1 ;*****************************************************************************
3 ;*****************************************************************************
4 ;* Copyright (C) 2008-2010 x264 project
6 ;* Authors: Loren Merritt <lorenm@u.washington.edu>
7 ;* Holger Lubitz <holger@lubitz.org>
9 ;* This file is part of Libav.
11 ;* Libav is free software; you can redistribute it and/or
12 ;* modify it under the terms of the GNU Lesser General Public
13 ;* License as published by the Free Software Foundation; either
14 ;* version 2.1 of the License, or (at your option) any later version.
16 ;* Libav is distributed in the hope that it will be useful,
17 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
18 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 ;* Lesser General Public License for more details.
21 ;* You should have received a copy of the GNU Lesser General Public
22 ;* License along with Libav; if not, write to the Free Software
23 ;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 ;******************************************************************************
40 %macro TRANSPOSE4x4B 5
41 SBUTTERFLY bw, %1, %2, %5
42 SBUTTERFLY bw, %3, %4, %5
43 SBUTTERFLY wd, %1, %3, %5
44 SBUTTERFLY wd, %2, %4, %5
48 %macro TRANSPOSE4x4W 5
49 SBUTTERFLY wd, %1, %2, %5
50 SBUTTERFLY wd, %3, %4, %5
51 SBUTTERFLY dq, %1, %3, %5
52 SBUTTERFLY dq, %2, %4, %5
56 %macro TRANSPOSE2x4x4W 5
57 SBUTTERFLY wd, %1, %2, %5
58 SBUTTERFLY wd, %3, %4, %5
59 SBUTTERFLY dq, %1, %3, %5
60 SBUTTERFLY dq, %2, %4, %5
61 SBUTTERFLY qdq, %1, %2, %5
62 SBUTTERFLY qdq, %3, %4, %5
65 %macro TRANSPOSE4x4D 5
66 SBUTTERFLY dq, %1, %2, %5
67 SBUTTERFLY dq, %3, %4, %5
68 SBUTTERFLY qdq, %1, %3, %5
69 SBUTTERFLY qdq, %2, %4, %5
73 %macro TRANSPOSE8x8W 9-11
75 SBUTTERFLY wd, %1, %2, %9
76 SBUTTERFLY wd, %3, %4, %9
77 SBUTTERFLY wd, %5, %6, %9
78 SBUTTERFLY wd, %7, %8, %9
79 SBUTTERFLY dq, %1, %3, %9
80 SBUTTERFLY dq, %2, %4, %9
81 SBUTTERFLY dq, %5, %7, %9
82 SBUTTERFLY dq, %6, %8, %9
83 SBUTTERFLY qdq, %1, %5, %9
84 SBUTTERFLY qdq, %2, %6, %9
85 SBUTTERFLY qdq, %3, %7, %9
86 SBUTTERFLY qdq, %4, %8, %9
90 ; in: m0..m7, unless %11 in which case m6 is in %9
91 ; out: m0..m7, unless %11 in which case m4 is in %10
92 ; spills into %9 and %10
96 SBUTTERFLY wd, %1, %2, %7
99 SBUTTERFLY wd, %3, %4, %2
100 SBUTTERFLY wd, %5, %6, %2
101 SBUTTERFLY wd, %7, %8, %2
102 SBUTTERFLY dq, %1, %3, %2
105 SBUTTERFLY dq, %2, %4, %3
106 SBUTTERFLY dq, %5, %7, %3
107 SBUTTERFLY dq, %6, %8, %3
108 SBUTTERFLY qdq, %1, %5, %3
109 SBUTTERFLY qdq, %2, %6, %3
112 SBUTTERFLY qdq, %3, %7, %2
113 SBUTTERFLY qdq, %4, %8, %2
122 ; PABSW macros assume %1 != %2, while ABS1/2 macros work in-place
146 %macro PSIGNW_SSSE3 2
150 %macro ABS1_MMX 2 ; a, tmp
157 %macro ABS2_MMX 4 ; a, b, tmp0, tmp1
168 %macro ABS1_MMX2 2 ; a, tmp
174 %macro ABS2_MMX2 4 ; a, b, tmp0, tmp1
221 %define ABS1 ABS1_MMX
222 %define ABS2 ABS2_MMX
223 %define ABSB ABSB_MMX
224 %define ABSB2 ABSB2_MMX
227 movd %1, [%2-3] ;to avoid crossing a cacheline
237 %macro SPLATB_SSSE3 3
256 %macro PALIGNR_SSSE3 4
260 %macro DEINTB 5 ; mask, reg1, mask, reg2, optional src to fill masks from
268 pand m%1, m%2 ; dst .. y6 .. y4
269 pand m%3, m%4 ; src .. y6 .. y4
270 psrlw m%2, 8 ; dst .. y7 .. y5
271 psrlw m%4, 8 ; src .. y7 .. y5
286 %macro SUMSUB_BADC 4-5
319 psraw %2, 1 ; %2: %2>>1
320 psraw %1, 1 ; %1: %1>>1
321 paddw %2, %4 ; %2: %2>>1+%1
322 psubw %1, %3 ; %1: %1>>1-%2
327 SUMSUB_BADC m%4, m%1, m%3, m%2; m%5
328 SUMSUB_BA m%3, m%4, m%5
329 SUMSUB2_AB m%1, m%2, m%5
330 SWAP %1, %3, %4, %5, %2
332 SUMSUB_BADC m%4, m%1, m%3, m%2
335 SUMSUB2_AB m%1, [%5], m%2
342 SUMSUBD2_AB m%2, m%4, m%6, m%5
343 ; %2: %2>>1-%4 %4: %2+%4>>1
344 SUMSUB_BA m%3, m%1, m%6
345 ; %3: %1+%3 %1: %1-%3
346 SUMSUB_BADC m%4, m%3, m%2, m%1, m%6
347 ; %4: %1+%3 + (%2+%4>>1)
348 ; %3: %1+%3 - (%2+%4>>1)
349 ; %2: %1-%3 + (%2>>1-%4)
350 ; %1: %1-%3 - (%2>>1-%4)
352 SUMSUBD2_AB m%2, m%4, [%5], [%5+16]
354 SUMSUB_BADC m%4, m%3, m%2, m%1
357 ; %1: %1+%3 + (%2+%4>>1) row0
358 ; %2: %1-%3 + (%2>>1-%4) row1
359 ; %3: %1-%3 - (%2>>1-%4) row2
360 ; %4: %1+%3 - (%2+%4>>1) row3
385 movhps [%5+%6+32], m%1
386 movhps [%5+%6+40], m%2
387 movhps [%5+%6+48], m%3
388 movhps [%5+%6+56], m%4
391 %macro LOAD_DIFF_8x4P 7-10 r0,r2,0 ; 4x dest, 2x temp, 2x pointer, increment?
392 LOAD_DIFF m%1, m%5, m%7, [%8], [%9]
393 LOAD_DIFF m%2, m%6, m%7, [%8+r1], [%9+r3]
394 LOAD_DIFF m%3, m%5, m%7, [%8+2*r1], [%9+2*r3]
395 LOAD_DIFF m%4, m%6, m%7, [%8+r4], [%9+r5]
423 %macro STORE_DIFFx2 8 ; add1, add2, reg1, reg2, zero, shift, source, stride
438 %macro PMINUB_MMX 3 ; dst, src, tmp
444 %macro PMINUB_MMXEXT 3 ; dst, src, ignored