| 1 | /* |
| 2 | * Copyright (c) 2012 |
| 3 | * MIPS Technologies, Inc., California. |
| 4 | * |
| 5 | * Redistribution and use in source and binary forms, with or without |
| 6 | * modification, are permitted provided that the following conditions |
| 7 | * are met: |
| 8 | * 1. Redistributions of source code must retain the above copyright |
| 9 | * notice, this list of conditions and the following disclaimer. |
| 10 | * 2. Redistributions in binary form must reproduce the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer in the |
| 12 | * documentation and/or other materials provided with the distribution. |
| 13 | * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its |
| 14 | * contributors may be used to endorse or promote products derived from |
| 15 | * this software without specific prior written permission. |
| 16 | * |
| 17 | * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND |
| 18 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 19 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 20 | * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE |
| 21 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 22 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 23 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 24 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 25 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 26 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 27 | * SUCH DAMAGE. |
| 28 | * |
| 29 | * Author: Stanislav Ocovaj (socovaj@mips.com) |
| 30 | * Author: Zoran Lukic (zoranl@mips.com) |
| 31 | * |
| 32 | * Optimized MDCT/IMDCT and FFT transforms |
| 33 | * |
| 34 | * This file is part of FFmpeg. |
| 35 | * |
| 36 | * FFmpeg is free software; you can redistribute it and/or |
| 37 | * modify it under the terms of the GNU Lesser General Public |
| 38 | * License as published by the Free Software Foundation; either |
| 39 | * version 2.1 of the License, or (at your option) any later version. |
| 40 | * |
| 41 | * FFmpeg is distributed in the hope that it will be useful, |
| 42 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 43 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 44 | * Lesser General Public License for more details. |
| 45 | * |
| 46 | * You should have received a copy of the GNU Lesser General Public |
| 47 | * License along with FFmpeg; if not, write to the Free Software |
| 48 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 49 | */ |
| 50 | #include "config.h" |
| 51 | #include "libavcodec/fft.h" |
| 52 | #include "libavcodec/fft_table.h" |
| 53 | |
| 54 | /** |
| 55 | * FFT transform |
| 56 | */ |
| 57 | |
| 58 | #if HAVE_INLINE_ASM |
| 59 | static void ff_fft_calc_mips(FFTContext *s, FFTComplex *z) |
| 60 | { |
| 61 | int nbits, i, n, num_transforms, offset, step; |
| 62 | int n4, n2, n34; |
| 63 | FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8; |
| 64 | FFTComplex *tmpz; |
| 65 | float w_re, w_im; |
| 66 | float *w_re_ptr, *w_im_ptr; |
| 67 | const int fft_size = (1 << s->nbits); |
| 68 | int s_n = s->nbits; |
| 69 | int tem1, tem2; |
| 70 | float pom, pom1, pom2, pom3; |
| 71 | float temp, temp1, temp3, temp4; |
| 72 | FFTComplex * tmpz_n2, * tmpz_n34, * tmpz_n4; |
| 73 | FFTComplex * tmpz_n2_i, * tmpz_n34_i, * tmpz_n4_i, * tmpz_i; |
| 74 | |
| 75 | /** |
| 76 | *num_transforms = (0x2aab >> (16 - s->nbits)) | 1; |
| 77 | */ |
| 78 | __asm__ volatile ( |
| 79 | "li %[tem1], 16 \n\t" |
| 80 | "sub %[s_n], %[tem1], %[s_n] \n\t" |
| 81 | "li %[tem2], 10923 \n\t" |
| 82 | "srav %[tem2], %[tem2], %[s_n] \n\t" |
| 83 | "ori %[num_t],%[tem2], 1 \n\t" |
| 84 | : [num_t]"=r"(num_transforms), [s_n]"+r"(s_n), |
| 85 | [tem1]"=&r"(tem1), [tem2]"=&r"(tem2) |
| 86 | ); |
| 87 | |
| 88 | |
| 89 | for (n=0; n<num_transforms; n++) { |
| 90 | offset = ff_fft_offsets_lut[n] << 2; |
| 91 | tmpz = z + offset; |
| 92 | |
| 93 | tmp1 = tmpz[0].re + tmpz[1].re; |
| 94 | tmp5 = tmpz[2].re + tmpz[3].re; |
| 95 | tmp2 = tmpz[0].im + tmpz[1].im; |
| 96 | tmp6 = tmpz[2].im + tmpz[3].im; |
| 97 | tmp3 = tmpz[0].re - tmpz[1].re; |
| 98 | tmp8 = tmpz[2].im - tmpz[3].im; |
| 99 | tmp4 = tmpz[0].im - tmpz[1].im; |
| 100 | tmp7 = tmpz[2].re - tmpz[3].re; |
| 101 | |
| 102 | tmpz[0].re = tmp1 + tmp5; |
| 103 | tmpz[2].re = tmp1 - tmp5; |
| 104 | tmpz[0].im = tmp2 + tmp6; |
| 105 | tmpz[2].im = tmp2 - tmp6; |
| 106 | tmpz[1].re = tmp3 + tmp8; |
| 107 | tmpz[3].re = tmp3 - tmp8; |
| 108 | tmpz[1].im = tmp4 - tmp7; |
| 109 | tmpz[3].im = tmp4 + tmp7; |
| 110 | |
| 111 | } |
| 112 | |
| 113 | if (fft_size < 8) |
| 114 | return; |
| 115 | |
| 116 | num_transforms = (num_transforms >> 1) | 1; |
| 117 | |
| 118 | for (n=0; n<num_transforms; n++) { |
| 119 | offset = ff_fft_offsets_lut[n] << 3; |
| 120 | tmpz = z + offset; |
| 121 | |
| 122 | __asm__ volatile ( |
| 123 | "lwc1 %[tmp1], 32(%[tmpz]) \n\t" |
| 124 | "lwc1 %[pom], 40(%[tmpz]) \n\t" |
| 125 | "lwc1 %[tmp3], 48(%[tmpz]) \n\t" |
| 126 | "lwc1 %[pom1], 56(%[tmpz]) \n\t" |
| 127 | "lwc1 %[tmp2], 36(%[tmpz]) \n\t" |
| 128 | "lwc1 %[pom2], 44(%[tmpz]) \n\t" |
| 129 | "lwc1 %[pom3], 60(%[tmpz]) \n\t" |
| 130 | "lwc1 %[tmp4], 52(%[tmpz]) \n\t" |
| 131 | "add.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re + tmpz[5].re; |
| 132 | "add.s %[tmp3], %[tmp3], %[pom1] \n\t" // tmp3 = tmpz[6].re + tmpz[7].re; |
| 133 | "add.s %[tmp2], %[tmp2], %[pom2] \n\t" // tmp2 = tmpz[4].im + tmpz[5].im; |
| 134 | "lwc1 %[pom], 40(%[tmpz]) \n\t" |
| 135 | "add.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im + tmpz[7].im; |
| 136 | "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3; |
| 137 | "sub.s %[tmp7], %[tmp1], %[tmp3] \n\t" // tmp7 = tmp1 - tmp3; |
| 138 | "lwc1 %[tmp1], 32(%[tmpz]) \n\t" |
| 139 | "lwc1 %[pom1], 44(%[tmpz]) \n\t" |
| 140 | "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4; |
| 141 | "sub.s %[tmp8], %[tmp2], %[tmp4] \n\t" // tmp8 = tmp2 - tmp4; |
| 142 | "lwc1 %[tmp2], 36(%[tmpz]) \n\t" |
| 143 | "lwc1 %[pom2], 56(%[tmpz]) \n\t" |
| 144 | "lwc1 %[pom3], 60(%[tmpz]) \n\t" |
| 145 | "lwc1 %[tmp3], 48(%[tmpz]) \n\t" |
| 146 | "lwc1 %[tmp4], 52(%[tmpz]) \n\t" |
| 147 | "sub.s %[tmp1], %[tmp1], %[pom] \n\t" // tmp1 = tmpz[4].re - tmpz[5].re; |
| 148 | "lwc1 %[pom], 0(%[tmpz]) \n\t" |
| 149 | "sub.s %[tmp2], %[tmp2], %[pom1] \n\t" // tmp2 = tmpz[4].im - tmpz[5].im; |
| 150 | "sub.s %[tmp3], %[tmp3], %[pom2] \n\t" // tmp3 = tmpz[6].re - tmpz[7].re; |
| 151 | "lwc1 %[pom2], 4(%[tmpz]) \n\t" |
| 152 | "sub.s %[pom1], %[pom], %[tmp5] \n\t" |
| 153 | "sub.s %[tmp4], %[tmp4], %[pom3] \n\t" // tmp4 = tmpz[6].im - tmpz[7].im; |
| 154 | "add.s %[pom3], %[pom], %[tmp5] \n\t" |
| 155 | "sub.s %[pom], %[pom2], %[tmp6] \n\t" |
| 156 | "add.s %[pom2], %[pom2], %[tmp6] \n\t" |
| 157 | "swc1 %[pom1], 32(%[tmpz]) \n\t" // tmpz[4].re = tmpz[0].re - tmp5; |
| 158 | "swc1 %[pom3], 0(%[tmpz]) \n\t" // tmpz[0].re = tmpz[0].re + tmp5; |
| 159 | "swc1 %[pom], 36(%[tmpz]) \n\t" // tmpz[4].im = tmpz[0].im - tmp6; |
| 160 | "swc1 %[pom2], 4(%[tmpz]) \n\t" // tmpz[0].im = tmpz[0].im + tmp6; |
| 161 | "lwc1 %[pom1], 16(%[tmpz]) \n\t" |
| 162 | "lwc1 %[pom3], 20(%[tmpz]) \n\t" |
| 163 | "li.s %[pom], 0.7071067812 \n\t" // float pom = 0.7071067812f; |
| 164 | "add.s %[temp1],%[tmp1], %[tmp2] \n\t" |
| 165 | "sub.s %[temp], %[pom1], %[tmp8] \n\t" |
| 166 | "add.s %[pom2], %[pom3], %[tmp7] \n\t" |
| 167 | "sub.s %[temp3],%[tmp3], %[tmp4] \n\t" |
| 168 | "sub.s %[temp4],%[tmp2], %[tmp1] \n\t" |
| 169 | "swc1 %[temp], 48(%[tmpz]) \n\t" // tmpz[6].re = tmpz[2].re - tmp8; |
| 170 | "swc1 %[pom2], 52(%[tmpz]) \n\t" // tmpz[6].im = tmpz[2].im + tmp7; |
| 171 | "add.s %[pom1], %[pom1], %[tmp8] \n\t" |
| 172 | "sub.s %[pom3], %[pom3], %[tmp7] \n\t" |
| 173 | "add.s %[tmp3], %[tmp3], %[tmp4] \n\t" |
| 174 | "mul.s %[tmp5], %[pom], %[temp1] \n\t" // tmp5 = pom * (tmp1 + tmp2); |
| 175 | "mul.s %[tmp7], %[pom], %[temp3] \n\t" // tmp7 = pom * (tmp3 - tmp4); |
| 176 | "mul.s %[tmp6], %[pom], %[temp4] \n\t" // tmp6 = pom * (tmp2 - tmp1); |
| 177 | "mul.s %[tmp8], %[pom], %[tmp3] \n\t" // tmp8 = pom * (tmp3 + tmp4); |
| 178 | "swc1 %[pom1], 16(%[tmpz]) \n\t" // tmpz[2].re = tmpz[2].re + tmp8; |
| 179 | "swc1 %[pom3], 20(%[tmpz]) \n\t" // tmpz[2].im = tmpz[2].im - tmp7; |
| 180 | "add.s %[tmp1], %[tmp5], %[tmp7] \n\t" // tmp1 = tmp5 + tmp7; |
| 181 | "sub.s %[tmp3], %[tmp5], %[tmp7] \n\t" // tmp3 = tmp5 - tmp7; |
| 182 | "add.s %[tmp2], %[tmp6], %[tmp8] \n\t" // tmp2 = tmp6 + tmp8; |
| 183 | "sub.s %[tmp4], %[tmp6], %[tmp8] \n\t" // tmp4 = tmp6 - tmp8; |
| 184 | "lwc1 %[temp], 8(%[tmpz]) \n\t" |
| 185 | "lwc1 %[temp1],12(%[tmpz]) \n\t" |
| 186 | "lwc1 %[pom], 24(%[tmpz]) \n\t" |
| 187 | "lwc1 %[pom2], 28(%[tmpz]) \n\t" |
| 188 | "sub.s %[temp4],%[temp], %[tmp1] \n\t" |
| 189 | "sub.s %[temp3],%[temp1], %[tmp2] \n\t" |
| 190 | "add.s %[temp], %[temp], %[tmp1] \n\t" |
| 191 | "add.s %[temp1],%[temp1], %[tmp2] \n\t" |
| 192 | "sub.s %[pom1], %[pom], %[tmp4] \n\t" |
| 193 | "add.s %[pom3], %[pom2], %[tmp3] \n\t" |
| 194 | "add.s %[pom], %[pom], %[tmp4] \n\t" |
| 195 | "sub.s %[pom2], %[pom2], %[tmp3] \n\t" |
| 196 | "swc1 %[temp4],40(%[tmpz]) \n\t" // tmpz[5].re = tmpz[1].re - tmp1; |
| 197 | "swc1 %[temp3],44(%[tmpz]) \n\t" // tmpz[5].im = tmpz[1].im - tmp2; |
| 198 | "swc1 %[temp], 8(%[tmpz]) \n\t" // tmpz[1].re = tmpz[1].re + tmp1; |
| 199 | "swc1 %[temp1],12(%[tmpz]) \n\t" // tmpz[1].im = tmpz[1].im + tmp2; |
| 200 | "swc1 %[pom1], 56(%[tmpz]) \n\t" // tmpz[7].re = tmpz[3].re - tmp4; |
| 201 | "swc1 %[pom3], 60(%[tmpz]) \n\t" // tmpz[7].im = tmpz[3].im + tmp3; |
| 202 | "swc1 %[pom], 24(%[tmpz]) \n\t" // tmpz[3].re = tmpz[3].re + tmp4; |
| 203 | "swc1 %[pom2], 28(%[tmpz]) \n\t" // tmpz[3].im = tmpz[3].im - tmp3; |
| 204 | : [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), |
| 205 | [tmp3]"=&f"(tmp3), [tmp2]"=&f"(tmp2), [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp7]"=&f"(tmp7), |
| 206 | [tmp6]"=&f"(tmp6), [tmp8]"=&f"(tmp8), [pom3]"=&f"(pom3),[temp]"=&f"(temp), [temp1]"=&f"(temp1), |
| 207 | [temp3]"=&f"(temp3), [temp4]"=&f"(temp4) |
| 208 | : [tmpz]"r"(tmpz) |
| 209 | : "memory" |
| 210 | ); |
| 211 | } |
| 212 | |
| 213 | step = 1 << (MAX_LOG2_NFFT - 4); |
| 214 | n4 = 4; |
| 215 | |
| 216 | for (nbits=4; nbits<=s->nbits; nbits++) { |
| 217 | /* |
| 218 | * num_transforms = (num_transforms >> 1) | 1; |
| 219 | */ |
| 220 | __asm__ volatile ( |
| 221 | "sra %[num_t], %[num_t], 1 \n\t" |
| 222 | "ori %[num_t], %[num_t], 1 \n\t" |
| 223 | |
| 224 | : [num_t] "+r" (num_transforms) |
| 225 | ); |
| 226 | n2 = 2 * n4; |
| 227 | n34 = 3 * n4; |
| 228 | |
| 229 | for (n=0; n<num_transforms; n++) { |
| 230 | offset = ff_fft_offsets_lut[n] << nbits; |
| 231 | tmpz = z + offset; |
| 232 | |
| 233 | tmpz_n2 = tmpz + n2; |
| 234 | tmpz_n4 = tmpz + n4; |
| 235 | tmpz_n34 = tmpz + n34; |
| 236 | |
| 237 | __asm__ volatile ( |
| 238 | "lwc1 %[pom1], 0(%[tmpz_n2]) \n\t" |
| 239 | "lwc1 %[pom], 0(%[tmpz_n34]) \n\t" |
| 240 | "lwc1 %[pom2], 4(%[tmpz_n2]) \n\t" |
| 241 | "lwc1 %[pom3], 4(%[tmpz_n34]) \n\t" |
| 242 | "lwc1 %[temp1],0(%[tmpz]) \n\t" |
| 243 | "lwc1 %[temp3],4(%[tmpz]) \n\t" |
| 244 | "add.s %[tmp5], %[pom1], %[pom] \n\t" // tmp5 = tmpz[ n2].re + tmpz[n34].re; |
| 245 | "sub.s %[tmp1], %[pom1], %[pom] \n\t" // tmp1 = tmpz[ n2].re - tmpz[n34].re; |
| 246 | "add.s %[tmp6], %[pom2], %[pom3] \n\t" // tmp6 = tmpz[ n2].im + tmpz[n34].im; |
| 247 | "sub.s %[tmp2], %[pom2], %[pom3] \n\t" // tmp2 = tmpz[ n2].im - tmpz[n34].im; |
| 248 | "sub.s %[temp], %[temp1], %[tmp5] \n\t" |
| 249 | "add.s %[temp1],%[temp1], %[tmp5] \n\t" |
| 250 | "sub.s %[temp4],%[temp3], %[tmp6] \n\t" |
| 251 | "add.s %[temp3],%[temp3], %[tmp6] \n\t" |
| 252 | "swc1 %[temp], 0(%[tmpz_n2]) \n\t" // tmpz[ n2].re = tmpz[ 0].re - tmp5; |
| 253 | "swc1 %[temp1],0(%[tmpz]) \n\t" // tmpz[ 0].re = tmpz[ 0].re + tmp5; |
| 254 | "lwc1 %[pom1], 0(%[tmpz_n4]) \n\t" |
| 255 | "swc1 %[temp4],4(%[tmpz_n2]) \n\t" // tmpz[ n2].im = tmpz[ 0].im - tmp6; |
| 256 | "lwc1 %[temp], 4(%[tmpz_n4]) \n\t" |
| 257 | "swc1 %[temp3],4(%[tmpz]) \n\t" // tmpz[ 0].im = tmpz[ 0].im + tmp6; |
| 258 | "sub.s %[pom], %[pom1], %[tmp2] \n\t" |
| 259 | "add.s %[pom1], %[pom1], %[tmp2] \n\t" |
| 260 | "add.s %[temp1],%[temp], %[tmp1] \n\t" |
| 261 | "sub.s %[temp], %[temp], %[tmp1] \n\t" |
| 262 | "swc1 %[pom], 0(%[tmpz_n34]) \n\t" // tmpz[n34].re = tmpz[n4].re - tmp2; |
| 263 | "swc1 %[pom1], 0(%[tmpz_n4]) \n\t" // tmpz[ n4].re = tmpz[n4].re + tmp2; |
| 264 | "swc1 %[temp1],4(%[tmpz_n34]) \n\t" // tmpz[n34].im = tmpz[n4].im + tmp1; |
| 265 | "swc1 %[temp], 4(%[tmpz_n4]) \n\t" // tmpz[ n4].im = tmpz[n4].im - tmp1; |
| 266 | : [tmp5]"=&f"(tmp5), |
| 267 | [tmp1]"=&f"(tmp1), [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), |
| 268 | [tmp2]"=&f"(tmp2), [tmp6]"=&f"(tmp6), [pom3]"=&f"(pom3), |
| 269 | [temp]"=&f"(temp), [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4) |
| 270 | : [tmpz]"r"(tmpz), [tmpz_n2]"r"(tmpz_n2), [tmpz_n34]"r"(tmpz_n34), [tmpz_n4]"r"(tmpz_n4) |
| 271 | : "memory" |
| 272 | ); |
| 273 | |
| 274 | w_re_ptr = (float*)(ff_cos_65536 + step); |
| 275 | w_im_ptr = (float*)(ff_cos_65536 + MAX_FFT_SIZE/4 - step); |
| 276 | |
| 277 | for (i=1; i<n4; i++) { |
| 278 | w_re = w_re_ptr[0]; |
| 279 | w_im = w_im_ptr[0]; |
| 280 | tmpz_n2_i = tmpz_n2 + i; |
| 281 | tmpz_n4_i = tmpz_n4 + i; |
| 282 | tmpz_n34_i= tmpz_n34 + i; |
| 283 | tmpz_i = tmpz + i; |
| 284 | |
| 285 | __asm__ volatile ( |
| 286 | "lwc1 %[temp], 0(%[tmpz_n2_i]) \n\t" |
| 287 | "lwc1 %[temp1], 4(%[tmpz_n2_i]) \n\t" |
| 288 | "lwc1 %[pom], 0(%[tmpz_n34_i]) \n\t" |
| 289 | "lwc1 %[pom1], 4(%[tmpz_n34_i]) \n\t" |
| 290 | "mul.s %[temp3], %[w_im], %[temp] \n\t" |
| 291 | "mul.s %[temp4], %[w_im], %[temp1] \n\t" |
| 292 | "mul.s %[pom2], %[w_im], %[pom1] \n\t" |
| 293 | "mul.s %[pom3], %[w_im], %[pom] \n\t" |
| 294 | "msub.s %[tmp2], %[temp3], %[w_re], %[temp1] \n\t" // tmp2 = w_re * tmpz[ n2+i].im - w_im * tmpz[ n2+i].re; |
| 295 | "madd.s %[tmp1], %[temp4], %[w_re], %[temp] \n\t" // tmp1 = w_re * tmpz[ n2+i].re + w_im * tmpz[ n2+i].im; |
| 296 | "msub.s %[tmp3], %[pom2], %[w_re], %[pom] \n\t" // tmp3 = w_re * tmpz[n34+i].re - w_im * tmpz[n34+i].im; |
| 297 | "madd.s %[tmp4], %[pom3], %[w_re], %[pom1] \n\t" // tmp4 = w_re * tmpz[n34+i].im + w_im * tmpz[n34+i].re; |
| 298 | "lwc1 %[temp], 0(%[tmpz_i]) \n\t" |
| 299 | "lwc1 %[pom], 4(%[tmpz_i]) \n\t" |
| 300 | "add.s %[tmp5], %[tmp1], %[tmp3] \n\t" // tmp5 = tmp1 + tmp3; |
| 301 | "sub.s %[tmp1], %[tmp1], %[tmp3] \n\t" // tmp1 = tmp1 - tmp3; |
| 302 | "add.s %[tmp6], %[tmp2], %[tmp4] \n\t" // tmp6 = tmp2 + tmp4; |
| 303 | "sub.s %[tmp2], %[tmp2], %[tmp4] \n\t" // tmp2 = tmp2 - tmp4; |
| 304 | "sub.s %[temp1], %[temp], %[tmp5] \n\t" |
| 305 | "add.s %[temp], %[temp], %[tmp5] \n\t" |
| 306 | "sub.s %[pom1], %[pom], %[tmp6] \n\t" |
| 307 | "add.s %[pom], %[pom], %[tmp6] \n\t" |
| 308 | "lwc1 %[temp3], 0(%[tmpz_n4_i]) \n\t" |
| 309 | "lwc1 %[pom2], 4(%[tmpz_n4_i]) \n\t" |
| 310 | "swc1 %[temp1], 0(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].re = tmpz[ i].re - tmp5; |
| 311 | "swc1 %[temp], 0(%[tmpz_i]) \n\t" // tmpz[ i].re = tmpz[ i].re + tmp5; |
| 312 | "swc1 %[pom1], 4(%[tmpz_n2_i]) \n\t" // tmpz[ n2+i].im = tmpz[ i].im - tmp6; |
| 313 | "swc1 %[pom] , 4(%[tmpz_i]) \n\t" // tmpz[ i].im = tmpz[ i].im + tmp6; |
| 314 | "sub.s %[temp4], %[temp3], %[tmp2] \n\t" |
| 315 | "add.s %[pom3], %[pom2], %[tmp1] \n\t" |
| 316 | "add.s %[temp3], %[temp3], %[tmp2] \n\t" |
| 317 | "sub.s %[pom2], %[pom2], %[tmp1] \n\t" |
| 318 | "swc1 %[temp4], 0(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].re = tmpz[n4+i].re - tmp2; |
| 319 | "swc1 %[pom3], 4(%[tmpz_n34_i]) \n\t" // tmpz[n34+i].im = tmpz[n4+i].im + tmp1; |
| 320 | "swc1 %[temp3], 0(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].re = tmpz[n4+i].re + tmp2; |
| 321 | "swc1 %[pom2], 4(%[tmpz_n4_i]) \n\t" // tmpz[ n4+i].im = tmpz[n4+i].im - tmp1; |
| 322 | : [tmp1]"=&f"(tmp1), [tmp2]"=&f" (tmp2), [temp]"=&f"(temp), [tmp3]"=&f"(tmp3), |
| 323 | [tmp4]"=&f"(tmp4), [tmp5]"=&f"(tmp5), [tmp6]"=&f"(tmp6), |
| 324 | [temp1]"=&f"(temp1), [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), |
| 325 | [pom]"=&f"(pom), [pom1]"=&f"(pom1), [pom2]"=&f"(pom2), [pom3]"=&f"(pom3) |
| 326 | : [w_re]"f"(w_re), [w_im]"f"(w_im), |
| 327 | [tmpz_i]"r"(tmpz_i),[tmpz_n2_i]"r"(tmpz_n2_i), |
| 328 | [tmpz_n34_i]"r"(tmpz_n34_i), [tmpz_n4_i]"r"(tmpz_n4_i) |
| 329 | : "memory" |
| 330 | ); |
| 331 | w_re_ptr += step; |
| 332 | w_im_ptr -= step; |
| 333 | } |
| 334 | } |
| 335 | step >>= 1; |
| 336 | n4 <<= 1; |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | /** |
| 341 | * MDCT/IMDCT transforms. |
| 342 | */ |
| 343 | |
| 344 | static void ff_imdct_half_mips(FFTContext *s, FFTSample *output, const FFTSample *input) |
| 345 | { |
| 346 | int k, n8, n4, n2, n, j; |
| 347 | const uint16_t *revtab = s->revtab; |
| 348 | const FFTSample *tcos = s->tcos; |
| 349 | const FFTSample *tsin = s->tsin; |
| 350 | const FFTSample *in1, *in2, *in3, *in4; |
| 351 | FFTComplex *z = (FFTComplex *)output; |
| 352 | |
| 353 | int j1; |
| 354 | const float *tcos1, *tsin1, *tcos2, *tsin2; |
| 355 | float temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, |
| 356 | temp9, temp10, temp11, temp12, temp13, temp14, temp15, temp16; |
| 357 | FFTComplex *z1, *z2; |
| 358 | |
| 359 | n = 1 << s->mdct_bits; |
| 360 | n2 = n >> 1; |
| 361 | n4 = n >> 2; |
| 362 | n8 = n >> 3; |
| 363 | |
| 364 | /* pre rotation */ |
| 365 | in1 = input; |
| 366 | in2 = input + n2 - 1; |
| 367 | in3 = input + 2; |
| 368 | in4 = input + n2 - 3; |
| 369 | |
| 370 | tcos1 = tcos; |
| 371 | tsin1 = tsin; |
| 372 | |
| 373 | /* n4 = 64 or 128 */ |
| 374 | for(k = 0; k < n4; k += 2) { |
| 375 | j = revtab[k ]; |
| 376 | j1 = revtab[k + 1]; |
| 377 | |
| 378 | __asm__ volatile ( |
| 379 | "lwc1 %[temp1], 0(%[in2]) \t\n" |
| 380 | "lwc1 %[temp2], 0(%[tcos1]) \t\n" |
| 381 | "lwc1 %[temp3], 0(%[tsin1]) \t\n" |
| 382 | "lwc1 %[temp4], 0(%[in1]) \t\n" |
| 383 | "lwc1 %[temp5], 0(%[in4]) \t\n" |
| 384 | "mul.s %[temp9], %[temp1], %[temp2] \t\n" |
| 385 | "mul.s %[temp10], %[temp1], %[temp3] \t\n" |
| 386 | "lwc1 %[temp6], 4(%[tcos1]) \t\n" |
| 387 | "lwc1 %[temp7], 4(%[tsin1]) \t\n" |
| 388 | "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n" |
| 389 | "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n" |
| 390 | "mul.s %[temp11], %[temp5], %[temp6] \t\n" |
| 391 | "mul.s %[temp12], %[temp5], %[temp7] \t\n" |
| 392 | "lwc1 %[temp8], 0(%[in3]) \t\n" |
| 393 | "addiu %[tcos1], %[tcos1], 8 \t\n" |
| 394 | "addiu %[tsin1], %[tsin1], 8 \t\n" |
| 395 | "addiu %[in1], %[in1], 16 \t\n" |
| 396 | "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n" |
| 397 | "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n" |
| 398 | "addiu %[in2], %[in2], -16 \t\n" |
| 399 | "addiu %[in3], %[in3], 16 \t\n" |
| 400 | "addiu %[in4], %[in4], -16 \t\n" |
| 401 | |
| 402 | : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), |
| 403 | [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), |
| 404 | [temp5]"=&f"(temp5), [temp6]"=&f"(temp6), |
| 405 | [temp7]"=&f"(temp7), [temp8]"=&f"(temp8), |
| 406 | [temp9]"=&f"(temp9), [temp10]"=&f"(temp10), |
| 407 | [temp11]"=&f"(temp11), [temp12]"=&f"(temp12), |
| 408 | [tsin1]"+r"(tsin1), [tcos1]"+r"(tcos1), |
| 409 | [in1]"+r"(in1), [in2]"+r"(in2), |
| 410 | [in3]"+r"(in3), [in4]"+r"(in4) |
| 411 | : |
| 412 | : "memory" |
| 413 | ); |
| 414 | |
| 415 | z[j ].re = temp9; |
| 416 | z[j ].im = temp10; |
| 417 | z[j1].re = temp11; |
| 418 | z[j1].im = temp12; |
| 419 | } |
| 420 | |
| 421 | s->fft_calc(s, z); |
| 422 | |
| 423 | /* post rotation + reordering */ |
| 424 | /* n8 = 32 or 64 */ |
| 425 | for(k = 0; k < n8; k += 2) { |
| 426 | tcos1 = &tcos[n8 - k - 2]; |
| 427 | tsin1 = &tsin[n8 - k - 2]; |
| 428 | tcos2 = &tcos[n8 + k]; |
| 429 | tsin2 = &tsin[n8 + k]; |
| 430 | z1 = &z[n8 - k - 2]; |
| 431 | z2 = &z[n8 + k ]; |
| 432 | |
| 433 | __asm__ volatile ( |
| 434 | "lwc1 %[temp1], 12(%[z1]) \t\n" |
| 435 | "lwc1 %[temp2], 4(%[tsin1]) \t\n" |
| 436 | "lwc1 %[temp3], 4(%[tcos1]) \t\n" |
| 437 | "lwc1 %[temp4], 8(%[z1]) \t\n" |
| 438 | "lwc1 %[temp5], 4(%[z1]) \t\n" |
| 439 | "mul.s %[temp9], %[temp1], %[temp2] \t\n" |
| 440 | "mul.s %[temp10], %[temp1], %[temp3] \t\n" |
| 441 | "lwc1 %[temp6], 0(%[tsin1]) \t\n" |
| 442 | "lwc1 %[temp7], 0(%[tcos1]) \t\n" |
| 443 | "nmsub.s %[temp9], %[temp9], %[temp4], %[temp3] \t\n" |
| 444 | "madd.s %[temp10], %[temp10], %[temp4], %[temp2] \t\n" |
| 445 | "mul.s %[temp11], %[temp5], %[temp6] \t\n" |
| 446 | "mul.s %[temp12], %[temp5], %[temp7] \t\n" |
| 447 | "lwc1 %[temp8], 0(%[z1]) \t\n" |
| 448 | "lwc1 %[temp1], 4(%[z2]) \t\n" |
| 449 | "lwc1 %[temp2], 0(%[tsin2]) \t\n" |
| 450 | "lwc1 %[temp3], 0(%[tcos2]) \t\n" |
| 451 | "nmsub.s %[temp11], %[temp11], %[temp8], %[temp7] \t\n" |
| 452 | "madd.s %[temp12], %[temp12], %[temp8], %[temp6] \t\n" |
| 453 | "mul.s %[temp13], %[temp1], %[temp2] \t\n" |
| 454 | "mul.s %[temp14], %[temp1], %[temp3] \t\n" |
| 455 | "lwc1 %[temp4], 0(%[z2]) \t\n" |
| 456 | "lwc1 %[temp5], 12(%[z2]) \t\n" |
| 457 | "lwc1 %[temp6], 4(%[tsin2]) \t\n" |
| 458 | "lwc1 %[temp7], 4(%[tcos2]) \t\n" |
| 459 | "nmsub.s %[temp13], %[temp13], %[temp4], %[temp3] \t\n" |
| 460 | "madd.s %[temp14], %[temp14], %[temp4], %[temp2] \t\n" |
| 461 | "mul.s %[temp15], %[temp5], %[temp6] \t\n" |
| 462 | "mul.s %[temp16], %[temp5], %[temp7] \t\n" |
| 463 | "lwc1 %[temp8], 8(%[z2]) \t\n" |
| 464 | "nmsub.s %[temp15], %[temp15], %[temp8], %[temp7] \t\n" |
| 465 | "madd.s %[temp16], %[temp16], %[temp8], %[temp6] \t\n" |
| 466 | : [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), |
| 467 | [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), |
| 468 | [temp5]"=&f"(temp5), [temp6]"=&f"(temp6), |
| 469 | [temp7]"=&f"(temp7), [temp8]"=&f"(temp8), |
| 470 | [temp9]"=&f"(temp9), [temp10]"=&f"(temp10), |
| 471 | [temp11]"=&f"(temp11), [temp12]"=&f"(temp12), |
| 472 | [temp13]"=&f"(temp13), [temp14]"=&f"(temp14), |
| 473 | [temp15]"=&f"(temp15), [temp16]"=&f"(temp16) |
| 474 | : [z1]"r"(z1), [z2]"r"(z2), |
| 475 | [tsin1]"r"(tsin1), [tcos1]"r"(tcos1), |
| 476 | [tsin2]"r"(tsin2), [tcos2]"r"(tcos2) |
| 477 | : "memory" |
| 478 | ); |
| 479 | |
| 480 | z1[1].re = temp9; |
| 481 | z1[1].im = temp14; |
| 482 | z2[0].re = temp13; |
| 483 | z2[0].im = temp10; |
| 484 | |
| 485 | z1[0].re = temp11; |
| 486 | z1[0].im = temp16; |
| 487 | z2[1].re = temp15; |
| 488 | z2[1].im = temp12; |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | /** |
| 493 | * Compute inverse MDCT of size N = 2^nbits |
| 494 | * @param output N samples |
| 495 | * @param input N/2 samples |
| 496 | */ |
| 497 | static void ff_imdct_calc_mips(FFTContext *s, FFTSample *output, const FFTSample *input) |
| 498 | { |
| 499 | int k; |
| 500 | int n = 1 << s->mdct_bits; |
| 501 | int n2 = n >> 1; |
| 502 | int n4 = n >> 2; |
| 503 | |
| 504 | ff_imdct_half_mips(s, output+n4, input); |
| 505 | |
| 506 | for(k = 0; k < n4; k+=4) { |
| 507 | output[k] = -output[n2-k-1]; |
| 508 | output[k+1] = -output[n2-k-2]; |
| 509 | output[k+2] = -output[n2-k-3]; |
| 510 | output[k+3] = -output[n2-k-4]; |
| 511 | |
| 512 | output[n-k-1] = output[n2+k]; |
| 513 | output[n-k-2] = output[n2+k+1]; |
| 514 | output[n-k-3] = output[n2+k+2]; |
| 515 | output[n-k-4] = output[n2+k+3]; |
| 516 | } |
| 517 | } |
| 518 | #endif /* HAVE_INLINE_ASM */ |
| 519 | |
| 520 | av_cold void ff_fft_init_mips(FFTContext *s) |
| 521 | { |
| 522 | int n=0; |
| 523 | |
| 524 | ff_fft_lut_init(ff_fft_offsets_lut, 0, 1 << 16, &n); |
| 525 | ff_init_ff_cos_tabs(16); |
| 526 | |
| 527 | #if HAVE_INLINE_ASM |
| 528 | s->fft_calc = ff_fft_calc_mips; |
| 529 | #if CONFIG_MDCT |
| 530 | s->imdct_calc = ff_imdct_calc_mips; |
| 531 | s->imdct_half = ff_imdct_half_mips; |
| 532 | #endif |
| 533 | #endif |
| 534 | } |