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"
30 SCALE32 .req s28 @ use vector of 4 in place of 9th scalar when decifactor=32 / JMAX=8
31 SCALE64 .req s0 @ spare register in scalar bank when decifactor=64 / JMAX=4
40 COEF0 .req s8 @ coefficient elements
48 ACCUM0 .req s16 @ double-buffered multiply-accumulate results
50 POST0 .req s24 @ do long-latency post-multiply in this vector in parallel
56 .macro inner_loop decifactor, dir, tail, head
65 vldr COEF0, [PCOEF, #X + (0*JMAX + 0) * Y]
66 vldr COEF1, [PCOEF, #X + (1*JMAX + 0) * Y]
67 vldr COEF2, [PCOEF, #X + (2*JMAX + 0) * Y]
68 vldr COEF3, [PCOEF, #X + (3*JMAX + 0) * Y]
71 vadd.f POST0, ACCUM0, ACCUM4 @ vector operation
74 vmul.f ACCUM0, COEF0, IN0 @ vector = vector * scalar
75 vldr COEF4, [PCOEF, #X + (0*JMAX + 1) * Y]
76 vldr COEF5, [PCOEF, #X + (1*JMAX + 1) * Y]
77 vldr COEF6, [PCOEF, #X + (2*JMAX + 1) * Y]
80 vmul.f POST0, POST0, SCALE\decifactor @ vector operation (SCALE may be scalar)
83 vldr COEF7, [PCOEF, #X + (3*JMAX + 1) * Y]
85 vmul.f ACCUM4, COEF4, IN1 @ vector operation
87 vldr COEF0, [PCOEF, #X + (0*JMAX + 2) * Y]
88 vldr COEF1, [PCOEF, #X + (1*JMAX + 2) * Y]
90 vmul.f ACCUM4, COEF4, IN1 @ vector operation
92 vldr COEF2, [PCOEF, #X + (2*JMAX + 2) * Y]
93 vldr COEF3, [PCOEF, #X + (3*JMAX + 2) * Y]
96 vstmia POUT!, {POST0-POST3}
99 vmla.f ACCUM0, COEF0, IN2 @ vector = vector * scalar
100 vldr COEF4, [PCOEF, #X + (0*JMAX + 3) * Y]
101 vldr COEF5, [PCOEF, #X + (1*JMAX + 3) * Y]
102 vldr COEF6, [PCOEF, #X + (2*JMAX + 3) * Y]
103 vldr COEF7, [PCOEF, #X + (3*JMAX + 3) * Y]
104 vmla.f ACCUM4, COEF4, IN3 @ vector = vector * scalar
105 .if \decifactor == 32
106 vldr COEF0, [PCOEF, #X + (0*JMAX + 4) * Y]
107 vldr COEF1, [PCOEF, #X + (1*JMAX + 4) * Y]
108 vldr COEF2, [PCOEF, #X + (2*JMAX + 4) * Y]
109 vldr COEF3, [PCOEF, #X + (3*JMAX + 4) * Y]
110 vmla.f ACCUM0, COEF0, IN4 @ vector = vector * scalar
111 vldr COEF4, [PCOEF, #X + (0*JMAX + 5) * Y]
112 vldr COEF5, [PCOEF, #X + (1*JMAX + 5) * Y]
113 vldr COEF6, [PCOEF, #X + (2*JMAX + 5) * Y]
114 vldr COEF7, [PCOEF, #X + (3*JMAX + 5) * Y]
115 vmla.f ACCUM4, COEF4, IN5 @ vector = vector * scalar
116 vldr COEF0, [PCOEF, #X + (0*JMAX + 6) * Y]
117 vldr COEF1, [PCOEF, #X + (1*JMAX + 6) * Y]
118 vldr COEF2, [PCOEF, #X + (2*JMAX + 6) * Y]
119 vldr COEF3, [PCOEF, #X + (3*JMAX + 6) * Y]
120 vmla.f ACCUM0, COEF0, IN6 @ vector = vector * scalar
121 vldr COEF4, [PCOEF, #X + (0*JMAX + 7) * Y]
122 vldr COEF5, [PCOEF, #X + (1*JMAX + 7) * Y]
123 vldr COEF6, [PCOEF, #X + (2*JMAX + 7) * Y]
124 vldr COEF7, [PCOEF, #X + (3*JMAX + 7) * Y]
125 vmla.f ACCUM4, COEF4, IN7 @ vector = vector * scalar
130 .macro dca_lfe_fir decifactor
131 function ff_dca_lfe_fir\decifactor\()_vfp, export=1
134 ldr ip, =0x03030000 @ RunFast mode, short vectors of length 4, stride 1
136 vldr IN0, [PIN, #-0*4]
137 vldr IN1, [PIN, #-1*4]
138 vldr IN2, [PIN, #-2*4]
139 vldr IN3, [PIN, #-3*4]
140 .if \decifactor == 32
143 vmov SCALE32, s0 @ duplicate scalar across vector
144 vldr IN4, [PIN, #-4*4]
145 vldr IN5, [PIN, #-5*4]
146 vldr IN6, [PIN, #-6*4]
147 vldr IN7, [PIN, #-7*4]
153 mov COUNTER, #\decifactor/4 - 1
154 inner_loop \decifactor, up,, head
155 1: add PCOEF, PCOEF, #4*JMAX*4
156 subs COUNTER, COUNTER, #1
157 inner_loop \decifactor, up, tail, head
159 inner_loop \decifactor, up, tail
161 mov COUNTER, #\decifactor/4 - 1
162 inner_loop \decifactor, down,, head
163 1: sub PCOEF, PCOEF, #4*JMAX*4
164 subs COUNTER, COUNTER, #1
165 inner_loop \decifactor, down, tail, head
167 inner_loop \decifactor, down, tail
169 .if \decifactor == 32
222 SCALEINT .req v4 @ only used in softfp case
227 /* Stack layout differs in softfp and hardfp cases:
230 * fp -> 6 arg words saved by caller
231 * a3,a4,v1-v3,v5,fp,lr on entry (a3 just to pad to 8 bytes)
234 * buf -> 8*32*4 bytes buffer
236 * sp -> 3 arg words for callee
239 * fp -> 7 arg words saved by caller
240 * a4,v1-v5,fp,lr on entry
243 * buf -> 8*32*4 bytes buffer
244 * sp -> 4 arg words for callee
247 /* void ff_dca_qmf_32_subbands_vfp(float samples_in[32][8], int sb_act,
248 * SynthFilterContext *synth, FFTContext *imdct,
249 * float (*synth_buf_ptr)[512],
250 * int *synth_buf_offset, float (*synth_buf2)[32],
251 * const float (*window)[512], float *samples_out,
252 * float (*raXin)[32], float scale);
254 function ff_dca_qmf_32_subbands_vfp, export=1
255 VFP push {a3-a4,v1-v3,v5,fp,lr}
256 NOVFP push {a4,v1-v5,fp,lr}
259 @ The buffer pointed at by raXin isn't big enough for us to do a
260 @ complete matrix transposition as we want to, so allocate an
261 @ alternative buffer from the stack. Align to 4 words for speed.
265 ldr lr, =0x03330000 @ RunFast mode, short vectors of length 4, stride 2
268 @ COUNT is used to count down 2 things at once:
269 @ bits 0-4 are the number of word pairs remaining in the output row
270 @ bits 5-31 are the number of words to copy (with possible negation)
271 @ from the source matrix before we start zeroing the remainder
272 mov COUNT, #(-4 << 5) + 16
273 adds COUNT, COUNT, SBACT, lsl #5
276 vldr s8, [IN, #(0*8+0)*4]
277 vldr s10, [IN, #(0*8+1)*4]
278 vldr s12, [IN, #(0*8+2)*4]
279 vldr s14, [IN, #(0*8+3)*4]
280 vldr s16, [IN, #(0*8+4)*4]
281 vldr s18, [IN, #(0*8+5)*4]
282 vldr s20, [IN, #(0*8+6)*4]
283 vldr s22, [IN, #(0*8+7)*4]
285 vldr s9, [IN, #(1*8+0)*4]
286 vldr s11, [IN, #(1*8+1)*4]
287 vldr s13, [IN, #(1*8+2)*4]
288 vldr s15, [IN, #(1*8+3)*4]
290 vldr s17, [IN, #(1*8+4)*4]
291 vldr s19, [IN, #(1*8+5)*4]
292 vldr s21, [IN, #(1*8+6)*4]
293 vldr s23, [IN, #(1*8+7)*4]
294 vstr d4, [BUF, #(0*32+0)*4]
295 vstr d5, [BUF, #(1*32+0)*4]
296 vstr d6, [BUF, #(2*32+0)*4]
297 vstr d7, [BUF, #(3*32+0)*4]
298 vstr d8, [BUF, #(4*32+0)*4]
299 vstr d9, [BUF, #(5*32+0)*4]
300 vstr d10, [BUF, #(6*32+0)*4]
301 vstr d11, [BUF, #(7*32+0)*4]
302 vldr s9, [IN, #(3*8+0)*4]
303 vldr s11, [IN, #(3*8+1)*4]
304 vldr s13, [IN, #(3*8+2)*4]
305 vldr s15, [IN, #(3*8+3)*4]
306 vldr s17, [IN, #(3*8+4)*4]
307 vldr s19, [IN, #(3*8+5)*4]
308 vldr s21, [IN, #(3*8+6)*4]
309 vldr s23, [IN, #(3*8+7)*4]
311 vldr s8, [IN, #(2*8+0)*4]
312 vldr s10, [IN, #(2*8+1)*4]
313 vldr s12, [IN, #(2*8+2)*4]
314 vldr s14, [IN, #(2*8+3)*4]
316 vldr s16, [IN, #(2*8+4)*4]
317 vldr s18, [IN, #(2*8+5)*4]
318 vldr s20, [IN, #(2*8+6)*4]
319 vldr s22, [IN, #(2*8+7)*4]
320 vstr d4, [BUF, #(0*32+2)*4]
321 vstr d5, [BUF, #(1*32+2)*4]
322 vstr d6, [BUF, #(2*32+2)*4]
323 vstr d7, [BUF, #(3*32+2)*4]
324 vstr d8, [BUF, #(4*32+2)*4]
325 vstr d9, [BUF, #(5*32+2)*4]
326 vstr d10, [BUF, #(6*32+2)*4]
327 vstr d11, [BUF, #(7*32+2)*4]
330 subs COUNT, COUNT, #(4 << 5) + 2
332 2: @ Now deal with trailing < 4 samples
333 adds COUNT, COUNT, #3 << 5
334 bmi 4f @ sb_act was a multiple of 4
335 bics lr, COUNT, #0x1F
338 vldr s8, [IN, #(0*8+0)*4]
339 vldr s10, [IN, #(0*8+1)*4]
340 vldr s12, [IN, #(0*8+2)*4]
341 vldr s14, [IN, #(0*8+3)*4]
342 vldr s16, [IN, #(0*8+4)*4]
343 vldr s18, [IN, #(0*8+5)*4]
344 vldr s20, [IN, #(0*8+6)*4]
345 vldr s22, [IN, #(0*8+7)*4]
356 vstr d4, [BUF, #(0*32+0)*4]
357 vstr d5, [BUF, #(1*32+0)*4]
358 vstr d6, [BUF, #(2*32+0)*4]
359 vstr d7, [BUF, #(3*32+0)*4]
360 vstr d8, [BUF, #(4*32+0)*4]
361 vstr d9, [BUF, #(5*32+0)*4]
362 vstr d10, [BUF, #(6*32+0)*4]
363 vstr d11, [BUF, #(7*32+0)*4]
367 3: @ sb_act was n*4+2 or n*4+3, so do the first 2
368 vldr s8, [IN, #(0*8+0)*4]
369 vldr s10, [IN, #(0*8+1)*4]
370 vldr s12, [IN, #(0*8+2)*4]
371 vldr s14, [IN, #(0*8+3)*4]
372 vldr s16, [IN, #(0*8+4)*4]
373 vldr s18, [IN, #(0*8+5)*4]
374 vldr s20, [IN, #(0*8+6)*4]
375 vldr s22, [IN, #(0*8+7)*4]
377 vldr s9, [IN, #(1*8+0)*4]
378 vldr s11, [IN, #(1*8+1)*4]
379 vldr s13, [IN, #(1*8+2)*4]
380 vldr s15, [IN, #(1*8+3)*4]
382 vldr s17, [IN, #(1*8+4)*4]
383 vldr s19, [IN, #(1*8+5)*4]
384 vldr s21, [IN, #(1*8+6)*4]
385 vldr s23, [IN, #(1*8+7)*4]
386 vstr d4, [BUF, #(0*32+0)*4]
387 vstr d5, [BUF, #(1*32+0)*4]
388 vstr d6, [BUF, #(2*32+0)*4]
389 vstr d7, [BUF, #(3*32+0)*4]
390 vstr d8, [BUF, #(4*32+0)*4]
391 vstr d9, [BUF, #(5*32+0)*4]
392 vstr d10, [BUF, #(6*32+0)*4]
393 vstr d11, [BUF, #(7*32+0)*4]
395 sub COUNT, COUNT, #(2 << 5) + 1
396 bics lr, COUNT, #0x1F
399 vldr s8, [IN, #(2*8+0)*4]
400 vldr s10, [IN, #(2*8+1)*4]
401 vldr s12, [IN, #(2*8+2)*4]
402 vldr s14, [IN, #(2*8+3)*4]
403 vldr s16, [IN, #(2*8+4)*4]
404 vldr s18, [IN, #(2*8+5)*4]
405 vldr s20, [IN, #(2*8+6)*4]
406 vldr s22, [IN, #(2*8+7)*4]
415 vstr d4, [BUF, #(0*32+0)*4]
416 vstr d5, [BUF, #(1*32+0)*4]
417 vstr d6, [BUF, #(2*32+0)*4]
418 vstr d7, [BUF, #(3*32+0)*4]
419 vstr d8, [BUF, #(4*32+0)*4]
420 vstr d9, [BUF, #(5*32+0)*4]
421 vstr d10, [BUF, #(6*32+0)*4]
422 vstr d11, [BUF, #(7*32+0)*4]
425 4: @ Now fill the remainder with 0
428 ands COUNT, COUNT, #0x1F
430 5: vstr d4, [BUF, #(0*32+0)*4]
431 vstr d4, [BUF, #(1*32+0)*4]
432 vstr d4, [BUF, #(2*32+0)*4]
433 vstr d4, [BUF, #(3*32+0)*4]
434 vstr d4, [BUF, #(4*32+0)*4]
435 vstr d4, [BUF, #(5*32+0)*4]
436 vstr d4, [BUF, #(6*32+0)*4]
437 vstr d4, [BUF, #(7*32+0)*4]
439 subs COUNT, COUNT, #1
443 ldr WINDOW, [fp, #3*4]
446 NOVFP ldr SCALEINT, [fp, #6*4]
450 NOVFP sub sp, sp, #4*4
452 VFP ldr a1, [fp, #-7*4] @ imdct
453 NOVFP ldr a1, [fp, #-8*4]
455 VFP stmia sp, {WINDOW, OUT, BUF}
456 NOVFP stmia sp, {WINDOW, OUT, BUF, SCALEINT}
457 VFP vldr SCALE, [sp, #3*4]
458 bl X(ff_synth_filter_float_vfp)
461 subs COUNT, COUNT, #1
464 A sub sp, fp, #(8+8)*4
465 T sub fp, fp, #(8+8)*4
468 VFP pop {a3-a4,v1-v3,v5,fp,pc}
469 NOVFP pop {a4,v1-v5,fp,pc}