/*
tabinit.c: initialize tables...
copyright ?-2008 by the mpg123 project - free software under the terms of the LGPL 2.1
see COPYING and AUTHORS files in distribution or http://mpg123.org
initially written by Michael Hipp
*/
#include "mpg123lib_intern.h"
#include "debug.h"
/* That altivec alignment part here should not hurt generic code, I hope */
#ifdef OPT_ALTIVEC
static ALIGNED(16) real cos64[16];
static ALIGNED(16) real cos32[8];
static ALIGNED(16) real cos16[4];
static ALIGNED(16) real cos8[2];
static ALIGNED(16) real cos4[1];
#elif defined(REAL_IS_FIXED) && defined(PRECALC_TABLES)
static real cos64[16] =
{
8398725,8480395,8647771,8909416,9279544,9780026,10443886,11321405,
12491246,14081950,16316987,19619946,24900150,34523836,57170182,170959967
};
static real cos32[8] =
{
8429197,8766072,9511743,10851869,13223040,17795219,28897867,85583072
};
static real cos16[4] =
{
8552951,10088893,15099095,42998586
};
static real cos8[2] =
{
9079764,21920489
};
static real cos4[1] =
{
11863283
};
#else
static real cos64[16],cos32[8],cos16[4],cos8[2],cos4[1];
#endif
real *pnts[] = { cos64,cos32,cos16,cos8,cos4 };
static long intwinbase[] = {
0, -1, -1, -1, -1, -1, -1, -2, -2, -2,
-2, -3, -3, -4, -4, -5, -5, -6, -7, -7,
-8, -9, -10, -11, -13, -14, -16, -17, -19, -21,
-24, -26, -29, -31, -35, -38, -41, -45, -49, -53,
-58, -63, -68, -73, -79, -85, -91, -97, -104, -111,
-117, -125, -132, -139, -147, -154, -161, -169, -176, -183,
-190, -196, -202, -208, -213, -218, -222, -225, -227, -228,
-228, -227, -224, -221, -215, -208, -200, -189, -177, -163,
-146, -127, -106, -83, -57, -29, 2, 36, 72, 111,
153, 197, 244, 294, 347, 401, 459, 519, 581, 645,
711, 779, 848, 919, 991, 1064, 1137, 1210, 1283, 1356,
1428, 1498, 1567, 1634, 1698, 1759, 1817, 1870, 1919, 1962,
2001, 2032, 2057, 2075, 2085, 2087, 2080, 2063, 2037, 2000,
1952, 1893, 1822, 1739, 1644, 1535, 1414, 1280, 1131, 970,
794, 605, 402, 185, -45, -288, -545, -814, -1095, -1388,
-1692, -2006, -2330, -2663, -3004, -3351, -3705, -4063, -4425, -4788,
-5153, -5517, -5879, -6237, -6589, -6935, -7271, -7597, -7910, -8209,
-8491, -8755, -8998, -9219, -9416, -9585, -9727, -9838, -9916, -9959,
-9966, -9935, -9863, -9750, -9592, -9389, -9139, -8840, -8492, -8092,
-7640, -7134, -6574, -5959, -5288, -4561, -3776, -2935, -2037, -1082,
-70, 998, 2122, 3300, 4533, 5818, 7154, 8540, 9975, 11455,
12980, 14548, 16155, 17799, 19478, 21189, 22929, 24694, 26482, 28289,
30112, 31947, 33791, 35640, 37489, 39336, 41176, 43006, 44821, 46617,
48390, 50137, 51853, 53534, 55178, 56778, 58333, 59838, 61289, 62684,
64019, 65290, 66494, 67629, 68692, 69679, 70590, 71420, 72169, 72835,
73415, 73908, 74313, 74630, 74856, 74992, 75038 };
void prepare_decode_tables()
{
#if !defined(REAL_IS_FIXED) || !defined(PRECALC_TABLES)
int i,k,kr,divv;
real *costab;
for(i=0;i<5;i++)
{
kr=0x10>>i; divv=0x40>>i;
costab = pnts[i];
for(k=0;k<kr;k++)
costab[k] = DOUBLE_TO_REAL(1.0 / (2.0 * cos(M_PI * ((double) k * 2.0 + 1.0) / (double) divv)));
}
#endif
}
#ifdef OPT_MMXORSSE
#if !defined(OPT_X86_64) && !defined(OPT_NEON) && !defined(OPT_NEON64) && !defined(OPT_AVX)
void make_decode_tables_mmx_asm(long scaleval, float* decwin_mmx, float *decwins);
void make_decode_tables_mmx(mpg123_handle *fr)
{
debug("MMX decode tables");
/* Take care: The scale should be like before, when we didn't have float output all around. */
make_decode_tables_mmx_asm((long)((fr->lastscale < 0 ? fr->p.outscale : fr->lastscale)*SHORT_SCALE), fr->decwin_mmx, fr->decwins);
debug("MMX decode tables done");
}
#else
/* This mimics round() as defined in C99. We stay C89. */
static int rounded(double f)
{
return (int)(f>0 ? floor(f+0.5) : ceil(f-0.5));
}
/* x86-64 doesn't use asm version */
void make_decode_tables_mmx(mpg123_handle *fr)
{
int i,j,val;
int idx = 0;
short *ptr = (short *)fr->decwins;
/* Scale is always based on 1.0 . */
double scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
debug1("MMX decode tables with scaleval %g", scaleval);
for(i=0,j=0;i<256;i++,j++,idx+=32)
{
if(idx < 512+16)
fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
if(i % 32 == 31)
idx -= 1023;
if(i % 64 == 63)
scaleval = - scaleval;
}
for( /* i=256 */ ;i<512;i++,j--,idx+=32)
{
if(idx < 512+16)
fr->decwin_mmx[idx+16] = fr->decwin_mmx[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
if(i % 32 == 31)
idx -= 1023;
if(i % 64 == 63)
scaleval = - scaleval;
}
for(i=0; i<512; i++) {
if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
else val = rounded(fr->decwin_mmx[i]*-0.5);
if(val > 32767) val = 32767;
else if(val < -32768) val = -32768;
ptr[i] = val;
}
for(i=512; i<512+32; i++) {
if(i&1) val = rounded(fr->decwin_mmx[i]*0.5);
else val = 0;
if(val > 32767) val = 32767;
else if(val < -32768) val = -32768;
ptr[i] = val;
}
for(i=0; i<512; i++) {
val = rounded(fr->decwin_mmx[511-i]*-0.5);
if(val > 32767) val = 32767;
else if(val < -32768) val = -32768;
ptr[512+32+i] = val;
}
debug("decode tables done");
}
#endif
#endif
#ifdef REAL_IS_FIXED
/* Need saturating multiplication that keeps table values in 32 bit range,
with the option to swap sign at will (so -2**31 is out).
This code is far from the decoder core and so assembly optimization might
be overkill. */
static int32_t sat_mul32(int32_t a, int32_t b)
{
int64_t prod = (int64_t)a * (int64_t)b;
/* TODO: record the clipping? An extra flag? */
if(prod > 2147483647L) return 2147483647L;
if(prod < -2147483647L) return -2147483647L;
return (int32_t)prod;
}
#endif
void make_decode_tables(mpg123_handle *fr)
{
int i,j;
int idx = 0;
double scaleval;
#ifdef REAL_IS_FIXED
real scaleval_long;
#endif
/* Scale is always based on 1.0 . */
scaleval = -0.5*(fr->lastscale < 0 ? fr->p.outscale : fr->lastscale);
debug1("decode tables with scaleval %g", scaleval);
#ifdef REAL_IS_FIXED
scaleval_long = DOUBLE_TO_REAL_15(scaleval);
debug1("decode table with fixed scaleval %li", (long)scaleval_long);
if(scaleval_long > 28618 || scaleval_long < -28618)
{
/* TODO: Limit the scaleval itself or limit the multiplication afterwards?
The former basically disables significant amplification for fixed-point
decoders, but avoids (possibly subtle) distortion. */
/* This would limit the amplification instead:
scaleval_long = scaleval_long < 0 ? -28618 : 28618; */
if(NOQUIET) warning("Desired amplification may introduce distortion.");
}
#endif
for(i=0,j=0;i<256;i++,j++,idx+=32)
{
if(idx < 512+16)
#ifdef REAL_IS_FIXED
fr->decwin[idx+16] = fr->decwin[idx] =
REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
#else
fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
#endif
if(i % 32 == 31)
idx -= 1023;
if(i % 64 == 63)
#ifdef REAL_IS_FIXED
scaleval_long = - scaleval_long;
#else
scaleval = - scaleval;
#endif
}
for( /* i=256 */ ;i<512;i++,j--,idx+=32)
{
if(idx < 512+16)
#ifdef REAL_IS_FIXED
fr->decwin[idx+16] = fr->decwin[idx] =
REAL_SCALE_WINDOW(sat_mul32(intwinbase[j],scaleval_long));
#else
fr->decwin[idx+16] = fr->decwin[idx] = DOUBLE_TO_REAL((double) intwinbase[j] * scaleval);
#endif
if(i % 32 == 31)
idx -= 1023;
if(i % 64 == 63)
#ifdef REAL_IS_FIXED
scaleval_long = - scaleval_long;
#else
scaleval = - scaleval;
#endif
}
#if defined(OPT_X86_64) || defined(OPT_ALTIVEC) || defined(OPT_SSE) || defined(OPT_SSE_VINTAGE) || defined(OPT_ARM) || defined(OPT_NEON) || defined(OPT_NEON64) || defined(OPT_AVX)
if( fr->cpu_opts.type == x86_64
|| fr->cpu_opts.type == altivec
|| fr->cpu_opts.type == sse
|| fr->cpu_opts.type == sse_vintage
|| fr->cpu_opts.type == arm
|| fr->cpu_opts.type == neon
|| fr->cpu_opts.type == neon64
|| fr->cpu_opts.type == avx )
{ /* for float SSE / AltiVec / ARM decoder */
for(i=512; i<512+32; i++)
{
fr->decwin[i] = (i&1) ? fr->decwin[i] : 0;
}
for(i=0; i<512; i++)
{
fr->decwin[512+32+i] = -fr->decwin[511-i];
}
#if defined(OPT_NEON) || defined(OPT_NEON64)
if(fr->cpu_opts.type == neon || fr->cpu_opts.type == neon64)
{
for(i=0; i<512; i+=2)
{
fr->decwin[i] = -fr->decwin[i];
}
}
#endif
}
#endif
debug("decode tables done");
}
#ifndef NO_8BIT
int make_conv16to8_table(mpg123_handle *fr)
{
int i;
int mode = fr->af.dec_enc;
/*
* ????: 8.0 is right but on SB cards '2.0' is a better value ???
*/
const double mul = 8.0;
if(!fr->conv16to8_buf){
fr->conv16to8_buf = (unsigned char *) malloc(8192);
if(!fr->conv16to8_buf) {
fr->err = MPG123_ERR_16TO8TABLE;
if(NOQUIET) error("Can't allocate 16 to 8 converter table!");
return -1;
}
fr->conv16to8 = fr->conv16to8_buf + 4096;
}
switch(mode)
{
case MPG123_ENC_ULAW_8:
{
double m=127.0 / log(256.0);
int c1;
for(i=-4096;i<4096;i++)
{
/* dunno whether this is a valid transformation rule ?!?!? */
if(i < 0)
c1 = 127 - (int) (log( 1.0 - 255.0 * (double) i*mul / 32768.0 ) * m);
else
c1 = 255 - (int) (log( 1.0 + 255.0 * (double) i*mul / 32768.0 ) * m);
if(c1 < 0 || c1 > 255)
{
if(NOQUIET) error2("Converror %d %d",i,c1);
return -1;
}
if(c1 == 0)
c1 = 2;
fr->conv16to8[i] = (unsigned char) c1;
}
}
break;
case MPG123_ENC_SIGNED_8:
for(i=-4096;i<4096;i++)
fr->conv16to8[i] = i>>5;
break;
case MPG123_ENC_UNSIGNED_8:
for(i=-4096;i<4096;i++)
fr->conv16to8[i] = (i>>5)+128;
break;
case MPG123_ENC_ALAW_8:
{
/*
Let's believe Wikipedia (http://en.wikipedia.org/wiki/G.711) that this
is the correct table:
s0000000wxyza... n000wxyz [0-31] -> [0-15]
s0000001wxyza... n001wxyz [32-63] -> [16-31]
s000001wxyzab... n010wxyz [64-127] -> [32-47]
s00001wxyzabc... n011wxyz [128-255] -> [48-63]
s0001wxyzabcd... n100wxyz [256-511] -> [64-79]
s001wxyzabcde... n101wxyz [512-1023] -> [80-95]
s01wxyzabcdef... n110wxyz [1024-2047] -> [96-111]
s1wxyzabcdefg... n111wxyz [2048-4095] -> [112-127]
Let's extend to -4096, too.
Also, bytes are xored with 0x55 for transmission.
Since it sounds OK, I assume it is fine;-)
*/
for(i=0; i<64; ++i)
fr->conv16to8[i] = ((unsigned int)i)>>1;
for(i=64; i<128; ++i)
fr->conv16to8[i] = ((((unsigned int)i)>>2) & 0xf) | (2<<4);
for(i=128; i<256; ++i)
fr->conv16to8[i] = ((((unsigned int)i)>>3) & 0xf) | (3<<4);
for(i=256; i<512; ++i)
fr->conv16to8[i] = ((((unsigned int)i)>>4) & 0xf) | (4<<4);
for(i=512; i<1024; ++i)
fr->conv16to8[i] = ((((unsigned int)i)>>5) & 0xf) | (5<<4);
for(i=1024; i<2048; ++i)
fr->conv16to8[i] = ((((unsigned int)i)>>6) & 0xf) | (6<<4);
for(i=2048; i<4096; ++i)
fr->conv16to8[i] = ((((unsigned int)i)>>7) & 0xf) | (7<<4);
for(i=-4095; i<0; ++i)
fr->conv16to8[i] = fr->conv16to8[-i] | 0x80;
fr->conv16to8[-4096] = fr->conv16to8[-4095];
for(i=-4096;i<4096;i++)
{
/* fr->conv16to8[i] = - i>>5; */
/* fprintf(stderr, "table %i %i\n", i<<AUSHIFT, fr->conv16to8[i]); */
fr->conv16to8[i] ^= 0x55;
}
}
break;
default:
fr->err = MPG123_ERR_16TO8TABLE;
if(NOQUIET) error("Unknown 8 bit encoding choice.");
return -1;
break;
}
return 0;
}
#endif