Description
The Hardware Encoder Video Accelerator (HEVA), supports HEVC encoding low complexity with a flexible architecture targeting at least 1080p60 with minimal processing units and memory cuts and up to 2160p120 with large number of units and large memory cuts. Trade-off performance/area at design configuration: Reference cache size for 2160p30 at 350 MHz, 1 reference frame and bandwidth, overhead of 100% for references i.e. 1.5 GBytes/sec (minimal is 1.2 GB/s) Hardware interfaces :Host interface AXI3/AXI4 slave interface for the registers and command/status FIFO, Memory interface AXI3/AXI4 Streaming interfaces to External DRAM, Asynchronous AXI3/AXI4 128 bits interface, Synchronous DMA arbiter and memory interface Task sequencing modules: Manages communication and storage between processing modules, Control the shared memories and caches between the TPU modules and TSU/MIF, Defines the execution mode of the task processing units Task processing modules: Perform the pixel and bit-stream processing under control of TCR/TSU, The number of processing elements is defined at design configuration to sustain the required performance, A local reference cache is needed for performance for some processing units.
Features
-
Encoder acceleration
-
Performance up to 330 Mpixel/sec (2160p30 + 1080p30)
-
HEVC Main support, Level 4.2 (2160p30)
-
H.264 High Profile Progressive, Level 5.1 (2160p30, 1080p120)
-
HEVC Sample Adaptive Offset in-loop deblocking
-
Full coding unit support (from CU 64x64 to CU 8x8, PU 4x4)
-
No restriction on MV range allowed (X<8192, Y<4096)
-
Slice support: single slice or number of CTB lines per slice
-
Slice level IT programmable
-
Original input frame
-
Bottom/Right original padding on-the-fly
-
YUV 420 semi-planar: NV21
-
Reference frame usage
-
Internal 2D frame format
-
Optional proprietary lossless compression on reconstructed/reference frames
-
Up to two reference frames
-
Generalized P/B frames for low delay encoding
-
GOP up to 8 frames hierarchical B-frames for random access encoding
-
Motion vector range not limited, trade-off bandwidth versus MV range up to the application programming and design configuration.
-
Programmable quality/performance trade-off
-
Optional user defined input parameters
-
Deblocking slice parameters
-
User defined quantization scaling matrix tables
-
Chroma QP offsets
-
Region of interest input map table
-
Motion vector of Interest input map table
-
Optional user report output
-
Motion vector field
-
with/without Basic picture analysis
-
Latency tolerance at design configuration
Deliverables
-
RTL Source Code
-
HDL based test bench and behavioral models
-
Test cases
-
Protocol checkers, bus watchers and performance monitors
-
Configurable synthesis shell
-
Documentation & Design Guide
-
Verification Guide
-
Synthesis Guide
Applications
-
HEVC & H264 Recorder & Transcoder, for NAS application for Set Top Box application Best in Class HEVC Transcoder .
-
HEVC & H264 Video monitoring on video surveillance sequences –
-
On outdoor, non-static sequences ~5 to ~10% of bit-rate reduction
-
On outdoor, static sequences ~10 to 20% of bit-rate reduction
-
On indoor & lowlight sequences ~10% of bit-rate reduction
