CN106294276B - A kind of transmitting module and its working method suitable for coarseness multinuclear computing system - Google Patents
A kind of transmitting module and its working method suitable for coarseness multinuclear computing system Download PDFInfo
- Publication number
- CN106294276B CN106294276B CN201610645200.1A CN201610645200A CN106294276B CN 106294276 B CN106294276 B CN 106294276B CN 201610645200 A CN201610645200 A CN 201610645200A CN 106294276 B CN106294276 B CN 106294276B
- Authority
- CN
- China
- Prior art keywords
- functional unit
- module
- underlying task
- filling
- underlying
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
- G06F15/163—Interprocessor communication
- G06F15/173—Interprocessor communication using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
- G06F15/17306—Intercommunication techniques
- G06F15/17318—Parallel communications techniques, e.g. gather, scatter, reduce, roadcast, multicast, all to all
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multi Processors (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a kind of transmitting modules and its working method suitable for coarseness multinuclear computing system, it is characterized in that being to be arranged in the main control unit of coarseness multinuclear computing system, main control unit includes:Scheduler module, idle function Queue module, data network moderator and Configuration network moderator;Transmitting module includes:Underlying task command memory, top-level task command memory, controller module, functional unit required list, filling module, scheduler module interface, configuring network interface, data network interface.The present invention can efficiently realize the collaboration between two-stage program, realize that functional unit is allocated according to the demand of assignment instructions, to avoid the problem that the scheme for executing assignment instructions using functional unit determining in advance may cause structural hazards and reduce system-computed performance, also the problems such as avoiding network congestion caused by using following functional element coordinate update scheme possible and the excessive consumption of logical resource, and then improve the calculated performance of system.
Description
Technical field
It is calculated the present invention relates to high density and processor scheduling field, specifically one kind is suitable for programming using two-stage
The transmitting module and its working method of framework computing system.
Background technology
Multi-core technology is because low power consumption, strong parallel processing capability and excellent calculated performance have become processor and set
The mainstream of meter.However, realizing the efficient mapping of the high efficiency communication and task between operation core, it is directly related to the calculation of multiple nucleus system
Can power be played, and is the major issue that current multiple nucleus system faces, and is referred to as " programming wall ".
In the programming of multiple nucleus system, programming difficulty can be reduced using two-stage programming framework, improve programming efficiency, it is convenient
The task scheduling of multiple nucleus system.But how to realize the collaboration between two-stage program, and how according to system mode update bottom
Functional unit information in assignment instructions is that two-stage weaves into the problem faced in architecture processor design.
Invention content
The present invention is to overcome the shortcoming of existing invention, it is proposed that a kind of hair suitable for coarseness multinuclear computing system
Module and its working method are penetrated, to can efficiently realize the collaboration between two-stage program, realizes that functional unit refers to according to task
The demand of order is allocated, to avoid the scheme for executing assignment instructions using functional unit determining in advance that from may causing structure
The problem of taking a risk and reducing system-computed performance also avoids network caused by using following functional element coordinate update scheme possible
The problems such as excessive consumption of congestion and logical resource, and then improve the calculated performance of system.
Used technical solution is the present invention in order to achieve the above objectives:
A kind of the characteristics of transmitting module suitable for coarseness multinuclear computing system of the present invention, is provided in the coarseness
In the main control unit of multinuclear computing system, the main control unit includes:Scheduler module, idle functional unit Queue module,
Data network moderator and Configuration network moderator;
The transmitting module includes:Underlying task command memory, top-level task command memory, controller module, work(
It can unit requests table, filling module, scheduler module interface, configuring network interface, data network interface;
The scheduler module interface receives top-level task instruction from the scheduler module, and the top-level task is instructed
It is cached in the top-level task command memory, after top-level task instruction receives, the scheduler module interface
Transmitting and receiving finishes signal to the controller module;
The top-level task instructs:Needed for the top-level task instruction described in the coarseness multinuclear computing system
The amount field of various functions unit, the initial address and length word that corresponding underlying task instruction is instructed with the top-level task
Section;
After receiving signal described in the controller module reception, described in top-level task command memory acquisition
Top-level task instructs, and the amount field for being included is passed to the functional unit required list;Appointed simultaneously according to the top layer
The initial address and length information of underlying task instruction included in business instruction send bottom to the configuring network interface and appoint
Business instruction request information;
The amount field that the functional unit required list is provided according to the controller module, to the idle functional unit
Queue module sends idle functional unit request, and waits the functional unit of the idle functional unit Queue module feedback to be received
Coordinate information;The functional unit required list receives the functional unit coordinate information and is stored in the work(after being classified
In energy unit requests table;
The configuring network interface receives the underlying task instruction request information that the controller module transmits and carries out
After packing, the Configuration network moderator of the main control unit is transmitted to for asking underlying task to instruct;
The data network interface receives the instruction of requested underlying task from the data network moderator, and by institute
It states in underlying task instruction buffer to the underlying task command memory;The underlying task instructs:The coarseness
The configuration information of each functional unit in multinuclear computing system;
The underlying task command memory receives the underlying task that the data network interface transmits and instructs and delayed
After depositing, sends caching and finish signal to the controller module;
The controller module receives after the caching finishes signal, sends read signal and filling signal to described respectively
Underlying task command memory and the filling module;
After the underlying task command memory receives the reading signal, the underlying task is instructed and is read, and
Send the filling module to;
The filling module receives the underlying task that the filling signal and the underlying task command memory transmit and refers to
After order, the functional unit type that the configuration information of each functional unit is included in being instructed according to underlying task, by the configuration
Information is assigned in the filling module and is filled;Required functional unit coordinate in the filling process is filled out by described
Mold filling block sends inquiry request to the functional unit required list, and obtained from the functional unit required list;
After the configuration information end-of-fill of each functional unit of underlying task instruction, the filling module ties filling
Underlying task instruction after beam is sent to the configuring network interface, and waits each work(of next underlying task instruction to be received
The configuration information of energy unit;
The configuring network interface receives the underlying task after the end-of-fill and instructs and be packaged, to forward
To the Configuration network moderator of the main control unit, for completing issuing for underlying task instruction.
A kind of working method of transmitting module suitable for coarseness multinuclear computing system of the invention is applied to described thick
In the main control unit of granularity multinuclear computing system, the main control unit includes:Scheduler module, idle functional unit queue mould
Block, data network moderator and Configuration network moderator;The transmitting module includes:Functional unit required list, filling module;Its
Feature is,
The working method of the transmitting module is to carry out as follows:
Step 1, the transmitting module receive top-level task from the scheduler module and instruct and cached;The top
Layer assignment instructions include:The required various functions unit of top-level task instruction described in the coarseness multinuclear computing system
Amount field, the initial address and length field that corresponding underlying task instruction is instructed with the top-level task;
Step 2, when the top-level task instruction receive after, the transmitting module by the top-level task instruction wrapped
The amount field contained passes to the functional unit required list;Appointed simultaneously according to bottom included in top-level task instruction
The initial address and length information of business instruction send underlying task instruction request information for asking to the Configuration network moderator
Underlying task is asked to instruct;
The amount field that step 4, the functional unit required list are provided according to the transmitting module, to the idle function
Cell queue module sends idle functional unit request, and waits the function of the idle functional unit Queue module feedback to be received
Unit coordinate information;
Step 5, the functional unit required list receive the functional unit coordinate information and are stored in after being classified
In the functional unit required list;
Step 6, the transmitting module receive requested underlying task instruction from the data network moderator and go forward side by side
Row caching;The underlying task instructs:The configuration information of each functional unit in the coarseness multinuclear computing system;
Step 7, when the transmitting module complete underlying task instruction reception after, send filling signal to the fill mould
Block;Underlying task instruction is also sent to the filling module simultaneously;
After step 8, the filling module receive the filling signal and underlying task instruction, referred to according to underlying task
The configuration information is assigned to the fill mould by the functional unit type that the configuration information of each functional unit is included in order
It is filled in block;Required functional unit coordinate in the filling process is from the filling module to the functional unit
Required list sends inquiry request, and obtained from the functional unit required list;
Step 9, after the configuration information end-of-fill of each functional unit of underlying task instruction, filling module general
Underlying task instruction after end-of-fill is sent to the Configuration network moderator after being packaged, and refers to for completing underlying task
What is enabled issues.
Compared with prior art, advantageous effects of the invention are embodied in:
1, top-level task is instructed the functional unit coordinate obtained in task scheduling by the present invention in task instruction issue
Etc. information be directly filled into underlying task instruction in, to realize two-stage program refer between collaboration, make to refer to by top-level task
The process being scheduled is enabled to instruct the process for carrying out functional unit mission dispatching to connect with by underlying task, to make two
Top-level task instruction facilitates the advantage of task scheduling to emerge from grade programming framework;
2, compared with the scheme for executing assignment instructions using functional unit coordinate determining in advance in the prior art, the present invention
Module and functional unit required list are filled by design, realizes the dynamic allocation of functional unit, i.e., when having task to refer in system
Required for enabling when the functional unit of type, the assignment instructions can be both assigned them to, without waiting for work(determining in advance
Energy unit free, thus greatly reduces the possibility of computing system recurring structure venture;By the position of functional unit and function
The function and type of unit decouple, and programmer is made whether need not to pay close attention to each functional unit when writing program in idle shape
State improves programming efficiency to significantly reduce programming difficulty;
3, compared with using following functional element coordinate update scheme in the prior art, the present invention is issued assignment instructions
Merge in same step with the distribution of functional unit, and be spatially limited in main control unit, is thus led
Controller only needs to send a configuration information to corresponding functional unit in an assignment instructions, reduces network and gathers around
Stifled possibility;Functional unit in computing system does not need design update coordinate and operates required corresponding hardware logic,
And only need that respective logic is arranged in main control unit, to reduce the task amount that functional unit designs in computing system,
Consider that hardware logic resource consumption can control, and functional unit number in computing system there are one main control units simultaneously
Measure more, each functional unit, which increases flush logic, can greatly increase the resource consumption of entire computing system, and the present invention passes through
The design for filling module, can greatly reduce the resource consumption of computing system;Finally, functional unit can be with after receiving configuration information
It is immediately performed calculating task, without waiting for refreshing information, to improve the working efficiency of computing system.
Description of the drawings
Fig. 1 for the present invention towards the structure chart using two-stage programming framework computing system;
Fig. 2 is the system-oriented work flow diagram of the present invention;
Fig. 3 for the present invention the main control unit structure chart towards coarseness multinuclear computing system;
Fig. 4 is the structure chart of transmitting module of the present invention;
Fig. 5 is two-stage programming framework schematic diagram of the present invention.
Specific implementation mode
In this example implementation, a kind of transmitting module suitable for coarseness multinuclear computing system is arranged in coarseness multinuclear
In the main control unit of computing system, as shown in Figure 1, goal systems includes network-on-chip, main control unit, memory and a variety of
Functional unit.Wherein network-on-chip includes data network PCC for data exchange, the Configuration network that is issued for assignment instructions
The Cfg_Net and state network Status_Net collected for feedback signal, data network use packet-circuit switching mechanism, match
It sets network and uses top-down broadcast mode, state network uses summarizes pattern step by step from bottom to top;Memory MEM by
SDRAM is realized, is mainly used for instruction and data of the storage for calculating;Functional unit includes then input/output interface unit and more
Kind different type computing unit, such as conventional data batch arithmetic unit, general coprocessor and specialized hardware customize accelerator
Deng.Main control unit is mainly responsible for the scheduling of calculating task, and issues assignment instructions to memory and other functional units.Input
Output interface unit is mainly responsible for the output of the importing of source data and operation result in system, including peripheral hardware tell interface unit and
High-speed communication unit between plate.Computing unit is then responsible for completing various processor active tasks according to the assignment instructions that main control unit issues,
There can be the multiple types such as batch arithmetic unit, general processor, orientation hardware accelerator, for example wherein batch arithmetic unit use can
Reconfiguration technique is realized, large batch of multiply-add regular operation in system is responsible for;A variety of arithmetic units are integrated in general processor, are responsible for system
The operations such as the scattered operation of complexity or the impossible division of batch arithmetic unit, trigonometric function in system;Orient hardware accelerator then
The algorithm that can be frequently encountered according to system institute domain-oriented is oriented hardware-accelerated.
Whole system works under the control of main control unit, and memory and other functional units are required to from main control list
Member receives control command (being referred to as configuration information, including configuration information head and configuration information body), and is executed accordingly according to its content
Operation.After system power-on reset, master controller reads initialization information from built-in ROM, then from input/output interface list
The instruction of first incoming task and source data, are stored in memory, then take assignment instructions from memory by instruction sequences, and according to appointing
The demand of business instruction, selects corresponding functional unit to issue configuration information, it is made to execute calculating.Functional unit is executing configuration
After the work required in information, inform that master controller itself is in idle condition by state network, to be returned by master controller
It receives, can be used for continuing to execute other work.It is completed when all calculating tasks instruct, computing system passes through input/output interface list
Member output operation result.
Computing system workflow proposed in this paper includes fetching, decoding, register renaming, dynamic dispatching, emits, holds
Row writes back and submits 8 stages, as shown in Fig. 2, wherein in addition to the stage of execution completes in each functional unit, other stages are equal
It is completed in main control unit.
As shown in figure 3, main control unit includes:Fetching module, decoding module, register renaming module, branch restore
Module writes back module, submits module, free physical registers queue, free virtual register queue, state layer interface, scheduling
Module, idle functional unit Queue module, data network moderator and Configuration network moderator;
As shown in figure 4, transmitting module includes:Underlying task command memory, top-level task command memory, controller mould
Block, functional unit required list, filling module, scheduler module interface, configuring network interface, data network interface;
Functional unit required list includes memory requests table and various computing unit required lists;
Filling module includes memory filling module and various computing units filling module;
Scheduler module interface receives top-level task instruction from scheduler module, and top-level task instruction buffer to top layer is appointed
It is engaged in command memory, after top-level task instruction receives, scheduler module interface transmitting and receiving finishes signal to controller
Module;
Top-level task instructs:Various functions unit in coarseness multinuclear computing system needed for top-level task instruction
Amount field, instruct with top-level task the initial address and length field of the instruction of corresponding underlying task;As shown in figure 5, top
The initial address underlying task corresponding with length field direction that layer assignment instructions are instructed by the underlying task of storage instructs.
Controller module receives receive signal after, from top-level task command memory obtain top-level task instruction,
And the information field of the storage channel quantity for the required by task for being included and various computing unit quantity is passed into function
Unit requests table;Simultaneously according to top-level task instruct included in underlying task instruction initial address and length information, to
Configuring network interface sends underlying task instruction request information;
The storage channel quantity and various computing unit quantity words that functional unit required list is provided according to controller module
Section sends idle functional unit request to idle functional unit Queue module, and waits idle functional unit Queue module to be received
The functional unit coordinate information of feedback;Functional unit required list receives functional unit coordinate information and is stored in after being classified
In functional unit required list;
After configuring network interface receives the underlying task instruction request information that controller module transmits and is packaged, turn
The Configuration network moderator of main control unit is issued for asking underlying task to instruct;
Data network interface receives requested underlying task instruction from data network moderator, and underlying task is referred to
Order is cached in underlying task command memory;Underlying task instructs:Each function list in coarseness multinuclear computing system
The configuration information of member, as shown in figure 5, underlying task instruction be used in the assignment instructions to each functional unit match
Confidence breath is sequentially arranged, and finally with occupy-place coding line be filled into coarseness multinuclear computing system used in SDRAM
Burst-length integral multiple;The configuration information of each functional unit includes configuration information head and configuration information body two parts;
After the underlying task that underlying task command memory reception data network interface transmits is instructed and cached, send
Caching finishes signal to controller module;
Controller module receives after caching finishes signal, sends respectively and reads signal and filling signal and give underlying task instruction
Memory and filling module;
After underlying task command memory receives reading signal, underlying task is instructed and is read, and sends fill mould to
Block;
After filling module receives the underlying task instruction that filling signal and underlying task command memory transmit, according to bottom
The functional unit type that configuration information head is included in the configuration information of each functional unit in assignment instructions, by configuration information point
It sends in filling module and is filled;Required functional unit coordinate in the filling process is from filling module to function list
First required list sends inquiry request, and obtained from functional unit required list;
The filling process of the configuration information of each functional unit of underlying task instruction carries out successively in order, when one
After a functional unit completes filling process, next functional unit in underlying task instruction starts padding, Zhi Daoren
All functional units of business instruction complete filling process, then the configuration information padding of underlying task instruction terminates;
After the configuration information end-of-fill of each functional unit of underlying task instruction, filling module will be after end-of-fill
Underlying task instruction be sent to configuring network interface, and wait each functional unit of next underlying task instruction to be received
Configuration information;
Configuring network interface receives the underlying task after end-of-fill and instructs and be packaged, to be transmitted to main control
The Configuration network moderator of unit, for completing issuing for underlying task instruction.
In this example implementation, a kind of working method of transmitting module suitable for coarseness multinuclear computing system is application
In the main control unit of coarseness multinuclear computing system, main control unit includes:Scheduler module, idle functional unit queue mould
Block, data network moderator and Configuration network moderator;Transmitting module includes:Functional unit required list, filling module;
The working method of transmitting module is to carry out as follows:
Step 1, the slave scheduler module of transmitting module receive top-level task and instruct and cached;Top-level task instruction packet
Contain:The amount field and top-level task of various functions unit in coarseness multinuclear computing system needed for top-level task instruction
Instruct the initial address and length field of corresponding underlying task instruction;
Step 2, after top-level task instruction receives, top-level task is instructed included amount field by transmitting module
Pass to functional unit required list;Simultaneously according to top-level task instruct included in underlying task instruction initial address and length
Information is spent, underlying task instruction request information is sent for asking underlying task to instruct to Configuration network moderator;
The amount field that step 4, functional unit required list are provided according to transmitting module, to idle functional unit Queue module
Idle functional unit request is sent, and waits the functional unit coordinate information of idle functional unit Queue module feedback to be received;
Step 5, functional unit required list receive functional unit coordinate information and are stored in functional unit after being classified
In required list;
Step 6, transmitting module receive requested underlying task from data network moderator and instruct and cached;Bottom
Layer assignment instructions include:The configuration information of each functional unit in coarseness multinuclear computing system;
Step 7, when transmitting module complete underlying task instruction reception after, send filling signal give filling module;Simultaneously
Underlying task instruction is also sent to filling module;
After step 8, filling module receive filling signal and underlying task instruction, each function in being instructed according to underlying task
Configuration information is assigned in filling module and is filled by the functional unit type that the configuration information of unit is included;It is filling
Required functional unit coordinate in the process is to send inquiry request from filling module to functional unit required list, and from function
It is obtained in unit requests table;
Step 9, underlying task instruction each functional unit configuration information end-of-fill after, filling module will fill
After underlying task instruction be packaged after be sent to Configuration network moderator, for complete underlying task instruction under
Hair.
Claims (2)
1. a kind of transmitting module suitable for coarseness multinuclear computing system, it is characterized in that setting is calculated in the coarseness multinuclear
In the main control unit of system, the main control unit includes:Scheduler module, idle functional unit Queue module, data network
Moderator and Configuration network moderator;
The transmitting module includes:Underlying task command memory, top-level task command memory, controller module, function list
First required list, filling module, scheduler module interface, configuring network interface, data network interface;
The scheduler module interface receives top-level task instruction from the scheduler module, and by the top-level task instruction buffer
Into the top-level task command memory, after top-level task instruction receives, the scheduler module interface is sent
Signal is received to the controller module;
The top-level task instructs:It is various needed for the top-level task instruction described in the coarseness multinuclear computing system
The amount field of functional unit, the initial address and length field that corresponding underlying task instruction is instructed with the top-level task;
After receiving signal described in the controller module reception, the top layer is obtained from the top-level task command memory
Assignment instructions, and the amount field for being included is passed into the functional unit required list;Referred to simultaneously according to the top-level task
The initial address and length information that underlying task included in order instructs send underlying task to the configuring network interface and refer to
Enable solicited message;
The amount field that the functional unit required list is provided according to the controller module, to the idle functional unit queue
Module sends idle functional unit request, and waits the functional unit coordinate of the idle functional unit Queue module feedback to be received
Information;The functional unit required list receives the functional unit coordinate information and is stored in the function list after being classified
In first required list;
The configuring network interface receives the underlying task instruction request information that the controller module transmits and is packaged
Afterwards, the Configuration network moderator of the main control unit is transmitted to for asking underlying task to instruct;
The data network interface receives the instruction of requested underlying task from the data network moderator, and by the bottom
Layer assignment instructions are cached in the underlying task command memory;The underlying task instructs:The coarseness multinuclear
The configuration information of each functional unit in computing system;
The underlying task command memory receives after the underlying task that the data network interface transmits instructs and cached,
It sends caching and finishes signal to the controller module;
The controller module receives after the caching finishes signal, sends read signal and filling signal to the bottom respectively
Assignment instructions memory and the filling module;
After the underlying task command memory receives the reading signal, the underlying task is instructed and is read, and is transmitted
To the filling module;
After the filling module receives the underlying task instruction that the filling signal and the underlying task command memory transmit,
The functional unit type that the configuration information of each functional unit is included in being instructed according to underlying task, by the configuration information point
It sends in the filling module and is filled;Required functional unit coordinate in the filling process is by the filling module
Inquiry request is sent to the functional unit required list, and obtained from the functional unit required list;
After the configuration information end-of-fill of each functional unit of underlying task instruction, the filling module will be after end-of-fill
Underlying task instruction be sent to the configuring network interface, and wait each function list of next underlying task instruction to be received
The configuration information of member;
The configuring network interface receives the underlying task after the end-of-fill and instructs and be packaged, to be transmitted to
The Configuration network moderator for stating main control unit, for completing issuing for underlying task instruction.
2. a kind of working method of transmitting module suitable for coarseness multinuclear computing system is to be applied to the coarseness multinuclear
In the main control unit of computing system, the main control unit includes:Scheduler module, idle functional unit Queue module, data
Network arbiter and Configuration network moderator;The transmitting module includes:Functional unit required list, filling module;It is characterized in that
The working method of the transmitting module is to carry out as follows:
Step 1, the transmitting module receive top-level task from the scheduler module and instruct and cached;The top layer is appointed
Business instructs:The quantity of various functions unit needed for the top-level task instruction described in the coarseness multinuclear computing system
Field, the initial address and length field that corresponding underlying task instruction is instructed with the top-level task;
Step 2, when the top-level task instruction receive after, the transmitting module by the top-level task instruction is included
Amount field passes to the functional unit required list;Referred to simultaneously according to underlying task included in top-level task instruction
The initial address and length information of order send underlying task instruction request information for asking bottom to the Configuration network moderator
Layer assignment instructions;
The amount field that step 4, the functional unit required list are provided according to the transmitting module, to the idle functional unit
Queue module sends idle functional unit request, and waits the functional unit of the idle functional unit Queue module feedback to be received
Coordinate information;
Step 5, the functional unit required list receive the functional unit coordinate information and are stored in after being classified described
In functional unit required list;
Step 6, the transmitting module receive requested underlying task from the data network moderator and instruct and delayed
It deposits;The underlying task instructs:The configuration information of each functional unit in the coarseness multinuclear computing system;
Step 7, when the transmitting module complete underlying task instruction reception after, send filling signal to the filling module;
Underlying task instruction is also sent to the filling module simultaneously;
After step 8, the filling module receive the filling signal and underlying task instruction, in being instructed according to underlying task
The configuration information is assigned in the filling module by the functional unit type that the configuration information of each functional unit is included
It is filled;Required functional unit coordinate in the filling process is to be asked from the filling module to the functional unit
Table sends inquiry request, and obtained from the functional unit required list;
Step 9, underlying task instruction each functional unit configuration information end-of-fill after, the filling module will filling
After underlying task instruction be packaged after be sent to the Configuration network moderator, for complete underlying task instruction
It issues.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610645200.1A CN106294276B (en) | 2016-08-08 | 2016-08-08 | A kind of transmitting module and its working method suitable for coarseness multinuclear computing system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610645200.1A CN106294276B (en) | 2016-08-08 | 2016-08-08 | A kind of transmitting module and its working method suitable for coarseness multinuclear computing system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106294276A CN106294276A (en) | 2017-01-04 |
CN106294276B true CN106294276B (en) | 2018-11-06 |
Family
ID=57667243
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610645200.1A Active CN106294276B (en) | 2016-08-08 | 2016-08-08 | A kind of transmitting module and its working method suitable for coarseness multinuclear computing system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106294276B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6247122B1 (en) * | 1998-12-02 | 2001-06-12 | Ip-First, L.L.C. | Method and apparatus for performing branch prediction combining static and dynamic branch predictors |
US6898699B2 (en) * | 2001-12-21 | 2005-05-24 | Intel Corporation | Return address stack including speculative return address buffer with back pointers |
CN1675626A (en) * | 2002-08-12 | 2005-09-28 | 皇家飞利浦电子股份有限公司 | Instruction cache way prediction for jump targets |
CN101770358A (en) * | 2010-02-10 | 2010-07-07 | 北京龙芯中科技术服务中心有限公司 | System and method for processing jump instruction of microprocessor in branch prediction way |
-
2016
- 2016-08-08 CN CN201610645200.1A patent/CN106294276B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6247122B1 (en) * | 1998-12-02 | 2001-06-12 | Ip-First, L.L.C. | Method and apparatus for performing branch prediction combining static and dynamic branch predictors |
US6898699B2 (en) * | 2001-12-21 | 2005-05-24 | Intel Corporation | Return address stack including speculative return address buffer with back pointers |
CN1675626A (en) * | 2002-08-12 | 2005-09-28 | 皇家飞利浦电子股份有限公司 | Instruction cache way prediction for jump targets |
CN101770358A (en) * | 2010-02-10 | 2010-07-07 | 北京龙芯中科技术服务中心有限公司 | System and method for processing jump instruction of microprocessor in branch prediction way |
Also Published As
Publication number | Publication date |
---|---|
CN106294276A (en) | 2017-01-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10552222B2 (en) | Task scheduling method and apparatus on heterogeneous multi-core reconfigurable computing platform | |
US10120728B2 (en) | Graphical processing unit (GPU) implementing a plurality of virtual GPUs | |
CN103543954B (en) | A kind of data storage and management method and device | |
CN103778013B (en) | A kind of multichannel Nand Flash controllers and its control method | |
US20190354317A1 (en) | Operation instruction scheduling method and apparatus for nand flash memory device | |
CN106155814B (en) | A kind of reconfigurable arithmetic unit that supporting multiple-working mode and its working method | |
US11150720B2 (en) | Systems and methods for power management of hardware utilizing virtual multilane architecture | |
CN110347635A (en) | A kind of heterogeneous polynuclear microprocessor based on multilayer bus | |
CN103246625B (en) | A kind of method of data and address sharing pin self-adaptative adjustment memory access granularity | |
CN108279927B (en) | Multi-channel instruction control method and system capable of adjusting instruction priority and controller | |
CN109144710A (en) | Resource regulating method, device and computer readable storage medium | |
WO2012052773A1 (en) | Data processing systems | |
GB2575291A (en) | Gateway to Gateway Synchronisation | |
CN106557442B (en) | A kind of chip system | |
US11675633B2 (en) | Virtualised gateways | |
US11010208B1 (en) | Sync groupings | |
US11327813B2 (en) | Sync group selection | |
CN110399221A (en) | Data processing method, system and terminal device | |
US10659396B2 (en) | Joining data within a reconfigurable fabric | |
US20180212894A1 (en) | Fork transfer of data between multiple agents within a reconfigurable fabric | |
WO2021115149A1 (en) | Neural network processor, chip and electronic device | |
CN106294276B (en) | A kind of transmitting module and its working method suitable for coarseness multinuclear computing system | |
US20230063751A1 (en) | A processor system and method for increasing data-transfer bandwidth during execution of a scheduled parallel process | |
US11899970B2 (en) | Storage system and method to perform workload associated with a host | |
CN110222410A (en) | A kind of electromagnetic environment emulation method based on Hadoop MapReduce |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |