Summary of the invention
To be solved by this invention is that there are IP kernel testing efficiency in 3D NoC is low for existing 3D NoC test dispatching method
The problem of, a kind of 3D NoC test dispatching method based on Petri network and IFA is provided.
To solve the above problems, the present invention is achieved by the following technical solutions:
3D NoC test dispatching method based on Petri network and IFA comprising steps are as follows:
Step 1 establishes HCTPN model according to 3D NoC test dispatching process on the basis of prototype Petri network, determines just
The mark that begins is identified with termination, and calculates input matrix and output matrix;
Step 2, the coding mode according to formula 1., and a chaos sequence is generated by cube chaotic maps formula, then by chaos
Sequence maps to purpose-function space and forms initial firefly population, wherein each firefly individual in initial firefly population
Indicate a kind of scheduling scheme;
In formula, XTAMIndicate the coding of TAM selection individual, BiIndicate the selected TAM number of i-th of IP kernel, 1≤i≤N;
S indicates the coding of priority individual, AjbThe IP kernel for indicating that test prioritization is b on j-th strip TAM is numbered, and 1≤j≤M, 1≤b≤
N-M+1;N is the quantity of IP kernel, and M is the quantity of TAM;
Step 3, judge initial firefly population firefly individual validity:
If firefly individual all in initial firefly population is effective, i.e., there is no do not chosen by any IP kernel
TAM then goes to step 4;
If at least one firefly individual is invalid in initial firefly population, i.e. presence is not chosen by any IP kernel
TAM then goes to step 2;
All firefly individuals in firefly population are all transformed in HCTPN model transition generation accordingly by step 4
Sequence σ such as formula is 2. shown;
In formula, transitionIt indicates that IP kernel i starts to test on TAM j, changesIndicate that IP kernel i completes to survey on TAM j
Examination, 1≤i≤N, N are the quantity of IP kernel, and 1≤j≤M, M are the quantity of TAM;
Step 5, for each of HCTPN model transition firing sequence, the change is successively excited since initial marking
It moves and corresponding timed transition in sequence occurs, and calculate the time delay of timed transition according to fitness function and routing algorithm, it is each
After the completion of a timed transition excitation, status indicator and cumulative transition time delay are updated, until each tax time-varying in transition firing sequence
Excitation of moving the capital to another place is completed and reaches termination mark, and total transition time delay of transition firing sequence is that the system of the scheduling scheme is surveyed at this time
Try the time;
All transition firing sequences in HCTPN model are all inversely transformed into firefly individual by step 6, and by sequence
Absolute brightness of total transition time delay as firefly individual;
Step 7, the position vector that all firefly individuals are updated according to glowworm swarm algorithm, it may be assumed that calculating firefly kind first
Descartes's distance in group between any 2 fireflies individual, then by firefly individual absolute brightness comparison come
It determines firefly individual moving direction, then calculates the mutual attractive force between firefly individual, finally based on firefly individual
Between mutual attractive force, the firefly individual for making brightness big leans on according to location update formula to the small firefly individual of brightness
Hold together;
Step 8 judges whether the population diversity of current firefly population is greater than the lower limit value being previously set:
If so, going to step 9;
Otherwise, a variation individual is first generated all in accordance with differential evolution algorithm to it to each firefly individual, and makes firefly
Fireworm individual intersects generation test individual according to certain probability with the variation individual, then in target firefly individual and to examination
It tests between individual and carries out non-value-added greedy selection and retain to the next generation;
Step 9 judges whether current iteration number reaches the number of iterations being previously set:
If reaching, terminate searching process and export optimal firefly individual and its corresponding optimal transition firing sequence and
Best test dispatching scheme;
Otherwise, then the number of iterations is enabled to add 1, and return step 4 is iterated again.
In above-mentioned steps 2, before first time initializes, it is also necessary to which the glowworm swarm algorithm parameter of setting, the firefly are calculated
Method parameter includes maximum number of iterations, population scale, step factor, greatest attraction forces, the absorption coefficient of light and the coefficient of variation and friendship
Pitch coefficient.
In above-mentioned steps 5, fitness function are as follows:
In formula,Wi,jIt is upper in HCTPN model to indicate that i-th of IP kernel returns
Layer skeleton pattern reselects the delay of j-th strip TAM;DIi,jIt is expended when representing i-th of IP kernel selection j-th strip TAM transmission total
Duration, DIi,j=Tcore i+Ttrans i, Tcore iExpression carries out IP kernel itself to test spent duration, Ttrans iIndicate test
The routing duration of data packet;1≤i≤N, N are the quantity of IP kernel;1≤j≤M, M are the quantity of TAM.
In above-mentioned steps 5, detailed process is as follows for routing algorithm:
Step 5.1, the position coordinates for determining source node and destination node;
Step 5.2 transmits source node up to identical as the X-coordinate of destination node according to X-direction, if passing through during this
The fringe node being connected directly is crossed, then is jumped using the interconnection line between fringe node, is otherwise transmitted according to original route;
Step 5.3 transmits source node up to identical as the Y-coordinate of destination node according to Y direction, if passing through during this
The fringe node being connected directly is crossed, then is jumped using the interconnection line between fringe node, is otherwise transmitted according to original route;
Step 5.4 transmits source node up to identical as the Z coordinate of destination node according to Z-direction, if passing through during this
The fringe node being connected directly is crossed, then is jumped using the interconnection line between fringe node, is otherwise transmitted according to original route.
Compared with prior art, the present invention is based on level Colored Timed Petri Nets (HCTPN) and improvement glowworm swarm algorithm
(IFA) the 3D NoC test dispatching method combined, has a characteristic that
1, system call process and local test details are described by HCTPN model layers, makes the behavioral test body of IP kernel
The now circulation for coloring Tokken in a model has provided for observation in real time with analysis system test status, network congestion degree
Effect means.When coloring thought is also introduced in modeling process simultaneously and being assigned, the concept of inhibitor arc, successively play compressed web scale and
Enhance the effect of model descriptive power.
2, in order to optimize the route transmission process in HCTPN model, edge in present invention combination 3D Torus topological structure
There are the advantages of plurality of interconnected line between node, propose a kind of improved routing algorithm: when in the routed path of test vector comprising straight
When connecing connected fringe node, which can directly be transmitted by the long interconnection line between node, reduce data with this
Packet hop count simultaneously alleviates path conflict phenomenon, to realize system test time in terms of scheduling combined aspects and routing procedure
Double optimization.
3, in order to realize that the Efficient Solution of HCTPN model, present invention combination chaos optimization method and population diversity monitor
After strategy and differential variation strategy, improve glowworm swarm algorithm, the ability for making it have Continuous optimization, and use improves
IFA algorithm optimizing is carried out to the transition firing sequence of model, to obtain, the testing time is most short, the highest test of dispatching efficiency
Optimizing scheduling scheme effectively improves the testing efficiency of IP kernel in 3D NoC.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific example, to the present invention
It is further described.
The present invention is mainly from establishing the Visualization Model of test dispatching, realize road in conjunction with the characteristics of 3D Torus topological structure
By process optimization and model solution algorithm improvement these three aspect expansion research, distributed by double optimization test resource
With routing procedure, to solve the technical problem that IP kernel testing efficiency is low in 3D NoC.
Due to all there is interconnection line in the NoC of 3D Torus topological structure between any two end node, test available
Resource is more abundant compared with 3D Mesh topological structure, can reduce network congestion, therefore the 3D NoC of Torus structure is selected to be studied,
One scale is the Torus structure 3D NoC of 3*3*3 as shown in Figure 1, showing the composition and interconnection mode of 3D NoC in figure.
1, test dispatching strategy
In 3D NoC test process, in order to improve resource utilization, the intrinsic resource in NoC is multiplexed as test access
Mechanism (test access mechanism, TAM), is specifically exactly routing node, the communication link reused in NoC
Etc. come the test vector that transmits IP kernel to be measured.Show that two IP kernels transmit road based on the test vector of XYZ routing algorithm in Fig. 1
Diameter, a transmission paths are the TAM that can be considered that a test port is fixed.
Often there is contention for resource phenomenon during concurrent testing in IP kernel, and two IP kernels as shown in figure 1 need when testing
The support of the same router (picture × place).Therefore it is needed when a collision occurs according to certain priority to IP kernel testing sequence
It is selected, the present invention is using the simplest principle first tested that first reaches as IP kernel formulation test prioritization.When IP kernel occupies certain
After route resource, which will be occupied by lasting until the core completes test.
2, test dispatching problem describes
Based on the above Test Strategy, the present invention is studied for the test dispatching process of IP kernel, specifically to be solved to ask
Topic can be described as: how in overall arrangement system all IP kernels behavioral test, comprehensively consider resource allocation, routing conflict, survey
The influence of priority, power consumption limit to test is tried, each IP kernel is reasonably distributed parallel to progress in a limited number of TAM resources
Test, to reduce the total time of test.Test process also requires each IP kernel in system to be only tested once simultaneously, and single IP
The test process of core has continuity, and the system resource occupied unified at the end of the IP kernel is tested will discharge.
Above-mentioned problem to be solved is the NP problem in specific environment with Complex Constraints, therefore the present invention is directed to ask for this
Topic establishes test dispatching model, and improves the routing procedure of IP kernel, solves model built finally by intelligent algorithm, obtains system
Testing time is most short, the highest scheduling scheme of resource utilization.
3,3D NoC test dispatching Petri network models
A kind of research tool of the Petri network as distributed system describes sequence in system, synchronization, simultaneously by figure
Phenomena such as hair and conflict, while can be with the state change of dynamic simulator system.And it is generally existing during 3D NoC test dispatching
Parallel, serially, conflict, deadlock phenomena such as, therefore Petri network be suitable for 3D NoC test dispatching process is modeled.
It based on the complexity of 3D NoC test dispatching process, is modeled according to prototype Petri network, will lead to network planning mould
It is excessive, structure is complicated, specifically there are following difficult points: (1) IP kernel pair in terms of describing test dispatching process for basic petri net
There is randomness system interlayer relation to be caused to be difficult to describe for the selection of TAM;(2) the routing scheduling process in IP kernel test is cumbersome,
And required resource is various, causes net popularization;(3) test prioritization of IP kernel can also generate the testing time on same TAM
It influences, causes web frame complexity and scale excessive.
For the above difficult point, the present invention utilizes HCTPN model modeling, i.e., adds color on the basis of prototype Petri network
Simultaneously binding modules modeling pattern, specific idea about modeling are as follows with Delay Factor: (1) using top-down approach, initially set up
The system-level skeleton pattern of test dispatching process, then skeleton pattern is gradually decomposed into each test subnet model, and describe two layers
Correlation between model.(2) using coloring thought, different IP kernel, route resource etc., compressed web are distinguished by color set
Scale.(3) by restraining arc description IP kernel priority scheduling, effectively inhibit Petri network scale because of the uncertainty of testing sequence
And it expands.(4) corresponding time delay is changed by assigning, indicates the testing time of IP kernel.The HCTPN model that the present invention establishes is divided into
Upper layer skeleton pattern and lower layer's subnet system: skeleton pattern is for describing integrated testability scheduling process and ignoring test detail;Son
Net system describes the processes such as priority, the routing scheduling of each IP kernel in testing as a module in skeleton pattern.It is taking turns
Being changed in wide model by substitution indicates subnet system, and contact tie of the Abstract place between two-layer model.
As shown in Fig. 2, skeleton pattern is for describing selection and Parallel Scheduling process of the IP kernel to TAM.Wherein become with substitution
Moving enter_j indicates subnet system, connects and transmit information between skeleton pattern and subnet system by Abstract place.Profile die
Type operational process is as follows: having the token (i.e. different IP kernels) for carrying particular color, triggering transition initial when initial in P0
After completing initialization, token enters distribution library institute distribute, color-match is completed with corresponding TAM, to select to correspond to
TAM carry out follow-up test.The buffer area buffer_j for entering this TAM after selecting TAM_j, so that it is complete to go to subnet system
At specific test process.
As shown in figure 3, subnet system is the refinement to transition are substituted in upper layer model.The operational process of subnet system is such as
Under: the token with higher priority directly triggers transition critical and enters prior, to prevent lower priority
Token, which enters this TAM, to be tested.Changing calculate indicates router-level topology operation, for determining that the token tests institute
The route resource and transmission path needed, circular see below a trifle.TransitionIndicate that IP kernel i starts to survey on TAMj
Examination, IP kernel starts to call router resource router_s...router_k after triggering the transition.If required router is by other
IP kernel on TAM occupies, then triggers conflict transition conflict, and token returns to skeleton pattern weight from communication pool institute testfail
New distribution TAM, while token obtained by the library wait and resource being waited to discharge, relief is changed finally by triggering, routing is thought highly of
Token newly is obtained, other IP kernels that can be entered the TAM call.TransitionIndicate IP kernel i is completed on TAMj test and it is adjoint
Certain delay excites token after the transition to enter communication pool institute out return skeleton pattern, and finally reaching IP kernel test terminates
Library institute end is stored.
Transition firing sequence in HCTPN modelThere are one between test dispatching scheme
To one mapping relations.By running HCTPN model it can calculate the overall delay fitness (σ) of σ, i.e. test dispatching scheme
Testing time, it is subsequent using σ as firefly individual, can using IFA carry out optimizing, search out optimal transition firing sequence σ into
And obtain best test dispatching scheme.
4, improved routing algorithm
Above-mentioned HCTPN model indicates different IP kernels with the token image for carrying different colours, obtains test dispatching process
To intuitively portraying, at the same it is convenient describe that IP kernel is parallel, sequential testing, phenomena such as priority, conflict.
Transition calculate in model indicates router-level topology process.It will be obtained according to different routing policies different
Routed path, and the testing time of IP kernel by test vector route time and IP kernel itself testing time (fixed value) this two
Part forms, if can be reduced route time can shorten the IP kernel integrated testability time.Therefore, the present invention is in Petri net model
The router-level topology mode of router-level topology transition improves, that is, combines any two in 3D torus topological structure NoC (such as Fig. 1)
All there is the characteristic of interconnection line between end node, the improved routing algorithm based on devising a kind of routing algorithm by XYZ, to contract
The route time of short IP kernel test test vector, the process of improved routing algorithm is as shown in figure 4, specific implementation step are as follows:
Step 1, the position coordinates for determining source node and destination node;
Step 2 transmits source node up to identical as the X-coordinate of destination node according to X-direction, if passing through during this
The fringe node being connected directly then is jumped using the interconnection line between fringe node, is otherwise transmitted according to original route;
Step 3 transmits source node up to identical as the Y-coordinate of destination node according to Y direction, if passing through during this
The fringe node being connected directly then is jumped using the interconnection line between fringe node, is otherwise transmitted according to original route;
Step 4 transmits source node up to identical as the Z coordinate of destination node according to Z-direction, if passing through during this
The fringe node being connected directly then is jumped using the interconnection line between fringe node, is otherwise transmitted according to original route.
According to the above improved routing algorithm, as shown in figure 5, in the NoC for the 3D Torus structure that scale is 3*3*3, it is right
The routed path of IP kernel 1 to be measured, which is calculated, (for intuitive show improving effect, is not drawn into other end nodes that do not use in Fig. 5
Between interconnection line).Wherein grey arrow is the routed path being calculated according to XYZ routing algorithm, dotted line arrow be according to
The routed path that improved routing algorithm is calculated.Observable goes out dotted line arrow path compared to grey arrow path, reduces
Routing node (routing node R1~R5 in such as figure in dotted-line ellipse frame) number of process, and the unit on routing node passes
The defeated time is 3 times of the unit transmission time on generic link, therefore improved routing algorithm can effectively reduce 3D
The route transmission time between Torus structure end node.
IP kernel 1 to be measured can be by other IP by 5 routing node R1~R5 that improved routing algorithm saves in Fig. 5 simultaneously
Core calls.Such as IP kernel 2 to be measured can be tested simultaneously by occupying routing node R2 with IP kernel 1 to be measured.It can thus be seen that
There are the IP kernel to be measured 1 of resource contention and IP kernel 2 to be measured under XYZ routing algorithm, no longer conflict under improved routing algorithm, and
It is to be tested simultaneously, to improve the concurrency of test, shorten system test time.
Above by the router-level topology method mode for changing router-level topology transition calculate in HCTPN model, on the one hand subtract
The router number passed through in IP kernel test process is lacked, so as to shorten the route time of IP kernel;On the other hand, by improving road
Resource contention phenomenon is alleviated by the routing node that algorithm saves, reduces network congestion degree, to improve parallel survey
Examination efficiency reduces system test time.
5, the HCTPN model solution based on IFA
Glowworm swarm algorithm (FA) is used as a kind of high-level heuristics algorithm, and concept is readily understood, need to adjust that parameter is few, convergence rate
Fastly, it has been successfully applied to solve the NP problem in the fields such as image procossing, task schedule.Based on the above characteristic, the present invention selects FA
Optimizing is carried out to the transition firing sequence σ in HCTPN model, the testing time is most short, the highest test of dispatching efficiency to acquire
Scheduling scheme.
5.1, algorithm coding
Actual test space and HCTPN model and FA algorithm space, the element representation method in this three is different,
The organic series connection of three can just be made, by suitable coding mode only convenient for solution.Therefore first by IP kernel in the test space to TAM
Selection and same TAM on IP kernel test prioritization pass through formula (1) respectively, (2) map to algorithm space, then will be upper
It states two formula and is converted to sequence in HCTPN modelFinally using σ as firefly
Body, and optimizing is carried out to σ by FA.
XTAM={ B1 B2 ··· BN} (1)
In formula 1 and formula 2, XTAMIndicate the coding of TAM selection individual, BiIndicate that i-th of IP to be measured is selected
TAM number, 1≤Bi≤ M, 1≤i≤N;S indicates the coding of priority individual, AjbIndicate the test prioritization on j-th strip TAM
It is numbered for the IP kernel of b;1≤Ajb≤ N, 1≤j≤M, 1≤b≤n;
N is IP kernel sum, and M is TAM quantity;N is the upper limit for allowing the IP kernel quantity distributed on every TAM, n=N-M+1;
The element that numerical value is 0 in S represents and IP kernel is not present at this sequence on the TAM.
5.2, test dispatching fitness function
The purpose of test dispatching is to shorten the testing time, therefore using the testing time as the fitness function of FA, when test
Between mathematical model it is as follows:
In formula:M represents TAM quantity, and N represents IP kernel quantity, Wi,jIt represents when money
When source conflicts, priority is low or power consumption is excessive, the upper layer skeleton pattern that IP kernel returns in HCTPN model reselects prolonging for TAM
When, DIi,jThe total duration expended when the test vector selection TAM j transmission of IP kernel i is represented, comprising testing IP kernel itself
Spent duration TcoreiAnd the routing duration T of test data packettrans i, i.e. DIi,j=Tcore i+Ttrans i。
5.3, IFA solves HCTPN model flow
FA exists when solving larger np complete problem: firefly individual in the subregion divided automatically most
Excellent individual has dependence;Glowworm swarm algorithm itself lacks Variation mechanism, keeps population diversity insufficient, easily early so as to cause algorithm
Ripe convergence, the phenomenons such as low optimization accuracy is low.To alleviate above phenomenon of FA during HCTPN model solution, the present invention combines mixed
Ignorant optimization method improves FA with population diversity monitoring strategies and differential variation strategy, the overall performance of Lai Tigao FA,
And using improved IFA solve HCTPN model, thus the optimal transition firing sequence σ i.e. testing time for obtaining model it is most short,
The highest test dispatching scheme of dispatching efficiency, specific algorithm improved procedure and model solution process are as shown in Figure 7.
Based on the above analysis, the present invention discloses a kind of 3D NoC test based on HCTPN model and improvement glowworm swarm algorithm
Dispatching method, i.e. the 3D NoC test dispatching method based on Petri network and IFA, specifically include that steps are as follows:
1, HCTPN model is established according to the characteristics of 3D NoC test dispatching process, determines initial marking M0It is identified with terminating
Mf, and calculate input matrix I and output matrix O;
2, the parameter of IFA algorithm is set, and according to formula (1), the coding mode of (2), passes through cube chaos of formula (4)
Mapping generates chaos sequence, then chaos sequence is mapped to purpose-function space and forms initial firefly population, judges firefly
The validity of individual, the TAM not chosen if it exists by any IP kernel are then reinitialized until each firefly individual can be just
Really indicate a kind of reasonable scheduling scheme;
Y (n+1)=4y (n)3-3y(n) (4)
Wherein [- 1,1] y (n) ∈, and y (n) ≠ 0, n=0,1,2,.
3, all firefly individuals in population are all transformed to corresponding transition firing sequence in HCTPN model
4, for each sequence σ from initial marking M0Start, rule successively occurs according to transition and finds out and can excite
Transition, and according to the time delay of fitness function (formula 3) and improved routing algorithm calculating timed transition, each transition has excited
Cheng Hou needs that Policy Updates status indicator M' and cumulative transition time delay occur according to transition, until each transition have excited in σ
At and reach termination mark Mf, the overall delay of σ is the system test time of the scheduling scheme at this time.Step 4 is repeated until kind
The overall delay of all σ all calculates completion in group;
Wherein rule occurs for transition as follows:
For changing t ∈ T, if the mark quantity in its all input magazines institute, it is all larger than or equal to library to changing
Directed edge arc on weight, i.e. s ∈·T → M (s) >=W (s, t), then claim change t be in the case where identifying M it is enabled, be denoted as M [t
>.Change t after being excited under identify M, system will acquisition new logo M'(be denoted as M [t > M'), calculation formula is as follows:
5, all sequences σ in model is inversely transformed into firefly individual, and by the overall delay of sequence σ as firefly
The absolute brightness of individual;
6, the position vector of all firefly individuals is updated according to glowworm swarm algorithm;
6.1, Descartes's distance in population between any two individual is calculated according to formula (6);
In formula, xi,kWith xj,kRespectively indicate the d dimension position vector of firefly individual i and individual jWithIn kth dimension.
6.2, individual moving direction is determined by the comparison of absolute brightness between individual;
6.3, the mutual attractive force between individual is calculated according to formula (7);
In formula, β0The greatest attraction forces at firefly individual i are represented, β can be usually set as0=1;γ represents the absorption coefficient of light.
6.4, the firefly individual for making brightness big is drawn close according to location update formula (8) to the small firefly individual of brightness;
In formula, t indicates the number of iterations;WithFirefly individual i and individual j in the t times iteration is respectively indicated to exist
Position vector in problem space;βij(rij) it is attraction of the individual i to individual j;α represents step factor, usually can be taken as
[0,1] constant in section;Indicate the random number vector obtained by Gaussian Profile, Levy flight etc..
7, it is then gone to according to the population diversity that formula (9) calculate firefly population if more than the lower limit value being previously set
8, if being less than or equal to the lower limit value being previously set, each target firefly individual carries out it all in accordance with differential evolution algorithm
It updates;
Wherein, xij(t) numerical value that firefly individual i jth is tieed up in the t times iteration is represented;N is the population scale of FA algorithm;L
Indicate the cornerwise length of longest in purpose-function space to be optimized;D is the dimension of purpose-function space;For institute in FA
There is the average value of individual jth dimension value, calculation method is as follows:
Differential evolution algorithm renewal process is as follows:
7.1, using each firefly individual in population as a target individual Xi,g, and it is given birth to according to formula (11)
At a variation individual Vi,g;
Vi,g=Xr1,g+F·(Xr2,g-Xr3,g) (11)
Wherein, Xr1,g、Xr2,g、Xr3,gThree fireflies that are respectively different in firefly population and being not equal to target individual
The position vector of fireworm individual;F is the coefficient of variation.
7.2, make each target firefly individual Xi,gWith corresponding variation individual Vi,g, handed over according to formula (12)
Fork generates test individual Ui,g;
WhereinFor Ui,gJth dimension;CR is interaction coefficent.
7.3 between each target firefly individual and corresponding test individual, carries out non-increasing according to formula (13)
The greedy of value selects and retains to the next generation;
Wherein, Xi,g+1For the target individual X of current iterationi,g, the updated position that is obtained by differential evolution algorithm to
Amount.
8, judge whether the stopping criterion for iteration for meeting algorithm, terminate searching process if meeting and export optimal firefly
Individual and its corresponding optimal transition firing sequence and best test dispatching scheme, add the number of iterations if being unsatisfactory for and return together
3 are returned to recycle again.
For the validity for verifying the method for the present invention, with the typical case in international reference circuit ITC'02Test Benchmarks
System: emulation experiment is carried out for d695.Table 1 gives the test dispatching optimization method in the present invention in 2*d695 reference circuit
On certain suboptimum experimental result, when using the minimum value of full test time as measurement standard, the present invention in method can
More preferably global optimum: 27794.8 cycles is searched out, combines basic FA algorithm to obtain compared to by prototype Petri network
Testing time: 30524 cycles reduce 8.94%.
Certain suboptimum experimental result that context of methods obtains on 1 2*d695 of table
Optimal transition firing sequence σ in table 1 indicates a kind of optimal test dispatching side of 20 IP kernels in 2*d695 circuit
Case.Test dispatching Gantt figure corresponding with the sequence is as shown in Figure 6.Dashed rectangle indicates that IP kernel is unsatisfactory for test condition in figure
When wait state.
The present invention is open aiming at the problem that network on three-dimensional chip (3D NoC) test dispatching difficulty, concurrent testing low efficiency
A kind of level Colored Timed Petri Nets (HCTPN) are the same as the solution improving glowworm swarm algorithm (IFA) and combining.This method
It on the one hand the features such as similar according to IP kernel testing process each during test dispatching, TAM selection is at random, routing procedure is cumbersome, will
Petri net model is divided into upper layer and lower layer, simplifies modeling process, and reduce model element by coloring thought, has compressed network planning
Mould portrays the characteristics such as the test dispatching process, resource constraint, priority of IP kernel in 3D NoC accurately;On the other hand,
Router-level topology transition in model are improved, to realize system test time in scheduling combined aspects and routing procedure side
The double optimization in face;In addition, being embedded in IFA in a model, strong algorithm is provided for model solution and is supported, so as to efficient
Best test dispatching scheme is acquired, and the testing time about shortens 8.94% compared with other methods, effectively improves testing efficiency.
It should be noted that although the above embodiment of the present invention be it is illustrative, this be not be to the present invention
Limitation, therefore the invention is not limited in above-mentioned specific embodiment.Without departing from the principles of the present invention, all
The other embodiment that those skilled in the art obtain under the inspiration of the present invention is accordingly to be regarded as within protection of the invention.