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)
202 %macro INT16_TO_INT32_N 0
214 %macro INT32_TO_INT16_N 0
224 %macro INT32_TO_FLOAT_INIT 0
227 %macro INT32_TO_FLOAT_N 0
234 %macro FLOAT_TO_INT32_INIT 0
237 %macro FLOAT_TO_INT32_N 0
248 %macro INT16_TO_FLOAT_INIT 0
251 %macro INT16_TO_FLOAT_N 0
263 %macro FLOAT_TO_INT16_INIT 0
266 %macro FLOAT_TO_INT16_N 0
280 CONV int32, int16, u, 2, 1, INT16_TO_INT32_N
281 CONV int32, int16, a, 2, 1, INT16_TO_INT32_N
282 CONV int16, int32, u, 1, 2, INT32_TO_INT16_N
283 CONV int16, int32, a, 1, 2, INT32_TO_INT16_N
286 CONV int32, int16, u, 2, 1, INT16_TO_INT32_N
287 CONV int32, int16, a, 2, 1, INT16_TO_INT32_N
288 CONV int16, int32, u, 1, 2, INT32_TO_INT16_N
289 CONV int16, int32, a, 1, 2, INT32_TO_INT16_N
291 PACK_2CH int16, int16, u, 1, 1
292 PACK_2CH int16, int16, a, 1, 1
293 PACK_2CH int32, int32, u, 2, 2
294 PACK_2CH int32, int32, a, 2, 2
295 PACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N
296 PACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N
297 PACK_2CH int16, int32, u, 1, 2, INT32_TO_INT16_N
298 PACK_2CH int16, int32, a, 1, 2, INT32_TO_INT16_N
300 UNPACK_2CH int16, int16, u, 1, 1
301 UNPACK_2CH int16, int16, a, 1, 1
302 UNPACK_2CH int32, int32, u, 2, 2
303 UNPACK_2CH int32, int32, a, 2, 2
304 UNPACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N
305 UNPACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N
306 UNPACK_2CH int16, int32, u, 1, 2, INT32_TO_INT16_N
307 UNPACK_2CH int16, int32, a, 1, 2, INT32_TO_INT16_N
310 CONV float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
311 CONV float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
312 CONV int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
313 CONV int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
314 CONV float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
315 CONV float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
316 CONV int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
317 CONV int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
319 PACK_2CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
320 PACK_2CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
321 PACK_2CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
322 PACK_2CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
323 PACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
324 PACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
325 PACK_2CH int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
326 PACK_2CH int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
328 UNPACK_2CH float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
329 UNPACK_2CH float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
330 UNPACK_2CH int32, float, u, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
331 UNPACK_2CH int32, float, a, 2, 2, FLOAT_TO_INT32_N, FLOAT_TO_INT32_INIT
332 UNPACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
333 UNPACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
334 UNPACK_2CH int16, float, u, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
335 UNPACK_2CH int16, float, a, 1, 2, FLOAT_TO_INT16_N, FLOAT_TO_INT16_INIT
339 UNPACK_2CH int16, int16, u, 1, 1
340 UNPACK_2CH int16, int16, a, 1, 1
341 UNPACK_2CH int32, int16, u, 2, 1, INT16_TO_INT32_N
342 UNPACK_2CH int32, int16, a, 2, 1, INT16_TO_INT32_N
343 UNPACK_2CH float, int16, u, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
344 UNPACK_2CH float, int16, a, 2, 1, INT16_TO_FLOAT_N, INT16_TO_FLOAT_INIT
348 CONV float, int32, u, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT
349 CONV float, int32, a, 2, 2, INT32_TO_FLOAT_N, INT32_TO_FLOAT_INIT