Playground/1brc
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

final version for abeobk (#654)

Browse Source 
* final version

* Correct stupid mistake

* min/max trick does not help that much, setting initial value does.

* cut the tail
This commit is contained in:
Van Phu DO 2024-02-01 04:03:20 +09:00 committed by GitHub
parent f0f6570975
commit 3c454d0222
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 393 additions and 355 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
prepare_abeobk.sh
View File

@ -20,6 +20,6 @@ sdk use java 21.0.2-graal 1>&2
# ./mvnw clean verify removes target/ and will re-trigger native image creation.
if [ ! -f target/CalculateAverage_abeobk_image ]; then
NATIVE_IMAGE_OPTS="--gc=epsilon -O3 -dsa -march=native -H:InlineAllBonus=10 -H:-GenLoopSafepoints -H:-ParseRuntimeOptions --enable-preview --initialize-at-build-time=dev.morling.onebrc.CalculateAverage_abeobk"
NATIVE_IMAGE_OPTS="--gc=epsilon -O3 -march=native -H:InlineAllBonus=10 -H:-GenLoopSafepoints --enable-preview --initialize-at-build-time=dev.morling.onebrc.CalculateAverage_abeobk"
native-image $NATIVE_IMAGE_OPTS -cp target/average-1.0.0-SNAPSHOT.jar -o target/CalculateAverage_abeobk_image dev.morling.onebrc.CalculateAverage_abeobk
fi

746
src/main/java/dev/morling/onebrc/CalculateAverage_abeobk.java
View File

@ -34,7 +34,6 @@ import java.util.stream.IntStream;
import sun.misc.Unsafe;
public class CalculateAverage_abeobk {
private static final boolean SHOW_ANALYSIS = false;
private static final int CPU_CNT = Runtime.getRuntime().availableProcessors();
private static final String FILE = "./measurements.txt";
@ -42,7 +41,7 @@ public class CalculateAverage_abeobk {
private static final long BUCKET_MASK = BUCKET_SIZE - 1;
private static final int MAX_STR_LEN = 100;
private static final int MAX_STATIONS = 10000;
private static final long CHUNK_SZ = 1 << 22; // 4MB chunk
private static final long CHUNK_SZ = 1 << 22;
private static final Unsafe UNSAFE = initUnsafe();
private static final long[] HASH_MASKS = new long[]{
0x0L,
@ -60,10 +59,6 @@ public class CalculateAverage_abeobk {
private static int chunk_cnt;
private static long start_addr, end_addr;
private static final void debug(String s, Object... args) {
System.out.println(String.format(s, args));
}
private static Unsafe initUnsafe() {
try {
Field theUnsafe = Unsafe.class.getDeclaredField("theUnsafe");
@ -75,354 +70,9 @@ public class CalculateAverage_abeobk {
}
}
// use native type, less conversion
static class Node {
long addr;
long hash;
long word0;
long tail;
long sum;
long min, max;
int keylen;
int count;
public final String toString() {
return (min / 10.0) + "/"
+ (Math.round(((double) sum / count)) / 10.0) + "/"
+ (max / 10.0);
}
final String key() {
byte[] sbuf = new byte[MAX_STR_LEN];
UNSAFE.copyMemory(null, addr, sbuf, Unsafe.ARRAY_BYTE_BASE_OFFSET, keylen);
return new String(sbuf, 0, (int) keylen, StandardCharsets.UTF_8);
}
Node(long a, long t, int kl, long h) {
addr = a;
tail = t;
min = 999;
max = -999;
keylen = kl;
hash = h;
}
Node(long a, long w0, long t, int kl, long h) {
addr = a;
word0 = w0;
min = 999;
max = -999;
tail = t;
keylen = kl;
hash = h;
}
final void add(long val) {
sum += val;
count++;
if (val > max) {
max = val;
}
if (val < min) {
min = val;
}
}
final void merge(Node other) {
sum += other.sum;
count += other.count;
if (other.max > max) {
max = other.max;
}
if (other.min < min) {
min = other.min;
}
}
final boolean contentEquals(long other_addr, long other_word0, long other_tail, long kl) {
if (word0 != other_word0 || tail != other_tail)
return false;
// this is faster than comparision if key is short
long xsum = 0;
long n = kl & 0xF8;
for (long i = 8; i < n; i += 8) {
xsum |= (UNSAFE.getLong(addr + i) ^ UNSAFE.getLong(other_addr + i));
}
return xsum == 0;
}
final boolean contentEquals(Node other) {
if (tail != other.tail)
return false;
long n = keylen & 0xF8;
for (long i = 0; i < n; i += 8) {
if (UNSAFE.getLong(addr + i) != UNSAFE.getLong(other.addr + i))
return false;
}
return true;
}
}
// idea from royvanrijn
static final long getSemiPosCode(final long word) {
long xor_semi = word ^ 0x3b3b3b3b3b3b3b3bL; // xor with ;;;;;;;;
return (xor_semi - 0x0101010101010101L) & (~xor_semi & 0x8080808080808080L);
}
static final long getLFCode(final long word) {
long xor_semi = word ^ 0x0A0A0A0A0A0A0A0AL; // xor with \n\n\n\n\n\n\n\n
return (xor_semi - 0x0101010101010101L) & (~xor_semi & 0x8080808080808080L);
}
static final long nextLine(long addr) {
long word = UNSAFE.getLong(addr);
long lfpos_code = getLFCode(word);
while (lfpos_code == 0) {
addr += 8;
word = UNSAFE.getLong(addr);
lfpos_code = getLFCode(word);
}
return addr + (Long.numberOfTrailingZeros(lfpos_code) >>> 3) + 1;
}
// speed/collision balance
static final long xxh32(long hash) {
long h = hash * 37;
return (h ^ (h >>> 29));
}
static final class ChunkParser {
long addr;
long end;
Node[] map;
ChunkParser(Node[] m, long a, long e) {
map = m;
addr = a;
end = e;
}
final boolean ok() {
return addr < end;
}
final long word() {
return UNSAFE.getLong(addr);
}
final long val() {
long num_word = UNSAFE.getLong(addr);
int dot_pos = Long.numberOfTrailingZeros(~num_word & 0x10101000);
addr += (dot_pos >>> 3) + 3;
// great idea from merykitty (Quan Anh Mai)
int shift = 28 - dot_pos;
long signed = (~num_word << 59) >> 63;
long dsmask = ~(signed & 0xFF);
long digits = ((num_word & dsmask) << shift) & 0x0F000F0F00L;
long abs_val = ((digits * 0x640a0001) >>> 32) & 0x3FF;
return ((abs_val ^ signed) - signed);
}
// optimize for contest
// save as much slow memory access as possible
// about 50% key < 8chars, 25% key bettween 8-10 chars
// keylength histogram (%) = [0, 0, 0, 0, 4, 10, 21, 15, 13, 11, 6, 6, 4, 2...
final Node key(long word0, long semipos_code) {
long row_addr = addr;
// about 50% chance key < 8 chars
if (semipos_code != 0) {
int semi_pos = Long.numberOfTrailingZeros(semipos_code) >>> 3;
addr += semi_pos + 1;
long tail = word0 & HASH_MASKS[semi_pos];
long hash = xxh32(tail);
int bucket = (int) (hash & BUCKET_MASK);
while (true) {
Node node = map[bucket];
if (node == null) {
return (map[bucket] = new Node(row_addr, tail, semi_pos, hash));
}
if (node.tail == tail) {
return node;
}
bucket++;
}
}
addr += 8;
long word = UNSAFE.getLong(addr);
semipos_code = getSemiPosCode(word);
// 43% chance
if (semipos_code != 0) {
int semi_pos = Long.numberOfTrailingZeros(semipos_code) >>> 3;
addr += semi_pos + 1;
long tail = (word & HASH_MASKS[semi_pos]);
long hash = xxh32(word0 ^ tail);
int bucket = (int) (hash & BUCKET_MASK);
while (true) {
Node node = map[bucket];
if (node == null) {
return (map[bucket] = new Node(row_addr, word0, tail, semi_pos + 8, hash));
}
if (node.word0 == word0 && node.tail == tail) {
return node;
}
bucket++;
}
}
// why not going for more? tested, slower
long hash = word0;
while (semipos_code == 0) {
hash ^= word;
addr += 8;
word = UNSAFE.getLong(addr);
semipos_code = getSemiPosCode(word);
}
int semi_pos = Long.numberOfTrailingZeros(semipos_code) >>> 3;
addr += semi_pos;
long keylen = addr - row_addr;
addr++;
long tail = (word & HASH_MASKS[semi_pos]);
hash = xxh32(hash ^ tail);
int bucket = (int) (hash & BUCKET_MASK);
while (true) {
Node node = map[bucket];
if (node == null) {
return (map[bucket] = new Node(row_addr, word0, tail, (int) keylen, hash));
}
if (node.contentEquals(row_addr, word0, tail, keylen)) {
return node;
}
bucket++;
}
}
}
// Thread pool worker
static final class Worker extends Thread {
final int thread_id; // for debug use only
int cls = 0;
Worker(int i) {
thread_id = i;
this.start();
}
@Override
public void run() {
var map = new Node[BUCKET_SIZE + MAX_STATIONS]; // extra space for collisions
int id;
// process in small chunk to maintain disk locality (artsiomkorzun trick)
while ((id = chunk_id.getAndIncrement()) < chunk_cnt) {
long addr = start_addr + id * CHUNK_SZ;
long end = Math.min(addr + CHUNK_SZ, end_addr);
// find start of line
if (id > 0) {
addr = nextLine(addr);
}
final int num_segs = 3;
long seglen = (end - addr) / num_segs;
long a0 = addr;
long a1 = nextLine(addr + 1 * seglen);
long a2 = nextLine(addr + 2 * seglen);
ChunkParser p0 = new ChunkParser(map, a0, a1);
ChunkParser p1 = new ChunkParser(map, a1, a2);
ChunkParser p2 = new ChunkParser(map, a2, end);
while (p0.ok() && p1.ok() && p2.ok()) {
long w0 = p0.word();
long w1 = p1.word();
long w2 = p2.word();
long sc0 = getSemiPosCode(w0);
long sc1 = getSemiPosCode(w1);
long sc2 = getSemiPosCode(w2);
Node n0 = p0.key(w0, sc0);
Node n1 = p1.key(w1, sc1);
Node n2 = p2.key(w2, sc2);
long v0 = p0.val();
long v1 = p1.val();
long v2 = p2.val();
n0.add(v0);
n1.add(v1);
n2.add(v2);
}
while (p0.ok()) {
long w = p0.word();
long sc = getSemiPosCode(w);
Node n = p0.key(w, sc);
long v = p0.val();
n.add(v);
}
while (p1.ok()) {
long w = p1.word();
long sc = getSemiPosCode(w);
Node n = p1.key(w, sc);
long v = p1.val();
n.add(v);
}
while (p2.ok()) {
long w = p2.word();
long sc = getSemiPosCode(w);
Node n = p2.key(w, sc);
long v = p2.val();
n.add(v);
}
}
// merge is cheaper than string casting (artsiomkorzun)
while (!mapref.compareAndSet(null, map)) {
var other_map = mapref.getAndSet(null);
if (other_map != null) {
for (int i = 0; i < other_map.length; i++) {
var other = other_map[i];
if (other == null)
continue;
int bucket = (int) (other.hash & BUCKET_MASK);
while (true) {
var node = map[bucket];
if (node == null) {
map[bucket] = other;
break;
}
if (node.contentEquals(other)) {
node.merge(other);
break;
}
bucket++;
if (SHOW_ANALYSIS)
cls++;
}
}
}
}
if (SHOW_ANALYSIS) {
debug("Thread %d collision = %d", thread_id, cls);
}
}
}
// thomaswue trick
private static void spawnWorker() throws IOException {
ProcessHandle.Info info = ProcessHandle.current().info();
ArrayList<String> workerCommand = new ArrayList<>();
info.command().ifPresent(workerCommand::add);
info.arguments().ifPresent(args -> workerCommand.addAll(Arrays.asList(args)));
workerCommand.add("--worker");
new ProcessBuilder()
.command(workerCommand)
.start()
.getInputStream()
.transferTo(System.out);
}
/*
* MAIN FUNCTION
*/
public static void main(String[] args) throws InterruptedException, IOException {
// thomaswue trick
if (args.length == 0 || !("--worker".equals(args[0]))) {
@ -457,4 +107,392 @@ public class CalculateAverage_abeobk {
System.out.println(ms);
System.out.close();
}
/*
* HELPER FUNCTIONS
*/
// Get semicolon pos code
static final long getSemiCode(final long w) {
long x = w ^ 0x3b3b3b3b3b3b3b3bL; // xor with ;;;;;;;;
return (x - 0x0101010101010101L) & (~x & 0x8080808080808080L);
}
// Get new line pos code
static final long getLFCode(final long w) {
long x = w ^ 0x0A0A0A0A0A0A0A0AL; // xor with \n\n\n\n\n\n\n\n
return (x - 0x0101010101010101L) & (~x & 0x8080808080808080L);
}
// Get decimal point pos code
static final int getDotCode(final long w) {
return Long.numberOfTrailingZeros(~w & 0x10101000);
}
// Convert semicolon pos code to position
static final int getSemiPos(final long spc) {
return Long.numberOfTrailingZeros(spc) >>> 3;
}
// Find next line address
static final long nextLF(long addr) {
long word = UNSAFE.getLong(addr);
long lfpos_code = getLFCode(word);
while (lfpos_code == 0) {
addr += 8;
word = UNSAFE.getLong(addr);
lfpos_code = getLFCode(word);
}
return addr + (Long.numberOfTrailingZeros(lfpos_code) >>> 3) + 1;
}
// Parse number
// great idea from merykitty (Quan Anh Mai)
static final long num(long w, int d) {
int shift = 28 - d;
long signed = (~w << 59) >> 63;
long dsmask = ~(signed & 0xFF);
long digits = ((w & dsmask) << shift) & 0x0F000F0F00L;
long abs_val = ((digits * 0x640a0001) >>> 32) & 0x3FF;
return ((abs_val ^ signed) - signed);
}
// Hash mixer
static final long mix(long hash) {
long h = hash * 37;
return (h ^ (h >>> 29));
}
// Spawn worker (thomaswue trick
private static void spawnWorker() throws IOException {
ProcessHandle.Info info = ProcessHandle.current().info();
ArrayList<String> workerCommand = new ArrayList<>();
info.command().ifPresent(workerCommand::add);
info.arguments().ifPresent(args -> workerCommand.addAll(Arrays.asList(args)));
workerCommand.add("--worker");
new ProcessBuilder()
.command(workerCommand)
.start()
.getInputStream()
.transferTo(System.out);
}
final static class Node {
long addr;
long hash;
long word0;
long sum;
long min, max;
int keylen;
int count;
public final String toString() {
return (min / 10.0) + "/"
+ (Math.round(((double) sum / count)) / 10.0) + "/"
+ (max / 10.0);
}
final String key() {
byte[] sbuf = new byte[MAX_STR_LEN];
UNSAFE.copyMemory(null, addr, sbuf, Unsafe.ARRAY_BYTE_BASE_OFFSET, keylen);
return new String(sbuf, 0, (int) keylen, StandardCharsets.UTF_8);
}
Node(long a, long h, int kl, long v) {
addr = a;
min = max = v;
keylen = kl;
hash = h;
}
Node(long a, long h, int kl) {
addr = a;
hash = h;
min = 999;
max = -999;
keylen = kl;
}
Node(long a, long w0, long h, int kl, long v) {
addr = a;
word0 = w0;
hash = h;
min = max = v;
keylen = kl;
}
Node(long a, long w0, long h, int kl) {
addr = a;
word0 = w0;
hash = h;
min = 999;
max = -999;
keylen = kl;
}
final void add(long val) {
sum += val;
count++;
if (val > max) {
max = val;
}
if (val < min) {
min = val;
}
}
final void merge(Node other) {
sum += other.sum;
count += other.count;
if (other.max > max) {
max = other.max;
}
if (other.min < min) {
min = other.min;
}
}
final boolean contentEquals(long other_addr, long other_word0, long other_hash, long kl) {
if (word0 != other_word0 || hash != other_hash)
return false;
// this is faster than comparision if key is short
long xsum = 0;
long n = kl & 0xF8;
for (long i = 8; i < n; i += 8) {
xsum |= (UNSAFE.getLong(addr + i) ^ UNSAFE.getLong(other_addr + i));
}
return xsum == 0;
}
final boolean contentEquals(Node other) {
if (hash != other.hash)
return false;
long n = keylen & 0xF8;
for (long i = 0; i < n; i += 8) {
if (UNSAFE.getLong(addr + i) != UNSAFE.getLong(other.addr + i))
return false;
}
return true;
}
}
// Thread pool worker
static final class Worker extends Thread {
final int thread_id; // for debug use only
Worker(int i) {
thread_id = i;
this.setPriority(Thread.MAX_PRIORITY);
this.start();
}
@Override
public void run() {
var map = new Node[BUCKET_SIZE + MAX_STATIONS]; // extra space for collisions
int id;
// process in small chunk to maintain disk locality (artsiomkorzun trick)
while ((id = chunk_id.getAndIncrement()) < chunk_cnt) {
long addr = start_addr + id * CHUNK_SZ;
long end = Math.min(addr + CHUNK_SZ, end_addr);
// find start of line
if (id > 0) {
addr = nextLF(addr);
}
final int num_segs = 3;
long seglen = (end - addr) / num_segs;
long a0 = addr;
long a1 = nextLF(addr + 1 * seglen);
long a2 = nextLF(addr + 2 * seglen);
ChunkParser p0 = new ChunkParser(map, a0, a1);
ChunkParser p1 = new ChunkParser(map, a1, a2);
ChunkParser p2 = new ChunkParser(map, a2, end);
while (p0.ok() && p1.ok() && p2.ok()) {
long w0 = p0.word();
long w1 = p1.word();
long w2 = p2.word();
long sc0 = getSemiCode(w0);
long sc1 = getSemiCode(w1);
long sc2 = getSemiCode(w2);
Node n0 = p0.key(w0, sc0);
Node n1 = p1.key(w1, sc1);
Node n2 = p2.key(w2, sc2);
long v0 = p0.val();
long v1 = p1.val();
long v2 = p2.val();
n0.add(v0);
n1.add(v1);
n2.add(v2);
}
while (p0.ok()) {
long w = p0.word();
long sc = getSemiCode(w);
Node n = p0.key(w, sc);
long v = p0.val();
n.add(v);
}
while (p1.ok()) {
long w = p1.word();
long sc = getSemiCode(w);
Node n = p1.key(w, sc);
long v = p1.val();
n.add(v);
}
while (p2.ok()) {
long w = p2.word();
long sc = getSemiCode(w);
Node n = p2.key(w, sc);
long v = p2.val();
n.add(v);
}
}
// merge is cheaper than string casting (artsiomkorzun)
while (!mapref.compareAndSet(null, map)) {
var other_map = mapref.getAndSet(null);
if (other_map != null) {
for (int i = 0; i < other_map.length; i++) {
var other = other_map[i];
if (other == null)
continue;
int bucket = (int) (other.hash & BUCKET_MASK);
while (true) {
var node = map[bucket];
if (node == null) {
map[bucket] = other;
break;
}
if (node.contentEquals(other)) {
node.merge(other);
break;
}
bucket++;
}
}
}
}
}
}
static final class ChunkParser {
long addr;
long end;
Node[] map;
ChunkParser(Node[] m, long a, long e) {
map = m;
addr = a;
end = e;
}
final boolean ok() {
return addr < end;
}
final long word() {
return UNSAFE.getLong(addr);
}
final void skip(int n) {
addr += n;
}
final void skip(long n) {
addr += n;
}
final long val0() {
long w = word();
int d = getDotCode(w);
return num(w, d);
}
final long val() {
long w = word();
int d = getDotCode(w);
skip((d >>> 3) + 3);
return num(w, d);
}
// optimize for contest
// save as much slow memory access as possible
// about 50% key < 8chars, 25% key bettween 8-10 chars
// keylength histogram (%) = [0, 0, 0, 0, 4, 10, 21, 15, 13, 11, 6, 6, 4, 2...
final Node key(long word0, long semipos_code) {
long row_addr = addr;
// about 50% chance key < 8 chars
if (semipos_code != 0) {
int semi_pos = Long.numberOfTrailingZeros(semipos_code) >>> 3;
skip(semi_pos + 1);
long tail = word0 & HASH_MASKS[semi_pos];
long hash = mix(tail);
int bucket = (int) (hash & BUCKET_MASK);
while (true) {
Node node = map[bucket];
if (node == null) {
return (map[bucket] = new Node(row_addr, hash, semi_pos));
}
if (node.hash == hash) {
return node;
}
bucket++;
}
}
skip(8);
long word = UNSAFE.getLong(addr);
semipos_code = getSemiCode(word);
// 43% chance
if (semipos_code != 0) {
int semi_pos = Long.numberOfTrailingZeros(semipos_code) >>> 3;
skip(semi_pos + 1);
long tail = word0 ^ (word & HASH_MASKS[semi_pos]);
long hash = mix(tail);
int bucket = (int) (hash & BUCKET_MASK);
while (true) {
Node node = map[bucket];
if (node == null) {
return (map[bucket] = new Node(row_addr, word0, hash, semi_pos + 8));
}
if (node.word0 == word0 && node.hash == hash) {
return node;
}
bucket++;
}
}
// why not going for more? tested, slower
long hash = word0;
while (semipos_code == 0) {
hash ^= word;
skip(8);
word = UNSAFE.getLong(addr);
semipos_code = getSemiCode(word);
}
int semi_pos = Long.numberOfTrailingZeros(semipos_code) >>> 3;
skip(semi_pos);
long keylen = addr - row_addr;
skip(1);
long tail = hash ^ (word & HASH_MASKS[semi_pos]);
hash = mix(tail);
int bucket = (int) (hash & BUCKET_MASK);
while (true) {
Node node = map[bucket];
if (node == null) {
return (map[bucket] = new Node(row_addr, word0, hash, (int) keylen));
}
if (node.contentEquals(row_addr, word0, hash, keylen)) {
return node;
}
bucket++;
}
}
}
}