1 ;******************************************************************************
2 ;* Copyright (c) 2012 Michael Niedermayer
4 ;* This file is part of FFmpeg.
6 ;* FFmpeg is free software; you can redistribute it and/or
7 ;* modify it under the terms of the GNU Lesser General Public
8 ;* License as published by the Free Software Foundation; either
9 ;* version 2.1 of the License, or (at your option) any later version.
11 ;* FFmpeg is distributed in the hope that it will be useful,
12 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
13 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 ;* Lesser General Public License for more details.
16 ;* You should have received a copy of the GNU Lesser General Public
17 ;* License along with FFmpeg; if not, write to the Free Software
18 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 ;******************************************************************************
21 %include "libavutil/x86/x86util.asm"
24 flt2pm31: times 8 dd 4.6566129e-10
25 flt2p31 : times 8 dd 2147483648.0
26 flt2p15 : times 8 dd 32768.0
28 word_unpack_shuf : db 0, 1, 4, 5, 8, 9,12,13, 2, 3, 6, 7,10,11,14,15
33 ;to, from, a/u, log2_outsize, log_intsize, const
35 cglobal pack_2ch_%2_to_%1_%3, 3, 4, 6, dst, src, len, src2
36 mov src2q , [srcq+gprsize]
41 jne pack_2ch_%2_to_%1_u_int %+ SUFFIX
43 jne pack_2ch_%2_to_%1_u_int %+ SUFFIX
45 jne pack_2ch_%2_to_%1_u_int %+ SUFFIX
47 pack_2ch_%2_to_%1_u_int %+ SUFFIX
49 lea srcq , [srcq + (1<<%5)*lenq]
50 lea src2q, [src2q + (1<<%5)*lenq]
51 lea dstq , [dstq + (2<<%4)*lenq]
56 mov%3 m0, [ srcq +(1<<%5)*lenq]
58 mov%3 m2, [ src2q+(1<<%5)*lenq]
68 mov%3 m0, [ srcq +(1<<%5)*lenq]
69 mov%3 m1, [mmsize + srcq +(1<<%5)*lenq]
70 mov%3 m2, [ src2q+(1<<%5)*lenq]
71 mov%3 m3, [mmsize + src2q+(1<<%5)*lenq]
78 mov%3 [ dstq+(2<<%4)*lenq], m0
79 mov%3 [ mmsize + dstq+(2<<%4)*lenq], m1
81 mov%3 [2*mmsize + dstq+(2<<%4)*lenq], m2
82 mov%3 [3*mmsize + dstq+(2<<%4)*lenq], m3
83 add lenq, 4*mmsize/(2<<%4)
85 add lenq, 2*mmsize/(2<<%4)
92 cglobal unpack_2ch_%2_to_%1_%3, 3, 4, 7, dst, src, len, dst2
93 mov dst2q , [dstq+gprsize]
98 jne unpack_2ch_%2_to_%1_u_int %+ SUFFIX
100 jne unpack_2ch_%2_to_%1_u_int %+ SUFFIX
102 jne unpack_2ch_%2_to_%1_u_int %+ SUFFIX
104 unpack_2ch_%2_to_%1_u_int %+ SUFFIX
106 lea srcq , [srcq + (2<<%5)*lenq]
107 lea dstq , [dstq + (1<<%4)*lenq]
108 lea dst2q, [dst2q + (1<<%4)*lenq]
111 mova m6, [word_unpack_shuf]
113 mov%3 m0, [ srcq +(2<<%5)*lenq]
114 mov%3 m2, [ mmsize + srcq +(2<<%5)*lenq]
116 %ifidn SUFFIX, _ssse3
137 shufps m0, m2, 10001000b
138 shufps m1, m2, 11011101b
141 mov%3 m2, [2*mmsize + srcq +(2<<%5)*lenq]
143 mov%3 m4, [3*mmsize + srcq +(2<<%5)*lenq]
144 shufps m2, m4, 10001000b
145 shufps m3, m4, 11011101b
149 mov%3 [ dstq+(1<<%4)*lenq], m0
151 mov%3 [ dst2q+(1<<%4)*lenq], m2
152 mov%3 [ mmsize + dstq+(1<<%4)*lenq], m1
153 mov%3 [ mmsize + dst2q+(1<<%4)*lenq], m3
154 add lenq, 2*mmsize/(1<<%4)
156 mov%3 [ dst2q+(1<<%4)*lenq], m1
157 add lenq, mmsize/(1<<%4)
164 cglobal %2_to_%1_%3, 3, 3, 6, dst, src, len
169 jne %2_to_%1_u_int %+ SUFFIX
171 jne %2_to_%1_u_int %+ SUFFIX
173 %2_to_%1_u_int %+ SUFFIX
175 lea srcq , [srcq + (1<<%5)*lenq]
176 lea dstq , [dstq + (1<<%4)*lenq]
180 mov%3 m0, [ srcq +(1<<%5)*lenq]
181 mov%3 m1, [ mmsize + srcq +(1<<%5)*lenq]
183 mov%3 m2, [2*mmsize + srcq +(1<<%5)*lenq]
184 mov%3 m3, [3*mmsize + srcq +(1<<%5)*lenq]
187 mov%3 [ dstq+(1<<%4)*lenq], m0
188 mov%3 [ mmsize + dstq+(1<<%4)*lenq], m1
190 mov%3 [2*mmsize + dstq+(1<<%4)*lenq], m2
191 mov%3 [3*mmsize + dstq+(1<<%4)*lenq], m3
192 add lenq, 4*mmsize/(1<<%4)
194 add lenq, 2*mmsize/(1<<%4)
206 cglobal pack_6ch_%2_to_%1_%3, 2,8,7, dst, src, src1, src2, src3, src4, src5, len
210 %define lend dword r2m
212 mov src1q, [srcq+1*gprsize]
213 mov src2q, [srcq+2*gprsize]
214 mov src3q, [srcq+3*gprsize]
215 mov src4q, [srcq+4*gprsize]
216 mov src5q, [srcq+5*gprsize]
221 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
223 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
225 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
227 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
229 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
231 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
233 pack_6ch_%2_to_%1_u_int %+ SUFFIX
242 mov%3 m1, [srcq+src1q]
243 mov%3 m2, [srcq+src2q]
244 mov%3 m3, [srcq+src3q]
245 mov%3 m4, [srcq+src4q]
246 mov%3 m5, [srcq+src5q]
254 blendps m6, m4, m0, 1100b
263 blendps m2, m5, m1, 1100b
276 mov %+ %3 %+ ps [dstq ], m0
277 mov %+ %3 %+ ps [dstq+16], m6
278 mov %+ %3 %+ ps [dstq+32], m4
279 mov %+ %3 %+ ps [dstq+48], m1
280 mov %+ %3 %+ ps [dstq+64], m2
281 mov %+ %3 %+ ps [dstq+80], m5
283 SBUTTERFLY dq, 0, 1, 6
284 SBUTTERFLY dq, 2, 3, 6
285 SBUTTERFLY dq, 4, 5, 6
306 %macro INT16_TO_INT32_N 6
318 %macro INT32_TO_INT16_N 6
328 %macro INT32_TO_FLOAT_INIT 6
331 %macro INT32_TO_FLOAT_N 6
338 %macro FLOAT_TO_INT32_INIT 6
341 %macro FLOAT_TO_INT32_N 6
352 %macro INT16_TO_FLOAT_INIT 6
355 %macro INT16_TO_FLOAT_N 6
356 INT16_TO_INT32_N %1,%2,%3,%4,%5,%6
367 %macro FLOAT_TO_INT16_INIT 6
370 %macro FLOAT_TO_INT16_N 6
387 CONV int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
388 CONV int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
389 CONV int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
390 CONV int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
392 PACK_6CH float, float, u, 2, 2, NOP_N, NOP_N
393 PACK_6CH float, float, a, 2, 2, NOP_N, NOP_N
396 PACK_6CH float, float, u, 2, 2, NOP_N, NOP_N
397 PACK_6CH float, float, a, 2, 2, NOP_N, NOP_N
400 CONV int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
401 CONV int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
402 CONV int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
403 CONV int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
405 PACK_2CH int16, int16, u, 1, 1, NOP_N, NOP_N
406 PACK_2CH int16, int16, a, 1, 1, NOP_N, NOP_N
407 PACK_2CH int32, int32, u, 2, 2, NOP_N, NOP_N
408 PACK_2CH int32, int32, a, 2, 2, NOP_N, NOP_N
409 PACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
410 PACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
411 PACK_2CH int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
412 PACK_2CH int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
414 UNPACK_2CH int16, int16, u, 1, 1, NOP_N, NOP_N
415 UNPACK_2CH int16, int16, a, 1, 1, NOP_N, NOP_N
416 UNPACK_2CH int32, int32, u, 2, 2, NOP_N, NOP_N
417 UNPACK_2CH int32, int32, a, 2, 2, NOP_N, NOP_N
418 UNPACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
419 UNPACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
420 UNPACK_2CH int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
421 UNPACK_2CH int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
423 CONV float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
424 CONV float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
425 CONV int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
426 CONV int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
427 CONV float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
428 CONV float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
429 CONV int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
430 CONV int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
432 PACK_2CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
433 PACK_2CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
434 PACK_2CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
435 PACK_2CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
436 PACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
437 PACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
438 PACK_2CH int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
439 PACK_2CH int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
441 UNPACK_2CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
442 UNPACK_2CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
443 UNPACK_2CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
444 UNPACK_2CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
445 UNPACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
446 UNPACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
447 UNPACK_2CH int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
448 UNPACK_2CH int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
450 PACK_6CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
451 PACK_6CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
452 PACK_6CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
453 PACK_6CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
456 UNPACK_2CH int16, int16, u, 1, 1, NOP_N, NOP_N
457 UNPACK_2CH int16, int16, a, 1, 1, NOP_N, NOP_N
458 UNPACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
459 UNPACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
460 UNPACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
461 UNPACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
463 %if HAVE_AVX_EXTERNAL
465 PACK_6CH float, float, u, 2, 2, NOP_N, NOP_N
466 PACK_6CH float, float, a, 2, 2, NOP_N, NOP_N
468 PACK_6CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
469 PACK_6CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
470 PACK_6CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
471 PACK_6CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
474 CONV float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
475 CONV float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
478 %if HAVE_AVX2_EXTERNAL
480 CONV int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
481 CONV int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT