Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(169)

Unified Diff: src/ic/stub-cache.h

Issue 483683005: Move IC code into a subdir and move ic-compilation related code from stub-cache into ic-compiler (Closed) Base URL: https://v8.googlecode.com/svn/branches/bleeding_edge
Patch Set: Fix BUILD.gn Created 6 years, 4 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « src/ic/ic-inl.h ('k') | src/ic/stub-cache.cc » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: src/ic/stub-cache.h
diff --git a/src/ic/stub-cache.h b/src/ic/stub-cache.h
new file mode 100644
index 0000000000000000000000000000000000000000..eb5343f85610f7d0b0f87f8612178568584aeab8
--- /dev/null
+++ b/src/ic/stub-cache.h
@@ -0,0 +1,168 @@
+// Copyright 2012 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_STUB_CACHE_H_
+#define V8_STUB_CACHE_H_
+
+#include "src/macro-assembler.h"
+
+namespace v8 {
+namespace internal {
+
+
+// The stub cache is used for megamorphic property accesses.
+// It maps (map, name, type) to property access handlers. The cache does not
+// need explicit invalidation when a prototype chain is modified, since the
+// handlers verify the chain.
+
+
+class SCTableReference {
+ public:
+ Address address() const { return address_; }
+
+ private:
+ explicit SCTableReference(Address address) : address_(address) {}
+
+ Address address_;
+
+ friend class StubCache;
+};
+
+
+class StubCache {
+ public:
+ struct Entry {
+ Name* key;
+ Code* value;
+ Map* map;
+ };
+
+ void Initialize();
+ // Access cache for entry hash(name, map).
+ Code* Set(Name* name, Map* map, Code* code);
+ Code* Get(Name* name, Map* map, Code::Flags flags);
+ // Clear the lookup table (@ mark compact collection).
+ void Clear();
+ // Collect all maps that match the name and flags.
+ void CollectMatchingMaps(SmallMapList* types, Handle<Name> name,
+ Code::Flags flags, Handle<Context> native_context,
+ Zone* zone);
+ // Generate code for probing the stub cache table.
+ // Arguments extra, extra2 and extra3 may be used to pass additional scratch
+ // registers. Set to no_reg if not needed.
+ void GenerateProbe(MacroAssembler* masm, Code::Flags flags, Register receiver,
+ Register name, Register scratch, Register extra,
+ Register extra2 = no_reg, Register extra3 = no_reg);
+
+ enum Table { kPrimary, kSecondary };
+
+ SCTableReference key_reference(StubCache::Table table) {
+ return SCTableReference(
+ reinterpret_cast<Address>(&first_entry(table)->key));
+ }
+
+ SCTableReference map_reference(StubCache::Table table) {
+ return SCTableReference(
+ reinterpret_cast<Address>(&first_entry(table)->map));
+ }
+
+ SCTableReference value_reference(StubCache::Table table) {
+ return SCTableReference(
+ reinterpret_cast<Address>(&first_entry(table)->value));
+ }
+
+ StubCache::Entry* first_entry(StubCache::Table table) {
+ switch (table) {
+ case StubCache::kPrimary:
+ return StubCache::primary_;
+ case StubCache::kSecondary:
+ return StubCache::secondary_;
+ }
+ UNREACHABLE();
+ return NULL;
+ }
+
+ Isolate* isolate() { return isolate_; }
+
+ // Setting the entry size such that the index is shifted by Name::kHashShift
+ // is convenient; shifting down the length field (to extract the hash code)
+ // automatically discards the hash bit field.
+ static const int kCacheIndexShift = Name::kHashShift;
+
+ private:
+ explicit StubCache(Isolate* isolate);
+
+ // The stub cache has a primary and secondary level. The two levels have
+ // different hashing algorithms in order to avoid simultaneous collisions
+ // in both caches. Unlike a probing strategy (quadratic or otherwise) the
+ // update strategy on updates is fairly clear and simple: Any existing entry
+ // in the primary cache is moved to the secondary cache, and secondary cache
+ // entries are overwritten.
+
+ // Hash algorithm for the primary table. This algorithm is replicated in
+ // assembler for every architecture. Returns an index into the table that
+ // is scaled by 1 << kCacheIndexShift.
+ static int PrimaryOffset(Name* name, Code::Flags flags, Map* map) {
+ STATIC_ASSERT(kCacheIndexShift == Name::kHashShift);
+ // Compute the hash of the name (use entire hash field).
+ DCHECK(name->HasHashCode());
+ uint32_t field = name->hash_field();
+ // Using only the low bits in 64-bit mode is unlikely to increase the
+ // risk of collision even if the heap is spread over an area larger than
+ // 4Gb (and not at all if it isn't).
+ uint32_t map_low32bits =
+ static_cast<uint32_t>(reinterpret_cast<uintptr_t>(map));
+ // We always set the in_loop bit to zero when generating the lookup code
+ // so do it here too so the hash codes match.
+ uint32_t iflags =
+ (static_cast<uint32_t>(flags) & ~Code::kFlagsNotUsedInLookup);
+ // Base the offset on a simple combination of name, flags, and map.
+ uint32_t key = (map_low32bits + field) ^ iflags;
+ return key & ((kPrimaryTableSize - 1) << kCacheIndexShift);
+ }
+
+ // Hash algorithm for the secondary table. This algorithm is replicated in
+ // assembler for every architecture. Returns an index into the table that
+ // is scaled by 1 << kCacheIndexShift.
+ static int SecondaryOffset(Name* name, Code::Flags flags, int seed) {
+ // Use the seed from the primary cache in the secondary cache.
+ uint32_t name_low32bits =
+ static_cast<uint32_t>(reinterpret_cast<uintptr_t>(name));
+ // We always set the in_loop bit to zero when generating the lookup code
+ // so do it here too so the hash codes match.
+ uint32_t iflags =
+ (static_cast<uint32_t>(flags) & ~Code::kFlagsNotUsedInLookup);
+ uint32_t key = (seed - name_low32bits) + iflags;
+ return key & ((kSecondaryTableSize - 1) << kCacheIndexShift);
+ }
+
+ // Compute the entry for a given offset in exactly the same way as
+ // we do in generated code. We generate an hash code that already
+ // ends in Name::kHashShift 0s. Then we multiply it so it is a multiple
+ // of sizeof(Entry). This makes it easier to avoid making mistakes
+ // in the hashed offset computations.
+ static Entry* entry(Entry* table, int offset) {
+ const int multiplier = sizeof(*table) >> Name::kHashShift;
+ return reinterpret_cast<Entry*>(reinterpret_cast<Address>(table) +
+ offset * multiplier);
+ }
+
+ static const int kPrimaryTableBits = 11;
+ static const int kPrimaryTableSize = (1 << kPrimaryTableBits);
+ static const int kSecondaryTableBits = 9;
+ static const int kSecondaryTableSize = (1 << kSecondaryTableBits);
+
+ Entry primary_[kPrimaryTableSize];
+ Entry secondary_[kSecondaryTableSize];
+ Isolate* isolate_;
+
+ friend class Isolate;
+ friend class SCTableReference;
+
+ DISALLOW_COPY_AND_ASSIGN(StubCache);
+};
+}
+} // namespace v8::internal
+
+#endif // V8_STUB_CACHE_H_
« no previous file with comments | « src/ic/ic-inl.h ('k') | src/ic/stub-cache.cc » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698