Carview!

GL:
- Support debug output, via GLRenderAPI.debug = true or env var GL_DEBUG=1
- Prefer EGL platform window
- Improve EGL init for GBM
- Prefer YUV sampler except android
- Fix immutable storage texture leak when rendering DRM objects
CUDA:
- Use a stream for output
- Fix blank lines at bottom if coded height != height
VAAPI:
- Support wayland display via SetGlobalOption("wl_display*"). It's optional because x11 and drm also work in wayland environment
- Fix dri3 interop was disabled and unusable long times ago
- va display is shared by default to avoid an potential crash in x11, can be disabled by decoder option shared=0 or SetGlobalOption("vaapi.shared", 0)
D3D11:
- Fix RTV leak in 0.33.0
- Disable IDXSync introduced in 0.33.0 for D3D11RenderAPI.rtv because a leak issue not yet fixed
Fix high cpu load on linux when EOF is read
Fix global options not applied for decoders output drm objects, except vaapi
Fix a mem leak: drmGetFormatModifierName, pulse audio init
Apple: fix audio renderer recursive init
BRAW: support 4.5+ SDK
R3D: support 9.0+ SDK and N-RAW. But still built with 8.6 SDK because of a bug in 9.0
FFmpeg:
- Fix missing rotation data for when avfilter is used
- Fix can't load avdevice on android

API:
- Audio/VideoFrame is ref counted
- Add VideoFrame.get(DX11Resource*, ID3D11Device*) to get underlying dx11 resource, or convert to dx11 textures if device is provided(cuda only)
Rockchip rkmpp decoder 10bit output can be directly imported by EGLImage w/o rga format conversion. Maybe it's the only player supports this feature.
MFT:
- Support async and hardware mft via property hardware=1 or async=1, hardware implies async. jpeg and vp8 can be async decoder
- hardware property value can be a vendor name, e.g. nvidia, intel, amd, qcom
- packed yuv format is not preferred
D3D11:
- Render target D3D11RenderAPI.rtv can be an array texture. array index starts from 0, and increases if renderVideo() draws a new frame, i.e. returns a new timestamp
- Expose IDXSync for user, which can be used to sync between contexts if render decoded dx11 resource from VideoFrame.get(DX11Resource*) or encode/process dx11 texture rendered by renderVideo(). It's also used internally.
- Fix signal on a wrong fence
- Improve cross device rendering w/o cpu wait
OpenGL:
- Fix yuv sampler shader
- Don't use GL_OES_EGL_image_external for ES3, fix error on some devices(e.g. rk3588 mali driver)
- More EGL platform types, and platform extension is used when possible.
Metal: render target supports array texture
CUDA:
- Add decoder property array=N or global option cuda.array for dx11 to use an array texture as output
- Add decoder property persistent or global option cuda.persistent for gl and dx11 to map as cuda resource only once
Dolby vision: support profile 20, used by MV-HEVC
Use rgba format for GBM to fix no EGLConfig for rgb. env var GBM_FORMAT can force another format, the value is a GBM fourcc, e.g. AR24
record() without any decoder will remux the media. steps: setMedia(..., {}) => record(file,...) => set(State::Playing) => waitFor(State::Stopped)
Fix avfilter processed audio format
Fix bitmap subtitle display area
Fix some subtitle crash
Fix mirror for rotated video used by OpenGL FBO
build with libc++21
lto is disabled for apple platforms because of compiler bug.
FFmpeg: support 8.0 abi
- Add global option ffmpeg and libffmpeg to reload all av modules, value can be libffmpeg path or handle

API:
- Add AudioFrame and Player.onFrame<AudioFrame>(). Example: obs
- VideoFrame.from() can import CUDA and VAAPI resources
NEW: VP8/9, HEVC with Alpha support. Maybe it's the 1st player engine supports those standard alpha videos.
- VP8/9, HEVC alpha layer decoding for all decoders.
- alpha layer rendering in all renderers. Alpha and base layer can be from different decoders and have different strides, and sampler types can be different.
- alpha=0 can disable alpha layer decoding, e.g. player.setProperty("video.decoder", "alpha=0")
non-standard alpha videos whose alpha data is at bottom or right, enable via decoder property alpha=right or alpha=bottom. sw decoder or hw decoder copy mode only.
MFT: support HEVC base layer decoding if has multiple layers(alpha and MV-HEVC). By default such bitstream is rejected by MFT
D3D11: fix downloading from stage texture, previously may crash on amd, arm64
CUDA:
- Only decode base layer of has multiple layers. By default it decodes all layers(alpha and MV-HEVC).
- Add properties device, stream, pbo etc. see https://github.com/wang-bin/mdk-sdk/wiki/Decoders#cuda
- Fix a crash when destroying the last frame
VT: RequestedMVHEVCVideoLayerIDs bug reported by me in 2024 is fixed in macOS 15.4, so remove workaroud for macOS 15.4+
VAAPI: add "vpp" property, 1 to convert to rgb via vpp
EGL:
- Prefer storage extension for desktop gl. external only formats support is vendor dependent. Works well on all gpus including raspberry pi. previously only OpenGL ES + external image can be used for extenal only formats.
- Fix no hdr metadata warnings in system log on android
- Support android platform display
- Fix HDR display if no colorspace extension(regression in 0.31.0)
- Fix crash if no client extension(regression in 0.31.0)
Add value "fit" to property "subtitle.size" to scale subtitle to renderer size and keep aspect ratio
CMake: add rpath-link for linux libc++
Load libshaderc.so instead of libshaderc-shared.so on linux
global option videoout.hdr10_metadata=0 can disable hdr10 metadata passthru to display
Fix old subtitle rendered if track switched
Fix point map if render api changed(default opengl, to another)
Don't clear render target if failed to open a decoder
FFmpeg:
- Add "format" option for sw decoder. used by hevc alpha, for example format=yuva420p
- "sw_fallback" property applies for codecs w/o hwaccel

API:
- Add color space and dolby vision profile in MediaInfo
- Add VideoEffect::ScaleChannels and ShiftChannels
- Add Player.subtitleText() to get subtitle text(except bitmap subtitles) with ass style or not, even if subtitle rendering is disabled by setProperty("subtitle", "0"). see https://github.com/wang-bin/mdk-sdk/wiki/Player-APIs#stdstring-subtitletextdouble-time---1-int-style--0-const
EGL:
- Support HDR color spaces. player.set(ColorSpaceUnknown) will switch between HDR and SDR color space depending on current rendering video frame. Tested on android, may work on linux
- Enable HDR10 metadata pass through, can be disabled by env var EGL_HDR_METADATA=0
- Prefer rgb10a2 to avoid recreating context when switching between SDR and HDR
- Support KHR_no_config_context to avoid recreating context when switching between SDR and HDR
Subtitle:
- Fix properties not applied if set before play
- Support splitting ASS rendered rgba image into several smaller regions when possible to reduce total size and gpu bandwidth. controlled by player property subtitle.ass.regions.max or env ASS_REGIONS_MAX, value is int. use subtitle.ass.regions.debug or env ASS_REGIONS_DEBUG to visualize regions and show debug messages.
- Keep colorspace the same as video, fix dark in bt2020 pq
- Add player property subtitle.scale to scale all kind of subtitles. default is 1.0, i.e. no scale
- Fix subtitle.blur is not applied because of a libass bug
- Add player property subtitle.size, default value is video to render in video frame size, other values will render in video renderer resolution
- Redraw subtitle if output resolution changed even if playback is paused, e.g. rendering in renderer resolution, and resize the renderer
- Redraw subtitle for any subtitle property change even if playback is paused.
- Render earlier
- Fix dead lock
- Fix unexpected null texture for some timestamp values and blinking
- Set ass storage size
Dolby vision: Fix shader error if yuv sampler is used for MediaCodec decoder
GL: add glFinish() before context destroyed
D3D11: Improve UMA support check. UMA optimized texture download. can be disabled by env GPU_OPTIMAL_DOWNLOAD=0
DXGI: Use the 1st screen's HDR capability
Metal: Fix rendering mutiple overlayes
MFT: return error if output type is not supported
Fix wrong start time if record() multiple times.
Fix buffered packets out of range of the value set by setBufferRange()
Fix audio not decoded after seek if paused and result in a/v desync(apple only).
Fix seek(if not started) callback may be called twice.
Fix global subtitle options not applied, regression in 0.30.1
FFmpeg:
- Support webvtt in mkv. Patch is from avbuild, not merged by upstream.
- Workaround aac packets from rtsp have no keyframe if ffmpeg > 7.1

Fix a crash in audio renderer if get delay
renderVideo() before the 1st frame will clear render target with background color. useful if render target content is undefined
Fix some global options not applied immediately
FFmpeg: prefer new avcodec_get_supported_config()

API:
- Add Player.appendBuffer() to play streams with user provided data, for example data is from websocket. media protocol is stream, i.e. player.setMedia("stream:empty_or_any_string"). setTimeout() can abort current playback if timedout to read data from user.
Plain text subtitle(srt for example) style can be changed by user via subtitle properties
Multi channels audio rendering, no downmix to stereo
Support dolby vision in mpeg ts
PulseAudio is the default audio render for linux
Improve audio renderer delay and AV sync
VT: Fix hardware decoder disabled if hvcC has no PS NALUs
D3D11:
- Fix UMA check if run x64 on arm64 and failed to create textures
- Disable cross context 0-copy for AMD gpus to avoid corrupt result on many gpus
- Set usage for swapchain, fix back buffer RTV create error on some win7 devices
VAAPI:
- Map YUYV fourcc, fix jpeg yuv 422 rendering
- Convert to RGB with VPP if DRM to EGLImage is not possible(desktop gl + external only formats)
DRM Prime:
- Error if no OpenGL ES2/3 context for external only formats, then VAAPI can fallback to suboptimal rendering instead of blank
- Map more formats
Player property "audio.tracks", "video.tracks" and "subtitle.tracks" is same as setActiveTracks(), value is a int list string, e.g. "0,1"
Reset more callbacks before deleting player object to simplify user code
Fix undefined symbol on wayland < 1.10, compatible with more linux distributions, e.g. ubuntu 14.04
More packed yuv formats
MFT: fix d3d9 not tried if d3d11 is not supported
Fix null callback invoked if reset by user
Will try to load Ass.framework on apple OS except macOS, to be compatible with media-kit's frameworks
FFmpeg: Fix wromg abi version if avfilter is disabled when swresample is loaded and crash, e.g. qt6's ffmpeg dlls
Add nvjpeg2000 plugin for windows x64, requires cudart64_12.dll and nvjpeg2k_0.dll
Compatible with libc++20

Android 64bit is built with 16KB page size support
Supprt GL_EXT_EGL_image_storage for drm, via global option "eglimage.storage=1"
VAAPI: prefer x11 display to support both EGL and GLX
Add privacy manifest for apple platforms
Improve demuxer cache
Enable http reconnect
Fix OpenGL symbols not resolved on android 15 emulator
Fix snapshot in OpenGL renderers
Fix snapshot callback not invoked if failed
Fix muxer write after close
Fix a decoder crash when stopping playback
Fix waitFor(State::Stopped)
Stop demuxer immediately if io is aborted.
Apply ffmmpeg muxer options
Prefer ffmpeg dynamic library even if statically linked. mainly used by iOS user provided FFmpeg.framework

Support demuxer cache for http(s), can be enabled via player.setProperty("demux.buffer.ranges", "8"/*or other positive int*/). {count, "cache.ranges"} event will be triggered when cached time ranges are created, dropped(via LRU cache) or merged. Other protocols can be enabled via player property player.setProperty("demux.buffer.ranges", "http,https,proto1,proto2");.
Add Player.bufferedTimeRanges()
Dolby vision: support reading metadata from av1 in all decoders, previously only supported by ffmpeg hwaccels and software decoder.
Support pause/resume rtsp stream in Player.set(State). Enable via player.setProperty("reader.pause", "1");.
Record video from non-key frame if property "reader.starts_with_key" is "0"
Support nvJPEG2000 as a plugin
Support planar xyz format, produced by nvJPEG2000
Support p012, p212, p412, can be produced by VT decoder(can be tested with VT:fourcc=tv20)
OpenGL:
- Detect(guess) invalid foreign context, usually in old OSes(ubuntu 20.04) and with GLRenderAPI.getProcAddress set by user.
- Fix blank display if failed update textures
D3D11: fix crash if decoder and renderer adapters are different
Metal:
- Set layer scale, fix low resolution
- Build highest shader version
VT: support ProRes without fourcc
VAAPI:
- Prefer drm display over x11, less errors, better performance
- Fix upload unmapped host memory crash
Fix wrong channels in recorded pcm
Fix prepare callback not invoked if decoder changed
Fix rgb48be rendering
FFmpeg
- Improve abi compatibility, found in x86 32bit ffmpeg 4.x
- Fix flipped frame in avfilter
Compatible with msvcp 14.40+

Support rendering in a foreign render pass for all renderers. Previously only foreign opengl context are implemented. The performance should be better than rendering to texture. Now it's able to render in a Qt RHI window
Support decoding A53 closed caption in all video decoders. Previously only supported by ffmpeg hwaccels and software decoder. Can be disabled by player property "cc=0".
Add new player properties: "buffer", "audio.decoders", "video.decoders". Convenient to control player in flutter.
Support video frames with negative stride, frames are vertically flipped
BRAW: support 4.0 SDK, compatible with 3.0
Improve NALU start code and EBP search
VT decoder:
- fix hevc mpegts decode error after seek
- max continuous session restart on error is 1 by default, controlled by property "error=1"
MFT: fix mf shutdown too early and crash when releasing a sample after stop
Fix CUDA decoder error after seek. It's a regression in the previous release.
OpenGL: Fix UBO error, requires glsl150(OpenGL 3.2, ES 3.0)
Fix alpha channel value of render target when blending subtitle, now output alpha >= background alpha
Fix multiple frames rendered if prepare() without play
Fix wrong position() value after prepare() without play
Fix buffering progress value
Fix play a new media if previously an individual audio track file/url is used
Fix endless wait when buffering on apple
Fix audio channels on apple
Fix no frame decoded if decoder changed when seeking
Vulkan: fix wsi extension check
Metal: fix layer colorspace and disable EDR for SDR
Fix wrong frame height if copy DX9/11 textures to host memory
Fix negative duration and start time for live streams, should be 0

Dolby Vision profile 5 support:
- Works for all renderers: OpenGL, D3D11/12, Metal, Vulkan
- Support SDR and HDR10 display
- Works for all decoders. For ffmpeg software decoder and hwaccels, dolby vision metadata are parsed by ffmpeg. For other decoders including ffmpeg external decoders and mdk hardware decoders, metadata is parsed in mdk
AMediaCodec system dolby vision decoder is disabled by default(previously was enabled). Can turn on via property "dv=1".
Improve gpu const buffer/uniform update for all renderers
OpenGL: Support UBO. Enabled for OpenGLES >= 3.0, OpenGL >= 3.1 or extension GL_ARB_uniform_buffer_object is supported. Can be disabled via global option "gl.ubo=0" or env var "GL_UBO=0"
Fix blank display when switching to some videos
Fix dx swapchain colorspace support check
Reset some callbacks when destroying player
Add option "reader.starts_with_key", "0" can decode from non-key. Can be used to reduce latency of network stream
C++ API:
- Fix callbacks are not thread safe
- Fix MediaStatus callback register only once
- Fix seek, prepare, snapshot callbacks
Fix video with a single frame not decoded until EOS
Fix blank display if play another video with only different color primaries
Fix packet drop
Fix seek after EOS demuxed
Use display p3 if hdr10 metadata is invalid
Read duration from metadata if stream duration is unknown
Set android application context automatically for ffmpeg 7.0+
Prefer AudioTrack over OpenSLES
FFmpeg: Add avformat.video/audio/subtitle_codec_id options

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Releases: wang-bin/mdk-sdk

v0.33.1

Uh oh!

v0.33.0

Uh oh!

Version 0.32.0

Uh oh!

v0.31.0

Uh oh!

v0.30.1

Uh oh!

Version 0.30.0

Uh oh!

Version 0.29.1

Uh oh!

Version 0.29.0

Uh oh!

Version 0.28.0

Uh oh!

Version 0.27.0

Uh oh!