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/x86inc.asm"
22 %include "libavutil/x86/x86util.asm"
26 flt2pm31: times 8 dd 4.6566129e-10
27 flt2p31 : times 8 dd 2147483648.0
28 flt2p15 : times 8 dd 32768.0
30 word_unpack_shuf : db 0, 1, 4, 5, 8, 9,12,13, 2, 3, 6, 7,10,11,14,15
35 ;to, from, a/u, log2_outsize, log_intsize, const
37 cglobal pack_2ch_%2_to_%1_%3, 3, 4, 6, dst, src, len, src2
38 mov src2q , [srcq+gprsize]
43 jne pack_2ch_%2_to_%1_u_int %+ SUFFIX
45 jne pack_2ch_%2_to_%1_u_int %+ SUFFIX
47 jne pack_2ch_%2_to_%1_u_int %+ SUFFIX
49 pack_2ch_%2_to_%1_u_int %+ SUFFIX
51 lea srcq , [srcq + (1<<%5)*lenq]
52 lea src2q, [src2q + (1<<%5)*lenq]
53 lea dstq , [dstq + (2<<%4)*lenq]
58 mov%3 m0, [ srcq +(1<<%5)*lenq]
60 mov%3 m2, [ src2q+(1<<%5)*lenq]
70 mov%3 m0, [ srcq +(1<<%5)*lenq]
71 mov%3 m1, [mmsize + srcq +(1<<%5)*lenq]
72 mov%3 m2, [ src2q+(1<<%5)*lenq]
73 mov%3 m3, [mmsize + src2q+(1<<%5)*lenq]
80 mov%3 [ dstq+(2<<%4)*lenq], m0
81 mov%3 [ mmsize + dstq+(2<<%4)*lenq], m1
83 mov%3 [2*mmsize + dstq+(2<<%4)*lenq], m2
84 mov%3 [3*mmsize + dstq+(2<<%4)*lenq], m3
85 add lenq, 4*mmsize/(2<<%4)
87 add lenq, 2*mmsize/(2<<%4)
94 cglobal unpack_2ch_%2_to_%1_%3, 3, 4, 7, dst, src, len, dst2
95 mov dst2q , [dstq+gprsize]
100 jne unpack_2ch_%2_to_%1_u_int %+ SUFFIX
102 jne unpack_2ch_%2_to_%1_u_int %+ SUFFIX
104 jne unpack_2ch_%2_to_%1_u_int %+ SUFFIX
106 unpack_2ch_%2_to_%1_u_int %+ SUFFIX
108 lea srcq , [srcq + (2<<%5)*lenq]
109 lea dstq , [dstq + (1<<%4)*lenq]
110 lea dst2q, [dst2q + (1<<%4)*lenq]
113 mova m6, [word_unpack_shuf]
115 mov%3 m0, [ srcq +(2<<%5)*lenq]
116 mov%3 m2, [ mmsize + srcq +(2<<%5)*lenq]
118 %ifidn SUFFIX, _ssse3
139 shufps m0, m2, 10001000b
140 shufps m1, m2, 11011101b
143 mov%3 m2, [2*mmsize + srcq +(2<<%5)*lenq]
145 mov%3 m4, [3*mmsize + srcq +(2<<%5)*lenq]
146 shufps m2, m4, 10001000b
147 shufps m3, m4, 11011101b
151 mov%3 [ dstq+(1<<%4)*lenq], m0
153 mov%3 [ dst2q+(1<<%4)*lenq], m2
154 mov%3 [ mmsize + dstq+(1<<%4)*lenq], m1
155 mov%3 [ mmsize + dst2q+(1<<%4)*lenq], m3
156 add lenq, 2*mmsize/(1<<%4)
158 mov%3 [ dst2q+(1<<%4)*lenq], m1
159 add lenq, mmsize/(1<<%4)
166 cglobal %2_to_%1_%3, 3, 3, 6, dst, src, len
171 jne %2_to_%1_u_int %+ SUFFIX
173 jne %2_to_%1_u_int %+ SUFFIX
175 %2_to_%1_u_int %+ SUFFIX
177 lea srcq , [srcq + (1<<%5)*lenq]
178 lea dstq , [dstq + (1<<%4)*lenq]
182 mov%3 m0, [ srcq +(1<<%5)*lenq]
183 mov%3 m1, [ mmsize + srcq +(1<<%5)*lenq]
185 mov%3 m2, [2*mmsize + srcq +(1<<%5)*lenq]
186 mov%3 m3, [3*mmsize + srcq +(1<<%5)*lenq]
189 mov%3 [ dstq+(1<<%4)*lenq], m0
190 mov%3 [ mmsize + dstq+(1<<%4)*lenq], m1
192 mov%3 [2*mmsize + dstq+(1<<%4)*lenq], m2
193 mov%3 [3*mmsize + dstq+(1<<%4)*lenq], m3
194 add lenq, 4*mmsize/(1<<%4)
196 add lenq, 2*mmsize/(1<<%4)
203 cglobal pack_6ch_%2_to_%1_%3, 2,8,7, dst, src, src1, src2, src3, src4, src5, len
207 %define lend dword r2m
209 mov src1q, [srcq+1*gprsize]
210 mov src2q, [srcq+2*gprsize]
211 mov src3q, [srcq+3*gprsize]
212 mov src4q, [srcq+4*gprsize]
213 mov src5q, [srcq+5*gprsize]
218 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
220 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
222 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
224 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
226 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
228 jne pack_6ch_%2_to_%1_u_int %+ SUFFIX
230 pack_6ch_%2_to_%1_u_int %+ SUFFIX
239 mov%3 m1, [srcq+src1q]
240 mov%3 m2, [srcq+src2q]
241 mov%3 m3, [srcq+src3q]
242 mov%3 m4, [srcq+src4q]
243 mov%3 m5, [srcq+src5q]
249 blendps m6, m4, m0, 1100b
252 blendps m2, m5, m1, 1100b
256 mov %+ %3 %+ ps [dstq ], m0
257 mov %+ %3 %+ ps [dstq+16], m6
258 mov %+ %3 %+ ps [dstq+32], m4
259 mov %+ %3 %+ ps [dstq+48], m1
260 mov %+ %3 %+ ps [dstq+64], m2
261 mov %+ %3 %+ ps [dstq+80], m5
263 SBUTTERFLY dq, 0, 1, 6
264 SBUTTERFLY dq, 2, 3, 6
265 SBUTTERFLY dq, 4, 5, 6
286 %macro INT16_TO_INT32_N 6
298 %macro INT32_TO_INT16_N 6
308 %macro INT32_TO_FLOAT_INIT 6
311 %macro INT32_TO_FLOAT_N 6
318 %macro FLOAT_TO_INT32_INIT 6
321 %macro FLOAT_TO_INT32_N 6
332 %macro INT16_TO_FLOAT_INIT 6
335 %macro INT16_TO_FLOAT_N 6
336 INT16_TO_INT32_N %1,%2,%3,%4,%5,%6
347 %macro FLOAT_TO_INT16_INIT 6
350 %macro FLOAT_TO_INT16_N 6
367 CONV int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
368 CONV int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
369 CONV int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
370 CONV int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
372 PACK_6CH float,float,u
373 PACK_6CH float,float,a
376 CONV int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
377 CONV int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
378 CONV int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
379 CONV int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
381 PACK_2CH int16, int16, u, 1, 1, NOP_N, NOP_N
382 PACK_2CH int16, int16, a, 1, 1, NOP_N, NOP_N
383 PACK_2CH int32, int32, u, 2, 2, NOP_N, NOP_N
384 PACK_2CH int32, int32, a, 2, 2, NOP_N, NOP_N
385 PACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
386 PACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
387 PACK_2CH int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
388 PACK_2CH int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
390 UNPACK_2CH int16, int16, u, 1, 1, NOP_N, NOP_N
391 UNPACK_2CH int16, int16, a, 1, 1, NOP_N, NOP_N
392 UNPACK_2CH int32, int32, u, 2, 2, NOP_N, NOP_N
393 UNPACK_2CH int32, int32, a, 2, 2, NOP_N, NOP_N
394 UNPACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
395 UNPACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
396 UNPACK_2CH int16, int32, u, 1, 2, INT32_TO_INT16_N, NOP_N
397 UNPACK_2CH int16, int32, a, 1, 2, INT32_TO_INT16_N, NOP_N
400 CONV float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
401 CONV float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
402 CONV int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
403 CONV int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
404 CONV float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
405 CONV float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
406 CONV int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
407 CONV int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
409 PACK_2CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
410 PACK_2CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
411 PACK_2CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
412 PACK_2CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
413 PACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
414 PACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
415 PACK_2CH int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
416 PACK_2CH int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
418 UNPACK_2CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
419 UNPACK_2CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
420 UNPACK_2CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
421 UNPACK_2CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
422 UNPACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
423 UNPACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
424 UNPACK_2CH int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
425 UNPACK_2CH int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
429 UNPACK_2CH int16, int16, u, 1, 1, NOP_N, NOP_N
430 UNPACK_2CH int16, int16, a, 1, 1, NOP_N, NOP_N
431 UNPACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N, NOP_N
432 UNPACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N, NOP_N
433 UNPACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
434 UNPACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
437 PACK_6CH float,float,u
438 PACK_6CH float,float,a
442 PACK_6CH float,float,u
443 PACK_6CH float,float,a
446 CONV float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
447 CONV float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT