CN103974081A - HEVC coding method based on multi-core processor Tilera - Google Patents

HEVC coding method based on multi-core processor Tilera Download PDF

Info

Publication number
CN103974081A
CN103974081A CN201410194175.0A CN201410194175A CN103974081A CN 103974081 A CN103974081 A CN 103974081A CN 201410194175 A CN201410194175 A CN 201410194175A CN 103974081 A CN103974081 A CN 103974081A
Authority
CN
China
Prior art keywords
task
coding
tilera
filter
row
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201410194175.0A
Other languages
Chinese (zh)
Other versions
CN103974081B (en
Inventor
郑顺利
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
HANGZHOU TOPZEN INFORMATION TECHNOLOGY Co Ltd
Original Assignee
HANGZHOU TOPZEN INFORMATION TECHNOLOGY Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by HANGZHOU TOPZEN INFORMATION TECHNOLOGY Co Ltd filed Critical HANGZHOU TOPZEN INFORMATION TECHNOLOGY Co Ltd
Priority to CN201410194175.0A priority Critical patent/CN103974081B/en
Publication of CN103974081A publication Critical patent/CN103974081A/en
Application granted granted Critical
Publication of CN103974081B publication Critical patent/CN103974081B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

The invention discloses an HEVC coding method based on a multi-core processor Tilera. The HEVC coding method based on the multi-core processor Tilera comprises the steps that according to the number of cores of the multi-core processor Tilera, a top coder allocates multiple frame coder threads, and meanwhile, multiple work threads are allocated for coding and filtering; each frame coder thread issues the coding tasks and the filtering tasks of each line of a current frame to a task queue, the work threads read the task queue and process the read tasks with CTU as a unit, wavefront parallel processing is carried out on the coding tasks and the filtering tasks of all lines, and the HEVC coding of the frame is completed. According to the HEVC coding method based on the multi-core processor Tilera, the tasks in each line of each frame are divided into the coding tasks and the filtering tasks, balancing optimization of loads among cores can be achieved in a task detailed mode, the wavefront parallel processing is carried out on the coding tasks and the filtering tasks of each frame through the work threads, the parallel processing capability by the adoption of the multi-core processor Tilera is utilized fully, and the efficiency of the HEVC coding is improved greatly.

Description

A kind of HEVC coding method based on polycaryon processor Tilera
Technical field
The present invention relates to HD video coding field, be specifically related to a kind of HEVC coding method based on polycaryon processor Tilera.
Background technology
Along with the development of high-definition video technology is with universal, H.264/AVC video encoding standard cannot meet the compression requirements of video data gradually.In order to tackle this situation, the Video coding integration and cooperation group (JCT-VC) being set up by MPEG and VCEG has developed video encoding standard of new generation, is called efficient video coding (HEVC:High Efficiency Video Coding).
HEVC with H.264/AVC compare, on the basis of hybrid coding basic framework, used piece partition structure more flexibly, the new technologies such as more accurate predictive mode and sample self adaptation skew (SAO:SampleAdaptive Offset).Can guarantee image quality with H.264/AVC identical in, half by Compression to the latter.But, being accompanied by high compression rate and what come is high computation complexity, this has caused difficulty for the application and popularization of HEVC.At present there have been many researchers to propose the optimized algorithm of HEVC, improved code efficiency.But except from algorithm aspect is optimized, can also accelerate coding rate by the mode of hardware parallel processing.
Parallel processing generally has data parallel and function parallelization two classes.Data parallel refers to data to be divided into different masses, and they are carried out to same operation simultaneously.Function parallelization refers to, according to the Function Classification of processing, reach the object of parallel processing by stream treatment.HEVC, when design, has considered the optimization method of many parallel processings, wherein more typically has the division of wavefront parallel processing and Tiles.In existing patent, the mode that a kind of Tiles of dependence divides to carry out parallel processing is suggested.This scheme is started with from parallel data processing, has realized the object of accelerating HEVC coding.
On the other hand, the two-dimentional bus architecture of Tilera polycaryon processor innovation, has broken through the technical bottleneck of traditional unidirectional bus, has improved greatly the handling property of polycaryon processor.Between each core of polycaryon processor Tilera, there are independently Cache system and memory system, can effectively avoid the data access conflict of traditional shared drive mode, can give full play to the ability of each core, improve program parallelization treatment effeciency.And in prior art, do not occur utilizing polycaryon processor Tilera to carry out HEVC coding, and therefore need a kind of HEVC coding method of the Tilera of utilization polycaryon processor badly, improve the efficiency of existing HEVC coding.
Summary of the invention
For the deficiencies in the prior art, the invention provides a kind of HEVC coding method based on polycaryon processor Tilera.
A HEVC coding method based on polycaryon processor Tilera, comprising:
(1) according to the check figure of polycaryon processor Tilera, by top layer encoder, distribute several frame encoder threads, for parallel processing different frame, distribute a plurality of worker threads to be used for encoding and filtering simultaneously;
(2) for the present frame of each frame encoder thread process; frame encoder thread is distributed to task queue by coding task and the filter task of each row in present frame; worker thread reads task queue Bing YiCTUWei unit reading of task is processed; and the coding task of each row and filter task carry out wavefront parallel processing, completes the HEVC coding of this frame.
Top layer encoder is software upper layer module, mainly carries out thread distribution and initialization, and the operation such as coding progress feedback.
The present invention is coding task and filter task by the task division of every a line in each frame, to be applicable to the two-dimentional bus architecture of Tilera polycaryon processor innovation, by worker thread, the coding task in each frame and filter task are carried out to wavefront parallel processing, be conducive to improve the code efficiency of HEVC coding.
In described step (1), the number of frame encoder thread is 1~6.
In described step (1), in polycaryon processor Tilera, each core distribution obtains a worker thread.
The number of frame encoder thread is relevant with the check figure of polycaryon processor Tilera with the number of worker thread.Under default situations, the quantity of frame encoder thread is determined according to the check figure of polycaryon processor Tilera voluntarily by top layer encoder.When check figure is greater than 32, distribute 6 frame encoder threads; Otherwise, when check figure is more than or equal to 16, distribute 5 frame encoder threads; Otherwise, when check figure is more than or equal to 12, distribute 4 frame encoder threads; Otherwise, when check figure is more than or equal to 4, distribute 2 frame encoder threads; Otherwise, only distribute 1 frame encoder thread.And under default situations, each core distribution obtains a worker thread.In addition, also can be artificially the quantity of prescribed coding device thread and worker thread based on experience value.
The coding task process of processing each row in described step (2) in the process of wavefront parallel processing is as follows:
(2-1) judge that whether current line is the first row, proceeds as follows according to judged result:
(a), if current line is the first row, if the CTU quantity that current line completes coding task is more than or equal to the threshold value of setting, the coding task of next line is distributed in task queue;
(b) if current line is not the first row, make the following judgment:
If current line completes the threshold value that the CTU quantity of coding task and the difference of the CTU quantity that lastrow completes coding task are less than or equal to setting, stop the coding task of current line, worker thread reads task queue again;
If current line completes the threshold value that the CTU quantity of coding task is more than or equal to setting, and not last column, the coding task of next line is distributed in task queue, and continue to process the coding task of this row;
If current line completes the CTU quantity of coding task and is more than or equal to the threshold value of setting, and is last column, continue to process the coding task of this row;
(2-2), after the coding task of current line is finished dealing with, compare the size of the line number of current line and the filter delay line number of setting:
If the line number of current line is greater than the filter delay line number of setting, the filter task that the line number that is current line by line number deducts the row of filter delay line number is added into task queue, and worker thread reads task queue again;
Otherwise worker thread reads task queue again.
The filter task process of processing each row in described step (2) in the process of wavefront parallel processing is as follows:
If current line completes the threshold value that the CTU quantity of filter task and the difference of the CTU quantity that lastrow completes filter task are less than setting, stop the filter task of current line, worker thread reads task queue again;
If current line completes the threshold value that the CTU quantity of filter task is more than or equal to setting, and not last column, the filter task of next line is distributed in task queue, and continue to process the filter task of this row;
If current line completes the CTU quantity of coding task and is more than or equal to the threshold value of setting, and is last column, continue to process the filter task of this row.
When worker thread is processed the coding task of every a line or filter task, be with CTU (Coding TreeUnit, code tree unit) for processing unit, from the coding task of every a line or filter task, first uncompleted CTU starts to process.In the processing procedure of each frame, by present frame according to CTU apportion.When certain a line processing progress is greater than the some row of next line (threshold value of setting), worker thread can add task queue by the Processing tasks of next line; Meanwhile, when certain a line is processed progress and is less than some row with lastrow difference, current line can be locked, waits for lastrow activation this journey, again starts processing.
These two rules that rely on worker thread realize HEVC cataloged procedure and process according to WPP (English spelling, translator of Chinese) form.Thereby reach the object of parallel processing, and utilized fully the superiority of Tilera polycaryon processor.When carrying out wavefront parallel processing; frame encoder thread is first directly distributed to the coding task of present frame the first row in task queue; worker thread is carried out the coding task of the first row; in the process of implementation according to above processing rule; constantly the coding task of each row and filter task are added in task queue, during interpolation, thread is each task in Processing tasks queue ceaselessly also; so form software flow processing procedure, greatly improved code efficiency.
Described filter delay line number is 0~2.
As preferably, described filter delay line number is 2.
Described threshold value is 2~4.
The line number postponing and the size of establishing threshold value have directly determined the accuracy rate of HEVC coding.Postpone line number and threshold value larger, the accuracy rate of HEVC coding is higher, but can cause code efficiency to decline.Therefore as preferred, in the present invention, filter delay line number is 2, and the threshold value of setting is also 2.
In addition, associative processor platform, the optimisation strategy of using Tilera to provide, comprises compile option optimization, recirculating network optimization, Intrinsic optimization, Feedback optimization etc., further improve code efficiency.
HEVC coding method based on polycaryon processor Tilera of the present invention, by the task division of every a line in each frame, be coding task and filter task, in the mode of refinement task, realize the balance optimizing of inter-core load, by worker thread, the coding task in each frame and filter task are carried out to wavefront parallel processing, take full advantage of two-dimentional bus architecture and internuclear independently Cache and the memory system of the innovation of Tilera polycaryon processor, realized the HEVC coding of two-dimentional bus, this HEVC coding method has broken through the technical bottleneck that traditional polycaryon processor adopts unidirectional bus and shared drive, by the parallel processing capability of application polycaryon processor Tilera, greatly improved the code efficiency of HEVC coding.
Embodiment
Below in conjunction with specific embodiment, the present invention is described in detail.
The HEVC coding method based on polycaryon processor Tilera of the present embodiment, comprising:
(1) according to the check figure of polycaryon processor Tilera, by top layer encoder, distribute several frame encoder threads, for parallel processing different frame, distribute a plurality of worker threads to be used for encoding and filtering simultaneously;
The number of frame encoder thread is 1~6, and in polycaryon processor Tilera, each core distribution obtains a worker thread.
In the present embodiment, the check figure of polycaryon processor Tilera is 36, and top layer encoder distributes 6 frame encoder threads, and each core distribution obtains a worker thread, therefore one has 36 worker threads.
(2) for the present frame of each frame encoder thread process; frame encoder thread is distributed to task queue by coding task and the filter task of each row in present frame; worker thread reads task queue Bing YiCTUWei unit reading of task is processed; and the coding task of each row and filter task carry out wavefront parallel processing, completes the HEVC coding of this frame.
The coding task process of processing each row in step (2) in the process of wavefront parallel processing is as follows:
(2-1) judge that whether current line is the first row, proceeds as follows according to judged result:
(a), if current line is the first row, if the CTU quantity that current line completes coding task is more than or equal to the threshold value of setting, the coding task of next line is distributed in task queue;
(b) if current line is not the first row, make the following judgment:
If current line completes the threshold value that the CTU quantity of coding task and the difference of the CTU quantity that lastrow completes coding task are less than or equal to setting, stop the coding task of current line, worker thread reads task queue again;
If current line completes the threshold value that the CTU quantity of coding task is more than or equal to setting, and not last column, the coding task of next line is distributed in task queue, and continue to process the coding task of this row;
If current line completes the CTU quantity of coding task and is more than or equal to the threshold value of setting, and is last column, continue to process the coding task of this row;
(2-2), after the coding task of current line is finished dealing with, compare the size of the line number of current line and the filter delay line number of setting:
If the line number of current line is greater than the filter delay line number of setting, the filter task that the line number that is current line by line number deducts the row of filter delay line number is added into task queue, and worker thread reads task queue again;
Otherwise worker thread reads task queue again.
The filter task process of processing each row in step (2) in the process of wavefront parallel processing is as follows:
If current line completes the threshold value that the CTU quantity of filter task and the difference of the CTU quantity that lastrow completes filter task are less than setting, stop the filter task of current line, worker thread reads task queue again;
If current line completes the threshold value that the CTU quantity of filter task is more than or equal to setting, and not last column, the filter task of next line is distributed in task queue, and continue to process the filter task of this row;
If current line completes the CTU quantity of coding task and is more than or equal to the threshold value of setting, and is last column, continue to process the filter task of this row.
Wherein, filter delay line number is 0~2 (in the present embodiment for be 2).
The threshold value of setting is 2~4 (being 2 in the present embodiment).
Choose the sequence of 5 different resolutions representative in the cycle tests that JCTVC provides:
A_Traffic_2560x1600_30 (being Traffic sequence), B_Kimono1_1920x1080_24 (being designated as Kimono1 sequence), C_BasketballDrill_832x480_50 (being designated as BasketballDrill sequence), D_RaceHorses_416x240_30 (being designated as RaceHorses sequence) and vidyo1_720p_60 (being designated as vidyo1 sequence) test, under same the encoding option (determining QP32), use respectively TileraGX (1.2GHz) monokaryon serial code and 36 core parallel encodings, efficiency contrast is as shown in table 1, wherein PSNR (PeakSignal to Noise Ratio) is Y-PSNR, the quality of presentation code, the ratio of frame per second when frame per second when speed-up ratio is parallel and serial.
Table 1
As can be seen from Table 1, while adopting the HEVC coding method of the present embodiment, code check slightly raises, and in the situation that quality slightly declines, coding rate (frame per second) has had large increase, average speedup to be about 10 times.Multiframe parallel processing and WPP work in coordination, and can fully play in the ideal case the advantage of Tilera polycaryon processor, greatly improve the efficiency of HEVC coding.
Disclosed is above only specific embodiments of the invention; but protection scope of the present invention is not limited to this; any those of ordinary skill in the art can carry out various changes and modification and not depart from the spirit and scope of the present invention the present invention, within all should being encompassed in the protection range of this practicality invention.

Claims (8)

1. the HEVC coding method based on polycaryon processor Tilera, is characterized in that, comprising:
(1) according to the check figure of polycaryon processor Tilera, by top layer encoder, distribute several frame encoder threads, for parallel processing different frame, distribute a plurality of worker threads to be used for encoding and filtering simultaneously;
(2) for the present frame of each frame encoder thread process; frame encoder thread is distributed to task queue by coding task and the filter task of each row in present frame; worker thread reads task queue Bing YiCTUWei unit reading of task is processed; and the coding task of each row and filter task carry out wavefront parallel processing, completes the HEVC coding of this frame.
2. the HEVC coding method based on polycaryon processor Tilera as claimed in claim 1, is characterized in that, in described step (1), the number of frame encoder thread is 1~6.
3. the HEVC coding method based on polycaryon processor Tilera as claimed in claim 2, is characterized in that, in described step (1), in polycaryon processor Tilera, each core distribution obtains a worker thread.
4. the HEVC coding method based on polycaryon processor Tilera as claimed in claim 3, is characterized in that, the coding task process of processing each row in described step (2) in the process of wavefront parallel processing is as follows:
(2-1) judge that whether current line is the first row, proceeds as follows according to judged result:
(a), if current line is the first row, if the CTU quantity that current line completes coding task is more than or equal to the threshold value of setting, the coding task of next line is distributed in task queue;
(b) if current line is not the first row, make the following judgment:
If current line completes the threshold value that the CTU quantity of coding task and the difference of the CTU quantity that lastrow completes coding task are less than or equal to setting, stop the coding task of current line, worker thread reads task queue again;
If current line completes the threshold value that the CTU quantity of coding task is more than or equal to setting, and not last column, the coding task of next line is distributed in task queue, and continue to process the coding task of this row;
If current line completes the CTU quantity of coding task and is more than or equal to the threshold value of setting, and is last column, continue to process the coding task of this row;
(2-2), after the coding task of current line is finished dealing with, compare the size of the line number of current line and the filter delay line number of setting:
If the line number of current line is greater than the filter delay line number of setting, the filter task that the line number that is current line by line number deducts the row of filter delay line number is added into task queue, and worker thread reads task queue again;
Otherwise worker thread reads task queue again.
5. the HEVC coding method based on polycaryon processor Tilera as claimed in claim 4, is characterized in that, the filter task process of processing each row in described step (2) in the process of wavefront parallel processing is as follows:
If current line completes the threshold value that the CTU quantity of filter task and the difference of the CTU quantity that lastrow completes filter task are less than setting, stop the filter task of current line, worker thread reads task queue again;
If current line completes the threshold value that the CTU quantity of filter task is more than or equal to setting, and not last column, the filter task of next line is distributed in task queue, and continue to process the filter task of this row;
If current line completes the CTU quantity of coding task and is more than or equal to the threshold value of setting, and is last column, continue to process the filter task of this row.
6. the HEVC coding method based on polycaryon processor Tilera as claimed in claim 4, is characterized in that, described filter delay line number is 0~2.
7. the HEVC coding method based on polycaryon processor Tilera as claimed in claim 6, is characterized in that, described filter delay line number is 2.
8. the HEVC coding method based on polycaryon processor Tilera as described in any one claim in claim 4~7, is characterized in that, described threshold value is 2~4.
CN201410194175.0A 2014-05-08 2014-05-08 HEVC coding method based on multi-core processor Tilera Expired - Fee Related CN103974081B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410194175.0A CN103974081B (en) 2014-05-08 2014-05-08 HEVC coding method based on multi-core processor Tilera

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410194175.0A CN103974081B (en) 2014-05-08 2014-05-08 HEVC coding method based on multi-core processor Tilera

Publications (2)

Publication Number Publication Date
CN103974081A true CN103974081A (en) 2014-08-06
CN103974081B CN103974081B (en) 2017-03-22

Family

ID=51243052

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410194175.0A Expired - Fee Related CN103974081B (en) 2014-05-08 2014-05-08 HEVC coding method based on multi-core processor Tilera

Country Status (1)

Country Link
CN (1) CN103974081B (en)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104202602A (en) * 2014-08-18 2014-12-10 三星电子(中国)研发中心 Device and method of executing video coding
CN104506867A (en) * 2014-12-01 2015-04-08 北京大学 Sample adaptive offset parameter estimation method and device
CN104519360A (en) * 2014-12-30 2015-04-15 中国科学院重庆绿色智能技术研究院 Compression method based on HEVC
CN105791829A (en) * 2016-03-30 2016-07-20 南京邮电大学 HEVC parallel intra-frame prediction method based on multi-core platform
CN105992008A (en) * 2016-03-30 2016-10-05 南京邮电大学 Multilevel multitask parallel decoding algorithm on multicore processor platform
CN107197296A (en) * 2017-06-22 2017-09-22 华中科技大学 A kind of HEVC parallel encoding method and systems based on COStream
CN107483950A (en) * 2016-06-07 2017-12-15 北京大学 Picture parallel encoding method and system
CN108449603A (en) * 2018-03-22 2018-08-24 南京邮电大学 Based on the multi-level task level of multi-core platform and the parallel HEVC coding/decoding methods of data level
CN110337002A (en) * 2019-08-15 2019-10-15 南京邮电大学 The multi-level efficient parallel decoding algorithm of one kind HEVC in multi-core processor platform
CN110446043A (en) * 2019-08-08 2019-11-12 南京邮电大学 A kind of HEVC fine grained parallel coding method based on multi-core platform
WO2024098821A1 (en) * 2022-11-11 2024-05-16 上海哔哩哔哩科技有限公司 Av1 filtering method and apparatus

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090304087A1 (en) * 2008-06-04 2009-12-10 Youji Shibahara Frame coding and field coding judgment method, image coding method, image coding apparatus, and program
CN102098503A (en) * 2009-12-14 2011-06-15 中兴通讯股份有限公司 Method and device for decoding image in parallel by multi-core processor
EP2445211A1 (en) * 2009-06-18 2012-04-25 ZTE Corporation Multi-core image encoding processing device and image filtering method thereof

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090304087A1 (en) * 2008-06-04 2009-12-10 Youji Shibahara Frame coding and field coding judgment method, image coding method, image coding apparatus, and program
EP2445211A1 (en) * 2009-06-18 2012-04-25 ZTE Corporation Multi-core image encoding processing device and image filtering method thereof
CN102098503A (en) * 2009-12-14 2011-06-15 中兴通讯股份有限公司 Method and device for decoding image in parallel by multi-core processor

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104202602A (en) * 2014-08-18 2014-12-10 三星电子(中国)研发中心 Device and method of executing video coding
CN104506867B (en) * 2014-12-01 2017-07-21 北京大学 Sample point self-adapted offset parameter method of estimation and device
CN104506867A (en) * 2014-12-01 2015-04-08 北京大学 Sample adaptive offset parameter estimation method and device
CN104519360A (en) * 2014-12-30 2015-04-15 中国科学院重庆绿色智能技术研究院 Compression method based on HEVC
CN104519360B (en) * 2014-12-30 2019-03-01 中国科学院重庆绿色智能技术研究院 A kind of compression method based on HEVC
CN105992008B (en) * 2016-03-30 2019-08-30 南京邮电大学 A kind of multi-level multi-task parallel coding/decoding method in multi-core processor platform
CN105992008A (en) * 2016-03-30 2016-10-05 南京邮电大学 Multilevel multitask parallel decoding algorithm on multicore processor platform
CN105791829B (en) * 2016-03-30 2019-05-03 南京邮电大学 A kind of parallel intra-frame prediction method of HEVC based on multi-core platform
CN105791829A (en) * 2016-03-30 2016-07-20 南京邮电大学 HEVC parallel intra-frame prediction method based on multi-core platform
CN107483950A (en) * 2016-06-07 2017-12-15 北京大学 Picture parallel encoding method and system
CN107197296A (en) * 2017-06-22 2017-09-22 华中科技大学 A kind of HEVC parallel encoding method and systems based on COStream
CN107197296B (en) * 2017-06-22 2019-08-13 华中科技大学 A kind of HEVC parallel encoding method and system based on COStream
CN108449603A (en) * 2018-03-22 2018-08-24 南京邮电大学 Based on the multi-level task level of multi-core platform and the parallel HEVC coding/decoding methods of data level
CN108449603B (en) * 2018-03-22 2019-11-22 南京邮电大学 Based on the multi-level task level of multi-core platform and the parallel HEVC coding/decoding method of data level
CN110446043A (en) * 2019-08-08 2019-11-12 南京邮电大学 A kind of HEVC fine grained parallel coding method based on multi-core platform
CN110337002A (en) * 2019-08-15 2019-10-15 南京邮电大学 The multi-level efficient parallel decoding algorithm of one kind HEVC in multi-core processor platform
CN110337002B (en) * 2019-08-15 2022-03-29 南京邮电大学 HEVC (high efficiency video coding) multi-level parallel decoding method on multi-core processor platform
WO2024098821A1 (en) * 2022-11-11 2024-05-16 上海哔哩哔哩科技有限公司 Av1 filtering method and apparatus

Also Published As

Publication number Publication date
CN103974081B (en) 2017-03-22

Similar Documents

Publication Publication Date Title
CN103974081A (en) HEVC coding method based on multi-core processor Tilera
Das et al. Distributed deep learning using synchronous stochastic gradient descent
CN105491377B (en) A kind of video decoded macroblock grade Method of Scheduling Parallel of computation complexity perception
WO2021139173A1 (en) Ai video processing method and apparatus
JP2010527194A (en) Dynamic motion vector analysis method
CN102547289B (en) Fast motion estimation method realized based on GPU (Graphics Processing Unit) parallel
CN104615498B (en) A kind of group system dynamic load balancing method of task based access control migration
CN107357661A (en) A kind of fine granularity GPU resource management method for mixed load
US20130166749A1 (en) Optimization Of Resource Utilization In A Collection Of Devices
CN101860752B (en) Video code stream parallelization method for embedded multi-core system
CN102497550A (en) Parallel acceleration method and device for motion compensation interpolation in H.264 encoding
CN114329327B (en) Sparse matrix parallel solving method and device based on upper and lower triangular decomposition
CN105407356B (en) The real-time JPEG2000 coding/decoding methods of high speed
CN104469396A (en) Distributed transcoding system and method
CN110990154A (en) Big data application optimization method and device and storage medium
CN105791829B (en) A kind of parallel intra-frame prediction method of HEVC based on multi-core platform
CN109495743A (en) A kind of parallelization method for video coding based on isomery many places platform
CN102625108A (en) Multi-core-processor-based H.264 decoding method
WO2023160236A1 (en) Slicing method and apparatus for multi-output neural network, and chip and storage medium
CN105868000B (en) A kind of expansible data processing method of parallelization for network I/O virtualization
US10713096B2 (en) System and method for handling data skew at run time
CN104036141B (en) Open computing language (OpenCL)-based red-black tree acceleration method
CN114461356B (en) Control method for number of processes of scheduler and IaaS cloud platform scheduling system
CN110446043A (en) A kind of HEVC fine grained parallel coding method based on multi-core platform
CN102075753B (en) Method for deblocking filtration in video coding and decoding

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20170322