1 ;******************************************************************************
2 ;* SSE-optimized functions for the DCA decoder
3 ;* Copyright (C) 2012-2014 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 pf_inv16: times 4 dd 0x3D800000 ; 1/16
29 ; void decode_hf(float dst[DCA_SUBBANDS][8], const int32_t vq_num[DCA_SUBBANDS],
30 ; const int8_t hf_vq[1024][32], intptr_t vq_offset,
31 ; int32_t scale[DCA_SUBBANDS][2], intptr_t start, intptr_t end)
34 cglobal decode_hf, 6,6,5, dst, num, src, offset, scale, start, end
35 lea srcq, [srcq + offsetq]
43 mov offsetd, [scaleq + 2 * startq]
46 cvtsi2ss m0, [scaleq + 2 * startq]
48 mov offsetd, [numq + startq]
54 pmovsxbd m1, [srcq + DICT + 0]
55 pmovsxbd m2, [srcq + DICT + 4]
57 movq m1, [srcq + DICT]
68 movd mm0, [srcq + DICT + 0]
69 movd mm1, [srcq + DICT + 4]
92 mova [dstq + 8 * startq + 0], m1
93 mova [dstq + 8 * startq + 16], m2
115 ; %1=v0/v1 %2=in1 %3=in2
123 %define OFFSET NUM_COEF*count
125 ; for v0, incrementing and for v1, decrementing
126 mova va, [cf0q + OFFSET]
127 mova vb, [cf0q + OFFSET + 4*NUM_COEF]
129 mova m4, [cf0q + OFFSET + mmsize]
130 mova m0, [cf0q + OFFSET + 4*NUM_COEF + mmsize]
136 fmaddps va, m4, %3, va
137 fmaddps vb, m0, %3, vb
145 ; va = va1 va2 va3 va4
146 ; vb = vb1 vb2 vb3 vb4
151 unpcklps va, vb ; va3 vb3 va4 vb4
152 unpckhps m4, vb ; va1 vb1 va2 vb2
153 addps m4, va ; va1+3 vb1+3 va2+4 vb2+4
154 movhlps vb, m4 ; va1+3 vb1+3
155 addps vb, m4 ; va0..4 vb0..4
156 movlps [outq + count], vb
164 ; void dca_lfe_fir(float *out, float *in, float *coefs)
166 cglobal dca_lfe_fir%1, 3,3,6-%1, out, in, cf0
170 %define NUM_COEF 4*(2-%1)
171 %define NUM_OUT 32*(%1+1)
173 movu IN1, [inq + 4 - 1*mmsize]
174 shufps IN1, IN1, q0123
176 movu IN2, [inq + 4 - 2*mmsize]
177 shufps IN2, IN2, q0123
180 mov count, -4*NUM_OUT
181 add cf0q, 4*NUM_COEF*NUM_OUT
189 shufps IN1, IN1, q0123
190 mov count, -4*NUM_OUT
191 ; cf1 already correctly positioned
192 add outq, 4*NUM_OUT ; outq now at out2
195 shufps IN2, IN2, q0123
206 %if HAVE_FMA3_EXTERNAL
212 %if cpuflag(sse2) && notcpuflag(avx)
222 vperm2f128 %1, %3, %3, 1
223 vshufps %1, %1, %1, q0123
225 pshufd %1, [%2], q0123
233 ; reading backwards: ptr1 = synth_buf + j + i; ptr2 = synth_buf + j - i
234 ;~ a += window[i + j] * (-synth_buf[15 - i + j])
235 ;~ b += window[i + j + 16] * (synth_buf[i + j])
236 SHUF m5, ptr2 + j + (15 - 3) * 4, m6
239 SHUF m11, ptr2 + j + (15 - 3) * 4 - mmsize, m12
240 mova m12, [ptr1 + j + mmsize]
243 fmaddps m2, m6, [win + %1 + j + 16 * 4], m2
244 fnmaddps m1, m5, [win + %1 + j], m1
246 fmaddps m8, m12, [win + %1 + j + mmsize + 16 * 4], m8
247 fnmaddps m7, m11, [win + %1 + j + mmsize], m7
250 mulps m6, m6, [win + %1 + j + 16 * 4]
251 mulps m5, m5, [win + %1 + j]
253 mulps m12, m12, [win + %1 + j + mmsize + 16 * 4]
254 mulps m11, m11, [win + %1 + j + mmsize]
262 %endif ; cpuflag(fma3)
263 ;~ c += window[i + j + 32] * (synth_buf[16 + i + j])
264 ;~ d += window[i + j + 48] * (synth_buf[31 - i + j])
265 SHUF m6, ptr2 + j + (31 - 3) * 4, m5
266 mova m5, [ptr1 + j + 16 * 4]
268 SHUF m12, ptr2 + j + (31 - 3) * 4 - mmsize, m11
269 mova m11, [ptr1 + j + mmsize + 16 * 4]
272 fmaddps m3, m5, [win + %1 + j + 32 * 4], m3
273 fmaddps m4, m6, [win + %1 + j + 48 * 4], m4
275 fmaddps m9, m11, [win + %1 + j + mmsize + 32 * 4], m9
276 fmaddps m10, m12, [win + %1 + j + mmsize + 48 * 4], m10
279 mulps m5, m5, [win + %1 + j + 32 * 4]
280 mulps m6, m6, [win + %1 + j + 48 * 4]
282 mulps m11, m11, [win + %1 + j + mmsize + 32 * 4]
283 mulps m12, m12, [win + %1 + j + mmsize + 48 * 4]
291 %endif ; cpuflag(fma3)
295 ; void ff_synth_filter_inner_<opt>(float *synth_buf, float synth_buf2[32],
296 ; const float window[512], float out[32],
297 ; intptr_t offset, float scale)
298 %macro SYNTH_FILTER 0
299 cglobal synth_filter_inner, 0, 6 + 4 * ARCH_X86_64, 7 + 6 * ARCH_X86_64, \
300 synth_buf, synth_buf2, window, out, off, scale
302 %if ARCH_X86_32 || WIN64
303 %if cpuflag(sse2) && notcpuflag(avx)
307 VBROADCASTSS m0, scalem
309 ; Make sure offset is in a register and not on the stack
314 vinsertf128 m0, m0, xmm0, 1
318 ; prepare inner counter limit 1
323 %if ARCH_X86_32 || notcpuflag(avx)
326 mov i, 16 * 4 - (ARCH_X86_64 + 1) * mmsize ; main loop counter
332 %define buf2 synth_buf2q
334 mov buf2, synth_buf2mp
337 ; m1 = a m2 = b m3 = c m4 = d
341 mova m2, [buf2 + i + 16 * 4]
349 %if ARCH_X86_32 || notcpuflag(avx)
355 %define ptr2 r7q ; must be loaded
360 mova m7, [buf2 + i + mmsize]
361 mova m8, [buf2 + i + mmsize + 16 * 4]
362 lea win, [windowq + i]
363 lea ptr1, [synth_bufq + i]
365 mov ptr2, synth_bufmp
366 ; prepare the inner loop counter
368 %if ARCH_X86_32 || notcpuflag(avx)
380 add win, OFFQ ; now at j-64, so define OFFSET
388 mov buf2, synth_buf2m ; needed for next iteration anyway
389 mov outq, outmp ; j, which will be set again during it
391 ;~ out[i] = a * scale;
392 ;~ out[i + 16] = b * scale;
399 ;~ synth_buf2[i] = c;
400 ;~ synth_buf2[i + 16] = d;
401 mova [buf2 + i + 0 * 4], m3
402 mova [buf2 + i + 16 * 4], m4
404 mova [buf2 + i + 0 * 4 + mmsize], m9
405 mova [buf2 + i + 16 * 4 + mmsize], m10
409 mova [outq + i + 0 * 4], m1
410 mova [outq + i + 16 * 4], m2
412 mova [outq + i + 0 * 4 + mmsize], m7
413 mova [outq + i + 16 * 4 + mmsize], m8
415 %if ARCH_X86_32 || notcpuflag(avx)
416 sub i, (ARCH_X86_64 + 1) * mmsize