptrmap_start_pos());
     bm.iterate(&patcher);
   }
 }

src/hotspot/share/cds/archiveHeapWriter.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2024, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -62,6 +62,7 @@ address ArchiveHeapWriter::_requested_top;
 
 GrowableArrayCHeap<ArchiveHeapWriter::NativePointerInfo, mtClassShared>* ArchiveHeapWriter::_native_pointers;
 GrowableArrayCHeap<oop, mtClassShared>* ArchiveHeapWriter::_source_objs;
+GrowableArrayCHeap<int, mtClassShared>* ArchiveHeapWriter::_source_objs_order;
 
 ArchiveHeapWriter::BufferOffsetToSourceObjectTable*
   ArchiveHeapWriter::_buffer_offset_to_source_obj_table = nullptr;
@@ -72,6 +73,7 @@ typedef ResourceHashtable<address, size_t,
       AnyObj::C_HEAP,
       mtClassShared> FillersTable;
 static FillersTable* _fillers;
+static int _num_native_ptrs = 0;
 
 void ArchiveHeapWriter::init() {
   if (HeapShared::can_write()) {
@@ -84,13 +86,15 @@ void ArchiveHeapWriter::init() {
 
     _native_pointers = new GrowableArrayCHeap<NativePointerInfo, mtClassShared>(2048);
     _source_objs = new GrowableArrayCHeap<oop, mtClassShared>(10000);
+    _source_objs_order = new GrowableArrayCHeap<int, mtClassShared>(10000);
 
     guarantee(UseG1GC, "implementation limitation");
     guarantee(MIN_GC_REGION_ALIGNMENT <= /*G1*/HeapRegion::min_region_size_in_words() * HeapWordSize, "must be");
   }
 }
 
 void ArchiveHeapWriter::add_source_obj(oop src_obj) {
+  _source_objs_order->append(_source_objs->length());
   _source_objs->append(src_obj);
 }
 
@@ -226,9 +230,54 @@ void ArchiveHeapWriter::copy_roots_to_buffer(GrowableArrayCHeap<oop, mtClassShar
   _buffer_used = new_used;
 }
 
+static int oop_sorting_rank(oop o) {
+  bool has_o_ptr = HeapShared::has_oop_pointers(o);
+  bool has_n_ptr = HeapShared::has_native_pointers(o);
+
+  if (!has_o_ptr) {
+    if (!has_n_ptr) {
+      return 0;
+    } else {
+      return 1;
+    }
+  } else {
+    if (has_n_ptr) {
+      return 2;
+    } else {
+      return 3;
+    }
+  }
+}
+
+// The goal is to sort the objects in increasing order of:
+// - objects that have no pointers
+// - objects that have only native pointers
+// - objects that have both native and oop pointers
+// - objects that have only oop pointers
+int ArchiveHeapWriter::compare_objs_by_oop_fields(int* a, int* b) {
+  oop oa = _source_objs->at(*a);
+  oop ob = _source_objs->at(*b);
+
+  int rank_a = oop_sorting_rank(oa);
+  int rank_b = oop_sorting_rank(ob);
+
+  if (rank_a != rank_b) {
+    return rank_a - rank_b;
+  } else {
+    // If they are the same rank, sort them by their position in the _source_objs array
+    return *a - *b;
+  }
+}
+
+void ArchiveHeapWriter::sort_source_objs() {
+  _source_objs_order->sort(compare_objs_by_oop_fields);
+}
+
 void ArchiveHeapWriter::copy_source_objs_to_buffer(GrowableArrayCHeap<oop, mtClassShared>* roots) {
-  for (int i = 0; i < _source_objs->length(); i++) {
-    oop src_obj = _source_objs->at(i);
+  sort_source_objs();
+  for (int i = 0; i < _source_objs_order->length(); i++) {
+    int src_obj_index = _source_objs_order->at(i);
+    oop src_obj = _source_objs->at(src_obj_index);
     HeapShared::CachedOopInfo* info = HeapShared::archived_object_cache()->get(src_obj);
     assert(info != nullptr, "must be");
     size_t buffer_offset = copy_one_source_obj_to_buffer(src_obj);
@@ -239,8 +288,8 @@ void ArchiveHeapWriter::copy_source_objs_to_buffer(GrowableArrayCHeap<oop, mtCla
 
   copy_roots_to_buffer(roots);
 
-  log_info(cds)("Size of heap region = " SIZE_FORMAT " bytes, %d objects, %d roots",
-                _buffer_used, _source_objs->length() + 1, roots->length());
+  log_info(cds)("Size of heap region = " SIZE_FORMAT " bytes, %d objects, %d roots, %d native ptrs",
+                _buffer_used, _source_objs->length() + 1, roots->length(), _num_native_ptrs);
 }
 
 size_t ArchiveHeapWriter::filler_array_byte_size(int length) {
@@ -512,21 +561,35 @@ class ArchiveHeapWriter::EmbeddedOopRelocator: public BasicOopIterateClosure {
   }
 };
 
+static void log_bitmap_usage(const char* which, BitMap* bitmap, size_t total_bits) {
+  // The whole heap is covered by total_bits, but there are only non-zero bits within [start ... end).
+  size_t start = bitmap->find_first_set_bit(0);
+  size_t end = bitmap->size();
+  log_info(cds)("%s = " SIZE_FORMAT_W(7) " ... " SIZE_FORMAT_W(7) " (%3zu%% ... %3zu%% = %3zu%%)", which,
+                start, end,
+                start * 100 / total_bits,
+                end * 100 / total_bits,
+                (end - start) * 100 / total_bits);
+}
+
 // Update all oop fields embedded in the buffered objects
 void ArchiveHeapWriter::relocate_embedded_oops(GrowableArrayCHeap<oop, mtClassShared>* roots,
                                                ArchiveHeapInfo* heap_info) {
   size_t oopmap_unit = (UseCompressedOops ? sizeof(narrowOop) : sizeof(oop));
   size_t heap_region_byte_size = _buffer_used;
   heap_info->oopmap()->resize(heap_region_byte_size   / oopmap_unit);
 
-  auto iterator = [&] (oop src_obj, HeapShared::CachedOopInfo& info) {
-    oop requested_obj = requested_obj_from_buffer_offset(info.buffer_offset());
+  for (int i = 0; i < _source_objs_order->length(); i++) {
+    int src_obj_index = _source_objs_order->at(i);
+    oop src_obj = _source_objs->at(src_obj_index);
+    HeapShared::CachedOopInfo* info = HeapShared::archived_object_cache()->get(src_obj);
+    assert(info != nullptr, "must be");
+    oop requested_obj = requested_obj_from_buffer_offset(info->buffer_offset());
     update_header_for_requested_obj(requested_obj, src_obj, src_obj->klass());
-    address buffered_obj = offset_to_buffered_address<address>(info.buffer_offset());
+    address buffered_obj = offset_to_buffered_address<address>(info->buffer_offset());
     EmbeddedOopRelocator relocator(src_obj, buffered_obj, heap_info->oopmap());
     src_obj->oop_iterate(&relocator);
   };
-  HeapShared::archived_object_cache()->iterate_all(iterator);
 
   // Relocate HeapShared::roots(), which is created in copy_roots_to_buffer() and
   // doesn't have a corresponding src_obj, so we can't use EmbeddedOopRelocator on it.
@@ -542,6 +605,10 @@ void ArchiveHeapWriter::relocate_embedded_oops(GrowableArrayCHeap<oop, mtClassSh
   }
 
   compute_ptrmap(heap_info);
+
+  size_t total_bytes = (size_t)_buffer->length();
+  log_bitmap_usage("oopmap", heap_info->oopmap(), total_bytes / (UseCompressedOops ? sizeof(narrowOop) : sizeof(oop)));
+  log_bitmap_usage("ptrmap", heap_info->ptrmap(), total_bytes / sizeof(address));
 }
 
 void ArchiveHeapWriter::mark_native_pointer(oop src_obj, int field_offset) {
@@ -551,6 +618,8 @@ void ArchiveHeapWriter::mark_native_pointer(oop src_obj, int field_offset) {
     info._src_obj = src_obj;
     info._field_offset = field_offset;
     _native_pointers->append(info);
+    HeapShared::set_has_native_pointers(src_obj);
+    _num_native_ptrs ++;
   }
 }
 
@@ -565,6 +634,13 @@ bool ArchiveHeapWriter::is_marked_as_native_pointer(ArchiveHeapInfo* heap_info,
   assert((Metadata**)_requested_bottom <= requested_field_addr && requested_field_addr < (Metadata**) _requested_top, "range check");
 
   BitMap::idx_t idx = requested_field_addr - (Metadata**) _requested_bottom;
+  // Leading zeros have been removed so some addresses may not be in the ptrmap
+  size_t start_pos = FileMapInfo::current_info()->heap_ptrmap_start_pos();
+  if (idx < start_pos) {
+    return false;
+  } else {
+    idx -= start_pos;
+  }
   return (idx < heap_info->ptrmap()->size()) && (heap_info->ptrmap()->at(idx) == true);
 }
 

src/hotspot/share/cds/archiveHeapWriter.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2024, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -140,6 +140,7 @@ class ArchiveHeapWriter : AllStatic {
 
   static GrowableArrayCHeap<NativePointerInfo, mtClassShared>* _native_pointers;
   static GrowableArrayCHeap<oop, mtClassShared>* _source_objs;
+  static GrowableArrayCHeap<int, mtClassShared>* _source_objs_order;
 
   typedef ResourceHashtable<size_t, oop,
       36137, // prime number
@@ -210,6 +211,10 @@ class ArchiveHeapWriter : AllStatic {
   template <typename T> static void relocate_root_at(oop requested_roots, int index, CHeapBitMap* oopmap);
 
   static void update_header_for_requested_obj(oop requested_obj, oop src_obj, Klass* src_klass);
+
+  static int compare_objs_by_oop_fields(int* a, int* b);
+  static void sort_source_objs();
+
 public:
   static void init() NOT_CDS_JAVA_HEAP_RETURN;
   static void add_source_obj(oop src_obj);

src/hotspot/share/cds/filemap.cpp

@@ -289,6 +289,8 @@ void FileMapHeader::print(outputStream* st) {
   st->print_cr("- requested_base_address:         " INTPTR_FORMAT, p2i(_requested_base_address));
   st->print_cr("- mapped_base_address:            " INTPTR_FORMAT, p2i(_mapped_base_address));
   st->print_cr("- heap_roots_offset:              " SIZE_FORMAT, _heap_roots_offset);
+  st->print_cr("- _heap_oopmap_start_pos:         " SIZE_FORMAT, _heap_oopmap_start_pos);
+  st->print_cr("- _heap_ptrmap_start_pos:         " SIZE_FORMAT, _heap_ptrmap_start_pos);
   st->print_cr("- allow_archiving_with_java_agent:%d", _allow_archiving_with_java_agent);
   st->print_cr("- use_optimized_module_handling:  %d", _use_optimized_module_handling);
   st->print_cr("- has_full_module_graph           %d", _has_full_module_graph);
@@ -1565,11 +1567,37 @@ static size_t write_bitmap(const CHeapBitMap* map, char* output, size_t offset)
   return offset + size_in_bytes;
 }
 
+// The start of the archived heap has many primitive arrays (String
+// bodies) that are not marked by the oop/ptr maps. So we must have
+// lots of leading zeros.
+size_t FileMapInfo::remove_bitmap_leading_zeros(CHeapBitMap* map) {
+  size_t old_zeros = map->find_first_set_bit(0);
+  size_t old_size = map->size_in_bytes();
+
+  // Slice and resize bitmap
+  map->truncate(old_zeros, map->size());
+
+  DEBUG_ONLY(
+    size_t new_zeros = map->find_first_set_bit(0);
+    assert(new_zeros == 0, "Should have removed leading zeros");
+  )
+
+  assert(map->size_in_bytes() < old_size, "Map size should have decreased");
+  return old_zeros;
+}
+
 char* FileMapInfo::write_bitmap_region(const CHeapBitMap* ptrmap, ArchiveHeapInfo* heap_info,
                                        size_t &size_in_bytes) {
   size_in_bytes = ptrmap->size_in_bytes();
 
   if (heap_info->is_used()) {
+    // Remove leading zeros
+    size_t removed_oop_zeros = remove_bitmap_leading_zeros(heap_info->oopmap());
+    size_t removed_ptr_zeros = remove_bitmap_leading_zeros(heap_info->ptrmap());
+
+    header()->set_heap_oopmap_start_pos(removed_oop_zeros);
+    header()->set_heap_ptrmap_start_pos(removed_ptr_zeros);
+
     size_in_bytes += heap_info->oopmap()->size_in_bytes();
     size_in_bytes += heap_info->ptrmap()->size_in_bytes();
   }

src/hotspot/share/cds/filemap.hpp

@@ -228,6 +228,8 @@ class FileMapHeader: private CDSFileMapHeaderBase {
   size_t _ptrmap_size_in_bits;          // Size of pointer relocation bitmap
   size_t _heap_roots_offset;            // Offset of the HeapShared::roots() object, from the bottom
                                         // of the archived heap objects, in bytes.
+  size_t _heap_oopmap_start_pos;        // The first bit in the oopmap corresponds to this position in the heap.
+  size_t _heap_ptrmap_start_pos;        // The first bit in the ptrmap corresponds to this position in the heap.
   char* from_mapped_offset(size_t offset) const {
     return mapped_base_address() + offset;
   }
@@ -269,6 +271,8 @@ class FileMapHeader: private CDSFileMapHeaderBase {
   bool compressed_oops()                   const { return _compressed_oops; }
   bool compressed_class_pointers()         const { return _compressed_class_ptrs; }
   size_t heap_roots_offset()               const { return _heap_roots_offset; }
+  size_t heap_oopmap_start_pos()           const { return _heap_oopmap_start_pos;}
+  size_t heap_ptrmap_start_pos()           const { return _heap_ptrmap_start_pos;}
   // FIXME: These should really return int
   jshort max_used_path_index()             const { return _max_used_path_index; }
   jshort app_module_paths_start_index()    const { return _app_module_paths_start_index; }
@@ -281,6 +285,8 @@ class FileMapHeader: private CDSFileMapHeaderBase {
   void set_ptrmap_size_in_bits(size_t s)         { _ptrmap_size_in_bits = s; }
   void set_mapped_base_address(char* p)          { _mapped_base_address = p; }
   void set_heap_roots_offset(size_t n)           { _heap_roots_offset = n; }
+  void set_heap_oopmap_start_pos(size_t n)       { _heap_oopmap_start_pos = n; }
+  void set_heap_ptrmap_start_pos(size_t n)       { _heap_ptrmap_start_pos = n; }
   void copy_base_archive_name(const char* name);
 
   void set_shared_path_table(SharedPathTable table) {
@@ -378,6 +384,8 @@ class FileMapInfo : public CHeapObj<mtInternal> {
   uintx   max_heap_size()      const { return header()->max_heap_size(); }
   size_t  heap_roots_offset()  const { return header()->heap_roots_offset(); }
   size_t  core_region_alignment() const { return header()->core_region_alignment(); }
+  size_t  heap_oopmap_start_pos() const { return header()->heap_oopmap_start_pos(); }
+  size_t  heap_ptrmap_start_pos() const { return header()->heap_ptrmap_start_pos(); }
 
   CompressedOops::Mode narrow_oop_mode()      const { return header()->narrow_oop_mode(); }
   jshort app_module_paths_start_index()       const { return header()->app_module_paths_start_index(); }
@@ -434,6 +442,7 @@ class FileMapInfo : public CHeapObj<mtInternal> {
   void  write_header();
   void  write_region(int region, char* base, size_t size,
                      bool read_only, bool allow_exec);
+  size_t remove_bitmap_leading_zeros(CHeapBitMap* map);
   char* write_bitmap_region(const CHeapBitMap* ptrmap, ArchiveHeapInfo* heap_info,
                             size_t &size_in_bytes);
   size_t write_heap_region(ArchiveHeapInfo* heap_info);