Patent Grant
8166430
The present invention generally relates to a procedure for determining the quality of a quantity of properties describing a machine. The machine of the type under consideration can be both software and hardware. Within the meaning of this application a machine is suitable to be described by means of properties. Such properties are generally known to the person skilled in the art of functional verification of digital circuits. An exemplary machine is thus a digital circuit.
When performing functional verification, an RTL-description of a circuit is verified as to whether it will function properly. The quality of the RTL-description after functional verification is decisive for the success of a circuit design project. Remaining errors are likely to cause high costs, loss of time and reputation.
In addition, using simulation as the principal work horse in functional verification is standard. For this, input patterns are preset, either automatically or by the user, and the output patterns generated by simulation compared with expectations. Errors are detected if an output pattern does not meet with expectations. Automatic mechanisms for error detection in simulation are hereinafter called monitors.
In order to detect an error, there must be given a suitable input pattern for which the simulation produces an output pattern in which the error shows up. Even in small circuits the quantity of all possible input patterns is very large. In addition, the factor between the run time in the real circuit and its simulation is in the range of 106. Therefore only a few input patterns can be simulated in comparison with the total number, so that many functional errors remain undetected.
Increasingly so-called property checkers are used for functional verification. Property checkers are given, as something input something called a “property”, i.e. a relationship concerning certain aspects of the circuit, that is described in a formal language. Within surprisingly short time, the property checker yields a result in the form of
either the information that the property has been proven by the property checker. In this case the relationship between the behavior aspects which is described in the property will always occur, irrespective of how the circuit is operated
a rebuttal in which an input pattern for the circuit is described that violates the relationship between the behaviour aspects. This rebuttal is generated by the property checker.
The automatic generation of counter-examples and the rapid checking of the property for all possible input patterns constitute a considerable advancement in the functional verification vis-à-vis simulation.
For this reason, property checkers are also being implemented today by innovative verification groups. However, they only verify selectively as behavioural associations that are considered critical, which were chosen by a verification engineer based on his/her intuition and experience. This nonetheless only enables gradual improvements in the quality of the RTL description after verification. A relatively high number of errors in the design are still present in the result.
In the area of circuit verification by means of simulation, suggestions have already been made concerning procedures for the evaluation of the quality of the verification environment, which even calculate measures, which are supposed to be related to this quality. In this sense, the number and coverage of the input patterns used for simulation is, on the one hand, of the greatest importance for the quality of the verification environment and, on the other hand, the effort to achieve a multiplicity of different types of input patterns is also cause for a large resource requirement.
The corresponding procedures have already been in use for years and are supported by commercial EDG tools (see e.g. D. Dempster, M. Stuart: “Verification Methodology Manual—Techniques for Verifying HDL Designs”, Teamwork International, 2nd edition, 2001; and S. Tarisan, K. Keutzer: “Coverage Metrics for Functional Validation of Hardware Designs”, IEEE Design & Test of Computers. 2001.) The metrics for determining the measures are often code-based such as the line coverage, path coverage, assignment coverage and coverage of states for explicit state machines from the RTL code.
Such methods for measuring the quality of a verification were first developed for software verification. Unlike sequential software, however, hardware operates in the form of parallel processes. The dependencies generated by this are therefore not taken into account by the methods, so that the resulting quality measures are not reliable. Therefore, alternative measures have been suggested, for example those based on signals and require, for example, the assignment of all possible values. In addition, there are functional metrics, which display how often a certain functionality has been executed.
An inherent weakness in all of these approaches exists, however, in that attaining a level of 100% is far from securing the entire verification of all functionality. On the one hand, these approaches already achieve 100% coverage even when not all functionality and consequently all possible error sources in the circuit tested have been injected; on the other hand, there is no systematic test which shows whether the verification environment is capable of actually identifying every error that shows up.
The property check is described in exemplary terms in: A. Biere, A. Cimatti, M. Fujita, Y. Zhu: “Symbolic Model Checking using SAT procedures instead of BDDs” Proc. of 36th Design Automation Conference, 1999; and in: “Formale Verifikation für Nicht-Formalisten (Formal Verification for Non-Formalists)”, Bormann, J. and Spalinger, C: (2001), Informationstechnik und Technische Informatik, Vol. 43, Issue January 2001, Oldenburg Verlag.
When performing the quality assurance of a verification environment based on the property check, the issue is not necessarily the multiplicity and diversity of the input patterns, since a property checker functions as if it were inspecting all input patterns. It is rather the integrity of the property set, that is, whether each error in the circuit is detected when proving at least one of the properties.
Papers in this field are documented, for example, in: Hojati, R.: “Determining Verification Coverage Using Circuit Properties” U.S. Pat. No. 6,594,804, granted Jul. 15, 2003; Hoskote, Kam, Ho, and Zhao: “Coverage Estimation for Symbolic Model Checking” Proc. of 36th Design Automation Conference, 1999; and in: Hoskote, Y. “Property Coverage in Formal Verification”, Patent WO200079421-A2. These approaches are all based on sequentially injecting errors in the circuit and then checking whether at least one of the properties is disproved by the property checker in the circuit that has been modified in this way. If each of the errors injected ends up being disproved, the quality of the property set will be designated as sufficiently high. Otherwise, a measure concerning the proportion of those injected errors will be determined which were detected by the property set.
With the known procedures, it is either injected as an error that signals are inverted in various achievable states or the gate, at whose output the signal to be covered is picked up, is changed on the net list.
These procedures are heuristic. There is no guarantee that errors other than those injected will be detected by the property set. In addition to the quantity of properties, the procedures also require the circuit description.
Task
Since so far the properties, as described above, have only been applied in a selective manner and based on experience, a need exists for a procedure, which can reliably specify the quality of a quantity of properties in a reproducible manner, particularly in the form of a measure which can then be brought into accordance with a target value by expanding or adjusting the volume.
Furthermore, a need exists for procedures for more reliable verification and the specification of circuits based on the reproducible quality of a quantity of properties.
Consequently, one object of the invention in question is to specify procedures which determine and utilize the quality of a quantity of properties describing a machine without having to resort to excessive use of resources or experience values.
The above tasks are fulfilled by the invention by means of the procedures of the independent claim, preferred exemplary embodiments constituting the subject matter of dependent claims. By means of determining the quality of a quantity of properties describing a machine it is possible to utilize the properties in a more efficient manner.
In particular, the invention suggests a method for determining the quality of a quantity of properties describing a machine, comprising the determination of the existence of at least one sub-set of interrelated properties (P0, P1, . . . Pn); and the verification as to whether, by the interaction of (P0, P1, . . . Pn), for at least one input pattern of the machine, the value of a given term Q(t) is unambiguously/uniquely determined at least at one point in time at which it is not unambiguously/uniquely determined by the observation of the individual properties, Q(t) being only dependent on values of input and output parameters at points in time relative to t.
Here, Q(t) can be specified by the user or drawn from a reference table such as the list of the output parameters for the machine.
Q(t) is assumed as unambiguously/uniquely determined a the quantity of properties in an input pattern at a point in time T if for two sequences of value, on the output and internal parameters which supplement the input pattern in such a way that all properties of the quantity on it are valid, the value of Q(T) on the one supplement is equal to the value of Q(T) on the other supplement.
It is typical for the property check that the individual properties do not clearly determine an output parameter and consequently no Q(t) value since the properties are often in the form for all t:
For this, the symbol => stands for the logical implication. The property itself does not determine any output parameters clearly because the complements to an input pattern can always be selected such that they violate condition 1 and then the values of the output parameters are able to be selected entirely contrary to one another. This type of complements completes the property and shows that no output value has been unambiguously/uniquely determined. The output values would be unambiguously/uniquely determined for an input pattern by the above property, if it is combined with a reset property, which ensures that the assignment of the internal signals following the activation of the reset input fulfils condition 1.
As such, the invention is based on checking whether individual properties, which are only utilized and considered on a singular basis pursuant to the state of the art, exist in relation to one another.
Using this basic concept, properties can be systematically identified and utilized and for example, can also form multiple series of properties, which constitute a statement independent of internal signals concerning the input/output behaviour of the machine over a longer period of time.
In accordance with a preferred exemplary embodiment, the interrelated properties form a series of properties, which starts with a property that describes the initialization behaviour of the machine and constitutes a statement concerning the input/output behaviour of the machine or also of a circuit over a longer period of time, whereupon this statement is no longer dependant on internal signals.
Those properties not considered in the formation of the series of properties are conveniently reported to the user. In this process, it is possible to adjust the volume of properties in an efficient way or to complete them since either the property in question has been reformulated or a predecessor and/or successor property is to be provided for it.
If the properties have the form Pi=forall t: Ai=>Zi(t) with implicatively associated Boolean expressions Ai and Zi, which depend on values from the input, output and internal parameters at points in time relative to t, it can be conveniently checked during the determination step whether a series of reference points in time T1, T2, . . . Tn exists, such that the property given by the chaining of the properties with respect to the reference points in time
already describes the input/output behaviour such that at least one input pattern, in which E clearly describes the value of Q(t) at a minimum of one point in time.
The procedure corresponding to the section “Automatic determining of all redundant specifications” proves this and the previous claims. For a quantity of properties, it is automatically determined, which properties are allowed to follow one another and how the reference points in time are to be formed during for this. Through this process, a graph is determined, whose only source is the property concerning the initializing behavior of the machine and whose finite paths, which start in the reset property, form the quantities or series of interrelated properties. Temporal domains, based on which the output signal is unambiguously/uniquely defined insofar the property is part of a series starting with the reset property and for which the reference points in time are selected in accordance with the values calculated, are calculated for every property and every output signal. Based on these temporal domains, it can then also be decided as to for which point in time Q(t) is unambiguously/uniquely defined.
The quantities of input patterns are conveniently described, on which Q(t) is unambiguously/uniquely determined for at least one point in time for each one.
If the properties are given in the implicative form described above, the quantity of input patterns, on which Q(t) is unambiguously/uniquely determined for at least one point in time for each one, consists of the input patterns for each of which a value series of the internal and output parameters is given such that there is at least one point in time t, for which the following applies based on these value series A0(t) and Z0(t) and A1(T1) and Z1(T1) and . . . and An(Tn).
In general, the reference points in time are dependant on t and on values of the input parameters at points in time relative to t. It is beneficial, however, if the reference points in time are only dependant on t.
The interrelated properties are arranged preferentially in a sequential relation, whereupon the values based on the output and internal signals, which are able to fulfill Zi−1(t) for the suitable selection of the input pattern especially for i≠0, are also able to fulfill the Boolean expression Ai(t′) for a suitable input series and for a point in time t′. This condition is also utilized in the procedure corresponding to the section “Automatic determining of all redundant specifications”.
The condition is closely related to the condition that Ai(t) and Zi−1(t) except for temporal displacement contain syntactically equal partial expressions concerning output and internal parameters.
By means of a sequential relation, which ensures between the elements of a finite quantity of properties that a series of properties can always be extended, it is possible to also check time ranges on an input pattern of unlimited length and to make statements for which points in time Q(t) is unambiguously/uniquely defined. In particular, it can determined that Q(t) on an input pattern is unambiguously/uniquely determined for all points in time following the first point in time.
The procedure according to the section “Calculation of all redundant specifications” sets the foundation by setting a graph whose finite paths set all series of properties possible.
In order to be able to identify additional sub-quantities, the step of identifying the existence of at least one sub-quantity of interrelated properties is repeated. By means of this process, it is e.g. possible to determine all possible property chains based on the properties.
The check preferred is whether the output parameters of the machine are unambiguously/uniquely determined for all input patterns and for all output parameters of a machine and for all points in time following the first point in time. This claim is proven by the combination of the procedure from the section “Automatic determination of all redundant specifications” together with the procedure for determining the completeness of a set of properties that is revealed in this patent specification.
A value measurement is identified and issued on a preferential basis, which corresponds to the probability that the value of this output parameter is unambiguously/uniquely determined at this point in time by the property for any input pattern starting with the circuit for any point in time and for any output parameter. This claim is proven by the combination of the procedure from the section “Automatic determination of all redundant specifications” together with the procedure revealed in this patent specification that is used to calculate the measure of completeness.
The invention also proposes a procedure for the functional verification of digital circuits, in which the circuit is checked using a quantity of properties whose value was determined as described previously and corresponds to a defined designation. In terms of the defined designation, a set threshold value may be involved or a value specified by the user, for which the value is believed to be sufficient because e.g. a further improvement can only be achieved by means of a disproportionate extra amount of work. The value can be monitored during the successive development of the properties and inferences can be made regarding the work to be incurred based on the difference between the designated value and the actual value.
The invention also proposes a procedure for the specification of digital circuits, in which the specification is created, based on a quantity of properties whose value was determined as described previously and corresponds to a defined designation. The value can be monitored during the successive development of the properties and inferences can be made regarding the work to be incurred based on the difference between the designated value and the actual value.
The invention proposes a procedure for the simulative verification of digital circuits, in which monitors are utilized, whereupon the monitors represent machines and the quality of the monitors is determined using the value of a quantity of properties described for the monitor. This quantity of properties is determined as previously described. In this manner, the quality of the monitor can be systematically tracked.
Finally, the invention proposes a procedure for the simulative verification of digital circuits, in which the coverage is determined by monitors, which are acquired from the assumptions of a set of properties whose value was determined pursuant to a procedure of the claims and which correspond to a defined designation. In this way, it can particularly be ensured that the coverage is derived from a complete set of properties, thereby expressing functional aspects on the one hand and on the other hand, covering the entire functionality of a module. This coverage therefore combines the suitability of the functional coverage with the breadth of the coverage of code-based coverage and is thereby particularly suited for the system simulation.
In summary, the invention therefore enables checking a quantity of properties in terms of whether they cover the entire function of the circuit to be checked rather than only covering critical behavioural associations on a punctual basis. If the properties have all been proven on the machine (e.g. software or circuit) by a property checker, only errors also existent in the properties remain uncovered. The main problem of the simulation that errors remain undetected because they do not occur in any of the output patterns generated is avoided. As a result, the quality, which has the RTL description after completion of the verification, increases. As such, the invention allows a circuit verification to be conducted exclusively using a property check for the first time.
The fact that the mechanisms for the detection of errors (that is, the monitors or properties) could themselves be flawed and conceal circuit errors as a result thereof is a characteristic of every functional verification. This problem also exists in the simulation, which is recorded therein that the monitors have not detected any errors although they occur in an output pattern. If that happens, the monitor responsible for this was either coded incompletely, that is, it fails to conduct any check at all in one of the situations generated by the simulation or the monitor actually expects an incorrect value.
It can be ensured in an invention-pursuant manner that the properties check this behaviour at any point in time. Incompleteness in the properties is therefore automatically identified. As a result, errors that would otherwise remain undetected due to the incompleteness of the verification environment are also avoided.
In this context, it should be noted that the invention is not limited, in terms of its use, to the vicinity of a property checker. It can also identify incompleteness in monitors. For this, the monitors must be generated from properties, which were appropriately checked. The generation of monitors from properties is familiar to professionals. Procedures for the quality assurance of monitors to date are only known in terms of code review and recycling up to now, which however is not comparable with the invention-pursuant approach.
The properties commonly describe the machine or circuit behaviour to a more abstract extent than the RTL description. They mostly describe the machine resp. circuit behaviour over several cycles such that the processes in a machine/a circuit are presented universally and closed. As a result, they provide the verification engineer with a larger overview than the RTL description such that additional errors in the properties are avoided. The invention checks whether a quantity of properties has the quality of a machine/circuit description at a greater level of abstraction.
With the state-of-the-art vis-à-vis the simulation, the analysis of certain conditions with the objective of analyzing the number and variety of the input patterns simulated and consequently the probability that the flawed machine/circuit components were indeed conducted (coverage) is to be evaluated. The conditions implemented for this are commonly oriented with the text of the machine/circuit description. In this case, conditions are being worked with, which are fulfilled by the completion of a line (or also only one part of the line) in the machine/circuit description. This approach has the advantage that these kind of conditions consider each component of the machine respectively circuit in a wide-reaching manner. It is, however, regarded as inadequate because it does not take the parallel design of machine/circuit components into consideration. As a result, the text-oriented conditions are supplemented by conditions, which are fulfilled in specific partial processes (functional coverage). Nowadays, these kinds of conditions are only defined on a punctual basis, where they are necessary according to the intuition and experience of the verification engineer. The invention allows the definition of a set of such conditions, which are each fulfilled, if individual properties describe the machine/circuit behaviour. Such a set of conditions considers on the one hand all machine/circuit components and on the other also their parallel design. No procedures are yet known, with which sets of conditions of comparable quality can be generated.
All this helps keep the number of errors remaining in a design at a much lower level following the completion of a verification that utilizes said invention than following a simulation-based verification. Through the efficiency of the property checker, such verification most often saves an incredible amount of work.
In so doing, the invention delivers a specific criterion for the completion of the verification using the property checker in a preferred manner. If a property set is complete or suffices for the target value, the verification will end, if this is not the case, then it must be continued. The procedures corresponding to the state-of-the-art are not familiar with such criteria. In this case, the verification will be continued until the time or financial budget has been consumed or until certain specifications have been fulfilled, which heuristically suggest that now the residual risk has dropped below a tolerable value.
The invention can also be applied even before a set of properties has been verified with a property checker vis-à-vis a machine respectively circuit. This becomes important when such a set of properties is intended to play the role of a specification. Therefore, the invention is able to provide a necessary condition for the quality of a specification. This condition is not sufficient: There are sets of properties, which can be designated as being complete, although no circuit is present, which fulfil all the properties of such a property set.
For the purpose of control, the invention delivers a measure of completeness on a preferred basis, which allows the progress of the verification to be presented during the course of the project. This measurement increases in correlation to a growing number of properties and only returns the value 1 once the number of properties is complete. By means of this, the verification can be monitored much more precisely using a property checker than with all other procedures known to present.
Additional advantages and features of the invention and a detailed understanding can be obtained from the following description of the current execution forms preferred at this point in time, whereupon the description is intended to be purely illustrative and should not be regarded as being of a limiting nature. In this description, reference will be made to the attached diagrams, for which the following applies:
In this description, reference will be made to the attached diagrams, for which the following applies.
The examples refer to circuits as one of the possible machines and are subsequently presented in the temporal language ITL. However, the invention is neither limited to this temporal logic nor to circuits. The procedures can also be completed using all other linear, temporal logics such as LTL, PSL or SVA and also be applied to software and other machines. The procedures described can be implemented both in terms of hardware and software.
The temporal logic is used for the description of the relationship between behavioural aspects in the properties. In the following representation, a slight modification of the ITL language, the input language of a property checker [it+ti] will be used. In this language, temporal aspects are described by referring to reference to points in time. Points in time correspond to cycles in synchronous digital circuits.
The core of ITL is comprised of logical lines of one of the following three forms at <point in time>:<Boolean condition>;
during [<point in time1>, <point in time2>]:<Boolean condition>;
within [<point in time1>, <point in time2>]:<Boolean condition>;
The first form requires the validity of the Boolean condition at the point in time specified, the second for all points in time in the interval and the third for at least one point in time in the interval. If the interval is empty, the second form is fulfilled trivially and the third is not fulfilled trivially.
The analysis of a Boolean condition at a specific point in time T for all standard operators occurs through analysis of partial expressions at this point in time T until a partial expression is issued by means of a signal, whose value is then implemented at the point in time T. Term at the point in time T+n respectively T−n is evaluated in order to evaluate a partial expression of the form next(Term, n) or prev(Term, n). In order to evaluate a partial expression of the form Term @ T′, term at the point in time T′ is evaluated.
Points in time are specified relative to a point in time t of choice, but one that is fixed or relative to temporal variables that have been derived from it, which are for their part to be declared again relative to t. As such, it should be expressed that at any point in time a property of a conceived, unlimited simulation session should be checked.
Temporal variables are declared by means of a list of expressions
For this, the Bi describe either t or another temporal variable Tj declared in the above list and are called basis temporal variables of Ti. A permitted allocation of the temporal variables must fulfil the conditions given above, that is Bi+ni<=Ti<=Bi+mi must be true. Temporal variables are generally used in order to describe how the observed circuit synchronizes with its environment. For this, signal edges are frequently utilized in the assumption by means of conditions in the form
in order to set a temporal value Ti. These conditions are referred to as localization conditions.
Logical lines form conditions as follows: In the simplest of cases, a condition represents a logical line. The succession of two conditions corresponds to their conjunction.
Disjunction is expressed by
Temporal-logical operators between the conditions are also used unless this condition is not to be directly proven using a property checker.
A property has the form
Condition 1 is to be referred to as assumption, condition 2 as the proof. The property is fulfilled, if the affirmation also applies to each permitted allocation of t and all temporal variables, for which the assumption is fulfilled. Vice versa, a counter example for the property is represented by a process, in which any permitted allocation of t and the temporal variables has been fulfilled for the assumption, but not the proof.
Partial expressions up to conditions can be summarized in macros, which can be parameterized if necessary.
The shortest interval [t−n, t+m] where n, m>=0 is herein referred to as the inspection window such that any point in time, at which any signal is evaluated, lies between t−n and t+m for any allocation of the temporal values.
t_first and t_last should describe the inspection window that results when all logical lines are replaced by at t: true, which for their part reference t_first or t_last.
Two synchronous digital circuits are to exist with a reset input. A similarity is assumed when both circuits produce the same output patterns following collective resetting to the same input patterns.
A quantity of properties is viewed to be complete when any two circuits, each of which fulfilling of all properties of this quantity, are similar to one another.
The completeness checker inspects properties of a specific type referred to as operation properties. First of all, the theory for the properties is presented, which are strictly limited in relation to their expressiveness. Afterwards, generalization occurs based on this more demonstrative representation.
Properties arising from the following basic form shall be assumed:
For this, initial state, triggeri, output descriptioni and final state are all to be considered as macros. The property essentially describes: “If the circuit is in a certain state, which is described by the macro initial state and if it recognizes the input patterns, which are described by the macro's triggeri, then it will return an output descriptioni and proceed to the final state”.
The basic idea behind these properties is to form chains composed of chain links. The chain links have ends, which each correspond to the initial and final state of a property in the above form. Under certain circumstances, one can form chains from properties in a corresponding manner. An objective of the invention is, by means of local inspection of a limited number of properties, to ensure that an unlimited chain of properties can be formed for every possible unlimited input pattern through the stacking of start and end states that correspond to one another and whose trigger is fulfilled by the input pattern and whose output descriptions clearly and completely describe the output pattern.
This is formalized as follows:
A logical expression P is designated to be dependent on a parameter s, if an allocation of the remaining syntactic parameters occurring in P exists such that the value P can be made true or false through the different selection of a value for s.
Given that a series of properties P0, P1, P2, . . . Pn of properties of the form Pi=for all t exists: ai(t)=>zi(t), with implicatively associated Boolean expressions ai(t) and zi(t), which are dependant on values of the input, output and internal parameters of a machine at points in time relative to t and a series of reference points in time T1, T2, . . . , Tn, which may be dependant on t and values of the input parameters relative to t. We designate the following as an applicable property for all t: (a0(t)=>z0(t)) and
((a0(t) and a1(T1(t)))=>z1(T1(t))) and
(a0(t) and a1(T1(t)) and a2(T2(t)))=>z2(T2(t))) and . . . .
We describe the series P0, P1, P2, . . . Pn with reference points in time T1, T2, . . . Tn as a chain when P′0, P′1, P′2, . . . P′n exists in the form
P′i=for all t:a′i(t)=>zi(t),
whose a′i are only dependant on values based on input values relative to t and whose applicable property is equivalent to the property, which belongs to the output series. In a chain, it only depends on the input signals as to whether and at which site the behaviour of the output and internal signals is described by the zi. With this requirement, the series P0, P1, P2, . . . . Pn with the reference points in time T1, T2, . . . , Tn, are trivially chains, if the ai are only dependant on input signals. But also one ai, which is dependant on internal or output signals, can belong to a chain, if e.g. zi−1 provides sufficient information concerning these signals. The manner in which this can be tested is part of the invention described in this text.
An internal or output signal s is, with respect to a chain P0, P1, P2, . . . . Pn featuring reference points in time T1, T2, . . . Tn, determined at a point in time Z, which has been issued by a function, which may depend on t and input signals, if its value is unambiguously/uniquely determined at the point in time Z by the chain. With a chain, it is consequently ensured that s always has the same value at the point in time Z(t0), whenever a certain series of input parameters occurs following a specific t0 and this series of input parameters sequentially fulfils a′0(t0), a′1(T1(t0)), a′2(T2(t0)), . . . a′n(Tn(t0)).
If a chain determines an output signal at a point in time, then there is a corresponding function, which only depends on input signals, which sets the value of the output signal at this point in time. Chains, which determine the output parameters at some points in time, thereby specify input/output behaviour independent of any one individual circuit. Part of this invention encompasses the manner in which it can be checked as to when a chain determines the behaviour based on output and internal signals.
We describe a series K0, K1, K2, . . . of properties, whereupon Kn represents each property belonging to P0, P1, . . . Pn as belonging to an infinite series P0, P1, P2, . . . of properties (which very well may result from a finite quantity of properties and then occur on multiple occasions) and T1, T2, . . . . An infinite series P0, P1, P2, . . . featuring corresponding reference points in time is called (infinite) K, if every finite partial series P0, P1, . . . Pn is a chain Kn. An infinite chain K determines an output signal s at one point in time Z, if an index K exists such that all Ki with i>k determine the output signal at the point in time Z.
If an infinite chain, which covers this input stimulus and which determines all output signals at all points in time with the exception of the initial partial component, can be formed from a quantity of properties for every input stimulus possible, then this quantity is complete in the above sense.
The conditions for checking completeness of properties of the above form “basic form” are:
If this property can be proven, it is known that a property Q can always be specified, with which it continues, if the final state of P has been reached. If this property has been proven for the quantity of successors of every property, one already becomes aware of the possibility of forming a chain for every input pattern that covers said input pattern.
is to be designated with Out(t). A point in time ZPo for every output signal o is determined such that a type-correct constant w exists, where Out(t)=> at ZPo:o=w applies.
A property Q′ is formed for every Q from NP, when t is replaced by t+nP such that the initial state of Q′ is assumed at the point in time, at which the final state of P is affirmed. t+nP is designated as a reference point in time of P for Q in the following section. OutP(t) and OutQ(t+nP) represents the output description for the event that P is followed by Q. This requires that there is a type-correct value w for every output signal o at every point in time T from the interval [ZPo, ZQo], where
OutP(t) and OutQ(t+nP)=>at T:o=w applies.
This then ensures that the output values along the chain are unambiguously/uniquely determined, possibly with the exception of some points in time, which will be replaced by the reset property.
If the properties have been developed while taking the environment conditions into account, the consistency property case_split_very_simple can no longer be proven. Input patterns, with which the environment conditions are violated, are then generated as counter examples. Therefore, the property must be proven instead of the property case_split_very_simple with the environment conditions
The term of partial properties, with which the way the observed module communicates with the adjacent modules is described in terms of phases, shall be introduced. A partial property represents an initial component of a longer property. If A and C are the assumption and affirmation of the longer property and a and c the corresponding components of the partial property, then A=>a und C=>c must apply. Furthermore, the partial property must apply to the circuit checked.
The following quantity of properties shall be considered in order to motivate and illustrate the necessity of partial properties. It describes a module, which communicates with an adjacent module. This neighbouring module is supposed to be described in an environment condition.
This adjacent module affirms that it produces “10” or “01”, whenever out_sig is set to “001”. This is described in the environment condition
E=(out_sig=“001”=>next(in_sig)=“10” or next(in_sig)=“01”)
The following specified properties t2 and t3 also actually describe the communication with the adjacent module: A request leads to the transmission of “001”. The adjacent module returns in_sig=“10” or “01” and an additional value based on out_sig is produced depending on this.
These properties are all to be contained in a quantity of successors. The property state_split_simple for this quantity of successors cannot be proven even under the environment condition E. A counter example to state_split_simple will e.g. at the point in time t show that state is set to idle and that request is also active, but set in_sig to “11” at the point in time t+3, such that none of the assumptions of t1, t2 or t3 are fulfilled.
This problem arises because out_sig=“001” in every case following a request is not included in the check and for this reason, only select values based on in_sig are allowed to occur at the point in time t+3. The communication described in t2 and t3, however, has two phases: on the one hand, the generation of “001” and on the other hand, the generation of “101” respectively “110”. The check of the communication is therefore conducted in phases. First of all, the generation of “001” is described:
This property originates when t2 or t3 are truncated at the point in time t+2. This kind of truncated property must generally be proven separately with the property checker.
The property case_split_simple can be proven with the assumptions of partial and t1.
The entry in_sig is first evaluated in the next phase. For the verification, that is partially continued by t2 or t3, the following property is automatically assembled and proven:
This property is assembled such that assumption and affirmation of the partial property are adopted into the assumption of the consistency property. This is possible because the property was proven on a partially separated basis. As a result, however, the environment condition also becomes applicable such that the property can actually be proven as a result.
Partial properties are often the initial components of other properties. The longest properties, namely those describing the entire communication, are designated as main properties. It is requested that the partial properties are formed for every main property P such that a finite, unambiguously/uniquely determined series Qn, for which Qn is each partial property of Qn+1 and P is the last element in this series.
As exemplified above, a partial property Q of a main property P can also be a partial property of another main property P′. For this, no displacements should result: Under the assumption that the point in time t in P′ is replaced by a t+i with a whole number i, before the check is conducted as to whether Q is also a partial property of P′. In this case, it must apply that Q occurs in the partial property series of P and P′. In this way, the partial properties of all main properties form a graph, which is characterized as follows:
directed and cycle-free
for every partial property, there is a path to at least one main property
every main or partial property has at most one predecessor in the graph.
Such a graph is referred to as the partial property graph.
The completeness checker receives an annotated quantity of properties (1) with the following information:
Name, mode (input, output, internal) and type of each signal that is referenced in the properties.
A property Preset is to be specially issued as the reset property.
A main property, which is described in the following text. The elements of this quantity of properties are called main properties.
A graph of partial properties (partial property graph), which is described in the following text.
Every element from the quantity of main properties from (1) provides the following information: Quantity of temporal variables TT, consists of temporal variables Ti:
T0 is to correspond to t, for i>0, every temporal variable is characterized by
The designations “requirement” and “conclusion” were selected in order to differentiate the portions of these specially annotated properties from the assumptions and affirmations, which can occur in every ITL property. The requirement becomes a component of the assumption and the conclusion for a component of the affirmation of an ITL property for the proof of such an annotated property on a circuit.
An environment condition E concerning the behaviour of the input signals in the connection to sites, at which the property checked was assumed. A part of the environment condition should be that the reset is never activated. The collaboration of environment and the observed circuit must, however, generally suffice for the property
However, E can frequently be proven without additional requirements.
Assumption of a determined range [Ki, ki] for every output oi: In this temporal range, the property is intended to unambiguously/uniquely describe the behaviour of the outputs.
Assumption of data interpretation functions D1, D2, . . . with evaluation points in time p1, p2, . . . : The evaluation points in time for this are typically located close to the reference points in time from the quantity of successors (see below). With the specification of the data interpretation function, the user designates which data stored in the state are of such importance to the function of the circuit that they must be relayed onto the following property. Through the sequence of the values described by the data interpretation functions along a property chain, it should be apparent as to how these data are calculated from inputs.
Quantity of successors NN of properties N1, N2, . . . each together with a reference point in time R1, R2, . . . Every reference point in time must be greater than T0. Furthermore, Preset may not occur in a quantity of successors.
The partial property graph is to correspond to the characterization in the previous section. A property Q in this graph is to be described by a quantity of temporal variables TTQ, an assumption AQ and an affirmation CQ. TTQ should also contain its basic temporal variable with every temporal variable and be a partial quantity of a quantity of temporal values TT from the quantity of main properties from (1). If there is an edge q−>Q in this graph of partial properties, then TTq should be a sub-quantity of TTQ, where AQ=>Aq and CQ=>Cq apply. Isolated nodes from the graph can be both sources as well as sinks of this graph. The properties from the quantity of main properties from (1) should be located at the sinks in this graph. If there are paths from a source to two main properties, then they can only occur together or not at all in a quantity of successors. If they occur in a quantity of successors, the reference points in time R of the two properties must be the same.
The verification of the properties on the circuit to be checked is described in
This is checked with respect to the circuit description (32) (which is given e.g. in VHDL RTL) through the use (33) of a property checker. As a result, the property is either proved or disproved. The positive proof result (34) or the disproving sequence (35) is relayed to the user. The verification of these properties corresponding to the procedure complies with the standard practice of verification using property checkers. Both the main and partial properties must be verified.
The steps (30) to (35) can be conducted prior to or following steps (2) to (25).
The elements of the annotated quantity of properties (1) not used in the circuit property are required for the formation of the consistency properties, which are generated during the completeness check by the completeness checker and proven by a property checker. A trivial model (3) is generated from the annotated quantity of properties (1) by the model generator (2). This trivial model is a special final machine, in which all signals presented in the annotated quantity of properties (1) are presented twice and can be differentiated from input signals. Each signals that belong together are to be specified sequentially with s and so. In this way, though the trivial model (3) contains the type information, it does not contain any information regarding the circuit behaviour itself. As a result, the completeness checker is only familiar with the signal behaviour that is described by the main properties.
A description as to how consistency properties are generated from the main and partial properties, which are proven based on the trivial model (3) by means of a property checker, will now follow. If this is successful for all consistency properties generated, the completeness of the set of main properties has been proven. First of all, copies in the different sets of variables of the trivial model (3) are generated for all main and partial properties by means of the property copier (4) in order to assemble the consistency properties. The two copies of a property are designated as P and Po and at the same time, for all components A, Ao, C, Co, . . . and so on. With this copying, the sets of temporal variables should be made differentiable such that the temporal variables To0, To1, . . . originate. The information Bi, pi, Ni and Ri are not modified by the copying process (4).
Under the loop control (5), a primary consistency property is determined for every main property P and for every successor property P′ from the quantity of successors of P. For this, the properties P′ and Po′ in (5′) are first of all copied such that everything is adopted except for the temporal variables and the temporal variables are replaced such that they can be differentiated from the temporal variables in P and Po. These “fresh” temporal variables are specified with T′i and To′i. In the generator for the first test (6), a primary consistency property is generated from the above properties according to the following scheme.
For this, R is the reference point in time, which is specified together with P′ in the quantity of successors of P.
In detail, the property is formed as follows: All temporal variable declarations from P are entered in lines 4 to 6. In lines 7 to 9, the corresponding temporal variables from Po are equated with their partner. In line 10, the initial point in time of P′ is set to the reference point in time from P and the temporal variables are then entered in lines 11 to 13 in accordance with their declaration in P′. The temporal variables from P′o corresponding to this are equated with them in lines 14 to 16. The line 19 indicates that the signals classified as input signals in the annotated quantity of properties (1) and their respective copies will be equated with one another. In lines 20 and 22, all localization conditions from P and Po are specified, while the corresponding assumptions and proof targets are entered in lines 21 and 23. The equality of all data interpretation functions for their respective evaluating points in time are summed in lines 24 to 26. In lines 27 and 28, the equality of all output values during the time frame, in which P resp. Po are responsible for their determination, is furthermore assumed. In lines 30 and 31, the assumption of all localization conditions from P′ and the requirement from P′ occurs. All localization conditions and the requirement from P′o should be located in lines 33 and 34.
The validity of the property based on the trivial model (3) that is acquired in this way is proven by the implementation (7) of a property checker. This provides either a proof or a disproof (8), which is relayed to the user.
An additional property is created in the generator (9) for a secondary consistency property in accordance with the following scheme.
Up to line 30, this property is equal to the primary consistency property. The requirements and sequences are now entered in lines 31 and 33 here and all localization conditions from P′ and P′o are entered in lines 30 and 32. Now the equality of the values for all output signals are to be proven in lines 35 to 37. The range, in which this equality is requested, is directly connected to the range, in which it was assumed in lines 27 to 29. Furthermore, the equality of the data interpretation functions is requested at its evaluation points (lines 38 to 40).
The property is again proven by means of the implementation (10) of the property checker or a disproof (11) is relayed to the user. After the loop control (4) has processed all pairs of sequential properties P, P′ in compliance with the specification in the properties from the main quantity of properties (1), it relays the control to a second loop control (12), which sorts through all properties P from the quantity of properties (2) and assembles a property of the following scheme first_case_split for every P. The following notation is used for the description of the scheme: While regarding all relevant entries of every partial or main property as being parameterized by the temporal variables. All sources of the partial property graph, namely those properties, for which no shorter partial properties exist, are processed. For this, one limits oneself to the sources, from which a path through the partial property graph exists, which leads to a main property N belonging to one of the quantities of successors of P. These sources are referred to as P′j and designate their components with A′j, T′ji, Z′ji, z′ji, L′ji and so on, Rj being the reference point in time, which is linked to N in the quantity of successors of P. Under the request that Rj is unambiguously/uniquely determined, also in the event several N exist for P′j.
A series Belj1, Belj2, . . . is formed for every source P′j, whereupon every element of the series is to have the form Beljn=(T′j0=Rj, T′j1=Rj+ujn1, T′j2=Rj+ujn2, . . . ) with whole numbers ujn1, ujn2, . . . and the series Beljn fathoms all possible allocations, which are permitted by the quantity of temporal variables of P′j. With A′j(Bel), a copy of A′j is described, whereupon all temporal variables corresponding to the allocation Bel are allocated. In the property generator (13), a property is to be formed in accordance with the following scheme:
The temporal variable declarations, localization conditions, requirements and sequences of P are located in the lines 4 to 7, 11 and 12. Line 10 requests the validity of the environment condition E belonging to P. Lines 14 to 26 list the requirements and localization conditions of all sources P′j with paths to the quantity of successors of P and instantiates all possible allocations of the respective temporal variables from the respective P′j. Through application of the property checker (14), this property is proven or a disproof (15) is relayed to the user.
Through the loop control (16), every node P′ of the partial property graph, which itself is not a sink in the graph and for which a path through the partial property graph exists which ends with a property N of the quantity of successors of P, is now processed for every main property P. P″j is to be the successor of P′ in the partial property graph. Belji is to be again a list of all permitted allocations of the temporal variables of P″j, starting with T′0=R, the reference point in time belonging to N. A property is formed by means of the property generator (17) according to the following scheme:
This property differentiates from the property created by the property generator (13) as follows: In essence, the check is limited to those properties P″, which represent a continuation of P′ in the sense of the partial property graph. Therefore, the temporal variables T′j are added to the lines 8 to 11 and the requirements, sequences and localization conditions are added to the lines 17 and 18.
The property is proven through the use (18) of the property checker based on the trivial model (3) or a disproof (19) is generated and relayed to the user. In this way, the loop control (16) is abandoned and an additional two consistency properties are formed and proven via Preset.
The reset property Preset is transferred to a consistency property by the property generator (20) in accordance with the following pattern:
The lines 4 to 6 herein designate the temporal variables from Preset, the lines 7 to 9 temporal variables from Poreset, which are equated with the corresponding temporal variables from Preset. Line 12 again ensures equal entries in both sets of variables and in lines 13 and 14, the validity of the localization conditions and the requirement from Preset are assumed. Under these assumptions, it is proven that the localization conditions and the requirement also apply to the copied set of parameters.
This theorem is proven through the application (21) of the property checker or its disproof (22) is relayed to the user. In the property generator (23), an additional consistency property for Preset is formed in accordance with the following scheme:
Up to line 12, the property corresponds to the output from the property generator (20). In the lines 13 to 16, the validity of the reset property based on both sets of variables is required. In lines 18 to 20, the equality of the output signals based on both sets of variables is requested and in lines 21 to 23, the equality of the data interpretation functions to their evaluation points is also requested.
The property acquired in this way is proven through the application (24) of the property checker or a disproof (25) is relayed to the user.
If all applications (7, 10, 14, 18, 21, 24) of the property checker return a proof for the respective consistency properties, the user is issued a confirmation of success (26).
Proof
It should be shown that the behaviour of the output signals is unambiguously/uniquely determined, in any case following the points in time Ki specified in the reset property and if the environment of the circuit is sufficient for the environment condition corresponding to the property environment-constrain (see below), in the event of a confirmation of success (26) through the properties from the quantity of main properties from (1).
For this, it is assumed that two circuits S and So are present. S is to fulfil all main and partial properties P and So is to fulfil all copied main and partial properties Po. It should be shown that S and So are similar to one another. Where In0, In1, . . . is a series of inputs, in which the reset at the beginning is activated corresponding to the reset property and which fulfils the environment conditions E corresponding to the property environment_constraints. Where Ino0, Ino1, is the copy of the series of inputs in the input signals of So.
The idea behind the proof is that a series Pi of properties and a series si of simulation points in time can be determined, from which it is possible for the proofs,
Therefore, an allocation of the temporal variables Ti from the reset property exists such that its requirement A and the localization conditions L1, L2, . . . and so on are valid. According to this, the assumption of the property “first_reset_test” is fulfilled by the property transformer (16) for this allocation of temporal variables. Since this property has also been proven, all corresponding localization conditions Lo1, Lo2, . . . and the requirement Ao of the copied reset property apply accordingly. In this way, the assumption of Poreset is fulfilled. Since Preset itself on the circuit So applies, the validity of the conclusion Co results. Since A and L1, L2, . . . apply, the validity of the conclusion C also results. Thus, the assumption of the property “second_reset_test” generated by the property generator (23) can be applied. This ensures the equality of the output signals within a certain temporal range. Furthermore, the equality of the data interpretation functions is affirmed at the respective evaluation point in time.
Inductive step: Under the assumption that the applicability of the properties P0 . . . Pn to the series of inputs has been ensured, whereupon each t is to be instantiated by the elements of the purely monotonously increasing series s0, s1, . . . sn. It is furthermore assumed that the usability of the properties Po0 to Pon is ensured for the series of inputs. In addition, it is assumed that t can also be set to sn for the use of Pon and that, assuming this, the temporal variable allocations Bel and Belo in Pn und Pon can be equally selected. This returns the equality of the output signals oi and ooi up to the point in time, which results from ki under the allocation Bel.
The following is written in order to provide a simplified representation: P=Pn and Po=Pon. Since P on the circuit applies, the affirmation C also applies and the property first_case_split, which was created by the property generator (13) for P can accordingly be used. According to this, a partial property P′0 exists, which represents a source in the partial property graph and an allocation Bel′0, under which the requirement A′0 and the localization conditions L′01, L′02, . . . apply based on the series of inputs In. Since P′0 applies to S, C′0 also applies. As a result, the property second_case_split from the property transformer (17) can also be used. This is the property (17), which affirms the existence of a partial property P′1 for us, which is longer than P′0 and for which, an allocation Bel′1 exists such that A′1 and L′11, L′12, . . . apply based on the series of inputs In. This consideration can be applied repeatedly up to a main property P′ and an allocation Bel′ of the temporal variables of P′ have been found such that A′ and the localization conditions L′1, L′2 of P′ apply based on the series of inputs In. The value for sn+1 results from the first element of Bel′, which is larger than sn due to a request concerning the reference points in time such that the series s0, s1, . . . sn+1 is now also recognized as strictly being of a monotonous progression.
According to the considerations in the previous section, the assumption based on In of the property first_test created by the property generator (6) is fulfilled for P and P′. According to this, the requirement A′o and the localization conditions L′oi of the copied P′o are also fulfilled, if the temporal variations of P′o are selected as in Bel′. In this way, the requirement and the localization conditions of P′ and P′o are fulfilled. Since P′ applies to S and P′o to So, the validity of C′ resp. C′o is consequently shown under the allocation Bel′. As a result, the assumption of the consistency property second_test created by the property generator (9) for P and P′ is fulfilled. The equality of the output signals oi and ooi up to the point in time k′i and the equality of the data interpretation functions D′i and D′oi can thus be determined at their evaluation points in time p′i. This proves the inductive step.
It is to be shown that the value of the output signals is equal at any point in time: Where o is an output signal, P any property and kP the upper bound belonging to o of the determinedness range of o in the property P. P. kP results through T+vP, where T is any temporal variable, vP an offset dependant on the property. Through the constraints concerning the selection of the temporal variables, T>=t and therefore kP>=t+vP. Where V is the minimum of all vP from a quantity of properties. Due to the strict monotony of the series of si, sn>n and consequently kP>=n+V. It correspondingly follows, no later than following the X-Vth inductive step, that oi @X=ooi @ X applies to an arbitrary point in time X. In this way, the proof is completed.
Applicable example of an interrupt controller. An interrupt controller is intended to process 8 interrupt inputs, which are to be numbered from 0 to 7. They should have set priorities corresponding to the numbering, whereupon 7 is the highest. Interrupts may appear as pulses. They are temporarily stored in the register pending_int. With the register mask, individual interrupts can be deactivated, in the event their assigned bit is set to 0.
If an interrupt is present, then the request signal int_req will be activated. Then the acknowledged is awaited, which is coded as a pulse in the signal int_ack. This acknowledgement is to arise between the 2nd and 11th cycle following the increasing edge of int_req. The interrupt controller issues a cycle following the arrival of the acknowledgement via the bus d_out as to which interrupt has the highest priority unmasked. At the same time, it blocks this said interrupt and all interrupts of lower priority.
The processor can describe the mask register via a bus access operation. For this, it activates the signal write_mask and returns the desired value in the signal d_in.
The setting of the unmasked interrupt of highest priority is to be occur in the function prioritize. The function reduce_mask is to modify the mask register as requested. The annotated quantity of properties (1) is to contain:
For this, the type state_t represents a listing type, which contains the value idle along with other values, which are necessary as interim steps for the RTL implementation, but which are not referenced in the properties.
Reset property: Name is “reset” and Treset=t+1
Macros for the definition of the environment conditions:
e1 affirms that the processor acknowledges the interrupt by the activation of the signal int_ack following no more than 10 cycles.
Quantity of Properties
Property Reset:
The quantity of temporal values TTreset only consists of the temporal variable T0, for which no processing range declaration and localization information exist.
Property Idle Concerning that No Operation is Executed:
Property wm Concerning the Modification of the Mask Register Mask:
Quantity of temporal variables TTwm only consists of the variable T0
Partial Property Graph
In this case, the partial property graph only additionally contains the partial property int_p with a quantity of temporal variables TTint
and a conclusion Cint
The graph consists of the isolated nodes idle and wm and of an edge between the additional nodes int_p and int.
The properties in the property description language ITL are listed in Annex I. The macros a_reset, a_int, a_idle, a_wm, a_int_p designate the requirements and if applicable, also the localization information for the temporal variable t_ack. c_reset, c_int, c_idle, c_wm and c_int_p designate the affirmations.
The trivial model (3) will be created through the translation of the following VHDL description into the internal format of a property checker.
This represents the description by the model generator (2).
The property copier (4) creates copies of the properties for various parameter sets. This is represented through the macro definitions a_int_1, a_int_2, a_int_p_1, a_int_p_2, and so on in the Annex. The actions of the property generators are represented by means of the properties created, which can be checked using the property checker based on the trivial model.
The formal verification of properties via the model and the corresponding check for completeness can be conducted in increments. A property is proven based on the model and subsequently demonstrated using the procedures presented here that it functions together with the other properties.
Automatic Determination of all Redundant Specifications:
In the following text, a stage previous to the invention-pursuant procedure will be described, in which only the quantity of properties and a signal declaration must be specified. It determines all user specifications resulting from the main procedure that can be determined such that only the environment assumptions have to be specified as an additional user specification for the main procedure.
This representation is limited to the environment assumptions, which are not dependant on output or internal signals.
A process diagram is provided in
The procedure itself operates with internal parameters, which have nothing to do with the internal parameters of the tested circuit. In step 2 of the procedure, the internal parameters of the procedure are pre-allocated in the following manner: The quantity of the successors NNP is pre-occupied by the quantity of all properties for every property P. For every pair P, Q, the quantity MRefPQ of all possible reference points in time is pre-occupied with a list of all points in time of the form T+u, for which T is a temporal variable of the property P and an entire number u is selected such that T+u is located in the inspection window of P for at least one allocation of T. The order of the elements in the list is arbitrary.
For every output and internal signal s, the amount of the temporal ranges DZPs, in which the definiteness of s is to be checked, is pre-occupied as follows: All temporal ranges, at which s is evaluated in the property P corresponding to the following definition, are adopted in a first pre-occupation step:
In a second pre-occupation step, each temporal range [a, b] is inspected, which has been adopted into DZPs in the first pre-occupation step. If there are points in time therein of the form T+u (T being a temporal variable and u a whole number I), which are contained in this temporal range [a, b] for all possible allocations of the temporal variables of P, then the point in time T+u represented by the temporal range [T+u, T+u] will also be adopted in DZPs.
In the steps 10 to 15, the initialization or reset property is identified. For this, the quantity MReset is pre-occupied in process step ten as an empty quantity, in which the reset property is to be subsequently located. In the loop 11, all properties P are run and transferred to a consistency property using the property generator 12 for the property scheme “first_reset_test”. This is checked in step 13 by the property checker based on the trivial model. In the case of a proof, the property P is added to MReset in step 14, in all other cases it is not added. In this way, the loop 11 is completed. In step 15, a check is conducted as to whether the MReset only contains one element, that is, whether the reset property PReset has been unambiguously/uniquely determined. If not, the procedure is interrupted. In step 16, the reset property from all quantities of successors NNP is removed.
In the steps 20 to 22, the quantity of the determinedness temporal ranges of the reset property is determined. For this, a consistency property similar to second_reset_test is assembled in the property generator 20. The lines 18 to 23 are, in deviation to this scheme, formed by
during[a,b]:s=so
whereupon s is every output and internal signal and [a,b] every temporal range from DZPReset s.
This property is subject to the property test 21 based on the trivial model. In the case of a counter example, it is determined in the counter example analyzer 22, which partial conditions for the affirmation of the consistency property have been violated and the corresponding temporal ranges are removed from DZPReset s. The property generator 20 is then restarted. This loop is completed once the property check 21 has found a proof.
In step 30, copies of the current allocation of the internal parameters of the procedure are created. All properties P are run in the loop managed by the loop control 31. All properties Q from the quantity of successors NNP to P are run in the loop managed by the loop control 32. Consistency properties for P and Q are created in the property generator 33 according to the scheme “first_test”. For this, the first value in MRefPQ is selected as the reference point in time. Furthermore, the lines 24 to 29 are replaced in the manner described above for the property generator 20 by means of statements concerning all output and internal signals for the quantities DZPs in the temporal ranges. In the property test 34 based on the trivial model, the assumption may arise as being self-contradictory or the property is disproved or the property is proved. In the event of a disproved assumption or disproved consistency property, the first element of MRefPQ is removed in step 35 and all DZRs are occupied with their initial value for all properties. In step 36, it is checked whether MRefPQ has become empty as a result of this. In this case, the property Q is removed from NNP in step 36 and continued with the next process step corresponding to the loop administration 32. If it was found in step 36 that MRefPQ is not yet empty, then it will be continued with process step 33 on the basis of the modified parameters.
If the consistency property is proven in the property check 33, then a property corresponding to “second_test” is created by the property generator 40. In turn, the lines 24 to 29 of this property scheme are replaced by conditions concerning the temporal ranges of DZPs and the lines 35 to 40 by conditions concerning the temporal ranges of DZQs. During the property check 41 based on the trivial model, either a counter example may arise or the consistency property is proven. If a counter example is returned, then the lines in violation are determined in the counter example analyzer 42 and the corresponding temporal ranges are removed from DZQs. Afterwards, it continues with process 40 on the basis of the modified parameters.
If the consistency property in step 41 is proven, then it continues on with the next process step corresponding to the loop administration 32. If the loops 32 and 31 are finished, it is then determined in process step 50 as to whether the current copies of the internal parameters of the process and those saved in the process step 30 have stayed the same. If they are not identical, then step 30 is continued on the basis of the new allocation of the internal parameters.
If they have stayed the same, then it is ensured that the series of properties described in the first elements of the MRefPQ by NNp and the reference points in time are chains. Furthermore, the DZQs clarify at which times the output signals in these chains have been determined. In step 51, the entries for the main procedure are now determined and issued as follows:
In this way, all redundant information is determined which is required for the completeness check and for determining the measure of completeness.
The request oi=ooi for the determinedness of the output signals can be diminished. This is particularly useful, if output signals are only allowed to be evaluated once they have been designated as valid by means of a validation signal. For this, a quantity of determinedness conditions Det1, Det2, . . . uniform for all properties is defined in the annotated set of properties. Said conditions are described on the basis of all signals s and so of both signal quantities. The determinedness time frames will then no longer be specified for every output signal, but rather for every determinedness condition.
By means of limited adjustment of the procedure, it can be ensured that the model properties prove the validity of a Boolean formula (which may also evaluate signal values at other points in time using prev and next) at any point in time.
The determinedness conditions oi=ooi are supplemented for this purpose by means of requests of the form cond=‘1’, whereupon cond represents the questionable Boolean formula. Justification: Based on the proof of the procedure invented, it is apparent that a condition of the form cond=‘1’ and condo=‘1’ proves that cond is determined and is always ‘1’. In the case of a counter example, either cond=‘0’ is true. This counter example would therefore also contradict the request cond=‘1’. In the event condo=‘0’ is true, an allocation of the copied variables exists, which was the cause of this counter example. This allocation, however, can be transferred to the uncopied variables, which then returns cond=‘0’.
In the lines 21 and 23 of the property scheme first_test and second_test, the line 12 in the property scheme first_case_split and the lines 16 and 18 of the property scheme second_case_split, it is planned that the requirement and sequence of a predecessor property P is completely adopted in the assumption of the respective consistency property. In many cases, however, it is sufficient to adopt only components a, ao, c and co of the assumptions A, Ao, C and Co of the properties P and Po, which are formed such that A=>a, Ao=>ao, C=>c and Co=>co apply. The quantity of temporal variables tt resp. tto, based on which these components are defined, can then be smaller than the original quantities of temporal variables TT resp. TTo. tt must be formed such that there is a signed element to from TT and that with every temporal variable unequal to t0, which is used in a, c, in the definition of the right limits ki for the determinedness intervals or in setting of the evaluation points in time pi of the data interpretation function Di also contains every corresponding basic temporal variable. tto is to contain call copies of the temporal variables of tt. The declarations for the temporal variables Ti and Toi in the property schemes can then be reduced to the elements of tt and tto as well as the lists of the localization information Li. t0=t is declared in deviation to this. The temporal range, in which the equality oi=ooi of the output signals is assumed, can be likewise reduced or these assumptions will be completely erased. All reductions described reduce the run times of the proofs for the consistency properties.
Examples for the formation of a from A and c from C (and in a series thereof also ao and co) are:
a=true and c is specified by the user. The user specifications can e.g. describe the final state in terms of how it was used in a simple introductory example. In this case, identical consistency properties can arise for various properties and must then also only be inspected once.
Syntactic analysis: The overlapping set of the signals for every pair P, P′ is formed, which on the one hand occur in A or C and occur on the other hand in A′. a and c are then formed from A and C, whereupon all logical lines are erased, in which none of the signals from the overlapping set occurs. tt is then selected corresponding to the above mentioned conditions.
Second Example of Execution:
The invention-pursuant procedure of the preferred form of execution described previously is not limited to the explicit specification as to which properties follow one another with which reference points in time. The basic idea of the interlinking of properties is also utilized in the following procedure. However, consistency properties are assembled in this procedure, for whose proof these chains are implicitly formed.
A significant component of these consistency properties is manifested by implications, into which the main and partial properties are transferred. The following implication is formed for a property P with the requirement A, the affirmation C, the localization conditions Li and the environment assumption E:
P′(t,Bel)=(A and L1 and L2 and . . . )implies(C and E);
P′ is instantiated with the point in time t and the allocation Bel of the temporal variables. The following is formed for the properties P1, P2, P3, . . . :
The point in time t_anf and t_end are to be selected in a sufficient size. assume_props is to also be transferred to the copied variable set, this is to return the condition assume_propso(t_anf, t_end).
assume_props and assume_propso are to be given in the assumption of a consistency property. t_anf and t_end are to cover a time frame of sufficient size. The prover will select an instance of the implications, whose left side is fulfilled by the start of the chain and the input series for a suitable start of a chain (that is for a condition concerning the value of the internal signals) and for an arbitrary series of input signals. The right side then provides the information concerning outputs and internal signals, which in turn can fulfil a left side of an implication, thereby providing further information and so on.
This is the basis of the following procedure: In the property graph, which is presented in
The condition is parameterized through the allocation Bel of the temporal variables in the manner described for first_case_split. Furthermore, the point in time t of the model property is emphasized in order to be able to instantiate it in a suitable manner. The start condition is formed by
For this, the P1, P2, . . . run through all start nodes. The end condition is formed by
For this, the Q1, Q2, . . . run through all properties, which have been marked as end nodes.
assume_props is formed from the following properties in the form described above:
For this, the variable w must be chosen sufficiently large in order to cover the longest path in terms of time through the property graph and m should be selected such that it covers the largest inspection window length of the properties, from which assume_props was formed. environment_constraints describe the environment assumptions, which are assigned to the properties, which occur in det_cond_start.
Following the proof of both consistency properties, it is ensured that the properties in the partial graph of the property graph defined by the start and end points are able to cover every input series such that the values of the output signals are unambiguously/uniquely determined based on the respective intervals det_intervali.
If the quantity of the start and end nodes is identical during the disassembly of the property graph and the intervals det_interval have been chosen in a suitable manner, both of these properties are sufficient for proving the completeness of the corresponding property set with the exception of the behaviour in the proximity of the reset. The completeness then results from induction: First of all, det_cond_start(t)=det_cond_end(t) applies. reach_det_cond shows that det_cond_start can be reached after a finite period of time and check_det_cond shows that the outputs in the interim period are unambiguously/uniquely determined.
It can be similarly argued that when the property graph is broken into multiple partial graphs such that all end points of a partial graph are also the start points of said graph or of another graph and vice-versa, every start point is also an end point.
This procedure presents an extension of the previously described procedure because it has the significant basic idea in common with it, that is, the formation of chains of properties and the inspection that the values of the output signals are unambiguously/uniquely determined along these chains.
Third Example of Execution
The invention is not limited to properties, which have been processed by means of formal verification. Rather, the properties can also be checked with a simulation instead of with the procedure as per
The execution example described here monitors that a quantity of properties is able to actually recognize every error in the output pattern of a simulation. In this way, it is the first known aid for the automatic quality assurance of error detection processes during the simulation.
Fourth Example of Execution
In the procedure of the fourth example of execution, the quality of a quantity of measurement conditions for determining the diversity and variety of input patterns for the simulation is to be determined. In this manner, it can be ensured that a quantity of “functional coverage points” illuminates every component of the circuit.
To present, conditions have been formulated, whose fulfillment is counted over the course of a simulation. These numbers are used to evaluate the diversity and variety of input patterns for a simulation (coverage). The conclusiveness of this evaluative procedure is dependant on the selection of conditions. There are two classes of conditions: The one class recognizes the execution of components of the circuit description (structural coverage), while the other recognizes specific behaviour based on signals from the circuit. The conditions of the first class can be determined in an automatic and consequently highly efficient manner. An important class of such evaluative procedures arises from conditions, which are oriented with the code of the circuit description. Such conditions are e.g. then fulfilled, if a line in the circuit description is executed. These conditions are automatically generated for every line and the subsequent evaluation of quality determines the relation between the lines, which have been executed at least once and all lines. This procedure ensures that every circuit component has been checked at least once. The quality measurement, however, does not consider various functional effects, e.g. the inherent parallel nature of circuit descriptions. If the verification engineers recognize necessity, they then manually insert temporally logical conditions, which are fulfilled for such problematic processes. The specification of such conditions is time-consuming and remains therefore reserved as critically recognized components of the circuit functionality.
The assumptions from a complete quantity of properties combine both advantages: They code functional aspects and check each component of the circuit. To this extent, the invention also enables the provision of a particularly critical set of such provisions. The affirmations concerning the output signals are of no significance for this such that the properties need not be generated in accordance with the scheme “second_test” and not be checked.
If an annotated quantity of properties has not been fully recognized as being complete, the possibility of generating a measure related to remaining work to be performed and can thereby play a similar role in the planning and monitoring of a verification process such as the (coverage) measure for diversity and variety for the input patterns in the context of the simulation.
In contrast to the context of the simulation, where the measures are certainly able to give a false impression of security, the procedure described in the following section only returns the value 1 once the quantity of properties is complete. In addition, the measure, which is determined for an upper quantity B of a quantity of properties A, is always greater than or equal to the measure for A.
In the following representation, ITL conditions are regarded as Boolean expressions concerning variables, which consider the value of a signal at one point in time. The variables themselves are not subject to time, rather several variables result when a signal is evaluated at different points in time. In actuality, the variables are pairs (<signal>, n). Their allocation with a value stipulates the value of the signal at the point in time t+n. As a result, <signal>@t+n applies to these variables, whereupon t is the arbitrary, but fixed point in time from ITL, relative to which all time data are generated.
Where B is a condition, in which no temporal variables except for t occur. Then, the carrier sup(B) is the quantity of all variables s@t+n such that s is a signal that is reference in the condition B at the point in time t+n. If B also contains temporal values, sup(B) is defined as sup(B(Bel1) or B(Bel2) or B(Bel3) . . . ), whereupon the Beli fathom all values of the temporal variables that are permitted. sup(B) is a finite quantity of variables, which is to be arranged linearly in some manner, perhaps through lexographic arrangement. In this way, the elements of this quantity of variables form a finite tuple. Where the quantity of definitions def(V) of a quantity of variables is the quantity of all value tuples, which can correctly to these variables occupy the variable tuple from V based on their type. Instead of def(sup(B)), one can also write def(B).
Example: For a 2-element bit vector sig1 and a sign-bit signal sig2, the ITL condition B corresponds to=
from the Boolean expression
sig1@t=“00” and sig2@t=‘1’ and sig2@t+1=‘1’,
has the support sup(B)=(sig1 @ t), (sig2 @ t), (sig2 @ t+1)) and the quantity of definitions
def(sup(B))={(“00”, ‘0’, ‘0’), (“00”, ‘0’, ‘1’), (“00”, ‘1’, ‘0’) (“00”, ‘1’), (“01”, ‘0’, ‘0’), . . . (“10”, ‘0’, ‘0’), . . . (“11”, ‘1’, ‘1’)};
Where B is a condition, whose carrier is a partial quantity of a quantity of variables V. Then, the cardinality card (B, V) of B based on V is to fulfil the number of elements from def(V), if one correspondingly allocates the variables B. card(B, V) can be automatically determined e.g. when the BDD for B is generated and then the applicable algorithms are utilized for this. Instead of card(B, sup(A)), one can write card(B, A).
For any quantity M, card(M) is to designate the number of the elements in M.
The basic idea of the procedure of the fourth example of execution lies in considering the properties and their quantity of successors as transitions in a machine, which is referred to as the property machine. The property machine for the first example of execution is represented in
The property machine of an incomplete quantity of properties will not clearly produce an output pattern for every input pattern. The basic idea of the procedure lies in accounting for the portion of the input patterns, for which the outputs can be determined.
An example for a property machine for an incomplete quantity of properties is specified in
In an important step of the procedure, the property machine is processed. As a result, a tree starting with Preset originates. A property is assigned to each node and the properties from the quantity of successors are assigned directly to its sons. The branches of the tree are cut as soon as a property is assigned to a node, which already appears on the path between Preset and said node. The node with the re-occurring property is then called a complete end node in contrast to empty end nodes, whose property has an empty quantity of successors. Such a tree is called an execution tree of the set of properties.
It is determined for every path of Preset for a full or empty end node, which portion of all possible stimuli has been covered by it. This value is reduced, if the path leads to an empty end node or if the outputs along the path are not precisely described. The sum of these values adjusted in this manner returns the measure of completeness.
It is only equal to 1, if the quantity of properties is complete in the sense of this patent application. In all other cases, it is less than 1 and increases with every property, which increases the property machine.
The procedure for determining the completeness measure is a particularly preferred procedure in terms of the invention in question. The properties, which are formed in accordance with the scheme first_test, check the accuracy of the quantity of successors. By means of the repeated application of properties, which are formed according to the scheme second_test and successive reduction of the affirmations concerning the output determinedness, it can be determined for a pair of properties P, P′ as to how complete the output description is by means of P′. Furthermore, it is determined which portion of the input patterns covers a requirement in the disjunction according to first_case_split and second_case_split.
For the following representation, it is to be assumed that the assumptions of the main and partial properties are selected such that the individual lines
A′i(Belij); L′i1(Belij); L′i2(Belij); . . . .
in the properties mutually exclude one another in accordance with the scheme first_case_split. That is, the conjunction for every two such lines is to be empty under the assumption issued by the scheme. Furthermore, the individual lines
A′i(Belij); L′i1(Belij); L′i2(Belij); . . . .
in the properties are to mutually exclude one another in accordance with the scheme second_case_split. If this is not the case, a priority must be established among the properties involved, for example by means of numeration P1, P2, . . . of the properties. The requirement of the original property Pn must then be replaced by
In the loop control (102) of
In the copier (104), the temporal variables of P′ are differentiated from those in P in the manner from the first part of the invention and subsequently designated with T′i. At the same time, T′0=R is set and the temporal variably is set to T0 by t in P. In the first coverage value determiner (105), a coverage value w(P, P′) is then assigned to the sequence from P and P′ as follows:
Where
G=A and C and E and L1 and L2 and . . . .
the assumption corresponding to the scheme first_case_split. Where
H=G and A′ and L′1 and L′2 and
the condition for this that the property P′ with its portions A′, C′, L′1, L′2 is utilized in a property corresponding to first_case_split. The coverage value for the sequence from P and P′ is determined as
w(P,P′)=card(H,H)/card(G,H)
As a result, the loop (103) is completed via the sources P′.
The loop control (106) is to now run through every edge P′→P″ in the partial property graph, which is located on a path to a successor property of P.
In the copier (107), the temporal values T′i of P′ and T″i of P″ are, on the one hand, differentiated from those in P and on the other, T0=t, T′0=R and T′i=T″i is set for all temporal variables of P′.
Parallel to the above representation, the coverage value w(P′, P″) is determined in the second coverage value determiner (108) as follows:
Where:
G′=G and A′ and L′1 and L′2 and . . . .
the assumption corresponding to the scheme second_case_split apply.
Where:
H′=G′ and A″ and L″1 and L″2 and . . .
the condition for this that the property P″ with its portions A″, C″, L″1, L″2 is utilized in a property corresponding to second_case_split. Then, the coverage value w(P′, P″) is defined for
w(P′,P″)=card(G′,H′)/card(H′,H′)
In this way, the loop checked by the loop control (106) is abandoned.
The loop control (109) now runs through all main properties P′, which are successors of P.
In the main property coverage determiner (110), the coverage value w(P, P′) is determined as follows: Where Q0, Q1, Q2, . . . Qn is the path from a source of the partial property graph up to the main property P′=Qn. Afterwards, the following is calculated:
w(P,Q)=w(P,Q0)*w(Q0,Q1)*w(Q1,Q2)* . . . *w(Qn−1,P′)
These individual factors are all known since w(P, Q0) was calculated in the first coverage value determiner (105) and the remaining w(Qi, Qi+1) in the second coverage value determiner (108).
In the output determinedness determiner (111), it is evaluated whether and how many of the lines
from the properties corresponding to the scheme second_test must be deleted in order to make the remaining property true. This is conducted in an iterative procedure, in which counter examples are automatically evaluated and lines violated by the counter example are deleted prior to the next application of the property checker. The output determinedness value w′(P, P′) is determined as the quotient of the number of remaining lines opposite the number of all output signals.
Where u is a constant in the closed interval [0, 1] In the determiner for the local measurement (112), the local measurement is now determined by
W(P,Q)=(u+(1−u)*w′(P,Q))*w(P,Q)
As a result, the inner loop (109) and the outer loop (102) end.
In the execution tree generator (113), the execution tree is now to be generated for the annotated set of properties (1).
Where v is a constant from the interval open on the right [0, 1[. The loop control (114) is to run through the nodes of the execution tree such that it starts with the end nodes and first processes an inner node once all successors have already been processed.
The node evaluator (115) now calculates a completeness value V(P) for every node P. If P is an empty end node, then
V(P)=v
With the constants v and u, the completeness measure can be adjusted to the distribution of work among individual partial tasks during the course of the formal verification.
If u=0, then the output determinedness is reflected to a very high degree in the final result. If u=1, then no consideration is given to the output determinedness. In this way, u can be used in order to allow the portion of processing load based on experience for ensuring the output determinedness to flow into the determination of the completeness measure during the course of the verification and consequently to bring the measure calculated in contact with the anticipated work. In any case, W(P,Q)=w(P,Q) applies, if the tasks have been completely determined.
v=0 results in all nodes that are located on paths to empty end nodes being evaluated with 0. The completeness measure then accordingly remains at 0 as long as the transition graph of the property machine does not contain a loop. This is potentially an unwanted effect since properties have obviously already been described such that a value/=0 is suggested. However, since the closing of the loop is a time-consuming verification step based on experience, the completeness measure should remain small as long as the loops are not closed and therefore, v should also remain close to 0.
It is to be shown that the procedure has the following properties:
For the proof of 1), it is first shown that the validity of a property corresponding to the scheme first_case_split has the consequence that the following applies in the situation of the properties corresponding to the scheme first_case_split:
w(P,P′1)+w(P,P′2)+w(P,P′3)+ . . . =1
For this, the P′i describe all sources of the partial property graph, from which a path leads to one of the properties from the quantity of successors.
In advance, some rules for calculating with card(A,B) are proven: Where H is any condition, V a finite quantity of variables, which is the upper quantity of sup(H) and W a finite quantity of variables disjoint to V. + is to designate the unification of quantities. Then
card(H,V+W)=card(def(W))*card(H,V)
for every upper quantity V of sup(H).
Where H1, H2, H3, . . . are the conditions, which are formed in the first coverage value determiner (105) from P′1, P′2, P′3. The quantity
V=sup(H1)+sup(H2)+sup(H3)+ . . .
(+ stands for the addition of quantities) is in any case an upper quantity of every sup(Hi). Then, the above-mentioned can be used and one acquires
The last reformulation is possible because it was assumed that the requirements of P′i are disjoint and therefore, the conjunction of any pairs of Hi is empty. The following reformulation is therefore possible:
The second last reformulation is valid because the validity of the property was required in accordance with scheme first_case_split.
In a similar manner, one can see that the following applies to a partial property P′ and its sons P″1, P″2, P″3, . . . in the partial property graph
w(P′,P″1)+w(P′,P″2)+ . . . =1
For every main property P and any arbitrary node Q′ in the partial property graph of P, s(P, Q′) is to be the sum of all w(P, Q), whereupon Q runs through all successor properties of P, which are to be reached by means of a path from Q′. Where Q0, Q1, . . . Qn is a path from a source to Q′ and Qn=Q′. This path is unambiguously/uniquely determined by means of the stipulations concerning the partial property graph. By means of the complete induction, it is proven that
s(P,Q′)=w(P,Q0)*w(Q0,Q1)*w(Q1,Q2)* . . . *w(Qn−1,Q′)
Embedding of induction: The claim is trivial if a sink of the partial property graph for Q′ is selected because the sinks indeed represent the main properties.
Inductive step: The induction is to run through the node of the sinks up to the source in the breadth-first search such that the claim already applies to all successors R1, R2, . . . of Q′ in the partial property graph for determining s(P, Q′). First of all, all paths lead through P and Q′ to main properties, which are located in the quantity of successors of P, due to the requests of the partial property graph. As a result, the following applies
The contents of the brackets in the last line were proven to result for 1 further up. Therefore the following applies
s(P,Q′)=w(P,Q0)*w(Q0,Q1)* . . . *w(Qn−1,Q′)
With this, the inductive claim is proven.
The main properties are designated with P1, P2, . . . from the quantity of successors of P.
Furthermore, the sources of the partial property graph, from which paths lead to properties from the quantity of successors of P, are designated with Q1, Q2, Q3, . . . .
The above consideration has demonstrated that
This value is equal to 1 as per the above considerations.
It will now be shown that V(P)=1 is valid for every node P in the execution tree of a complete quantity of properties. This is demonstrated by means of complete induction via the execution tree:
Embedding of induction: There is no property without successors in a complete quantity of properties. As a result, every end node is full and therefore, the inductive claim applies to all end nodes.
Inductive step: The induction is to traverse all nodes in the reversed breadth-first search, that is a node P is first selected, when the inductive assumption has already been proven for its sons Q1, Q2, . . . . Then the following applies
Under the assumption that the quantity of properties is complete, all output signals have also all been tested by means of the property second_test. As a result, W(P, Q)=w(P, Q) applies and consequently
V(P)=v+(1−v)*(w(P,Q1)+w(P,Q2)+ . . . )
The sum in the brackets is 1 due to the considerations above, therefore
V(P)=v+1−v=1
As a result, the inductive claim is proven. For this, it also particularly applies that V(Preset)=1 and 1 is equal to the completeness measure.
Where M1 and M2 are an annotated quantity of properties and M1 a sub-quantity of M2. The properties also contained from M2 should not have any contradictory assumptions. The consistency conditions concerning the quantities of successors require of M2, that the execution tree of M2 then also contains the addition properties. Furthermore, v/=1, v/=0 and u/=0 apply.
The execution tree of M1 is a sub-tree of M2. First of all, it is demonstrated that the following applies to every node P of the execution tree of M1 and to the relation between the evaluations V1 and V2 for M1 resp. M2:
V1(P)<=V2(P)
and even
V1(P)<V2(P), in the event the partial tree of the execution tree beginning in P contains properties, which arise in M2, but not in M1.
The proof is conducted again by means of induction. This follows a reversed breadth-first search.
Embedding of Induction:
If P is a full end node in M1, then it is also a full end node in M2. Therefore, the inductive claim applies.
If P is an empty end node in M1, then it can be an inner node in M2. If it is an inner node, then the following applies:
V1(P)<V2(P)
This results from the definition of V1 resp. V2, from v<1, from V(P)>0 for every node and from W(P, Q)>0 because u/=0 was required. This corresponds to the inductive claim.
If P is an empty end node in M1 and not an inner node in M2, then it is also an empty end node in M2 such that the inductive claim once again applies.
Induction:
In a reversed breadth-first search, the execution tree is once again traversed. An inner node P in the execution tree of M1 is to be observed. It is to have the sons Q1, Q2, . . . , to which the inductive claim applies. If the partial trees of the two execution trees beginning with Qi are equal, then V1(Qi)=V2(Qi) applies for all i. As a result, V1(P)=V2(P) also applies because the same numbers are inserted in the definition.
In all other cases, there is at least one Q among Qi with various partial trees and therefore V1(Q)<V2(Q). If these numbers are inserted in the definitions of V1(P) and V2(P), the following is returned as a result
V1(P)<V2(P)
because multiplication is conducted using positive numbers. As a result, the inductive claim is also demonstrated for the inductive step.
As a consequence thereof, the following applies
V1(Preset)<V2(Preset).
Measure for an Incomplete Quantity of Properties
An incomplete, annotated quantity of properties is to be given. As a result, it can be furthered by at least one property, which is disjoint to all the other ones. This proof shows that the furthered quantity of properties has a significantly larger completeness measure than the original quantity of properties. At the same time, the completeness measure is limited to a maximum of 1 as shown by the first part of the proof. Thus, the completeness measure of an incomplete quantity of properties is less than 1.
Application Example of the Fourth Example of Execution
Reference will be made to the interrupt controller corresponding to the application example for the first example of execution and one observes an incomplete quantity of properties, which consist of the properties reset, wm and idle. All output descriptions should be precise. The quantities of successors are to be correspondingly reduced by int; in the partial property graph, int_op appears and int does not appear. v=0.1 is to be given.
The graph of the resulting property machine is provided in
Conjuncts, which already occur in Greset, were omitted for the last reformulation. For the following reformulation, some rules concerning card arising from: A and B are two conditions, V an upper quantity of sup(A), W an upper quantity of sup(B) and V and W are utilized disjoint to one another. Then card(A and B, V+W)=card(A, V)*card(B, W). If A and B are two conditions and V an upper quantity of sup(A+B), then card(A or B, V)=card(A, V)+card(B and not A, V) applies.
The elements of sup(H) can be easily read from the above representation. sup(G) does not contain write_mask @ t+1 such that
Parallel to above, the following therefore results
w(reset,idle)=0.5
w(reset, idle) and w(reset, wm) supplementing one another for 1 complies with the validity of the property corresponding to first_case_split.
Where G′wm is defined as Gwm, whereupon however the line via pending_int is skipped. Then the following applies:
With g=(pending_int @ t+1=(pending_int @ t or int_in @ t) and
((pending_int @ t or int_req @ t) and d_in @ t=0)) applying
First of all, card(pending_int @ t+1=(pending_int @ t or int_in @ t), g)=2**32 applies since from the five 8-bit signals in sup(g), all of them except of pending_int @ t+1 can be freely selected and for each selection, there is a directly corresponding pending_int @ t+1.
For the determining of card(g,g), a case differentiation is conducted concerning number of zeros in d_in@t. For each one there are several 8!/((8−n)!*n!) possibilities to generate a bit array with n zeros. At the positions of these zeros, ones may occur in pending_int @ t and in int_req @ t, all other positions must be 0. For every variable, there are 2**n possibilities for filling the allowed positions with 0 or 1. Therefore, there are (8!2**2n)/((8−n)!*n!) possibilities of occupying these three variables such that on the one hand, d_in @ t contains exactly n zeros and on the other hand, ((pending_int @ t or int_req @ t) and d_in @ t=0) applies. The sum concerning this expression for n=0, 1, . . . 8 returns 390625. For every combination, int_in@t can be freely selected and then there is only exactly one suitable pending_int t+1. Thus, the following applies
card(g′, g) is calculated as following: Based on assumptions similar to those above it follows that there are 6561 possibilities of fulfilling (pending_int @ t and mask @ t)=0. For each of these possibilities, int_in can be selected freely and it then returns exactly one mask @ t+1 respectively pending_int @ t+1, which can be suitably selected for it. Consequently card(g′, g)=1679616 applies.
For the calculation of card(g, g), it can be concluded as above that there are 390625 possibilities of selecting int_in t, pending_int@t and mask@t. From this selection, mask @ t+1 und pending_int @ t+1 clearly result. As a result, the following applies
w(idle,idle)=0.5*390625/1679616=0.116284
The completeness values of the nodes now result as in
Extensions of the two to four examples of execution:
Coverage Value Calculation
The coverage values can be calculated according to various algorithms. A conceivable alternative algorithm approximates it by generating random values, which fulfil G and it is checked as to how many of these random values fulfil the assumptions of the possible successor properties.
Determining Output Determinedness
One can skip to more general determinedness criteria. Furthermore, the criteria for determining the value w′(P, Q) can be selected in a considerable number of ways: One procedure sets w′(P,Q)=u as soon as the property has not been fulfilled according to the scheme second_test and is equal to 1 in all other cases. This completely abstains from iterations. On the other extreme, the determinedness of a signal is checked separately at every point in time.
Reduced Assumptions Concerning the Predecessor Property
As in the first example of execution, one can skip to simplified assumptions concerning the predecessor property. This has the advantage that, under certain circumstances, several properties produce the same reduced assumptions and in this manner, the calculations can be shortened.
The invention is not limited to the temporal logic described. It can be transferred to all linear temporal logics, that is, particularly to the standards PSL and SVA. Finally, it is crucial that the properties no longer enable a structured application on a singular basis, but rather in terms of forming chains.
| Number | Date | Country | Kind |
|---|---|---|---|
| 05020124 | Sep 2005 | EP | regional |
The computer program listing appendix submitted with this application, which is contained in the file SN12488672Code.txt having 29 KB, created on Feb. 23, 2009, is incorporated by reference in its entirety. The present application is a continuation of and claims the benefit of the filing of U.S. patent application Ser. No. 11/459,433, filed on Jul. 24, 2006, now U.S. Pat. No. 7,571,398, which claims the benefit of the filing date of European Patent Application EP 05 020 124.3, filed on Sep. 15, 2005, both of which are hereby incorporated by reference in their entirety.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5065335 | Yokota et al. | Nov 1991 | A |
| 5689685 | Feldmann et al. | Nov 1997 | A |
| 5880966 | Hogan | Mar 1999 | A |
| 6131078 | Plaisted | Oct 2000 | A |
| 6324678 | Dangelo et al. | Nov 2001 | B1 |
| 6530065 | McDonald et al. | Mar 2003 | B1 |
| 6594804 | Hojati | Jul 2003 | B1 |
| 6728939 | Johannsen | Apr 2004 | B2 |
| 6848088 | Levitt et al. | Jan 2005 | B1 |
| 7020856 | Singhal et al. | Mar 2006 | B2 |
| 7124383 | Chen et al. | Oct 2006 | B2 |
| 7159198 | Ip et al. | Jan 2007 | B1 |
| 7571398 | Bormann et al. | Aug 2009 | B2 |
| 20050010882 | Winkelmann et al. | Jan 2005 | A1 |
| 20050114809 | Lu | May 2005 | A1 |
| 20050159835 | Yamada et al. | Jul 2005 | A1 |
| 20050183048 | Kinzelbach | Aug 2005 | A1 |
| 20060156261 | Farkash et al. | Jul 2006 | A1 |
| 20060161413 | Wei et al. | Jul 2006 | A1 |
| 20060212837 | Prasad | Sep 2006 | A1 |
| 20060225022 | Ezaki | Oct 2006 | A1 |
| Number | Date | Country |
|---|---|---|
| 0079421 | Dec 2000 | WO |
| Number | Date | Country | |
|---|---|---|---|
| 20090327984 A1 | Dec 2009 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11459433 | Jul 2006 | US |
| Child | 12488672 | US |
