compressedStream.cpp revision 0:a61af66fc99e
1/* 2 * Copyright 1997-2006 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 20 * CA 95054 USA or visit www.sun.com if you need additional information or 21 * have any questions. 22 * 23 */ 24 25#include "incls/_precompiled.incl" 26#include "incls/_compressedStream.cpp.incl" 27 28// 32-bit one-to-one sign encoding taken from Pack200 29// converts leading sign bits into leading zeroes with trailing sign bit 30inline juint CompressedStream::encode_sign(jint value) { 31 return (value << 1) ^ (value >> 31); 32} 33inline jint CompressedStream::decode_sign(juint value) { 34 return (value >> 1) ^ -(jint)(value & 1); 35} 36 37// 32-bit self-inverse encoding of float bits 38// converts trailing zeroes (common in floats) to leading zeroes 39inline juint CompressedStream::reverse_int(juint i) { 40 // Hacker's Delight, Figure 7-1 41 i = (i & 0x55555555) << 1 | (i >> 1) & 0x55555555; 42 i = (i & 0x33333333) << 2 | (i >> 2) & 0x33333333; 43 i = (i & 0x0f0f0f0f) << 4 | (i >> 4) & 0x0f0f0f0f; 44 i = (i << 24) | ((i & 0xff00) << 8) | ((i >> 8) & 0xff00) | (i >> 24); 45 return i; 46} 47 48 49jint CompressedReadStream::read_signed_int() { 50 return decode_sign(read_int()); 51} 52 53// Compressing floats is simple, because the only common pattern 54// is trailing zeroes. (Compare leading sign bits on ints.) 55// Since floats are left-justified, as opposed to right-justified 56// ints, we can bit-reverse them in order to take advantage of int 57// compression. 58 59jfloat CompressedReadStream::read_float() { 60 int rf = read_int(); 61 int f = reverse_int(rf); 62 return jfloat_cast(f); 63} 64 65jdouble CompressedReadStream::read_double() { 66 jint rh = read_int(); 67 jint rl = read_int(); 68 jint h = reverse_int(rh); 69 jint l = reverse_int(rl); 70 return jdouble_cast(jlong_from(h, l)); 71} 72 73jlong CompressedReadStream::read_long() { 74 jint low = read_signed_int(); 75 jint high = read_signed_int(); 76 return jlong_from(high, low); 77} 78 79CompressedWriteStream::CompressedWriteStream(int initial_size) : CompressedStream(NULL, 0) { 80 _buffer = NEW_RESOURCE_ARRAY(u_char, initial_size); 81 _size = initial_size; 82 _position = 0; 83} 84 85void CompressedWriteStream::grow() { 86 u_char* _new_buffer = NEW_RESOURCE_ARRAY(u_char, _size * 2); 87 memcpy(_new_buffer, _buffer, _position); 88 _buffer = _new_buffer; 89 _size = _size * 2; 90} 91 92void CompressedWriteStream::write_signed_int(jint value) { 93 // this encoding, called SIGNED5, is taken from Pack200 94 write_int(encode_sign(value)); 95} 96 97void CompressedWriteStream::write_float(jfloat value) { 98 juint f = jint_cast(value); 99 juint rf = reverse_int(f); 100 assert(f == reverse_int(rf), "can re-read same bits"); 101 write_int(rf); 102} 103 104void CompressedWriteStream::write_double(jdouble value) { 105 juint h = high(jlong_cast(value)); 106 juint l = low( jlong_cast(value)); 107 juint rh = reverse_int(h); 108 juint rl = reverse_int(l); 109 assert(h == reverse_int(rh), "can re-read same bits"); 110 assert(l == reverse_int(rl), "can re-read same bits"); 111 write_int(rh); 112 write_int(rl); 113} 114 115void CompressedWriteStream::write_long(jlong value) { 116 write_signed_int(low(value)); 117 write_signed_int(high(value)); 118} 119 120 121/// The remaining details 122 123#ifndef PRODUCT 124// set this to trigger unit test 125void test_compressed_stream(int trace); 126bool test_compressed_stream_enabled = false; 127#endif 128 129// This encoding, called UNSIGNED5, is taken from J2SE Pack200. 130// It assumes that most values have lots of leading zeroes. 131// Very small values, in the range [0..191], code in one byte. 132// Any 32-bit value (including negatives) can be coded, in 133// up to five bytes. The grammar is: 134// low_byte = [0..191] 135// high_byte = [192..255] 136// any_byte = low_byte | high_byte 137// coding = low_byte 138// | high_byte low_byte 139// | high_byte high_byte low_byte 140// | high_byte high_byte high_byte low_byte 141// | high_byte high_byte high_byte high_byte any_byte 142// Each high_byte contributes six bits of payload. 143// The encoding is one-to-one (except for integer overflow) 144// and easy to parse and unparse. 145 146jint CompressedReadStream::read_int_mb(jint b0) { 147 int pos = position() - 1; 148 u_char* buf = buffer() + pos; 149 assert(buf[0] == b0 && b0 >= L, "correctly called"); 150 jint sum = b0; 151 // must collect more bytes: b[1]...b[4] 152 int lg_H_i = lg_H; 153 for (int i = 0; ; ) { 154 jint b_i = buf[++i]; // b_i = read(); ++i; 155 sum += b_i << lg_H_i; // sum += b[i]*(64**i) 156 if (b_i < L || i == MAX_i) { 157 set_position(pos+i+1); 158 return sum; 159 } 160 lg_H_i += lg_H; 161 } 162} 163 164void CompressedWriteStream::write_int_mb(jint value) { 165 debug_only(int pos1 = position()); 166 juint sum = value; 167 for (int i = 0; ; ) { 168 if (sum < L || i == MAX_i) { 169 // remainder is either a "low code" or the 5th byte 170 assert(sum == (u_char)sum, "valid byte"); 171 write((u_char)sum); 172 break; 173 } 174 sum -= L; 175 int b_i = L + (sum % H); // this is a "high code" 176 sum >>= lg_H; // extracted 6 bits 177 write(b_i); ++i; 178 } 179 180#ifndef PRODUCT 181 if (test_compressed_stream_enabled) { // hack to enable this stress test 182 test_compressed_stream_enabled = false; 183 test_compressed_stream(0); 184 } 185#endif 186} 187 188 189#ifndef PRODUCT 190/// a unit test (can be run by hand from a debugger) 191 192// Avoid a VS2005 compiler stack overflow w/ fastdebug build. 193// The following pragma optimize turns off optimization ONLY 194// for this block (a matching directive turns it back on later). 195// These directives can be removed once the MS VS.NET 2005 196// compiler stack overflow is fixed. 197#if _MSC_VER >=1400 && !defined(_WIN64) 198#pragma optimize("", off) 199#endif 200 201// generator for an "interesting" set of critical values 202enum { stretch_limit = (1<<16) * (64-16+1) }; 203static jlong stretch(jint x, int bits) { 204 // put x[high 4] into place 205 jlong h = (jlong)((x >> (16-4))) << (bits - 4); 206 // put x[low 12] into place, sign extended 207 jlong l = ((jlong)x << (64-12)) >> (64-12); 208 // move l upwards, maybe 209 l <<= (x >> 16); 210 return h ^ l; 211} 212 213void test_compressed_stream(int trace) { 214 CompressedWriteStream bytes(stretch_limit * 100); 215 jint n; 216 int step = 0, fails = 0; 217#define CHECKXY(x, y, fmt) { \ 218 ++step; \ 219 int xlen = (pos = decode.position()) - lastpos; lastpos = pos; \ 220 if (trace > 0 && (step % trace) == 0) { \ 221 tty->print_cr("step %d, n=%08x: value=" fmt " (len=%d)", \ 222 step, n, x, xlen); } \ 223 if (x != y) { \ 224 tty->print_cr("step %d, n=%d: " fmt " != " fmt, step, n, x, y); \ 225 fails++; \ 226 } } 227 for (n = 0; n < (1<<8); n++) { 228 jbyte x = (jbyte)n; 229 bytes.write_byte(x); ++step; 230 } 231 for (n = 0; n < stretch_limit; n++) { 232 jint x = (jint)stretch(n, 32); 233 bytes.write_int(x); ++step; 234 bytes.write_signed_int(x); ++step; 235 bytes.write_float(jfloat_cast(x)); ++step; 236 } 237 for (n = 0; n < stretch_limit; n++) { 238 jlong x = stretch(n, 64); 239 bytes.write_long(x); ++step; 240 bytes.write_double(jdouble_cast(x)); ++step; 241 } 242 int length = bytes.position(); 243 if (trace != 0) 244 tty->print_cr("set up test of %d stream values, size %d", step, length); 245 step = 0; 246 // now decode it all 247 CompressedReadStream decode(bytes.buffer()); 248 int pos, lastpos = decode.position(); 249 for (n = 0; n < (1<<8); n++) { 250 jbyte x = (jbyte)n; 251 jbyte y = decode.read_byte(); 252 CHECKXY(x, y, "%db"); 253 } 254 for (n = 0; n < stretch_limit; n++) { 255 jint x = (jint)stretch(n, 32); 256 jint y1 = decode.read_int(); 257 CHECKXY(x, y1, "%du"); 258 jint y2 = decode.read_signed_int(); 259 CHECKXY(x, y2, "%di"); 260 jint y3 = jint_cast(decode.read_float()); 261 CHECKXY(x, y3, "%df"); 262 } 263 for (n = 0; n < stretch_limit; n++) { 264 jlong x = stretch(n, 64); 265 jlong y1 = decode.read_long(); 266 CHECKXY(x, y1, INT64_FORMAT "l"); 267 jlong y2 = jlong_cast(decode.read_double()); 268 CHECKXY(x, y2, INT64_FORMAT "d"); 269 } 270 int length2 = decode.position(); 271 if (trace != 0) 272 tty->print_cr("finished test of %d stream values, size %d", step, length2); 273 guarantee(length == length2, "bad length"); 274 guarantee(fails == 0, "test failures"); 275} 276 277#if _MSC_VER >=1400 && !defined(_WIN64) 278#pragma optimize("", on) 279#endif 280 281#endif // PRODUCT 282