Chris@69: /* Copyright (c) 2001-2011 Timothy B. Terriberry
Chris@69: */
Chris@69: /*
Chris@69:    Redistribution and use in source and binary forms, with or without
Chris@69:    modification, are permitted provided that the following conditions
Chris@69:    are met:
Chris@69: 
Chris@69:    - Redistributions of source code must retain the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer.
Chris@69: 
Chris@69:    - Redistributions in binary form must reproduce the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer in the
Chris@69:    documentation and/or other materials provided with the distribution.
Chris@69: 
Chris@69:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
Chris@69:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
Chris@69:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
Chris@69:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
Chris@69:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
Chris@69:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
Chris@69:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
Chris@69:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
Chris@69:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
Chris@69:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
Chris@69:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Chris@69: */
Chris@69: 
Chris@69: #ifdef HAVE_CONFIG_H
Chris@69: #include "config.h"
Chris@69: #endif
Chris@69: 
Chris@69: #include "entcode.h"
Chris@69: #include "arch.h"
Chris@69: 
Chris@69: #if !defined(EC_CLZ)
Chris@69: /*This is a fallback for systems where we don't know how to access
Chris@69:    a BSR or CLZ instruction (see ecintrin.h).
Chris@69:   If you are optimizing Opus on a new platform and it has a native CLZ or
Chris@69:    BZR (e.g. cell, MIPS, x86, etc) then making it available to Opus will be
Chris@69:    an easy performance win.*/
Chris@69: int ec_ilog(opus_uint32 _v){
Chris@69:   /*On a Pentium M, this branchless version tested as the fastest on
Chris@69:      1,000,000,000 random 32-bit integers, edging out a similar version with
Chris@69:      branches, and a 256-entry LUT version.*/
Chris@69:   int ret;
Chris@69:   int m;
Chris@69:   ret=!!_v;
Chris@69:   m=!!(_v&0xFFFF0000)<<4;
Chris@69:   _v>>=m;
Chris@69:   ret|=m;
Chris@69:   m=!!(_v&0xFF00)<<3;
Chris@69:   _v>>=m;
Chris@69:   ret|=m;
Chris@69:   m=!!(_v&0xF0)<<2;
Chris@69:   _v>>=m;
Chris@69:   ret|=m;
Chris@69:   m=!!(_v&0xC)<<1;
Chris@69:   _v>>=m;
Chris@69:   ret|=m;
Chris@69:   ret+=!!(_v&0x2);
Chris@69:   return ret;
Chris@69: }
Chris@69: #endif
Chris@69: 
Chris@69: #if 1
Chris@69: /* This is a faster version of ec_tell_frac() that takes advantage
Chris@69:    of the low (1/8 bit) resolution to use just a linear function
Chris@69:    followed by a lookup to determine the exact transition thresholds. */
Chris@69: opus_uint32 ec_tell_frac(ec_ctx *_this){
Chris@69:   static const unsigned correction[8] =
Chris@69:     {35733, 38967, 42495, 46340,
Chris@69:      50535, 55109, 60097, 65535};
Chris@69:   opus_uint32 nbits;
Chris@69:   opus_uint32 r;
Chris@69:   int         l;
Chris@69:   unsigned    b;
Chris@69:   nbits=_this->nbits_total<<BITRES;
Chris@69:   l=EC_ILOG(_this->rng);
Chris@69:   r=_this->rng>>(l-16);
Chris@69:   b = (r>>12)-8;
Chris@69:   b += r>correction[b];
Chris@69:   l = (l<<3)+b;
Chris@69:   return nbits-l;
Chris@69: }
Chris@69: #else
Chris@69: opus_uint32 ec_tell_frac(ec_ctx *_this){
Chris@69:   opus_uint32 nbits;
Chris@69:   opus_uint32 r;
Chris@69:   int         l;
Chris@69:   int         i;
Chris@69:   /*To handle the non-integral number of bits still left in the encoder/decoder
Chris@69:      state, we compute the worst-case number of bits of val that must be
Chris@69:      encoded to ensure that the value is inside the range for any possible
Chris@69:      subsequent bits.
Chris@69:     The computation here is independent of val itself (the decoder does not
Chris@69:      even track that value), even though the real number of bits used after
Chris@69:      ec_enc_done() may be 1 smaller if rng is a power of two and the
Chris@69:      corresponding trailing bits of val are all zeros.
Chris@69:     If we did try to track that special case, then coding a value with a
Chris@69:      probability of 1/(1<<n) might sometimes appear to use more than n bits.
Chris@69:     This may help explain the surprising result that a newly initialized
Chris@69:      encoder or decoder claims to have used 1 bit.*/
Chris@69:   nbits=_this->nbits_total<<BITRES;
Chris@69:   l=EC_ILOG(_this->rng);
Chris@69:   r=_this->rng>>(l-16);
Chris@69:   for(i=BITRES;i-->0;){
Chris@69:     int b;
Chris@69:     r=r*r>>15;
Chris@69:     b=(int)(r>>16);
Chris@69:     l=l<<1|b;
Chris@69:     r>>=b;
Chris@69:   }
Chris@69:   return nbits-l;
Chris@69: }
Chris@69: #endif
Chris@69: 
Chris@69: #ifdef USE_SMALL_DIV_TABLE
Chris@69: /* Result of 2^32/(2*i+1), except for i=0. */
Chris@69: const opus_uint32 SMALL_DIV_TABLE[129] = {
Chris@69:    0xFFFFFFFF, 0x55555555, 0x33333333, 0x24924924,
Chris@69:    0x1C71C71C, 0x1745D174, 0x13B13B13, 0x11111111,
Chris@69:    0x0F0F0F0F, 0x0D79435E, 0x0C30C30C, 0x0B21642C,
Chris@69:    0x0A3D70A3, 0x097B425E, 0x08D3DCB0, 0x08421084,
Chris@69:    0x07C1F07C, 0x07507507, 0x06EB3E45, 0x06906906,
Chris@69:    0x063E7063, 0x05F417D0, 0x05B05B05, 0x0572620A,
Chris@69:    0x05397829, 0x05050505, 0x04D4873E, 0x04A7904A,
Chris@69:    0x047DC11F, 0x0456C797, 0x04325C53, 0x04104104,
Chris@69:    0x03F03F03, 0x03D22635, 0x03B5CC0E, 0x039B0AD1,
Chris@69:    0x0381C0E0, 0x0369D036, 0x03531DEC, 0x033D91D2,
Chris@69:    0x0329161F, 0x03159721, 0x03030303, 0x02F14990,
Chris@69:    0x02E05C0B, 0x02D02D02, 0x02C0B02C, 0x02B1DA46,
Chris@69:    0x02A3A0FD, 0x0295FAD4, 0x0288DF0C, 0x027C4597,
Chris@69:    0x02702702, 0x02647C69, 0x02593F69, 0x024E6A17,
Chris@69:    0x0243F6F0, 0x0239E0D5, 0x02302302, 0x0226B902,
Chris@69:    0x021D9EAD, 0x0214D021, 0x020C49BA, 0x02040810,
Chris@69:    0x01FC07F0, 0x01F44659, 0x01ECC07B, 0x01E573AC,
Chris@69:    0x01DE5D6E, 0x01D77B65, 0x01D0CB58, 0x01CA4B30,
Chris@69:    0x01C3F8F0, 0x01BDD2B8, 0x01B7D6C3, 0x01B20364,
Chris@69:    0x01AC5701, 0x01A6D01A, 0x01A16D3F, 0x019C2D14,
Chris@69:    0x01970E4F, 0x01920FB4, 0x018D3018, 0x01886E5F,
Chris@69:    0x0183C977, 0x017F405F, 0x017AD220, 0x01767DCE,
Chris@69:    0x01724287, 0x016E1F76, 0x016A13CD, 0x01661EC6,
Chris@69:    0x01623FA7, 0x015E75BB, 0x015AC056, 0x01571ED3,
Chris@69:    0x01539094, 0x01501501, 0x014CAB88, 0x0149539E,
Chris@69:    0x01460CBC, 0x0142D662, 0x013FB013, 0x013C995A,
Chris@69:    0x013991C2, 0x013698DF, 0x0133AE45, 0x0130D190,
Chris@69:    0x012E025C, 0x012B404A, 0x01288B01, 0x0125E227,
Chris@69:    0x01234567, 0x0120B470, 0x011E2EF3, 0x011BB4A4,
Chris@69:    0x01194538, 0x0116E068, 0x011485F0, 0x0112358E,
Chris@69:    0x010FEF01, 0x010DB20A, 0x010B7E6E, 0x010953F3,
Chris@69:    0x01073260, 0x0105197F, 0x0103091B, 0x01010101
Chris@69: };
Chris@69: #endif