annotate ppc/mathops.h @ 4980:800a543a2513 libavcodec

tweak mmx2 sad. 40% faster on core2, 18% faster on k8, 5% faster on p4.
author lorenm
date Fri, 11 May 2007 00:45:07 +0000
parents c8c591fe26f8
children 9ecbfc0c82bf
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
1 /*
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
2 * simple math operations
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
3 * Copyright (c) 2001, 2002 Fabrice Bellard.
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
4 * Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at> et al
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
5 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3733
diff changeset
6 * This file is part of FFmpeg.
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3733
diff changeset
7 *
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3733
diff changeset
8 * FFmpeg is free software; you can redistribute it and/or
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
9 * modify it under the terms of the GNU Lesser General Public
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
10 * License as published by the Free Software Foundation; either
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3733
diff changeset
11 * version 2.1 of the License, or (at your option) any later version.
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
12 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3733
diff changeset
13 * FFmpeg is distributed in the hope that it will be useful,
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
16 * Lesser General Public License for more details.
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
17 *
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
18 * You should have received a copy of the GNU Lesser General Public
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3733
diff changeset
19 * License along with FFmpeg; if not, write to the Free Software
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
21 */
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
22
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
23 #if defined(ARCH_POWERPC_405)
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
24 /* signed 16x16 -> 32 multiply add accumulate */
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
25 # define MAC16(rt, ra, rb) \
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
26 asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
27
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
28 /* signed 16x16 -> 32 multiply */
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
29 # define MUL16(ra, rb) \
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
30 ({ int __rt;
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
31 asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb));
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
32 __rt; })
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
33 #endif