Wed, 02 Jun 2010 22:45:42 -0700
Merge
duke@435 | 1 | /* |
trims@1907 | 2 | * Copyright (c) 1997, 2006, Oracle and/or its affiliates. All rights reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
trims@1907 | 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
trims@1907 | 20 | * or visit www.oracle.com if you need additional information or have any |
trims@1907 | 21 | * questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
duke@435 | 25 | #include "incls/_precompiled.incl" |
duke@435 | 26 | #include "incls/_compressedStream.cpp.incl" |
duke@435 | 27 | |
duke@435 | 28 | // 32-bit one-to-one sign encoding taken from Pack200 |
duke@435 | 29 | // converts leading sign bits into leading zeroes with trailing sign bit |
duke@435 | 30 | inline juint CompressedStream::encode_sign(jint value) { |
duke@435 | 31 | return (value << 1) ^ (value >> 31); |
duke@435 | 32 | } |
duke@435 | 33 | inline jint CompressedStream::decode_sign(juint value) { |
duke@435 | 34 | return (value >> 1) ^ -(jint)(value & 1); |
duke@435 | 35 | } |
duke@435 | 36 | |
duke@435 | 37 | // 32-bit self-inverse encoding of float bits |
duke@435 | 38 | // converts trailing zeroes (common in floats) to leading zeroes |
duke@435 | 39 | inline juint CompressedStream::reverse_int(juint i) { |
duke@435 | 40 | // Hacker's Delight, Figure 7-1 |
duke@435 | 41 | i = (i & 0x55555555) << 1 | (i >> 1) & 0x55555555; |
duke@435 | 42 | i = (i & 0x33333333) << 2 | (i >> 2) & 0x33333333; |
duke@435 | 43 | i = (i & 0x0f0f0f0f) << 4 | (i >> 4) & 0x0f0f0f0f; |
duke@435 | 44 | i = (i << 24) | ((i & 0xff00) << 8) | ((i >> 8) & 0xff00) | (i >> 24); |
duke@435 | 45 | return i; |
duke@435 | 46 | } |
duke@435 | 47 | |
duke@435 | 48 | |
duke@435 | 49 | jint CompressedReadStream::read_signed_int() { |
duke@435 | 50 | return decode_sign(read_int()); |
duke@435 | 51 | } |
duke@435 | 52 | |
duke@435 | 53 | // Compressing floats is simple, because the only common pattern |
duke@435 | 54 | // is trailing zeroes. (Compare leading sign bits on ints.) |
duke@435 | 55 | // Since floats are left-justified, as opposed to right-justified |
duke@435 | 56 | // ints, we can bit-reverse them in order to take advantage of int |
duke@435 | 57 | // compression. |
duke@435 | 58 | |
duke@435 | 59 | jfloat CompressedReadStream::read_float() { |
duke@435 | 60 | int rf = read_int(); |
duke@435 | 61 | int f = reverse_int(rf); |
duke@435 | 62 | return jfloat_cast(f); |
duke@435 | 63 | } |
duke@435 | 64 | |
duke@435 | 65 | jdouble CompressedReadStream::read_double() { |
duke@435 | 66 | jint rh = read_int(); |
duke@435 | 67 | jint rl = read_int(); |
duke@435 | 68 | jint h = reverse_int(rh); |
duke@435 | 69 | jint l = reverse_int(rl); |
duke@435 | 70 | return jdouble_cast(jlong_from(h, l)); |
duke@435 | 71 | } |
duke@435 | 72 | |
duke@435 | 73 | jlong CompressedReadStream::read_long() { |
duke@435 | 74 | jint low = read_signed_int(); |
duke@435 | 75 | jint high = read_signed_int(); |
duke@435 | 76 | return jlong_from(high, low); |
duke@435 | 77 | } |
duke@435 | 78 | |
duke@435 | 79 | CompressedWriteStream::CompressedWriteStream(int initial_size) : CompressedStream(NULL, 0) { |
duke@435 | 80 | _buffer = NEW_RESOURCE_ARRAY(u_char, initial_size); |
duke@435 | 81 | _size = initial_size; |
duke@435 | 82 | _position = 0; |
duke@435 | 83 | } |
duke@435 | 84 | |
duke@435 | 85 | void CompressedWriteStream::grow() { |
duke@435 | 86 | u_char* _new_buffer = NEW_RESOURCE_ARRAY(u_char, _size * 2); |
duke@435 | 87 | memcpy(_new_buffer, _buffer, _position); |
duke@435 | 88 | _buffer = _new_buffer; |
duke@435 | 89 | _size = _size * 2; |
duke@435 | 90 | } |
duke@435 | 91 | |
duke@435 | 92 | void CompressedWriteStream::write_signed_int(jint value) { |
duke@435 | 93 | // this encoding, called SIGNED5, is taken from Pack200 |
duke@435 | 94 | write_int(encode_sign(value)); |
duke@435 | 95 | } |
duke@435 | 96 | |
duke@435 | 97 | void CompressedWriteStream::write_float(jfloat value) { |
duke@435 | 98 | juint f = jint_cast(value); |
duke@435 | 99 | juint rf = reverse_int(f); |
duke@435 | 100 | assert(f == reverse_int(rf), "can re-read same bits"); |
duke@435 | 101 | write_int(rf); |
duke@435 | 102 | } |
duke@435 | 103 | |
duke@435 | 104 | void CompressedWriteStream::write_double(jdouble value) { |
duke@435 | 105 | juint h = high(jlong_cast(value)); |
duke@435 | 106 | juint l = low( jlong_cast(value)); |
duke@435 | 107 | juint rh = reverse_int(h); |
duke@435 | 108 | juint rl = reverse_int(l); |
duke@435 | 109 | assert(h == reverse_int(rh), "can re-read same bits"); |
duke@435 | 110 | assert(l == reverse_int(rl), "can re-read same bits"); |
duke@435 | 111 | write_int(rh); |
duke@435 | 112 | write_int(rl); |
duke@435 | 113 | } |
duke@435 | 114 | |
duke@435 | 115 | void CompressedWriteStream::write_long(jlong value) { |
duke@435 | 116 | write_signed_int(low(value)); |
duke@435 | 117 | write_signed_int(high(value)); |
duke@435 | 118 | } |
duke@435 | 119 | |
duke@435 | 120 | |
duke@435 | 121 | /// The remaining details |
duke@435 | 122 | |
duke@435 | 123 | #ifndef PRODUCT |
duke@435 | 124 | // set this to trigger unit test |
duke@435 | 125 | void test_compressed_stream(int trace); |
duke@435 | 126 | bool test_compressed_stream_enabled = false; |
duke@435 | 127 | #endif |
duke@435 | 128 | |
duke@435 | 129 | // This encoding, called UNSIGNED5, is taken from J2SE Pack200. |
duke@435 | 130 | // It assumes that most values have lots of leading zeroes. |
duke@435 | 131 | // Very small values, in the range [0..191], code in one byte. |
duke@435 | 132 | // Any 32-bit value (including negatives) can be coded, in |
duke@435 | 133 | // up to five bytes. The grammar is: |
duke@435 | 134 | // low_byte = [0..191] |
duke@435 | 135 | // high_byte = [192..255] |
duke@435 | 136 | // any_byte = low_byte | high_byte |
duke@435 | 137 | // coding = low_byte |
duke@435 | 138 | // | high_byte low_byte |
duke@435 | 139 | // | high_byte high_byte low_byte |
duke@435 | 140 | // | high_byte high_byte high_byte low_byte |
duke@435 | 141 | // | high_byte high_byte high_byte high_byte any_byte |
duke@435 | 142 | // Each high_byte contributes six bits of payload. |
duke@435 | 143 | // The encoding is one-to-one (except for integer overflow) |
duke@435 | 144 | // and easy to parse and unparse. |
duke@435 | 145 | |
duke@435 | 146 | jint CompressedReadStream::read_int_mb(jint b0) { |
duke@435 | 147 | int pos = position() - 1; |
duke@435 | 148 | u_char* buf = buffer() + pos; |
duke@435 | 149 | assert(buf[0] == b0 && b0 >= L, "correctly called"); |
duke@435 | 150 | jint sum = b0; |
duke@435 | 151 | // must collect more bytes: b[1]...b[4] |
duke@435 | 152 | int lg_H_i = lg_H; |
duke@435 | 153 | for (int i = 0; ; ) { |
duke@435 | 154 | jint b_i = buf[++i]; // b_i = read(); ++i; |
duke@435 | 155 | sum += b_i << lg_H_i; // sum += b[i]*(64**i) |
duke@435 | 156 | if (b_i < L || i == MAX_i) { |
duke@435 | 157 | set_position(pos+i+1); |
duke@435 | 158 | return sum; |
duke@435 | 159 | } |
duke@435 | 160 | lg_H_i += lg_H; |
duke@435 | 161 | } |
duke@435 | 162 | } |
duke@435 | 163 | |
duke@435 | 164 | void CompressedWriteStream::write_int_mb(jint value) { |
duke@435 | 165 | debug_only(int pos1 = position()); |
duke@435 | 166 | juint sum = value; |
duke@435 | 167 | for (int i = 0; ; ) { |
duke@435 | 168 | if (sum < L || i == MAX_i) { |
duke@435 | 169 | // remainder is either a "low code" or the 5th byte |
duke@435 | 170 | assert(sum == (u_char)sum, "valid byte"); |
duke@435 | 171 | write((u_char)sum); |
duke@435 | 172 | break; |
duke@435 | 173 | } |
duke@435 | 174 | sum -= L; |
duke@435 | 175 | int b_i = L + (sum % H); // this is a "high code" |
duke@435 | 176 | sum >>= lg_H; // extracted 6 bits |
duke@435 | 177 | write(b_i); ++i; |
duke@435 | 178 | } |
duke@435 | 179 | |
duke@435 | 180 | #ifndef PRODUCT |
duke@435 | 181 | if (test_compressed_stream_enabled) { // hack to enable this stress test |
duke@435 | 182 | test_compressed_stream_enabled = false; |
duke@435 | 183 | test_compressed_stream(0); |
duke@435 | 184 | } |
duke@435 | 185 | #endif |
duke@435 | 186 | } |
duke@435 | 187 | |
duke@435 | 188 | |
duke@435 | 189 | #ifndef PRODUCT |
duke@435 | 190 | /// a unit test (can be run by hand from a debugger) |
duke@435 | 191 | |
duke@435 | 192 | // Avoid a VS2005 compiler stack overflow w/ fastdebug build. |
duke@435 | 193 | // The following pragma optimize turns off optimization ONLY |
duke@435 | 194 | // for this block (a matching directive turns it back on later). |
duke@435 | 195 | // These directives can be removed once the MS VS.NET 2005 |
duke@435 | 196 | // compiler stack overflow is fixed. |
duke@435 | 197 | #if _MSC_VER >=1400 && !defined(_WIN64) |
duke@435 | 198 | #pragma optimize("", off) |
duke@435 | 199 | #endif |
duke@435 | 200 | |
duke@435 | 201 | // generator for an "interesting" set of critical values |
duke@435 | 202 | enum { stretch_limit = (1<<16) * (64-16+1) }; |
duke@435 | 203 | static jlong stretch(jint x, int bits) { |
duke@435 | 204 | // put x[high 4] into place |
duke@435 | 205 | jlong h = (jlong)((x >> (16-4))) << (bits - 4); |
duke@435 | 206 | // put x[low 12] into place, sign extended |
duke@435 | 207 | jlong l = ((jlong)x << (64-12)) >> (64-12); |
duke@435 | 208 | // move l upwards, maybe |
duke@435 | 209 | l <<= (x >> 16); |
duke@435 | 210 | return h ^ l; |
duke@435 | 211 | } |
duke@435 | 212 | |
duke@435 | 213 | void test_compressed_stream(int trace) { |
duke@435 | 214 | CompressedWriteStream bytes(stretch_limit * 100); |
duke@435 | 215 | jint n; |
duke@435 | 216 | int step = 0, fails = 0; |
duke@435 | 217 | #define CHECKXY(x, y, fmt) { \ |
duke@435 | 218 | ++step; \ |
duke@435 | 219 | int xlen = (pos = decode.position()) - lastpos; lastpos = pos; \ |
duke@435 | 220 | if (trace > 0 && (step % trace) == 0) { \ |
duke@435 | 221 | tty->print_cr("step %d, n=%08x: value=" fmt " (len=%d)", \ |
duke@435 | 222 | step, n, x, xlen); } \ |
duke@435 | 223 | if (x != y) { \ |
duke@435 | 224 | tty->print_cr("step %d, n=%d: " fmt " != " fmt, step, n, x, y); \ |
duke@435 | 225 | fails++; \ |
duke@435 | 226 | } } |
duke@435 | 227 | for (n = 0; n < (1<<8); n++) { |
duke@435 | 228 | jbyte x = (jbyte)n; |
duke@435 | 229 | bytes.write_byte(x); ++step; |
duke@435 | 230 | } |
duke@435 | 231 | for (n = 0; n < stretch_limit; n++) { |
duke@435 | 232 | jint x = (jint)stretch(n, 32); |
duke@435 | 233 | bytes.write_int(x); ++step; |
duke@435 | 234 | bytes.write_signed_int(x); ++step; |
duke@435 | 235 | bytes.write_float(jfloat_cast(x)); ++step; |
duke@435 | 236 | } |
duke@435 | 237 | for (n = 0; n < stretch_limit; n++) { |
duke@435 | 238 | jlong x = stretch(n, 64); |
duke@435 | 239 | bytes.write_long(x); ++step; |
duke@435 | 240 | bytes.write_double(jdouble_cast(x)); ++step; |
duke@435 | 241 | } |
duke@435 | 242 | int length = bytes.position(); |
duke@435 | 243 | if (trace != 0) |
duke@435 | 244 | tty->print_cr("set up test of %d stream values, size %d", step, length); |
duke@435 | 245 | step = 0; |
duke@435 | 246 | // now decode it all |
duke@435 | 247 | CompressedReadStream decode(bytes.buffer()); |
duke@435 | 248 | int pos, lastpos = decode.position(); |
duke@435 | 249 | for (n = 0; n < (1<<8); n++) { |
duke@435 | 250 | jbyte x = (jbyte)n; |
duke@435 | 251 | jbyte y = decode.read_byte(); |
duke@435 | 252 | CHECKXY(x, y, "%db"); |
duke@435 | 253 | } |
duke@435 | 254 | for (n = 0; n < stretch_limit; n++) { |
duke@435 | 255 | jint x = (jint)stretch(n, 32); |
duke@435 | 256 | jint y1 = decode.read_int(); |
duke@435 | 257 | CHECKXY(x, y1, "%du"); |
duke@435 | 258 | jint y2 = decode.read_signed_int(); |
duke@435 | 259 | CHECKXY(x, y2, "%di"); |
duke@435 | 260 | jint y3 = jint_cast(decode.read_float()); |
duke@435 | 261 | CHECKXY(x, y3, "%df"); |
duke@435 | 262 | } |
duke@435 | 263 | for (n = 0; n < stretch_limit; n++) { |
duke@435 | 264 | jlong x = stretch(n, 64); |
duke@435 | 265 | jlong y1 = decode.read_long(); |
duke@435 | 266 | CHECKXY(x, y1, INT64_FORMAT "l"); |
duke@435 | 267 | jlong y2 = jlong_cast(decode.read_double()); |
duke@435 | 268 | CHECKXY(x, y2, INT64_FORMAT "d"); |
duke@435 | 269 | } |
duke@435 | 270 | int length2 = decode.position(); |
duke@435 | 271 | if (trace != 0) |
duke@435 | 272 | tty->print_cr("finished test of %d stream values, size %d", step, length2); |
duke@435 | 273 | guarantee(length == length2, "bad length"); |
duke@435 | 274 | guarantee(fails == 0, "test failures"); |
duke@435 | 275 | } |
duke@435 | 276 | |
duke@435 | 277 | #if _MSC_VER >=1400 && !defined(_WIN64) |
duke@435 | 278 | #pragma optimize("", on) |
duke@435 | 279 | #endif |
duke@435 | 280 | |
duke@435 | 281 | #endif // PRODUCT |