/* * Copyright (C) 2016 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. AND ITS CONTRIBUTORS ``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 ITS 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. */ #pragma once #if ENABLE(DFG_JIT) namespace JSC { namespace DFG { class Graph; // Picks up groups of barriers that could be executed in any order with respect to each other and // places then at the earliest point in the program where the cluster would be correct. This phase // makes only the first of the cluster be a FencedStoreBarrier while the rest are normal // StoreBarriers. This phase also removes redundant barriers - for example, the cluster may end up // with two or more barriers on the same object, in which case it is totally safe for us to drop // one of them. The reason why this is sound hinges on the "earliest point where the cluster would // be correct" property. For example, take this input: // // a: Call() // b: PutByOffset(@o, @o, @x) // c: FencedStoreBarrier(@o) // d: PutByOffset(@o, @o, @y) // e: FencedStoreBarrier(@o) // f: PutByOffset(@p, @p, @z) // g: FencedStoreBarrier(@p) // h: GetByOffset(@q) // i: Call() // // The cluster of barriers is @c, @e, and @g. All of the barriers are between two doesGC effects: // the calls at @a and @i. Because there are no doesGC effects between @a and @i and there is no // possible control flow entry into this sequence between @ and @i, we could could just execute all // of the barriers just before @i in any order. The earliest point where the cluster would be // correct is just after @f, since that's the last operation that needs a barrier. We use the // earliest to reduce register pressure. When the barriers are clustered just after @f, we get: // // a: Call() // b: PutByOffset(@o, @o, @x) // d: PutByOffset(@o, @o, @y) // f: PutByOffset(@p, @p, @z) // c: FencedStoreBarrier(@o) // e: FencedStoreBarrier(@o) // g: FencedStoreBarrier(@p) // h: GetByOffset(@q) // i: Call() // // This phase does more. It takes advantage of the clustering to remove fences and remove redundant // barriers. So this phase will output this: // // a: Call() // b: PutByOffset(@o, @o, @x) // d: PutByOffset(@o, @o, @y) // f: PutByOffset(@p, @p, @z) // c: FencedStoreBarrier(@o) // g: StoreBarrier(@p) // h: GetByOffset(@q) // i: Call() // // This optimization improves both overall throughput and the throughput while the concurrent GC is // running. In the former, we are simplifying instruction selection for all but the first fence. In // the latter, we are reducing the cost of all but the first barrier. The first barrier will awlays // take slow path when there is concurrent GC activity, since the slow path contains the fence. But // all of the other barriers will only take slow path if they really need to remember the object. bool performStoreBarrierClustering(Graph&); } } // namespace JSC::DFG #endif // ENABLE(DFG_JIT)