http://blog.csdn.net/leixiaohua1020/article/details/44279329 (Structural Membership Management System-AVOption)
http://blog.csdn.net/leixiaohua1020/article/details/44268323 (Structural Membership Management System-AVClass)
//Need to understand the definition of the option structure - ------------------------------- simplified version
typedef struct AVOption {
const char *name;
const char *help;
int offset; -----------------Recording offset
enum AVOptionType type; -------------Parameter type, calling different interfaces according to type
union { -------------------Generally, the default values are saved.
int64_t i64;
double dbl;
const char *str;
AVRational q;
} default_val;
double min; ///< minimum valid value for the option
double max; ///< maximum valid value for the option
int flags;
const char *unit; --------------Signs are of the same kind
} AVOption;- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
Common commands in modules: Take nvenc for example
#define OFFSET(x) offsetof(NvencContext, x)- 1
- 1
{ "preset", "Set the encoding preset", OFFSET(preset), AV_OPT_TYPE_INT, { .i64 = PRESET_MEDIUM }, PRESET_DEFAULT, PRESET_LOSSLESS_HP, VE, "preset" },
{ "default", "", 0, AV_OPT_TYPE_CONST, { .i64 = PRESET_DEFAULT }, 0, 0, VE, "preset" },
{ "slow", "hq 2 passes", 0, AV_OPT_TYPE_CONST, { .i64 = PRESET_SLOW }, 0, 0, VE, "preset" },
......
{ "medium", "hq 1 pass", 0, AV_OPT_TYPE_CONST, { .i64 = PRESET_MEDIUM }, 0, 0, VE, "preset" },
{ "rc", "Override the preset rate-control", OFFSET(rc), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, VE, "rc" },
{ "constqp", "Constant QP mode", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_CONSTQP }, 0, 0, VE, "rc" },
....
{ NULL }- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
In fact, the mechanism of ffmpeg is modular. Commands are sent to a specific module, and then the module performs corresponding operations according to these commands. So the question is, how exactly do these commands go down to the module? That is, how does the upper layer relate to the private data of the module? This is part of the command mechanism. AVClass, a key structure, is used to perform an undertaking.
Generally speaking, the basic idea of setting commands is to use AVClass, associate the specified structure, search and match through the option array, and then use private data members to set command values. The key to setting the command is to save Option in AVClass, where the offset of a member record is important. For example, the h264_nvenc encoder is assigned to the private pointer priv in the upper layer.
(const AVClass*)s->priv_data = codec->priv_class;
Among them, s is the structure AVCodecContext, codec is the structure AVCodec, codec - > priv_class is h264_nvenc_class. The key of this class is to save the options information.
Depending on the offset, i.e. recording the relative offset address of each member of the NvencContext, assigning the address, and then in the specific underlying (plug-in) code:
NvencContext *ctx = avctx->priv_data;
This determines the relative address offset of each NvencContext member and the data member value of each NvencContext member.
Why is priv_data what NvencContext wants? A key technique here is to use pointer conversions such as const AVClass*. That's why one member of the NvencContext structure is AVClass. Almost many structures with similar inheritance relationships have such structures. AVClass can be said to be a structure of management type.
So how are the parameters of the command set? Here you need to understand the command parameter mechanism, such as structure, type, etc.
Simply, numerical class commands generally have two ways --- gpu 3 or --- gpu ls. There are several interfaces for setting commands. Here's a brief analysis of av_opt_set
int av_opt_set(void *obj, const char *name, const char *val, int search_flags)
{
int ret = 0;
void *dst, *target_obj;
/*Check Matching Command: Call av_opt_next inside*/
const AVOption *o = av_opt_find2(obj, name, NULL, 0, search_flags, &target_obj);
if (!o || !target_obj)
return AVERROR_OPTION_NOT_FOUND;
if (!val && (o->type != AV_OPT_TYPE_STRING &&
o->type != AV_OPT_TYPE_PIXEL_FMT && o->type != AV_OPT_TYPE_SAMPLE_FMT &&
o->type != AV_OPT_TYPE_IMAGE_SIZE && o->type != AV_OPT_TYPE_VIDEO_RATE &&
o->type != AV_OPT_TYPE_DURATION && o->type != AV_OPT_TYPE_COLOR &&
o->type != AV_OPT_TYPE_CHANNEL_LAYOUT && o->type != AV_OPT_TYPE_BOOL))
return AVERROR(EINVAL);
if (o->flags & AV_OPT_FLAG_READONLY)
return AVERROR(EINVAL);
/*Determine the offset position. The conversion is for one byte offset.*/
dst = ((uint8_t *)target_obj) + o->offset;
/*Call different parameter interfaces according to different parameter types*/
switch (o->type) {
case AV_OPT_TYPE_BOOL:
return set_string_bool(obj, o, val, dst);
case AV_OPT_TYPE_STRING:
return set_string(obj, o, val, dst);
case AV_OPT_TYPE_BINARY:
return set_string_binary(obj, o, val, dst);
case AV_OPT_TYPE_FLAGS:
case AV_OPT_TYPE_INT:
case AV_OPT_TYPE_INT64:
case AV_OPT_TYPE_FLOAT:
case AV_OPT_TYPE_DOUBLE:
case AV_OPT_TYPE_RATIONAL:
return set_string_number(obj, target_obj, o, val, dst);
case AV_OPT_TYPE_IMAGE_SIZE:
return set_string_image_size(obj, o, val, dst);
case AV_OPT_TYPE_VIDEO_RATE: {
AVRational tmp;
ret = set_string_video_rate(obj, o, val, &tmp);
if (ret < 0)
return ret;
return write_number(obj, o, dst, 1, tmp.den, tmp.num);
}
case AV_OPT_TYPE_PIXEL_FMT:
return set_string_pixel_fmt(obj, o, val, dst);
case AV_OPT_TYPE_SAMPLE_FMT:
return set_string_sample_fmt(obj, o, val, dst);
case AV_OPT_TYPE_DURATION:
if (!val) {
*(int64_t *)dst = 0;
return 0;
} else {
if ((ret = av_parse_time(dst, val, 1)) < 0)
av_log(obj, AV_LOG_ERROR, "Unable to parse option value \"%s\" as duration\n", val);
return ret;
}
break;
case AV_OPT_TYPE_COLOR:
return set_string_color(obj, o, val, dst);
case AV_OPT_TYPE_CHANNEL_LAYOUT:
if (!val || !strcmp(val, "none")) {
*(int64_t *)dst = 0;
} else {
int64_t cl = av_get_channel_layout(val);
if (!cl) {
av_log(obj, AV_LOG_ERROR, "Unable to parse option value \"%s\" as channel layout\n", val);
ret = AVERROR(EINVAL);
}
*(int64_t *)dst = cl;
return ret;
}
break;
}
av_log(obj, AV_LOG_ERROR, "Invalid option type.\n");
return AVERROR(EINVAL);
}- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
- 35
- 36
- 37
- 38
- 39
- 40
- 41
- 42
- 43
- 44
- 45
- 46
- 47
- 48
- 49
- 50
- 51
- 52
- 53
- 54
- 55
- 56
- 57
- 58
- 59
- 60
- 61
- 62
- 63
- 64
- 65
- 66
- 67
- 68
- 69
- 70
- 71
- 72
- 73
- 74
- 75
- 76
- 77
- 78
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
- 35
- 36
- 37
- 38
- 39
- 40
- 41
- 42
- 43
- 44
- 45
- 46
- 47
- 48
- 49
- 50
- 51
- 52
- 53
- 54
- 55
- 56
- 57
- 58
- 59
- 60
- 61
- 62
- 63
- 64
- 65
- 66
- 67
- 68
- 69
- 70
- 71
- 72
- 73
- 74
- 75
- 76
- 77
- 78
const AVOption *av_opt_next(const void *obj, const AVOption *last)
{
const AVClass *class;
if (!obj)
/*Connect - to get the address of the first member of the structure*/
/*Analysis: Getting the content of the first-level pointer is the de-reference of the second-level pointer.
The pointer itself is a variable and requires an address store.
If the first member of a structure is a pointer, then the first address of the structure is to point to that member. That is, to store the address of this member.*/
class = *(const AVClass**)obj;
if (!last && class && class->option && class->option[0].name)
return class->option;
if (last && last[1].name)
return ++last;
return NULL;
}- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
static int set_string_number(void *obj, void *target_obj, const AVOption *o, const char *val, void *dst)
{
int ret = 0;
int num, den;
char c;
/*The role of regular expressions here is not yet understood.*/
if (sscanf(val, "%d%*1[:/]%d%c", &num, &den, &c) == 2) {
if ((ret = write_number(obj, o, dst, 1, den, num)) >= 0)
return ret;
ret = 0;
}
for (;;) {
int i = 0;
char buf[256];
int cmd = 0;
double d;
int64_t intnum = 1;
if (o->type == AV_OPT_TYPE_FLAGS) {
if (*val == '+' || *val == '-')
cmd = *(val++);
for (; i < sizeof(buf) - 1 && val[i] && val[i] != '+' && val[i] != '-'; i++)
buf[i] = val[i];
buf[i] = 0;
}
{
/*Essentially, it introduces unit processing, converts value into name, o - > unit as value, makes a transit, and continues to call av_opt_find2. */
const AVOption *o_named = av_opt_find(target_obj, i ? buf : val, o->unit, 0, 0);
int res;
int ci = 0;
double const_values[64];
const char * const_names[64];
if (o_named && o_named->type == AV_OPT_TYPE_CONST)
d = DEFAULT_NUMVAL(o_named);
else {
if (o->unit) {
for (o_named = NULL; o_named = av_opt_next(target_obj, o_named); ) {
if (o_named->type == AV_OPT_TYPE_CONST &&
o_named->unit &&
!strcmp(o_named->unit, o->unit)) {
if (ci + 6 >= FF_ARRAY_ELEMS(const_values)) {
av_log(obj, AV_LOG_ERROR, "const_values array too small for %s\n", o->unit);
return AVERROR_PATCHWELCOME;
}
const_names [ci ] = o_named->name;
const_values[ci++] = DEFAULT_NUMVAL(o_named);
}
}
}
const_names [ci ] = "default";
const_values[ci++] = DEFAULT_NUMVAL(o);
const_names [ci ] = "max";
const_values[ci++] = o->max;
const_names [ci ] = "min";
const_values[ci++] = o->min;
const_names [ci ] = "none";
const_values[ci++] = 0;
const_names [ci ] = "all";
const_values[ci++] = ~0;
const_names [ci] = NULL;
const_values[ci] = 0;
res = av_expr_parse_and_eval(&d, i ? buf : val, const_names,
const_values, NULL, NULL, NULL, NULL, NULL, 0, obj);
if (res < 0) {
av_log(obj, AV_LOG_ERROR, "Unable to parse option value \"%s\"\n", val);
return res;
}
}
}
if (o->type == AV_OPT_TYPE_FLAGS) {
read_number(o, dst, NULL, NULL, &intnum);
if (cmd == '+')
d = intnum | (int64_t)d;
else if (cmd == '-')
d = intnum &~(int64_t)d;
}
if ((ret = write_number(obj, o, dst, d, 1, 1)) < 0)
return ret;
val += i;
if (!i || !*val)
return 0;
}
return 0;
}
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
- 35
- 36
- 37
- 38
- 39
- 40
- 41
- 42
- 43
- 44
- 45
- 46
- 47
- 48
- 49
- 50
- 51
- 52
- 53
- 54
- 55
- 56
- 57
- 58
- 59
- 60
- 61
- 62
- 63
- 64
- 65
- 66
- 67
- 68
- 69
- 70
- 71
- 72
- 73
- 74
- 75
- 76
- 77
- 78
- 79
- 80
- 81
- 82
- 83
- 84
- 85
- 86
- 87
- 88
- 89
- 90
- 91
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
- 35
- 36
- 37
- 38
- 39
- 40
- 41
- 42
- 43
- 44
- 45
- 46
- 47
- 48
- 49
- 50
- 51
- 52
- 53
- 54
- 55
- 56
- 57
- 58
- 59
- 60
- 61
- 62
- 63
- 64
- 65
- 66
- 67
- 68
- 69
- 70
- 71
- 72
- 73
- 74
- 75
- 76
- 77
- 78
- 79
- 80
- 81
- 82
- 83
- 84
- 85
- 86
- 87
- 88
- 89
- 90
- 91
Take away the above procedure and summarize the following procedure briefly:
av_opt_set
av_opt_find2 --- Check and Rematch Command
av_opt_next -- Called in a loop until NULL is encountered
If it is the first lookup, the first option will be returned.
class = (const AVClass*)obj; -- Establish a connection - in order to get the address of the first member of the structure
DST = ((uint8_t*) target_obj) + o-> offset; -- Determine offset, and the conversion is for a byte offset.
According to type, call corresponding processing interfaces, such as numerical classes.
set_string_number(obj, target_obj, o, val, dst);
If it is of - gpu ls type, further processing is needed.
Specific treatment methods:
(sscanf(val, "% d%*1[:/]%d%c", & num, & den, & c) -- Regular expression matching, if not ls, call write_number
If the above conditions are not satisfied,
Perform av_opt_find (target_obj, i?Buf: val, o-> unit, 0, 0);-
In essence, it introduces unit processing, converts value to name, o - > unit to value, makes a transit, and continues to call av_opt_find2.
Write_number (obj, o, dst, d, 1, 1) - assignment
ret = av_opt_set(s->priv_data, key, value, 0))
av_opt_set(void *obj, const char *name, const char *val, int search_flags)
Okay, contact the above (const AVClass*) s - > priv_data = codec - > priv_class;
Overall idea: Pass in a private data s - > priv_data, through conversion class = const AVClass*)obj, pointer conversion type AVClass, which contains option structure. This option holds the specific offset. Find the matching command through this option. Find the command, determine the offset, and assign the address. That is to say, private data holds the offset and the value corresponding to this address.
Summarize briefly with a diagram:
(const AVClass*)s->priv_data = codec->priv_class;
NvencContext *ctx = avctx->priv_data;
Original address: http://blog.csdn.net/sand_wiliam/article/details/53377721?locationNum=5&fps=1