Encoder Integration Layer (EIL)
Last updated
Last updated
This article serves as a reference to the V-Nova LCEVC Encoder Integration Layer and should be read in conjunction with the Encoder Integration Layer (EIL) header, where each function is well documented as to its behaviour, arguments and return conditions. This document outlines the design, flow of functionality, and the properties exposed through the Integration Layers configuration system.
The following figure illustrates the EIL software stack.
The EIL combines V-Nova LCEVC with the base codec, and wrappers for supported base codecs. V-Nova provides example source code below to show how to use the Integration Layer.
The EIL provides a simple yet powerful API, and is similar in usage to working with other video encoding libraries.
The steps for encoding streams using the V-Nova LCEVC EIL are as follows:
Initialise the Integration Layer
While there are frames to encode:
a. Obtain an input frame and pass ownership to the encoder.
b. Wait for the encoder to complete, then fetch the decoded frame.
c. Release ownership of the output frame back to the encoder.
Shut down the Integration Layer
The integration of the EIL can be summarised in several steps:
Creation of the EIL
Initialisation of the EIL
Passing video frames to the EIL
Fetching the encoded bitstream from the EIL
Flushing the EIL
Destruction of the EIL
The above steps, as well as the various options available to integrations, are described below.
An instance of the EIL is created using EIL_Open
. An EILOpenSettings
object is required, which at minimum, must specify the name of a base encoder. This instance of the EIL should be passed to all other functions. The EILOpenSettings
can be initialised to a sensible default state using EIL_OpenSettingsDefault
.
The EIL issues internally generated logging messages via a callback. The callback is registered on the EILOpenSettings
struct. You must provide a callback function pointer before calling Open to receive messages.
The EIL is initialised using EIL_Initialise
. An EILInitSettings
object is required, in which the encoding properties, such as resolution, framerate and bitrate, are specified. properties_json
should be set to a json object string with extra optional encoder options.
Upon successful initialisation, the user can then proceed to encoding YUV input.
If the previous steps succeeded, then the EIL is now ready to receive frames.
If the EIL has been configured to not use external input, then a preallocated frame can be fetched from the EIL using EIL_GetPicture
; otherwise, an EILPicture
object should be created at the integration layer and passed to EIL_Encode
.
To encode YUV frames, perform the following steps:
Call EIL_GetPicture
to obtain an EILPicture
object for the type of frame to be encoded. The caller now temporarily owns the picture object.
Load YUV data into the pre-allocated memory of the plane member of the picture object.
Call EIL_Encode
using the EILPicture
object to begin encoding the picture. The encoder retakes ownership of the EILPicture
object, and it is no longer valid to manipulate it. This is a blocking call for the duration of the encoding of a single frame.
It is invalid behaviour to hold onto the EILOutput
object across multiple frames. The caller must copy and buffer the data if that is the required behaviour.
After receiving enough frames, the EIL is ready to output encoded data. The encoded data is fetched using the EIL_GetOutput
function.
If there is any output data available when EIL_GetOutput
is invoked, then EIL_GetOutput
returns EIL_RC_Success
.
If there is no data available and no errors occurred, then EIL_GetOutput
returns EIL_RC_Finished
.
To fetch the encoded bitstream from the EIL, perform the following steps:
The loop call EIL_GetOutput
returns the following, depending upon the presence of output data:
- If there is output data, then EILGetOutput
returns EIL_RC_Success
and populates EILOutput*
with a valid pointer. The caller now owns the output data object.
- If there is no more output data, then the EIL_GetOutput
returns EIL_RC_Finished
, unless an error is generated.
Please be aware this step is important, as the EIL may generate more than one output per single input, for example, when generating an interlaced encode.
Utilise the output data, either by writing it directly out to file or copying it if it must be buffered for output in a muxing system.
Call EIL_ReleaseOutput
to release the output object back to the ownership of the encoder. It is then no longer valid to read from this object.
Once the integration has finished with the output data, it should return it to the EIL using EIL_ReleaseOutput
. It is perfectly fine to call this method from a separate thread other than EIL_Encode
.
To signal to the EIL that no more input data is available, simply pass NULL
to EIL_Encode
. The EIL then finishes processing any queued video frames. This should only ever be performed at the end of the stream.
To complete encoding, perform the following steps:
Call EIL_Encode
once using a null pointer for the picture object argument, to flush the entire pipeline. This is a blocking call whilst the remaining pictures are encoded. Subsequent calls will have no effect.
Repeat step 4 from the encoding process IL_RC_Finished
is returned.
Once encoding has finished, then the EIL can be destroyed. Call EIL_Close
to safely shut down and release the encoding context.
The EIL can be configured to accept externally allocated input frames. To do this, the integration should set external_input
in EILInitSettings
to true. When configured in this mode, the EIL will not allocate any pictures itself, and so EIL_GetPicture
will always fail.
In external input mode, the integration must create the EILPicture
object itself. The object can be default initialised using EIL_InitPictureDefault
. The integration then must set the plane
pointers to memory it has allocated and stride
to the number of bytes per line. The other fields are then used in the same manner as when external input is disabled.
The planes must be valid from the point that the picture is passed to EIL_Encode
, until the associated output is produced.
The EIL can also accept Android hardware buffers in this mode when the EIL is configured for GPU mode. To do this the picture's memory_type
is set to EIL_MT_AHardwareBuffer
and the plane pointers are set to AHardwareBuffer
.
The EIL is able to reconfigure the target bitrate and framerate, if the base encoder also supports dynamic reconfiguration. This is done by passing EILReconfigureSettings
, with the new values, to the EIL_Reconfigure
function.
Almost all of the APIs available from the EIL return EILReturnCode
. The documentation for each function Error Handling outlines the potential return conditions. Every effort has been made to ensure that invalid usage of the IL is caught and presented with a meaningful return code. Additionally, a log error or warning will typically be issued via the logging mechanism.
Base layer encoding is implemented via a plugin system. Base encoder plugins, along with their corresponding wrappers, are dynamically loaded at runtime, based upon the name of the Base encoder specified in the EILOpenSettings
structure.
The name of the base encoder that the EIL uses for base encoding should not contain the preceding lcevc_eilp_
(liblcevc_eilp_
on Linux) or the platform's dynamic library extension. For example, if a plugin is called liblcevc_eilp_x264.so
or lcevc_eilp_x264.dll
, then the base_encoder
string should be x264
.
The EIL utilises several search paths to locate plugins. This is to assist with different program usage behaviours and EIL integration techniques. The order of search is as follows:
The location of the application process.
The current working directory.
System-defined PATH environment variable.
An API is provided for the EIL to interact with plugins. This allows a user to optionally implement their own base encoder through the plugin system without the need to be embedded within the EIL, and provides the benefits of modularity. Note: This requires the base plugin API and further documentation on how to develop your own base encoder plugin.
The EIL provides a very powerful configuration system. All configuration properties, except for the basic frame information, are set via a JSON configuration string in the EILInitSettings
structure.
The configuration system is supported by a comprehensive built-in system of optimised parameter defaults. The EIL employs ideal settings for both the LCEVC encoder and base encoder, based upon key parameters passed to it by the application, e.g., resolution, bitrate, tune.
Properties can be anonymously queried by the EIL_QueryPropertyGroups
API, and, if the user knows the name of properties, by the EIL_QueryProperty
API.
Properties specified within the JSON string override any setting provided by the database. Therefore, it is important that an application takes care to specify only properties within the JSON string that the user has explicitly set. For example, in a command-line program, only the options specified by the user in the command line should be included in the JSON string, or, from a GUI, only the options that the user has manually specified a choice for should be included in the JSON string.
The JSON string is processed utilising strong typing correlating to the target property type. If the types are not adhered to, then the EIL will fail to initialise. The user must take care to generate a JSON string that utilises the correct JSON type for each property type. The following table outlines the types and their JSON equivalents:
EILPropertyType
JSON type
EIL_PT_Int32
Number
EIL_PT_Int64
Number
EIL_PT_Uint32
Number
EIL_PT_Uint64
Number
EIL_PT_Float
Number
EIL_PT_Double
Number
EIL_PT_String
String
EIL_PT_Boolean
Boolean, string (‘0’ = false), or number (0 = false)
EIL_PT_Blob
Not used for properties
The JSON string is formatted with a top-level object, and a member per property, as shown in the following example:
For additional convenience, properties have been assigned to logical groupings, and can optionally contain a description about what the property does. This is so a user may dynamically generate a command-line parser, or a UI, for the publicly available properties.
All publicly available encoding configuration properties for the base encoder and LCEVC enhancement can be queried through the API using EIL_QueryPropertyGroups
. This will return an array of all property groups which each contain an array of all properties in that group. Alternatively, if the name of the parameter is known then it can be directly queried through EIL_QueryProperty
.
Each property contains a name, description, the expected type of that property and its current value. Below is an example that queries the API for all properties and prints them.
Metadata works much in the same way as properties, except that they are queried only from the system, and they cannot be set externally. They can be treated as read-only properties.
Metadata can only be queried by name using the EIL_QueryMetadata
API.
Metadata additionally implements an extension to properties allowing access to a contiguous block of memory (EIL_PT_Blob
). This system is incorporated with the Base encoder plugin, such that a Base encoder can output some data to the integrator of the IL, with the IL being ignorant to its usage.
This system is used to allow the EIL to report to the user some internal values, or data that are either not interpreted as properties, or that cannot be expressed as properties.
For example, the ffmpeg sample integration utilises this feature to obtain the h.264 global headers from the libx264 Base encoder plugin to be used for HLS.
This section describes the metadata values available from the EIL. It is broken down into sections for each discrete component of the EIL. The following tables indicate the name of the metadata, its associated property type, and a description of what it is and how it is used.
Metadata
Type
Description
base_type
EIL_PT_String
Describes the underlying technology of the base encoder plugin utilised. For example, the libx264 plugin will output h264 in this metadata.
base_width
EIL_PT_Uint32
Width of the base encoded video.
base_height
EIL_PT_Uint32
Height of the base encoded video.
Metadata
Type
Description
h264_header_data
EIL_PT_Blob
This is available when the ‘global-header’ x264 property is set to 1. This is a contiguous block of memory containing the H.264 headers (PPS, SPS, etc.), and can be used when the output container is segmented, so that the headers can be injected at the start of each segment.
Utility method for getting a string representation of an EIL return code in English.
Initialises the members of an EILOpenSettings
object to sensible default state. Some fields of the settings must still be populated by the integration before the settings can be used by the EIL.
This is the entry point for creating an instance of the EIL. The open function will have the version of the API appended at the end of the function name to ensure that a newer or older version of the EIL that contains breaking changes cannot be mistakenly used. As such, the EIL_Open
define is provided to aid in calling the correct open function.
If the EIL fails to open, then it is typically due to not being able to find and open the requested base encoder. A detailed message as to why the encoder failed to open should be sent to the log callback, if it is provided.
The EILContext
is an opaque handle to the EIL's encoder instance. An EILContext
is created when the EIL is opened using EIL_Open
, and the same instance should be passed to all other functions. It is destroyed when EIL_Close
is called.
EILOpenSettings Overview
EILOpenSettings
contains the information required to open the EIL. It can be default initialised using EIL_OpenSettingsDefault
.
base_encoder
[ Required ]
The name of the base encoder that the EIL uses for base encoding should not contain the preceding lcevc_eilp_
(liblcevc_eilp_
on Linux) or the platform's dynamic library extension. For example, if a plugin is called liblcevc_eilp_x264.so
or lcevc_eilp_x264.dll
, then the base_encoder
string should be x264
.
There are two special strings for base_encoder
: none
and inject
:
none
is a pass-through plugin that sets the output recon to the input.
inject
allows for reserved2
to reference an EILPAPI
object that contains function pointers for all of a plugin's methods.
base_encoder_userdata
[ Dependent on the plugin | Default: NULL
]
The base_encoder_userdata
is a handle that allows some arbitrary data to be passed through to a base plugin.
log_callback
[ Optional | Default: NULL
]
log_callback
can be set to a method that receives all log messages.
userdata
is the value set in log_userdata
, and level
is one of the levels in the enumeration EILLogLevel
. If left unset, then all logging is disabled.
All log messages are in the format level | group | message\n
, and are limited to 4096 bytes.
log_filepath
[ Optional | Default: NULL
]
log_filepath
is an optional file path, that when specified , contains all log messages.
reserved, reserved2
Reserved fields that should not be modified, unless specified elsewhere.
Initialises the members of an EILInitSettings
object to sensible defaults. Some fields of the structure must still be populated by the integration before the settings can be used by the EIL.
Initialises the EIL so that it is ready to receive pictures for encoding. The context must have previously been created using EIL_Open
.
EILInitSettings
contains the information required to initialise the EIL. It can be default initialised using EIL_InitSettingsDefault
.
[ Optional | Default: EIL_YUV_420P
]
The colour format of the input image. If it is not one of the YUV formats, then there will be an extra conversion step that converts the input into the YUV format. The matrix used for the colour conversion can be controlled using the picture_colorspace
config option.
[ Required ]
The resolution of the input image. Once the EIL is initialised. then these are modified by the EIL to become the resolution of the base video. These modified values should be used by any operation or process that consumes the bitstream. Consult the LCEVC standard for supported resolutions.
[ Required ]
The framerate's numerator and denominator, e.g., 25/1
for 25fps or 30000/1001
for 29.97fps
[ Required ]
The target bitrate, in kilobits/s, at which to encode.
[ Optional | Default: EIL_GOPLengthAuto
]
The GOP length in frames. There are three values to which gop_length
can be set, subject to the base encoder's abilities:
EIL_GOPLengthAuto
: The EIL determines which GOP length to use.
EIL_GOPLengthInfinite
: Only the first frame is an I-frame.
EIL_GOPLengthIntraOnly
: Every frame is an I-frame.
[ Required ]
A string containing json properties for the EIL's user-specified options.
[ Optional | Default: 0 ]
Whether or not the input image's planes should be allocated by the integration or the EIL.
When set to 0
, the EIL allocates and manages the lifetimes of an image's planes.
When set to 1
, the planes are expected to be allocated and managed external to the EIL.
When the EIL is configured to operate in internal input mode (see external_input
under EIL_InitSettings
), this method is used to obtain a picture object with preallocated planes. The planes can then be populated with frame data and passed to EIL_Encode
. A picture obtained using this method is owned by the integration until that picture is passed back through EIL_Encode
.
The EIL internally creates a pool of pictures from which this method extracts. The most common reason for the failure of this method, is that the pool of pictures is empty. This can occur if this function is called too frequently without returning the pictures through EIL_Encode
. The size of the pool depends on the base encoder's frame delay plus the EIL's frame delay, where frame delay is the number of frames that must be provided to the encoder before any output is produced.
EILPicture Overview
An EILPicture
is in the input to the encoder, containing all information needed to encode a single frame of video. If the EIL is configured to operate in external input mode, then it is up to the integration to set the fields of this struct, in which case it can be default initialised using EIL_InitPictureDefault
. Otherwise, the EIL will populate the fields of the struct, and the integration should populate the preallocated planes.
memory_type
[ Optional if externally allocated | Default EIL_MT_Host
]
memory_type
specifies the type of memory stored in the picture's planes. This can currently only be either:
EIL_MT_Host
is memory that is accessible from the device's CPU, i.e., allocated using malloc
or some device memory that has been mapped. If the memory is externally allocated, then the memory must be aligned for 8- or 16-bit access, depending upon the bitdepth.
EIL_MT_AHardwareBuffer
indicates that the planes are Android hardware buffers
(AHardwareBuffer
). The encoder must be configured to use the GPU for this to be accepted, when used in conjunction with external input.
Other memory types:
EIL_MT_TextureGL
is currently unsupported as input to the EIL. The EIL may pass this type of
memory to the base encoder, depending upon the configuration.
EIL_MT_Unknown
is used only to indicate that the memory type is not yet defined. memory_type
should not be set to EIL_MT_Unknown
by the point of ingest.
num_planes, plane[EIL_MaxPlanes]
[ Required if externally allocated ]
Pointers to the appropriate structures holding the plane data, as defined by the memory_type
. These planes can be interleaved in the cases of the RGB formats. As such, num_planes
indicates the number of plane pointers rather than the number of channels. In the case of contiguous memory, each pointer should be set to the starting address of each plane, and the number of planes should be set as if the memory is not contiguous.
stride[EIL_MaxPlanes]
[ Required if externally allocated ]
The stride, in bytes, for each plane defined in plane
.
base_type
[ Optional | Default: EIL_BT_Uknown
]
If set to a value other than the default, then when this picture is encoded, the EIL will attempt to force the requested type of picture from the base encoder. Whether or not the requested type of frame is produced, is subject to a base encoder's capabilities.
frame_type, field_type
[ Optional if externally allocated | Default: EIL_FrameType_Progressive
]
Whether this frame is progressive, a single field, or interlaced, and if it is a single field, then whether the frame is the top or bottom field.
pts
[ Optional | Defaut: -1
]
The presentation timestamp of the picture. If left as the default, then it is determined by the EIL.
user_data
[ Optional | Default: NULL
]
A handle to some arbitrary data that is set on the EILOutput
produced by the picture.
reserved
Internally used field. Do not modify.
Passes a picture that is ready for encode to the EIL. A picture can be obtained from the EI, either:
Using EIL_GetPicture
, if external input is disabled; or,
Created and populated by the integration, if external input is enabled.
Gets the next output available from the EIL. The integration has ownership of the EILOutput
until it has been passed back through EIL_ReleaseOutput
. This can return one of two return codes upon success:
EIL_RC_Success
: Successfully got the next frame of output.
EIL_RC_Finished
: No errors occurred but there there is currently no output available.
Returns an output obtained using EIL_GetOutput
back to the encoder.
EILOutput
EILOutput Overview
The EILOutput
struct contains all information for an encoded frame of video. These should never have to be created manually, and should be treated as read-only.
frame_type, field_type
Whether the frame that produced this output is progressive, interlaced or a field. field_type
is applicable, only if the type is EIL_FT_Field
.
base_type, keyframe
The type of frame that the base encoder produced, and whether or not it is flagged as a keyframe. Refer to the appropriate base encoder's standard for their meanings.
qp
The Quantisation Parameter (QP) of the frame, as reported by the base encoder.
pts
The presentation timestamp of the encoded frame. This is the input frame's pts, if it was specified; otherwise, it is the pts that the base encoder produced.
dts
The decode timestamp of the encoded frame, as produced by the base encoder.
config
Whether the data in this output is the encoder's config. For example, if the base encoder is an H264 encoder and config
is 1
, then the output contains only the SPS and PPS NAL units. The config output is always presented before any frame data.
data, data_length, lcevc, lcevc_length
The encoded bitstream for the frame and its length. If the separate_output
config option is set, then the LCEVC data is on the lcevc
field; otherwise, it is contained in the data
field, along with the base bitstream.
user_data
The user data that was set on the picture that produced this frame.
reserved
Internally used field. Do not modify.
This group of methods is used to query the options available through the EIL. These are usually controlled by the user.
EIL_QueryPropertyGroups
returns a collection of all properties in their logical groupings.
EIL_QueryProperty
obtains an individual property by its name.
The properties and their values are passed to the EIL in JSON format on the EILInitSettings
structure through the properties_json
field.
This structure contains an array of all property groups.
The property group contains a logical grouping of all properties available through the EIL and base encoders.
EILProperty Overview
This structure is used to describe config properties or metadata of the EIL and the base encoder.
name, description
A short name and description of the property that was queried.
type
The type of value contained in this property. This value should be used to determine which field of the value
union is set:
EIL_PT_Int32
: value.i32
EIL_PT_Int64
: value.i64
EIL_PT_Uint32
: value.u32
EIL_PT_Uint64
: value.u64
EIL_PT_Float
: value.f
EIL_PT_Double
: value.d
EIL_PT_String
: value.str
EIL_PT_Boolean
: value.i32
blob, blob_size
blob
and blob_size
are used when the property type is EIL_PT_Blob
. This is only ever the case when querying metadata.