2 * Copyright (c) 2013 RISC OS Open Ltd
3 * Author: Ben Avison <bavison@riscosopen.org>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/arm/asm.S"
38 COEF0 .req s8 @ coefficient elements
46 ACCUM0 .req s16 @ double-buffered multiply-accumulate results
48 POST0 .req s24 @ do long-latency post-multiply in this vector in parallel
54 .macro inner_loop decifactor, dir, tail, head
63 vldr COEF0, [PCOEF, #X + (0*JMAX + 0) * Y]
64 vldr COEF1, [PCOEF, #X + (1*JMAX + 0) * Y]
65 vldr COEF2, [PCOEF, #X + (2*JMAX + 0) * Y]
66 vldr COEF3, [PCOEF, #X + (3*JMAX + 0) * Y]
69 vadd.f POST0, ACCUM0, ACCUM4 @ vector operation
72 vmul.f ACCUM0, COEF0, IN0 @ vector = vector * scalar
73 vldr COEF4, [PCOEF, #X + (0*JMAX + 1) * Y]
74 vldr COEF5, [PCOEF, #X + (1*JMAX + 1) * Y]
75 vldr COEF6, [PCOEF, #X + (2*JMAX + 1) * Y]
78 vldr COEF7, [PCOEF, #X + (3*JMAX + 1) * Y]
80 vmul.f ACCUM4, COEF4, IN1 @ vector operation
82 vldr COEF0, [PCOEF, #X + (0*JMAX + 2) * Y]
83 vldr COEF1, [PCOEF, #X + (1*JMAX + 2) * Y]
85 vmul.f ACCUM4, COEF4, IN1 @ vector operation
87 vldr COEF2, [PCOEF, #X + (2*JMAX + 2) * Y]
88 vldr COEF3, [PCOEF, #X + (3*JMAX + 2) * Y]
91 vstmia POUT!, {POST0-POST3}
94 vmla.f ACCUM0, COEF0, IN2 @ vector = vector * scalar
95 vldr COEF4, [PCOEF, #X + (0*JMAX + 3) * Y]
96 vldr COEF5, [PCOEF, #X + (1*JMAX + 3) * Y]
97 vldr COEF6, [PCOEF, #X + (2*JMAX + 3) * Y]
98 vldr COEF7, [PCOEF, #X + (3*JMAX + 3) * Y]
99 vmla.f ACCUM4, COEF4, IN3 @ vector = vector * scalar
100 .if \decifactor == 32
101 vldr COEF0, [PCOEF, #X + (0*JMAX + 4) * Y]
102 vldr COEF1, [PCOEF, #X + (1*JMAX + 4) * Y]
103 vldr COEF2, [PCOEF, #X + (2*JMAX + 4) * Y]
104 vldr COEF3, [PCOEF, #X + (3*JMAX + 4) * Y]
105 vmla.f ACCUM0, COEF0, IN4 @ vector = vector * scalar
106 vldr COEF4, [PCOEF, #X + (0*JMAX + 5) * Y]
107 vldr COEF5, [PCOEF, #X + (1*JMAX + 5) * Y]
108 vldr COEF6, [PCOEF, #X + (2*JMAX + 5) * Y]
109 vldr COEF7, [PCOEF, #X + (3*JMAX + 5) * Y]
110 vmla.f ACCUM4, COEF4, IN5 @ vector = vector * scalar
111 vldr COEF0, [PCOEF, #X + (0*JMAX + 6) * Y]
112 vldr COEF1, [PCOEF, #X + (1*JMAX + 6) * Y]
113 vldr COEF2, [PCOEF, #X + (2*JMAX + 6) * Y]
114 vldr COEF3, [PCOEF, #X + (3*JMAX + 6) * Y]
115 vmla.f ACCUM0, COEF0, IN6 @ vector = vector * scalar
116 vldr COEF4, [PCOEF, #X + (0*JMAX + 7) * Y]
117 vldr COEF5, [PCOEF, #X + (1*JMAX + 7) * Y]
118 vldr COEF6, [PCOEF, #X + (2*JMAX + 7) * Y]
119 vldr COEF7, [PCOEF, #X + (3*JMAX + 7) * Y]
120 vmla.f ACCUM4, COEF4, IN7 @ vector = vector * scalar
125 .macro dca_lfe_fir decifactor
126 function ff_dca_lfe_fir\decifactor\()_vfp, export=1
128 ldr ip, =0x03030000 @ RunFast mode, short vectors of length 4, stride 1
130 vldr IN0, [PIN, #-0*4]
131 vldr IN1, [PIN, #-1*4]
132 vldr IN2, [PIN, #-2*4]
133 vldr IN3, [PIN, #-3*4]
134 .if \decifactor == 32
137 vldr IN4, [PIN, #-4*4]
138 vldr IN5, [PIN, #-5*4]
139 vldr IN6, [PIN, #-6*4]
140 vldr IN7, [PIN, #-7*4]
146 mov COUNTER, #\decifactor/4 - 1
147 inner_loop \decifactor, up,, head
148 1: add PCOEF, PCOEF, #4*JMAX*4
149 subs COUNTER, COUNTER, #1
150 inner_loop \decifactor, up, tail, head
152 inner_loop \decifactor, up, tail
154 mov COUNTER, #\decifactor/4 - 1
155 inner_loop \decifactor, down,, head
156 1: sub PCOEF, PCOEF, #4*JMAX*4
157 subs COUNTER, COUNTER, #1
158 inner_loop \decifactor, down, tail, head
160 inner_loop \decifactor, down, tail
162 .if \decifactor == 32
213 SCALEINT .req v4 @ only used in softfp case
218 /* Stack layout differs in softfp and hardfp cases:
221 * fp -> 6 arg words saved by caller
222 * a3,a4,v1-v3,v5,fp,lr on entry (a3 just to pad to 8 bytes)
225 * buf -> 8*32*4 bytes buffer
227 * sp -> 3 arg words for callee
230 * fp -> 7 arg words saved by caller
231 * a4,v1-v5,fp,lr on entry
234 * buf -> 8*32*4 bytes buffer
235 * sp -> 4 arg words for callee
238 /* void ff_dca_qmf_32_subbands_vfp(float samples_in[32][8], int sb_act,
239 * SynthFilterContext *synth, FFTContext *imdct,
240 * float (*synth_buf_ptr)[512],
241 * int *synth_buf_offset, float (*synth_buf2)[32],
242 * const float (*window)[512], float *samples_out,
243 * float (*raXin)[32], float scale);
245 function ff_dca_qmf_32_subbands_vfp, export=1
246 VFP push {a3-a4,v1-v3,v5,fp,lr}
247 NOVFP push {a4,v1-v5,fp,lr}
250 @ The buffer pointed at by raXin isn't big enough for us to do a
251 @ complete matrix transposition as we want to, so allocate an
252 @ alternative buffer from the stack. Align to 4 words for speed.
256 ldr lr, =0x03330000 @ RunFast mode, short vectors of length 4, stride 2
259 @ COUNT is used to count down 2 things at once:
260 @ bits 0-4 are the number of word pairs remaining in the output row
261 @ bits 5-31 are the number of words to copy (with possible negation)
262 @ from the source matrix before we start zeroing the remainder
263 mov COUNT, #(-4 << 5) + 16
264 adds COUNT, COUNT, SBACT, lsl #5
267 vldr s8, [IN, #(0*8+0)*4]
268 vldr s10, [IN, #(0*8+1)*4]
269 vldr s12, [IN, #(0*8+2)*4]
270 vldr s14, [IN, #(0*8+3)*4]
271 vldr s16, [IN, #(0*8+4)*4]
272 vldr s18, [IN, #(0*8+5)*4]
273 vldr s20, [IN, #(0*8+6)*4]
274 vldr s22, [IN, #(0*8+7)*4]
276 vldr s9, [IN, #(1*8+0)*4]
277 vldr s11, [IN, #(1*8+1)*4]
278 vldr s13, [IN, #(1*8+2)*4]
279 vldr s15, [IN, #(1*8+3)*4]
281 vldr s17, [IN, #(1*8+4)*4]
282 vldr s19, [IN, #(1*8+5)*4]
283 vldr s21, [IN, #(1*8+6)*4]
284 vldr s23, [IN, #(1*8+7)*4]
285 vstr d4, [BUF, #(0*32+0)*4]
286 vstr d5, [BUF, #(1*32+0)*4]
287 vstr d6, [BUF, #(2*32+0)*4]
288 vstr d7, [BUF, #(3*32+0)*4]
289 vstr d8, [BUF, #(4*32+0)*4]
290 vstr d9, [BUF, #(5*32+0)*4]
291 vstr d10, [BUF, #(6*32+0)*4]
292 vstr d11, [BUF, #(7*32+0)*4]
293 vldr s9, [IN, #(3*8+0)*4]
294 vldr s11, [IN, #(3*8+1)*4]
295 vldr s13, [IN, #(3*8+2)*4]
296 vldr s15, [IN, #(3*8+3)*4]
297 vldr s17, [IN, #(3*8+4)*4]
298 vldr s19, [IN, #(3*8+5)*4]
299 vldr s21, [IN, #(3*8+6)*4]
300 vldr s23, [IN, #(3*8+7)*4]
302 vldr s8, [IN, #(2*8+0)*4]
303 vldr s10, [IN, #(2*8+1)*4]
304 vldr s12, [IN, #(2*8+2)*4]
305 vldr s14, [IN, #(2*8+3)*4]
307 vldr s16, [IN, #(2*8+4)*4]
308 vldr s18, [IN, #(2*8+5)*4]
309 vldr s20, [IN, #(2*8+6)*4]
310 vldr s22, [IN, #(2*8+7)*4]
311 vstr d4, [BUF, #(0*32+2)*4]
312 vstr d5, [BUF, #(1*32+2)*4]
313 vstr d6, [BUF, #(2*32+2)*4]
314 vstr d7, [BUF, #(3*32+2)*4]
315 vstr d8, [BUF, #(4*32+2)*4]
316 vstr d9, [BUF, #(5*32+2)*4]
317 vstr d10, [BUF, #(6*32+2)*4]
318 vstr d11, [BUF, #(7*32+2)*4]
321 subs COUNT, COUNT, #(4 << 5) + 2
323 2: @ Now deal with trailing < 4 samples
324 adds COUNT, COUNT, #3 << 5
325 bmi 4f @ sb_act was a multiple of 4
326 bics lr, COUNT, #0x1F
329 vldr s8, [IN, #(0*8+0)*4]
330 vldr s10, [IN, #(0*8+1)*4]
331 vldr s12, [IN, #(0*8+2)*4]
332 vldr s14, [IN, #(0*8+3)*4]
333 vldr s16, [IN, #(0*8+4)*4]
334 vldr s18, [IN, #(0*8+5)*4]
335 vldr s20, [IN, #(0*8+6)*4]
336 vldr s22, [IN, #(0*8+7)*4]
347 vstr d4, [BUF, #(0*32+0)*4]
348 vstr d5, [BUF, #(1*32+0)*4]
349 vstr d6, [BUF, #(2*32+0)*4]
350 vstr d7, [BUF, #(3*32+0)*4]
351 vstr d8, [BUF, #(4*32+0)*4]
352 vstr d9, [BUF, #(5*32+0)*4]
353 vstr d10, [BUF, #(6*32+0)*4]
354 vstr d11, [BUF, #(7*32+0)*4]
358 3: @ sb_act was n*4+2 or n*4+3, so do the first 2
359 vldr s8, [IN, #(0*8+0)*4]
360 vldr s10, [IN, #(0*8+1)*4]
361 vldr s12, [IN, #(0*8+2)*4]
362 vldr s14, [IN, #(0*8+3)*4]
363 vldr s16, [IN, #(0*8+4)*4]
364 vldr s18, [IN, #(0*8+5)*4]
365 vldr s20, [IN, #(0*8+6)*4]
366 vldr s22, [IN, #(0*8+7)*4]
368 vldr s9, [IN, #(1*8+0)*4]
369 vldr s11, [IN, #(1*8+1)*4]
370 vldr s13, [IN, #(1*8+2)*4]
371 vldr s15, [IN, #(1*8+3)*4]
373 vldr s17, [IN, #(1*8+4)*4]
374 vldr s19, [IN, #(1*8+5)*4]
375 vldr s21, [IN, #(1*8+6)*4]
376 vldr s23, [IN, #(1*8+7)*4]
377 vstr d4, [BUF, #(0*32+0)*4]
378 vstr d5, [BUF, #(1*32+0)*4]
379 vstr d6, [BUF, #(2*32+0)*4]
380 vstr d7, [BUF, #(3*32+0)*4]
381 vstr d8, [BUF, #(4*32+0)*4]
382 vstr d9, [BUF, #(5*32+0)*4]
383 vstr d10, [BUF, #(6*32+0)*4]
384 vstr d11, [BUF, #(7*32+0)*4]
386 sub COUNT, COUNT, #(2 << 5) + 1
387 bics lr, COUNT, #0x1F
390 vldr s8, [IN, #(2*8+0)*4]
391 vldr s10, [IN, #(2*8+1)*4]
392 vldr s12, [IN, #(2*8+2)*4]
393 vldr s14, [IN, #(2*8+3)*4]
394 vldr s16, [IN, #(2*8+4)*4]
395 vldr s18, [IN, #(2*8+5)*4]
396 vldr s20, [IN, #(2*8+6)*4]
397 vldr s22, [IN, #(2*8+7)*4]
406 vstr d4, [BUF, #(0*32+0)*4]
407 vstr d5, [BUF, #(1*32+0)*4]
408 vstr d6, [BUF, #(2*32+0)*4]
409 vstr d7, [BUF, #(3*32+0)*4]
410 vstr d8, [BUF, #(4*32+0)*4]
411 vstr d9, [BUF, #(5*32+0)*4]
412 vstr d10, [BUF, #(6*32+0)*4]
413 vstr d11, [BUF, #(7*32+0)*4]
416 4: @ Now fill the remainder with 0
419 ands COUNT, COUNT, #0x1F
421 5: vstr d4, [BUF, #(0*32+0)*4]
422 vstr d4, [BUF, #(1*32+0)*4]
423 vstr d4, [BUF, #(2*32+0)*4]
424 vstr d4, [BUF, #(3*32+0)*4]
425 vstr d4, [BUF, #(4*32+0)*4]
426 vstr d4, [BUF, #(5*32+0)*4]
427 vstr d4, [BUF, #(6*32+0)*4]
428 vstr d4, [BUF, #(7*32+0)*4]
430 subs COUNT, COUNT, #1
434 ldr WINDOW, [fp, #3*4]
437 NOVFP ldr SCALEINT, [fp, #6*4]
441 NOVFP sub sp, sp, #4*4
443 VFP ldr a1, [fp, #-7*4] @ imdct
444 NOVFP ldr a1, [fp, #-8*4]
446 VFP stmia sp, {WINDOW, OUT, BUF}
447 NOVFP stmia sp, {WINDOW, OUT, BUF, SCALEINT}
448 VFP vldr SCALE, [sp, #3*4]
449 bl X(ff_synth_filter_float_vfp)
452 subs COUNT, COUNT, #1
455 A sub sp, fp, #(8+8)*4
456 T sub fp, fp, #(8+8)*4
459 VFP pop {a3-a4,v1-v3,v5,fp,pc}
460 NOVFP pop {a4,v1-v5,fp,pc}