introduce bitloops

This commit is contained in:
mike
2026-01-11 21:06:09 +01:00
parent 9b9a6e3550
commit 85e6f32b3c
5 changed files with 308 additions and 87 deletions

View File

@@ -0,0 +1,244 @@
package puzzle;
import com.google.gson.Gson;
import puzzle.SwedishGenerator.Lemma;
import java.io.*;
import java.nio.ByteBuffer;
import java.nio.channels.FileChannel;
import java.nio.charset.StandardCharsets;
import java.nio.file.*;
import java.util.Arrays;
import java.util.Locale;
import java.util.function.Consumer;
public final class CsvIndexService
implements Closeable {
static final ScopedValue<CsvIndexService> SC = ScopedValue.newInstance();
static final Gson GSON = new Gson();
private static final int MAGIC = 0x4C494458; // "LIDX"
private static final int VERSION = 1;
private final Path csvPath;
private final Path idxPath;
private volatile long[] offsets; // lazy
private volatile FileChannel csvChannel; // open once
private final Object lock = new Object();
public CsvIndexService(Path csvPath, Path idxPath) {
this.csvPath = csvPath;
this.idxPath = idxPath;
}
public static String[] lineToClue(String line) {
if (line.isBlank()) throw new RuntimeException("Empty line");
var parts = line.split(",", 5);
var rawClue = parts[4].trim();
if (rawClue.startsWith("\"") && rawClue.endsWith("\"")) {
rawClue = rawClue.substring(1, rawClue.length() - 1).replace("\"\"", "\"");
}
return GSON.fromJson(rawClue, String[].class);
}
public static void lineToLemma(String line, Consumer<Lemma> ok) {
if (line.isBlank()) {
throw new RuntimeException("Empty line");
}
var parts = line.split(",", 5);
var id = Integer.parseInt(parts[0].trim());
var word = parts[1].trim();
if (!word.matches("^[A-Z]{2,8}$")) {
throw new RuntimeException("Invalid word:" + line);
}
// CSV has level 1-10. llmScores use 10-level.
int score = Integer.parseInt(parts[2].trim());
if (score < 1) {
if (Main.VERBOSE) System.err.println("Word too complex: " + line);
return;
}
int simpel = Integer.parseInt(parts[3].trim());
ok.accept(new Lemma(id, word, simpel));
}
public static String[] clues(int index) {
try {
if (SC.isBound())
return lineToClue(SC.get().getLine(index));
return new String[0];
} catch (Exception e) {
throw new RuntimeException("Failed to get clues for index " + index, e);
}
}
/** Haal één regel op (0-based line index), met self-healing index (1x rebuild). */
public String getLine(int lineIndex) throws IOException {
ensureLoaded();
var line = readLineAt(lineIndex);
if (startsWithIndex(line, lineIndex)) return line;
// mismatch => rebuild index en nog 1x proberen
synchronized (lock) {
rebuildIndexLocked();
line = readLineAt(lineIndex);
if (startsWithIndex(line, lineIndex)) return line;
}
throw new RuntimeException("Index mismatch after rebuild. Requested=" + lineIndex + ", got line=" + preview(line));
}
private void ensureLoaded() throws IOException {
if (offsets != null && csvChannel != null && csvChannel.isOpen()) return;
synchronized (lock) {
if (offsets != null && csvChannel != null && csvChannel.isOpen()) return;
csvChannel = FileChannel.open(csvPath, StandardOpenOption.READ);
if (Files.exists(idxPath)) {
try {
offsets = readIndex(idxPath);
return;
} catch (IOException badIndex) {
// fall-through -> rebuild
}
}
rebuildIndexLocked();
}
}
private void rebuildIndexLocked() throws IOException {
var built = buildOffsets(csvPath);
writeIndex(idxPath, built);
offsets = built;
}
private String readLineAt(int lineIndex) throws IOException {
var local = offsets;
if (lineIndex < 0 || lineIndex >= local.length) {
throw new IndexOutOfBoundsException("lineIndex=" + lineIndex + ", max=" + (local.length - 1));
}
var start = local[lineIndex];
csvChannel.position(start);
// lees in blokjes (sneller dan 1 byte) tot newline
var buf = new byte[8192];
var total = 0;
var out = new byte[256];
while (true) {
var bb = ByteBuffer.wrap(buf);
var n = csvChannel.read(bb);
if (n < 0) break; // EOF
var end = n;
for (var i = 0; i < end; i++) {
var b = buf[i];
if (b == (byte) '\n') {
// reposition kanaal op byte na newline
long back = (end - i - 1);
csvChannel.position(csvChannel.position() - back);
return new String(out, 0, total, StandardCharsets.UTF_8);
}
if (b == (byte) '\r') continue;
if (total == out.length) out = Arrays.copyOf(out, out.length * 2);
out[total++] = b;
}
}
return new String(out, 0, total, StandardCharsets.UTF_8);
}
/** Check: begint de regel met "<lineIndex>," */
private static boolean startsWithIndex(String line, int lineIndex) {
if (line == null || line.isEmpty()) return false;
var comma = line.indexOf(',');
if (comma <= 0) return false;
// snelle parse zonder split
long v = 0;
for (var i = 0; i < comma; i++) {
var c = line.charAt(i);
if (c < '0' || c > '9') return false;
v = (v * 10) + (c - '0');
if (v > Integer.MAX_VALUE) return false;
}
return v == lineIndex;
}
private static String preview(String s) {
if (s == null) return "null";
return s.length() <= 120 ? s : s.substring(0, 120) + "...";
}
/** Bouw offsets door newlines te scannen. Resultaat is exact getrimd. */
public static long[] buildOffsets(Path path) throws IOException {
try (var ch = FileChannel.open(path, StandardOpenOption.READ)) {
var offs = new long[131072]; // start-capacity, groeit indien nodig
var c = 0;
offs[c++] = 0L;
var buf = ByteBuffer.allocateDirect(1 << 20);
long pos = 0;
while (true) {
buf.clear();
var n = ch.read(buf);
if (n < 0) break;
buf.flip();
for (var i = 0; i < n; i++) {
if (buf.get(i) == (byte) '\n') {
if (c == offs.length) offs = Arrays.copyOf(offs, offs.length * 2);
offs[c++] = pos + i + 1;
}
}
pos += n;
}
return Arrays.copyOf(offs, c);
}
}
public static void writeIndex(Path out, long[] offsets) throws IOException {
try (var dos = new DataOutputStream(new BufferedOutputStream(Files.newOutputStream(
out, StandardOpenOption.CREATE, StandardOpenOption.TRUNCATE_EXISTING, StandardOpenOption.WRITE)))) {
dos.writeInt(MAGIC);
dos.writeInt(VERSION);
dos.writeInt(offsets.length);
for (var v : offsets) dos.writeLong(v);
}
}
public static long[] readIndex(Path in) throws IOException {
try (var dis = new DataInputStream(new BufferedInputStream(Files.newInputStream(in)))) {
var magic = dis.readInt();
if (magic != MAGIC) throw new IOException("Not a LIDX file");
var version = dis.readInt();
if (version != VERSION) throw new IOException("Unsupported version: " + version);
var n = dis.readInt();
if (n < 0) throw new IOException("Corrupt length: " + n);
var offsets = new long[n];
for (var i = 0; i < n; i++) offsets[i] = dis.readLong();
return offsets;
}
}
@Override
public void close() throws IOException {
synchronized (lock) {
if (csvChannel != null) csvChannel.close();
csvChannel = null;
offsets = null;
}
}
}

View File

@@ -13,9 +13,9 @@ import java.util.*;
import java.util.concurrent.*;
import java.util.concurrent.atomic.AtomicLong;
import static puzzle.CsvIndexService.SC;
import static puzzle.Export.*;
import static puzzle.SwedishGenerator.*;
import static puzzle.SwedishGenerator.Dict.GSON;
import static puzzle.SwedishGenerator.Dict.loadDict;
public class Main {
@@ -53,6 +53,11 @@ public class Main {
}
public void main(String[] args) {
var csv = Paths.get("nl_score_hints_v3.csv");
var idx = Paths.get("nl_score_hints_v3.idx");
ScopedValue.where(SC, new CsvIndexService(csv, idx)).run(() -> _main(args));
}
public void _main(String[] args) {
var opts = parseArgs(args);
if (opts.reindex) {
@@ -69,6 +74,8 @@ public class Main {
section("Settings");
printSettings(opts);
var csv = Paths.get("nl_score_hints_v3.csv");
var idx = Paths.get("nl_score_hints_v3.idx");
var res = generatePuzzle(opts);
if (res == null) {
@@ -377,7 +384,7 @@ public class Main {
record JsonExportedPuzzle(String date, String theme, int difficulty, Rewards rewards, String[] grid, WordOut[] words) { }
private static String toJson(ExportedPuzzle puzzle, String date, String theme) {
return GSON.toJson(new JsonExportedPuzzle(date, theme, puzzle.difficulty(), puzzle.rewards(), puzzle.gridv2(), puzzle.words()));
return CsvIndexService.GSON.toJson(new JsonExportedPuzzle(date, theme, puzzle.difficulty(), puzzle.rewards(), puzzle.gridv2(), puzzle.words()));
}
private static String escapeJson(String s) {

View File

@@ -1,6 +1,5 @@
package puzzle;
import com.google.gson.Gson;
import lombok.Getter;
import lombok.val;
import precomp.Neighbors9x8;
@@ -36,6 +35,7 @@ public record SwedishGenerator(Rng rng) {
//@formatter:off
@FunctionalInterface interface SlotVisitor { void visit(int key, long packedPos, int len); }
//@formatter:on
static final long GT_1_OFFSET_53_BIT = 0x3E00000000000000L;
static final long X = 0L;
static final int LOG_EVERY_MS = 200;
static final int BAR_LEN = 22;
@@ -208,7 +208,7 @@ public record SwedishGenerator(Rng rng) {
if (idx < 64) lo &= ~(1L << idx);
else hi &= ~(1L << (idx & 63));
}
static boolean isDigit(byte b) { return (b & 48) == 48; }
static boolean isDigit(byte b) { return (b & B48) == B48; }
boolean isDigitAt(int index) { return isDigit(g[index]); }
boolean isClue(long index) {
if (index < 64) return ((lo >> index) & 1L) != X;
@@ -242,9 +242,12 @@ public record SwedishGenerator(Rng rng) {
}
return true;
}
static boolean isLetter(byte b) { return (b & 64) != 0; }
static final byte B0 = (byte) 0;
static final byte B64 = (byte) 64;
static final byte B48 = (byte) 48;
static boolean isLetter(byte b) { return (b & B64) != B0; }
public boolean isLetterSet(int idx) { return isLetter(g[idx]); }
static boolean notDigit(byte b) { return (b & 48) != 48; }
static boolean notDigit(byte b) { return (b & B48) != B48; }
public boolean isLetterAt(int index) { return notDigit(g[index]); }
public double similarity(Grid b) {
@@ -252,17 +255,11 @@ public record SwedishGenerator(Rng rng) {
for (int i = 0; i < SIZE; i++) if (g[i] == b.g[i]) same++;
return same / SIZED;
}
int clueCount() { return Long.bitCount(lo) + Long.bitCount(hi); }
/* for (int k = 0, n = Math.min(MAX_WORD_LENGTH7, (int) (packed >>> 56) * 7); k < n; ) {
if (isClue((int) ((packed >>> k) & 0x7F))) break;
k += 7;
if (k >= MIN_LEN7) return true;
}
return false;*/
boolean hasRoomForClue(long packed) { return ((packed >>> 56)) > 1L && notClue(packed & 0x7FL) && notClue((packed >>> 7) & 0x7FL); }
int clueCount() { return Long.bitCount(lo) + Long.bitCount(hi); }
boolean hasRoomForClue(long packed) { return (packed & GT_1_OFFSET_53_BIT) != X && notClue(packed & 0x7FL) && notClue((packed >>> 7) & 0x7FL); }
void forEachSlot(SlotVisitor visitor) {
for (var l = lo; l != X; l &= l - 1) processSlot(this, visitor, Long.numberOfTrailingZeros(l));
for (var h = hi; h != X; h &= h - 1) processSlot(this, visitor, 64 + Long.numberOfTrailingZeros(h));
for (var h = hi; h != X; h &= h - 1) processSlot(this, visitor, 64 | Long.numberOfTrailingZeros(h));
}
}
@@ -286,21 +283,21 @@ public record SwedishGenerator(Rng rng) {
}
}
public static record Lemma(int index, byte[] word, int simpel, String[] clue) {
public static record Lemma(int index, byte[] word, int simpel) {
static int LEMMA_COUNTER = 0;
public Lemma(int index, String word, int simpel, String[] clu) { this(index, word.getBytes(StandardCharsets.US_ASCII), simpel, clu); }
public Lemma(String word, int simpel, String clue) { this(LEMMA_COUNTER++, word, simpel, new String[]{ clue }); }
byte byteAt(int idx) { return word[idx]; }
@Override public int hashCode() { return index; }
@Override public boolean equals(Object o) { return (o == this) || (o instanceof Lemma l && l.index == index); }
public Lemma(int index, String word, int simpel) { this(index, word.getBytes(StandardCharsets.US_ASCII), simpel); }
public Lemma(String word, int simpel) { this(LEMMA_COUNTER++, word, simpel); }
byte byteAt(int idx) { return word[idx]; }
@Override public int hashCode() { return index; }
@Override public boolean equals(Object o) { return (o == this) || (o instanceof Lemma l && l.index == index); }
String[] clue() { return CsvIndexService.clues(index); }
}
public static record Dict(
DictEntry[] index,
int length) {
static final Gson GSON = new Gson();
public Dict(Lemma[] wordz) {
var index = new DictEntry[MAX_WORD_LENGTH_PLUS_ONE];
Arrays.setAll(index, i -> new DictEntry(i));
@@ -317,59 +314,18 @@ public record SwedishGenerator(Rng rng) {
entry.pos[i][letter].add(idx);
}
}
for (int i = MIN_LEN; i < index.length; i++) {
var len = index[i].words.size();
if (len <= 0) {
throw new RuntimeException("No words for length " + i);
}
}
for (int i = MIN_LEN; i < index.length; i++) if (index[i].words.size() <= 0) throw new RuntimeException("No words for length " + i);
this(index, Arrays.stream(index).mapToInt(i -> i.words.size()).sum());
}
static Dict loadDict(String wordsPath) {
String raw;
try {
raw = Files.readString(Path.of(wordsPath), StandardCharsets.UTF_8);
var map = new ArrayList<Lemma>();
Files.lines(Path.of(wordsPath), StandardCharsets.UTF_8).forEach(line -> CsvIndexService.lineToLemma(line, map::add));
return new Dict(map.toArray(Lemma[]::new));
} catch (IOException e) {
e.printStackTrace();
throw new RuntimeException("Failed to load dictionary from " + wordsPath, e);
}
var map = new ArrayList<Lemma>();
var first = true;
for (var line : raw.split("\\R")) {
if (line.isBlank()) {
System.err.println("Empty line: " + line);
continue;
}
var parts = line.split(",", 5);
var id = Integer.parseInt(parts[0].trim());
var word = parts[1].trim();
if (first && word.equalsIgnoreCase("WOORD")) {
first = false;
continue;
}
first = false;
var s = word.toUpperCase(Locale.ROOT);
if (!s.matches("^[A-Z]{2,8}$")) {
System.err.println("Invalid word: " + line);
continue;
}
// CSV has level 1-10. llmScores use 10-level.
int score = Integer.parseInt(parts[2].trim());
if (score < 1) {
if (Main.VERBOSE) System.err.println("Word too complex: " + line);
continue;
}
int simpel = Integer.parseInt(parts[3].trim());
var rawClue = parts[4].trim();
if (rawClue.startsWith("\"") && rawClue.endsWith("\"")) {
rawClue = rawClue.substring(1, rawClue.length() - 1).replace("\"\"", "\"");
}
map.add(new Lemma(id, s, simpel, GSON.fromJson(rawClue, String[].class)));
}
return new Dict(map.toArray(Lemma[]::new));
}
}

View File

@@ -10,6 +10,8 @@ import puzzle.SwedishGenerator.Grid;
import puzzle.SwedishGenerator.Lemma;
import puzzle.SwedishGenerator.Rng;
import java.io.IOException;
import java.nio.file.Paths;
import java.util.HashMap;
import static org.junit.jupiter.api.Assertions.*;
@@ -29,7 +31,7 @@ public class ExportFormatTest {
var clueMap = new HashMap<Integer, Lemma>();
// key = (cellIndex << 4) | direction
var key = (0 << 4) | 2;
var lemma = new Lemma("TEST", 1, "A test word");
var lemma = new Lemma("TEST", 1);
clueMap.put(key, lemma);
// Manually fill the grid letters for "TEST" at (0,1), (0,2), (0,3), (0,4)
@@ -94,4 +96,16 @@ public class ExportFormatTest {
assertTrue(row.matches("#+"));
}
}
@Test
void testIndex() {
var csv = Paths.get("nl_score_hints_v3.csv");
var idx = Paths.get("nl_score_hints_v3.idx");
try (var svc = new CsvIndexService(csv, idx)) {
System.out.println(svc.getLine(1319));
} catch (IOException e) {
throw new RuntimeException(e);
}
}
}

View File

@@ -103,19 +103,19 @@ public class SwedishGeneratorTest {
@Test
void testLemmaAndDict() {
var l2a = new Lemma("IN", 1, "BIJ");
var l4a = new Lemma("INER", 1, "BIJER");
var l6a = new Lemma("INEREN", 1, "BIJERE");
var l7a = new Lemma("INERENA", 1, "BIJERE");
var l8a = new Lemma("INERENAE", 1, "BIJERE");
var l2a = new Lemma("IN", 1);
var l4a = new Lemma("INER", 1);
var l6a = new Lemma("INEREN", 1);
var l7a = new Lemma("INERENA", 1);
var l8a = new Lemma("INERENAE", 1);
var l1 = new Lemma("APPLE", 5, "A fruit");
var l1 = new Lemma("APPLE", 5);
Assertions.assertArrayEquals("APPLE".getBytes(StandardCharsets.US_ASCII), l1.word());
assertEquals(5, l1.word().length);
assertEquals(5, l1.simpel());
assertEquals((byte) 'A', l1.byteAt(0));
var l2 = new Lemma("AXE", 2, "A tool");
var l2 = new Lemma("AXE", 2);
var dict = new Dict(new Lemma[]{ l1, l2, l2a, l4a, l6a, l7a, l8a });
assertEquals(1, dict.index()[3].words().size());
@@ -179,15 +179,15 @@ public class SwedishGeneratorTest {
@Test
void testCandidateInfoForPattern() {
var l0 = new Lemma("IN", 1, "BIJ");
var l3a = new Lemma("INE", 1, "BIJE");
var l4a = new Lemma("INER", 1, "BIJER");
var l6a = new Lemma("INEREN", 1, "BIJERE");
var l7a = new Lemma("INERENA", 1, "BIJERE");
var l8a = new Lemma("INERENAE", 1, "BIJERE");
var l1 = new Lemma("APPLE", 1, "fruit");
var l2 = new Lemma("APPLY", 1, "verb");
var l3 = new Lemma("BANAN", 1, "fruit");
var l0 = new Lemma("IN", 1);
var l3a = new Lemma("INE", 1);
var l4a = new Lemma("INER", 1);
var l6a = new Lemma("INEREN", 1);
var l7a = new Lemma("INERENA", 1);
var l8a = new Lemma("INERENAE", 1);
var l1 = new Lemma("APPLE", 1);
var l2 = new Lemma("APPLY", 1);
var l3 = new Lemma("BANAN", 1);
var dict = new Dict(new Lemma[]{ l0, l1, l2, l3, l3a, l4a, l6a, l7a, l8a });
// Pattern "APP--" for length 5
@@ -274,7 +274,7 @@ public class SwedishGeneratorTest {
// r(i) and c(i) are used by placeWord.
var packedPos = ((long) Grid.offset(0, 0)) | (((long) Grid.offset(0, 1)) << 7) | (((long) Grid.offset(0, 2)) << 14);
var s = Slot.from(0, packedPos, 3);
var w1 = new Lemma("ABC", 1, "test");
var w1 = new Lemma("ABC", 1);
var undoBuffer = new int[10];
// 1. Successful placement in empty grid
@@ -289,7 +289,7 @@ public class SwedishGeneratorTest {
assertEquals(0L, undoBuffer[1]); // 0 new characters placed
// 3. Conflict: place "ABD" where "ABC" is
var w2 = new Lemma("ABD", 1, "conflict");
var w2 = new Lemma("ABD", 1);
assertFalse(placeWord(grid, s, w2, undoBuffer, 2));
// Verify grid is unchanged (still "ABC")
assertEquals('A', grid.byteAt(Grid.offset(0, 0)));
@@ -312,7 +312,7 @@ public class SwedishGeneratorTest {
// Slot at 0,1 length 2
var packedPos = ((long) Grid.offset(0, 1)) | (((long) Grid.offset(0, 2)) << 7);
var s = Slot.from((0 << 8) | (1 << 4) | 2, packedPos, 2);
var w = new Lemma("AZ", 1, "A to Z");
var w = new Lemma("AZ", 1);
var undoBuffer = new int[10];
var placed = placeWord(grid, s, w, undoBuffer, 0);