| /* | 
 |  * Copyright (C) 2014, 2015 Apple Inc. All rights reserved. | 
 |  * | 
 |  * Redistribution and use in source and binary forms, with or without | 
 |  * modification, are permitted provided that the following conditions | 
 |  * are met: | 
 |  * 1. Redistributions of source code must retain the above copyright | 
 |  *    notice, this list of conditions and the following disclaimer. | 
 |  * 2. 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 with the distribution. | 
 |  * | 
 |  * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY | 
 |  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | 
 |  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | 
 |  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL APPLE INC. OR | 
 |  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, | 
 |  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | 
 |  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR | 
 |  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY | 
 |  * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | 
 |  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | 
 |  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.  | 
 |  */ | 
 |  | 
 | #include "config.h" | 
 | #include "DFGStructureAbstractValue.h" | 
 |  | 
 | #if ENABLE(DFG_JIT) | 
 |  | 
 | #include "DFGGraph.h" | 
 | #include "JSCJSValueInlines.h" | 
 |  | 
 | namespace JSC { namespace DFG { | 
 |  | 
 | #if ASSERT_ENABLED | 
 | void StructureAbstractValue::assertIsRegistered(Graph& graph) const | 
 | { | 
 |     if (isTop()) | 
 |         return; | 
 |      | 
 |     for (unsigned i = size(); i--;) | 
 |         graph.assertIsRegistered(at(i).get()); | 
 | } | 
 | #endif // ASSERT_ENABLED | 
 |  | 
 | void StructureAbstractValue::clobber() | 
 | { | 
 |     // The premise of this approach to clobbering is that anytime we introduce | 
 |     // a watchable structure into an abstract value, we watchpoint it. You can assert | 
 |     // that this holds by calling assertIsWatched(). | 
 |          | 
 |     if (isTop()) | 
 |         return; | 
 |  | 
 |     setClobbered(true); | 
 |          | 
 |     if (m_set.isThin()) { | 
 |         if (!m_set.singleEntry()) | 
 |             return; | 
 |         if (!m_set.singleEntry()->dfgShouldWatch()) | 
 |             makeTopWhenThin(); | 
 |         return; | 
 |     } | 
 |      | 
 |     RegisteredStructureSet::OutOfLineList* list = m_set.list(); | 
 |     for (unsigned i = list->m_length; i--;) { | 
 |         if (!list->list()[i]->dfgShouldWatch()) { | 
 |             makeTop(); | 
 |             return; | 
 |         } | 
 |     } | 
 | } | 
 |  | 
 | void StructureAbstractValue::observeTransition(RegisteredStructure from, RegisteredStructure to) | 
 | { | 
 |     ASSERT(!from->dfgShouldWatch()); | 
 |  | 
 |     if (isTop()) | 
 |         return; | 
 |      | 
 |     if (!m_set.contains(from)) | 
 |         return; | 
 |      | 
 |     if (!m_set.add(to)) | 
 |         return; | 
 |      | 
 |     if (m_set.size() > polymorphismLimit) | 
 |         makeTop(); | 
 | } | 
 |  | 
 | void StructureAbstractValue::observeTransitions(const TransitionVector& vector) | 
 | { | 
 |     if (isTop()) | 
 |         return; | 
 |      | 
 |     RegisteredStructureSet newStructures; | 
 |     for (unsigned i = vector.size(); i--;) { | 
 |         ASSERT(!vector[i].previous->dfgShouldWatch()); | 
 |  | 
 |         if (!m_set.contains(vector[i].previous)) | 
 |             continue; | 
 |          | 
 |         newStructures.add(vector[i].next); | 
 |     } | 
 |      | 
 |     if (!m_set.merge(newStructures)) | 
 |         return; | 
 |      | 
 |     if (m_set.size() > polymorphismLimit) | 
 |         makeTop(); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::add(RegisteredStructure structure) | 
 | { | 
 |     if (isTop()) | 
 |         return false; | 
 |      | 
 |     if (!m_set.add(structure)) | 
 |         return false; | 
 |      | 
 |     if (m_set.size() > polymorphismLimit) | 
 |         makeTop(); | 
 |      | 
 |     return true; | 
 | } | 
 |  | 
 | bool StructureAbstractValue::merge(const RegisteredStructureSet& other) | 
 | { | 
 |     if (isTop()) | 
 |         return false; | 
 |      | 
 |     return mergeNotTop(other); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::mergeSlow(const StructureAbstractValue& other) | 
 | { | 
 |     // It isn't immediately obvious that the code below is doing the right thing, so let's go | 
 |     // through it. | 
 |     // | 
 |     // This not clobbered, other not clobbered: Clearly, we don't want to make anything clobbered | 
 |     // since we just have two sets and we are merging them. mergeNotTop() can handle this just | 
 |     // fine. | 
 |     // | 
 |     // This clobbered, other clobbered: Clobbered means that we have a set of things, plus we | 
 |     // temporarily have the set of all things but the latter will go away once we hit the next | 
 |     // invalidation point. This allows us to merge two clobbered sets the natural way. For now | 
 |     // the set will still be TOP (and so we keep the clobbered bit set), but we know that after | 
 |     // invalidation, we will have the union of the this and other. | 
 |     // | 
 |     // This clobbered, other not clobbered: It's safe to merge in other for both before and after | 
 |     // invalidation, so long as we leave the clobbered bit set. Before invalidation this has no | 
 |     // effect since the set will still appear to have all things in it. The way to think about | 
 |     // what invalidation would do is imagine if we had a set A that was clobbered and a set B | 
 |     // that wasn't and we considered the following two cases. Note that we expect A to be the | 
 |     // same at the end in both cases: | 
 |     // | 
 |     //            A.merge(B)                             InvalidationPoint | 
 |     //            InvalidationPoint                      A.merge(B) | 
 |     // | 
 |     // The fact that we expect A to be the same in both cases means that we want to merge other | 
 |     // into this but keep the clobbered bit. | 
 |     // | 
 |     // This not clobbered, other clobbered: This is just the converse of the previous case. We | 
 |     // want to merge other into this and set the clobbered bit. | 
 |      | 
 |     bool changed = false; | 
 |      | 
 |     if (!isClobbered() && other.isClobbered()) { | 
 |         setClobbered(true); | 
 |         changed = true; | 
 |     } | 
 |      | 
 |     changed |= mergeNotTop(other.m_set); | 
 |      | 
 |     return changed; | 
 | } | 
 |  | 
 | bool StructureAbstractValue::mergeNotTop(const RegisteredStructureSet& other) | 
 | { | 
 |     if (!m_set.merge(other)) | 
 |         return false; | 
 |      | 
 |     if (m_set.size() > polymorphismLimit) | 
 |         makeTop(); | 
 |      | 
 |     return true; | 
 | } | 
 |  | 
 | void StructureAbstractValue::filter(const RegisteredStructureSet& other) | 
 | { | 
 |     if (isTop()) { | 
 |         m_set = other; | 
 |         return; | 
 |     } | 
 |      | 
 |     if (isClobbered()) { | 
 |         // We have two choices here: | 
 |         // | 
 |         // Do nothing: It's legal to keep our set intact, which would essentially mean that for | 
 |         // now, our set would behave like TOP but after the next invalidation point it wold be | 
 |         // a finite set again. This may be a good choice if 'other' is much bigger than our | 
 |         // m_set. | 
 |         // | 
 |         // Replace m_set with other and clear the clobber bit: This is also legal, and means that | 
 |         // we're no longer clobbered. This is usually better because it immediately gives us a | 
 |         // smaller set. | 
 |         // | 
 |         // This scenario should come up rarely. We usually don't do anything to an abstract value | 
 |         // after it is clobbered. But we apply some heuristics. | 
 |          | 
 |         if (other.size() > m_set.size() + clobberedSupremacyThreshold) | 
 |             return; // Keep the clobbered set. | 
 |          | 
 |         m_set = other; | 
 |         setClobbered(false); | 
 |         return; | 
 |     } | 
 |      | 
 |     m_set.filter(other); | 
 | } | 
 |  | 
 | void StructureAbstractValue::filter(const StructureAbstractValue& other) | 
 | { | 
 |     if (other.isTop()) | 
 |         return; | 
 |      | 
 |     if (other.isClobbered()) { | 
 |         if (isTop()) | 
 |             return; | 
 |          | 
 |         if (!isClobbered()) { | 
 |             // See justification in filter(const RegisteredStructureSet&), above. An unclobbered set is | 
 |             // almost always better. | 
 |             if (m_set.size() > other.m_set.size() + clobberedSupremacyThreshold) | 
 |                 *this = other; // Keep the clobbered set. | 
 |             return; | 
 |         } | 
 |  | 
 |         m_set.filter(other.m_set); | 
 |         return; | 
 |     } | 
 |      | 
 |     filter(other.m_set); | 
 | } | 
 |  | 
 | void StructureAbstractValue::filterSlow(SpeculatedType type) | 
 | { | 
 |     if (!(type & SpecCell)) { | 
 |         clear(); | 
 |         return; | 
 |     } | 
 |      | 
 |     ASSERT(!isTop()); | 
 |      | 
 |     m_set.genericFilter( | 
 |         [&] (RegisteredStructure structure) { | 
 |             return !!(speculationFromStructure(structure.get()) & type); | 
 |         }); | 
 | } | 
 |  | 
 | void StructureAbstractValue::filterClassInfoSlow(const ClassInfo* classInfo) | 
 | { | 
 |     ASSERT(!isTop()); | 
 |     m_set.genericFilter( | 
 |         [&] (RegisteredStructure structure) { | 
 |             return structure->classInfo()->isSubClassOf(classInfo); | 
 |         }); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::contains(RegisteredStructure structure) const | 
 | { | 
 |     if (isInfinite()) | 
 |         return true; | 
 |      | 
 |     return m_set.contains(structure); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::contains(Structure* structure) const | 
 | { | 
 |     if (isInfinite()) | 
 |         return true; | 
 |      | 
 |     return m_set.toStructureSet().contains(structure); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::isSubsetOf(const RegisteredStructureSet& other) const | 
 | { | 
 |     if (isInfinite()) | 
 |         return false; | 
 |      | 
 |     return m_set.isSubsetOf(other); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::isSubsetOf(const StructureAbstractValue& other) const | 
 | { | 
 |     if (isTop()) | 
 |         return false; | 
 |      | 
 |     if (other.isTop()) | 
 |         return true; | 
 |      | 
 |     if (isClobbered() == other.isClobbered()) | 
 |         return m_set.isSubsetOf(other.m_set); | 
 |      | 
 |     // Here it gets tricky. If in doubt, return false! | 
 |      | 
 |     if (isClobbered()) | 
 |         return false; // A clobbered set is never a subset of an unclobbered set. | 
 |      | 
 |     // An unclobbered set is currently a subset of a clobbered set, but it may not be so after | 
 |     // invalidation. | 
 |     return m_set.isSubsetOf(other.m_set); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::isSupersetOf(const RegisteredStructureSet& other) const | 
 | { | 
 |     if (isInfinite()) | 
 |         return true; | 
 |      | 
 |     return m_set.isSupersetOf(other); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::overlaps(const RegisteredStructureSet& other) const | 
 | { | 
 |     if (isInfinite()) | 
 |         return true; | 
 |      | 
 |     return m_set.overlaps(other); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::overlaps(const StructureAbstractValue& other) const | 
 | { | 
 |     if (other.isInfinite()) | 
 |         return true; | 
 |      | 
 |     return overlaps(other.m_set); | 
 | } | 
 |  | 
 | bool StructureAbstractValue::isSubClassOf(const ClassInfo* classInfo) const | 
 | { | 
 |     if (isInfinite()) | 
 |         return false; | 
 |  | 
 |     // Note that this function returns true if the structure set is empty. | 
 |     for (const RegisteredStructure structure : m_set) { | 
 |         if (!structure->classInfo()->isSubClassOf(classInfo)) | 
 |             return false; | 
 |     } | 
 |     return true; | 
 | } | 
 |  | 
 | bool StructureAbstractValue::isNotSubClassOf(const ClassInfo* classInfo) const | 
 | { | 
 |     if (isInfinite()) | 
 |         return false; | 
 |  | 
 |     // Note that this function returns true if the structure set is empty. | 
 |     for (const RegisteredStructure structure : m_set) { | 
 |         if (structure->classInfo()->isSubClassOf(classInfo)) | 
 |             return false; | 
 |     } | 
 |     return true; | 
 | } | 
 |  | 
 | bool StructureAbstractValue::equalsSlow(const StructureAbstractValue& other) const | 
 | { | 
 |     ASSERT(m_set.m_pointer != other.m_set.m_pointer); | 
 |     ASSERT(!isTop()); | 
 |     ASSERT(!other.isTop()); | 
 |      | 
 |     return m_set == other.m_set | 
 |         && isClobbered() == other.isClobbered(); | 
 | } | 
 |  | 
 | void StructureAbstractValue::dumpInContext(PrintStream& out, DumpContext* context) const | 
 | { | 
 |     if (isClobbered()) | 
 |         out.print("Clobbered:"); | 
 |      | 
 |     if (isTop()) | 
 |         out.print("TOP"); | 
 |     else | 
 |         out.print(inContext(m_set.toStructureSet(), context)); | 
 | } | 
 |  | 
 | void StructureAbstractValue::dump(PrintStream& out) const | 
 | { | 
 |     dumpInContext(out, nullptr); | 
 | } | 
 |  | 
 | void StructureAbstractValue::validateReferences(const TrackedReferences& trackedReferences) const | 
 | { | 
 |     if (isTop()) | 
 |         return; | 
 |     m_set.validateReferences(trackedReferences); | 
 | } | 
 |  | 
 | } } // namespace JSC::DFG | 
 |  | 
 | #endif // ENABLE(DFG_JIT) | 
 |  |