repositories / deb_ffmpeg.git / blame
| Commit | Line | Data |
|---|---|---|
| 2ba45a60 DM |
1 | /* |
| 2 | * Copyright (c) 2009 David Conrad | |
| 3 | * | |
| 4 | * This file is part of FFmpeg. | |
| 5 | * | |
| 6 | * FFmpeg is free software; you can redistribute it and/or | |
| 7 | * modify it under the terms of the GNU Lesser General Public | |
| 8 | * License as published by the Free Software Foundation; either | |
| 9 | * version 2.1 of the License, or (at your option) any later version. | |
| 10 | * | |
| 11 | * FFmpeg is distributed in the hope that it will be useful, | |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 14 | * Lesser General Public License for more details. | |
| 15 | * | |
| 16 | * You should have received a copy of the GNU Lesser General Public | |
| 17 | * License along with FFmpeg; if not, write to the Free Software | |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 19 | */ | |
| 20 | ||
| 21 | #include "libavutil/arm/asm.S" | |
| 22 | ||
| 23 | const vp3_idct_constants, align=4 | |
| 24 | .short 64277, 60547, 54491, 46341, 36410, 25080, 12785 | |
| 25 | endconst | |
| 26 | ||
| 27 | #define xC1S7 d0[0] | |
| 28 | #define xC2S6 d0[1] | |
| 29 | #define xC3S5 d0[2] | |
| 30 | #define xC4S4 d0[3] | |
| 31 | #define xC5S3 d1[0] | |
| 32 | #define xC6S2 d1[1] | |
| 33 | #define xC7S1 d1[2] | |
| 34 | ||
| 35 | .macro vp3_loop_filter | |
| 36 | vsubl.u8 q3, d18, d17 | |
| 37 | vsubl.u8 q2, d16, d19 | |
| 38 | vadd.i16 q1, q3, q3 | |
| 39 | vadd.i16 q2, q2, q3 | |
| 40 | vadd.i16 q0, q1, q2 | |
| 41 | vrshr.s16 q0, q0, #3 | |
| 42 | vmovl.u8 q9, d18 | |
| 43 | vdup.u16 q15, r2 | |
| 44 | ||
| 45 | vabs.s16 q1, q0 | |
| 46 | vshr.s16 q0, q0, #15 | |
| 47 | vqsub.u16 q2, q15, q1 | |
| 48 | vqsub.u16 q3, q2, q1 | |
| 49 | vsub.i16 q1, q2, q3 | |
| 50 | veor q1, q1, q0 | |
| 51 | vsub.i16 q0, q1, q0 | |
| 52 | ||
| 53 | vaddw.u8 q2, q0, d17 | |
| 54 | vsub.i16 q3, q9, q0 | |
| 55 | vqmovun.s16 d0, q2 | |
| 56 | vqmovun.s16 d1, q3 | |
| 57 | .endm | |
| 58 | ||
| 59 | function ff_vp3_v_loop_filter_neon, export=1 | |
| 60 | sub ip, r0, r1 | |
| 61 | sub r0, r0, r1, lsl #1 | |
| 62 | vld1.64 {d16}, [r0,:64], r1 | |
| 63 | vld1.64 {d17}, [r0,:64], r1 | |
| 64 | vld1.64 {d18}, [r0,:64], r1 | |
| 65 | vld1.64 {d19}, [r0,:64], r1 | |
| 66 | ldrb r2, [r2, #129*4] | |
| 67 | ||
| 68 | vp3_loop_filter | |
| 69 | ||
| 70 | vst1.64 {d0}, [ip,:64], r1 | |
| 71 | vst1.64 {d1}, [ip,:64], r1 | |
| 72 | bx lr | |
| 73 | endfunc | |
| 74 | ||
| 75 | function ff_vp3_h_loop_filter_neon, export=1 | |
| 76 | sub ip, r0, #1 | |
| 77 | sub r0, r0, #2 | |
| 78 | vld1.32 {d16[]}, [r0], r1 | |
| 79 | vld1.32 {d17[]}, [r0], r1 | |
| 80 | vld1.32 {d18[]}, [r0], r1 | |
| 81 | vld1.32 {d19[]}, [r0], r1 | |
| 82 | vld1.32 {d16[1]}, [r0], r1 | |
| 83 | vld1.32 {d17[1]}, [r0], r1 | |
| 84 | vld1.32 {d18[1]}, [r0], r1 | |
| 85 | vld1.32 {d19[1]}, [r0], r1 | |
| 86 | ldrb r2, [r2, #129*4] | |
| 87 | ||
| 88 | vtrn.8 d16, d17 | |
| 89 | vtrn.8 d18, d19 | |
| 90 | vtrn.16 d16, d18 | |
| 91 | vtrn.16 d17, d19 | |
| 92 | ||
| 93 | vp3_loop_filter | |
| 94 | ||
| 95 | vtrn.8 d0, d1 | |
| 96 | ||
| 97 | vst1.16 {d0[0]}, [ip], r1 | |
| 98 | vst1.16 {d1[0]}, [ip], r1 | |
| 99 | vst1.16 {d0[1]}, [ip], r1 | |
| 100 | vst1.16 {d1[1]}, [ip], r1 | |
| 101 | vst1.16 {d0[2]}, [ip], r1 | |
| 102 | vst1.16 {d1[2]}, [ip], r1 | |
| 103 | vst1.16 {d0[3]}, [ip], r1 | |
| 104 | vst1.16 {d1[3]}, [ip], r1 | |
| 105 | bx lr | |
| 106 | endfunc | |
| 107 | ||
| 108 | ||
| 109 | function vp3_idct_start_neon | |
| 110 | vpush {d8-d15} | |
| 111 | vmov.i16 q4, #0 | |
| 112 | vmov.i16 q5, #0 | |
| 113 | movrel r3, vp3_idct_constants | |
| 114 | vld1.64 {d0-d1}, [r3,:128] | |
| 115 | vld1.64 {d16-d19}, [r2,:128] | |
| 116 | vst1.64 {q4-q5}, [r2,:128]! | |
| 117 | vld1.64 {d20-d23}, [r2,:128] | |
| 118 | vst1.64 {q4-q5}, [r2,:128]! | |
| 119 | vld1.64 {d24-d27}, [r2,:128] | |
| 120 | vst1.64 {q4-q5}, [r2,:128]! | |
| 121 | vadd.s16 q1, q8, q12 | |
| 122 | vsub.s16 q8, q8, q12 | |
| 123 | vld1.64 {d28-d31}, [r2,:128] | |
| 124 | vst1.64 {q4-q5}, [r2,:128]! | |
| 125 | ||
| 126 | vp3_idct_core_neon: | |
| 127 | vmull.s16 q2, d18, xC1S7 // (ip[1] * C1) << 16 | |
| 128 | vmull.s16 q3, d19, xC1S7 | |
| 129 | vmull.s16 q4, d2, xC4S4 // ((ip[0] + ip[4]) * C4) << 16 | |
| 130 | vmull.s16 q5, d3, xC4S4 | |
| 131 | vmull.s16 q6, d16, xC4S4 // ((ip[0] - ip[4]) * C4) << 16 | |
| 132 | vmull.s16 q7, d17, xC4S4 | |
| 133 | vshrn.s32 d4, q2, #16 | |
| 134 | vshrn.s32 d5, q3, #16 | |
| 135 | vshrn.s32 d6, q4, #16 | |
| 136 | vshrn.s32 d7, q5, #16 | |
| 137 | vshrn.s32 d8, q6, #16 | |
| 138 | vshrn.s32 d9, q7, #16 | |
| 139 | vadd.s16 q12, q1, q3 // E = (ip[0] + ip[4]) * C4 | |
| 140 | vadd.s16 q8, q8, q4 // F = (ip[0] - ip[4]) * C4 | |
| 141 | vadd.s16 q1, q2, q9 // ip[1] * C1 | |
| 142 | ||
| 143 | vmull.s16 q2, d30, xC1S7 // (ip[7] * C1) << 16 | |
| 144 | vmull.s16 q3, d31, xC1S7 | |
| 145 | vmull.s16 q4, d30, xC7S1 // (ip[7] * C7) << 16 | |
| 146 | vmull.s16 q5, d31, xC7S1 | |
| 147 | vmull.s16 q6, d18, xC7S1 // (ip[1] * C7) << 16 | |
| 148 | vmull.s16 q7, d19, xC7S1 | |
| 149 | vshrn.s32 d4, q2, #16 | |
| 150 | vshrn.s32 d5, q3, #16 | |
| 151 | vshrn.s32 d6, q4, #16 // ip[7] * C7 | |
| 152 | vshrn.s32 d7, q5, #16 | |
| 153 | vshrn.s32 d8, q6, #16 // ip[1] * C7 | |
| 154 | vshrn.s32 d9, q7, #16 | |
| 155 | vadd.s16 q2, q2, q15 // ip[7] * C1 | |
| 156 | vadd.s16 q9, q1, q3 // A = ip[1] * C1 + ip[7] * C7 | |
| 157 | vsub.s16 q15, q4, q2 // B = ip[1] * C7 - ip[7] * C1 | |
| 158 | ||
| 159 | vmull.s16 q2, d22, xC5S3 // (ip[3] * C5) << 16 | |
| 160 | vmull.s16 q3, d23, xC5S3 | |
| 161 | vmull.s16 q4, d22, xC3S5 // (ip[3] * C3) << 16 | |
| 162 | vmull.s16 q5, d23, xC3S5 | |
| 163 | vmull.s16 q6, d26, xC5S3 // (ip[5] * C5) << 16 | |
| 164 | vmull.s16 q7, d27, xC5S3 | |
| 165 | vshrn.s32 d4, q2, #16 | |
| 166 | vshrn.s32 d5, q3, #16 | |
| 167 | vshrn.s32 d6, q4, #16 | |
| 168 | vshrn.s32 d7, q5, #16 | |
| 169 | vshrn.s32 d8, q6, #16 | |
| 170 | vshrn.s32 d9, q7, #16 | |
| 171 | vadd.s16 q3, q3, q11 // ip[3] * C3 | |
| 172 | vadd.s16 q4, q4, q13 // ip[5] * C5 | |
| 173 | vadd.s16 q1, q2, q11 // ip[3] * C5 | |
| 174 | vadd.s16 q11, q3, q4 // C = ip[3] * C3 + ip[5] * C5 | |
| 175 | ||
| 176 | vmull.s16 q2, d26, xC3S5 // (ip[5] * C3) << 16 | |
| 177 | vmull.s16 q3, d27, xC3S5 | |
| 178 | vmull.s16 q4, d20, xC2S6 // (ip[2] * C2) << 16 | |
| 179 | vmull.s16 q5, d21, xC2S6 | |
| 180 | vmull.s16 q6, d28, xC6S2 // (ip[6] * C6) << 16 | |
| 181 | vmull.s16 q7, d29, xC6S2 | |
| 182 | vshrn.s32 d4, q2, #16 | |
| 183 | vshrn.s32 d5, q3, #16 | |
| 184 | vshrn.s32 d6, q4, #16 | |
| 185 | vshrn.s32 d7, q5, #16 | |
| 186 | vshrn.s32 d8, q6, #16 // ip[6] * C6 | |
| 187 | vshrn.s32 d9, q7, #16 | |
| 188 | vadd.s16 q2, q2, q13 // ip[5] * C3 | |
| 189 | vadd.s16 q3, q3, q10 // ip[2] * C2 | |
| 190 | vsub.s16 q13, q2, q1 // D = ip[5] * C3 - ip[3] * C5 | |
| 191 | vsub.s16 q1, q9, q11 // (A - C) | |
| 192 | vadd.s16 q11, q9, q11 // Cd = A + C | |
| 193 | vsub.s16 q9, q15, q13 // (B - D) | |
| 194 | vadd.s16 q13, q15, q13 // Dd = B + D | |
| 195 | vadd.s16 q15, q3, q4 // G = ip[2] * C2 + ip[6] * C6 | |
| 196 | ||
| 197 | vmull.s16 q2, d2, xC4S4 // ((A - C) * C4) << 16 | |
| 198 | vmull.s16 q3, d3, xC4S4 | |
| 199 | vmull.s16 q4, d28, xC2S6 // (ip[6] * C2) << 16 | |
| 200 | vmull.s16 q5, d29, xC2S6 | |
| 201 | vmull.s16 q6, d20, xC6S2 // (ip[2] * C6) << 16 | |
| 202 | vmull.s16 q7, d21, xC6S2 | |
| 203 | vshrn.s32 d4, q2, #16 | |
| 204 | vshrn.s32 d5, q3, #16 | |
| 205 | vshrn.s32 d6, q4, #16 | |
| 206 | vshrn.s32 d7, q5, #16 | |
| 207 | vshrn.s32 d8, q6, #16 // ip[2] * C6 | |
| 208 | vmull.s16 q5, d18, xC4S4 // ((B - D) * C4) << 16 | |
| 209 | vmull.s16 q6, d19, xC4S4 | |
| 210 | vshrn.s32 d9, q7, #16 | |
| 211 | vadd.s16 q3, q3, q14 // ip[6] * C2 | |
| 212 | vadd.s16 q10, q1, q2 // Ad = (A - C) * C4 | |
| 213 | vsub.s16 q14, q4, q3 // H = ip[2] * C6 - ip[6] * C2 | |
| 214 | bx lr | |
| 215 | endfunc | |
| 216 | ||
| 217 | .macro VP3_IDCT_END type | |
| 218 | function vp3_idct_end_\type\()_neon | |
| 219 | .ifc \type, col | |
| 220 | vdup.16 q0, r3 | |
| 221 | vadd.s16 q12, q12, q0 | |
| 222 | vadd.s16 q8, q8, q0 | |
| 223 | .endif | |
| 224 | ||
| 225 | vshrn.s32 d2, q5, #16 | |
| 226 | vshrn.s32 d3, q6, #16 | |
| 227 | vadd.s16 q2, q12, q15 // Gd = E + G | |
| 228 | vadd.s16 q9, q1, q9 // (B - D) * C4 | |
| 229 | vsub.s16 q12, q12, q15 // Ed = E - G | |
| 230 | vsub.s16 q3, q8, q10 // Fd = F - Ad | |
| 231 | vadd.s16 q10, q8, q10 // Add = F + Ad | |
| 232 | vadd.s16 q4, q9, q14 // Hd = Bd + H | |
| 233 | vsub.s16 q14, q9, q14 // Bdd = Bd - H | |
| 234 | vadd.s16 q8, q2, q11 // [0] = Gd + Cd | |
| 235 | vsub.s16 q15, q2, q11 // [7] = Gd - Cd | |
| 236 | vadd.s16 q9, q10, q4 // [1] = Add + Hd | |
| 237 | vsub.s16 q10, q10, q4 // [2] = Add - Hd | |
| 238 | vadd.s16 q11, q12, q13 // [3] = Ed + Dd | |
| 239 | vsub.s16 q12, q12, q13 // [4] = Ed - Dd | |
| 240 | .ifc \type, row | |
| 241 | vtrn.16 q8, q9 | |
| 242 | .endif | |
| 243 | vadd.s16 q13, q3, q14 // [5] = Fd + Bdd | |
| 244 | vsub.s16 q14, q3, q14 // [6] = Fd - Bdd | |
| 245 | ||
| 246 | .ifc \type, row | |
| 247 | // 8x8 transpose | |
| 248 | vtrn.16 q10, q11 | |
| 249 | vtrn.16 q12, q13 | |
| 250 | vtrn.16 q14, q15 | |
| 251 | vtrn.32 q8, q10 | |
| 252 | vtrn.32 q9, q11 | |
| 253 | vtrn.32 q12, q14 | |
| 254 | vtrn.32 q13, q15 | |
| 255 | vswp d17, d24 | |
| 256 | vswp d19, d26 | |
| 257 | vadd.s16 q1, q8, q12 | |
| 258 | vswp d21, d28 | |
| 259 | vsub.s16 q8, q8, q12 | |
| 260 | vswp d23, d30 | |
| 261 | .endif | |
| 262 | bx lr | |
| 263 | endfunc | |
| 264 | .endm | |
| 265 | ||
| 266 | VP3_IDCT_END row | |
| 267 | VP3_IDCT_END col | |
| 268 | ||
| 269 | function ff_vp3_idct_put_neon, export=1 | |
| 270 | mov ip, lr | |
| 271 | bl vp3_idct_start_neon | |
| 272 | bl vp3_idct_end_row_neon | |
| 273 | mov r3, #8 | |
| 274 | add r3, r3, #2048 // convert signed pixel to unsigned | |
| 275 | bl vp3_idct_core_neon | |
| 276 | bl vp3_idct_end_col_neon | |
| 277 | mov lr, ip | |
| 278 | vpop {d8-d15} | |
| 279 | ||
| 280 | vqshrun.s16 d0, q8, #4 | |
| 281 | vqshrun.s16 d1, q9, #4 | |
| 282 | vqshrun.s16 d2, q10, #4 | |
| 283 | vqshrun.s16 d3, q11, #4 | |
| 284 | vst1.64 {d0}, [r0,:64], r1 | |
| 285 | vqshrun.s16 d4, q12, #4 | |
| 286 | vst1.64 {d1}, [r0,:64], r1 | |
| 287 | vqshrun.s16 d5, q13, #4 | |
| 288 | vst1.64 {d2}, [r0,:64], r1 | |
| 289 | vqshrun.s16 d6, q14, #4 | |
| 290 | vst1.64 {d3}, [r0,:64], r1 | |
| 291 | vqshrun.s16 d7, q15, #4 | |
| 292 | vst1.64 {d4}, [r0,:64], r1 | |
| 293 | vst1.64 {d5}, [r0,:64], r1 | |
| 294 | vst1.64 {d6}, [r0,:64], r1 | |
| 295 | vst1.64 {d7}, [r0,:64], r1 | |
| 296 | bx lr | |
| 297 | endfunc | |
| 298 | ||
| 299 | function ff_vp3_idct_add_neon, export=1 | |
| 300 | mov ip, lr | |
| 301 | bl vp3_idct_start_neon | |
| 302 | bl vp3_idct_end_row_neon | |
| 303 | mov r3, #8 | |
| 304 | bl vp3_idct_core_neon | |
| 305 | bl vp3_idct_end_col_neon | |
| 306 | mov lr, ip | |
| 307 | vpop {d8-d15} | |
| 308 | mov r2, r0 | |
| 309 | ||
| 310 | vld1.64 {d0}, [r0,:64], r1 | |
| 311 | vshr.s16 q8, q8, #4 | |
| 312 | vld1.64 {d1}, [r0,:64], r1 | |
| 313 | vshr.s16 q9, q9, #4 | |
| 314 | vld1.64 {d2}, [r0,:64], r1 | |
| 315 | vaddw.u8 q8, q8, d0 | |
| 316 | vld1.64 {d3}, [r0,:64], r1 | |
| 317 | vaddw.u8 q9, q9, d1 | |
| 318 | vld1.64 {d4}, [r0,:64], r1 | |
| 319 | vshr.s16 q10, q10, #4 | |
| 320 | vld1.64 {d5}, [r0,:64], r1 | |
| 321 | vshr.s16 q11, q11, #4 | |
| 322 | vld1.64 {d6}, [r0,:64], r1 | |
| 323 | vqmovun.s16 d0, q8 | |
| 324 | vld1.64 {d7}, [r0,:64], r1 | |
| 325 | vqmovun.s16 d1, q9 | |
| 326 | vaddw.u8 q10, q10, d2 | |
| 327 | vaddw.u8 q11, q11, d3 | |
| 328 | vshr.s16 q12, q12, #4 | |
| 329 | vshr.s16 q13, q13, #4 | |
| 330 | vqmovun.s16 d2, q10 | |
| 331 | vqmovun.s16 d3, q11 | |
| 332 | vaddw.u8 q12, q12, d4 | |
| 333 | vaddw.u8 q13, q13, d5 | |
| 334 | vshr.s16 q14, q14, #4 | |
| 335 | vshr.s16 q15, q15, #4 | |
| 336 | vst1.64 {d0}, [r2,:64], r1 | |
| 337 | vqmovun.s16 d4, q12 | |
| 338 | vst1.64 {d1}, [r2,:64], r1 | |
| 339 | vqmovun.s16 d5, q13 | |
| 340 | vst1.64 {d2}, [r2,:64], r1 | |
| 341 | vaddw.u8 q14, q14, d6 | |
| 342 | vst1.64 {d3}, [r2,:64], r1 | |
| 343 | vaddw.u8 q15, q15, d7 | |
| 344 | vst1.64 {d4}, [r2,:64], r1 | |
| 345 | vqmovun.s16 d6, q14 | |
| 346 | vst1.64 {d5}, [r2,:64], r1 | |
| 347 | vqmovun.s16 d7, q15 | |
| 348 | vst1.64 {d6}, [r2,:64], r1 | |
| 349 | vst1.64 {d7}, [r2,:64], r1 | |
| 350 | bx lr | |
| 351 | endfunc | |
| 352 | ||
| 353 | function ff_vp3_idct_dc_add_neon, export=1 | |
| 354 | ldrsh r12, [r2] | |
| 355 | mov r3, r0 | |
| 356 | add r12, r12, #15 | |
| 357 | vdup.16 q15, r12 | |
| 358 | mov r12, #0 | |
| 359 | strh r12, [r2] | |
| 360 | vshr.s16 q15, q15, #5 | |
| 361 | ||
| 362 | vld1.8 {d0}, [r0,:64], r1 | |
| 363 | vld1.8 {d1}, [r0,:64], r1 | |
| 364 | vld1.8 {d2}, [r0,:64], r1 | |
| 365 | vaddw.u8 q8, q15, d0 | |
| 366 | vld1.8 {d3}, [r0,:64], r1 | |
| 367 | vaddw.u8 q9, q15, d1 | |
| 368 | vld1.8 {d4}, [r0,:64], r1 | |
| 369 | vaddw.u8 q10, q15, d2 | |
| 370 | vld1.8 {d5}, [r0,:64], r1 | |
| 371 | vaddw.u8 q11, q15, d3 | |
| 372 | vld1.8 {d6}, [r0,:64], r1 | |
| 373 | vaddw.u8 q12, q15, d4 | |
| 374 | vld1.8 {d7}, [r0,:64], r1 | |
| 375 | vaddw.u8 q13, q15, d5 | |
| 376 | vqmovun.s16 d0, q8 | |
| 377 | vaddw.u8 q14, q15, d6 | |
| 378 | vqmovun.s16 d1, q9 | |
| 379 | vaddw.u8 q15, q15, d7 | |
| 380 | vqmovun.s16 d2, q10 | |
| 381 | vst1.8 {d0}, [r3,:64], r1 | |
| 382 | vqmovun.s16 d3, q11 | |
| 383 | vst1.8 {d1}, [r3,:64], r1 | |
| 384 | vqmovun.s16 d4, q12 | |
| 385 | vst1.8 {d2}, [r3,:64], r1 | |
| 386 | vqmovun.s16 d5, q13 | |
| 387 | vst1.8 {d3}, [r3,:64], r1 | |
| 388 | vqmovun.s16 d6, q14 | |
| 389 | vst1.8 {d4}, [r3,:64], r1 | |
| 390 | vqmovun.s16 d7, q15 | |
| 391 | vst1.8 {d5}, [r3,:64], r1 | |
| 392 | vst1.8 {d6}, [r3,:64], r1 | |
| 393 | vst1.8 {d7}, [r3,:64], r1 | |
| 394 | bx lr | |
| 395 | endfunc |