fmtconvert_neon.S 14.5 KB
Newer Older
1 2 3 4
/*
 * ARM NEON optimised Format Conversion Utils
 * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav is free software; you can redistribute it and/or
8 9 10 11
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
12
 * Libav is distributed in the hope that it will be useful,
13 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with Libav; if not, write to the Free Software
19 20 21 22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "config.h"
23
#include "libavutil/arm/asm.S"
24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70

function ff_float_to_int16_neon, export=1
        subs            r2,  r2,  #8
        vld1.64         {d0-d1},  [r1,:128]!
        vcvt.s32.f32    q8,  q0,  #16
        vld1.64         {d2-d3},  [r1,:128]!
        vcvt.s32.f32    q9,  q1,  #16
        beq             3f
        bics            ip,  r2,  #15
        beq             2f
1:      subs            ip,  ip,  #16
        vshrn.s32       d4,  q8,  #16
        vld1.64         {d0-d1},  [r1,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vshrn.s32       d5,  q9,  #16
        vld1.64         {d2-d3},  [r1,:128]!
        vcvt.s32.f32    q1,  q1,  #16
        vshrn.s32       d6,  q0,  #16
        vst1.64         {d4-d5},  [r0,:128]!
        vshrn.s32       d7,  q1,  #16
        vld1.64         {d16-d17},[r1,:128]!
        vcvt.s32.f32    q8,  q8,  #16
        vld1.64         {d18-d19},[r1,:128]!
        vcvt.s32.f32    q9,  q9,  #16
        vst1.64         {d6-d7},  [r0,:128]!
        bne             1b
        ands            r2,  r2,  #15
        beq             3f
2:      vld1.64         {d0-d1},  [r1,:128]!
        vshrn.s32       d4,  q8,  #16
        vcvt.s32.f32    q0,  q0,  #16
        vld1.64         {d2-d3},  [r1,:128]!
        vshrn.s32       d5,  q9,  #16
        vcvt.s32.f32    q1,  q1,  #16
        vshrn.s32       d6,  q0,  #16
        vst1.64         {d4-d5},  [r0,:128]!
        vshrn.s32       d7,  q1,  #16
        vst1.64         {d6-d7},  [r0,:128]!
        bx              lr
3:      vshrn.s32       d4,  q8,  #16
        vshrn.s32       d5,  q9,  #16
        vst1.64         {d4-d5},  [r0,:128]!
        bx              lr
endfunc

function ff_float_to_int16_interleave_neon, export=1
        cmp             r3, #2
71
        itt             lt
72
        ldrlt           r1, [r1]
73
        blt             X(ff_float_to_int16_neon)
74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196
        bne             4f

        ldr             r3, [r1]
        ldr             r1, [r1, #4]

        subs            r2,  r2,  #8
        vld1.64         {d0-d1},  [r3,:128]!
        vcvt.s32.f32    q8,  q0,  #16
        vld1.64         {d2-d3},  [r3,:128]!
        vcvt.s32.f32    q9,  q1,  #16
        vld1.64         {d20-d21},[r1,:128]!
        vcvt.s32.f32    q10, q10, #16
        vld1.64         {d22-d23},[r1,:128]!
        vcvt.s32.f32    q11, q11, #16
        beq             3f
        bics            ip,  r2,  #15
        beq             2f
1:      subs            ip,  ip,  #16
        vld1.64         {d0-d1},  [r3,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vsri.32         q10, q8,  #16
        vld1.64         {d2-d3},  [r3,:128]!
        vcvt.s32.f32    q1,  q1,  #16
        vld1.64         {d24-d25},[r1,:128]!
        vcvt.s32.f32    q12, q12, #16
        vld1.64         {d26-d27},[r1,:128]!
        vsri.32         q11, q9,  #16
        vst1.64         {d20-d21},[r0,:128]!
        vcvt.s32.f32    q13, q13, #16
        vst1.64         {d22-d23},[r0,:128]!
        vsri.32         q12, q0,  #16
        vld1.64         {d16-d17},[r3,:128]!
        vsri.32         q13, q1,  #16
        vst1.64         {d24-d25},[r0,:128]!
        vcvt.s32.f32    q8,  q8,  #16
        vld1.64         {d18-d19},[r3,:128]!
        vcvt.s32.f32    q9,  q9,  #16
        vld1.64         {d20-d21},[r1,:128]!
        vcvt.s32.f32    q10, q10, #16
        vld1.64         {d22-d23},[r1,:128]!
        vcvt.s32.f32    q11, q11, #16
        vst1.64         {d26-d27},[r0,:128]!
        bne             1b
        ands            r2,  r2,  #15
        beq             3f
2:      vsri.32         q10, q8,  #16
        vld1.64         {d0-d1},  [r3,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vld1.64         {d2-d3},  [r3,:128]!
        vcvt.s32.f32    q1,  q1,  #16
        vld1.64         {d24-d25},[r1,:128]!
        vcvt.s32.f32    q12, q12, #16
        vsri.32         q11, q9,  #16
        vld1.64         {d26-d27},[r1,:128]!
        vcvt.s32.f32    q13, q13, #16
        vst1.64         {d20-d21},[r0,:128]!
        vsri.32         q12, q0,  #16
        vst1.64         {d22-d23},[r0,:128]!
        vsri.32         q13, q1,  #16
        vst1.64         {d24-d27},[r0,:128]!
        bx              lr
3:      vsri.32         q10, q8,  #16
        vsri.32         q11, q9,  #16
        vst1.64         {d20-d23},[r0,:128]!
        bx              lr

4:      push            {r4-r8,lr}
        cmp             r3,  #4
        lsl             ip,  r3,  #1
        blt             4f

        @ 4 channels
5:      ldmia           r1!, {r4-r7}
        mov             lr,  r2
        mov             r8,  r0
        vld1.64         {d16-d17},[r4,:128]!
        vcvt.s32.f32    q8,  q8,  #16
        vld1.64         {d18-d19},[r5,:128]!
        vcvt.s32.f32    q9,  q9,  #16
        vld1.64         {d20-d21},[r6,:128]!
        vcvt.s32.f32    q10, q10, #16
        vld1.64         {d22-d23},[r7,:128]!
        vcvt.s32.f32    q11, q11, #16
6:      subs            lr,  lr,  #8
        vld1.64         {d0-d1},  [r4,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vsri.32         q9,  q8,  #16
        vld1.64         {d2-d3},  [r5,:128]!
        vcvt.s32.f32    q1,  q1,  #16
        vsri.32         q11, q10, #16
        vld1.64         {d4-d5},  [r6,:128]!
        vcvt.s32.f32    q2,  q2,  #16
        vzip.32         d18, d22
        vld1.64         {d6-d7},  [r7,:128]!
        vcvt.s32.f32    q3,  q3,  #16
        vzip.32         d19, d23
        vst1.64         {d18},    [r8], ip
        vsri.32         q1,  q0,  #16
        vst1.64         {d22},    [r8], ip
        vsri.32         q3,  q2,  #16
        vst1.64         {d19},    [r8], ip
        vzip.32         d2,  d6
        vst1.64         {d23},    [r8], ip
        vzip.32         d3,  d7
        beq             7f
        vld1.64         {d16-d17},[r4,:128]!
        vcvt.s32.f32    q8,  q8,  #16
        vst1.64         {d2},     [r8], ip
        vld1.64         {d18-d19},[r5,:128]!
        vcvt.s32.f32    q9,  q9,  #16
        vst1.64         {d6},     [r8], ip
        vld1.64         {d20-d21},[r6,:128]!
        vcvt.s32.f32    q10, q10, #16
        vst1.64         {d3},     [r8], ip
        vld1.64         {d22-d23},[r7,:128]!
        vcvt.s32.f32    q11, q11, #16
        vst1.64         {d7},     [r8], ip
        b               6b
7:      vst1.64         {d2},     [r8], ip
        vst1.64         {d6},     [r8], ip
        vst1.64         {d3},     [r8], ip
        vst1.64         {d7},     [r8], ip
        subs            r3,  r3,  #4
197
        it              eq
198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306
        popeq           {r4-r8,pc}
        cmp             r3,  #4
        add             r0,  r0,  #8
        bge             5b

        @ 2 channels
4:      cmp             r3,  #2
        blt             4f
        ldmia           r1!, {r4-r5}
        mov             lr,  r2
        mov             r8,  r0
        tst             lr,  #8
        vld1.64         {d16-d17},[r4,:128]!
        vcvt.s32.f32    q8,  q8,  #16
        vld1.64         {d18-d19},[r5,:128]!
        vcvt.s32.f32    q9,  q9,  #16
        vld1.64         {d20-d21},[r4,:128]!
        vcvt.s32.f32    q10, q10, #16
        vld1.64         {d22-d23},[r5,:128]!
        vcvt.s32.f32    q11, q11, #16
        beq             6f
        subs            lr,  lr,  #8
        beq             7f
        vsri.32         d18, d16, #16
        vsri.32         d19, d17, #16
        vld1.64         {d16-d17},[r4,:128]!
        vcvt.s32.f32    q8,  q8,  #16
        vst1.32         {d18[0]}, [r8], ip
        vsri.32         d22, d20, #16
        vst1.32         {d18[1]}, [r8], ip
        vsri.32         d23, d21, #16
        vst1.32         {d19[0]}, [r8], ip
        vst1.32         {d19[1]}, [r8], ip
        vld1.64         {d18-d19},[r5,:128]!
        vcvt.s32.f32    q9,  q9,  #16
        vst1.32         {d22[0]}, [r8], ip
        vst1.32         {d22[1]}, [r8], ip
        vld1.64         {d20-d21},[r4,:128]!
        vcvt.s32.f32    q10, q10, #16
        vst1.32         {d23[0]}, [r8], ip
        vst1.32         {d23[1]}, [r8], ip
        vld1.64         {d22-d23},[r5,:128]!
        vcvt.s32.f32    q11, q11, #16
6:      subs            lr,  lr,  #16
        vld1.64         {d0-d1},  [r4,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vsri.32         d18, d16, #16
        vld1.64         {d2-d3},  [r5,:128]!
        vcvt.s32.f32    q1,  q1,  #16
        vsri.32         d19, d17, #16
        vld1.64         {d4-d5},  [r4,:128]!
        vcvt.s32.f32    q2,  q2,  #16
        vld1.64         {d6-d7},  [r5,:128]!
        vcvt.s32.f32    q3,  q3,  #16
        vst1.32         {d18[0]}, [r8], ip
        vsri.32         d22, d20, #16
        vst1.32         {d18[1]}, [r8], ip
        vsri.32         d23, d21, #16
        vst1.32         {d19[0]}, [r8], ip
        vsri.32         d2,  d0,  #16
        vst1.32         {d19[1]}, [r8], ip
        vsri.32         d3,  d1,  #16
        vst1.32         {d22[0]}, [r8], ip
        vsri.32         d6,  d4,  #16
        vst1.32         {d22[1]}, [r8], ip
        vsri.32         d7,  d5,  #16
        vst1.32         {d23[0]}, [r8], ip
        vst1.32         {d23[1]}, [r8], ip
        beq             6f
        vld1.64         {d16-d17},[r4,:128]!
        vcvt.s32.f32    q8,  q8,  #16
        vst1.32         {d2[0]},  [r8], ip
        vst1.32         {d2[1]},  [r8], ip
        vld1.64         {d18-d19},[r5,:128]!
        vcvt.s32.f32    q9,  q9,  #16
        vst1.32         {d3[0]},  [r8], ip
        vst1.32         {d3[1]},  [r8], ip
        vld1.64         {d20-d21},[r4,:128]!
        vcvt.s32.f32    q10, q10, #16
        vst1.32         {d6[0]},  [r8], ip
        vst1.32         {d6[1]},  [r8], ip
        vld1.64         {d22-d23},[r5,:128]!
        vcvt.s32.f32    q11, q11, #16
        vst1.32         {d7[0]},  [r8], ip
        vst1.32         {d7[1]},  [r8], ip
        bgt             6b
6:      vst1.32         {d2[0]},  [r8], ip
        vst1.32         {d2[1]},  [r8], ip
        vst1.32         {d3[0]},  [r8], ip
        vst1.32         {d3[1]},  [r8], ip
        vst1.32         {d6[0]},  [r8], ip
        vst1.32         {d6[1]},  [r8], ip
        vst1.32         {d7[0]},  [r8], ip
        vst1.32         {d7[1]},  [r8], ip
        b               8f
7:      vsri.32         d18, d16, #16
        vsri.32         d19, d17, #16
        vst1.32         {d18[0]}, [r8], ip
        vsri.32         d22, d20, #16
        vst1.32         {d18[1]}, [r8], ip
        vsri.32         d23, d21, #16
        vst1.32         {d19[0]}, [r8], ip
        vst1.32         {d19[1]}, [r8], ip
        vst1.32         {d22[0]}, [r8], ip
        vst1.32         {d22[1]}, [r8], ip
        vst1.32         {d23[0]}, [r8], ip
        vst1.32         {d23[1]}, [r8], ip
8:      subs            r3,  r3,  #2
        add             r0,  r0,  #4
307
        it              eq
308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
        popeq           {r4-r8,pc}

        @ 1 channel
4:      ldr             r4,  [r1],#4
        tst             r2,  #8
        mov             lr,  r2
        mov             r5,  r0
        vld1.64         {d0-d1},  [r4,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vld1.64         {d2-d3},  [r4,:128]!
        vcvt.s32.f32    q1,  q1,  #16
        bne             8f
6:      subs            lr,  lr,  #16
        vld1.64         {d4-d5},  [r4,:128]!
        vcvt.s32.f32    q2,  q2,  #16
        vld1.64         {d6-d7},  [r4,:128]!
        vcvt.s32.f32    q3,  q3,  #16
        vst1.16         {d0[1]},  [r5,:16], ip
        vst1.16         {d0[3]},  [r5,:16], ip
        vst1.16         {d1[1]},  [r5,:16], ip
        vst1.16         {d1[3]},  [r5,:16], ip
        vst1.16         {d2[1]},  [r5,:16], ip
        vst1.16         {d2[3]},  [r5,:16], ip
        vst1.16         {d3[1]},  [r5,:16], ip
        vst1.16         {d3[3]},  [r5,:16], ip
        beq             7f
        vld1.64         {d0-d1},  [r4,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vld1.64         {d2-d3},  [r4,:128]!
        vcvt.s32.f32    q1,  q1,  #16
7:      vst1.16         {d4[1]},  [r5,:16], ip
        vst1.16         {d4[3]},  [r5,:16], ip
        vst1.16         {d5[1]},  [r5,:16], ip
        vst1.16         {d5[3]},  [r5,:16], ip
        vst1.16         {d6[1]},  [r5,:16], ip
        vst1.16         {d6[3]},  [r5,:16], ip
        vst1.16         {d7[1]},  [r5,:16], ip
        vst1.16         {d7[3]},  [r5,:16], ip
        bgt             6b
        pop             {r4-r8,pc}
8:      subs            lr,  lr,  #8
        vst1.16         {d0[1]},  [r5,:16], ip
        vst1.16         {d0[3]},  [r5,:16], ip
        vst1.16         {d1[1]},  [r5,:16], ip
        vst1.16         {d1[3]},  [r5,:16], ip
        vst1.16         {d2[1]},  [r5,:16], ip
        vst1.16         {d2[3]},  [r5,:16], ip
        vst1.16         {d3[1]},  [r5,:16], ip
        vst1.16         {d3[3]},  [r5,:16], ip
357
        it              eq
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
        popeq           {r4-r8,pc}
        vld1.64         {d0-d1},  [r4,:128]!
        vcvt.s32.f32    q0,  q0,  #16
        vld1.64         {d2-d3},  [r4,:128]!
        vcvt.s32.f32    q1,  q1,  #16
        b               6b
endfunc

function ff_int32_to_float_fmul_scalar_neon, export=1
VFP     vdup.32         q0,  d0[0]
VFP     len     .req    r2
NOVFP   vdup.32         q0,  r2
NOVFP   len     .req    r3

        vld1.32         {q1},[r1,:128]!
        vcvt.f32.s32    q3,  q1
        vld1.32         {q2},[r1,:128]!
        vcvt.f32.s32    q8,  q2
1:      subs            len, len, #8
        pld             [r1, #16]
        vmul.f32        q9,  q3,  q0
        vmul.f32        q10, q8,  q0
        beq             2f
        vld1.32         {q1},[r1,:128]!
        vcvt.f32.s32    q3,  q1
        vld1.32         {q2},[r1,:128]!
        vcvt.f32.s32    q8,  q2
        vst1.32         {q9}, [r0,:128]!
        vst1.32         {q10},[r0,:128]!
        b               1b
2:      vst1.32         {q9}, [r0,:128]!
        vst1.32         {q10},[r0,:128]!
        bx              lr
        .unreq  len
endfunc