1 /*
2 * Copyright (c) 2020, 2022, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 */
23 package org.openjdk.bench.java.lang.foreign;
24
25 import java.lang.foreign.*;
26
27 import org.openjdk.jmh.annotations.Benchmark;
28 import org.openjdk.jmh.annotations.BenchmarkMode;
29 import org.openjdk.jmh.annotations.Fork;
30 import org.openjdk.jmh.annotations.Measurement;
31 import org.openjdk.jmh.annotations.Mode;
32 import org.openjdk.jmh.annotations.OutputTimeUnit;
33 import org.openjdk.jmh.annotations.Setup;
34 import org.openjdk.jmh.annotations.State;
35 import org.openjdk.jmh.annotations.TearDown;
36 import org.openjdk.jmh.annotations.Warmup;
37 import sun.misc.Unsafe;
38
39 import java.nio.ByteBuffer;
40 import java.nio.ByteOrder;
41 import java.util.concurrent.TimeUnit;
42
43 import static java.lang.foreign.ValueLayout.*;
44
45 @BenchmarkMode(Mode.AverageTime)
46 @Warmup(iterations = 5, time = 500, timeUnit = TimeUnit.MILLISECONDS)
47 @Measurement(iterations = 10, time = 500, timeUnit = TimeUnit.MILLISECONDS)
48 @State(org.openjdk.jmh.annotations.Scope.Thread)
49 @OutputTimeUnit(TimeUnit.MILLISECONDS)
50 @Fork(3)
51 public class LoopOverNew extends JavaLayouts {
52
53 static final Unsafe unsafe = Utils.unsafe;
54
55 static final int ELEM_SIZE = 1_000_000;
56 static final int CARRIER_SIZE = (int)JAVA_INT.byteSize();
57 static final int ALLOC_SIZE = ELEM_SIZE * CARRIER_SIZE;
58 static final MemoryLayout ALLOC_LAYOUT = MemoryLayout.sequenceLayout(ELEM_SIZE, JAVA_INT);
59 final Arena arena = Arena.ofConfined();
60 final SegmentAllocator recyclingAlloc = SegmentAllocator.prefixAllocator(arena.allocate(ALLOC_LAYOUT));
61
62 @TearDown
63 public void tearDown() throws Throwable {
64 arena.close();
65 }
66
67 @Benchmark
68 public void unsafe_loop() {
69 long unsafe_addr = unsafe.allocateMemory(ALLOC_SIZE);
70 for (int i = 0; i < ELEM_SIZE; i++) {
71 unsafe.putInt(unsafe_addr + (i * CARRIER_SIZE) , i);
72 }
73 unsafe.freeMemory(unsafe_addr);
74 }
75
76 @Benchmark
77 public void segment_loop_confined() {
78 try (Arena arena = Arena.ofConfined()) {
79 MemorySegment segment = arena.allocate(ALLOC_SIZE, 4);
80 for (int i = 0; i < ELEM_SIZE; i++) {
81 VH_INT.set(segment, (long) i, i);
82 }
83 }
84 }
85
86 @Benchmark
87 public void segment_loop_shared() {
88 try (Arena arena = Arena.ofShared()) {
89 MemorySegment segment = arena.allocate(ALLOC_SIZE, 4);
90 for (int i = 0; i < ELEM_SIZE; i++) {
91 VH_INT.set(segment, (long) i, i);
92 }
93 }
94 }
95
96 @Benchmark
97 public void segment_loop_recycle() {
98 MemorySegment segment = recyclingAlloc.allocate(ALLOC_SIZE, 4);
99 for (int i = 0; i < ELEM_SIZE; i++) {
100 VH_INT.set(segment, (long) i, i);
101 }
102 }
103
104 @Benchmark
105 public void buffer_loop() {
106 ByteBuffer byteBuffer = ByteBuffer.allocateDirect(ALLOC_SIZE).order(ByteOrder.nativeOrder());
107 for (int i = 0; i < ELEM_SIZE; i++) {
108 byteBuffer.putInt(i * CARRIER_SIZE , i);
109 }
110 unsafe.invokeCleaner(byteBuffer);
111 }
112
113 // hack to even out calls to System::gc, which allows us to compare how the implicit segment deallocation
114 // fares compared with ByteBuffer; if there's no call to System.gc() we end up comparing how well the two
115 // act under significant native memory pressure, and here the ByteBuffer API has more juice, since it features
116 // a complex exponential back off with multiple GC retries (see ByteBuffer::allocateDirect). Of course, we
117 // don't care about those cases with segments, as if clients need to allocate/free very frequently
118 // they should just use deterministic deallocation (with confined session) instead, which delivers much
119 // better performances anyway.
120 static byte gcCount = 0;
121
122 @Benchmark
123 public void buffer_loop_implicit() {
124 if (gcCount++ == 0) System.gc(); // GC when we overflow
125 ByteBuffer byteBuffer = ByteBuffer.allocateDirect(ALLOC_SIZE).order(ByteOrder.nativeOrder());
126 for (int i = 0; i < ELEM_SIZE; i++) {
127 byteBuffer.putInt(i * CARRIER_SIZE , i);
128 }
129 }
130
131 @Benchmark
132 public void segment_loop_implicit() {
133 if (gcCount++ == 0) System.gc(); // GC when we overflow
134 Arena scope = Arena.ofAuto();
135 MemorySegment segment = scope.allocate(ALLOC_SIZE, 4);
136 for (int i = 0; i < ELEM_SIZE; i++) {
137 VH_INT.set(segment, (long) i, i);
138 }
139 }
140 }