FFmpeg
mathops.h
Go to the documentation of this file.
1 /*
2  * simple math operations
3  * Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at> et al
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #ifndef AVCODEC_ARM_MATHOPS_H
23 #define AVCODEC_ARM_MATHOPS_H
24 
25 #include <stdint.h>
26 #include "config.h"
27 #include "libavutil/common.h"
28 
29 #if HAVE_INLINE_ASM
30 
31 #if HAVE_ARMV6_INLINE
32 #define MULH MULH
33 static inline av_const int MULH(int a, int b)
34 {
35  int r;
36  __asm__ ("smmul %0, %1, %2" : "=r"(r) : "r"(a), "r"(b));
37  return r;
38 }
39 
40 #define FASTDIV FASTDIV
41 static av_always_inline av_const int FASTDIV(int a, int b)
42 {
43  int r;
44  __asm__ ("cmp %2, #2 \n\t"
45  "ldr %0, [%3, %2, lsl #2] \n\t"
46  "ite le \n\t"
47  "lsrle %0, %1, #1 \n\t"
48  "smmulgt %0, %0, %1 \n\t"
49  : "=&r"(r) : "r"(a), "r"(b), "r"(ff_inverse) : "cc");
50  return r;
51 }
52 
53 #else /* HAVE_ARMV6_INLINE */
54 
55 #define FASTDIV FASTDIV
56 static av_always_inline av_const int FASTDIV(int a, int b)
57 {
58  int r, t;
59  __asm__ ("umull %1, %0, %2, %3"
60  : "=&r"(r), "=&r"(t) : "r"(a), "r"(ff_inverse[b]));
61  return r;
62 }
63 #endif
64 
65 #define MLS64(d, a, b) MAC64(d, -(a), b)
66 
67 #if HAVE_ARMV5TE_INLINE
68 
69 /* signed 16x16 -> 32 multiply add accumulate */
70 # define MAC16(rt, ra, rb) \
71  __asm__ ("smlabb %0, %1, %2, %0" : "+r"(rt) : "r"(ra), "r"(rb));
72 
73 /* signed 16x16 -> 32 multiply */
74 # define MUL16 MUL16
75 static inline av_const int MUL16(int ra, int rb)
76 {
77  int rt;
78  __asm__ ("smulbb %0, %1, %2" : "=r"(rt) : "r"(ra), "r"(rb));
79  return rt;
80 }
81 
82 #endif
83 
84 #define mid_pred mid_pred
85 static inline av_const int mid_pred(int a, int b, int c)
86 {
87  int m;
88  __asm__ (
89  "mov %0, %2 \n\t"
90  "cmp %1, %2 \n\t"
91  "itt gt \n\t"
92  "movgt %0, %1 \n\t"
93  "movgt %1, %2 \n\t"
94  "cmp %1, %3 \n\t"
95  "it le \n\t"
96  "movle %1, %3 \n\t"
97  "cmp %0, %1 \n\t"
98  "it gt \n\t"
99  "movgt %0, %1 \n\t"
100  : "=&r"(m), "+r"(a)
101  : "r"(b), "r"(c)
102  : "cc");
103  return m;
104 }
105 
106 #endif /* HAVE_INLINE_ASM */
107 
108 #endif /* AVCODEC_ARM_MATHOPS_H */
r
const char * r
Definition: vf_curves.c:126
av_const
#define av_const
Definition: attributes.h:84
b
#define b
Definition: input.c:41
ff_inverse
const uint32_t ff_inverse[257]
Definition: mathtables.c:27
ra
#define ra
Definition: regdef.h:57
MULH
#define MULH
Definition: mathops.h:42
MUL16
#define MUL16(ra, rb)
Definition: mathops.h:89
FASTDIV
#define FASTDIV(a, b)
Definition: mathops.h:214
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
a
The reader does not expect b to be semantically here and if the code is changed by maybe adding a a division or other the signedness will almost certainly be mistaken To avoid this confusion a new type was SUINT is the C unsigned type but it holds a signed int to use the same example SUINT a
Definition: undefined.txt:41
common.h
av_always_inline
#define av_always_inline
Definition: attributes.h:49
mid_pred
#define mid_pred
Definition: mathops.h:98
__asm__
__asm__(".macro parse_r var r\n\t" "\\var = -1\n\t" _IFC_REG(0) _IFC_REG(1) _IFC_REG(2) _IFC_REG(3) _IFC_REG(4) _IFC_REG(5) _IFC_REG(6) _IFC_REG(7) _IFC_REG(8) _IFC_REG(9) _IFC_REG(10) _IFC_REG(11) _IFC_REG(12) _IFC_REG(13) _IFC_REG(14) _IFC_REG(15) _IFC_REG(16) _IFC_REG(17) _IFC_REG(18) _IFC_REG(19) _IFC_REG(20) _IFC_REG(21) _IFC_REG(22) _IFC_REG(23) _IFC_REG(24) _IFC_REG(25) _IFC_REG(26) _IFC_REG(27) _IFC_REG(28) _IFC_REG(29) _IFC_REG(30) _IFC_REG(31) ".iflt \\var\n\t" ".error \"Unable to parse register name \\r\"\n\t" ".endif\n\t" ".endm")