Java / Kotlin 极限性能指南：省内存与跑得快

# JDK 25：正式但默认关闭，可以直接打开
java -XX:+UseCompactObjectHeaders -jar app.jar
# JDK 27：默认开启，如需关闭
java -XX:-UseCompactObjectHeaders -jar app.jar

【本机实测】JolSizes$Mixed（默认对象头）
OFF  SZ     TYPE DESCRIPTION
  0   8          (object header: mark)
  8   4          (object header: class)
 12   4      int Mixed.i
 16   8     long Mixed.l
 24   1     byte Mixed.b
 25   1  boolean Mixed.z
 26   2          (alignment/padding gap)
 28   4   Object Mixed.ref
Instance size: 32 bytes

import org.openjdk.jol.info.ClassLayout;
import org.openjdk.jol.info.GraphLayout;
import org.openjdk.jol.vm.VM;
import java.util.*;

/** 用 JOL 实测对象大小。运行：java -cp jol-core.jar:. [-XX:+UseCompactObjectHeaders] JolSizes */
public class JolSizes {
    enum Fuel { ELECTRIC, BENZINE, DIESEL }
    // 视频里的四步
    static final class HybridTwoInts { int electric; int benzine; }   // C: struct { int; int; } = 8 字节
    static final class FlagsInt { int flags; }                         // C: int flags = 4 字节
    static final class FlagsByte { byte flags; }                       // C: uint8_t 位域 = 1 字节
    static final class FlagsBools { boolean electric, benzine, diesel; }
    static final class FlagsEnumSet { final EnumSet<Fuel> fuels = EnumSet.noneOf(Fuel.class); }
    static final class Point { int x; int y; }
    static final class Mixed { byte b; long l; int i; Object ref; boolean z; }

    static void cls(Class<?> c) {
        System.out.printf("%-28s instance=%d%n", c.getSimpleName(), ClassLayout.parseClass(c).instanceSize());
    }
    static void graph(String what, Object o) {
        System.out.printf("%-40s total=%d bytes, objects=%d%n", what, GraphLayout.parseInstance(o).totalSize(),
                GraphLayout.parseInstance(o).totalCount());
    }

    public static void main(String[] args) {
        System.out.println("java.version=" + System.getProperty("java.version"));
        System.out.println(VM.current().details());
        System.out.println(ClassLayout.parseClass(Mixed.class).toPrintable());
        for (Class<?> c : List.of(Object.class, Integer.class, Long.class, Boolean.class, Point.class,
                HybridTwoInts.class, FlagsInt.class, FlagsByte.class, FlagsBools.class, FlagsEnumSet.class, Mixed.class)) cls(c);
        // EnumSet/EnumMap 引用的 Class 对象、枚举常量和 universe 数组是全类型共享的，这里只计每个实例独有的部分
        EnumSet<Fuel> es = EnumSet.of(Fuel.ELECTRIC, Fuel.BENZINE);
        long esSize = ClassLayout.parseInstance(es).instanceSize();
        System.out.printf("%-40s %s=%d bytes (独有部分)%n", "EnumSet.of(ELECTRIC,BENZINE)", es.getClass().getSimpleName(), esSize);
        System.out.printf("%-40s total=%d bytes (持有者 + RegularEnumSet)%n", "FlagsEnumSet 对象", ClassLayout.parseClass(FlagsEnumSet.class).instanceSize() + esSize);
        graph("BitSet(3)", new BitSet(3));
        graph("20 个 ASCII 字符的 String (Latin1)", new String("abcdefghijklmnopqrst"));
        graph("20 个汉字的 String (UTF16)", new String("一二三四五六七八九十一二三四五六七八九十"));
        graph("new int[0]", new int[0]);
        graph("new byte[1000]", new byte[1000]);
        graph("new int[1000]", new int[1000]);
        graph("new long[1000]", new long[1000]);
        HybridTwoInts[] cars = new HybridTwoInts[1000];
        for (int i = 0; i < cars.length; i++) cars[i] = new HybridTwoInts();
        graph("HybridTwoInts[1000] + 1000 objects", cars);
        FlagsByte[] fb = new FlagsByte[1000];
        for (int i = 0; i < fb.length; i++) fb[i] = new FlagsByte();
        graph("FlagsByte[1000] + 1000 objects", fb);
        graph("BitSet(3000) (1000 辆车 × 3 位)", new BitSet(3000));
        List<Integer> boxed = new ArrayList<>(1000);
        for (int i = 0; i < 1000; i++) boxed.add(i + 1000); // 超出 Integer 缓存，确保都是新对象
        graph("ArrayList<Integer> 1000 个 (>=1000)", boxed);
        int[] prim = new int[1000];
        graph("int[1000]", prim);
        Map<Integer, Integer> hm = new HashMap<>();
        for (int i = 0; i < 1000; i++) hm.put(i + 1000, i + 1000);
        graph("HashMap<Integer,Integer> 1000 项", hm);
        var fu = new it.unimi.dsi.fastutil.ints.Int2IntOpenHashMap();
        for (int i = 0; i < 1000; i++) fu.put(i + 1000, i + 1000);
        graph("fastutil Int2IntOpenHashMap 1000 项", fu);
        EnumMap<Fuel, Integer> em = new EnumMap<>(Fuel.class);
        System.out.printf("%-40s total=%d bytes (EnumMap 对象 + 长度 3 的 vals 数组，独有部分)%n", "EnumMap<Fuel,Integer>",
                ClassLayout.parseInstance(em).instanceSize() + ClassLayout.parseInstance(new Object[3]).instanceSize());
    }
}

import java.util.ArrayList;
import java.util.List;
import java.util.stream.IntStream;

/** Boxing costs and pitfalls: List<Integer> vs int[], the Integer cache, and reference equality. */
public class J01_Boxing {
    // jvmperf-lint: ignore jp-boxed-primitive-collection -- Intentional counterexample.
    static long sumBoxed(List<Integer> xs) {
        long s = 0;
        for (Integer x : xs) s += x;          // Unboxes each value; throws NPE if x is null.
        return s;
    }
    static long sumPrimitive(int[] xs) {
        long s = 0;
        for (int x : xs) s += x;              // Contiguous storage, no pointer chasing; C2 may vectorize it.
        return s;
    }
    public static void main(String[] args) {
        int n = 10_000;
        // jvmperf-lint: ignore jp-boxed-primitive-collection -- Intentional counterexample.
        List<Integer> boxed = new ArrayList<>(n);
        int[] prim = new int[n];
        for (int i = 0; i < n; i++) { boxed.add(i); prim[i] = i; }
        check(sumBoxed(boxed) == sumPrimitive(prim), "sum");

        // Integer.valueOf caches -128..127; raise the upper bound with -XX:AutoBoxCacheMax.
        Integer a = 127, b = 127, c = 128, d = 128;
        check(a == b, "cached");                   // The same cached object.
        check(c.equals(d), "equals");              // Always use equals for wrapper values.
        // c == d is false with default settings; do not rely on reference identity.

        // Streams: mapToInt keeps primitive values; map(...).reduce(Integer::sum) boxes each step.
        long s1 = IntStream.range(0, n).asLongStream().sum();
        int s2 = boxed.stream().mapToInt(Integer::intValue).sum();
        check(s1 == sumPrimitive(prim) && s2 == (int) s1, "streams");
        System.out.println("J01 ok");
    }
    static void check(boolean ok, String what) { if (!ok) throw new AssertionError(what); }
}

import it.unimi.dsi.fastutil.ints.Int2IntOpenHashMap;
import it.unimi.dsi.fastutil.ints.IntArrayList;
import org.eclipse.collections.api.list.primitive.MutableIntList;
import org.eclipse.collections.impl.factory.primitive.IntLists;
import org.eclipse.collections.impl.map.mutable.primitive.IntIntHashMap;

/** Primitive collections from fastutil and Eclipse Collections avoid boxing and per-entry Node objects. */
public class J02_PrimitiveCollections {
    public static void main(String[] args) {
        int n = 1000;
        Int2IntOpenHashMap counts = new Int2IntOpenHashMap(n);   // Reserve capacity.
        counts.defaultReturnValue(0);
        for (int i = 0; i < n; i++) counts.addTo(i % 10, 1);     // Increment in place without boxing.
        check(counts.get(3) == 100, "fastutil map");

        IntArrayList list = new IntArrayList(n);
        for (int i = 0; i < n; i++) list.add(i);
        check(list.getInt(999) == 999, "fastutil list");

        IntIntHashMap ec = new IntIntHashMap(n);
        for (int i = 0; i < n; i++) ec.addToValue(i % 10, 1);
        MutableIntList ecl = IntLists.mutable.withInitialCapacity(n);
        for (int i = 0; i < n; i++) ecl.add(i);
        check(ec.get(3) == 100 && ecl.sum() == 499500L, "eclipse collections");
        System.out.println("J02 ok");
    }
    static void check(boolean ok, String what) { if (!ok) throw new AssertionError(what); }
}

import java.util.BitSet;
import java.util.EnumSet;

/** Flags using bit operations, EnumSet, and BitSet; this is the hybrid-car case study. */
public class J03_Flags {
    enum Fuel { ELECTRIC, BENZINE, DIESEL }

    // 1) One field per flag: three booleans.
    static final class CarBools { boolean electric, benzine, diesel; }

    // 2) Store three flags in one byte; Java requires explicit bit operations.
    static final class CarByte {
        static final int ELECTRIC = 1, BENZINE = 1 << 1, DIESEL = 1 << 2;
        byte flags;
        boolean has(int f) { return (flags & f) != 0; }
        void set(int f) { flags = (byte) (flags | f); }
        void clear(int f) { flags = (byte) (flags & ~f); }
    }

    // 3) Packing saves meaningful memory in arrays: one byte or three bits per car.
    static final class Fleet {
        final byte[] flags;               // Primitive array: one byte per car.
        Fleet(int n) { flags = new byte[n]; }
    }
    static final class PackedFleet {
        final long[] bits;                // Three densely packed bits per car.
        PackedFleet(int n) { bits = new long[(n * 3 + 63) >>> 6]; }
        void set(int car, int fuelBit) { int i = car * 3 + fuelBit; bits[i >>> 6] |= 1L << i; } // long shifts use the low six bits of the distance.
        boolean has(int car, int fuelBit) { int i = car * 3 + fuelBit; return (bits[i >>> 6] & (1L << i)) != 0; }
    }

    public static void main(String[] args) {
        CarByte hybrid = new CarByte();
        hybrid.set(CarByte.ELECTRIC | CarByte.BENZINE);
        check(hybrid.has(CarByte.ELECTRIC) && !hybrid.has(CarByte.DIESEL) && hybrid.flags == 0b011, "byte flags");

        EnumSet<Fuel> fuels = EnumSet.of(Fuel.ELECTRIC, Fuel.BENZINE);   // Up to 64 constants fit in one long.
        check(fuels.contains(Fuel.BENZINE) && fuels.size() == 2, "EnumSet");

        BitSet bs = new BitSet(3000);
        bs.set(5 * 3 + Fuel.BENZINE.ordinal());
        check(bs.get(16) && bs.cardinality() == 1, "BitSet");

        PackedFleet pf = new PackedFleet(1000);
        check(pf.bits.length == 47, "47 longs = 376 bytes of data"); // ceil(3000/64)
        pf.set(999, 2);
        check(pf.has(999, 2) && !pf.has(999, 1), "packed");
        Fleet f = new Fleet(1000);
        f.flags[0] = CarByte.ELECTRIC | CarByte.BENZINE;
        check(Integer.bitCount(f.flags[0]) == 2, "fleet");
        System.out.println("J03 ok");
    }
    static void check(boolean ok, String what) { if (!ok) throw new AssertionError(what); }
}

// Kotlin version of the flag case study: boolean fields, bit operations, a value class, and arrays.
package k05

import java.util.EnumSet

enum class Fuel { ELECTRIC, BENZINE, DIESEL }

data class HybridBools(val electric: Boolean, val benzine: Boolean, val diesel: Boolean)

// The value class adds a type and operations to Byte; the runtime representation remains a byte.
@JvmInline
value class FuelFlags(val bits: Byte) {
    operator fun contains(f: Fuel): Boolean = (bits.toInt() and (1 shl f.ordinal)) != 0
    operator fun plus(f: Fuel): FuelFlags = FuelFlags((bits.toInt() or (1 shl f.ordinal)).toByte())
    operator fun minus(f: Fuel): FuelFlags = FuelFlags((bits.toInt() and (1 shl f.ordinal).inv()).toByte())
    companion object {
        val NONE = FuelFlags(0)
    }
}

class Car(var fuels: FuelFlags)      // The field is represented as a byte.

fun main() {
    val hybrid = FuelFlags.NONE + Fuel.ELECTRIC + Fuel.BENZINE
    check(Fuel.ELECTRIC in hybrid && Fuel.DIESEL !in hybrid)
    check(hybrid.bits.toInt() == 0b011)
    check((hybrid - Fuel.BENZINE).bits.toInt() == 0b001)
    check(Car::class.java.getDeclaredField("fuels").type == Byte::class.javaPrimitiveType)
    // Kotlin uses infix functions for bit operations: shl, shr, ushr, and, or, xor, inv.
    check((1 shl 2) or 1 == 5 && (0b110 and 0b011) == 0b010 && 5.inv() == -6)
    // For 1,000 cars, ByteArray uses one byte each; Array<FuelFlags> and List<FuelFlags> box every value.
    val fleet = ByteArray(1000)
    fleet[7] = hybrid.bits
    check(Fuel.BENZINE in FuelFlags(fleet[7]))
    val es: EnumSet<Fuel> = EnumSet.of(Fuel.ELECTRIC, Fuel.BENZINE)
    check(es.size == 2 && HybridBools(true, true, false).electric)
    println("K05 ok")
}

/** Array of Structures vs Structure of Arrays: SoA is contiguous and header-free when scanning one field. */
public class J04_SoA {
    static final class Particle { double x, y, vx, vy; boolean alive; }   // AoS: each particle is a separate heap object.

    static final class Particles {                                       // SoA: one primitive array per field.
        final double[] x, y, vx, vy;
        final long[] alive;                                              // Store liveness flags out of band in a bitmap.
        int size;
        Particles(int cap) {
            x = new double[cap]; y = new double[cap]; vx = new double[cap]; vy = new double[cap];
            alive = new long[(cap + 63) >>> 6];
        }
        int add(double px, double py, double pvx, double pvy) {
            int i = size++;
            x[i] = px; y[i] = py; vx[i] = pvx; vy[i] = pvy; alive[i >>> 6] |= 1L << i;
            return i;
        }
        void step(double dt) {                                           // The hot loop touches only the needed arrays.
            for (int i = 0; i < size; i++) { x[i] += vx[i] * dt; y[i] += vy[i] * dt; }
        }
    }

    public static void main(String[] args) {
        Particles ps = new Particles(1024);
        for (int i = 0; i < 1000; i++) ps.add(i, 0, 1, 2);
        ps.step(0.5);
        if (ps.x[10] != 10.5 || ps.y[10] != 1.0) throw new AssertionError();
        Particle[] aos = new Particle[1000];
        for (int i = 0; i < aos.length; i++) { aos[i] = new Particle(); aos[i].x = i; aos[i].vx = 1; }
        for (Particle p : aos) p.x += p.vx * 0.5;
        if (aos[10].x != ps.x[10]) throw new AssertionError();
        System.out.println("J04 ok");
    }
}

import java.util.ArrayList;
import java.util.EnumMap;
import java.util.EnumSet;
import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
import java.util.Map;

/** Reserve collection capacity and use EnumMap/EnumSet for enum keys. */
public class J13_Presize {
    enum Color { RED, GREEN, BLUE }
    public static void main(String[] args) {
        int n = 10_000;
        List<String> names = new ArrayList<>(n);                  // Avoid repeated growth and array copies.
        // jvmperf-lint: ignore jp-boxed-primitive-collection -- Intentional counterexample.
        Map<String, Integer> idx = HashMap.newHashMap(n);           // JDK 19+: reserve by element count, accounting for the load factor.
        HashSet<String> seen = HashSet.newHashSet(n);               // JDK 19+
        for (int i = 0; i < n; i++) { String s = "k" + i; names.add(s); idx.put(s, i); seen.add(s); }
        // Note: new HashMap<>(n) takes table capacity; inserting n elements may still resize it.
        EnumMap<Color, Integer> hits = new EnumMap<>(Color.class);  // Internally an array indexed by ordinal.
        hits.merge(Color.RED, 1, Integer::sum);
        EnumSet<Color> warm = EnumSet.of(Color.RED, Color.GREEN);
        if (idx.get("k42") != 42 || !warm.contains(Color.RED) || hits.get(Color.RED) != 1 || seen.size() != n)
            throw new AssertionError();
        System.out.println("J13 ok");
    }
}

import java.util.Arrays;

/** Immutable values and bulk operations: record arrays vs primitive arrays; Arrays methods use intrinsics. */
public class J18_Records {
    record Sample(long ts, double value) {}
    public static void main(String[] args) {
        int n = 1000;
        long[] ts = new long[n];
        double[] vs = new double[n];
        Arrays.setAll(ts, i -> 1_000L * i);
        Arrays.fill(vs, 1.5);
        double sum = 0;
        for (double v : vs) sum += v;
        long[] copy = Arrays.copyOf(ts, n * 2);                // System.arraycopy intrinsic
        if (sum != 1500 || copy[999] != 999_000L || Arrays.binarySearch(ts, 5_000L) != 5) throw new AssertionError();
        Sample s = new Sample(ts[1], vs[1]);
        if (s.ts() != 1000) throw new AssertionError();
        System.out.println("J18 ok");
    }
}

import java.util.List;
import java.util.StringJoiner;

/** Strings: reserve StringBuilder capacity, avoid += in loops, and use join or intern appropriately. */
public class J05_Strings {
    static String joinSlow(List<String> parts) {
        String s = "";
        // jvmperf-lint: ignore jp-string-concat-hot -- Intentional counterexample.
        for (String p : parts) s += p + ",";          // Creates and copies a new String each iteration: O(n²).
        return s;
    }
    static String joinFast(List<String> parts) {
        int cap = 0;
        for (String p : parts) cap += p.length() + 1;
        StringBuilder sb = new StringBuilder(cap);     // Allocate enough capacity once.
        for (String p : parts) sb.append(p).append(',');
        return sb.toString();
    }
    public static void main(String[] args) {
        List<String> parts = List.of("a", "bb", "ccc");
        check(joinSlow(parts).equals(joinFast(parts)), "join");
        check(String.join(",", parts).equals("a,bb,ccc"), "String.join");
        StringJoiner sj = new StringJoiner(",", "[", "]");
        parts.forEach(sj::add);
        check(sj.toString().equals("[a,bb,ccc]"), "joiner");
        // A single + expression is fine: javac 9+ uses invokedynamic and StringConcatFactory (JEP 280).
        String name = "x"; int n = 3;
        check((name + "=" + n).equals("x=3"), "indy concat");
        // Use intern for a small set of long-lived, repeated values; UseStringDeduplication shares backing arrays only.
        String k1 = new String("status-OK").intern(), k2 = "status-OK";
        check(k1 == k2, "intern");
        System.out.println("J05 ok");
    }
    static void check(boolean ok, String what) { if (!ok) throw new AssertionError(what); }
}

// Build strings with reserved capacity; avoid += concatenation in loops.
package k11

fun render(items: List<Pair<String, Int>>): String =
    buildString(items.size * 16) {                    // Estimate capacity to reduce growth.
        for ((k, v) in items) { append(k); append('='); append(v); append('\n') }
    }

fun renderSlow(items: List<Pair<String, Int>>): String {
    var s = ""
    // jvmperf-lint: ignore jp-string-concat-hot -- Intentional counterexample.
    for ((k, v) in items) s += "$k=$v\n"              // Copies the entire string on each iteration.
    return s
}

fun main() {
    val items = List(100) { "k$it" to it }
    check(render(items) == renderSlow(items))
    check(items.joinToString(",", limit = 2) { it.first } == "k0,k1,...")
    println("K11 ok")
}

import java.util.regex.Pattern;

/** Compile and reuse Pattern; String.split has a single-character fast path; replace(CharSequence) is literal. */
public class J15_Regex {
    private static final Pattern WS = Pattern.compile("\\s+");     // Compile once.

    static int countWordsSlow(String[] lines) {
        int n = 0;
        // jvmperf-lint: ignore jp-regex-hot -- Intentional counterexample.
        for (String l : lines) n += l.trim().split("\\s+").length;   // Compiles the regex on each call.
        return n;
    }
    static int countWordsFast(String[] lines) {
        int n = 0;
        for (String l : lines) n += WS.split(l.trim()).length;
        return n;
    }
    public static void main(String[] args) {
        String[] lines = {"a b  c", " d e "};
        if (countWordsSlow(lines) != countWordsFast(lines)) throw new AssertionError();
        // A single non-metacharacter such as "," takes String.split's fast path without regex compilation.
        if ("a,b,c".split(",").length != 3) throw new AssertionError();
        // replace(CharSequence, CharSequence) is literal; replaceAll treats its first argument as a regex.
        if (!"a.b.c".replace(".", "/").equals("a/b/c")) throw new AssertionError();
        System.out.println("J15 ok");
    }
}

// IntArray maps to JVM int[]; Array<Int> and List<Int> store boxed Integer values.
package k01

fun sumIntArray(xs: IntArray): Long {
    var s = 0L
    for (x in xs) s += x            // Compiles to an indexed loop with no iterator or boxing.
    return s
}

// jvmperf-lint: ignore jp-boxed-primitive-collection -- Intentional counterexample.
fun sumList(xs: List<Int>): Long {
    var s = 0L
    for (x in xs) s += x            // Uses an iterator and unboxes each element.
    return s
}

fun main() {
    val n = 1000
    val prim = IntArray(n) { it }                 // int[]
    // jvmperf-lint: ignore jp-boxed-primitive-collection -- Intentional counterexample.
    val boxedArr: Array<Int> = Array(n) { it }    // Integer[]
    // jvmperf-lint: ignore jp-boxed-primitive-collection -- Intentional counterexample.
    val list: List<Int> = List(n) { it }          // ArrayList<Integer>
    check(prim.javaClass.name == "[I")
    check(boxedArr.javaClass.name == "[Ljava.lang.Integer;")
    check(sumIntArray(prim) == sumList(list))
    check(prim.sum() == boxedArr.sum())
    // Range loops also avoid allocation: 0 until n and xs.indices compile to integer loops.
    var s = 0
    for (i in prim.indices) s += prim[i]
    check(s == 499500)
    // (0 until n).map { } creates a boxed List<Int>; use IntArray(n) { } for primitive results.
    val squares = IntArray(n) { it * it }
    check(squares[3] == 9)
    println("K01 ok")
}

// A value class stays unboxed as its own type, but boxes as a generic, nullable, or interface value.
package k02

@JvmInline
value class UserId(val raw: Int)

class Holder(val id: UserId)            // The field is represented as int.
// jvmperf-lint: ignore jp-value-class-boxing -- Intentional counterexample.
class NullableHolder(val id: UserId?)   // Nullable use boxes the field as UserId.

// Use a sentinel instead of null to avoid boxing.
@JvmInline
value class OptUserId(val raw: Int) {
    val isNone: Boolean get() = raw == NONE
    companion object {
        const val NONE = -1
        val None = OptUserId(NONE)
    }
}
class OptHolder(val id: OptUserId)

fun main() {
    check(Holder::class.java.getDeclaredField("id").type == Int::class.javaPrimitiveType)
    check(NullableHolder::class.java.getDeclaredField("id").type == UserId::class.java)
    check(OptHolder::class.java.getDeclaredField("id").type == Int::class.javaPrimitiveType)
    // jvmperf-lint: ignore jp-value-class-boxing -- Intentional counterexample.
    val ids: List<UserId> = listOf(UserId(1), UserId(2))   // Generic elements are boxed UserId objects.
    check((ids[0] as Any).javaClass == UserId::class.java)
    // Store many IDs in IntArray and wrap them as UserId only when needed.
    val raw = intArrayOf(1, 2, 3)
    val first = UserId(raw[0])
    check(first == UserId(1) && OptHolder(OptUserId.None).id.isNone)
    println("K02 ok")
}

// Inline functions, reified types, and lambda allocations.
package k03

// inline expands the lambda at the call site, avoiding a function object and Int boxing.
inline fun repeatSum(n: Int, f: (Int) -> Int): Int {
    var s = 0
    for (i in 0 until n) s += f(i)
    return s
}

// A non-inline function receives a lambda object; Function1.invoke(Object) boxes the Int argument and result.
fun repeatSumBoxed(n: Int, f: (Int) -> Int): Int {
    var s = 0
    for (i in 0 until n) s += f(i)
    return s
}

// A primitive-signature fun interface avoids boxing through generic Function1.
fun interface IntOp { fun apply(x: Int): Int }
fun repeatSumPrim(n: Int, f: IntOp): Int {
    var s = 0
    for (i in 0 until n) s += f.apply(i)
    return s
}

// reified lets an inline function access its type parameter without a Class argument.
inline fun <reified T> List<Any>.countOf(): Int = count { it is T }

fun nonCapturing(): (Int) -> Int = { it + 1 }
fun capturing(k: Int): (Int) -> Int = { it + k }

fun main() {
    val k = 3
    check(repeatSum(10) { it * k } == repeatSumBoxed(10) { it * k })
    check(repeatSumPrim(10) { it * k } == 135)
    check(listOf<Any>(1, "a", 2).countOf<Int>() == 2)
    // Since Kotlin 2.0, lambdas use invokedynamic (LambdaMetafactory) by default.
    // A non-capturing lambda reuses an instance at one call site; a capturing lambda creates a new object each time.
    check(nonCapturing() === nonCapturing())
    check(capturing(1) !== capturing(1))
    println("K03 ok")
}

// Use asSequence() for multi-step chains over large data or with early termination; direct operations suit small collections.
package k04

fun main() {
    val data = List(100_000) { it }
    // Collection version: map creates a 100,000-element list, filter creates another, then take returns five.
    // jvmperf-lint: ignore jp-collection-chain -- Intentional counterexample.
    val eager = data.map { it * 2 }.filter { it % 3 == 0 }.take(5)
    // Sequence version: process one element at a time and stop after five matches, with no intermediate lists.
    var mapped = 0
    val lazy = data.asSequence().map { mapped++; it * 2 }.filter { it % 3 == 0 }.take(5).toList()
    check(eager == lazy)
    check(mapped < 20)              // Only the first few elements were processed.
    // Primitive operations such as IntArray.sumOf/fold are inline and do not box values.
    val arr = IntArray(1000) { it }
    check(arr.sumOf { it.toLong() } == 499500L)
    // Sequence<Int> still boxes elements; use a loop or IntStream for numeric pipelines.
    println("K04 ok")
}

// by lazy defaults to SYNCHRONIZED (with locking and an extra object); use NONE for single-threaded access.
package k06

class Config {
    // Accessed only on the main thread or another single thread:
    val parsed: Map<String, String> by lazy(LazyThreadSafetyMode.NONE) { mapOf("a" to "1") }
    // Multiple threads may access it and initialization is cheap and idempotent: PUBLICATION (may initialize more than once).
    val table: IntArray by lazy(LazyThreadSafetyMode.PUBLICATION) { IntArray(256) { it * it } }
    // Default mode: SYNCHRONIZED.
    // jvmperf-lint: ignore jp-lazy-default-mode -- Intentional counterexample.
    val heavy: String by lazy { "computed" }
}

// Each by lazy property adds a Lazy object and a lambda, which costs memory across many instances.
// For many instances, use a nullable field with a manual check or move the cache elsewhere.
class Row(private val raw: String) {
    private var cachedLen: Int = -1
    val len: Int get() { if (cachedLen < 0) cachedLen = raw.length; return cachedLen }
}

fun main() {
    val c = Config()
    check(c.parsed["a"] == "1" && c.table[3] == 9 && c.heavy == "computed")
    check(Row("abc").len == 3)
    println("K06 ok")
}

// const val and @JvmField change generated getters and companion-object indirection.
package k07

import java.lang.reflect.Modifier

class Limits {
    companion object {
        const val MAX = 1024                      // Compile-time constant, inlined at use sites.
        @JvmField val DEFAULT_NAMES = listOf("a")  // Public static final field on the outer class; no getter.
        val viaGetter = listOf("b")                // Private static field plus Companion.getViaGetter().
    }
}

object Registry {
    @JvmField val items = ArrayList<String>()     // @JvmField lets Java access Registry.items directly.
}

class Point(@JvmField val x: Int, @JvmField val y: Int)   // Java callers access the fields directly.

fun main() {
    val f = Limits::class.java.getDeclaredField("DEFAULT_NAMES")
    check(Modifier.isPublic(f.modifiers) && Modifier.isStatic(f.modifiers))
    check(!Modifier.isPublic(Limits::class.java.getDeclaredField("viaGetter").modifiers))
    check(Limits.MAX == 1024 && Point(1, 2).x == 1)
    Registry.items.add("x")
    println("K07 ok")
}

// Each data class copy() allocates an object; repeated copies in a hot loop create many short-lived objects.
package k08

data class State(val x: Int, val y: Int, val steps: Int)

fun simulateCopy(n: Int): State {
    var s = State(0, 0, 0)
    // jvmperf-lint: ignore jp-data-copy-hot -- Intentional counterexample.
    repeat(n) { s = s.copy(x = s.x + 1, steps = s.steps + 1) }   // n allocations; escape analysis may not remove them.
    return s
}

// On a hot path, accumulate in locals (or a mutable helper) and construct the immutable result once.
fun simulateLocals(n: Int): State {
    var x = 0; var steps = 0
    repeat(n) { x++; steps++ }
    return State(x, 0, steps)
}

fun main() {
    check(simulateCopy(1000) == simulateLocals(1000))
    println("K08 ok")
}

// vararg and the spread operator *: spreading an array makes a copy.
package k09

fun sum(vararg xs: Int): Int { var s = 0; for (x in xs) s += x; return s }
fun identity(vararg xs: Int): IntArray = xs

fun sumArray(xs: IntArray): Int { var s = 0; for (x in xs) s += x; return s }   // Accept arrays directly on hot paths.

fun main() {
    val a = intArrayOf(1, 2, 3)
    check(identity(*a) !== a)            // The spread operator creates a copy.
    check(sum(*a) == sumArray(a))
    check(sum(1, 2, 3) == 6)             // Creates a new int[3] for each call.
    println("K09 ok")
}

// Each suspend function compiles to a state machine and needs a continuation when suspended; launch/async also allocate objects.
// Split small CPU-bound tasks into chunks instead of creating one coroutine per element.
package k10

import kotlinx.coroutines.Dispatchers
import kotlinx.coroutines.async
import kotlinx.coroutines.awaitAll
import kotlinx.coroutines.coroutineScope
import kotlinx.coroutines.runBlocking

suspend fun parallelSum(xs: IntArray, chunks: Int = Runtime.getRuntime().availableProcessors()): Long = coroutineScope {
    val size = (xs.size + chunks - 1) / chunks
    (0 until chunks).map { c ->
        async(Dispatchers.Default) {
            var s = 0L
            val end = minOf(xs.size, (c + 1) * size)
            for (i in c * size until end) s += xs[i]     // A regular loop with no suspension point or allocation inside the chunk.
            s
        }
    }.awaitAll().sum()
}

fun main() = runBlocking {
    val xs = IntArray(1_000_000) { 1 }
    check(parallelSum(xs) == 1_000_000L)
    println("K10 ok")
}

// Primitive collections in androidx.collection 1.4+ work with Kotlin Multiplatform and regular JVM projects.
package k12

import androidx.collection.MutableIntIntMap
import androidx.collection.MutableIntList
import androidx.collection.MutableIntObjectMap
import androidx.collection.SparseArrayCompat
import androidx.collection.mutableIntListOf

fun main() {
    val counts = MutableIntIntMap(64)                 // Open addressing; keys and values are ints.
    for (i in 0 until 1000) counts[i % 10] = counts.getOrDefault(i % 10, 0) + 1
    check(counts[3] == 100)

    val ids: MutableIntList = mutableIntListOf()
    for (i in 0 until 100) ids.add(i)
    var total = 0
    ids.forEach { total += it }                        // Inline; elements are passed as ints.
    check(total == 4950 && ids.size == 100)

    val names = MutableIntObjectMap<String>()
    names[42] = "answer"
    check(names[42] == "answer")

    val sparse = SparseArrayCompat<String>()            // Like android.util.SparseArray, uses sorted arrays and binary search.
    sparse.put(7, "seven")
    check(sparse[7] == "seven")
    println("K12 ok")
}

import java.lang.foreign.Arena;
import java.lang.foreign.MemoryLayout;
import java.lang.foreign.MemorySegment;
import java.lang.foreign.StructLayout;
import java.lang.foreign.ValueLayout;
import java.lang.invoke.VarHandle;
import java.nio.ByteBuffer;
import java.nio.ByteOrder;

/** Off-heap memory with the FFM API (final in JDK 22, JEP 454) and direct ByteBuffer. */
public class J06_OffHeapFfm {
    // struct Car { int id; byte flags; [3 bytes padding] } -- explicit layout with no object header.
    static final StructLayout CAR = MemoryLayout.structLayout(
            ValueLayout.JAVA_INT.withName("id"),
            ValueLayout.JAVA_BYTE.withName("flags"),
            MemoryLayout.paddingLayout(3));
    static final VarHandle ID = CAR.varHandle(MemoryLayout.PathElement.groupElement("id"));
    static final VarHandle FLAGS = CAR.varHandle(MemoryLayout.PathElement.groupElement("flags"));

    public static void main(String[] args) {
        check(CAR.byteSize() == 8, "layout size");
        try (Arena arena = Arena.ofConfined()) {               // Deterministic release when the try block ends.
            MemorySegment cars = arena.allocate(CAR, 1000);    // One contiguous block of 1,000 × 8 bytes.
            for (int i = 0; i < 1000; i++) {
                MemorySegment c = cars.asSlice(i * CAR.byteSize(), CAR);
                ID.set(c, 0L, i);
                FLAGS.set(c, 0L, (byte) 0b011);
            }
            MemorySegment c42 = cars.asSlice(42 * CAR.byteSize(), CAR);
            check((int) ID.get(c42, 0L) == 42 && (byte) FLAGS.get(c42, 0L) == 3, "segment");
            check(cars.byteSize() == 8000, "8000 bytes");
        }
        // Legacy API: GC reclaims direct ByteBuffers through Cleaner; capacity is limited by -XX:MaxDirectMemorySize.
        ByteBuffer bb = ByteBuffer.allocateDirect(8 * 1000).order(ByteOrder.nativeOrder());
        bb.putInt(42 * 8, 42).put(42 * 8 + 4, (byte) 3);
        check(bb.getInt(42 * 8) == 42, "direct buffer");
        System.out.println("J06 ok");
    }
    static void check(boolean ok, String what) { if (!ok) throw new AssertionError(what); }
}

import java.io.BufferedWriter;
import java.io.IOException;
import java.lang.foreign.Arena;
import java.lang.foreign.MemorySegment;
import java.lang.foreign.ValueLayout;
import java.nio.channels.FileChannel;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.StandardOpenOption;

/** I/O: buffered writes, line-by-line reads, and mapping to an unrestricted-size MemorySegment. */
public class J12_Io {
    public static void main(String[] args) throws IOException {
        Path p = Files.createTempFile("jvmperf", ".txt");
        try (BufferedWriter w = Files.newBufferedWriter(p, StandardCharsets.UTF_8)) {  // 8 KiB buffer by default.
            for (int i = 0; i < 1000; i++) { w.write(Integer.toString(i)); w.newLine(); }
        }
        long lines;
        try (var s = Files.lines(p)) { lines = s.count(); }
        if (lines != 1000) throw new AssertionError();

        try (FileChannel ch = FileChannel.open(p, StandardOpenOption.READ); Arena arena = Arena.ofConfined()) {
            MemorySegment m = ch.map(FileChannel.MapMode.READ_ONLY, 0, ch.size(), arena);  // JDK 22+
            long newlines = 0;
            for (long i = 0; i < m.byteSize(); i++) if (m.get(ValueLayout.JAVA_BYTE, i) == '\n') newlines++;
            if (newlines != 1000) throw new AssertionError();
        }
        Files.delete(p);
        System.out.println("J12 ok");
    }
}

/**
 * Escape analysis and scalar replacement: C2 may avoid heap allocation for small objects that do not escape.
 * Verify with -XX:-EliminateAllocations or JMH -prof gc; inspect gc.alloc.rate.norm.
 */
public class J08_EscapeAnalysis {
    record Vec(double x, double y) {
        Vec plus(Vec o) { return new Vec(x + o.x, y + o.y); }
        double len2() { return x * x + y * y; }
    }
    // Vec does not escape and its calls can inline, so C2 may replace it with two local doubles.
    static double sumLen2(double[] xs, double[] ys) {
        double s = 0;
        for (int i = 0; i < xs.length; i++) s += new Vec(xs[i], ys[i]).plus(new Vec(1, 1)).len2();
        return s;
    }
    static Vec last;                          // Storing into a static field makes the object escape.
    static void escapes(double x) { last = new Vec(x, x); }

    public static void main(String[] args) {
        double[] xs = {0, 1, 2}, ys = {0, 1, 2};
        if (sumLen2(xs, ys) != 2 + 8 + 18) throw new AssertionError();
        escapes(1);
        System.out.println("J08 ok");
    }
}

/**
 * JIT-friendly code: sealed types, switch pattern matching, small methods, and intrinsics.
 * Inspect inlining with: java -XX:+UnlockDiagnosticVMOptions -XX:+PrintInlining
 */
public class J09_JitFriendly {
    sealed interface Shape permits Circle, Rect {}
    record Circle(double r) implements Shape {}
    record Rect(double w, double h) implements Shape {}

    static double area(Shape s) {
        return switch (s) {                       // Exhaustive switch; no virtual call.
            case Circle c -> Math.PI * c.r() * c.r();
            case Rect r -> r.w() * r.h();
        };
    }
    public static void main(String[] args) {
        Shape[] shapes = {new Circle(1), new Rect(2, 3)};
        double total = 0;
        for (Shape s : shapes) total += area(s);
        if (Math.abs(total - (Math.PI + 6)) > 1e-9) throw new AssertionError();
        // These HotSpot intrinsics compile to a small number of specialized instructions.
        if (Integer.bitCount(0b1011) != 3 || Long.numberOfTrailingZeros(8) != 3) throw new AssertionError();
        if (Math.fma(2.0, 3.0, 1.0) != 7.0) throw new AssertionError();
        if (java.util.Arrays.mismatch(new byte[]{1, 2, 3}, new byte[]{1, 2, 4}) != 2) throw new AssertionError();
        System.out.println("J09 ok");
    }
}

import jdk.incubator.vector.FloatVector;
import jdk.incubator.vector.VectorOperators;
import jdk.incubator.vector.VectorSpecies;

/**
 * Vector API (JDK 25, tenth incubator round, JEP 508). Compile and run with --add-modules jdk.incubator.vector.
 * javac --add-modules jdk.incubator.vector J07_VectorApi.java
 */
public class J07_VectorApi {
    static final VectorSpecies<Float> S = FloatVector.SPECIES_PREFERRED;   // Select a width supported by the CPU.

    static float dot(float[] a, float[] b) {
        var acc = FloatVector.zero(S);
        int i = 0, bound = S.loopBound(a.length);
        for (; i < bound; i += S.length()) {
            var va = FloatVector.fromArray(S, a, i);
            var vb = FloatVector.fromArray(S, b, i);
            acc = va.fma(vb, acc);
        }
        float sum = acc.reduceLanes(VectorOperators.ADD);
        for (; i < a.length; i++) sum += a[i] * b[i];                    // Handle the tail with scalar operations.
        return sum;
    }
    public static void main(String[] args) {
        float[] a = new float[1003], b = new float[1003];
        for (int i = 0; i < a.length; i++) { a[i] = 1; b[i] = 2; }
        if (dot(a, b) != 2006f) throw new AssertionError(dot(a, b));
        System.out.println("J07 ok, lanes=" + S.length());
    }
}

/** Do not use exceptions for control flow; frequent exceptions can disable writable stack traces. */
public class J14_Exceptions {
    static final class ParseFailure extends RuntimeException {
        private static final long serialVersionUID = 1L;
        ParseFailure(String msg) { super(msg, null, false, false); }   // Skips stack capture; cheaper but harder to debug.
    }
    // Bad: use an exception to check whether a value is numeric.
    static boolean isIntSlow(String s) {
        try { Integer.parseInt(s); return true; } catch (NumberFormatException e) { return false; }
    }
    // Better: validate before parsing.
    static boolean isIntFast(String s) {
        if (s.isEmpty() || s.length() > 11) return false;          // "-2147483648" is 11 characters.
        int i = (s.charAt(0) == '-' || s.charAt(0) == '+') ? 1 : 0;
        if (i == s.length()) return false;
        for (; i < s.length(); i++) if (!Character.isDigit(s.charAt(i))) return false;
        try { Integer.parseInt(s); return true; } catch (NumberFormatException e) { return false; } // Only rare overflow cases reach the exception.
    }
    public static void main(String[] args) {
        for (String s : new String[]{"12", "-3", "x", "", "-", "99999999999", "-2147483648", "+5", "2147483648"})
            if (isIntSlow(s) != isIntFast(s)) throw new AssertionError(s);
        ParseFailure f = new ParseFailure("bad");
        if (f.getStackTrace().length != 0) throw new AssertionError();
        System.out.println("J14 ok");
    }
}

import java.util.concurrent.atomic.LongAdder;

/**
 * False sharing occurs when fields frequently written by different threads share a cache line.
 * - The JDK annotation @jdk.internal.vm.annotation.Contended requires the compile-time --add-exports flag
 *   and the runtime option -XX:-RestrictContended in user code.
 * - LongAdder is a ready-made option; its internal Cell counters use @Contended.
 * - Manual padding avoids JVM flags, but field layout is not guaranteed.
 */
public class J10_FalseSharing {
    static final class PaddedCounter {
        @jdk.internal.vm.annotation.Contended volatile long a;
        @jdk.internal.vm.annotation.Contended volatile long b;
    }
    public static void main(String[] args) throws Exception {
        LongAdder adder = new LongAdder();
        Thread[] ts = new Thread[4];
        for (int t = 0; t < ts.length; t++) {
            ts[t] = Thread.ofPlatform().start(() -> { for (int i = 0; i < 100_000; i++) adder.increment(); });
        }
        for (Thread t : ts) t.join();
        if (adder.sum() != 400_000) throw new AssertionError();
        PaddedCounter pc = new PaddedCounter();
        pc.a = 1; pc.b = 2;
        System.out.println("J10 ok");
    }
}

import java.util.concurrent.Executors;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.stream.IntStream;

/** Virtual threads (JDK 21, JEP 444) suit many blocking I/O tasks, not CPU-bound work. */
public class J11_VirtualThreads {
    public static void main(String[] args) {
        AtomicInteger done = new AtomicInteger();
        try (var exec = Executors.newVirtualThreadPerTaskExecutor()) {   // One virtual thread per task; do not pool them.
            IntStream.range(0, 10_000).forEach(i -> exec.submit(() -> {
                try { Thread.sleep(10); } catch (InterruptedException e) { Thread.currentThread().interrupt(); }
                done.incrementAndGet();
            }));
        } // close() waits for all tasks to finish.
        if (done.get() != 10_000) throw new AssertionError();
        System.out.println("J11 ok");
    }
}

import java.nio.ByteBuffer;
import java.util.ArrayDeque;

/**
 * Pool only expensive resources such as direct ByteBuffers, large arrays, connections, and threads.
 * Pooling small objects is often slower: TLAB allocation is cheap, while pools add synchronization and leak risks.
 */
public class J16_Pool {
    static final class BufferPool {                       // Single-threaded; use one per thread or add locking for shared use.
        private final ArrayDeque<ByteBuffer> free = new ArrayDeque<>();
        private final int size;
        BufferPool(int size, int prealloc) {
            this.size = size;
            // jvmperf-lint: ignore jp-expensive-alloc-hot -- Pool is preallocated once during construction.
            for (int i = 0; i < prealloc; i++) free.push(ByteBuffer.allocateDirect(size));
        }
        ByteBuffer acquire() { ByteBuffer b = free.poll(); return b != null ? b.clear() : ByteBuffer.allocateDirect(size); }
        void release(ByteBuffer b) { free.push(b); }
    }
    public static void main(String[] args) {
        BufferPool pool = new BufferPool(64 * 1024, 4);
        ByteBuffer b = pool.acquire();
        b.putLong(42L);
        pool.release(b);
        if (pool.acquire() != b) throw new AssertionError();   // Reuses the same off-heap buffer.
        System.out.println("J16 ok");
    }
}