|
| 1 | +package me.lemire.longcompression; |
| 2 | + |
| 3 | +import me.lemire.integercompression.BinaryPacking; |
| 4 | +import me.lemire.integercompression.IntWrapper; |
| 5 | +import me.lemire.integercompression.Util; |
| 6 | + |
| 7 | +/** |
| 8 | + * Scheme based on a commonly used idea: can be extremely fast. |
| 9 | + * It encodes integers in blocks of 64 longs. For arrays containing |
| 10 | + * an arbitrary number of longs, you should use it in conjunction |
| 11 | + * with another CODEC: |
| 12 | + * |
| 13 | + * <pre>LongCODEC ic = |
| 14 | + * new Composition(new LongBinaryPacking(), new LongVariableByte()).</pre> |
| 15 | + * |
| 16 | + * Note that this does not use differential coding: if you are working on sorted |
| 17 | + * lists, you must compute the deltas separately. |
| 18 | + * |
| 19 | + * <p> |
| 20 | + * For details, please see {@link BinaryPacking} |
| 21 | + * </p> |
| 22 | + * |
| 23 | + * @author Benoit Lacelle |
| 24 | + */ |
| 25 | +public final class LongBinaryPacking implements LongCODEC, SkippableLongCODEC { |
| 26 | + final static int BLOCK_SIZE = 64; |
| 27 | + |
| 28 | + @Override |
| 29 | + public void compress(long[] in, IntWrapper inpos, int inlength, |
| 30 | + long[] out, IntWrapper outpos) { |
| 31 | + inlength = Util.greatestMultiple(inlength, BLOCK_SIZE); |
| 32 | + if (inlength == 0) |
| 33 | + return; |
| 34 | + out[outpos.get()] = inlength; |
| 35 | + outpos.increment(); |
| 36 | + headlessCompress(in, inpos, inlength, out, outpos); |
| 37 | + } |
| 38 | + |
| 39 | + @Override |
| 40 | + public void headlessCompress(long[] in, IntWrapper inpos, int inlength, |
| 41 | + long[] out, IntWrapper outpos) { |
| 42 | + inlength = Util.greatestMultiple(inlength, BLOCK_SIZE); |
| 43 | + int tmpoutpos = outpos.get(); |
| 44 | + int s = inpos.get(); |
| 45 | + // Compress by block of 8 * 64 longs as much as possible |
| 46 | + for (; s + BLOCK_SIZE * 8 - 1 < inpos.get() + inlength; s += BLOCK_SIZE * 8) { |
| 47 | + // maxbits can be anything between 0 and 64 included: expressed within a byte (1 << 6) |
| 48 | + final int mbits1 = LongUtil.maxbits(in, s + 0 * BLOCK_SIZE, BLOCK_SIZE); |
| 49 | + final int mbits2 = LongUtil.maxbits(in, s + 1 * BLOCK_SIZE, BLOCK_SIZE); |
| 50 | + final int mbits3 = LongUtil.maxbits(in, s + 2 * BLOCK_SIZE, BLOCK_SIZE); |
| 51 | + final int mbits4 = LongUtil.maxbits(in, s + 3 * BLOCK_SIZE, BLOCK_SIZE); |
| 52 | + final int mbits5 = LongUtil.maxbits(in, s + 4 * BLOCK_SIZE, BLOCK_SIZE); |
| 53 | + final int mbits6 = LongUtil.maxbits(in, s + 5 * BLOCK_SIZE, BLOCK_SIZE); |
| 54 | + final int mbits7 = LongUtil.maxbits(in, s + 6 * BLOCK_SIZE, BLOCK_SIZE); |
| 55 | + final int mbits8 = LongUtil.maxbits(in, s + 7 * BLOCK_SIZE, BLOCK_SIZE); |
| 56 | + // The first long expressed the maxbits for the 8 buckets |
| 57 | + out[tmpoutpos++] = ((long) mbits1 << 56) | ((long) mbits2 << 48) | ((long) mbits3 << 40) | ((long) mbits4 << 32) | (mbits5 << 24) | (mbits6 << 16) | (mbits7 << 8) | (mbits8); |
| 58 | + LongBitPacking.fastpackwithoutmask(in, s + 0 * BLOCK_SIZE, out, tmpoutpos, (int) mbits1); |
| 59 | + tmpoutpos += mbits1; |
| 60 | + LongBitPacking.fastpackwithoutmask(in, s + 1 * BLOCK_SIZE, out, tmpoutpos, (int) mbits2); |
| 61 | + tmpoutpos += mbits2; |
| 62 | + LongBitPacking.fastpackwithoutmask(in, s + 2 * BLOCK_SIZE, out, tmpoutpos, (int) mbits3); |
| 63 | + tmpoutpos += mbits3; |
| 64 | + LongBitPacking.fastpackwithoutmask(in, s + 3 * BLOCK_SIZE, out, tmpoutpos, (int) mbits4); |
| 65 | + tmpoutpos += mbits4; |
| 66 | + LongBitPacking.fastpackwithoutmask(in, s + 4 * BLOCK_SIZE, out, tmpoutpos, (int) mbits5); |
| 67 | + tmpoutpos += mbits5; |
| 68 | + LongBitPacking.fastpackwithoutmask(in, s + 5 * BLOCK_SIZE, out, tmpoutpos, (int) mbits6); |
| 69 | + tmpoutpos += mbits6; |
| 70 | + LongBitPacking.fastpackwithoutmask(in, s + 6 * BLOCK_SIZE, out, tmpoutpos, (int) mbits7); |
| 71 | + tmpoutpos += mbits7; |
| 72 | + LongBitPacking.fastpackwithoutmask(in, s + 7 * BLOCK_SIZE, out, tmpoutpos, (int) mbits8); |
| 73 | + tmpoutpos += mbits8; |
| 74 | + } |
| 75 | + // Then we compress up to 7 blocks of 64 longs |
| 76 | + for (; s < inpos.get() + inlength; s += BLOCK_SIZE ) { |
| 77 | + final int mbits = LongUtil.maxbits(in, s, BLOCK_SIZE); |
| 78 | + out[tmpoutpos++] = mbits; |
| 79 | + LongBitPacking.fastpackwithoutmask(in, s, out, tmpoutpos, mbits); |
| 80 | + tmpoutpos += mbits; |
| 81 | + } |
| 82 | + inpos.add(inlength); |
| 83 | + outpos.set(tmpoutpos); |
| 84 | + } |
| 85 | + |
| 86 | + @Override |
| 87 | + public void uncompress(long[] in, IntWrapper inpos, int inlength, |
| 88 | + long[] out, IntWrapper outpos) { |
| 89 | + if (inlength == 0) |
| 90 | + return; |
| 91 | + final int outlength = (int) in[inpos.get()]; |
| 92 | + inpos.increment(); |
| 93 | + headlessUncompress(in,inpos, inlength,out,outpos,outlength); |
| 94 | + } |
| 95 | + |
| 96 | + @Override |
| 97 | + public void headlessUncompress(long[] in, IntWrapper inpos, int inlength, |
| 98 | + long[] out, IntWrapper outpos, int num) { |
| 99 | + final int outlength = Util.greatestMultiple(num, BLOCK_SIZE); |
| 100 | + int tmpinpos = inpos.get(); |
| 101 | + int s = outpos.get(); |
| 102 | + for (; s + BLOCK_SIZE * 8 - 1 < outpos.get() + outlength; s += BLOCK_SIZE * 8) { |
| 103 | + final int mbits1 = (int) ((in[tmpinpos] >>> 56)); |
| 104 | + final int mbits2 = (int) ((in[tmpinpos] >>> 48) & 0xFF); |
| 105 | + final int mbits3 = (int) ((in[tmpinpos] >>> 40) & 0xFF); |
| 106 | + final int mbits4 = (int) ((in[tmpinpos] >>> 32) & 0xFF); |
| 107 | + final int mbits5 = (int) ((in[tmpinpos] >>> 24) & 0xFF); |
| 108 | + final int mbits6 = (int) ((in[tmpinpos] >>> 16) & 0xFF); |
| 109 | + final int mbits7 = (int) ((in[tmpinpos] >>> 8) & 0xFF); |
| 110 | + final int mbits8 = (int) ((in[tmpinpos]) & 0xFF); |
| 111 | + ++tmpinpos; |
| 112 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 0 * BLOCK_SIZE, mbits1); |
| 113 | + tmpinpos += mbits1; |
| 114 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 1 * BLOCK_SIZE, mbits2); |
| 115 | + tmpinpos += mbits2; |
| 116 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 2 * BLOCK_SIZE, mbits3); |
| 117 | + tmpinpos += mbits3; |
| 118 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 3 * BLOCK_SIZE, mbits4); |
| 119 | + tmpinpos += mbits4; |
| 120 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 4 * BLOCK_SIZE, mbits5); |
| 121 | + tmpinpos += mbits5; |
| 122 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 5 * BLOCK_SIZE, mbits6); |
| 123 | + tmpinpos += mbits6; |
| 124 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 6 * BLOCK_SIZE, mbits7); |
| 125 | + tmpinpos += mbits7; |
| 126 | + LongBitPacking.fastunpack(in, tmpinpos, out, s + 7 * BLOCK_SIZE, mbits8); |
| 127 | + tmpinpos += mbits8; |
| 128 | + } |
| 129 | + for (; s < outpos.get() + outlength; s += BLOCK_SIZE ) { |
| 130 | + final int mbits = (int) in[tmpinpos]; |
| 131 | + ++tmpinpos; |
| 132 | + LongBitPacking.fastunpack(in, tmpinpos, out, s, mbits); |
| 133 | + tmpinpos += mbits; |
| 134 | + } |
| 135 | + outpos.add(outlength); |
| 136 | + inpos.set(tmpinpos); |
| 137 | + } |
| 138 | + |
| 139 | + @Override |
| 140 | + public String toString() { |
| 141 | + return this.getClass().getSimpleName(); |
| 142 | + } |
| 143 | +} |
0 commit comments