annotate ppc/mathops.h @ 9905:f3ff182e9ecf libavcodec

Fix potentially unaligned accesses in ff_copy_bits() A pointer should never be assigned a value which may have less than the required alignment of the target type. Compilers may assume pointer values have the required alignment, and emit normal load/store instructions. Unaligned pointers should use a character type or compiler-specific type modifiers.
author mru
date Wed, 01 Jul 2009 17:48:22 +0000
parents d762f9979193
children 5bf12cdf0cb6
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
8629
04423b2f6e0b cosmetics: Remove pointless period after copyright statement non-sentences.
diego
parents: 8590
diff changeset
3 * Copyright (c) 2001, 2002 Fabrice Bellard
3733
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
7760
c4a4495715dd Globally rename the header inclusion guard names.
stefano
parents: 7333
diff changeset
23 #ifndef AVCODEC_PPC_MATHOPS_H
c4a4495715dd Globally rename the header inclusion guard names.
stefano
parents: 7333
diff changeset
24 #define AVCODEC_PPC_MATHOPS_H
5163
9ecbfc0c82bf add multiple inclusion guards to headers
mru
parents: 3947
diff changeset
25
8662
31bdc2da2e80 Add required headers to fix warnings during 'make checkheaders'.
diego
parents: 8655
diff changeset
26 #include "config.h"
31bdc2da2e80 Add required headers to fix warnings during 'make checkheaders'.
diego
parents: 8655
diff changeset
27
8655
14ccee231565 Add a check for ppc4xx instructions; rename preprocessor directive accordingly.
diego
parents: 8629
diff changeset
28 #if HAVE_PPC4XX
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
29 /* signed 16x16 -> 32 multiply add accumulate */
7333
a8a79f5385f6 cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents: 6501
diff changeset
30 #define MAC16(rt, ra, rb) \
8031
eebc7209c47f Convert asm keyword into __asm__.
flameeyes
parents: 7760
diff changeset
31 __asm__ ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
32
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
33 /* signed 16x16 -> 32 multiply */
7333
a8a79f5385f6 cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents: 6501
diff changeset
34 #define MUL16(ra, rb) \
a8a79f5385f6 cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents: 6501
diff changeset
35 ({ int __rt; \
8031
eebc7209c47f Convert asm keyword into __asm__.
flameeyes
parents: 7760
diff changeset
36 __asm__ ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); \
7333
a8a79f5385f6 cosmetics: Reformat PPC code in libavcodec according to style guidelines.
diego
parents: 6501
diff changeset
37 __rt; })
3733
d1b5acd0b680 New single instruction math operation header
lu_zero
parents:
diff changeset
38 #endif
5163
9ecbfc0c82bf add multiple inclusion guards to headers
mru
parents: 3947
diff changeset
39
9596
182f35c8eaed PPC: implement MULH() in assembler
mru
parents: 8662
diff changeset
40 #define MULH MULH
182f35c8eaed PPC: implement MULH() in assembler
mru
parents: 8662
diff changeset
41 static inline av_const int MULH(int a, int b){
182f35c8eaed PPC: implement MULH() in assembler
mru
parents: 8662
diff changeset
42 int r;
182f35c8eaed PPC: implement MULH() in assembler
mru
parents: 8662
diff changeset
43 __asm__ ("mulhw %0, %1, %2" : "=r"(r) : "r"(a), "r"(b));
182f35c8eaed PPC: implement MULH() in assembler
mru
parents: 8662
diff changeset
44 return r;
182f35c8eaed PPC: implement MULH() in assembler
mru
parents: 8662
diff changeset
45 }
182f35c8eaed PPC: implement MULH() in assembler
mru
parents: 8662
diff changeset
46
9713
d762f9979193 Change HAVE_PPC64 to ARCH_PPC64.
reimar
parents: 9625
diff changeset
47 #if !ARCH_PPC64
9625
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
48 static inline av_const int64_t MAC64(int64_t d, int a, int b)
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
49 {
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
50 union { uint64_t x; unsigned hl[2]; } x = { d };
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
51 int h, l;
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
52 __asm__ ("mullw %3, %4, %5 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
53 "mulhw %2, %4, %5 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
54 "addc %1, %1, %3 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
55 "adde %0, %0, %2 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
56 : "+r"(x.hl[0]), "+r"(x.hl[1]), "=&r"(h), "=&r"(l)
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
57 : "r"(a), "r"(b));
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
58 return x.x;
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
59 }
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
60 #define MAC64(d, a, b) ((d) = MAC64(d, a, b))
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
61
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
62 static inline av_const int64_t MLS64(int64_t d, int a, int b)
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
63 {
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
64 union { uint64_t x; unsigned hl[2]; } x = { d };
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
65 int h, l;
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
66 __asm__ ("mullw %3, %4, %5 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
67 "mulhw %2, %4, %5 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
68 "subfc %1, %3, %1 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
69 "subfe %0, %2, %0 \n\t"
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
70 : "+r"(x.hl[0]), "+r"(x.hl[1]), "=&r"(h), "=&r"(l)
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
71 : "r"(a), "r"(b));
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
72 return x.x;
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
73 }
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
74 #define MLS64(d, a, b) ((d) = MLS64(d, a, b))
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
75 #endif
7c4ab94a82d8 PPC: 32-bit asm for MAC64 and MLS64
mru
parents: 9596
diff changeset
76
7760
c4a4495715dd Globally rename the header inclusion guard names.
stefano
parents: 7333
diff changeset
77 #endif /* AVCODEC_PPC_MATHOPS_H */