| @@ -1,2503 +0,0 @@ |
| 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 | * Szabolcs Pal (sabolc@mips.com) | | |
| 31 | * | | |
| 32 | * AAC coefficients encoder optimized for MIPS floating-point architecture | | |
| 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 | | | |
| 51 | /** | | |
| 52 | * @file | | |
| 53 | * Reference: libavcodec/aaccoder.c | | |
| 54 | */ | | |
| 55 | | | |
| 56 | #include "libavutil/libm.h" | | |
| 57 | | | |
| 58 | #include <float.h> | | |
| 59 | #include "libavutil/mathematics.h" | | |
| 60 | #include "libavcodec/avcodec.h" | | |
| 61 | #include "libavcodec/put_bits.h" | | |
| 62 | #include "libavcodec/aac.h" | | |
| 63 | #include "libavcodec/aacenc.h" | | |
| 64 | #include "libavcodec/aacencdsp.h" | | |
| 65 | #include "libavcodec/aactab.h" | | |
| 66 | #include "libavcodec/aacenctab.h" | | |
| 67 | #include "libavcodec/aacenc_utils.h" | | |
| 68 | | | |
| 69 | #if HAVE_INLINE_ASM | | |
| 70 | #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6 | | |
| 71 | typedef struct BandCodingPath { | | |
| 72 | int prev_idx; | | |
| 73 | float cost; | | |
| 74 | int run; | | |
| 75 | } BandCodingPath; | | |
| 76 | | | |
| 77 | static const uint8_t uquad_sign_bits[81] = { | | |
| 78 | 0, 1, 1, 1, 2, 2, 1, 2, 2, | | |
| 79 | 1, 2, 2, 2, 3, 3, 2, 3, 3, | | |
| 80 | 1, 2, 2, 2, 3, 3, 2, 3, 3, | | |
| 81 | 1, 2, 2, 2, 3, 3, 2, 3, 3, | | |
| 82 | 2, 3, 3, 3, 4, 4, 3, 4, 4, | | |
| 83 | 2, 3, 3, 3, 4, 4, 3, 4, 4, | | |
| 84 | 1, 2, 2, 2, 3, 3, 2, 3, 3, | | |
| 85 | 2, 3, 3, 3, 4, 4, 3, 4, 4, | | |
| 86 | 2, 3, 3, 3, 4, 4, 3, 4, 4 | | |
| 87 | }; | | |
| 88 | | | |
| 89 | static const uint8_t upair7_sign_bits[64] = { | | |
| 90 | 0, 1, 1, 1, 1, 1, 1, 1, | | |
| 91 | 1, 2, 2, 2, 2, 2, 2, 2, | | |
| 92 | 1, 2, 2, 2, 2, 2, 2, 2, | | |
| 93 | 1, 2, 2, 2, 2, 2, 2, 2, | | |
| 94 | 1, 2, 2, 2, 2, 2, 2, 2, | | |
| 95 | 1, 2, 2, 2, 2, 2, 2, 2, | | |
| 96 | 1, 2, 2, 2, 2, 2, 2, 2, | | |
| 97 | 1, 2, 2, 2, 2, 2, 2, 2, | | |
| 98 | }; | | |
| 99 | | | |
| 100 | static const uint8_t upair12_sign_bits[169] = { | | |
| 101 | 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | | |
| 102 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 103 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 104 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 105 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 106 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 107 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 108 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 109 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 110 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 111 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 112 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 113 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 | | |
| 114 | }; | | |
| 115 | | | |
| 116 | static const uint8_t esc_sign_bits[289] = { | | |
| 117 | 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | | |
| 118 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 119 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 120 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 121 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 122 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 123 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 124 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 125 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 126 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 127 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 128 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 129 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 130 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 131 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 132 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | | |
| 133 | 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 | | |
| 134 | }; | | |
| 135 | | | |
| 136 | /** | | |
| 137 | * Functions developed from template function and optimized for quantizing and encoding band | | |
| 138 | */ | | |
| 139 | static void quantize_and_encode_band_cost_SQUAD_mips(struct AACEncContext *s, | | |
| 140 | PutBitContext *pb, const float *in, float *out, | | |
| 141 | const float *scaled, int size, int scale_idx, | | |
| 142 | int cb, const float lambda, const float uplim, | | |
| 143 | int *bits, float *energy, const float ROUNDING) | | |
| 144 | { | | |
| 145 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 146 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 147 | int i; | | |
| 148 | int qc1, qc2, qc3, qc4; | | |
| 149 | float qenergy = 0.0f; | | |
| 150 | | | |
| 151 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 152 | uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1]; | | |
| 153 | float *p_vec = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 154 | | | |
| 155 | abs_pow34_v(s->scoefs, in, size); | | |
| 156 | scaled = s->scoefs; | | |
| 157 | for (i = 0; i < size; i += 4) { | | |
| 158 | int curidx; | | |
| 159 | int *in_int = (int *)&in[i]; | | |
| 160 | int t0, t1, t2, t3, t4, t5, t6, t7; | | |
| 161 | const float *vec; | | |
| 162 | | | |
| 163 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 164 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 165 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 166 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 167 | | | |
| 168 | __asm__ volatile ( | | |
| 169 | ".set push \n\t" | | |
| 170 | ".set noreorder \n\t" | | |
| 171 | | | |
| 172 | "slt %[qc1], $zero, %[qc1] \n\t" | | |
| 173 | "slt %[qc2], $zero, %[qc2] \n\t" | | |
| 174 | "slt %[qc3], $zero, %[qc3] \n\t" | | |
| 175 | "slt %[qc4], $zero, %[qc4] \n\t" | | |
| 176 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 177 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 178 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 179 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 180 | "srl %[t0], %[t0], 31 \n\t" | | |
| 181 | "srl %[t1], %[t1], 31 \n\t" | | |
| 182 | "srl %[t2], %[t2], 31 \n\t" | | |
| 183 | "srl %[t3], %[t3], 31 \n\t" | | |
| 184 | "subu %[t4], $zero, %[qc1] \n\t" | | |
| 185 | "subu %[t5], $zero, %[qc2] \n\t" | | |
| 186 | "subu %[t6], $zero, %[qc3] \n\t" | | |
| 187 | "subu %[t7], $zero, %[qc4] \n\t" | | |
| 188 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 189 | "movn %[qc2], %[t5], %[t1] \n\t" | | |
| 190 | "movn %[qc3], %[t6], %[t2] \n\t" | | |
| 191 | "movn %[qc4], %[t7], %[t3] \n\t" | | |
| 192 | | | |
| 193 | ".set pop \n\t" | | |
| 194 | | | |
| 195 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 196 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 197 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 198 | [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7) | | |
| 199 | : [in_int]"r"(in_int) | | |
| 200 | : "memory" | | |
| 201 | ); | | |
| 202 | | | |
| 203 | curidx = qc1; | | |
| 204 | curidx *= 3; | | |
| 205 | curidx += qc2; | | |
| 206 | curidx *= 3; | | |
| 207 | curidx += qc3; | | |
| 208 | curidx *= 3; | | |
| 209 | curidx += qc4; | | |
| 210 | curidx += 40; | | |
| 211 | | | |
| 212 | put_bits(pb, p_bits[curidx], p_codes[curidx]); | | |
| 213 | | | |
| 214 | if (out || energy) { | | |
| 215 | float e1,e2,e3,e4; | | |
| 216 | vec = &p_vec[curidx*4]; | | |
| 217 | e1 = vec[0] * IQ; | | |
| 218 | e2 = vec[1] * IQ; | | |
| 219 | e3 = vec[2] * IQ; | | |
| 220 | e4 = vec[3] * IQ; | | |
| 221 | if (out) { | | |
| 222 | out[i+0] = e1; | | |
| 223 | out[i+1] = e2; | | |
| 224 | out[i+2] = e3; | | |
| 225 | out[i+3] = e4; | | |
| 226 | } | | |
| 227 | if (energy) | | |
| 228 | qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4); | | |
| 229 | } | | |
| 230 | } | | |
| 231 | if (energy) | | |
| 232 | *energy = qenergy; | | |
| 233 | } | | |
| 234 | | | |
| 235 | static void quantize_and_encode_band_cost_UQUAD_mips(struct AACEncContext *s, | | |
| 236 | PutBitContext *pb, const float *in, float *out, | | |
| 237 | const float *scaled, int size, int scale_idx, | | |
| 238 | int cb, const float lambda, const float uplim, | | |
| 239 | int *bits, float *energy, const float ROUNDING) | | |
| 240 | { | | |
| 241 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 242 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 243 | int i; | | |
| 244 | int qc1, qc2, qc3, qc4; | | |
| 245 | float qenergy = 0.0f; | | |
| 246 | | | |
| 247 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 248 | uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1]; | | |
| 249 | float *p_vec = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 250 | | | |
| 251 | abs_pow34_v(s->scoefs, in, size); | | |
| 252 | scaled = s->scoefs; | | |
| 253 | for (i = 0; i < size; i += 4) { | | |
| 254 | int curidx, sign, count; | | |
| 255 | int *in_int = (int *)&in[i]; | | |
| 256 | uint8_t v_bits; | | |
| 257 | unsigned int v_codes; | | |
| 258 | int t0, t1, t2, t3, t4; | | |
| 259 | const float *vec; | | |
| 260 | | | |
| 261 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 262 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 263 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 264 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 265 | | | |
| 266 | __asm__ volatile ( | | |
| 267 | ".set push \n\t" | | |
| 268 | ".set noreorder \n\t" | | |
| 269 | | | |
| 270 | "ori %[t4], $zero, 2 \n\t" | | |
| 271 | "ori %[sign], $zero, 0 \n\t" | | |
| 272 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 273 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 274 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 275 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 276 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 277 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 278 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 279 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 280 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 281 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 282 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 283 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 284 | "slt %[t0], %[t0], $zero \n\t" | | |
| 285 | "movn %[sign], %[t0], %[qc1] \n\t" | | |
| 286 | "slt %[t1], %[t1], $zero \n\t" | | |
| 287 | "slt %[t2], %[t2], $zero \n\t" | | |
| 288 | "slt %[t3], %[t3], $zero \n\t" | | |
| 289 | "sll %[t0], %[sign], 1 \n\t" | | |
| 290 | "or %[t0], %[t0], %[t1] \n\t" | | |
| 291 | "movn %[sign], %[t0], %[qc2] \n\t" | | |
| 292 | "slt %[t4], $zero, %[qc1] \n\t" | | |
| 293 | "slt %[t1], $zero, %[qc2] \n\t" | | |
| 294 | "slt %[count], $zero, %[qc3] \n\t" | | |
| 295 | "sll %[t0], %[sign], 1 \n\t" | | |
| 296 | "or %[t0], %[t0], %[t2] \n\t" | | |
| 297 | "movn %[sign], %[t0], %[qc3] \n\t" | | |
| 298 | "slt %[t2], $zero, %[qc4] \n\t" | | |
| 299 | "addu %[count], %[count], %[t4] \n\t" | | |
| 300 | "addu %[count], %[count], %[t1] \n\t" | | |
| 301 | "sll %[t0], %[sign], 1 \n\t" | | |
| 302 | "or %[t0], %[t0], %[t3] \n\t" | | |
| 303 | "movn %[sign], %[t0], %[qc4] \n\t" | | |
| 304 | "addu %[count], %[count], %[t2] \n\t" | | |
| 305 | | | |
| 306 | ".set pop \n\t" | | |
| 307 | | | |
| 308 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 309 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 310 | [sign]"=&r"(sign), [count]"=&r"(count), | | |
| 311 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 312 | [t4]"=&r"(t4) | | |
| 313 | : [in_int]"r"(in_int) | | |
| 314 | : "memory" | | |
| 315 | ); | | |
| 316 | | | |
| 317 | curidx = qc1; | | |
| 318 | curidx *= 3; | | |
| 319 | curidx += qc2; | | |
| 320 | curidx *= 3; | | |
| 321 | curidx += qc3; | | |
| 322 | curidx *= 3; | | |
| 323 | curidx += qc4; | | |
| 324 | | | |
| 325 | v_codes = (p_codes[curidx] << count) | (sign & ((1 << count) - 1)); | | |
| 326 | v_bits = p_bits[curidx] + count; | | |
| 327 | put_bits(pb, v_bits, v_codes); | | |
| 328 | | | |
| 329 | if (out || energy) { | | |
| 330 | float e1,e2,e3,e4; | | |
| 331 | vec = &p_vec[curidx*4]; | | |
| 332 | e1 = copysignf(vec[0] * IQ, in[i+0]); | | |
| 333 | e2 = copysignf(vec[1] * IQ, in[i+1]); | | |
| 334 | e3 = copysignf(vec[2] * IQ, in[i+2]); | | |
| 335 | e4 = copysignf(vec[3] * IQ, in[i+3]); | | |
| 336 | if (out) { | | |
| 337 | out[i+0] = e1; | | |
| 338 | out[i+1] = e2; | | |
| 339 | out[i+2] = e3; | | |
| 340 | out[i+3] = e4; | | |
| 341 | } | | |
| 342 | if (energy) | | |
| 343 | qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4); | | |
| 344 | } | | |
| 345 | } | | |
| 346 | if (energy) | | |
| 347 | *energy = qenergy; | | |
| 348 | } | | |
| 349 | | | |
| 350 | static void quantize_and_encode_band_cost_SPAIR_mips(struct AACEncContext *s, | | |
| 351 | PutBitContext *pb, const float *in, float *out, | | |
| 352 | const float *scaled, int size, int scale_idx, | | |
| 353 | int cb, const float lambda, const float uplim, | | |
| 354 | int *bits, float *energy, const float ROUNDING) | | |
| 355 | { | | |
| 356 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 357 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 358 | int i; | | |
| 359 | int qc1, qc2, qc3, qc4; | | |
| 360 | float qenergy = 0.0f; | | |
| 361 | | | |
| 362 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 363 | uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1]; | | |
| 364 | float *p_vec = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 365 | | | |
| 366 | abs_pow34_v(s->scoefs, in, size); | | |
| 367 | scaled = s->scoefs; | | |
| 368 | for (i = 0; i < size; i += 4) { | | |
| 369 | int curidx, curidx2; | | |
| 370 | int *in_int = (int *)&in[i]; | | |
| 371 | uint8_t v_bits; | | |
| 372 | unsigned int v_codes; | | |
| 373 | int t0, t1, t2, t3, t4, t5, t6, t7; | | |
| 374 | const float *vec1, *vec2; | | |
| 375 | | | |
| 376 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 377 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 378 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 379 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 380 | | | |
| 381 | __asm__ volatile ( | | |
| 382 | ".set push \n\t" | | |
| 383 | ".set noreorder \n\t" | | |
| 384 | | | |
| 385 | "ori %[t4], $zero, 4 \n\t" | | |
| 386 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 387 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 388 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 389 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 390 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 391 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 392 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 393 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 394 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 395 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 396 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 397 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 398 | "srl %[t0], %[t0], 31 \n\t" | | |
| 399 | "srl %[t1], %[t1], 31 \n\t" | | |
| 400 | "srl %[t2], %[t2], 31 \n\t" | | |
| 401 | "srl %[t3], %[t3], 31 \n\t" | | |
| 402 | "subu %[t4], $zero, %[qc1] \n\t" | | |
| 403 | "subu %[t5], $zero, %[qc2] \n\t" | | |
| 404 | "subu %[t6], $zero, %[qc3] \n\t" | | |
| 405 | "subu %[t7], $zero, %[qc4] \n\t" | | |
| 406 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 407 | "movn %[qc2], %[t5], %[t1] \n\t" | | |
| 408 | "movn %[qc3], %[t6], %[t2] \n\t" | | |
| 409 | "movn %[qc4], %[t7], %[t3] \n\t" | | |
| 410 | | | |
| 411 | ".set pop \n\t" | | |
| 412 | | | |
| 413 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 414 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 415 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 416 | [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7) | | |
| 417 | : [in_int]"r"(in_int) | | |
| 418 | : "memory" | | |
| 419 | ); | | |
| 420 | | | |
| 421 | curidx = 9 * qc1; | | |
| 422 | curidx += qc2 + 40; | | |
| 423 | | | |
| 424 | curidx2 = 9 * qc3; | | |
| 425 | curidx2 += qc4 + 40; | | |
| 426 | | | |
| 427 | v_codes = (p_codes[curidx] << p_bits[curidx2]) | (p_codes[curidx2]); | | |
| 428 | v_bits = p_bits[curidx] + p_bits[curidx2]; | | |
| 429 | put_bits(pb, v_bits, v_codes); | | |
| 430 | | | |
| 431 | if (out || energy) { | | |
| 432 | float e1,e2,e3,e4; | | |
| 433 | vec1 = &p_vec[curidx*2 ]; | | |
| 434 | vec2 = &p_vec[curidx2*2]; | | |
| 435 | e1 = vec1[0] * IQ; | | |
| 436 | e2 = vec1[1] * IQ; | | |
| 437 | e3 = vec2[0] * IQ; | | |
| 438 | e4 = vec2[1] * IQ; | | |
| 439 | if (out) { | | |
| 440 | out[i+0] = e1; | | |
| 441 | out[i+1] = e2; | | |
| 442 | out[i+2] = e3; | | |
| 443 | out[i+3] = e4; | | |
| 444 | } | | |
| 445 | if (energy) | | |
| 446 | qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4); | | |
| 447 | } | | |
| 448 | } | | |
| 449 | if (energy) | | |
| 450 | *energy = qenergy; | | |
| 451 | } | | |
| 452 | | | |
| 453 | static void quantize_and_encode_band_cost_UPAIR7_mips(struct AACEncContext *s, | | |
| 454 | PutBitContext *pb, const float *in, float *out, | | |
| 455 | const float *scaled, int size, int scale_idx, | | |
| 456 | int cb, const float lambda, const float uplim, | | |
| 457 | int *bits, float *energy, const float ROUNDING) | | |
| 458 | { | | |
| 459 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 460 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 461 | int i; | | |
| 462 | int qc1, qc2, qc3, qc4; | | |
| 463 | float qenergy = 0.0f; | | |
| 464 | | | |
| 465 | uint8_t *p_bits = (uint8_t*) ff_aac_spectral_bits[cb-1]; | | |
| 466 | uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1]; | | |
| 467 | float *p_vec = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 468 | | | |
| 469 | abs_pow34_v(s->scoefs, in, size); | | |
| 470 | scaled = s->scoefs; | | |
| 471 | for (i = 0; i < size; i += 4) { | | |
| 472 | int curidx1, curidx2, sign1, count1, sign2, count2; | | |
| 473 | int *in_int = (int *)&in[i]; | | |
| 474 | uint8_t v_bits; | | |
| 475 | unsigned int v_codes; | | |
| 476 | int t0, t1, t2, t3, t4; | | |
| 477 | const float *vec1, *vec2; | | |
| 478 | | | |
| 479 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 480 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 481 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 482 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 483 | | | |
| 484 | __asm__ volatile ( | | |
| 485 | ".set push \n\t" | | |
| 486 | ".set noreorder \n\t" | | |
| 487 | | | |
| 488 | "ori %[t4], $zero, 7 \n\t" | | |
| 489 | "ori %[sign1], $zero, 0 \n\t" | | |
| 490 | "ori %[sign2], $zero, 0 \n\t" | | |
| 491 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 492 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 493 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 494 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 495 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 496 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 497 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 498 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 499 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 500 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 501 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 502 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 503 | "slt %[t0], %[t0], $zero \n\t" | | |
| 504 | "movn %[sign1], %[t0], %[qc1] \n\t" | | |
| 505 | "slt %[t2], %[t2], $zero \n\t" | | |
| 506 | "movn %[sign2], %[t2], %[qc3] \n\t" | | |
| 507 | "slt %[t1], %[t1], $zero \n\t" | | |
| 508 | "sll %[t0], %[sign1], 1 \n\t" | | |
| 509 | "or %[t0], %[t0], %[t1] \n\t" | | |
| 510 | "movn %[sign1], %[t0], %[qc2] \n\t" | | |
| 511 | "slt %[t3], %[t3], $zero \n\t" | | |
| 512 | "sll %[t0], %[sign2], 1 \n\t" | | |
| 513 | "or %[t0], %[t0], %[t3] \n\t" | | |
| 514 | "movn %[sign2], %[t0], %[qc4] \n\t" | | |
| 515 | "slt %[count1], $zero, %[qc1] \n\t" | | |
| 516 | "slt %[t1], $zero, %[qc2] \n\t" | | |
| 517 | "slt %[count2], $zero, %[qc3] \n\t" | | |
| 518 | "slt %[t2], $zero, %[qc4] \n\t" | | |
| 519 | "addu %[count1], %[count1], %[t1] \n\t" | | |
| 520 | "addu %[count2], %[count2], %[t2] \n\t" | | |
| 521 | | | |
| 522 | ".set pop \n\t" | | |
| 523 | | | |
| 524 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 525 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 526 | [sign1]"=&r"(sign1), [count1]"=&r"(count1), | | |
| 527 | [sign2]"=&r"(sign2), [count2]"=&r"(count2), | | |
| 528 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 529 | [t4]"=&r"(t4) | | |
| 530 | : [in_int]"r"(in_int) | | |
| 531 | : "t0", "t1", "t2", "t3", "t4", | | |
| 532 | "memory" | | |
| 533 | ); | | |
| 534 | | | |
| 535 | curidx1 = 8 * qc1; | | |
| 536 | curidx1 += qc2; | | |
| 537 | | | |
| 538 | v_codes = (p_codes[curidx1] << count1) | sign1; | | |
| 539 | v_bits = p_bits[curidx1] + count1; | | |
| 540 | put_bits(pb, v_bits, v_codes); | | |
| 541 | | | |
| 542 | curidx2 = 8 * qc3; | | |
| 543 | curidx2 += qc4; | | |
| 544 | | | |
| 545 | v_codes = (p_codes[curidx2] << count2) | sign2; | | |
| 546 | v_bits = p_bits[curidx2] + count2; | | |
| 547 | put_bits(pb, v_bits, v_codes); | | |
| 548 | | | |
| 549 | if (out || energy) { | | |
| 550 | float e1,e2,e3,e4; | | |
| 551 | vec1 = &p_vec[curidx1*2]; | | |
| 552 | vec2 = &p_vec[curidx2*2]; | | |
| 553 | e1 = copysignf(vec1[0] * IQ, in[i+0]); | | |
| 554 | e2 = copysignf(vec1[1] * IQ, in[i+1]); | | |
| 555 | e3 = copysignf(vec2[0] * IQ, in[i+2]); | | |
| 556 | e4 = copysignf(vec2[1] * IQ, in[i+3]); | | |
| 557 | if (out) { | | |
| 558 | out[i+0] = e1; | | |
| 559 | out[i+1] = e2; | | |
| 560 | out[i+2] = e3; | | |
| 561 | out[i+3] = e4; | | |
| 562 | } | | |
| 563 | if (energy) | | |
| 564 | qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4); | | |
| 565 | } | | |
| 566 | } | | |
| 567 | if (energy) | | |
| 568 | *energy = qenergy; | | |
| 569 | } | | |
| 570 | | | |
| 571 | static void quantize_and_encode_band_cost_UPAIR12_mips(struct AACEncContext *s, | | |
| 572 | PutBitContext *pb, const float *in, float *out, | | |
| 573 | const float *scaled, int size, int scale_idx, | | |
| 574 | int cb, const float lambda, const float uplim, | | |
| 575 | int *bits, float *energy, const float ROUNDING) | | |
| 576 | { | | |
| 577 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 578 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 579 | int i; | | |
| 580 | int qc1, qc2, qc3, qc4; | | |
| 581 | float qenergy = 0.0f; | | |
| 582 | | | |
| 583 | uint8_t *p_bits = (uint8_t*) ff_aac_spectral_bits[cb-1]; | | |
| 584 | uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1]; | | |
| 585 | float *p_vec = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 586 | | | |
| 587 | abs_pow34_v(s->scoefs, in, size); | | |
| 588 | scaled = s->scoefs; | | |
| 589 | for (i = 0; i < size; i += 4) { | | |
| 590 | int curidx1, curidx2, sign1, count1, sign2, count2; | | |
| 591 | int *in_int = (int *)&in[i]; | | |
| 592 | uint8_t v_bits; | | |
| 593 | unsigned int v_codes; | | |
| 594 | int t0, t1, t2, t3, t4; | | |
| 595 | const float *vec1, *vec2; | | |
| 596 | | | |
| 597 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 598 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 599 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 600 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 601 | | | |
| 602 | __asm__ volatile ( | | |
| 603 | ".set push \n\t" | | |
| 604 | ".set noreorder \n\t" | | |
| 605 | | | |
| 606 | "ori %[t4], $zero, 12 \n\t" | | |
| 607 | "ori %[sign1], $zero, 0 \n\t" | | |
| 608 | "ori %[sign2], $zero, 0 \n\t" | | |
| 609 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 610 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 611 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 612 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 613 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 614 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 615 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 616 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 617 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 618 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 619 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 620 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 621 | "slt %[t0], %[t0], $zero \n\t" | | |
| 622 | "movn %[sign1], %[t0], %[qc1] \n\t" | | |
| 623 | "slt %[t2], %[t2], $zero \n\t" | | |
| 624 | "movn %[sign2], %[t2], %[qc3] \n\t" | | |
| 625 | "slt %[t1], %[t1], $zero \n\t" | | |
| 626 | "sll %[t0], %[sign1], 1 \n\t" | | |
| 627 | "or %[t0], %[t0], %[t1] \n\t" | | |
| 628 | "movn %[sign1], %[t0], %[qc2] \n\t" | | |
| 629 | "slt %[t3], %[t3], $zero \n\t" | | |
| 630 | "sll %[t0], %[sign2], 1 \n\t" | | |
| 631 | "or %[t0], %[t0], %[t3] \n\t" | | |
| 632 | "movn %[sign2], %[t0], %[qc4] \n\t" | | |
| 633 | "slt %[count1], $zero, %[qc1] \n\t" | | |
| 634 | "slt %[t1], $zero, %[qc2] \n\t" | | |
| 635 | "slt %[count2], $zero, %[qc3] \n\t" | | |
| 636 | "slt %[t2], $zero, %[qc4] \n\t" | | |
| 637 | "addu %[count1], %[count1], %[t1] \n\t" | | |
| 638 | "addu %[count2], %[count2], %[t2] \n\t" | | |
| 639 | | | |
| 640 | ".set pop \n\t" | | |
| 641 | | | |
| 642 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 643 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 644 | [sign1]"=&r"(sign1), [count1]"=&r"(count1), | | |
| 645 | [sign2]"=&r"(sign2), [count2]"=&r"(count2), | | |
| 646 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 647 | [t4]"=&r"(t4) | | |
| 648 | : [in_int]"r"(in_int) | | |
| 649 | : "memory" | | |
| 650 | ); | | |
| 651 | | | |
| 652 | curidx1 = 13 * qc1; | | |
| 653 | curidx1 += qc2; | | |
| 654 | | | |
| 655 | v_codes = (p_codes[curidx1] << count1) | sign1; | | |
| 656 | v_bits = p_bits[curidx1] + count1; | | |
| 657 | put_bits(pb, v_bits, v_codes); | | |
| 658 | | | |
| 659 | curidx2 = 13 * qc3; | | |
| 660 | curidx2 += qc4; | | |
| 661 | | | |
| 662 | v_codes = (p_codes[curidx2] << count2) | sign2; | | |
| 663 | v_bits = p_bits[curidx2] + count2; | | |
| 664 | put_bits(pb, v_bits, v_codes); | | |
| 665 | | | |
| 666 | if (out || energy) { | | |
| 667 | float e1,e2,e3,e4; | | |
| 668 | vec1 = &p_vec[curidx1*2]; | | |
| 669 | vec2 = &p_vec[curidx2*2]; | | |
| 670 | e1 = copysignf(vec1[0] * IQ, in[i+0]); | | |
| 671 | e2 = copysignf(vec1[1] * IQ, in[i+1]); | | |
| 672 | e3 = copysignf(vec2[0] * IQ, in[i+2]); | | |
| 673 | e4 = copysignf(vec2[1] * IQ, in[i+3]); | | |
| 674 | if (out) { | | |
| 675 | out[i+0] = e1; | | |
| 676 | out[i+1] = e2; | | |
| 677 | out[i+2] = e3; | | |
| 678 | out[i+3] = e4; | | |
| 679 | } | | |
| 680 | if (energy) | | |
| 681 | qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4); | | |
| 682 | } | | |
| 683 | } | | |
| 684 | if (energy) | | |
| 685 | *energy = qenergy; | | |
| 686 | } | | |
| 687 | | | |
| 688 | static void quantize_and_encode_band_cost_ESC_mips(struct AACEncContext *s, | | |
| 689 | PutBitContext *pb, const float *in, float *out, | | |
| 690 | const float *scaled, int size, int scale_idx, | | |
| 691 | int cb, const float lambda, const float uplim, | | |
| 692 | int *bits, float *energy, const float ROUNDING) | | |
| 693 | { | | |
| 694 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 695 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 696 | int i; | | |
| 697 | int qc1, qc2, qc3, qc4; | | |
| 698 | float qenergy = 0.0f; | | |
| 699 | | | |
| 700 | uint8_t *p_bits = (uint8_t* )ff_aac_spectral_bits[cb-1]; | | |
| 701 | uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1]; | | |
| 702 | float *p_vectors = (float* )ff_aac_codebook_vectors[cb-1]; | | |
| 703 | | | |
| 704 | abs_pow34_v(s->scoefs, in, size); | | |
| 705 | scaled = s->scoefs; | | |
| 706 | | | |
| 707 | if (cb < 11) { | | |
| 708 | for (i = 0; i < size; i += 4) { | | |
| 709 | int curidx, curidx2, sign1, count1, sign2, count2; | | |
| 710 | int *in_int = (int *)&in[i]; | | |
| 711 | uint8_t v_bits; | | |
| 712 | unsigned int v_codes; | | |
| 713 | int t0, t1, t2, t3, t4; | | |
| 714 | const float *vec1, *vec2; | | |
| 715 | | | |
| 716 | qc1 = scaled[i ] * Q34 + ROUNDING; | | |
| 717 | qc2 = scaled[i+1] * Q34 + ROUNDING; | | |
| 718 | qc3 = scaled[i+2] * Q34 + ROUNDING; | | |
| 719 | qc4 = scaled[i+3] * Q34 + ROUNDING; | | |
| 720 | | | |
| 721 | __asm__ volatile ( | | |
| 722 | ".set push \n\t" | | |
| 723 | ".set noreorder \n\t" | | |
| 724 | | | |
| 725 | "ori %[t4], $zero, 16 \n\t" | | |
| 726 | "ori %[sign1], $zero, 0 \n\t" | | |
| 727 | "ori %[sign2], $zero, 0 \n\t" | | |
| 728 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 729 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 730 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 731 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 732 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 733 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 734 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 735 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 736 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 737 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 738 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 739 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 740 | "slt %[t0], %[t0], $zero \n\t" | | |
| 741 | "movn %[sign1], %[t0], %[qc1] \n\t" | | |
| 742 | "slt %[t2], %[t2], $zero \n\t" | | |
| 743 | "movn %[sign2], %[t2], %[qc3] \n\t" | | |
| 744 | "slt %[t1], %[t1], $zero \n\t" | | |
| 745 | "sll %[t0], %[sign1], 1 \n\t" | | |
| 746 | "or %[t0], %[t0], %[t1] \n\t" | | |
| 747 | "movn %[sign1], %[t0], %[qc2] \n\t" | | |
| 748 | "slt %[t3], %[t3], $zero \n\t" | | |
| 749 | "sll %[t0], %[sign2], 1 \n\t" | | |
| 750 | "or %[t0], %[t0], %[t3] \n\t" | | |
| 751 | "movn %[sign2], %[t0], %[qc4] \n\t" | | |
| 752 | "slt %[count1], $zero, %[qc1] \n\t" | | |
| 753 | "slt %[t1], $zero, %[qc2] \n\t" | | |
| 754 | "slt %[count2], $zero, %[qc3] \n\t" | | |
| 755 | "slt %[t2], $zero, %[qc4] \n\t" | | |
| 756 | "addu %[count1], %[count1], %[t1] \n\t" | | |
| 757 | "addu %[count2], %[count2], %[t2] \n\t" | | |
| 758 | | | |
| 759 | ".set pop \n\t" | | |
| 760 | | | |
| 761 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 762 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 763 | [sign1]"=&r"(sign1), [count1]"=&r"(count1), | | |
| 764 | [sign2]"=&r"(sign2), [count2]"=&r"(count2), | | |
| 765 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 766 | [t4]"=&r"(t4) | | |
| 767 | : [in_int]"r"(in_int) | | |
| 768 | : "memory" | | |
| 769 | ); | | |
| 770 | | | |
| 771 | curidx = 17 * qc1; | | |
| 772 | curidx += qc2; | | |
| 773 | curidx2 = 17 * qc3; | | |
| 774 | curidx2 += qc4; | | |
| 775 | | | |
| 776 | v_codes = (p_codes[curidx] << count1) | sign1; | | |
| 777 | v_bits = p_bits[curidx] + count1; | | |
| 778 | put_bits(pb, v_bits, v_codes); | | |
| 779 | | | |
| 780 | v_codes = (p_codes[curidx2] << count2) | sign2; | | |
| 781 | v_bits = p_bits[curidx2] + count2; | | |
| 782 | put_bits(pb, v_bits, v_codes); | | |
| 783 | | | |
| 784 | if (out || energy) { | | |
| 785 | float e1,e2,e3,e4; | | |
| 786 | vec1 = &p_vectors[curidx*2 ]; | | |
| 787 | vec2 = &p_vectors[curidx2*2]; | | |
| 788 | e1 = copysignf(vec1[0] * IQ, in[i+0]); | | |
| 789 | e2 = copysignf(vec1[1] * IQ, in[i+1]); | | |
| 790 | e3 = copysignf(vec2[0] * IQ, in[i+2]); | | |
| 791 | e4 = copysignf(vec2[1] * IQ, in[i+3]); | | |
| 792 | if (out) { | | |
| 793 | out[i+0] = e1; | | |
| 794 | out[i+1] = e2; | | |
| 795 | out[i+2] = e3; | | |
| 796 | out[i+3] = e4; | | |
| 797 | } | | |
| 798 | if (energy) | | |
| 799 | qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4); | | |
| 800 | } | | |
| 801 | } | | |
| 802 | } else { | | |
| 803 | for (i = 0; i < size; i += 4) { | | |
| 804 | int curidx, curidx2, sign1, count1, sign2, count2; | | |
| 805 | int *in_int = (int *)&in[i]; | | |
| 806 | uint8_t v_bits; | | |
| 807 | unsigned int v_codes; | | |
| 808 | int c1, c2, c3, c4; | | |
| 809 | int t0, t1, t2, t3, t4; | | |
| 810 | | | |
| 811 | qc1 = scaled[i ] * Q34 + ROUNDING; | | |
| 812 | qc2 = scaled[i+1] * Q34 + ROUNDING; | | |
| 813 | qc3 = scaled[i+2] * Q34 + ROUNDING; | | |
| 814 | qc4 = scaled[i+3] * Q34 + ROUNDING; | | |
| 815 | | | |
| 816 | __asm__ volatile ( | | |
| 817 | ".set push \n\t" | | |
| 818 | ".set noreorder \n\t" | | |
| 819 | | | |
| 820 | "ori %[t4], $zero, 16 \n\t" | | |
| 821 | "ori %[sign1], $zero, 0 \n\t" | | |
| 822 | "ori %[sign2], $zero, 0 \n\t" | | |
| 823 | "shll_s.w %[c1], %[qc1], 18 \n\t" | | |
| 824 | "shll_s.w %[c2], %[qc2], 18 \n\t" | | |
| 825 | "shll_s.w %[c3], %[qc3], 18 \n\t" | | |
| 826 | "shll_s.w %[c4], %[qc4], 18 \n\t" | | |
| 827 | "srl %[c1], %[c1], 18 \n\t" | | |
| 828 | "srl %[c2], %[c2], 18 \n\t" | | |
| 829 | "srl %[c3], %[c3], 18 \n\t" | | |
| 830 | "srl %[c4], %[c4], 18 \n\t" | | |
| 831 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 832 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 833 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 834 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 835 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 836 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 837 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 838 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 839 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 840 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 841 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 842 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 843 | "slt %[t0], %[t0], $zero \n\t" | | |
| 844 | "movn %[sign1], %[t0], %[qc1] \n\t" | | |
| 845 | "slt %[t2], %[t2], $zero \n\t" | | |
| 846 | "movn %[sign2], %[t2], %[qc3] \n\t" | | |
| 847 | "slt %[t1], %[t1], $zero \n\t" | | |
| 848 | "sll %[t0], %[sign1], 1 \n\t" | | |
| 849 | "or %[t0], %[t0], %[t1] \n\t" | | |
| 850 | "movn %[sign1], %[t0], %[qc2] \n\t" | | |
| 851 | "slt %[t3], %[t3], $zero \n\t" | | |
| 852 | "sll %[t0], %[sign2], 1 \n\t" | | |
| 853 | "or %[t0], %[t0], %[t3] \n\t" | | |
| 854 | "movn %[sign2], %[t0], %[qc4] \n\t" | | |
| 855 | "slt %[count1], $zero, %[qc1] \n\t" | | |
| 856 | "slt %[t1], $zero, %[qc2] \n\t" | | |
| 857 | "slt %[count2], $zero, %[qc3] \n\t" | | |
| 858 | "slt %[t2], $zero, %[qc4] \n\t" | | |
| 859 | "addu %[count1], %[count1], %[t1] \n\t" | | |
| 860 | "addu %[count2], %[count2], %[t2] \n\t" | | |
| 861 | | | |
| 862 | ".set pop \n\t" | | |
| 863 | | | |
| 864 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 865 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 866 | [sign1]"=&r"(sign1), [count1]"=&r"(count1), | | |
| 867 | [sign2]"=&r"(sign2), [count2]"=&r"(count2), | | |
| 868 | [c1]"=&r"(c1), [c2]"=&r"(c2), | | |
| 869 | [c3]"=&r"(c3), [c4]"=&r"(c4), | | |
| 870 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 871 | [t4]"=&r"(t4) | | |
| 872 | : [in_int]"r"(in_int) | | |
| 873 | : "memory" | | |
| 874 | ); | | |
| 875 | | | |
| 876 | curidx = 17 * qc1; | | |
| 877 | curidx += qc2; | | |
| 878 | | | |
| 879 | curidx2 = 17 * qc3; | | |
| 880 | curidx2 += qc4; | | |
| 881 | | | |
| 882 | v_codes = (p_codes[curidx] << count1) | sign1; | | |
| 883 | v_bits = p_bits[curidx] + count1; | | |
| 884 | put_bits(pb, v_bits, v_codes); | | |
| 885 | | | |
| 886 | if (p_vectors[curidx*2 ] == 64.0f) { | | |
| 887 | int len = av_log2(c1); | | |
| 888 | v_codes = (((1 << (len - 3)) - 2) << len) | (c1 & ((1 << len) - 1)); | | |
| 889 | put_bits(pb, len * 2 - 3, v_codes); | | |
| 890 | } | | |
| 891 | if (p_vectors[curidx*2+1] == 64.0f) { | | |
| 892 | int len = av_log2(c2); | | |
| 893 | v_codes = (((1 << (len - 3)) - 2) << len) | (c2 & ((1 << len) - 1)); | | |
| 894 | put_bits(pb, len*2-3, v_codes); | | |
| 895 | } | | |
| 896 | | | |
| 897 | v_codes = (p_codes[curidx2] << count2) | sign2; | | |
| 898 | v_bits = p_bits[curidx2] + count2; | | |
| 899 | put_bits(pb, v_bits, v_codes); | | |
| 900 | | | |
| 901 | if (p_vectors[curidx2*2 ] == 64.0f) { | | |
| 902 | int len = av_log2(c3); | | |
| 903 | v_codes = (((1 << (len - 3)) - 2) << len) | (c3 & ((1 << len) - 1)); | | |
| 904 | put_bits(pb, len* 2 - 3, v_codes); | | |
| 905 | } | | |
| 906 | if (p_vectors[curidx2*2+1] == 64.0f) { | | |
| 907 | int len = av_log2(c4); | | |
| 908 | v_codes = (((1 << (len - 3)) - 2) << len) | (c4 & ((1 << len) - 1)); | | |
| 909 | put_bits(pb, len * 2 - 3, v_codes); | | |
| 910 | } | | |
| 911 | | | |
| 912 | if (out || energy) { | | |
| 913 | float e1, e2, e3, e4; | | |
| 914 | e1 = copysignf(c1 * cbrtf(c1) * IQ, in[i+0]); | | |
| 915 | e2 = copysignf(c2 * cbrtf(c2) * IQ, in[i+1]); | | |
| 916 | e3 = copysignf(c3 * cbrtf(c3) * IQ, in[i+2]); | | |
| 917 | e4 = copysignf(c4 * cbrtf(c4) * IQ, in[i+3]); | | |
| 918 | if (out) { | | |
| 919 | out[i+0] = e1; | | |
| 920 | out[i+1] = e2; | | |
| 921 | out[i+2] = e3; | | |
| 922 | out[i+3] = e4; | | |
| 923 | } | | |
| 924 | if (energy) | | |
| 925 | qenergy += (e1*e1 + e2*e2) + (e3*e3 + e4*e4); | | |
| 926 | } | | |
| 927 | } | | |
| 928 | } | | |
| 929 | if (energy) | | |
| 930 | *energy = qenergy; | | |
| 931 | } | | |
| 932 | | | |
| 933 | static void quantize_and_encode_band_cost_NONE_mips(struct AACEncContext *s, | | |
| 934 | PutBitContext *pb, const float *in, float *out, | | |
| 935 | const float *scaled, int size, int scale_idx, | | |
| 936 | int cb, const float lambda, const float uplim, | | |
| 937 | int *bits, float *energy, const float ROUNDING) { | | |
| 938 | av_assert0(0); | | |
| 939 | } | | |
| 940 | | | |
| 941 | static void quantize_and_encode_band_cost_ZERO_mips(struct AACEncContext *s, | | |
| 942 | PutBitContext *pb, const float *in, float *out, | | |
| 943 | const float *scaled, int size, int scale_idx, | | |
| 944 | int cb, const float lambda, const float uplim, | | |
| 945 | int *bits, float *energy, const float ROUNDING) { | | |
| 946 | int i; | | |
| 947 | if (bits) | | |
| 948 | *bits = 0; | | |
| 949 | if (out) { | | |
| 950 | for (i = 0; i < size; i += 4) { | | |
| 951 | out[i ] = 0.0f; | | |
| 952 | out[i+1] = 0.0f; | | |
| 953 | out[i+2] = 0.0f; | | |
| 954 | out[i+3] = 0.0f; | | |
| 955 | } | | |
| 956 | } | | |
| 957 | if (energy) | | |
| 958 | *energy = 0.0f; | | |
| 959 | } | | |
| 960 | | | |
| 961 | static void (*const quantize_and_encode_band_cost_arr[])(struct AACEncContext *s, | | |
| 962 | PutBitContext *pb, const float *in, float *out, | | |
| 963 | const float *scaled, int size, int scale_idx, | | |
| 964 | int cb, const float lambda, const float uplim, | | |
| 965 | int *bits, float *energy, const float ROUNDING) = { | | |
| 966 | quantize_and_encode_band_cost_ZERO_mips, | | |
| 967 | quantize_and_encode_band_cost_SQUAD_mips, | | |
| 968 | quantize_and_encode_band_cost_SQUAD_mips, | | |
| 969 | quantize_and_encode_band_cost_UQUAD_mips, | | |
| 970 | quantize_and_encode_band_cost_UQUAD_mips, | | |
| 971 | quantize_and_encode_band_cost_SPAIR_mips, | | |
| 972 | quantize_and_encode_band_cost_SPAIR_mips, | | |
| 973 | quantize_and_encode_band_cost_UPAIR7_mips, | | |
| 974 | quantize_and_encode_band_cost_UPAIR7_mips, | | |
| 975 | quantize_and_encode_band_cost_UPAIR12_mips, | | |
| 976 | quantize_and_encode_band_cost_UPAIR12_mips, | | |
| 977 | quantize_and_encode_band_cost_ESC_mips, | | |
| 978 | quantize_and_encode_band_cost_NONE_mips, /* cb 12 doesn't exist */ | | |
| 979 | quantize_and_encode_band_cost_ZERO_mips, | | |
| 980 | quantize_and_encode_band_cost_ZERO_mips, | | |
| 981 | quantize_and_encode_band_cost_ZERO_mips, | | |
| 982 | }; | | |
| 983 | | | |
| 984 | #define quantize_and_encode_band_cost( \ | | |
| 985 | s, pb, in, out, scaled, size, scale_idx, cb, \ | | |
| 986 | lambda, uplim, bits, energy, ROUNDING) \ | | |
| 987 | quantize_and_encode_band_cost_arr[cb]( \ | | |
| 988 | s, pb, in, out, scaled, size, scale_idx, cb, \ | | |
| 989 | lambda, uplim, bits, energy, ROUNDING) | | |
| 990 | | | |
| 991 | static void quantize_and_encode_band_mips(struct AACEncContext *s, PutBitContext *pb, | | |
| 992 | const float *in, float *out, int size, int scale_idx, | | |
| 993 | int cb, const float lambda, int rtz) | | |
| 994 | { | | |
| 995 | quantize_and_encode_band_cost(s, pb, in, out, NULL, size, scale_idx, cb, lambda, | | |
| 996 | INFINITY, NULL, NULL, (rtz) ? ROUND_TO_ZERO : ROUND_STANDARD); | | |
| 997 | } | | |
| 998 | | | |
| 999 | /** | | |
| 1000 | * Functions developed from template function and optimized for getting the number of bits | | |
| 1001 | */ | | |
| 1002 | static float get_band_numbits_ZERO_mips(struct AACEncContext *s, | | |
| 1003 | PutBitContext *pb, const float *in, | | |
| 1004 | const float *scaled, int size, int scale_idx, | | |
| 1005 | int cb, const float lambda, const float uplim, | | |
| 1006 | int *bits) | | |
| 1007 | { | | |
| 1008 | return 0; | | |
| 1009 | } | | |
| 1010 | | | |
| 1011 | static float get_band_numbits_NONE_mips(struct AACEncContext *s, | | |
| 1012 | PutBitContext *pb, const float *in, | | |
| 1013 | const float *scaled, int size, int scale_idx, | | |
| 1014 | int cb, const float lambda, const float uplim, | | |
| 1015 | int *bits) | | |
| 1016 | { | | |
| 1017 | av_assert0(0); | | |
| 1018 | return 0; | | |
| 1019 | } | | |
| 1020 | | | |
| 1021 | static float get_band_numbits_SQUAD_mips(struct AACEncContext *s, | | |
| 1022 | PutBitContext *pb, const float *in, | | |
| 1023 | const float *scaled, int size, int scale_idx, | | |
| 1024 | int cb, const float lambda, const float uplim, | | |
| 1025 | int *bits) | | |
| 1026 | { | | |
| 1027 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1028 | int i; | | |
| 1029 | int qc1, qc2, qc3, qc4; | | |
| 1030 | int curbits = 0; | | |
| 1031 | | | |
| 1032 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 1033 | | | |
| 1034 | for (i = 0; i < size; i += 4) { | | |
| 1035 | int curidx; | | |
| 1036 | int *in_int = (int *)&in[i]; | | |
| 1037 | int t0, t1, t2, t3, t4, t5, t6, t7; | | |
| 1038 | | | |
| 1039 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1040 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1041 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1042 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1043 | | | |
| 1044 | __asm__ volatile ( | | |
| 1045 | ".set push \n\t" | | |
| 1046 | ".set noreorder \n\t" | | |
| 1047 | | | |
| 1048 | "slt %[qc1], $zero, %[qc1] \n\t" | | |
| 1049 | "slt %[qc2], $zero, %[qc2] \n\t" | | |
| 1050 | "slt %[qc3], $zero, %[qc3] \n\t" | | |
| 1051 | "slt %[qc4], $zero, %[qc4] \n\t" | | |
| 1052 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 1053 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 1054 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 1055 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 1056 | "srl %[t0], %[t0], 31 \n\t" | | |
| 1057 | "srl %[t1], %[t1], 31 \n\t" | | |
| 1058 | "srl %[t2], %[t2], 31 \n\t" | | |
| 1059 | "srl %[t3], %[t3], 31 \n\t" | | |
| 1060 | "subu %[t4], $zero, %[qc1] \n\t" | | |
| 1061 | "subu %[t5], $zero, %[qc2] \n\t" | | |
| 1062 | "subu %[t6], $zero, %[qc3] \n\t" | | |
| 1063 | "subu %[t7], $zero, %[qc4] \n\t" | | |
| 1064 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1065 | "movn %[qc2], %[t5], %[t1] \n\t" | | |
| 1066 | "movn %[qc3], %[t6], %[t2] \n\t" | | |
| 1067 | "movn %[qc4], %[t7], %[t3] \n\t" | | |
| 1068 | | | |
| 1069 | ".set pop \n\t" | | |
| 1070 | | | |
| 1071 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1072 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1073 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1074 | [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7) | | |
| 1075 | : [in_int]"r"(in_int) | | |
| 1076 | : "memory" | | |
| 1077 | ); | | |
| 1078 | | | |
| 1079 | curidx = qc1; | | |
| 1080 | curidx *= 3; | | |
| 1081 | curidx += qc2; | | |
| 1082 | curidx *= 3; | | |
| 1083 | curidx += qc3; | | |
| 1084 | curidx *= 3; | | |
| 1085 | curidx += qc4; | | |
| 1086 | curidx += 40; | | |
| 1087 | | | |
| 1088 | curbits += p_bits[curidx]; | | |
| 1089 | } | | |
| 1090 | return curbits; | | |
| 1091 | } | | |
| 1092 | | | |
| 1093 | static float get_band_numbits_UQUAD_mips(struct AACEncContext *s, | | |
| 1094 | PutBitContext *pb, const float *in, | | |
| 1095 | const float *scaled, int size, int scale_idx, | | |
| 1096 | int cb, const float lambda, const float uplim, | | |
| 1097 | int *bits) | | |
| 1098 | { | | |
| 1099 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1100 | int i; | | |
| 1101 | int curbits = 0; | | |
| 1102 | int qc1, qc2, qc3, qc4; | | |
| 1103 | | | |
| 1104 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 1105 | | | |
| 1106 | for (i = 0; i < size; i += 4) { | | |
| 1107 | int curidx; | | |
| 1108 | int t0, t1, t2, t3, t4; | | |
| 1109 | | | |
| 1110 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1111 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1112 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1113 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1114 | | | |
| 1115 | __asm__ volatile ( | | |
| 1116 | ".set push \n\t" | | |
| 1117 | ".set noreorder \n\t" | | |
| 1118 | | | |
| 1119 | "ori %[t4], $zero, 2 \n\t" | | |
| 1120 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 1121 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 1122 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 1123 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 1124 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1125 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 1126 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 1127 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 1128 | | | |
| 1129 | ".set pop \n\t" | | |
| 1130 | | | |
| 1131 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1132 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1133 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1134 | [t4]"=&r"(t4) | | |
| 1135 | ); | | |
| 1136 | | | |
| 1137 | curidx = qc1; | | |
| 1138 | curidx *= 3; | | |
| 1139 | curidx += qc2; | | |
| 1140 | curidx *= 3; | | |
| 1141 | curidx += qc3; | | |
| 1142 | curidx *= 3; | | |
| 1143 | curidx += qc4; | | |
| 1144 | | | |
| 1145 | curbits += p_bits[curidx]; | | |
| 1146 | curbits += uquad_sign_bits[curidx]; | | |
| 1147 | } | | |
| 1148 | return curbits; | | |
| 1149 | } | | |
| 1150 | | | |
| 1151 | static float get_band_numbits_SPAIR_mips(struct AACEncContext *s, | | |
| 1152 | PutBitContext *pb, const float *in, | | |
| 1153 | const float *scaled, int size, int scale_idx, | | |
| 1154 | int cb, const float lambda, const float uplim, | | |
| 1155 | int *bits) | | |
| 1156 | { | | |
| 1157 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1158 | int i; | | |
| 1159 | int qc1, qc2, qc3, qc4; | | |
| 1160 | int curbits = 0; | | |
| 1161 | | | |
| 1162 | uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1]; | | |
| 1163 | | | |
| 1164 | for (i = 0; i < size; i += 4) { | | |
| 1165 | int curidx, curidx2; | | |
| 1166 | int *in_int = (int *)&in[i]; | | |
| 1167 | int t0, t1, t2, t3, t4, t5, t6, t7; | | |
| 1168 | | | |
| 1169 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1170 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1171 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1172 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1173 | | | |
| 1174 | __asm__ volatile ( | | |
| 1175 | ".set push \n\t" | | |
| 1176 | ".set noreorder \n\t" | | |
| 1177 | | | |
| 1178 | "ori %[t4], $zero, 4 \n\t" | | |
| 1179 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 1180 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 1181 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 1182 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 1183 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1184 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 1185 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 1186 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 1187 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 1188 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 1189 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 1190 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 1191 | "srl %[t0], %[t0], 31 \n\t" | | |
| 1192 | "srl %[t1], %[t1], 31 \n\t" | | |
| 1193 | "srl %[t2], %[t2], 31 \n\t" | | |
| 1194 | "srl %[t3], %[t3], 31 \n\t" | | |
| 1195 | "subu %[t4], $zero, %[qc1] \n\t" | | |
| 1196 | "subu %[t5], $zero, %[qc2] \n\t" | | |
| 1197 | "subu %[t6], $zero, %[qc3] \n\t" | | |
| 1198 | "subu %[t7], $zero, %[qc4] \n\t" | | |
| 1199 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1200 | "movn %[qc2], %[t5], %[t1] \n\t" | | |
| 1201 | "movn %[qc3], %[t6], %[t2] \n\t" | | |
| 1202 | "movn %[qc4], %[t7], %[t3] \n\t" | | |
| 1203 | | | |
| 1204 | ".set pop \n\t" | | |
| 1205 | | | |
| 1206 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1207 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1208 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1209 | [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7) | | |
| 1210 | : [in_int]"r"(in_int) | | |
| 1211 | : "memory" | | |
| 1212 | ); | | |
| 1213 | | | |
| 1214 | curidx = 9 * qc1; | | |
| 1215 | curidx += qc2 + 40; | | |
| 1216 | | | |
| 1217 | curidx2 = 9 * qc3; | | |
| 1218 | curidx2 += qc4 + 40; | | |
| 1219 | | | |
| 1220 | curbits += p_bits[curidx] + p_bits[curidx2]; | | |
| 1221 | } | | |
| 1222 | return curbits; | | |
| 1223 | } | | |
| 1224 | | | |
| 1225 | static float get_band_numbits_UPAIR7_mips(struct AACEncContext *s, | | |
| 1226 | PutBitContext *pb, const float *in, | | |
| 1227 | const float *scaled, int size, int scale_idx, | | |
| 1228 | int cb, const float lambda, const float uplim, | | |
| 1229 | int *bits) | | |
| 1230 | { | | |
| 1231 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1232 | int i; | | |
| 1233 | int qc1, qc2, qc3, qc4; | | |
| 1234 | int curbits = 0; | | |
| 1235 | | | |
| 1236 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 1237 | | | |
| 1238 | for (i = 0; i < size; i += 4) { | | |
| 1239 | int curidx, curidx2; | | |
| 1240 | int t0, t1, t2, t3, t4; | | |
| 1241 | | | |
| 1242 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1243 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1244 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1245 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1246 | | | |
| 1247 | __asm__ volatile ( | | |
| 1248 | ".set push \n\t" | | |
| 1249 | ".set noreorder \n\t" | | |
| 1250 | | | |
| 1251 | "ori %[t4], $zero, 7 \n\t" | | |
| 1252 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 1253 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 1254 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 1255 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 1256 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1257 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 1258 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 1259 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 1260 | | | |
| 1261 | ".set pop \n\t" | | |
| 1262 | | | |
| 1263 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1264 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1265 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1266 | [t4]"=&r"(t4) | | |
| 1267 | ); | | |
| 1268 | | | |
| 1269 | curidx = 8 * qc1; | | |
| 1270 | curidx += qc2; | | |
| 1271 | | | |
| 1272 | curidx2 = 8 * qc3; | | |
| 1273 | curidx2 += qc4; | | |
| 1274 | | | |
| 1275 | curbits += p_bits[curidx] + | | |
| 1276 | upair7_sign_bits[curidx] + | | |
| 1277 | p_bits[curidx2] + | | |
| 1278 | upair7_sign_bits[curidx2]; | | |
| 1279 | } | | |
| 1280 | return curbits; | | |
| 1281 | } | | |
| 1282 | | | |
| 1283 | static float get_band_numbits_UPAIR12_mips(struct AACEncContext *s, | | |
| 1284 | PutBitContext *pb, const float *in, | | |
| 1285 | const float *scaled, int size, int scale_idx, | | |
| 1286 | int cb, const float lambda, const float uplim, | | |
| 1287 | int *bits) | | |
| 1288 | { | | |
| 1289 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1290 | int i; | | |
| 1291 | int qc1, qc2, qc3, qc4; | | |
| 1292 | int curbits = 0; | | |
| 1293 | | | |
| 1294 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 1295 | | | |
| 1296 | for (i = 0; i < size; i += 4) { | | |
| 1297 | int curidx, curidx2; | | |
| 1298 | int t0, t1, t2, t3, t4; | | |
| 1299 | | | |
| 1300 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1301 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1302 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1303 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1304 | | | |
| 1305 | __asm__ volatile ( | | |
| 1306 | ".set push \n\t" | | |
| 1307 | ".set noreorder \n\t" | | |
| 1308 | | | |
| 1309 | "ori %[t4], $zero, 12 \n\t" | | |
| 1310 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 1311 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 1312 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 1313 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 1314 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1315 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 1316 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 1317 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 1318 | | | |
| 1319 | ".set pop \n\t" | | |
| 1320 | | | |
| 1321 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1322 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1323 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1324 | [t4]"=&r"(t4) | | |
| 1325 | ); | | |
| 1326 | | | |
| 1327 | curidx = 13 * qc1; | | |
| 1328 | curidx += qc2; | | |
| 1329 | | | |
| 1330 | curidx2 = 13 * qc3; | | |
| 1331 | curidx2 += qc4; | | |
| 1332 | | | |
| 1333 | curbits += p_bits[curidx] + | | |
| 1334 | p_bits[curidx2] + | | |
| 1335 | upair12_sign_bits[curidx] + | | |
| 1336 | upair12_sign_bits[curidx2]; | | |
| 1337 | } | | |
| 1338 | return curbits; | | |
| 1339 | } | | |
| 1340 | | | |
| 1341 | static float get_band_numbits_ESC_mips(struct AACEncContext *s, | | |
| 1342 | PutBitContext *pb, const float *in, | | |
| 1343 | const float *scaled, int size, int scale_idx, | | |
| 1344 | int cb, const float lambda, const float uplim, | | |
| 1345 | int *bits) | | |
| 1346 | { | | |
| 1347 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1348 | int i; | | |
| 1349 | int qc1, qc2, qc3, qc4; | | |
| 1350 | int curbits = 0; | | |
| 1351 | | | |
| 1352 | uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1]; | | |
| 1353 | | | |
| 1354 | for (i = 0; i < size; i += 4) { | | |
| 1355 | int curidx, curidx2; | | |
| 1356 | int cond0, cond1, cond2, cond3; | | |
| 1357 | int c1, c2, c3, c4; | | |
| 1358 | int t4, t5; | | |
| 1359 | | | |
| 1360 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1361 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1362 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1363 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1364 | | | |
| 1365 | __asm__ volatile ( | | |
| 1366 | ".set push \n\t" | | |
| 1367 | ".set noreorder \n\t" | | |
| 1368 | | | |
| 1369 | "ori %[t4], $zero, 15 \n\t" | | |
| 1370 | "ori %[t5], $zero, 16 \n\t" | | |
| 1371 | "shll_s.w %[c1], %[qc1], 18 \n\t" | | |
| 1372 | "shll_s.w %[c2], %[qc2], 18 \n\t" | | |
| 1373 | "shll_s.w %[c3], %[qc3], 18 \n\t" | | |
| 1374 | "shll_s.w %[c4], %[qc4], 18 \n\t" | | |
| 1375 | "srl %[c1], %[c1], 18 \n\t" | | |
| 1376 | "srl %[c2], %[c2], 18 \n\t" | | |
| 1377 | "srl %[c3], %[c3], 18 \n\t" | | |
| 1378 | "srl %[c4], %[c4], 18 \n\t" | | |
| 1379 | "slt %[cond0], %[t4], %[qc1] \n\t" | | |
| 1380 | "slt %[cond1], %[t4], %[qc2] \n\t" | | |
| 1381 | "slt %[cond2], %[t4], %[qc3] \n\t" | | |
| 1382 | "slt %[cond3], %[t4], %[qc4] \n\t" | | |
| 1383 | "movn %[qc1], %[t5], %[cond0] \n\t" | | |
| 1384 | "movn %[qc2], %[t5], %[cond1] \n\t" | | |
| 1385 | "movn %[qc3], %[t5], %[cond2] \n\t" | | |
| 1386 | "movn %[qc4], %[t5], %[cond3] \n\t" | | |
| 1387 | "ori %[t5], $zero, 31 \n\t" | | |
| 1388 | "clz %[c1], %[c1] \n\t" | | |
| 1389 | "clz %[c2], %[c2] \n\t" | | |
| 1390 | "clz %[c3], %[c3] \n\t" | | |
| 1391 | "clz %[c4], %[c4] \n\t" | | |
| 1392 | "subu %[c1], %[t5], %[c1] \n\t" | | |
| 1393 | "subu %[c2], %[t5], %[c2] \n\t" | | |
| 1394 | "subu %[c3], %[t5], %[c3] \n\t" | | |
| 1395 | "subu %[c4], %[t5], %[c4] \n\t" | | |
| 1396 | "sll %[c1], %[c1], 1 \n\t" | | |
| 1397 | "sll %[c2], %[c2], 1 \n\t" | | |
| 1398 | "sll %[c3], %[c3], 1 \n\t" | | |
| 1399 | "sll %[c4], %[c4], 1 \n\t" | | |
| 1400 | "addiu %[c1], %[c1], -3 \n\t" | | |
| 1401 | "addiu %[c2], %[c2], -3 \n\t" | | |
| 1402 | "addiu %[c3], %[c3], -3 \n\t" | | |
| 1403 | "addiu %[c4], %[c4], -3 \n\t" | | |
| 1404 | "subu %[cond0], $zero, %[cond0] \n\t" | | |
| 1405 | "subu %[cond1], $zero, %[cond1] \n\t" | | |
| 1406 | "subu %[cond2], $zero, %[cond2] \n\t" | | |
| 1407 | "subu %[cond3], $zero, %[cond3] \n\t" | | |
| 1408 | "and %[c1], %[c1], %[cond0] \n\t" | | |
| 1409 | "and %[c2], %[c2], %[cond1] \n\t" | | |
| 1410 | "and %[c3], %[c3], %[cond2] \n\t" | | |
| 1411 | "and %[c4], %[c4], %[cond3] \n\t" | | |
| 1412 | | | |
| 1413 | ".set pop \n\t" | | |
| 1414 | | | |
| 1415 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1416 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1417 | [cond0]"=&r"(cond0), [cond1]"=&r"(cond1), | | |
| 1418 | [cond2]"=&r"(cond2), [cond3]"=&r"(cond3), | | |
| 1419 | [c1]"=&r"(c1), [c2]"=&r"(c2), | | |
| 1420 | [c3]"=&r"(c3), [c4]"=&r"(c4), | | |
| 1421 | [t4]"=&r"(t4), [t5]"=&r"(t5) | | |
| 1422 | ); | | |
| 1423 | | | |
| 1424 | curidx = 17 * qc1; | | |
| 1425 | curidx += qc2; | | |
| 1426 | | | |
| 1427 | curidx2 = 17 * qc3; | | |
| 1428 | curidx2 += qc4; | | |
| 1429 | | | |
| 1430 | curbits += p_bits[curidx]; | | |
| 1431 | curbits += esc_sign_bits[curidx]; | | |
| 1432 | curbits += p_bits[curidx2]; | | |
| 1433 | curbits += esc_sign_bits[curidx2]; | | |
| 1434 | | | |
| 1435 | curbits += c1; | | |
| 1436 | curbits += c2; | | |
| 1437 | curbits += c3; | | |
| 1438 | curbits += c4; | | |
| 1439 | } | | |
| 1440 | return curbits; | | |
| 1441 | } | | |
| 1442 | | | |
| 1443 | static float (*const get_band_numbits_arr[])(struct AACEncContext *s, | | |
| 1444 | PutBitContext *pb, const float *in, | | |
| 1445 | const float *scaled, int size, int scale_idx, | | |
| 1446 | int cb, const float lambda, const float uplim, | | |
| 1447 | int *bits) = { | | |
| 1448 | get_band_numbits_ZERO_mips, | | |
| 1449 | get_band_numbits_SQUAD_mips, | | |
| 1450 | get_band_numbits_SQUAD_mips, | | |
| 1451 | get_band_numbits_UQUAD_mips, | | |
| 1452 | get_band_numbits_UQUAD_mips, | | |
| 1453 | get_band_numbits_SPAIR_mips, | | |
| 1454 | get_band_numbits_SPAIR_mips, | | |
| 1455 | get_band_numbits_UPAIR7_mips, | | |
| 1456 | get_band_numbits_UPAIR7_mips, | | |
| 1457 | get_band_numbits_UPAIR12_mips, | | |
| 1458 | get_band_numbits_UPAIR12_mips, | | |
| 1459 | get_band_numbits_ESC_mips, | | |
| 1460 | get_band_numbits_NONE_mips, /* cb 12 doesn't exist */ | | |
| 1461 | get_band_numbits_ZERO_mips, | | |
| 1462 | get_band_numbits_ZERO_mips, | | |
| 1463 | get_band_numbits_ZERO_mips, | | |
| 1464 | }; | | |
| 1465 | | | |
| 1466 | #define get_band_numbits( \ | | |
| 1467 | s, pb, in, scaled, size, scale_idx, cb, \ | | |
| 1468 | lambda, uplim, bits) \ | | |
| 1469 | get_band_numbits_arr[cb]( \ | | |
| 1470 | s, pb, in, scaled, size, scale_idx, cb, \ | | |
| 1471 | lambda, uplim, bits) | | |
| 1472 | | | |
| 1473 | static float quantize_band_cost_bits(struct AACEncContext *s, const float *in, | | |
| 1474 | const float *scaled, int size, int scale_idx, | | |
| 1475 | int cb, const float lambda, const float uplim, | | |
| 1476 | int *bits, float *energy) | | |
| 1477 | { | | |
| 1478 | return get_band_numbits(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits); | | |
| 1479 | } | | |
| 1480 | | | |
| 1481 | /** | | |
| 1482 | * Functions developed from template function and optimized for getting the band cost | | |
| 1483 | */ | | |
| 1484 | #if HAVE_MIPSFPU | | |
| 1485 | static float get_band_cost_ZERO_mips(struct AACEncContext *s, | | |
| 1486 | PutBitContext *pb, const float *in, | | |
| 1487 | const float *scaled, int size, int scale_idx, | | |
| 1488 | int cb, const float lambda, const float uplim, | | |
| 1489 | int *bits, float *energy) | | |
| 1490 | { | | |
| 1491 | int i; | | |
| 1492 | float cost = 0; | | |
| 1493 | | | |
| 1494 | for (i = 0; i < size; i += 4) { | | |
| 1495 | cost += in[i ] * in[i ]; | | |
| 1496 | cost += in[i+1] * in[i+1]; | | |
| 1497 | cost += in[i+2] * in[i+2]; | | |
| 1498 | cost += in[i+3] * in[i+3]; | | |
| 1499 | } | | |
| 1500 | if (bits) | | |
| 1501 | *bits = 0; | | |
| 1502 | if (energy) | | |
| 1503 | *energy = 0.0f; | | |
| 1504 | return cost * lambda; | | |
| 1505 | } | | |
| 1506 | | | |
| 1507 | static float get_band_cost_NONE_mips(struct AACEncContext *s, | | |
| 1508 | PutBitContext *pb, const float *in, | | |
| 1509 | const float *scaled, int size, int scale_idx, | | |
| 1510 | int cb, const float lambda, const float uplim, | | |
| 1511 | int *bits, float *energy) | | |
| 1512 | { | | |
| 1513 | av_assert0(0); | | |
| 1514 | return 0; | | |
| 1515 | } | | |
| 1516 | | | |
| 1517 | static float get_band_cost_SQUAD_mips(struct AACEncContext *s, | | |
| 1518 | PutBitContext *pb, const float *in, | | |
| 1519 | const float *scaled, int size, int scale_idx, | | |
| 1520 | int cb, const float lambda, const float uplim, | | |
| 1521 | int *bits, float *energy) | | |
| 1522 | { | | |
| 1523 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1524 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 1525 | int i; | | |
| 1526 | float cost = 0; | | |
| 1527 | float qenergy = 0.0f; | | |
| 1528 | int qc1, qc2, qc3, qc4; | | |
| 1529 | int curbits = 0; | | |
| 1530 | | | |
| 1531 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 1532 | float *p_codes = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 1533 | | | |
| 1534 | for (i = 0; i < size; i += 4) { | | |
| 1535 | const float *vec; | | |
| 1536 | int curidx; | | |
| 1537 | int *in_int = (int *)&in[i]; | | |
| 1538 | float *in_pos = (float *)&in[i]; | | |
| 1539 | float di0, di1, di2, di3; | | |
| 1540 | int t0, t1, t2, t3, t4, t5, t6, t7; | | |
| 1541 | | | |
| 1542 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1543 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1544 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1545 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1546 | | | |
| 1547 | __asm__ volatile ( | | |
| 1548 | ".set push \n\t" | | |
| 1549 | ".set noreorder \n\t" | | |
| 1550 | | | |
| 1551 | "slt %[qc1], $zero, %[qc1] \n\t" | | |
| 1552 | "slt %[qc2], $zero, %[qc2] \n\t" | | |
| 1553 | "slt %[qc3], $zero, %[qc3] \n\t" | | |
| 1554 | "slt %[qc4], $zero, %[qc4] \n\t" | | |
| 1555 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 1556 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 1557 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 1558 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 1559 | "srl %[t0], %[t0], 31 \n\t" | | |
| 1560 | "srl %[t1], %[t1], 31 \n\t" | | |
| 1561 | "srl %[t2], %[t2], 31 \n\t" | | |
| 1562 | "srl %[t3], %[t3], 31 \n\t" | | |
| 1563 | "subu %[t4], $zero, %[qc1] \n\t" | | |
| 1564 | "subu %[t5], $zero, %[qc2] \n\t" | | |
| 1565 | "subu %[t6], $zero, %[qc3] \n\t" | | |
| 1566 | "subu %[t7], $zero, %[qc4] \n\t" | | |
| 1567 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1568 | "movn %[qc2], %[t5], %[t1] \n\t" | | |
| 1569 | "movn %[qc3], %[t6], %[t2] \n\t" | | |
| 1570 | "movn %[qc4], %[t7], %[t3] \n\t" | | |
| 1571 | | | |
| 1572 | ".set pop \n\t" | | |
| 1573 | | | |
| 1574 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1575 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1576 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1577 | [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7) | | |
| 1578 | : [in_int]"r"(in_int) | | |
| 1579 | : "memory" | | |
| 1580 | ); | | |
| 1581 | | | |
| 1582 | curidx = qc1; | | |
| 1583 | curidx *= 3; | | |
| 1584 | curidx += qc2; | | |
| 1585 | curidx *= 3; | | |
| 1586 | curidx += qc3; | | |
| 1587 | curidx *= 3; | | |
| 1588 | curidx += qc4; | | |
| 1589 | curidx += 40; | | |
| 1590 | | | |
| 1591 | curbits += p_bits[curidx]; | | |
| 1592 | vec = &p_codes[curidx*4]; | | |
| 1593 | | | |
| 1594 | qenergy += vec[0]*vec[0] + vec[1]*vec[1] | | |
| 1595 | + vec[2]*vec[2] + vec[3]*vec[3]; | | |
| 1596 | | | |
| 1597 | __asm__ volatile ( | | |
| 1598 | ".set push \n\t" | | |
| 1599 | ".set noreorder \n\t" | | |
| 1600 | | | |
| 1601 | "lwc1 $f0, 0(%[in_pos]) \n\t" | | |
| 1602 | "lwc1 $f1, 0(%[vec]) \n\t" | | |
| 1603 | "lwc1 $f2, 4(%[in_pos]) \n\t" | | |
| 1604 | "lwc1 $f3, 4(%[vec]) \n\t" | | |
| 1605 | "lwc1 $f4, 8(%[in_pos]) \n\t" | | |
| 1606 | "lwc1 $f5, 8(%[vec]) \n\t" | | |
| 1607 | "lwc1 $f6, 12(%[in_pos]) \n\t" | | |
| 1608 | "lwc1 $f7, 12(%[vec]) \n\t" | | |
| 1609 | "nmsub.s %[di0], $f0, $f1, %[IQ] \n\t" | | |
| 1610 | "nmsub.s %[di1], $f2, $f3, %[IQ] \n\t" | | |
| 1611 | "nmsub.s %[di2], $f4, $f5, %[IQ] \n\t" | | |
| 1612 | "nmsub.s %[di3], $f6, $f7, %[IQ] \n\t" | | |
| 1613 | | | |
| 1614 | ".set pop \n\t" | | |
| 1615 | | | |
| 1616 | : [di0]"=&f"(di0), [di1]"=&f"(di1), | | |
| 1617 | [di2]"=&f"(di2), [di3]"=&f"(di3) | | |
| 1618 | : [in_pos]"r"(in_pos), [vec]"r"(vec), | | |
| 1619 | [IQ]"f"(IQ) | | |
| 1620 | : "$f0", "$f1", "$f2", "$f3", | | |
| 1621 | "$f4", "$f5", "$f6", "$f7", | | |
| 1622 | "memory" | | |
| 1623 | ); | | |
| 1624 | | | |
| 1625 | cost += di0 * di0 + di1 * di1 | | |
| 1626 | + di2 * di2 + di3 * di3; | | |
| 1627 | } | | |
| 1628 | | | |
| 1629 | if (bits) | | |
| 1630 | *bits = curbits; | | |
| 1631 | if (energy) | | |
| 1632 | *energy = qenergy * (IQ*IQ); | | |
| 1633 | return cost * lambda + curbits; | | |
| 1634 | } | | |
| 1635 | | | |
| 1636 | static float get_band_cost_UQUAD_mips(struct AACEncContext *s, | | |
| 1637 | PutBitContext *pb, const float *in, | | |
| 1638 | const float *scaled, int size, int scale_idx, | | |
| 1639 | int cb, const float lambda, const float uplim, | | |
| 1640 | int *bits, float *energy) | | |
| 1641 | { | | |
| 1642 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1643 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 1644 | int i; | | |
| 1645 | float cost = 0; | | |
| 1646 | float qenergy = 0.0f; | | |
| 1647 | int curbits = 0; | | |
| 1648 | int qc1, qc2, qc3, qc4; | | |
| 1649 | | | |
| 1650 | uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1]; | | |
| 1651 | float *p_codes = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 1652 | | | |
| 1653 | for (i = 0; i < size; i += 4) { | | |
| 1654 | const float *vec; | | |
| 1655 | int curidx; | | |
| 1656 | float *in_pos = (float *)&in[i]; | | |
| 1657 | float di0, di1, di2, di3; | | |
| 1658 | int t0, t1, t2, t3, t4; | | |
| 1659 | | | |
| 1660 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1661 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1662 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1663 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1664 | | | |
| 1665 | __asm__ volatile ( | | |
| 1666 | ".set push \n\t" | | |
| 1667 | ".set noreorder \n\t" | | |
| 1668 | | | |
| 1669 | "ori %[t4], $zero, 2 \n\t" | | |
| 1670 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 1671 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 1672 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 1673 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 1674 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1675 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 1676 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 1677 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 1678 | | | |
| 1679 | ".set pop \n\t" | | |
| 1680 | | | |
| 1681 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1682 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1683 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1684 | [t4]"=&r"(t4) | | |
| 1685 | ); | | |
| 1686 | | | |
| 1687 | curidx = qc1; | | |
| 1688 | curidx *= 3; | | |
| 1689 | curidx += qc2; | | |
| 1690 | curidx *= 3; | | |
| 1691 | curidx += qc3; | | |
| 1692 | curidx *= 3; | | |
| 1693 | curidx += qc4; | | |
| 1694 | | | |
| 1695 | curbits += p_bits[curidx]; | | |
| 1696 | curbits += uquad_sign_bits[curidx]; | | |
| 1697 | vec = &p_codes[curidx*4]; | | |
| 1698 | | | |
| 1699 | qenergy += vec[0]*vec[0] + vec[1]*vec[1] | | |
| 1700 | + vec[2]*vec[2] + vec[3]*vec[3]; | | |
| 1701 | | | |
| 1702 | __asm__ volatile ( | | |
| 1703 | ".set push \n\t" | | |
| 1704 | ".set noreorder \n\t" | | |
| 1705 | | | |
| 1706 | "lwc1 %[di0], 0(%[in_pos]) \n\t" | | |
| 1707 | "lwc1 %[di1], 4(%[in_pos]) \n\t" | | |
| 1708 | "lwc1 %[di2], 8(%[in_pos]) \n\t" | | |
| 1709 | "lwc1 %[di3], 12(%[in_pos]) \n\t" | | |
| 1710 | "abs.s %[di0], %[di0] \n\t" | | |
| 1711 | "abs.s %[di1], %[di1] \n\t" | | |
| 1712 | "abs.s %[di2], %[di2] \n\t" | | |
| 1713 | "abs.s %[di3], %[di3] \n\t" | | |
| 1714 | "lwc1 $f0, 0(%[vec]) \n\t" | | |
| 1715 | "lwc1 $f1, 4(%[vec]) \n\t" | | |
| 1716 | "lwc1 $f2, 8(%[vec]) \n\t" | | |
| 1717 | "lwc1 $f3, 12(%[vec]) \n\t" | | |
| 1718 | "nmsub.s %[di0], %[di0], $f0, %[IQ] \n\t" | | |
| 1719 | "nmsub.s %[di1], %[di1], $f1, %[IQ] \n\t" | | |
| 1720 | "nmsub.s %[di2], %[di2], $f2, %[IQ] \n\t" | | |
| 1721 | "nmsub.s %[di3], %[di3], $f3, %[IQ] \n\t" | | |
| 1722 | | | |
| 1723 | ".set pop \n\t" | | |
| 1724 | | | |
| 1725 | : [di0]"=&f"(di0), [di1]"=&f"(di1), | | |
| 1726 | [di2]"=&f"(di2), [di3]"=&f"(di3) | | |
| 1727 | : [in_pos]"r"(in_pos), [vec]"r"(vec), | | |
| 1728 | [IQ]"f"(IQ) | | |
| 1729 | : "$f0", "$f1", "$f2", "$f3", | | |
| 1730 | "memory" | | |
| 1731 | ); | | |
| 1732 | | | |
| 1733 | cost += di0 * di0 + di1 * di1 | | |
| 1734 | + di2 * di2 + di3 * di3; | | |
| 1735 | } | | |
| 1736 | | | |
| 1737 | if (bits) | | |
| 1738 | *bits = curbits; | | |
| 1739 | if (energy) | | |
| 1740 | *energy = qenergy * (IQ*IQ); | | |
| 1741 | return cost * lambda + curbits; | | |
| 1742 | } | | |
| 1743 | | | |
| 1744 | static float get_band_cost_SPAIR_mips(struct AACEncContext *s, | | |
| 1745 | PutBitContext *pb, const float *in, | | |
| 1746 | const float *scaled, int size, int scale_idx, | | |
| 1747 | int cb, const float lambda, const float uplim, | | |
| 1748 | int *bits, float *energy) | | |
| 1749 | { | | |
| 1750 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1751 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 1752 | int i; | | |
| 1753 | float cost = 0; | | |
| 1754 | float qenergy = 0.0f; | | |
| 1755 | int qc1, qc2, qc3, qc4; | | |
| 1756 | int curbits = 0; | | |
| 1757 | | | |
| 1758 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 1759 | float *p_codes = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 1760 | | | |
| 1761 | for (i = 0; i < size; i += 4) { | | |
| 1762 | const float *vec, *vec2; | | |
| 1763 | int curidx, curidx2; | | |
| 1764 | int *in_int = (int *)&in[i]; | | |
| 1765 | float *in_pos = (float *)&in[i]; | | |
| 1766 | float di0, di1, di2, di3; | | |
| 1767 | int t0, t1, t2, t3, t4, t5, t6, t7; | | |
| 1768 | | | |
| 1769 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1770 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1771 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1772 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1773 | | | |
| 1774 | __asm__ volatile ( | | |
| 1775 | ".set push \n\t" | | |
| 1776 | ".set noreorder \n\t" | | |
| 1777 | | | |
| 1778 | "ori %[t4], $zero, 4 \n\t" | | |
| 1779 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 1780 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 1781 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 1782 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 1783 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1784 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 1785 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 1786 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 1787 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 1788 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 1789 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 1790 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 1791 | "srl %[t0], %[t0], 31 \n\t" | | |
| 1792 | "srl %[t1], %[t1], 31 \n\t" | | |
| 1793 | "srl %[t2], %[t2], 31 \n\t" | | |
| 1794 | "srl %[t3], %[t3], 31 \n\t" | | |
| 1795 | "subu %[t4], $zero, %[qc1] \n\t" | | |
| 1796 | "subu %[t5], $zero, %[qc2] \n\t" | | |
| 1797 | "subu %[t6], $zero, %[qc3] \n\t" | | |
| 1798 | "subu %[t7], $zero, %[qc4] \n\t" | | |
| 1799 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1800 | "movn %[qc2], %[t5], %[t1] \n\t" | | |
| 1801 | "movn %[qc3], %[t6], %[t2] \n\t" | | |
| 1802 | "movn %[qc4], %[t7], %[t3] \n\t" | | |
| 1803 | | | |
| 1804 | ".set pop \n\t" | | |
| 1805 | | | |
| 1806 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1807 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1808 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1809 | [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7) | | |
| 1810 | : [in_int]"r"(in_int) | | |
| 1811 | : "memory" | | |
| 1812 | ); | | |
| 1813 | | | |
| 1814 | curidx = 9 * qc1; | | |
| 1815 | curidx += qc2 + 40; | | |
| 1816 | | | |
| 1817 | curidx2 = 9 * qc3; | | |
| 1818 | curidx2 += qc4 + 40; | | |
| 1819 | | | |
| 1820 | curbits += p_bits[curidx]; | | |
| 1821 | curbits += p_bits[curidx2]; | | |
| 1822 | | | |
| 1823 | vec = &p_codes[curidx*2]; | | |
| 1824 | vec2 = &p_codes[curidx2*2]; | | |
| 1825 | | | |
| 1826 | qenergy += vec[0]*vec[0] + vec[1]*vec[1] | | |
| 1827 | + vec2[0]*vec2[0] + vec2[1]*vec2[1]; | | |
| 1828 | | | |
| 1829 | __asm__ volatile ( | | |
| 1830 | ".set push \n\t" | | |
| 1831 | ".set noreorder \n\t" | | |
| 1832 | | | |
| 1833 | "lwc1 $f0, 0(%[in_pos]) \n\t" | | |
| 1834 | "lwc1 $f1, 0(%[vec]) \n\t" | | |
| 1835 | "lwc1 $f2, 4(%[in_pos]) \n\t" | | |
| 1836 | "lwc1 $f3, 4(%[vec]) \n\t" | | |
| 1837 | "lwc1 $f4, 8(%[in_pos]) \n\t" | | |
| 1838 | "lwc1 $f5, 0(%[vec2]) \n\t" | | |
| 1839 | "lwc1 $f6, 12(%[in_pos]) \n\t" | | |
| 1840 | "lwc1 $f7, 4(%[vec2]) \n\t" | | |
| 1841 | "nmsub.s %[di0], $f0, $f1, %[IQ] \n\t" | | |
| 1842 | "nmsub.s %[di1], $f2, $f3, %[IQ] \n\t" | | |
| 1843 | "nmsub.s %[di2], $f4, $f5, %[IQ] \n\t" | | |
| 1844 | "nmsub.s %[di3], $f6, $f7, %[IQ] \n\t" | | |
| 1845 | | | |
| 1846 | ".set pop \n\t" | | |
| 1847 | | | |
| 1848 | : [di0]"=&f"(di0), [di1]"=&f"(di1), | | |
| 1849 | [di2]"=&f"(di2), [di3]"=&f"(di3) | | |
| 1850 | : [in_pos]"r"(in_pos), [vec]"r"(vec), | | |
| 1851 | [vec2]"r"(vec2), [IQ]"f"(IQ) | | |
| 1852 | : "$f0", "$f1", "$f2", "$f3", | | |
| 1853 | "$f4", "$f5", "$f6", "$f7", | | |
| 1854 | "memory" | | |
| 1855 | ); | | |
| 1856 | | | |
| 1857 | cost += di0 * di0 + di1 * di1 | | |
| 1858 | + di2 * di2 + di3 * di3; | | |
| 1859 | } | | |
| 1860 | | | |
| 1861 | if (bits) | | |
| 1862 | *bits = curbits; | | |
| 1863 | if (energy) | | |
| 1864 | *energy = qenergy * (IQ*IQ); | | |
| 1865 | return cost * lambda + curbits; | | |
| 1866 | } | | |
| 1867 | | | |
| 1868 | static float get_band_cost_UPAIR7_mips(struct AACEncContext *s, | | |
| 1869 | PutBitContext *pb, const float *in, | | |
| 1870 | const float *scaled, int size, int scale_idx, | | |
| 1871 | int cb, const float lambda, const float uplim, | | |
| 1872 | int *bits, float *energy) | | |
| 1873 | { | | |
| 1874 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 1875 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 1876 | int i; | | |
| 1877 | float cost = 0; | | |
| 1878 | float qenergy = 0.0f; | | |
| 1879 | int qc1, qc2, qc3, qc4; | | |
| 1880 | int curbits = 0; | | |
| 1881 | | | |
| 1882 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 1883 | float *p_codes = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 1884 | | | |
| 1885 | for (i = 0; i < size; i += 4) { | | |
| 1886 | const float *vec, *vec2; | | |
| 1887 | int curidx, curidx2, sign1, count1, sign2, count2; | | |
| 1888 | int *in_int = (int *)&in[i]; | | |
| 1889 | float *in_pos = (float *)&in[i]; | | |
| 1890 | float di0, di1, di2, di3; | | |
| 1891 | int t0, t1, t2, t3, t4; | | |
| 1892 | | | |
| 1893 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 1894 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 1895 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 1896 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 1897 | | | |
| 1898 | __asm__ volatile ( | | |
| 1899 | ".set push \n\t" | | |
| 1900 | ".set noreorder \n\t" | | |
| 1901 | | | |
| 1902 | "ori %[t4], $zero, 7 \n\t" | | |
| 1903 | "ori %[sign1], $zero, 0 \n\t" | | |
| 1904 | "ori %[sign2], $zero, 0 \n\t" | | |
| 1905 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 1906 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 1907 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 1908 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 1909 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 1910 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 1911 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 1912 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 1913 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 1914 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 1915 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 1916 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 1917 | "slt %[t0], %[t0], $zero \n\t" | | |
| 1918 | "movn %[sign1], %[t0], %[qc1] \n\t" | | |
| 1919 | "slt %[t2], %[t2], $zero \n\t" | | |
| 1920 | "movn %[sign2], %[t2], %[qc3] \n\t" | | |
| 1921 | "slt %[t1], %[t1], $zero \n\t" | | |
| 1922 | "sll %[t0], %[sign1], 1 \n\t" | | |
| 1923 | "or %[t0], %[t0], %[t1] \n\t" | | |
| 1924 | "movn %[sign1], %[t0], %[qc2] \n\t" | | |
| 1925 | "slt %[t3], %[t3], $zero \n\t" | | |
| 1926 | "sll %[t0], %[sign2], 1 \n\t" | | |
| 1927 | "or %[t0], %[t0], %[t3] \n\t" | | |
| 1928 | "movn %[sign2], %[t0], %[qc4] \n\t" | | |
| 1929 | "slt %[count1], $zero, %[qc1] \n\t" | | |
| 1930 | "slt %[t1], $zero, %[qc2] \n\t" | | |
| 1931 | "slt %[count2], $zero, %[qc3] \n\t" | | |
| 1932 | "slt %[t2], $zero, %[qc4] \n\t" | | |
| 1933 | "addu %[count1], %[count1], %[t1] \n\t" | | |
| 1934 | "addu %[count2], %[count2], %[t2] \n\t" | | |
| 1935 | | | |
| 1936 | ".set pop \n\t" | | |
| 1937 | | | |
| 1938 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 1939 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 1940 | [sign1]"=&r"(sign1), [count1]"=&r"(count1), | | |
| 1941 | [sign2]"=&r"(sign2), [count2]"=&r"(count2), | | |
| 1942 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 1943 | [t4]"=&r"(t4) | | |
| 1944 | : [in_int]"r"(in_int) | | |
| 1945 | : "memory" | | |
| 1946 | ); | | |
| 1947 | | | |
| 1948 | curidx = 8 * qc1; | | |
| 1949 | curidx += qc2; | | |
| 1950 | | | |
| 1951 | curidx2 = 8 * qc3; | | |
| 1952 | curidx2 += qc4; | | |
| 1953 | | | |
| 1954 | curbits += p_bits[curidx]; | | |
| 1955 | curbits += upair7_sign_bits[curidx]; | | |
| 1956 | vec = &p_codes[curidx*2]; | | |
| 1957 | | | |
| 1958 | curbits += p_bits[curidx2]; | | |
| 1959 | curbits += upair7_sign_bits[curidx2]; | | |
| 1960 | vec2 = &p_codes[curidx2*2]; | | |
| 1961 | | | |
| 1962 | qenergy += vec[0]*vec[0] + vec[1]*vec[1] | | |
| 1963 | + vec2[0]*vec2[0] + vec2[1]*vec2[1]; | | |
| 1964 | | | |
| 1965 | __asm__ volatile ( | | |
| 1966 | ".set push \n\t" | | |
| 1967 | ".set noreorder \n\t" | | |
| 1968 | | | |
| 1969 | "lwc1 %[di0], 0(%[in_pos]) \n\t" | | |
| 1970 | "lwc1 %[di1], 4(%[in_pos]) \n\t" | | |
| 1971 | "lwc1 %[di2], 8(%[in_pos]) \n\t" | | |
| 1972 | "lwc1 %[di3], 12(%[in_pos]) \n\t" | | |
| 1973 | "abs.s %[di0], %[di0] \n\t" | | |
| 1974 | "abs.s %[di1], %[di1] \n\t" | | |
| 1975 | "abs.s %[di2], %[di2] \n\t" | | |
| 1976 | "abs.s %[di3], %[di3] \n\t" | | |
| 1977 | "lwc1 $f0, 0(%[vec]) \n\t" | | |
| 1978 | "lwc1 $f1, 4(%[vec]) \n\t" | | |
| 1979 | "lwc1 $f2, 0(%[vec2]) \n\t" | | |
| 1980 | "lwc1 $f3, 4(%[vec2]) \n\t" | | |
| 1981 | "nmsub.s %[di0], %[di0], $f0, %[IQ] \n\t" | | |
| 1982 | "nmsub.s %[di1], %[di1], $f1, %[IQ] \n\t" | | |
| 1983 | "nmsub.s %[di2], %[di2], $f2, %[IQ] \n\t" | | |
| 1984 | "nmsub.s %[di3], %[di3], $f3, %[IQ] \n\t" | | |
| 1985 | | | |
| 1986 | ".set pop \n\t" | | |
| 1987 | | | |
| 1988 | : [di0]"=&f"(di0), [di1]"=&f"(di1), | | |
| 1989 | [di2]"=&f"(di2), [di3]"=&f"(di3) | | |
| 1990 | : [in_pos]"r"(in_pos), [vec]"r"(vec), | | |
| 1991 | [vec2]"r"(vec2), [IQ]"f"(IQ) | | |
| 1992 | : "$f0", "$f1", "$f2", "$f3", | | |
| 1993 | "memory" | | |
| 1994 | ); | | |
| 1995 | | | |
| 1996 | cost += di0 * di0 + di1 * di1 | | |
| 1997 | + di2 * di2 + di3 * di3; | | |
| 1998 | } | | |
| 1999 | | | |
| 2000 | if (bits) | | |
| 2001 | *bits = curbits; | | |
| 2002 | if (energy) | | |
| 2003 | *energy = qenergy * (IQ*IQ); | | |
| 2004 | return cost * lambda + curbits; | | |
| 2005 | } | | |
| 2006 | | | |
| 2007 | static float get_band_cost_UPAIR12_mips(struct AACEncContext *s, | | |
| 2008 | PutBitContext *pb, const float *in, | | |
| 2009 | const float *scaled, int size, int scale_idx, | | |
| 2010 | int cb, const float lambda, const float uplim, | | |
| 2011 | int *bits, float *energy) | | |
| 2012 | { | | |
| 2013 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 2014 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 2015 | int i; | | |
| 2016 | float cost = 0; | | |
| 2017 | float qenergy = 0.0f; | | |
| 2018 | int qc1, qc2, qc3, qc4; | | |
| 2019 | int curbits = 0; | | |
| 2020 | | | |
| 2021 | uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1]; | | |
| 2022 | float *p_codes = (float *)ff_aac_codebook_vectors[cb-1]; | | |
| 2023 | | | |
| 2024 | for (i = 0; i < size; i += 4) { | | |
| 2025 | const float *vec, *vec2; | | |
| 2026 | int curidx, curidx2; | | |
| 2027 | int sign1, count1, sign2, count2; | | |
| 2028 | int *in_int = (int *)&in[i]; | | |
| 2029 | float *in_pos = (float *)&in[i]; | | |
| 2030 | float di0, di1, di2, di3; | | |
| 2031 | int t0, t1, t2, t3, t4; | | |
| 2032 | | | |
| 2033 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 2034 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 2035 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 2036 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 2037 | | | |
| 2038 | __asm__ volatile ( | | |
| 2039 | ".set push \n\t" | | |
| 2040 | ".set noreorder \n\t" | | |
| 2041 | | | |
| 2042 | "ori %[t4], $zero, 12 \n\t" | | |
| 2043 | "ori %[sign1], $zero, 0 \n\t" | | |
| 2044 | "ori %[sign2], $zero, 0 \n\t" | | |
| 2045 | "slt %[t0], %[t4], %[qc1] \n\t" | | |
| 2046 | "slt %[t1], %[t4], %[qc2] \n\t" | | |
| 2047 | "slt %[t2], %[t4], %[qc3] \n\t" | | |
| 2048 | "slt %[t3], %[t4], %[qc4] \n\t" | | |
| 2049 | "movn %[qc1], %[t4], %[t0] \n\t" | | |
| 2050 | "movn %[qc2], %[t4], %[t1] \n\t" | | |
| 2051 | "movn %[qc3], %[t4], %[t2] \n\t" | | |
| 2052 | "movn %[qc4], %[t4], %[t3] \n\t" | | |
| 2053 | "lw %[t0], 0(%[in_int]) \n\t" | | |
| 2054 | "lw %[t1], 4(%[in_int]) \n\t" | | |
| 2055 | "lw %[t2], 8(%[in_int]) \n\t" | | |
| 2056 | "lw %[t3], 12(%[in_int]) \n\t" | | |
| 2057 | "slt %[t0], %[t0], $zero \n\t" | | |
| 2058 | "movn %[sign1], %[t0], %[qc1] \n\t" | | |
| 2059 | "slt %[t2], %[t2], $zero \n\t" | | |
| 2060 | "movn %[sign2], %[t2], %[qc3] \n\t" | | |
| 2061 | "slt %[t1], %[t1], $zero \n\t" | | |
| 2062 | "sll %[t0], %[sign1], 1 \n\t" | | |
| 2063 | "or %[t0], %[t0], %[t1] \n\t" | | |
| 2064 | "movn %[sign1], %[t0], %[qc2] \n\t" | | |
| 2065 | "slt %[t3], %[t3], $zero \n\t" | | |
| 2066 | "sll %[t0], %[sign2], 1 \n\t" | | |
| 2067 | "or %[t0], %[t0], %[t3] \n\t" | | |
| 2068 | "movn %[sign2], %[t0], %[qc4] \n\t" | | |
| 2069 | "slt %[count1], $zero, %[qc1] \n\t" | | |
| 2070 | "slt %[t1], $zero, %[qc2] \n\t" | | |
| 2071 | "slt %[count2], $zero, %[qc3] \n\t" | | |
| 2072 | "slt %[t2], $zero, %[qc4] \n\t" | | |
| 2073 | "addu %[count1], %[count1], %[t1] \n\t" | | |
| 2074 | "addu %[count2], %[count2], %[t2] \n\t" | | |
| 2075 | | | |
| 2076 | ".set pop \n\t" | | |
| 2077 | | | |
| 2078 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 2079 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 2080 | [sign1]"=&r"(sign1), [count1]"=&r"(count1), | | |
| 2081 | [sign2]"=&r"(sign2), [count2]"=&r"(count2), | | |
| 2082 | [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3), | | |
| 2083 | [t4]"=&r"(t4) | | |
| 2084 | : [in_int]"r"(in_int) | | |
| 2085 | : "memory" | | |
| 2086 | ); | | |
| 2087 | | | |
| 2088 | curidx = 13 * qc1; | | |
| 2089 | curidx += qc2; | | |
| 2090 | | | |
| 2091 | curidx2 = 13 * qc3; | | |
| 2092 | curidx2 += qc4; | | |
| 2093 | | | |
| 2094 | curbits += p_bits[curidx]; | | |
| 2095 | curbits += p_bits[curidx2]; | | |
| 2096 | curbits += upair12_sign_bits[curidx]; | | |
| 2097 | curbits += upair12_sign_bits[curidx2]; | | |
| 2098 | vec = &p_codes[curidx*2]; | | |
| 2099 | vec2 = &p_codes[curidx2*2]; | | |
| 2100 | | | |
| 2101 | qenergy += vec[0]*vec[0] + vec[1]*vec[1] | | |
| 2102 | + vec2[0]*vec2[0] + vec2[1]*vec2[1]; | | |
| 2103 | | | |
| 2104 | __asm__ volatile ( | | |
| 2105 | ".set push \n\t" | | |
| 2106 | ".set noreorder \n\t" | | |
| 2107 | | | |
| 2108 | "lwc1 %[di0], 0(%[in_pos]) \n\t" | | |
| 2109 | "lwc1 %[di1], 4(%[in_pos]) \n\t" | | |
| 2110 | "lwc1 %[di2], 8(%[in_pos]) \n\t" | | |
| 2111 | "lwc1 %[di3], 12(%[in_pos]) \n\t" | | |
| 2112 | "abs.s %[di0], %[di0] \n\t" | | |
| 2113 | "abs.s %[di1], %[di1] \n\t" | | |
| 2114 | "abs.s %[di2], %[di2] \n\t" | | |
| 2115 | "abs.s %[di3], %[di3] \n\t" | | |
| 2116 | "lwc1 $f0, 0(%[vec]) \n\t" | | |
| 2117 | "lwc1 $f1, 4(%[vec]) \n\t" | | |
| 2118 | "lwc1 $f2, 0(%[vec2]) \n\t" | | |
| 2119 | "lwc1 $f3, 4(%[vec2]) \n\t" | | |
| 2120 | "nmsub.s %[di0], %[di0], $f0, %[IQ] \n\t" | | |
| 2121 | "nmsub.s %[di1], %[di1], $f1, %[IQ] \n\t" | | |
| 2122 | "nmsub.s %[di2], %[di2], $f2, %[IQ] \n\t" | | |
| 2123 | "nmsub.s %[di3], %[di3], $f3, %[IQ] \n\t" | | |
| 2124 | | | |
| 2125 | ".set pop \n\t" | | |
| 2126 | | | |
| 2127 | : [di0]"=&f"(di0), [di1]"=&f"(di1), | | |
| 2128 | [di2]"=&f"(di2), [di3]"=&f"(di3) | | |
| 2129 | : [in_pos]"r"(in_pos), [vec]"r"(vec), | | |
| 2130 | [vec2]"r"(vec2), [IQ]"f"(IQ) | | |
| 2131 | : "$f0", "$f1", "$f2", "$f3", | | |
| 2132 | "memory" | | |
| 2133 | ); | | |
| 2134 | | | |
| 2135 | cost += di0 * di0 + di1 * di1 | | |
| 2136 | + di2 * di2 + di3 * di3; | | |
| 2137 | } | | |
| 2138 | | | |
| 2139 | if (bits) | | |
| 2140 | *bits = curbits; | | |
| 2141 | if (energy) | | |
| 2142 | *energy = qenergy * (IQ*IQ); | | |
| 2143 | return cost * lambda + curbits; | | |
| 2144 | } | | |
| 2145 | | | |
| 2146 | static float get_band_cost_ESC_mips(struct AACEncContext *s, | | |
| 2147 | PutBitContext *pb, const float *in, | | |
| 2148 | const float *scaled, int size, int scale_idx, | | |
| 2149 | int cb, const float lambda, const float uplim, | | |
| 2150 | int *bits, float *energy) | | |
| 2151 | { | | |
| 2152 | const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; | | |
| 2153 | const float IQ = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; | | |
| 2154 | const float CLIPPED_ESCAPE = 165140.0f * IQ; | | |
| 2155 | int i; | | |
| 2156 | float cost = 0; | | |
| 2157 | float qenergy = 0.0f; | | |
| 2158 | int qc1, qc2, qc3, qc4; | | |
| 2159 | int curbits = 0; | | |
| 2160 | | | |
| 2161 | uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1]; | | |
| 2162 | float *p_codes = (float* )ff_aac_codebook_vectors[cb-1]; | | |
| 2163 | | | |
| 2164 | for (i = 0; i < size; i += 4) { | | |
| 2165 | const float *vec, *vec2; | | |
| 2166 | int curidx, curidx2; | | |
| 2167 | float t1, t2, t3, t4, V; | | |
| 2168 | float di1, di2, di3, di4; | | |
| 2169 | int cond0, cond1, cond2, cond3; | | |
| 2170 | int c1, c2, c3, c4; | | |
| 2171 | int t6, t7; | | |
| 2172 | | | |
| 2173 | qc1 = scaled[i ] * Q34 + ROUND_STANDARD; | | |
| 2174 | qc2 = scaled[i+1] * Q34 + ROUND_STANDARD; | | |
| 2175 | qc3 = scaled[i+2] * Q34 + ROUND_STANDARD; | | |
| 2176 | qc4 = scaled[i+3] * Q34 + ROUND_STANDARD; | | |
| 2177 | | | |
| 2178 | __asm__ volatile ( | | |
| 2179 | ".set push \n\t" | | |
| 2180 | ".set noreorder \n\t" | | |
| 2181 | | | |
| 2182 | "ori %[t6], $zero, 15 \n\t" | | |
| 2183 | "ori %[t7], $zero, 16 \n\t" | | |
| 2184 | "shll_s.w %[c1], %[qc1], 18 \n\t" | | |
| 2185 | "shll_s.w %[c2], %[qc2], 18 \n\t" | | |
| 2186 | "shll_s.w %[c3], %[qc3], 18 \n\t" | | |
| 2187 | "shll_s.w %[c4], %[qc4], 18 \n\t" | | |
| 2188 | "srl %[c1], %[c1], 18 \n\t" | | |
| 2189 | "srl %[c2], %[c2], 18 \n\t" | | |
| 2190 | "srl %[c3], %[c3], 18 \n\t" | | |
| 2191 | "srl %[c4], %[c4], 18 \n\t" | | |
| 2192 | "slt %[cond0], %[t6], %[qc1] \n\t" | | |
| 2193 | "slt %[cond1], %[t6], %[qc2] \n\t" | | |
| 2194 | "slt %[cond2], %[t6], %[qc3] \n\t" | | |
| 2195 | "slt %[cond3], %[t6], %[qc4] \n\t" | | |
| 2196 | "movn %[qc1], %[t7], %[cond0] \n\t" | | |
| 2197 | "movn %[qc2], %[t7], %[cond1] \n\t" | | |
| 2198 | "movn %[qc3], %[t7], %[cond2] \n\t" | | |
| 2199 | "movn %[qc4], %[t7], %[cond3] \n\t" | | |
| 2200 | | | |
| 2201 | ".set pop \n\t" | | |
| 2202 | | | |
| 2203 | : [qc1]"+r"(qc1), [qc2]"+r"(qc2), | | |
| 2204 | [qc3]"+r"(qc3), [qc4]"+r"(qc4), | | |
| 2205 | [cond0]"=&r"(cond0), [cond1]"=&r"(cond1), | | |
| 2206 | [cond2]"=&r"(cond2), [cond3]"=&r"(cond3), | | |
| 2207 | [c1]"=&r"(c1), [c2]"=&r"(c2), | | |
| 2208 | [c3]"=&r"(c3), [c4]"=&r"(c4), | | |
| 2209 | [t6]"=&r"(t6), [t7]"=&r"(t7) | | |
| 2210 | ); | | |
| 2211 | | | |
| 2212 | curidx = 17 * qc1; | | |
| 2213 | curidx += qc2; | | |
| 2214 | | | |
| 2215 | curidx2 = 17 * qc3; | | |
| 2216 | curidx2 += qc4; | | |
| 2217 | | | |
| 2218 | curbits += p_bits[curidx]; | | |
| 2219 | curbits += esc_sign_bits[curidx]; | | |
| 2220 | vec = &p_codes[curidx*2]; | | |
| 2221 | | | |
| 2222 | curbits += p_bits[curidx2]; | | |
| 2223 | curbits += esc_sign_bits[curidx2]; | | |
| 2224 | vec2 = &p_codes[curidx2*2]; | | |
| 2225 | | | |
| 2226 | curbits += (av_log2(c1) * 2 - 3) & (-cond0); | | |
| 2227 | curbits += (av_log2(c2) * 2 - 3) & (-cond1); | | |
| 2228 | curbits += (av_log2(c3) * 2 - 3) & (-cond2); | | |
| 2229 | curbits += (av_log2(c4) * 2 - 3) & (-cond3); | | |
| 2230 | | | |
| 2231 | t1 = fabsf(in[i ]); | | |
| 2232 | t2 = fabsf(in[i+1]); | | |
| 2233 | t3 = fabsf(in[i+2]); | | |
| 2234 | t4 = fabsf(in[i+3]); | | |
| 2235 | | | |
| 2236 | if (cond0) { | | |
| 2237 | if (t1 >= CLIPPED_ESCAPE) { | | |
| 2238 | di1 = t1 - CLIPPED_ESCAPE; | | |
| 2239 | qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE; | | |
| 2240 | } else { | | |
| 2241 | di1 = t1 - (V = c1 * cbrtf(c1) * IQ); | | |
| 2242 | qenergy += V*V; | | |
| 2243 | } | | |
| 2244 | } else { | | |
| 2245 | di1 = t1 - (V = vec[0] * IQ); | | |
| 2246 | qenergy += V*V; | | |
| 2247 | } | | |
| 2248 | | | |
| 2249 | if (cond1) { | | |
| 2250 | if (t2 >= CLIPPED_ESCAPE) { | | |
| 2251 | di2 = t2 - CLIPPED_ESCAPE; | | |
| 2252 | qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE; | | |
| 2253 | } else { | | |
| 2254 | di2 = t2 - (V = c2 * cbrtf(c2) * IQ); | | |
| 2255 | qenergy += V*V; | | |
| 2256 | } | | |
| 2257 | } else { | | |
| 2258 | di2 = t2 - (V = vec[1] * IQ); | | |
| 2259 | qenergy += V*V; | | |
| 2260 | } | | |
| 2261 | | | |
| 2262 | if (cond2) { | | |
| 2263 | if (t3 >= CLIPPED_ESCAPE) { | | |
| 2264 | di3 = t3 - CLIPPED_ESCAPE; | | |
| 2265 | qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE; | | |
| 2266 | } else { | | |
| 2267 | di3 = t3 - (V = c3 * cbrtf(c3) * IQ); | | |
| 2268 | qenergy += V*V; | | |
| 2269 | } | | |
| 2270 | } else { | | |
| 2271 | di3 = t3 - (V = vec2[0] * IQ); | | |
| 2272 | qenergy += V*V; | | |
| 2273 | } | | |
| 2274 | | | |
| 2275 | if (cond3) { | | |
| 2276 | if (t4 >= CLIPPED_ESCAPE) { | | |
| 2277 | di4 = t4 - CLIPPED_ESCAPE; | | |
| 2278 | qenergy += CLIPPED_ESCAPE*CLIPPED_ESCAPE; | | |
| 2279 | } else { | | |
| 2280 | di4 = t4 - (V = c4 * cbrtf(c4) * IQ); | | |
| 2281 | qenergy += V*V; | | |
| 2282 | } | | |
| 2283 | } else { | | |
| 2284 | di4 = t4 - (V = vec2[1]*IQ); | | |
| 2285 | qenergy += V*V; | | |
| 2286 | } | | |
| 2287 | | | |
| 2288 | cost += di1 * di1 + di2 * di2 | | |
| 2289 | + di3 * di3 + di4 * di4; | | |
| 2290 | } | | |
| 2291 | | | |
| 2292 | if (bits) | | |
| 2293 | *bits = curbits; | | |
| 2294 | return cost * lambda + curbits; | | |
| 2295 | } | | |
| 2296 | | | |
| 2297 | static float (*const get_band_cost_arr[])(struct AACEncContext *s, | | |
| 2298 | PutBitContext *pb, const float *in, | | |
| 2299 | const float *scaled, int size, int scale_idx, | | |
| 2300 | int cb, const float lambda, const float uplim, | | |
| 2301 | int *bits, float *energy) = { | | |
| 2302 | get_band_cost_ZERO_mips, | | |
| 2303 | get_band_cost_SQUAD_mips, | | |
| 2304 | get_band_cost_SQUAD_mips, | | |
| 2305 | get_band_cost_UQUAD_mips, | | |
| 2306 | get_band_cost_UQUAD_mips, | | |
| 2307 | get_band_cost_SPAIR_mips, | | |
| 2308 | get_band_cost_SPAIR_mips, | | |
| 2309 | get_band_cost_UPAIR7_mips, | | |
| 2310 | get_band_cost_UPAIR7_mips, | | |
| 2311 | get_band_cost_UPAIR12_mips, | | |
| 2312 | get_band_cost_UPAIR12_mips, | | |
| 2313 | get_band_cost_ESC_mips, | | |
| 2314 | get_band_cost_NONE_mips, /* cb 12 doesn't exist */ | | |
| 2315 | get_band_cost_ZERO_mips, | | |
| 2316 | get_band_cost_ZERO_mips, | | |
| 2317 | get_band_cost_ZERO_mips, | | |
| 2318 | }; | | |
| 2319 | | | |
| 2320 | #define get_band_cost( \ | | |
| 2321 | s, pb, in, scaled, size, scale_idx, cb, \ | | |
| 2322 | lambda, uplim, bits, energy) \ | | |
| 2323 | get_band_cost_arr[cb]( \ | | |
| 2324 | s, pb, in, scaled, size, scale_idx, cb, \ | | |
| 2325 | lambda, uplim, bits, energy) | | |
| 2326 | | | |
| 2327 | static float quantize_band_cost(struct AACEncContext *s, const float *in, | | |
| 2328 | const float *scaled, int size, int scale_idx, | | |
| 2329 | int cb, const float lambda, const float uplim, | | |
| 2330 | int *bits, float *energy) | | |
| 2331 | { | | |
| 2332 | return get_band_cost(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits, energy); | | |
| 2333 | } | | |
| 2334 | | | |
| 2335 | #include "libavcodec/aacenc_quantization_misc.h" | | |
| 2336 | | | |
| 2337 | #include "libavcodec/aaccoder_twoloop.h" | | |
| 2338 | | | |
| 2339 | static void search_for_ms_mips(AACEncContext *s, ChannelElement *cpe) | | |
| 2340 | { | | |
| 2341 | int start = 0, i, w, w2, g, sid_sf_boost, prev_mid, prev_side; | | |
| 2342 | uint8_t nextband0[128], nextband1[128]; | | |
| 2343 | float M[128], S[128]; | | |
| 2344 | float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3; | | |
| 2345 | const float lambda = s->lambda; | | |
| 2346 | const float mslambda = FFMIN(1.0f, lambda / 120.f); | | |
| 2347 | SingleChannelElement *sce0 = &cpe->ch[0]; | | |
| 2348 | SingleChannelElement *sce1 = &cpe->ch[1]; | | |
| 2349 | if (!cpe->common_window) | | |
| 2350 | return; | | |
| 2351 | | | |
| 2352 | /** Scout out next nonzero bands */ | | |
| 2353 | ff_init_nextband_map(sce0, nextband0); | | |
| 2354 | ff_init_nextband_map(sce1, nextband1); | | |
| 2355 | | | |
| 2356 | prev_mid = sce0->sf_idx[0]; | | |
| 2357 | prev_side = sce1->sf_idx[0]; | | |
| 2358 | for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) { | | |
| 2359 | start = 0; | | |
| 2360 | for (g = 0; g < sce0->ics.num_swb; g++) { | | |
| 2361 | float bmax = bval2bmax(g * 17.0f / sce0->ics.num_swb) / 0.0045f; | | |
| 2362 | if (!cpe->is_mask[w*16+g]) | | |
| 2363 | cpe->ms_mask[w*16+g] = 0; | | |
| 2364 | if (!sce0->zeroes[w*16+g] && !sce1->zeroes[w*16+g] && !cpe->is_mask[w*16+g]) { | | |
| 2365 | float Mmax = 0.0f, Smax = 0.0f; | | |
| 2366 | | | |
| 2367 | /* Must compute mid/side SF and book for the whole window group */ | | |
| 2368 | for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) { | | |
| 2369 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++) { | | |
| 2370 | M[i] = (sce0->coeffs[start+(w+w2)*128+i] | | |
| 2371 | + sce1->coeffs[start+(w+w2)*128+i]) * 0.5; | | |
| 2372 | S[i] = M[i] | | |
| 2373 | - sce1->coeffs[start+(w+w2)*128+i]; | | |
| 2374 | } | | |
| 2375 | abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]); | | |
| 2376 | abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]); | | |
| 2377 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++ ) { | | |
| 2378 | Mmax = FFMAX(Mmax, M34[i]); | | |
| 2379 | Smax = FFMAX(Smax, S34[i]); | | |
| 2380 | } | | |
| 2381 | } | | |
| 2382 | | | |
| 2383 | for (sid_sf_boost = 0; sid_sf_boost < 4; sid_sf_boost++) { | | |
| 2384 | float dist1 = 0.0f, dist2 = 0.0f; | | |
| 2385 | int B0 = 0, B1 = 0; | | |
| 2386 | int minidx; | | |
| 2387 | int mididx, sididx; | | |
| 2388 | int midcb, sidcb; | | |
| 2389 | | | |
| 2390 | minidx = FFMIN(sce0->sf_idx[w*16+g], sce1->sf_idx[w*16+g]); | | |
| 2391 | mididx = av_clip(minidx, 0, SCALE_MAX_POS - SCALE_DIV_512); | | |
| 2392 | sididx = av_clip(minidx - sid_sf_boost * 3, 0, SCALE_MAX_POS - SCALE_DIV_512); | | |
| 2393 | if (sce0->band_type[w*16+g] != NOISE_BT && sce1->band_type[w*16+g] != NOISE_BT | | |
| 2394 | && ( !ff_sfdelta_can_replace(sce0, nextband0, prev_mid, mididx, w*16+g) | | |
| 2395 | || !ff_sfdelta_can_replace(sce1, nextband1, prev_side, sididx, w*16+g))) { | | |
| 2396 | /* scalefactor range violation, bad stuff, will decrease quality unacceptably */ | | |
| 2397 | continue; | | |
| 2398 | } | | |
| 2399 | | | |
| 2400 | midcb = find_min_book(Mmax, mididx); | | |
| 2401 | sidcb = find_min_book(Smax, sididx); | | |
| 2402 | | | |
| 2403 | /* No CB can be zero */ | | |
| 2404 | midcb = FFMAX(1,midcb); | | |
| 2405 | sidcb = FFMAX(1,sidcb); | | |
| 2406 | | | |
| 2407 | for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) { | | |
| 2408 | FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g]; | | |
| 2409 | FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g]; | | |
| 2410 | float minthr = FFMIN(band0->threshold, band1->threshold); | | |
| 2411 | int b1,b2,b3,b4; | | |
| 2412 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++) { | | |
| 2413 | M[i] = (sce0->coeffs[start+(w+w2)*128+i] | | |
| 2414 | + sce1->coeffs[start+(w+w2)*128+i]) * 0.5; | | |
| 2415 | S[i] = M[i] | | |
| 2416 | - sce1->coeffs[start+(w+w2)*128+i]; | | |
| 2417 | } | | |
| 2418 | | | |
| 2419 | abs_pow34_v(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]); | | |
| 2420 | abs_pow34_v(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]); | | |
| 2421 | abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]); | | |
| 2422 | abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]); | | |
| 2423 | dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128], | | |
| 2424 | L34, | | |
| 2425 | sce0->ics.swb_sizes[g], | | |
| 2426 | sce0->sf_idx[w*16+g], | | |
| 2427 | sce0->band_type[w*16+g], | | |
| 2428 | lambda / band0->threshold, INFINITY, &b1, NULL); | | |
| 2429 | dist1 += quantize_band_cost(s, &sce1->coeffs[start + (w+w2)*128], | | |
| 2430 | R34, | | |
| 2431 | sce1->ics.swb_sizes[g], | | |
| 2432 | sce1->sf_idx[w*16+g], | | |
| 2433 | sce1->band_type[w*16+g], | | |
| 2434 | lambda / band1->threshold, INFINITY, &b2, NULL); | | |
| 2435 | dist2 += quantize_band_cost(s, M, | | |
| 2436 | M34, | | |
| 2437 | sce0->ics.swb_sizes[g], | | |
| 2438 | mididx, | | |
| 2439 | midcb, | | |
| 2440 | lambda / minthr, INFINITY, &b3, NULL); | | |
| 2441 | dist2 += quantize_band_cost(s, S, | | |
| 2442 | S34, | | |
| 2443 | sce1->ics.swb_sizes[g], | | |
| 2444 | sididx, | | |
| 2445 | sidcb, | | |
| 2446 | mslambda / (minthr * bmax), INFINITY, &b4, NULL); | | |
| 2447 | B0 += b1+b2; | | |
| 2448 | B1 += b3+b4; | | |
| 2449 | dist1 -= b1+b2; | | |
| 2450 | dist2 -= b3+b4; | | |
| 2451 | } | | |
| 2452 | cpe->ms_mask[w*16+g] = dist2 <= dist1 && B1 < B0; | | |
| 2453 | if (cpe->ms_mask[w*16+g]) { | | |
| 2454 | if (sce0->band_type[w*16+g] != NOISE_BT && sce1->band_type[w*16+g] != NOISE_BT) { | | |
| 2455 | sce0->sf_idx[w*16+g] = mididx; | | |
| 2456 | sce1->sf_idx[w*16+g] = sididx; | | |
| 2457 | sce0->band_type[w*16+g] = midcb; | | |
| 2458 | sce1->band_type[w*16+g] = sidcb; | | |
| 2459 | } else if ((sce0->band_type[w*16+g] != NOISE_BT) ^ (sce1->band_type[w*16+g] != NOISE_BT)) { | | |
| 2460 | /* ms_mask unneeded, and it confuses some decoders */ | | |
| 2461 | cpe->ms_mask[w*16+g] = 0; | | |
| 2462 | } | | |
| 2463 | break; | | |
| 2464 | } else if (B1 > B0) { | | |
| 2465 | /* More boost won't fix this */ | | |
| 2466 | break; | | |
| 2467 | } | | |
| 2468 | } | | |
| 2469 | } | | |
| 2470 | if (!sce0->zeroes[w*16+g] && sce0->band_type[w*16+g] < RESERVED_BT) | | |
| 2471 | prev_mid = sce0->sf_idx[w*16+g]; | | |
| 2472 | if (!sce1->zeroes[w*16+g] && !cpe->is_mask[w*16+g] && sce1->band_type[w*16+g] < RESERVED_BT) | | |
| 2473 | prev_side = sce1->sf_idx[w*16+g]; | | |
| 2474 | start += sce0->ics.swb_sizes[g]; | | |
| 2475 | } | | |
| 2476 | } | | |
| 2477 | } | | |
| 2478 | #endif /*HAVE_MIPSFPU */ | | |
| 2479 | | | |
| 2480 | #include "libavcodec/aaccoder_trellis.h" | | |
| 2481 | | | |
| 2482 | #endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */ | | |
| 2483 | #endif /* HAVE_INLINE_ASM */ | | |
| 2484 | | | |
| 2485 | void ff_aac_coder_init_mips(AACEncContext *c) { | | |
| 2486 | #if HAVE_INLINE_ASM | | |
| 2487 | #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6 | | |
| 2488 | AACCoefficientsEncoder *e = c->coder; | | |
| 2489 | int option = c->options.coder; | | |
| 2490 | | | |
| 2491 | if (option == 2) { | | |
| 2492 | e->quantize_and_encode_band = quantize_and_encode_band_mips; | | |
| 2493 | e->encode_window_bands_info = codebook_trellis_rate; | | |
| 2494 | #if HAVE_MIPSFPU | | |
| 2495 | e->search_for_quantizers = search_for_quantizers_twoloop; | | |
| 2496 | #endif /* HAVE_MIPSFPU */ | | |
| 2497 | } | | |
| 2498 | #if HAVE_MIPSFPU | | |
| 2499 | e->search_for_ms = search_for_ms_mips; | | |
| 2500 | #endif /* HAVE_MIPSFPU */ | | |
| 2501 | #endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */ | | |
| 2502 | #endif /* HAVE_INLINE_ASM */ | | |
| 2503 | } | | |