1 ;******************************************************************************
2 ;* AAC Spectral Band Replication decoding functions
3 ;* Copyright (C) 2012 Christophe Gisquet <christophe.gisquet@gmail.com>
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
20 ;******************************************************************************
22 %include "libavutil/x86/x86util.asm"
25 ; mask equivalent for multiply by -1.0 1.0
26 ps_mask times 2 dd 1<<31, 0
27 ps_mask2 times 2 dd 0, 1<<31
28 ps_mask3 dd 0, 0, 0, 1<<31
29 ps_noise0 times 2 dd 1.0, 0.0,
30 ps_noise2 times 2 dd -1.0, 0.0
31 ps_noise13 dd 0.0, 1.0, 0.0, -1.0
32 dd 0.0, -1.0, 0.0, 1.0
33 dd 0.0, 1.0, 0.0, -1.0
34 cextern sbr_noise_table
40 cglobal sbr_sum_square, 2, 3, 6
66 ; len is a multiple of 2, thus there are at least 4 elements to process
88 cglobal sbr_hf_g_filt, 5, 6, 5
89 lea r1, [r1 + 8*r4] ; offset by ixh elements into X_high
97 movlps m0, [r2 + 4*r3 + 0]
98 movlps m1, [r2 + 4*r3 + 8]
99 movlps m2, [r1 + 0*STEP]
100 movlps m3, [r1 + 2*STEP]
101 movhps m2, [r1 + 1*STEP]
102 movhps m3, [r1 + 3*STEP]
107 movu [r0 + 8*r3 + 0], m0
108 movu [r0 + 8*r3 + 16], m1
112 and r5, 3 ; number of single element loops
114 .loop1: ; element 0 and 1 can be computed at the same time
128 ; void ff_sbr_hf_gen_sse(float (*X_high)[2], const float (*X_low)[2],
129 ; const float alpha0[2], const float alpha1[2],
130 ; float bw, int start, int end)
132 cglobal sbr_hf_gen, 4,4,8, X_high, X_low, alpha0, alpha1, BW, S, E
135 %if ARCH_X86_64 == 0 || WIN64
141 mulps m2, bw ; (a1[0] a1[1])*bw
142 mulps m1, bw ; (a0[0] a0[1])*bw = (a2 a3)
143 mulps m2, bw ; (a1[0] a1[1])*bw*bw = (a0 a1)
148 %if ARCH_X86_64 == 0 || WIN64
149 ; start and end 6th and 7th args on stack
155 ; BW does not actually occupy a register, so shift by 1
159 sub start, end ; neg num of loops
160 lea X_highq, [X_highq + end*2*4]
161 lea X_lowq, [X_lowq + end*2*4 - 2*2*4]
162 shl start, 3 ; offset from num loops
164 mova m0, [X_lowq + start]
172 movu m7, [X_lowq + start + 8] ; BbCc
175 shufps m0, m0, q2301 ; aAbB
176 shufps m7, m7, q2301 ; bBcC
182 mova m0, [X_lowq + start +16] ; CcDd
186 mova [X_highq + start], m7
191 cglobal sbr_sum64x5, 1,2,4,z
214 cglobal sbr_qmf_post_shuffle, 2,3,4,W,z
215 lea r2q, [zq + (64-4)*4]
220 shufps m0, m0, m0, q0123
233 cglobal sbr_neg_odd_64, 1,2,4,z
253 ; void ff_sbr_qmf_deint_bfly_sse2(float *v, const float *src0, const float *src1)
254 %macro SBR_QMF_DEINT_BFLY 0
255 cglobal sbr_qmf_deint_bfly, 3,5,8, v,src0,src1,vrev,c
256 mov cq, 64*4-2*mmsize
257 lea vrevq, [vq + 64*4]
261 mova m4, [src0q+cq+mmsize]
262 mova m5, [src1q+mmsize]
269 shufps m2, m0, m0, q0123
270 shufps m3, m1, m1, q0123
271 shufps m6, m4, m4, q0123
272 shufps m7, m5, m5, q0123
279 mova [vrevq+mmsize], m5
281 mova [vq+cq+mmsize], m4
296 cglobal sbr_qmf_pre_shuffle, 1,4,6,z
297 %define OFFSET (32*4-2*mmsize)
299 lea r1q, [zq + (32+1)*4]
304 movu m2, [r1q + mmsize]
305 movu m1, [zq + r3q + 4 + mmsize]
306 movu m3, [zq + r3q + 4]
312 SBUTTERFLY dq, 2, 3, 4
313 SBUTTERFLY dq, 0, 1, 4
314 mova [r2q + 2*r3q + 0*mmsize], m2
315 mova [r2q + 2*r3q + 1*mmsize], m3
316 mova [r2q + 2*r3q + 2*mmsize], m0
317 mova [r2q + 2*r3q + 3*mmsize], m1
328 %define NOISE_TABLE r6q ; r5q is m_max
330 %define NOISE_TABLE r5q
334 %define NOISE_TABLE sbr_noise_table
339 lea NOISE_TABLE, [%1]
340 mova m0, [kxq + NOISE_TABLE]
347 ; sbr_hf_apply_noise_0(float (*Y)[2], const float *s_m,
348 ; const float *q_filt, int noise,
350 cglobal sbr_hf_apply_noise_0, 5,5+NREGS+UNIX64,8, Y,s_m,q_filt,noise,kx,m_max
354 ; sbr_hf_apply_noise_1(float (*Y)[2], const float *s_m,
355 ; const float *q_filt, int noise,
357 cglobal sbr_hf_apply_noise_1, 5,5+NREGS+UNIX64,8, Y,s_m,q_filt,noise,kx,m_max
363 ; sbr_hf_apply_noise_2(float (*Y)[2], const float *s_m,
364 ; const float *q_filt, int noise,
366 cglobal sbr_hf_apply_noise_2, 5,5+NREGS+UNIX64,8, Y,s_m,q_filt,noise,kx,m_max
370 ; sbr_hf_apply_noise_3(float (*Y)[2], const float *s_m,
371 ; const float *q_filt, int noise,
373 cglobal sbr_hf_apply_noise_3, 5,5+NREGS+UNIX64,8, Y,s_m,q_filt,noise,kx,m_max
376 LOAD_NST ps_noise13+16
379 %if ARCH_X86_64 == 0 || WIN64
385 movsxdifnidn noiseq, noised
389 lea NOISE_TABLE, [sbr_noise_table]
391 lea Yq, [Yq + 2*count]
398 mova m1, [q_filtq + count]
399 movu m3, [noiseq + NOISE_TABLE + 1*mmsize]
400 movu m4, [noiseq + NOISE_TABLE + 2*mmsize]
405 mulps m1, m3 ; m2 = q_filt[m] * ff_sbr_noise_table[noise]
406 mulps m2, m4 ; m2 = q_filt[m] * ff_sbr_noise_table[noise]
407 mova m3, [s_mq + count]
408 ; TODO: replace by a vpermd in AVX2
411 pcmpeqd m6, m3, m5 ; m6 == 0
412 pcmpeqd m7, m4, m5 ; m7 == 0
413 mulps m3, m0 ; s_m[m] * phi_sign
414 mulps m4, m0 ; s_m[m] * phi_sign
417 movu m6, [Yq + 2*count]
418 movu m7, [Yq + 2*count + mmsize]
423 movu [Yq + 2*count], m6
424 movu [Yq + 2*count + mmsize], m7
430 cglobal sbr_qmf_deint_neg, 2,4,4,v,src,vrev,c
434 lea vrevq, [vq + OFFSET + COUNT]
435 add vq, OFFSET-mmsize
439 mova m0, [srcq + 2*cq + 0*mmsize]
440 mova m1, [srcq + 2*cq + 1*mmsize]
441 shufps m2, m0, m1, q2020
445 mova [vrevq + cq], m2
451 %macro SBR_AUTOCORRELATE 0
452 cglobal sbr_autocorrelate, 2,3,8,32, x, phi, cnt
459 movddup m5, [xq+cntq]
465 MOVH m7, [xq+cntq+8 ]
466 MOVH m1, [xq+cntq+16]
469 mulps m3, m5, m7 ; x[0][0] * x[1][0], x[0][1] * x[1][1], x[0][0] * x[1][1], x[0][1] * x[1][0]
470 mulps m4, m5, m5 ; x[0][0] * x[0][0], x[0][1] * x[0][1];
471 mulps m5, m1 ; real_sum2 = x[0][0] * x[2][0], x[0][1] * x[2][1]; imag_sum2 = x[0][0] * x[2][1], x[0][1] * x[2][0]
476 MOVH m2, [xq+cntq+16]
479 mulps m6, m7, m1 ; real_sum1 = x[1][0] * x[2][0], x[1][1] * x[2][1]; imag_sum1 += x[1][0] * x[2][1], x[1][1] * x[2][0]
481 mulps m7, m7 ; real_sum0 = x[1][0] * x[1][0], x[1][1] * x[1][1];
482 addps m5, m4 ; real_sum2 += x[1][0] * x[3][0], x[1][1] * x[3][1]; imag_sum2 += x[1][0] * x[3][1], x[1][1] * x[3][0]
487 MOVH m0, [xq+cntq+16]
493 addps m6, m3 ; real_sum1 += x[i][0] * x[i + 1][0], x[i][1] * x[i + 1][1]; imag_sum1 += x[i][0] * x[i + 1][1], x[i][1] * x[i + 1][0];
494 addps m5, m4 ; real_sum2 += x[i][0] * x[i + 2][0], x[i][1] * x[i + 2][1]; imag_sum2 += x[i][0] * x[i + 2][1], x[i][1] * x[i + 2][0];
495 addps m7, m1 ; real_sum0 += x[i][0] * x[i][0], x[i][1] * x[i][1];
497 MOVH m1, [xq+cntq+16]
503 addps m6, m3 ; real_sum1 += x[i][0] * x[i + 1][0], x[i][1] * x[i + 1][1]; imag_sum1 += x[i][0] * x[i + 1][1], x[i][1] * x[i + 1][0];
504 addps m5, m4 ; real_sum2 += x[i][0] * x[i + 2][0], x[i][1] * x[i + 2][1]; imag_sum2 += x[i][0] * x[i + 2][1], x[i][1] * x[i + 2][0];
505 addps m7, m2 ; real_sum0 += x[i][0] * x[i][0], x[i][1] * x[i][1];
507 MOVH m2, [xq+cntq+16]
513 addps m6, m3 ; real_sum1 += x[i][0] * x[i + 1][0], x[i][1] * x[i + 1][1]; imag_sum1 += x[i][0] * x[i + 1][1], x[i][1] * x[i + 1][0];
514 addps m5, m4 ; real_sum2 += x[i][0] * x[i + 2][0], x[i][1] * x[i + 2][1]; imag_sum2 += x[i][0] * x[i + 2][1], x[i][1] * x[i + 2][0];
515 addps m7, m0 ; real_sum0 += x[i][0] * x[i][0], x[i][1] * x[i][1];
521 addps m2, m6 ; real_sum1 + x[38][0] * x[39][0], x[38][1] * x[39][1]; imag_sum1 + x[38][0] * x[39][1], x[38][1] * x[39][0];
522 addps m1, m7 ; real_sum0 + x[38][0] * x[38][0], x[38][1] * x[38][1];
523 addps m6, [rsp ] ; real_sum1 + x[ 0][0] * x[ 1][0], x[ 0][1] * x[ 1][1]; imag_sum1 + x[ 0][0] * x[ 1][1], x[ 0][1] * x[ 1][0];
524 addps m7, [rsp+16] ; real_sum0 + x[ 0][0] * x[ 0][0], x[ 0][1] * x[ 0][1];
539 movhps [phiq+0x18], m7
540 movss [phiq+0x28], m7
541 movss [phiq+0x10], m1