1 ;*****************************************************************************
3 ;*****************************************************************************
4 ;* Copyright (C) 2008 Loren Merritt <lorenm@u.washington.edu>
6 ;* This program is free software; you can redistribute it and/or modify
7 ;* it under the terms of the GNU General Public License as published by
8 ;* the Free Software Foundation; either version 2 of the License, or
9 ;* (at your option) any later version.
11 ;* This program 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
14 ;* GNU General Public License for more details.
16 ;* You should have received a copy of the GNU General Public License
17 ;* along with this program; if not, write to the Free Software
18 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
19 ;*****************************************************************************
28 %macro TRANSPOSE4x4W
5
29 SBUTTERFLY wd
, %1, %2, %5
30 SBUTTERFLY wd
, %3, %4, %5
31 SBUTTERFLY
dq, %1, %3, %5
32 SBUTTERFLY
dq, %2, %4, %5
36 %macro TRANSPOSE2x4x4W
5
37 SBUTTERFLY wd
, %1, %2, %5
38 SBUTTERFLY wd
, %3, %4, %5
39 SBUTTERFLY
dq, %1, %3, %5
40 SBUTTERFLY
dq, %2, %4, %5
41 SBUTTERFLY qdq
, %1, %2, %5
42 SBUTTERFLY qdq
, %3, %4, %5
45 %macro TRANSPOSE4x4D
5
46 SBUTTERFLY
dq, %1, %2, %5
47 SBUTTERFLY
dq, %3, %4, %5
48 SBUTTERFLY qdq
, %1, %3, %5
49 SBUTTERFLY qdq
, %2, %4, %5
53 %macro TRANSPOSE8x8W
9-11
55 SBUTTERFLY wd
, %1, %2, %9
56 SBUTTERFLY wd
, %3, %4, %9
57 SBUTTERFLY wd
, %5, %6, %9
58 SBUTTERFLY wd
, %7, %8, %9
59 SBUTTERFLY
dq, %1, %3, %9
60 SBUTTERFLY
dq, %2, %4, %9
61 SBUTTERFLY
dq, %5, %7, %9
62 SBUTTERFLY
dq, %6, %8, %9
63 SBUTTERFLY qdq
, %1, %5, %9
64 SBUTTERFLY qdq
, %2, %6, %9
65 SBUTTERFLY qdq
, %3, %7, %9
66 SBUTTERFLY qdq
, %4, %8, %9
70 ; in: m0..m7, unless %11 in which case m6 is in %9
71 ; out: m0..m7, unless %11 in which case m4 is in %10
72 ; spills into %9 and %10
76 SBUTTERFLY wd
, %1, %2, %7
79 SBUTTERFLY wd
, %3, %4, %2
80 SBUTTERFLY wd
, %5, %6, %2
81 SBUTTERFLY wd
, %7, %8, %2
82 SBUTTERFLY
dq, %1, %3, %2
85 SBUTTERFLY
dq, %2, %4, %3
86 SBUTTERFLY
dq, %5, %7, %3
87 SBUTTERFLY
dq, %6, %8, %3
88 SBUTTERFLY qdq
, %1, %5, %3
89 SBUTTERFLY qdq
, %2, %6, %3
92 SBUTTERFLY qdq
, %3, %7, %2
93 SBUTTERFLY qdq
, %4, %8, %2
102 %macro ABS1_MMX
2 ; a, tmp
108 %macro ABS2_MMX
4 ; a, b, tmp0, tmp1
126 %define ABS1 ABS1_MMX
127 %define ABS2 ABS2_MMX
135 movd
%1, [%2-3] ;to avoid crossing a cacheline
145 %macro SPLATB_SSSE3
3
164 %macro PALIGNR_SSSE3
4
183 %macro HADAMARD8_1D
8
184 SUMSUB_BADC
%1, %5, %2, %6
185 SUMSUB_BADC
%3, %7, %4, %8
186 SUMSUB_BADC
%1, %3, %2, %4
187 SUMSUB_BADC
%5, %7, %6, %8
188 SUMSUB_BADC
%1, %2, %3, %4
189 SUMSUB_BADC
%5, %6, %7, %8
225 %macro LOAD_DIFF_8x4P
6-8 r0
,r2
; 4x dest, 2x temp, 2x pointer
226 LOAD_DIFF
%1, %5, none
, [%7], [%8]
227 LOAD_DIFF
%2, %6, none
, [%7+r1
], [%8+r3
]
228 LOAD_DIFF
%3, %5, none
, [%7+2*r1
], [%8+2*r3
]
229 LOAD_DIFF
%4, %6, none
, [%7+r4
], [%8+r5
]