Systems and methods for efficient parallelized video encoding
Methods and systems for parallelized encoding of video are disclosed. According to one embodiment, a video encoder comprises a plurality of encoding engines. Each encoding engine is configured to receive a respective designated region of a video frame and produce respective quantized coefficients, the respective region having one or more unencoded frame blocks. Each encoding engine has a local symcoder for performing entropy-based encoding of the respective quantized coefficients. The video encoder has a rate control module, in communication with each encoding engine, for receiving from the respective local symcoder of each encoding engine a respective region-level bit count. The video encoder has a central buffer, in communication with each encoding engine, for receiving from each encoding engine the respective quantized coefficients. The video encoder has a final symcoder, in communication with the central buffer and the rate control module, wherein the final symcoder performs further entropy-based encoding of the respective quantized coefficients received in the central buffer, and transmits to the rate control module a frame-level bit count.
1. A video encoder, comprising:
a plurality of encoding engines, wherein each encoding engine is configured to receive a respective designated region of a video frame and produce respective quantized coefficients, the respective region having one or more unencoded frame blocks, and includes a local symcoder for performing, in parallel, entropy-based encoding of the respective quantized coefficients;
a rate control module, in communication with each encoding engine, for receiving from the respective local symcoder of each encoding engine a respective region-level bit count;
a central buffer, in communication with each encoding engine, for receiving from each encoding engine respective quantized coefficients; and
a final symcoder, in communication with the central buffer and the rate control module, wherein the final symcoder performs entropy-based encoding of the respective quantized coefficients received in the central buffer, and transmits to the rate control module a frame-level bit count, wherein at least one encoding engine is configured to select a context model used by the corresponding local symcoder based on a region-level bit count generated by the corresponding local symcoder or the frame-level bit count.
2. The video encoder of claim 1 , wherein the rate control module is configured to compute a rate for encoding an unencoded frame block based on one or more region-level bit counts received from the one or more local symcoders, or the frame-level bit count.
3. The video encoder of claim 1 , wherein at least one of the encoding engines further comprises one or more of:
a predictor for predicting pixel values of the one or more received unencoded frame blocks;
a residual computation and transformation module for generating residuals for predicted pixel values of the one or more received unencoded frame blocks, and transforming the residuals; and
a quantizer for quantizing the transformed residuals.
4. The video encoder of claim 3 , wherein the rate control module is configured to compute a quantization parameter based on one or more region-level bit counts received from the one or more local symcoders, or the frame-level bit count.
5. The video encoder of claim 1 , further comprising a frame divider configured to receive an unencoded video frame, divide the unencoded frame into a plurality of non-overlapping regions, each non-overlapping region comprising at least one frame block, and transmit at least one non-overlapping region to at least one encoding engine.
6. The video encoder of claim 5 , wherein in each encoding engine the local symcoder is configured to encode at least 2040 blocks of size 64×64 pixels in one second, whereby the video encoder can encode 3840 pixel×2160 pixel frames at a rate of 60 frames per second (fps).
7. The video encoder of claim 1 , wherein at least one encoding engine is one of an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).
8. The video encoder of claim 1 , wherein at least one encoding engine has a processor in communication with memory, and having instructions in the memory that configure the processor as the local symcoder.
9. A method of encoding video frames, the method comprising:
receiving by each of a plurality of encoding engines a respective designated region of a video frame, the respective region having one or more unencoded frame blocks;
generating by each encoding engine respective quantized coefficients for the respective region of the video frame;
performing by a respective local symcoder in each encoding engine, in parallel with other symcoders, entropy-based encoding of the respective quantized coefficients and providing a respective region-level bit count to a rate control module;
determining by the rate control module, using the region-level bit counts received from the local symcoder of the plurality of encoding engines, an approximate frame-level bit count;
receiving at a central buffer, from each encoding engine, respective quantized coefficients; and
performing entropy encoding by a final symcoder of the respective quantized coefficients received in the central buffer, and transmitting to the rate control module a frame-level bit count, wherein at least one encoding engine selects a context model used by the corresponding local symcoder based on a region-level bit count generated by the corresponding local symcoder or the frame-level bit count.
10. The method of claim 9 , wherein the rate control module computes a rate for encoding an unencoded frame block based on one or more region-level bit counts received from the one or more local symcoders, or the frame-level bit count.
11. The method of claim 9 , further comprising performing by at least one of the encoding engines:
prediction of pixel values of the one or more received unencoded frame blocks;
computation of residuals for the predicted pixel values;
transformation of the residuals to frequency domain; and
quantization of the transformed residuals.
12. The method of claim 11 , further comprising:
computing by the rate control module a quantization parameter based on one or more region-level bit counts received from the one or more local symcoders, or the frame-level bit count.
13. The method of claim 9 , further comprising:
dividing, using a frame divider, the unencoded frame into a plurality of non-overlapping regions, each non-overlapping region comprising at least one frame block; and
transmit at least one non-overlapping region to at least one encoding engine.
14. The method of claim 13 , wherein in each encoding engine the local symcoder is configured to encode at least 2040 blocks of size 64×64 pixels in one second, whereby the video encoder can encode 3840 pixel×2160 pixel frames at a rate of 60 frames per second (fps).
15. The method of claim 9 , wherein at least one encoding engine is one of an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).
16. The method of claim 9 , wherein at least one encoding engine has a processor in communication with memory, and having instructions in the memory that configure the processor as the local symcoder.