API Changes

None

Features

• Vulkan pipeline - Refactored to reduce complexity
• Documentation - Updates for Android integrations

Bug Fixes

• Ensure Vulkan frame is copied to ReceiveDecoderPicture when using an external picture defined by a PicturePlaneDesc
• Resolve possibility of race condition in Vulkan temporal buffer handling when multiple bases are submitted at once

API Changes

• Added new P010 and P016 enums to LCEVC_ColorFormat
• API additions are fully backwards compatible with 4.x.x API

Features

• Vulkan pipeline - Improved shaders to reduce power consumption on Android by 32%
• Vulkan pipeline - Merge horizontal and vertical upscalers for performance and efficiency
• Vulkan pipeline - Extend output conversion for all buffer types
• Vulkan pipeline - Added dynamic workgroup size selection at runtime
• CPU pipeline - Reduce processing overhead in temporal buffer for frames with low residual load
• Build system - Added VN_SDK_TRACING_SYSTEM_SINK option for system level perfetto tracing on Android
• Build system - Added build support for RISC-V in PR38

Bug Fixes

• Fix conformance issue at extreme chroma values with missing clamp
• passthrough_mode=scale on CPU pipeline behaves correctly
• Resolved issues in public GitHub CI

API Changes

None

Features

• Pixel processing - Refactored NEON and SSE upscalers to yield a 15% reduction in frame times
• Pixel processing - Added a new AVX2 upscaler to yield a 30% reduction in frame times on x86 systems
• JNI - Added support for CPU rendering and CPU pipeline in AndroidX.media
• Pipelines - New auto pipeline is added to automatically switch between CPU and Vulkan on Android depending on device support
• Vulkan pipeline - Added render support for 10bit
• Build system - FFmpeg dependency bumped to 9.0
• Test harness - Can now use hardware base decoders: MediaCodec, QSV and VideoToolbox when using an appropriate build of FFmpeg

Bug Fixes

• SSE build flags are only applied to SSE optimized source files - Issue #37
• Various stability improvements for Android in both CPU and Vulkan pipelines

API Changes

None

Features

• Enhancement - Added a bitstream fuzzing tool and hardened the bitstream parser against malformed input
• Enhancement - Various performance improvements to yield a 5% decrease in frame times
• Vulkan - Combine add and convert stages to improve performance
• Build System & Enhancement - Allow the enhancement and common modules to be built as a Linux kernel module. Various new CMake flags.
• Build System - Added VN_SDK_LDE_FLOATS CMake flag to allow builds without float usage
• Build System - Improved VN_SDK_MINIMUM_LIB_SIZE CMake flag to further reduce the file size of built shared libraries and the JNI aar

Bug Fixes

• Fixed default thread auto-detection on desktop platforms
• Improve Vulkan extension initialization checks
• Fix a Vulkan race condition
• Ensure passthrough behaves as expected in the Vulkan pipeline and JNI
• Ensure runtime SIMD acceleration flags are fully respected
• Fix ordering issues with out of order timestamp inputs
• Fix some conformance issues with min/max residual values

API Changes

• Added new Peek API call: LCEVC_PeekDecoderPictureDesc to peek all parameters of a future output picture
• API additions are fully backwards compatible with 4.x.x API

Features

• Vulkan - Support for Android hardware buffers on output including format conversion
• Common - Allow custom time implementation via clock_custom.h for kernel builds
• Build System - Added VN_SDK_STATIC_LIBS_AS_OBJECTS CMake flag to produce object libraries
• Build System - Added VN_SDK_MINIMUM_LIB_SIZE CMake flag to reduce the file size of build shared libraries

Bug Fixes

• Vulkan shader tweaks to support a wider range of Android GPUs
• LOQ1 residuals use correct inverse transform for level 1 1D scaling
• Improved edge case behavior of peek
• User data is correctly applied to residuals for conformance
• Improve error handling of bitstream parsing
• Builds in different build directories are fully reproducible and identical
• .y extension raw monochrome bases are correctly detected by the harness for monochrome tests
• VN_SDK_LTO CMake flag forces full mode in all compilers

API Changes

• Added new Render API calls: LCEVC_RenderInit, LCEVC_RenderSetWindow, LCEVC_RenderSendPicture and LCEVC_RenderReceivePicture along with associated structs and enums
• API additions are fully backwards compatible with 4.0.x API

Features

• Pipelines - Removed deprecated legacy pipeline
• Pipelines - Major refactor to move common code between cpu and vulkan pipelines to common base class
• Structure - WASM wrapper moved to new wrappers dir along with NEW Java Native Interface (JNI)
• Wrappers - Added JNI for consumption by AndroidX.media3
• Vulkan Pipeline - The Vulkan pipeline is now performant!
• Vulkan Pipeline - Added rendering support for use in AndroidX.media3
• Vulkan Pipeline - Memory handled by VMA
• Vulkan Pipeline - Test harness can render to the display with glfw via the test harness with new --render flag
• Enhancement - Simplified LdeKernel structure within LdeGlobalConfig
• Enhancement - Moved sharpen parameters from LdeFrameConfig to LdeGlobalConfig
• Pixel Processing - Refactored upscalers to reduce frame time by ~20% and memory usage by ~5% in CPU mode
• Pixel Processing - Added optimized sharpening filter functions
• API - De-duplicated disable_simd and force_scalar flags in favor of disable_simd
• API - Removed log_levels configuration option
• Func tests - Encoder version updated to SDK 3.16.4 for all stream generation
• Func tests - Improved and extended dithering tests
• Structure - Stop using git LFS for small in-repo unit test assets
• Documentation - Additional docs for pixel_processing and extract modules
• Documentation - Added docs for Vulkan pipeline options

Bug Fixes

• LCEVC_Log event type now triggers correctly on each log
• ldeApproximatePA could be called multiple times on the same kernel
• Dithering is applied correctly for a 0D 0D scaled stream in the CPU pipeline
• Compiling with VN_SDK_THREADING=OFF is truly single threaded with no references to threading libraries

API Changes

None

Features

• Enhancement - All temporal refresh residuals are treated as SET commands
• Enhancement - Huffman stack allocations can be configured through CMake
• Pixel Processing - Blit API split into 'Add' and 'Convert'
• Pixel Processing - Temporal add functions allow an implicit convert decreasing frame times by ~15%
• CPU Pipeline - Android thread count defaults to 1 for improved default behavior on Android phones
• Diagnostics - Tracing macros support metadata for Perfetto

Bug Fixes

None

API Changes

None

Features

• pkg-config - Main decoder API, utility and extract libraries are split into separate .pc files. Each .pc file only contains one library when builing shared.
• Build system - CMake & 32-bit Windows updates for Cerbero compatibility

Bug Fixes

• Fix crash when running concurrent instances of the decoder
• HDR additional info stream config is read correctly
• Fix undefined reference when building with ASYNC_DIAGNOSTICS=OFF - Issue #34
• Highlight residuals in CPU mode correctly clear setzero comands
• Fixes to github CI for linting from fork branches

API Changes

None

Features

• Common - Added memory pools to reduce system allocations
• CPU pipeline - Refactored to use new memory architecture to improve performance
• Vulkan pipeline - Refactored to decouple frame object from the pipeline

Bug Fixes

• Fix error in passthrough mode in FFmpeg
• Fix build error in acceleration.c when building with external toolchains - PR #32 & PR #33
• Libav base decoder correctly handles streams without enhancement (native mode)
• Further fixes to github CI

API Changes

None

Features

None

Bug Fixes

• pkg-config file is dynamically generated to allow easy builds of FFmpeg 7.1 with --enable-liblcevc-dec