ryan/mozjpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Re-factor and re-license under the libjpeg BSD-style license. Justification: the accelerated Huffman encoding optimizations in libjpeg-turbo were all developed by me as an independent developer. The structure of the inline Huffman encoding macros was originally borrowed from similar routines in the TurboJPEG/mediaLib codec, which is part of VirtualGL and TurboVNC. Thus, although the code for these macros was not copied verbatim, they were still thought to be a derivative work of TurboJPEG/mediaLib, and I assigned the copyright and license from TurboJPEG/mediaLib to them. I have re-written these routines from first principles by breaking down the libjpeg out-of-line routines. Although the new code bears algorithmic similarities to the TurboJPEG/mediaLib macros, it can now clearly be shown to be derived from the out-of-line routines and thus, in my opinion, it can no longer be considered a derivative of TurboJPEG/mediaLib. -- DRC

Browse Source
This commit is contained in:
DRC
2011-04-26 22:08:31 +00:00
parent 2540beb951
commit ed7ec8322c
1 changed files with 167 additions and 185 deletions
Download Patch File Download Diff File Expand all files Collapse all files

262
jchuff.c
View File

@@ -2,6 +2,7 @@
* jchuff.c * jchuff.c
* *
* Copyright (C) 1991-1997, Thomas G. Lane. * Copyright (C) 1991-1997, Thomas G. Lane.
* Copyright (C) 2009-2011, D. R. Commander.
* This file is part of the Independent JPEG Group's software. * This file is part of the Independent JPEG Group's software.
* For conditions of distribution and use, see the accompanying README file. * For conditions of distribution and use, see the accompanying README file.
* *
@@ -14,21 +15,6 @@
* permanent JPEG objects only upon successful completion of an MCU. * permanent JPEG objects only upon successful completion of an MCU.
*/ */
/* Modifications:
* Copyright (C)2007 Sun Microsystems, Inc.
* Copyright (C)2009 D. R. Commander
*
* This library is free software and may be redistributed and/or modified under
* the terms of the wxWindows Library License, Version 3.1 or (at your option)
* any later version. The full license is in the LICENSE.txt file included
* with this distribution.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* wxWindows Library License for more details.
*/
#define JPEG_INTERNALS #define JPEG_INTERNALS
#include "jinclude.h" #include "jinclude.h"
#include "jpeglib.h" #include "jpeglib.h"
@@ -36,12 +22,13 @@
#include <limits.h> #include <limits.h>
static unsigned char jpeg_first_bit_table[65536]; static unsigned char jpeg_first_bit_table[65536];
static int jpeg_first_bit_table_init=0; static int jpeg_first_bit_table_init = 0;
#ifndef min #ifndef min
#define min(a,b) ((a)<(b)?(a):(b)) #define min(a,b) ((a)<(b)?(a):(b))
#endif #endif
/* Expanded entropy encoder object for Huffman encoding. /* Expanded entropy encoder object for Huffman encoding.
* *
* The savable_state subrecord contains fields that change within an MCU, * The savable_state subrecord contains fields that change within an MCU,
@@ -181,7 +168,6 @@ start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
} }
/* Initialize bit buffer to empty */ /* Initialize bit buffer to empty */
entropy->saved.put_buffer = 0; entropy->saved.put_buffer = 0;
entropy->saved.put_bits = 0; entropy->saved.put_bits = 0;
@@ -325,30 +311,27 @@ dump_buffer (working_state * state)
/* Outputting bits to the file */ /* Outputting bits to the file */
/* Only the right 24 bits of put_buffer are used; the valid bits are /* These macros perform the same task as the emit_bits() function in the
* left-justified in this part. At most 16 bits can be passed to emit_bits * original libjpeg code. In addition to reducing overhead by explicitly
* in one call, and we never retain more than 7 bits in put_buffer * inlining the code, additional performance is achieved by taking into
* between calls, so 24 bits are sufficient. * account the size of the bit buffer and waiting until it is almost full
* before emptying it. This mostly benefits 64-bit platforms, since 6
* bytes can be stored in a 64-bit bit buffer before it has to be emptied.
*/ */
/***************************************************************/
#define EMIT_BYTE() { \ #define EMIT_BYTE() { \
if (0xFF == (*buffer++ = (unsigned char)(put_buffer >> (put_bits -= 8)))) \ JOCTET c; \
put_bits -= 8; \
c = (JOCTET)GETJOCTET(put_buffer >> put_bits); \
*buffer++ = c; \
if (c == 0xFF) /* need to stuff a zero byte? */ \
*buffer++ = 0; \ *buffer++ = 0; \
} }
/***************************************************************/ #define PUT_BITS(code, size) { \
#define DUMP_BITS_(code, size) { \
put_bits += size; \ put_bits += size; \
put_buffer = (put_buffer << size) | code; \ put_buffer = (put_buffer << size) | code; \
if (put_bits > 7) \ }
while(put_bits > 7) \
EMIT_BYTE() \
}
/***************************************************************/
#define CHECKBUF15() { \ #define CHECKBUF15() { \
if (put_bits > 15) { \ if (put_bits > 15) { \
@@ -357,6 +340,15 @@ dump_buffer (working_state * state)
} \ } \
} }
#define CHECKBUF31() { \
if (put_bits > 31) { \
EMIT_BYTE() \
EMIT_BYTE() \
EMIT_BYTE() \
EMIT_BYTE() \
} \
}
#define CHECKBUF47() { \ #define CHECKBUF47() { \
if (put_bits > 47) { \ if (put_bits > 47) { \
EMIT_BYTE() \ EMIT_BYTE() \
@@ -368,87 +360,37 @@ dump_buffer (working_state * state)
} \ } \
} }
#define CHECKBUF31() { \ #if __WORDSIZE==64 || defined(_WIN64)
if (put_bits > 31) { \
EMIT_BYTE() \ #define EMIT_BITS(code, size) { \
EMIT_BYTE() \ CHECKBUF47() \
EMIT_BYTE() \ PUT_BITS(code, size) \
EMIT_BYTE() \
} \
} }
/***************************************************************/ #define EMIT_CODE(code, size) { \
temp2 &= (((INT32) 1)<<nbits) - 1; \
#define DUMP_BITS_NOCHECK(code, size) { \
put_bits += size; \
put_buffer = (put_buffer << size) | code; \
}
#if __WORDSIZE==64 || defined(_WIN64)
#define DUMP_BITS(code, size) { \
CHECKBUF47() \
put_bits += size; \
put_buffer = (put_buffer << size) | code; \
}
#else
#define DUMP_BITS(code, size) { \
put_bits += size; \
put_buffer = (put_buffer << size) | code; \
CHECKBUF15() \
}
#endif
/***************************************************************/
#define DUMP_SINGLE_VALUE(ht, codevalue) { \
size = ht->ehufsi[codevalue]; \
code = ht->ehufco[codevalue]; \
\
DUMP_BITS(code, size) \
}
/***************************************************************/
#define DUMP_VALUE_SLOW(ht, codevalue, t, nbits) { \
size = ht->ehufsi[codevalue]; \
code = ht->ehufco[codevalue]; \
t &= ~(-1 << nbits); \
DUMP_BITS_NOCHECK(code, size) \
CHECKBUF15() \
DUMP_BITS_NOCHECK(t, nbits) \
CHECKBUF15() \
}
#if __WORDSIZE==64 || defined(_WIN64)
#define DUMP_VALUE(ht, codevalue, t, nbits) { \
size = ht->ehufsi[codevalue]; \
code = ht->ehufco[codevalue]; \
t &= ~(-1 << nbits); \
CHECKBUF31() \ CHECKBUF31() \
DUMP_BITS_NOCHECK(code, size) \ PUT_BITS(code, size) \
DUMP_BITS_NOCHECK(t, nbits) \ PUT_BITS(temp2, nbits) \
} }
#else #else
#define DUMP_VALUE(ht, codevalue, t, nbits) { \ #define EMIT_BITS(code, size) { \
size = ht->ehufsi[codevalue]; \ PUT_BITS(code, size) \
code = ht->ehufco[codevalue]; \
t &= ~(-1 << nbits); \
DUMP_BITS_NOCHECK(code, size) \
CHECKBUF15() \ CHECKBUF15() \
DUMP_BITS_NOCHECK(t, nbits) \ }
#define EMIT_CODE(code, size) { \
temp2 &= (((INT32) 1)<<nbits) - 1; \
PUT_BITS(code, size) \
CHECKBUF15() \
PUT_BITS(temp2, nbits) \
CHECKBUF15() \ CHECKBUF15() \
} }
#endif #endif
/***************************************************************/
#define BUFSIZE (DCTSIZE2 * 2) #define BUFSIZE (DCTSIZE2 * 2)
@@ -481,12 +423,11 @@ dump_buffer (working_state * state)
} \ } \
} }
/***************************************************************/
LOCAL(boolean) LOCAL(boolean)
flush_bits (working_state * state) flush_bits (working_state * state)
{ {
unsigned char _buffer[BUFSIZE], *buffer; JOCTET _buffer[BUFSIZE], *buffer;
size_t put_buffer; int put_bits; size_t put_buffer; int put_bits;
size_t bytes, bytestocopy; int localbuf = 0; size_t bytes, bytestocopy; int localbuf = 0;
@@ -494,7 +435,9 @@ flush_bits (working_state * state)
put_bits = state->cur.put_bits; put_bits = state->cur.put_bits;
LOAD_BUFFER() LOAD_BUFFER()
DUMP_BITS_(0x7F, 7) /* fill any partial byte with ones */
PUT_BITS(0x7F, 7)
while (put_bits >= 8) EMIT_BYTE()
state->cur.put_buffer = 0; /* and reset bit-buffer to empty */ state->cur.put_buffer = 0; /* and reset bit-buffer to empty */
state->cur.put_bits = 0; state->cur.put_bits = 0;
@@ -503,16 +446,17 @@ flush_bits (working_state * state)
return TRUE; return TRUE;
} }
/* Encode a single block's worth of coefficients */ /* Encode a single block's worth of coefficients */
LOCAL(boolean) LOCAL(boolean)
encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val, encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val,
c_derived_tbl *dctbl, c_derived_tbl *actbl) c_derived_tbl *dctbl, c_derived_tbl *actbl)
{ {
int temp, temp2; int temp, temp2, temp3;
int nbits; int nbits;
int r, sflag, size, code; int r, code, size;
unsigned char _buffer[BUFSIZE], *buffer; JOCTET _buffer[BUFSIZE], *buffer;
size_t put_buffer; int put_bits; size_t put_buffer; int put_bits;
int code_0xf0 = actbl->ehufco[0xf0], size_0xf0 = actbl->ehufsi[0xf0]; int code_0xf0 = actbl->ehufco[0xf0], size_0xf0 = actbl->ehufsi[0xf0];
size_t bytes, bytestocopy; int localbuf = 0; size_t bytes, bytestocopy; int localbuf = 0;
@@ -525,50 +469,88 @@ encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val,
temp = temp2 = block[0] - last_dc_val; temp = temp2 = block[0] - last_dc_val;
sflag = temp >> 31; /* This is a well-known technique for obtaining the absolute value without a
temp -= ((temp + temp) & sflag); * branch. It is derived from an assembly language technique presented in
temp2 += sflag; * "How to Optimize for the Pentium Processors", Copyright (c) 1996, 1997 by
* Agner Fog.
*/
temp3 = temp >> (CHAR_BIT * sizeof(int) - 1);
temp ^= temp3;
temp -= temp3;
/* For a negative input, want temp2 = bitwise complement of abs(input) */
/* This code assumes we are on a two's complement machine */
temp2 += temp3;
/* Find the number of bits needed for the magnitude of the coefficient */
nbits = jpeg_first_bit_table[temp]; nbits = jpeg_first_bit_table[temp];
DUMP_VALUE_SLOW(dctbl, nbits, temp2, nbits)
/* Emit the Huffman-coded symbol for the number of bits */
code = dctbl->ehufco[nbits];
size = dctbl->ehufsi[nbits];
PUT_BITS(code, size)
CHECKBUF15()
/* Mask off any extra bits in code */
temp2 &= (((INT32) 1)<<nbits) - 1;
/* Emit that number of bits of the value, if positive, */
/* or the complement of its magnitude, if negative. */
PUT_BITS(temp2, nbits)
CHECKBUF15()
/* Encode the AC coefficients per section F.1.2.2 */ /* Encode the AC coefficients per section F.1.2.2 */
r = 0; /* r = run length of zeros */ r = 0; /* r = run length of zeros */
#define innerloop(order) { \ /* Manually unroll the k loop to eliminate the counter variable. This
temp2 = *(JCOEF*)((unsigned char*)block + order); \ * improves performance greatly on systems with a limited number of
if(temp2 == 0) r++; \ * registers (such as x86.)
else { \ */
temp = (JCOEF)temp2; \ #define kloop(jpeg_natural_order_of_k) { \
sflag = temp >> 31; \ if ((temp = block[jpeg_natural_order_of_k]) == 0) { \
temp = (temp ^ sflag) - sflag; \ r++; \
temp2 += sflag; \ } else { \
temp2 = temp; \
/* Branch-less absolute value, bitwise complement, etc., same as above */ \
temp3 = temp >> (CHAR_BIT * sizeof(int) - 1); \
temp ^= temp3; \
temp -= temp3; \
temp2 += temp3; \
nbits = jpeg_first_bit_table[temp]; \ nbits = jpeg_first_bit_table[temp]; \
for(; r > 15; r -= 16) DUMP_BITS(code_0xf0, size_0xf0) \ /* if run length > 15, must emit special run-length-16 codes (0xF0) */ \
sflag = (r << 4) + nbits; \ while (r > 15) { \
DUMP_VALUE(actbl, sflag, temp2, nbits) \ EMIT_BITS(code_0xf0, size_0xf0) \
r -= 16; \
} \
/* Emit Huffman symbol for run length / number of bits */ \
temp3 = (r << 4) + nbits; \
code = actbl->ehufco[temp3]; \
size = actbl->ehufsi[temp3]; \
EMIT_CODE(code, size) \
r = 0; \ r = 0; \
}} } \
}
innerloop(2*1); innerloop(2*8); innerloop(2*16); innerloop(2*9); /* One iteration for each value in jpeg_natural_order[] */
innerloop(2*2); innerloop(2*3); innerloop(2*10); innerloop(2*17); kloop(1); kloop(8); kloop(16); kloop(9); kloop(2); kloop(3);
innerloop(2*24); innerloop(2*32); innerloop(2*25); innerloop(2*18); kloop(10); kloop(17); kloop(24); kloop(32); kloop(25); kloop(18);
innerloop(2*11); innerloop(2*4); innerloop(2*5); innerloop(2*12); kloop(11); kloop(4); kloop(5); kloop(12); kloop(19); kloop(26);
innerloop(2*19); innerloop(2*26); innerloop(2*33); innerloop(2*40); kloop(33); kloop(40); kloop(48); kloop(41); kloop(34); kloop(27);
innerloop(2*48); innerloop(2*41); innerloop(2*34); innerloop(2*27); kloop(20); kloop(13); kloop(6); kloop(7); kloop(14); kloop(21);
innerloop(2*20); innerloop(2*13); innerloop(2*6); innerloop(2*7); kloop(28); kloop(35); kloop(42); kloop(49); kloop(56); kloop(57);
innerloop(2*14); innerloop(2*21); innerloop(2*28); innerloop(2*35); kloop(50); kloop(43); kloop(36); kloop(29); kloop(22); kloop(15);
innerloop(2*42); innerloop(2*49); innerloop(2*56); innerloop(2*57); kloop(23); kloop(30); kloop(37); kloop(44); kloop(51); kloop(58);
innerloop(2*50); innerloop(2*43); innerloop(2*36); innerloop(2*29); kloop(59); kloop(52); kloop(45); kloop(38); kloop(31); kloop(39);
innerloop(2*22); innerloop(2*15); innerloop(2*23); innerloop(2*30); kloop(46); kloop(53); kloop(60); kloop(61); kloop(54); kloop(47);
innerloop(2*37); innerloop(2*44); innerloop(2*51); innerloop(2*58); kloop(55); kloop(62); kloop(63);
innerloop(2*59); innerloop(2*52); innerloop(2*45); innerloop(2*38);
innerloop(2*31); innerloop(2*39); innerloop(2*46); innerloop(2*53);
innerloop(2*60); innerloop(2*61); innerloop(2*54); innerloop(2*47);
innerloop(2*55); innerloop(2*62); innerloop(2*63);
/* If the last coef(s) were zero, emit an end-of-block code */ /* If the last coef(s) were zero, emit an end-of-block code */
if (r > 0) DUMP_SINGLE_VALUE(actbl, 0x0) if (r > 0) {
code = actbl->ehufco[0];
size = actbl->ehufsi[0];
EMIT_BITS(code, size)
}
state->cur.put_buffer = put_buffer; state->cur.put_buffer = put_buffer;
state->cur.put_bits = put_bits; state->cur.put_bits = put_bits;