Chris@0: /* Chris@0: * libmad - MPEG audio decoder library Chris@0: * Copyright (C) 2000-2004 Underbit Technologies, Inc. Chris@0: * Chris@0: * This program is free software; you can redistribute it and/or modify Chris@0: * it under the terms of the GNU General Public License as published by Chris@0: * the Free Software Foundation; either version 2 of the License, or Chris@0: * (at your option) any later version. Chris@0: * Chris@0: * This program is distributed in the hope that it will be useful, Chris@0: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@0: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@0: * GNU General Public License for more details. Chris@0: * Chris@0: * You should have received a copy of the GNU General Public License Chris@0: * along with this program; if not, write to the Free Software Chris@0: * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA Chris@0: * Chris@0: * $Id: synth.c,v 1.25 2004/01/23 09:41:33 rob Exp $ Chris@0: */ Chris@0: Chris@0: # ifdef HAVE_CONFIG_H Chris@0: # include "config.h" Chris@0: # endif Chris@0: Chris@0: # include "global.h" Chris@0: Chris@0: # include "fixed.h" Chris@0: # include "frame.h" Chris@0: # include "synth.h" Chris@0: Chris@0: /* Chris@0: * NAME: synth->init() Chris@0: * DESCRIPTION: initialize synth struct Chris@0: */ Chris@0: void mad_synth_init(struct mad_synth *synth) Chris@0: { Chris@0: mad_synth_mute(synth); Chris@0: Chris@0: synth->phase = 0; Chris@0: Chris@0: synth->pcm.samplerate = 0; Chris@0: synth->pcm.channels = 0; Chris@0: synth->pcm.length = 0; Chris@0: } Chris@0: Chris@0: /* Chris@0: * NAME: synth->mute() Chris@0: * DESCRIPTION: zero all polyphase filterbank values, resetting synthesis Chris@0: */ Chris@0: void mad_synth_mute(struct mad_synth *synth) Chris@0: { Chris@0: unsigned int ch, s, v; Chris@0: Chris@0: for (ch = 0; ch < 2; ++ch) { Chris@0: for (s = 0; s < 16; ++s) { Chris@0: for (v = 0; v < 8; ++v) { Chris@0: synth->filter[ch][0][0][s][v] = synth->filter[ch][0][1][s][v] = Chris@0: synth->filter[ch][1][0][s][v] = synth->filter[ch][1][1][s][v] = 0; Chris@0: } Chris@0: } Chris@0: } Chris@0: } Chris@0: Chris@0: /* Chris@0: * An optional optimization called here the Subband Synthesis Optimization Chris@0: * (SSO) improves the performance of subband synthesis at the expense of Chris@0: * accuracy. Chris@0: * Chris@0: * The idea is to simplify 32x32->64-bit multiplication to 32x32->32 such Chris@0: * that extra scaling and rounding are not necessary. This often allows the Chris@0: * compiler to use faster 32-bit multiply-accumulate instructions instead of Chris@0: * explicit 64-bit multiply, shift, and add instructions. Chris@0: * Chris@0: * SSO works like this: a full 32x32->64-bit multiply of two mad_fixed_t Chris@0: * values requires the result to be right-shifted 28 bits to be properly Chris@0: * scaled to the same fixed-point format. Right shifts can be applied at any Chris@0: * time to either operand or to the result, so the optimization involves Chris@0: * careful placement of these shifts to minimize the loss of accuracy. Chris@0: * Chris@0: * First, a 14-bit shift is applied with rounding at compile-time to the D[] Chris@0: * table of coefficients for the subband synthesis window. This only loses 2 Chris@0: * bits of accuracy because the lower 12 bits are always zero. A second Chris@0: * 12-bit shift occurs after the DCT calculation. This loses 12 bits of Chris@0: * accuracy. Finally, a third 2-bit shift occurs just before the sample is Chris@0: * saved in the PCM buffer. 14 + 12 + 2 == 28 bits. Chris@0: */ Chris@0: Chris@0: /* FPM_DEFAULT without OPT_SSO will actually lose accuracy and performance */ Chris@0: Chris@0: # if defined(FPM_DEFAULT) && !defined(OPT_SSO) Chris@0: # define OPT_SSO Chris@0: # endif Chris@0: Chris@0: /* second SSO shift, with rounding */ Chris@0: Chris@0: # if defined(OPT_SSO) Chris@0: # define SHIFT(x) (((x) + (1L << 11)) >> 12) Chris@0: # else Chris@0: # define SHIFT(x) (x) Chris@0: # endif Chris@0: Chris@0: /* possible DCT speed optimization */ Chris@0: Chris@0: # if defined(OPT_SPEED) && defined(MAD_F_MLX) Chris@0: # define OPT_DCTO Chris@0: # define MUL(x, y) \ Chris@0: ({ mad_fixed64hi_t hi; \ Chris@0: mad_fixed64lo_t lo; \ Chris@0: MAD_F_MLX(hi, lo, (x), (y)); \ Chris@0: hi << (32 - MAD_F_SCALEBITS - 3); \ Chris@0: }) Chris@0: # else Chris@0: # undef OPT_DCTO Chris@0: # define MUL(x, y) mad_f_mul((x), (y)) Chris@0: # endif Chris@0: Chris@0: /* Chris@0: * NAME: dct32() Chris@0: * DESCRIPTION: perform fast in[32]->out[32] DCT Chris@0: */ Chris@0: static Chris@0: void dct32(mad_fixed_t const in[32], unsigned int slot, Chris@0: mad_fixed_t lo[16][8], mad_fixed_t hi[16][8]) Chris@0: { Chris@0: mad_fixed_t t0, t1, t2, t3, t4, t5, t6, t7; Chris@0: mad_fixed_t t8, t9, t10, t11, t12, t13, t14, t15; Chris@0: mad_fixed_t t16, t17, t18, t19, t20, t21, t22, t23; Chris@0: mad_fixed_t t24, t25, t26, t27, t28, t29, t30, t31; Chris@0: mad_fixed_t t32, t33, t34, t35, t36, t37, t38, t39; Chris@0: mad_fixed_t t40, t41, t42, t43, t44, t45, t46, t47; Chris@0: mad_fixed_t t48, t49, t50, t51, t52, t53, t54, t55; Chris@0: mad_fixed_t t56, t57, t58, t59, t60, t61, t62, t63; Chris@0: mad_fixed_t t64, t65, t66, t67, t68, t69, t70, t71; Chris@0: mad_fixed_t t72, t73, t74, t75, t76, t77, t78, t79; Chris@0: mad_fixed_t t80, t81, t82, t83, t84, t85, t86, t87; Chris@0: mad_fixed_t t88, t89, t90, t91, t92, t93, t94, t95; Chris@0: mad_fixed_t t96, t97, t98, t99, t100, t101, t102, t103; Chris@0: mad_fixed_t t104, t105, t106, t107, t108, t109, t110, t111; Chris@0: mad_fixed_t t112, t113, t114, t115, t116, t117, t118, t119; Chris@0: mad_fixed_t t120, t121, t122, t123, t124, t125, t126, t127; Chris@0: mad_fixed_t t128, t129, t130, t131, t132, t133, t134, t135; Chris@0: mad_fixed_t t136, t137, t138, t139, t140, t141, t142, t143; Chris@0: mad_fixed_t t144, t145, t146, t147, t148, t149, t150, t151; Chris@0: mad_fixed_t t152, t153, t154, t155, t156, t157, t158, t159; Chris@0: mad_fixed_t t160, t161, t162, t163, t164, t165, t166, t167; Chris@0: mad_fixed_t t168, t169, t170, t171, t172, t173, t174, t175; Chris@0: mad_fixed_t t176; Chris@0: Chris@0: /* costab[i] = cos(PI / (2 * 32) * i) */ Chris@0: Chris@0: # if defined(OPT_DCTO) Chris@0: # define costab1 MAD_F(0x7fd8878e) Chris@0: # define costab2 MAD_F(0x7f62368f) Chris@0: # define costab3 MAD_F(0x7e9d55fc) Chris@0: # define costab4 MAD_F(0x7d8a5f40) Chris@0: # define costab5 MAD_F(0x7c29fbee) Chris@0: # define costab6 MAD_F(0x7a7d055b) Chris@0: # define costab7 MAD_F(0x78848414) Chris@0: # define costab8 MAD_F(0x7641af3d) Chris@0: # define costab9 MAD_F(0x73b5ebd1) Chris@0: # define costab10 MAD_F(0x70e2cbc6) Chris@0: # define costab11 MAD_F(0x6dca0d14) Chris@0: # define costab12 MAD_F(0x6a6d98a4) Chris@0: # define costab13 MAD_F(0x66cf8120) Chris@0: # define costab14 MAD_F(0x62f201ac) Chris@0: # define costab15 MAD_F(0x5ed77c8a) Chris@0: # define costab16 MAD_F(0x5a82799a) Chris@0: # define costab17 MAD_F(0x55f5a4d2) Chris@0: # define costab18 MAD_F(0x5133cc94) Chris@0: # define costab19 MAD_F(0x4c3fdff4) Chris@0: # define costab20 MAD_F(0x471cece7) Chris@0: # define costab21 MAD_F(0x41ce1e65) Chris@0: # define costab22 MAD_F(0x3c56ba70) Chris@0: # define costab23 MAD_F(0x36ba2014) Chris@0: # define costab24 MAD_F(0x30fbc54d) Chris@0: # define costab25 MAD_F(0x2b1f34eb) Chris@0: # define costab26 MAD_F(0x25280c5e) Chris@0: # define costab27 MAD_F(0x1f19f97b) Chris@0: # define costab28 MAD_F(0x18f8b83c) Chris@0: # define costab29 MAD_F(0x12c8106f) Chris@0: # define costab30 MAD_F(0x0c8bd35e) Chris@0: # define costab31 MAD_F(0x0647d97c) Chris@0: # else Chris@0: # define costab1 MAD_F(0x0ffb10f2) /* 0.998795456 */ Chris@0: # define costab2 MAD_F(0x0fec46d2) /* 0.995184727 */ Chris@0: # define costab3 MAD_F(0x0fd3aac0) /* 0.989176510 */ Chris@0: # define costab4 MAD_F(0x0fb14be8) /* 0.980785280 */ Chris@0: # define costab5 MAD_F(0x0f853f7e) /* 0.970031253 */ Chris@0: # define costab6 MAD_F(0x0f4fa0ab) /* 0.956940336 */ Chris@0: # define costab7 MAD_F(0x0f109082) /* 0.941544065 */ Chris@0: # define costab8 MAD_F(0x0ec835e8) /* 0.923879533 */ Chris@0: # define costab9 MAD_F(0x0e76bd7a) /* 0.903989293 */ Chris@0: # define costab10 MAD_F(0x0e1c5979) /* 0.881921264 */ Chris@0: # define costab11 MAD_F(0x0db941a3) /* 0.857728610 */ Chris@0: # define costab12 MAD_F(0x0d4db315) /* 0.831469612 */ Chris@0: # define costab13 MAD_F(0x0cd9f024) /* 0.803207531 */ Chris@0: # define costab14 MAD_F(0x0c5e4036) /* 0.773010453 */ Chris@0: # define costab15 MAD_F(0x0bdaef91) /* 0.740951125 */ Chris@0: # define costab16 MAD_F(0x0b504f33) /* 0.707106781 */ Chris@0: # define costab17 MAD_F(0x0abeb49a) /* 0.671558955 */ Chris@0: # define costab18 MAD_F(0x0a267993) /* 0.634393284 */ Chris@0: # define costab19 MAD_F(0x0987fbfe) /* 0.595699304 */ Chris@0: # define costab20 MAD_F(0x08e39d9d) /* 0.555570233 */ Chris@0: # define costab21 MAD_F(0x0839c3cd) /* 0.514102744 */ Chris@0: # define costab22 MAD_F(0x078ad74e) /* 0.471396737 */ Chris@0: # define costab23 MAD_F(0x06d74402) /* 0.427555093 */ Chris@0: # define costab24 MAD_F(0x061f78aa) /* 0.382683432 */ Chris@0: # define costab25 MAD_F(0x0563e69d) /* 0.336889853 */ Chris@0: # define costab26 MAD_F(0x04a5018c) /* 0.290284677 */ Chris@0: # define costab27 MAD_F(0x03e33f2f) /* 0.242980180 */ Chris@0: # define costab28 MAD_F(0x031f1708) /* 0.195090322 */ Chris@0: # define costab29 MAD_F(0x0259020e) /* 0.146730474 */ Chris@0: # define costab30 MAD_F(0x01917a6c) /* 0.098017140 */ Chris@0: # define costab31 MAD_F(0x00c8fb30) /* 0.049067674 */ Chris@0: # endif Chris@0: Chris@0: t0 = in[0] + in[31]; t16 = MUL(in[0] - in[31], costab1); Chris@0: t1 = in[15] + in[16]; t17 = MUL(in[15] - in[16], costab31); Chris@0: Chris@0: t41 = t16 + t17; Chris@0: t59 = MUL(t16 - t17, costab2); Chris@0: t33 = t0 + t1; Chris@0: t50 = MUL(t0 - t1, costab2); Chris@0: Chris@0: t2 = in[7] + in[24]; t18 = MUL(in[7] - in[24], costab15); Chris@0: t3 = in[8] + in[23]; t19 = MUL(in[8] - in[23], costab17); Chris@0: Chris@0: t42 = t18 + t19; Chris@0: t60 = MUL(t18 - t19, costab30); Chris@0: t34 = t2 + t3; Chris@0: t51 = MUL(t2 - t3, costab30); Chris@0: Chris@0: t4 = in[3] + in[28]; t20 = MUL(in[3] - in[28], costab7); Chris@0: t5 = in[12] + in[19]; t21 = MUL(in[12] - in[19], costab25); Chris@0: Chris@0: t43 = t20 + t21; Chris@0: t61 = MUL(t20 - t21, costab14); Chris@0: t35 = t4 + t5; Chris@0: t52 = MUL(t4 - t5, costab14); Chris@0: Chris@0: t6 = in[4] + in[27]; t22 = MUL(in[4] - in[27], costab9); Chris@0: t7 = in[11] + in[20]; t23 = MUL(in[11] - in[20], costab23); Chris@0: Chris@0: t44 = t22 + t23; Chris@0: t62 = MUL(t22 - t23, costab18); Chris@0: t36 = t6 + t7; Chris@0: t53 = MUL(t6 - t7, costab18); Chris@0: Chris@0: t8 = in[1] + in[30]; t24 = MUL(in[1] - in[30], costab3); Chris@0: t9 = in[14] + in[17]; t25 = MUL(in[14] - in[17], costab29); Chris@0: Chris@0: t45 = t24 + t25; Chris@0: t63 = MUL(t24 - t25, costab6); Chris@0: t37 = t8 + t9; Chris@0: t54 = MUL(t8 - t9, costab6); Chris@0: Chris@0: t10 = in[6] + in[25]; t26 = MUL(in[6] - in[25], costab13); Chris@0: t11 = in[9] + in[22]; t27 = MUL(in[9] - in[22], costab19); Chris@0: Chris@0: t46 = t26 + t27; Chris@0: t64 = MUL(t26 - t27, costab26); Chris@0: t38 = t10 + t11; Chris@0: t55 = MUL(t10 - t11, costab26); Chris@0: Chris@0: t12 = in[2] + in[29]; t28 = MUL(in[2] - in[29], costab5); Chris@0: t13 = in[13] + in[18]; t29 = MUL(in[13] - in[18], costab27); Chris@0: Chris@0: t47 = t28 + t29; Chris@0: t65 = MUL(t28 - t29, costab10); Chris@0: t39 = t12 + t13; Chris@0: t56 = MUL(t12 - t13, costab10); Chris@0: Chris@0: t14 = in[5] + in[26]; t30 = MUL(in[5] - in[26], costab11); Chris@0: t15 = in[10] + in[21]; t31 = MUL(in[10] - in[21], costab21); Chris@0: Chris@0: t48 = t30 + t31; Chris@0: t66 = MUL(t30 - t31, costab22); Chris@0: t40 = t14 + t15; Chris@0: t57 = MUL(t14 - t15, costab22); Chris@0: Chris@0: t69 = t33 + t34; t89 = MUL(t33 - t34, costab4); Chris@0: t70 = t35 + t36; t90 = MUL(t35 - t36, costab28); Chris@0: t71 = t37 + t38; t91 = MUL(t37 - t38, costab12); Chris@0: t72 = t39 + t40; t92 = MUL(t39 - t40, costab20); Chris@0: t73 = t41 + t42; t94 = MUL(t41 - t42, costab4); Chris@0: t74 = t43 + t44; t95 = MUL(t43 - t44, costab28); Chris@0: t75 = t45 + t46; t96 = MUL(t45 - t46, costab12); Chris@0: t76 = t47 + t48; t97 = MUL(t47 - t48, costab20); Chris@0: Chris@0: t78 = t50 + t51; t100 = MUL(t50 - t51, costab4); Chris@0: t79 = t52 + t53; t101 = MUL(t52 - t53, costab28); Chris@0: t80 = t54 + t55; t102 = MUL(t54 - t55, costab12); Chris@0: t81 = t56 + t57; t103 = MUL(t56 - t57, costab20); Chris@0: Chris@0: t83 = t59 + t60; t106 = MUL(t59 - t60, costab4); Chris@0: t84 = t61 + t62; t107 = MUL(t61 - t62, costab28); Chris@0: t85 = t63 + t64; t108 = MUL(t63 - t64, costab12); Chris@0: t86 = t65 + t66; t109 = MUL(t65 - t66, costab20); Chris@0: Chris@0: t113 = t69 + t70; Chris@0: t114 = t71 + t72; Chris@0: Chris@0: /* 0 */ hi[15][slot] = SHIFT(t113 + t114); Chris@0: /* 16 */ lo[ 0][slot] = SHIFT(MUL(t113 - t114, costab16)); Chris@0: Chris@0: t115 = t73 + t74; Chris@0: t116 = t75 + t76; Chris@0: Chris@0: t32 = t115 + t116; Chris@0: Chris@0: /* 1 */ hi[14][slot] = SHIFT(t32); Chris@0: Chris@0: t118 = t78 + t79; Chris@0: t119 = t80 + t81; Chris@0: Chris@0: t58 = t118 + t119; Chris@0: Chris@0: /* 2 */ hi[13][slot] = SHIFT(t58); Chris@0: Chris@0: t121 = t83 + t84; Chris@0: t122 = t85 + t86; Chris@0: Chris@0: t67 = t121 + t122; Chris@0: Chris@0: t49 = (t67 * 2) - t32; Chris@0: Chris@0: /* 3 */ hi[12][slot] = SHIFT(t49); Chris@0: Chris@0: t125 = t89 + t90; Chris@0: t126 = t91 + t92; Chris@0: Chris@0: t93 = t125 + t126; Chris@0: Chris@0: /* 4 */ hi[11][slot] = SHIFT(t93); Chris@0: Chris@0: t128 = t94 + t95; Chris@0: t129 = t96 + t97; Chris@0: Chris@0: t98 = t128 + t129; Chris@0: Chris@0: t68 = (t98 * 2) - t49; Chris@0: Chris@0: /* 5 */ hi[10][slot] = SHIFT(t68); Chris@0: Chris@0: t132 = t100 + t101; Chris@0: t133 = t102 + t103; Chris@0: Chris@0: t104 = t132 + t133; Chris@0: Chris@0: t82 = (t104 * 2) - t58; Chris@0: Chris@0: /* 6 */ hi[ 9][slot] = SHIFT(t82); Chris@0: Chris@0: t136 = t106 + t107; Chris@0: t137 = t108 + t109; Chris@0: Chris@0: t110 = t136 + t137; Chris@0: Chris@0: t87 = (t110 * 2) - t67; Chris@0: Chris@0: t77 = (t87 * 2) - t68; Chris@0: Chris@0: /* 7 */ hi[ 8][slot] = SHIFT(t77); Chris@0: Chris@0: t141 = MUL(t69 - t70, costab8); Chris@0: t142 = MUL(t71 - t72, costab24); Chris@0: t143 = t141 + t142; Chris@0: Chris@0: /* 8 */ hi[ 7][slot] = SHIFT(t143); Chris@0: /* 24 */ lo[ 8][slot] = Chris@0: SHIFT((MUL(t141 - t142, costab16) * 2) - t143); Chris@0: Chris@0: t144 = MUL(t73 - t74, costab8); Chris@0: t145 = MUL(t75 - t76, costab24); Chris@0: t146 = t144 + t145; Chris@0: Chris@0: t88 = (t146 * 2) - t77; Chris@0: Chris@0: /* 9 */ hi[ 6][slot] = SHIFT(t88); Chris@0: Chris@0: t148 = MUL(t78 - t79, costab8); Chris@0: t149 = MUL(t80 - t81, costab24); Chris@0: t150 = t148 + t149; Chris@0: Chris@0: t105 = (t150 * 2) - t82; Chris@0: Chris@0: /* 10 */ hi[ 5][slot] = SHIFT(t105); Chris@0: Chris@0: t152 = MUL(t83 - t84, costab8); Chris@0: t153 = MUL(t85 - t86, costab24); Chris@0: t154 = t152 + t153; Chris@0: Chris@0: t111 = (t154 * 2) - t87; Chris@0: Chris@0: t99 = (t111 * 2) - t88; Chris@0: Chris@0: /* 11 */ hi[ 4][slot] = SHIFT(t99); Chris@0: Chris@0: t157 = MUL(t89 - t90, costab8); Chris@0: t158 = MUL(t91 - t92, costab24); Chris@0: t159 = t157 + t158; Chris@0: Chris@0: t127 = (t159 * 2) - t93; Chris@0: Chris@0: /* 12 */ hi[ 3][slot] = SHIFT(t127); Chris@0: Chris@0: t160 = (MUL(t125 - t126, costab16) * 2) - t127; Chris@0: Chris@0: /* 20 */ lo[ 4][slot] = SHIFT(t160); Chris@0: /* 28 */ lo[12][slot] = Chris@0: SHIFT((((MUL(t157 - t158, costab16) * 2) - t159) * 2) - t160); Chris@0: Chris@0: t161 = MUL(t94 - t95, costab8); Chris@0: t162 = MUL(t96 - t97, costab24); Chris@0: t163 = t161 + t162; Chris@0: Chris@0: t130 = (t163 * 2) - t98; Chris@0: Chris@0: t112 = (t130 * 2) - t99; Chris@0: Chris@0: /* 13 */ hi[ 2][slot] = SHIFT(t112); Chris@0: Chris@0: t164 = (MUL(t128 - t129, costab16) * 2) - t130; Chris@0: Chris@0: t166 = MUL(t100 - t101, costab8); Chris@0: t167 = MUL(t102 - t103, costab24); Chris@0: t168 = t166 + t167; Chris@0: Chris@0: t134 = (t168 * 2) - t104; Chris@0: Chris@0: t120 = (t134 * 2) - t105; Chris@0: Chris@0: /* 14 */ hi[ 1][slot] = SHIFT(t120); Chris@0: Chris@0: t135 = (MUL(t118 - t119, costab16) * 2) - t120; Chris@0: Chris@0: /* 18 */ lo[ 2][slot] = SHIFT(t135); Chris@0: Chris@0: t169 = (MUL(t132 - t133, costab16) * 2) - t134; Chris@0: Chris@0: t151 = (t169 * 2) - t135; Chris@0: Chris@0: /* 22 */ lo[ 6][slot] = SHIFT(t151); Chris@0: Chris@0: t170 = (((MUL(t148 - t149, costab16) * 2) - t150) * 2) - t151; Chris@0: Chris@0: /* 26 */ lo[10][slot] = SHIFT(t170); Chris@0: /* 30 */ lo[14][slot] = Chris@0: SHIFT((((((MUL(t166 - t167, costab16) * 2) - Chris@0: t168) * 2) - t169) * 2) - t170); Chris@0: Chris@0: t171 = MUL(t106 - t107, costab8); Chris@0: t172 = MUL(t108 - t109, costab24); Chris@0: t173 = t171 + t172; Chris@0: Chris@0: t138 = (t173 * 2) - t110; Chris@0: Chris@0: t123 = (t138 * 2) - t111; Chris@0: Chris@0: t139 = (MUL(t121 - t122, costab16) * 2) - t123; Chris@0: Chris@0: t117 = (t123 * 2) - t112; Chris@0: Chris@0: /* 15 */ hi[ 0][slot] = SHIFT(t117); Chris@0: Chris@0: t124 = (MUL(t115 - t116, costab16) * 2) - t117; Chris@0: Chris@0: /* 17 */ lo[ 1][slot] = SHIFT(t124); Chris@0: Chris@0: t131 = (t139 * 2) - t124; Chris@0: Chris@0: /* 19 */ lo[ 3][slot] = SHIFT(t131); Chris@0: Chris@0: t140 = (t164 * 2) - t131; Chris@0: Chris@0: /* 21 */ lo[ 5][slot] = SHIFT(t140); Chris@0: Chris@0: t174 = (MUL(t136 - t137, costab16) * 2) - t138; Chris@0: Chris@0: t155 = (t174 * 2) - t139; Chris@0: Chris@0: t147 = (t155 * 2) - t140; Chris@0: Chris@0: /* 23 */ lo[ 7][slot] = SHIFT(t147); Chris@0: Chris@0: t156 = (((MUL(t144 - t145, costab16) * 2) - t146) * 2) - t147; Chris@0: Chris@0: /* 25 */ lo[ 9][slot] = SHIFT(t156); Chris@0: Chris@0: t175 = (((MUL(t152 - t153, costab16) * 2) - t154) * 2) - t155; Chris@0: Chris@0: t165 = (t175 * 2) - t156; Chris@0: Chris@0: /* 27 */ lo[11][slot] = SHIFT(t165); Chris@0: Chris@0: t176 = (((((MUL(t161 - t162, costab16) * 2) - Chris@0: t163) * 2) - t164) * 2) - t165; Chris@0: Chris@0: /* 29 */ lo[13][slot] = SHIFT(t176); Chris@0: /* 31 */ lo[15][slot] = Chris@0: SHIFT((((((((MUL(t171 - t172, costab16) * 2) - Chris@0: t173) * 2) - t174) * 2) - t175) * 2) - t176); Chris@0: Chris@0: /* Chris@0: * Totals: Chris@0: * 80 multiplies Chris@0: * 80 additions Chris@0: * 119 subtractions Chris@0: * 49 shifts (not counting SSO) Chris@0: */ Chris@0: } Chris@0: Chris@0: # undef MUL Chris@0: # undef SHIFT Chris@0: Chris@0: /* third SSO shift and/or D[] optimization preshift */ Chris@0: Chris@0: # if defined(OPT_SSO) Chris@0: # if MAD_F_FRACBITS != 28 Chris@0: # error "MAD_F_FRACBITS must be 28 to use OPT_SSO" Chris@0: # endif Chris@0: # define ML0(hi, lo, x, y) ((lo) = (x) * (y)) Chris@0: # define MLA(hi, lo, x, y) ((lo) += (x) * (y)) Chris@0: # define MLN(hi, lo) ((lo) = -(lo)) Chris@0: # define MLZ(hi, lo) ((void) (hi), (mad_fixed_t) (lo)) Chris@0: # define SHIFT(x) ((x) >> 2) Chris@0: # define PRESHIFT(x) ((MAD_F(x) + (1L << 13)) >> 14) Chris@0: # else Chris@0: # define ML0(hi, lo, x, y) MAD_F_ML0((hi), (lo), (x), (y)) Chris@0: # define MLA(hi, lo, x, y) MAD_F_MLA((hi), (lo), (x), (y)) Chris@0: # define MLN(hi, lo) MAD_F_MLN((hi), (lo)) Chris@0: # define MLZ(hi, lo) MAD_F_MLZ((hi), (lo)) Chris@0: # define SHIFT(x) (x) Chris@0: # if defined(MAD_F_SCALEBITS) Chris@0: # undef MAD_F_SCALEBITS Chris@0: # define MAD_F_SCALEBITS (MAD_F_FRACBITS - 12) Chris@0: # define PRESHIFT(x) (MAD_F(x) >> 12) Chris@0: # else Chris@0: # define PRESHIFT(x) MAD_F(x) Chris@0: # endif Chris@0: # endif Chris@0: Chris@0: static Chris@0: mad_fixed_t const D[17][32] = { Chris@0: # include "D.dat" Chris@0: }; Chris@0: Chris@0: # if defined(ASO_SYNTH) Chris@0: void synth_full(struct mad_synth *, struct mad_frame const *, Chris@0: unsigned int, unsigned int); Chris@0: # else Chris@0: /* Chris@0: * NAME: synth->full() Chris@0: * DESCRIPTION: perform full frequency PCM synthesis Chris@0: */ Chris@0: static Chris@0: void synth_full(struct mad_synth *synth, struct mad_frame const *frame, Chris@0: unsigned int nch, unsigned int ns) Chris@0: { Chris@0: unsigned int phase, ch, s, sb, pe, po; Chris@0: mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8]; Chris@0: mad_fixed_t const (*sbsample)[36][32]; Chris@0: register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8]; Chris@0: register mad_fixed_t const (*Dptr)[32], *ptr; Chris@0: register mad_fixed64hi_t hi; Chris@0: register mad_fixed64lo_t lo; Chris@0: Chris@0: for (ch = 0; ch < nch; ++ch) { Chris@0: sbsample = &frame->sbsample[ch]; Chris@0: filter = &synth->filter[ch]; Chris@0: phase = synth->phase; Chris@0: pcm1 = synth->pcm.samples[ch]; Chris@0: Chris@0: for (s = 0; s < ns; ++s) { Chris@0: dct32((*sbsample)[s], phase >> 1, Chris@0: (*filter)[0][phase & 1], (*filter)[1][phase & 1]); Chris@0: Chris@0: pe = phase & ~1; Chris@0: po = ((phase - 1) & 0xf) | 1; Chris@0: Chris@0: /* calculate 32 samples */ Chris@0: Chris@0: fe = &(*filter)[0][ phase & 1][0]; Chris@0: fx = &(*filter)[0][~phase & 1][0]; Chris@0: fo = &(*filter)[1][~phase & 1][0]; Chris@0: Chris@0: Dptr = &D[0]; Chris@0: Chris@0: ptr = *Dptr + po; Chris@0: ML0(hi, lo, (*fx)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fx)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fx)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fx)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fx)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fx)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fx)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fx)[7], ptr[ 2]); Chris@0: MLN(hi, lo); Chris@0: Chris@0: ptr = *Dptr + pe; Chris@0: MLA(hi, lo, (*fe)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fe)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fe)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fe)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fe)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fe)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fe)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fe)[7], ptr[ 2]); Chris@0: Chris@0: *pcm1++ = SHIFT(MLZ(hi, lo)); Chris@0: Chris@0: pcm2 = pcm1 + 30; Chris@0: Chris@0: for (sb = 1; sb < 16; ++sb) { Chris@0: ++fe; Chris@0: ++Dptr; Chris@0: Chris@0: /* D[32 - sb][i] == -D[sb][31 - i] */ Chris@0: Chris@0: ptr = *Dptr + po; Chris@0: ML0(hi, lo, (*fo)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fo)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fo)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fo)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fo)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fo)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fo)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fo)[7], ptr[ 2]); Chris@0: MLN(hi, lo); Chris@0: Chris@0: ptr = *Dptr + pe; Chris@0: MLA(hi, lo, (*fe)[7], ptr[ 2]); Chris@0: MLA(hi, lo, (*fe)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fe)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fe)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fe)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fe)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fe)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fe)[0], ptr[ 0]); Chris@0: Chris@0: *pcm1++ = SHIFT(MLZ(hi, lo)); Chris@0: Chris@0: ptr = *Dptr - pe; Chris@0: ML0(hi, lo, (*fe)[0], ptr[31 - 16]); Chris@0: MLA(hi, lo, (*fe)[1], ptr[31 - 14]); Chris@0: MLA(hi, lo, (*fe)[2], ptr[31 - 12]); Chris@0: MLA(hi, lo, (*fe)[3], ptr[31 - 10]); Chris@0: MLA(hi, lo, (*fe)[4], ptr[31 - 8]); Chris@0: MLA(hi, lo, (*fe)[5], ptr[31 - 6]); Chris@0: MLA(hi, lo, (*fe)[6], ptr[31 - 4]); Chris@0: MLA(hi, lo, (*fe)[7], ptr[31 - 2]); Chris@0: Chris@0: ptr = *Dptr - po; Chris@0: MLA(hi, lo, (*fo)[7], ptr[31 - 2]); Chris@0: MLA(hi, lo, (*fo)[6], ptr[31 - 4]); Chris@0: MLA(hi, lo, (*fo)[5], ptr[31 - 6]); Chris@0: MLA(hi, lo, (*fo)[4], ptr[31 - 8]); Chris@0: MLA(hi, lo, (*fo)[3], ptr[31 - 10]); Chris@0: MLA(hi, lo, (*fo)[2], ptr[31 - 12]); Chris@0: MLA(hi, lo, (*fo)[1], ptr[31 - 14]); Chris@0: MLA(hi, lo, (*fo)[0], ptr[31 - 16]); Chris@0: Chris@0: *pcm2-- = SHIFT(MLZ(hi, lo)); Chris@0: Chris@0: ++fo; Chris@0: } Chris@0: Chris@0: ++Dptr; Chris@0: Chris@0: ptr = *Dptr + po; Chris@0: ML0(hi, lo, (*fo)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fo)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fo)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fo)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fo)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fo)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fo)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fo)[7], ptr[ 2]); Chris@0: Chris@0: *pcm1 = SHIFT(-MLZ(hi, lo)); Chris@0: pcm1 += 16; Chris@0: Chris@0: phase = (phase + 1) % 16; Chris@0: } Chris@0: } Chris@0: } Chris@0: # endif Chris@0: Chris@0: /* Chris@0: * NAME: synth->half() Chris@0: * DESCRIPTION: perform half frequency PCM synthesis Chris@0: */ Chris@0: static Chris@0: void synth_half(struct mad_synth *synth, struct mad_frame const *frame, Chris@0: unsigned int nch, unsigned int ns) Chris@0: { Chris@0: unsigned int phase, ch, s, sb, pe, po; Chris@0: mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8]; Chris@0: mad_fixed_t const (*sbsample)[36][32]; Chris@0: register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8]; Chris@0: register mad_fixed_t const (*Dptr)[32], *ptr; Chris@0: register mad_fixed64hi_t hi; Chris@0: register mad_fixed64lo_t lo; Chris@0: Chris@0: for (ch = 0; ch < nch; ++ch) { Chris@0: sbsample = &frame->sbsample[ch]; Chris@0: filter = &synth->filter[ch]; Chris@0: phase = synth->phase; Chris@0: pcm1 = synth->pcm.samples[ch]; Chris@0: Chris@0: for (s = 0; s < ns; ++s) { Chris@0: dct32((*sbsample)[s], phase >> 1, Chris@0: (*filter)[0][phase & 1], (*filter)[1][phase & 1]); Chris@0: Chris@0: pe = phase & ~1; Chris@0: po = ((phase - 1) & 0xf) | 1; Chris@0: Chris@0: /* calculate 16 samples */ Chris@0: Chris@0: fe = &(*filter)[0][ phase & 1][0]; Chris@0: fx = &(*filter)[0][~phase & 1][0]; Chris@0: fo = &(*filter)[1][~phase & 1][0]; Chris@0: Chris@0: Dptr = &D[0]; Chris@0: Chris@0: ptr = *Dptr + po; Chris@0: ML0(hi, lo, (*fx)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fx)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fx)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fx)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fx)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fx)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fx)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fx)[7], ptr[ 2]); Chris@0: MLN(hi, lo); Chris@0: Chris@0: ptr = *Dptr + pe; Chris@0: MLA(hi, lo, (*fe)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fe)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fe)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fe)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fe)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fe)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fe)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fe)[7], ptr[ 2]); Chris@0: Chris@0: *pcm1++ = SHIFT(MLZ(hi, lo)); Chris@0: Chris@0: pcm2 = pcm1 + 14; Chris@0: Chris@0: for (sb = 1; sb < 16; ++sb) { Chris@0: ++fe; Chris@0: ++Dptr; Chris@0: Chris@0: /* D[32 - sb][i] == -D[sb][31 - i] */ Chris@0: Chris@0: if (!(sb & 1)) { Chris@0: ptr = *Dptr + po; Chris@0: ML0(hi, lo, (*fo)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fo)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fo)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fo)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fo)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fo)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fo)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fo)[7], ptr[ 2]); Chris@0: MLN(hi, lo); Chris@0: Chris@0: ptr = *Dptr + pe; Chris@0: MLA(hi, lo, (*fe)[7], ptr[ 2]); Chris@0: MLA(hi, lo, (*fe)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fe)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fe)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fe)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fe)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fe)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fe)[0], ptr[ 0]); Chris@0: Chris@0: *pcm1++ = SHIFT(MLZ(hi, lo)); Chris@0: Chris@0: ptr = *Dptr - po; Chris@0: ML0(hi, lo, (*fo)[7], ptr[31 - 2]); Chris@0: MLA(hi, lo, (*fo)[6], ptr[31 - 4]); Chris@0: MLA(hi, lo, (*fo)[5], ptr[31 - 6]); Chris@0: MLA(hi, lo, (*fo)[4], ptr[31 - 8]); Chris@0: MLA(hi, lo, (*fo)[3], ptr[31 - 10]); Chris@0: MLA(hi, lo, (*fo)[2], ptr[31 - 12]); Chris@0: MLA(hi, lo, (*fo)[1], ptr[31 - 14]); Chris@0: MLA(hi, lo, (*fo)[0], ptr[31 - 16]); Chris@0: Chris@0: ptr = *Dptr - pe; Chris@0: MLA(hi, lo, (*fe)[0], ptr[31 - 16]); Chris@0: MLA(hi, lo, (*fe)[1], ptr[31 - 14]); Chris@0: MLA(hi, lo, (*fe)[2], ptr[31 - 12]); Chris@0: MLA(hi, lo, (*fe)[3], ptr[31 - 10]); Chris@0: MLA(hi, lo, (*fe)[4], ptr[31 - 8]); Chris@0: MLA(hi, lo, (*fe)[5], ptr[31 - 6]); Chris@0: MLA(hi, lo, (*fe)[6], ptr[31 - 4]); Chris@0: MLA(hi, lo, (*fe)[7], ptr[31 - 2]); Chris@0: Chris@0: *pcm2-- = SHIFT(MLZ(hi, lo)); Chris@0: } Chris@0: Chris@0: ++fo; Chris@0: } Chris@0: Chris@0: ++Dptr; Chris@0: Chris@0: ptr = *Dptr + po; Chris@0: ML0(hi, lo, (*fo)[0], ptr[ 0]); Chris@0: MLA(hi, lo, (*fo)[1], ptr[14]); Chris@0: MLA(hi, lo, (*fo)[2], ptr[12]); Chris@0: MLA(hi, lo, (*fo)[3], ptr[10]); Chris@0: MLA(hi, lo, (*fo)[4], ptr[ 8]); Chris@0: MLA(hi, lo, (*fo)[5], ptr[ 6]); Chris@0: MLA(hi, lo, (*fo)[6], ptr[ 4]); Chris@0: MLA(hi, lo, (*fo)[7], ptr[ 2]); Chris@0: Chris@0: *pcm1 = SHIFT(-MLZ(hi, lo)); Chris@0: pcm1 += 8; Chris@0: Chris@0: phase = (phase + 1) % 16; Chris@0: } Chris@0: } Chris@0: } Chris@0: Chris@0: /* Chris@0: * NAME: synth->frame() Chris@0: * DESCRIPTION: perform PCM synthesis of frame subband samples Chris@0: */ Chris@0: void mad_synth_frame(struct mad_synth *synth, struct mad_frame const *frame) Chris@0: { Chris@0: unsigned int nch, ns; Chris@0: void (*synth_frame)(struct mad_synth *, struct mad_frame const *, Chris@0: unsigned int, unsigned int); Chris@0: Chris@0: nch = MAD_NCHANNELS(&frame->header); Chris@0: ns = MAD_NSBSAMPLES(&frame->header); Chris@0: Chris@0: synth->pcm.samplerate = frame->header.samplerate; Chris@0: synth->pcm.channels = nch; Chris@0: synth->pcm.length = 32 * ns; Chris@0: Chris@0: synth_frame = synth_full; Chris@0: Chris@0: if (frame->options & MAD_OPTION_HALFSAMPLERATE) { Chris@0: synth->pcm.samplerate /= 2; Chris@0: synth->pcm.length /= 2; Chris@0: Chris@0: synth_frame = synth_half; Chris@0: } Chris@0: Chris@0: synth_frame(synth, frame, nch, ns); Chris@0: Chris@0: synth->phase = (synth->phase + ns) % 16; Chris@0: }