1/* Copyright (C) 2007 Hong Zhiqian */ 2/** 3 @file kiss_fftr_tm.h 4 @author Hong Zhiqian 5 @brief Various compatibility routines for Speex (TriMedia version) 6*/ 7/* 8 Redistribution and use in source and binary forms, with or without 9 modification, are permitted provided that the following conditions 10 are met: 11 12 - Redistributions of source code must retain the above copyright 13 notice, this list of conditions and the following disclaimer. 14 15 - Redistributions in binary form must reproduce the above copyright 16 notice, this list of conditions and the following disclaimer in the 17 documentation and/or other materials provided with the distribution. 18 19 - Neither the name of the Xiph.org Foundation nor the names of its 20 contributors may be used to endorse or promote products derived from 21 this software without specific prior written permission. 22 23 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 26 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR 27 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 28 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 29 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 30 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 31 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 32 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 33 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 34*/ 35#include "_kiss_fft_guts_tm.h" 36 37#ifdef TM_ASM 38 39#include "profile_tm.h" 40 41#ifdef FIXED_POINT 42 43#define TM_NDIV(res,c,frac) \ 44 { register int c1, c0; \ 45 \ 46 c1 = -asri(16,(c)); \ 47 c0 = sex16((c)); \ 48 (res) = pack16lsb(sround(c1 * (32767/(frac))), sround(c0 * (32767/(frac))));\ 49 } 50 51 52#define OVERRIDE_KISS_FFTR 53void kiss_fftr(kiss_fftr_cfg st,const kiss_fft_scalar * restrict timedata, kiss_fft_cpx * restrict freqdata) 54{ 55 register int ncfft, ncfft2, k; 56 register int * restrict tmpbuf; 57 register int * restrict twiddles; 58 59 ncfft = st->substate->nfft; 60 ncfft2 = ncfft >> 1; 61 tmpbuf = (int*)st->tmpbuf; 62 twiddles = (int*)st->super_twiddles; 63 64 TMDEBUG_ALIGNMEM(timedata); 65 TMDEBUG_ALIGNMEM(freqdata); 66 TMDEBUG_ALIGNMEM(tmpbuf); 67 TMDEBUG_ALIGNMEM(twiddles); 68 69 kiss_fft(st->substate , (const kiss_fft_cpx*)timedata, st->tmpbuf); 70 71 { 72 register int tdcr, tdci; 73 tdcr = sround(st->tmpbuf[0].r * (32767/2)); 74 tdci = sround(st->tmpbuf[0].i * (32767/2)); 75 76 freqdata[0].r = tdcr + tdci; 77 freqdata[ncfft].r = tdcr - tdci; 78 freqdata[ncfft].i = freqdata[0].i = 0; 79 } 80 81 for ( k=1 ; k <= ncfft2 ; ++k ) 82 { 83 register int fpk, fpnk, i, tw, f1k, f2k; 84 register int fq1, fq2; 85 86 i = ncfft-k; 87 88 fpk = ld32x(tmpbuf,k); 89 tw = ld32x(twiddles,k); 90 fpnk = ld32x(tmpbuf,i); 91 92 TM_DIV(fpk, fpk, 2); 93 TM_NDIV(fpnk,fpnk,2); 94 95 TM_ADD( f1k, fpk , fpnk ); 96 TM_SUB( f2k, fpk , fpnk ); 97 TM_MUL( tw , f2k, tw ); 98 TM_ADD( fq1, f1k, tw ); 99 TM_SHR( fq1, fq1, 1 ); 100 TM_SUB( fq2, f1k, tw ); 101 TM_NEGMSB( fq2, fq2 ); 102 TM_SHR( fq2, fq2, 1 ); 103 104 105 st32d( k<<2, freqdata, fq1 ); 106 st32d( i<<2, freqdata, fq2 ); 107 } 108} 109 110#define OVERRIDE_KISS_FFTRI 111void kiss_fftri(kiss_fftr_cfg st,const kiss_fft_cpx * restrict freqdata,kiss_fft_scalar * restrict timedata) 112{ 113 register int k, ncfft, ncfft2; 114 register int * restrict tmpbuf; 115 register int * restrict twiddles; 116 117 ncfft = st->substate->nfft; 118 ncfft2 = ncfft >> 1; 119 tmpbuf = (int*)st->tmpbuf; 120 twiddles = (int*)st->super_twiddles; 121 122 TMDEBUG_ALIGNMEM(freqdata); 123 TMDEBUG_ALIGNMEM(timedata); 124 TMDEBUG_ALIGNMEM(tmpbuf); 125 TMDEBUG_ALIGNMEM(twiddles); 126 127 { 128 register int fqr, fqnr; 129 130 fqr = freqdata[0].r; 131 fqnr = freqdata[ncfft].r; 132 133 st->tmpbuf[0].r = fqr + fqnr; 134 st->tmpbuf[0].i = fqr - fqnr; 135 } 136 137 for ( k=1 ; k <= ncfft2 ; ++k ) 138 { 139 register int fk, fnkc, i, tw, fek, fok, tmp; 140 register int tbk, tbn; 141 142 i = ncfft-k; 143 144 fk = ld32x(freqdata,k); 145 tw = ld32x(twiddles,k); 146 fnkc = pack16lsb(-freqdata[i].i, freqdata[i].r); 147 148 TM_ADD (fek, fk, fnkc); 149 TM_SUB (tmp, fk, fnkc); 150 TM_MUL (fok, tmp, tw ); 151 TM_ADD (tbk, fek, fok); 152 TM_SUB (tbn, fek, fok); 153 TM_NEGMSB(tbn, tbn); 154 155 st32d(k<<2, tmpbuf, tbk); 156 st32d(i<<2, tmpbuf, tbn); 157 } 158 kiss_fft (st->substate, st->tmpbuf, (kiss_fft_cpx *) timedata); 159} 160 161#else 162 163#define OVERRIDE_KISS_FFTR 164void kiss_fftr(kiss_fftr_cfg st,const kiss_fft_scalar * restrict timedata,kiss_fft_cpx * restrict freqdata) 165{ 166 register kiss_fft_cpx fpnk, fpk, f1k, f2k, twk; 167 register int k, ncfft; 168 register kiss_fft_cpx * restrict tmpbuf, * restrict tw; 169 register float tdcr, tdci; 170 171 ncfft = st->substate->nfft; 172 tmpbuf= st->tmpbuf; 173 tw = st->super_twiddles; 174 175 kiss_fft( st->substate , (const kiss_fft_cpx*)timedata, tmpbuf ); 176 177 tdcr = tmpbuf[0].r; 178 tdci = tmpbuf[0].i; 179 180 freqdata[0].r = tdcr + tdci; 181 freqdata[ncfft].r = tdcr - tdci; 182 freqdata[ncfft].i = freqdata[0].i = 0; 183 184 for ( k=1;k <= ncfft/2 ; ++k ) 185 { 186 fpk = tmpbuf[k]; 187 fpnk.r = tmpbuf[ncfft-k].r; 188 fpnk.i = -tmpbuf[ncfft-k].i; 189 190 C_ADD( f1k, fpk , fpnk ); 191 C_SUB( f2k, fpk , fpnk ); 192 C_MUL( twk, f2k , tw[k]); 193 194 freqdata[k].r = HALF_OF(f1k.r + twk.r); 195 freqdata[k].i = HALF_OF(f1k.i + twk.i); 196 freqdata[ncfft-k].r = HALF_OF(f1k.r - twk.r); 197 freqdata[ncfft-k].i = HALF_OF(twk.i - f1k.i); 198 } 199} 200 201#define OVERRIDE_KISS_FFTRI 202void kiss_fftri(kiss_fftr_cfg st,const kiss_fft_cpx * restrict freqdata,kiss_fft_scalar * restrict timedata) 203{ 204 register int k, ncfft; 205 register kiss_fft_cpx * restrict tmpbuf, * restrict tw; 206 207 208 ncfft = st->substate->nfft; 209 tmpbuf= st->tmpbuf; 210 tw = st->super_twiddles; 211 212 tmpbuf[0].r = freqdata[0].r + freqdata[ncfft].r; 213 tmpbuf[0].i = freqdata[0].r - freqdata[ncfft].r; 214 215 for (k = 1; k <= ncfft / 2; ++k) 216 { 217 register kiss_fft_cpx fk, fnkc, fek, fok, tmp; 218 fk = freqdata[k]; 219 fnkc.r = freqdata[ncfft - k].r; 220 fnkc.i = -freqdata[ncfft - k].i; 221 222 C_ADD (fek, fk, fnkc); 223 C_SUB (tmp, fk, fnkc); 224 C_MUL (fok,tmp,tw[k]); 225 C_ADD (tmpbuf[k],fek, fok); 226 C_SUB (tmp, fek, fok); 227 tmpbuf[ncfft - k].r = tmp.r; 228 tmpbuf[ncfft - k].i = -tmp.i; 229 } 230 kiss_fft (st->substate, st->tmpbuf, (kiss_fft_cpx *) timedata); 231} 232 233#endif 234#endif 235 236