duke@435: /*
trims@1907:  * Copyright (c) 1997, 2006, Oracle and/or its affiliates. All rights reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
trims@1907:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907:  * or visit www.oracle.com if you need additional information or have any
trims@1907:  * questions.
duke@435:  *
duke@435:  */
duke@435: 
duke@435: #include "incls/_precompiled.incl"
duke@435: #include "incls/_compressedStream.cpp.incl"
duke@435: 
duke@435: // 32-bit one-to-one sign encoding taken from Pack200
duke@435: // converts leading sign bits into leading zeroes with trailing sign bit
duke@435: inline juint CompressedStream::encode_sign(jint  value) {
duke@435:   return (value << 1) ^ (value >> 31);
duke@435: }
duke@435: inline jint  CompressedStream::decode_sign(juint value) {
duke@435:   return (value >> 1) ^ -(jint)(value & 1);
duke@435: }
duke@435: 
duke@435: // 32-bit self-inverse encoding of float bits
duke@435: // converts trailing zeroes (common in floats) to leading zeroes
duke@435: inline juint CompressedStream::reverse_int(juint i) {
duke@435:   // Hacker's Delight, Figure 7-1
duke@435:   i = (i & 0x55555555) << 1 | (i >> 1) & 0x55555555;
duke@435:   i = (i & 0x33333333) << 2 | (i >> 2) & 0x33333333;
duke@435:   i = (i & 0x0f0f0f0f) << 4 | (i >> 4) & 0x0f0f0f0f;
duke@435:   i = (i << 24) | ((i & 0xff00) << 8) | ((i >> 8) & 0xff00) | (i >> 24);
duke@435:   return i;
duke@435: }
duke@435: 
duke@435: 
duke@435: jint CompressedReadStream::read_signed_int() {
duke@435:   return decode_sign(read_int());
duke@435: }
duke@435: 
duke@435: // Compressing floats is simple, because the only common pattern
duke@435: // is trailing zeroes.  (Compare leading sign bits on ints.)
duke@435: // Since floats are left-justified, as opposed to right-justified
duke@435: // ints, we can bit-reverse them in order to take advantage of int
duke@435: // compression.
duke@435: 
duke@435: jfloat CompressedReadStream::read_float() {
duke@435:   int rf = read_int();
duke@435:   int f  = reverse_int(rf);
duke@435:   return jfloat_cast(f);
duke@435: }
duke@435: 
duke@435: jdouble CompressedReadStream::read_double() {
duke@435:   jint rh = read_int();
duke@435:   jint rl = read_int();
duke@435:   jint h  = reverse_int(rh);
duke@435:   jint l  = reverse_int(rl);
duke@435:   return jdouble_cast(jlong_from(h, l));
duke@435: }
duke@435: 
duke@435: jlong CompressedReadStream::read_long() {
duke@435:   jint low  = read_signed_int();
duke@435:   jint high = read_signed_int();
duke@435:   return jlong_from(high, low);
duke@435: }
duke@435: 
duke@435: CompressedWriteStream::CompressedWriteStream(int initial_size) : CompressedStream(NULL, 0) {
duke@435:   _buffer   = NEW_RESOURCE_ARRAY(u_char, initial_size);
duke@435:   _size     = initial_size;
duke@435:   _position = 0;
duke@435: }
duke@435: 
duke@435: void CompressedWriteStream::grow() {
duke@435:   u_char* _new_buffer = NEW_RESOURCE_ARRAY(u_char, _size * 2);
duke@435:   memcpy(_new_buffer, _buffer, _position);
duke@435:   _buffer = _new_buffer;
duke@435:   _size   = _size * 2;
duke@435: }
duke@435: 
duke@435: void CompressedWriteStream::write_signed_int(jint value) {
duke@435:   // this encoding, called SIGNED5, is taken from Pack200
duke@435:   write_int(encode_sign(value));
duke@435: }
duke@435: 
duke@435: void CompressedWriteStream::write_float(jfloat value) {
duke@435:   juint f = jint_cast(value);
duke@435:   juint rf = reverse_int(f);
duke@435:   assert(f == reverse_int(rf), "can re-read same bits");
duke@435:   write_int(rf);
duke@435: }
duke@435: 
duke@435: void CompressedWriteStream::write_double(jdouble value) {
duke@435:   juint h  = high(jlong_cast(value));
duke@435:   juint l  = low( jlong_cast(value));
duke@435:   juint rh = reverse_int(h);
duke@435:   juint rl = reverse_int(l);
duke@435:   assert(h == reverse_int(rh), "can re-read same bits");
duke@435:   assert(l == reverse_int(rl), "can re-read same bits");
duke@435:   write_int(rh);
duke@435:   write_int(rl);
duke@435: }
duke@435: 
duke@435: void CompressedWriteStream::write_long(jlong value) {
duke@435:   write_signed_int(low(value));
duke@435:   write_signed_int(high(value));
duke@435: }
duke@435: 
duke@435: 
duke@435: /// The remaining details
duke@435: 
duke@435: #ifndef PRODUCT
duke@435: // set this to trigger unit test
duke@435: void test_compressed_stream(int trace);
duke@435: bool test_compressed_stream_enabled = false;
duke@435: #endif
duke@435: 
duke@435: // This encoding, called UNSIGNED5, is taken from J2SE Pack200.
duke@435: // It assumes that most values have lots of leading zeroes.
duke@435: // Very small values, in the range [0..191], code in one byte.
duke@435: // Any 32-bit value (including negatives) can be coded, in
duke@435: // up to five bytes.  The grammar is:
duke@435: //    low_byte  = [0..191]
duke@435: //    high_byte = [192..255]
duke@435: //    any_byte  = low_byte | high_byte
duke@435: //    coding = low_byte
duke@435: //           | high_byte low_byte
duke@435: //           | high_byte high_byte low_byte
duke@435: //           | high_byte high_byte high_byte low_byte
duke@435: //           | high_byte high_byte high_byte high_byte any_byte
duke@435: // Each high_byte contributes six bits of payload.
duke@435: // The encoding is one-to-one (except for integer overflow)
duke@435: // and easy to parse and unparse.
duke@435: 
duke@435: jint CompressedReadStream::read_int_mb(jint b0) {
duke@435:   int     pos = position() - 1;
duke@435:   u_char* buf = buffer() + pos;
duke@435:   assert(buf[0] == b0 && b0 >= L, "correctly called");
duke@435:   jint    sum = b0;
duke@435:   // must collect more bytes:  b[1]...b[4]
duke@435:   int lg_H_i = lg_H;
duke@435:   for (int i = 0; ; ) {
duke@435:     jint b_i = buf[++i]; // b_i = read(); ++i;
duke@435:     sum += b_i << lg_H_i;  // sum += b[i]*(64**i)
duke@435:     if (b_i < L || i == MAX_i) {
duke@435:       set_position(pos+i+1);
duke@435:       return sum;
duke@435:     }
duke@435:     lg_H_i += lg_H;
duke@435:   }
duke@435: }
duke@435: 
duke@435: void CompressedWriteStream::write_int_mb(jint value) {
duke@435:   debug_only(int pos1 = position());
duke@435:   juint sum = value;
duke@435:   for (int i = 0; ; ) {
duke@435:     if (sum < L || i == MAX_i) {
duke@435:       // remainder is either a "low code" or the 5th byte
duke@435:       assert(sum == (u_char)sum, "valid byte");
duke@435:       write((u_char)sum);
duke@435:       break;
duke@435:     }
duke@435:     sum -= L;
duke@435:     int b_i = L + (sum % H);  // this is a "high code"
duke@435:     sum >>= lg_H;             // extracted 6 bits
duke@435:     write(b_i); ++i;
duke@435:   }
duke@435: 
duke@435: #ifndef PRODUCT
duke@435:   if (test_compressed_stream_enabled) {  // hack to enable this stress test
duke@435:     test_compressed_stream_enabled = false;
duke@435:     test_compressed_stream(0);
duke@435:   }
duke@435: #endif
duke@435: }
duke@435: 
duke@435: 
duke@435: #ifndef PRODUCT
duke@435: /// a unit test (can be run by hand from a debugger)
duke@435: 
duke@435: // Avoid a VS2005 compiler stack overflow w/ fastdebug build.
duke@435: // The following pragma optimize turns off optimization ONLY
duke@435: // for this block (a matching directive turns it back on later).
duke@435: // These directives can be removed once the MS VS.NET 2005
duke@435: // compiler stack overflow is fixed.
duke@435: #if _MSC_VER >=1400 && !defined(_WIN64)
duke@435: #pragma optimize("", off)
duke@435: #endif
duke@435: 
duke@435: // generator for an "interesting" set of critical values
duke@435: enum { stretch_limit = (1<<16) * (64-16+1) };
duke@435: static jlong stretch(jint x, int bits) {
duke@435:   // put x[high 4] into place
duke@435:   jlong h = (jlong)((x >> (16-4))) << (bits - 4);
duke@435:   // put x[low 12] into place, sign extended
duke@435:   jlong l = ((jlong)x << (64-12)) >> (64-12);
duke@435:   // move l upwards, maybe
duke@435:   l <<= (x >> 16);
duke@435:   return h ^ l;
duke@435: }
duke@435: 
duke@435: void test_compressed_stream(int trace) {
duke@435:   CompressedWriteStream bytes(stretch_limit * 100);
duke@435:   jint n;
duke@435:   int step = 0, fails = 0;
duke@435: #define CHECKXY(x, y, fmt) { \
duke@435:     ++step; \
duke@435:     int xlen = (pos = decode.position()) - lastpos; lastpos = pos; \
duke@435:     if (trace > 0 && (step % trace) == 0) { \
duke@435:       tty->print_cr("step %d, n=%08x: value=" fmt " (len=%d)", \
duke@435:                     step, n, x, xlen); } \
duke@435:     if (x != y) {                                                     \
duke@435:       tty->print_cr("step %d, n=%d: " fmt " != " fmt, step, n, x, y); \
duke@435:       fails++; \
duke@435:     } }
duke@435:   for (n = 0; n < (1<<8); n++) {
duke@435:     jbyte x = (jbyte)n;
duke@435:     bytes.write_byte(x); ++step;
duke@435:   }
duke@435:   for (n = 0; n < stretch_limit; n++) {
duke@435:     jint x = (jint)stretch(n, 32);
duke@435:     bytes.write_int(x); ++step;
duke@435:     bytes.write_signed_int(x); ++step;
duke@435:     bytes.write_float(jfloat_cast(x)); ++step;
duke@435:   }
duke@435:   for (n = 0; n < stretch_limit; n++) {
duke@435:     jlong x = stretch(n, 64);
duke@435:     bytes.write_long(x); ++step;
duke@435:     bytes.write_double(jdouble_cast(x)); ++step;
duke@435:   }
duke@435:   int length = bytes.position();
duke@435:   if (trace != 0)
duke@435:     tty->print_cr("set up test of %d stream values, size %d", step, length);
duke@435:   step = 0;
duke@435:   // now decode it all
duke@435:   CompressedReadStream decode(bytes.buffer());
duke@435:   int pos, lastpos = decode.position();
duke@435:   for (n = 0; n < (1<<8); n++) {
duke@435:     jbyte x = (jbyte)n;
duke@435:     jbyte y = decode.read_byte();
duke@435:     CHECKXY(x, y, "%db");
duke@435:   }
duke@435:   for (n = 0; n < stretch_limit; n++) {
duke@435:     jint x = (jint)stretch(n, 32);
duke@435:     jint y1 = decode.read_int();
duke@435:     CHECKXY(x, y1, "%du");
duke@435:     jint y2 = decode.read_signed_int();
duke@435:     CHECKXY(x, y2, "%di");
duke@435:     jint y3 = jint_cast(decode.read_float());
duke@435:     CHECKXY(x, y3, "%df");
duke@435:   }
duke@435:   for (n = 0; n < stretch_limit; n++) {
duke@435:     jlong x = stretch(n, 64);
duke@435:     jlong y1 = decode.read_long();
duke@435:     CHECKXY(x, y1, INT64_FORMAT "l");
duke@435:     jlong y2 = jlong_cast(decode.read_double());
duke@435:     CHECKXY(x, y2, INT64_FORMAT "d");
duke@435:   }
duke@435:   int length2 = decode.position();
duke@435:   if (trace != 0)
duke@435:     tty->print_cr("finished test of %d stream values, size %d", step, length2);
duke@435:   guarantee(length == length2, "bad length");
duke@435:   guarantee(fails == 0, "test failures");
duke@435: }
duke@435: 
duke@435: #if _MSC_VER >=1400 && !defined(_WIN64)
duke@435: #pragma optimize("", on)
duke@435: #endif
duke@435: 
duke@435: #endif // PRODUCT