JP2009140222A - Power estimation method for lsi, and apparatus thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a power estimation method for estimating power consumption which corresponds to the operation of an LSI, in a short time, and to provide an apparatus. <P>SOLUTION: As an advance preparation, a function model is analyzed to extract function model parameters (a signal, a variation, or the like) affecting the power consumption, a net list model described in a net list is analyzed to extract net list parameters (a terminal name, a register name, or the like) corresponding to the function model parameters, and a correspondence relation between two sides is held. The net list model is operation-simulated to measure the power consumption, and a power model comprising a combination of the function model parameters affecting the power and the power consumption corresponding thereto is created. The function model is operation-simulated about the application program of a power estimation target, the power is extracted, in reference to the power model about a log of the function model parameters, when executing the application program, and the power consumption of the function model is estimated. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an LSI power estimation method and apparatus, and more particularly, to a method and apparatus that reduce power estimation time.

  In the process of designing an LSI, it is required to estimate power in advance. As an LSI power estimation method, there is a method of estimating power by performing a simulation at a transistor level. In this method, the number of ON / OFF times of all the transistors included in the LSI and the number of charge / discharge times of all the signal lines are counted, and the power consumption generated by ON / OFF of each individual transistor, Multiply the charge / discharge power by the charge / discharge capacity and estimate the power. In this method, it is necessary to simulate all of the enormous number of transistors and signal lines included in the LSI, and enormous calculation time is required for power estimation.

  In another method for estimating the LSI power, the power of the function macro (IP) included in the LSI is given in advance as a fixed value, and the power is estimated by integrating the IP power in the LSI. However, in this method, since a constant power value that does not depend on the operation corresponding to the LSI application is used, it is not possible to estimate the power difference due to the application program that instructs the function macro (IP) to operate.

  Various LSI power estimation methods have also been proposed in the following Patent Documents 1, 2, and 3. According to Patent Document 1, an LSI is divided into a clock circuit portion and other circuit portions, power consumption is estimated at the gate level or transistor level for the clock circuit portion, and power consumption is calculated at the RT level for other circuit portions. Estimate and add both.

  Also, according to Patent Document 2, the memory portion is estimated at the transistor level, and the average power consumption and peak power consumption of the LSI are calculated by reflecting the number of accesses of the memory portion when the entire LSI is simulated at the RT level. To do.

According to Patent Document 3, the combinational circuit is made into a block, the power consumption is calculated to make a library, and the power consumption of the entire chip is calculated together with the sequential circuit.
Japanese Patent Laid-Open No. 2004-287669 Japanese Patent Laid-Open No. 2003-256495 Japanese Patent Laid-Open No. 2002-15022

  Although there is a method for estimating the power consumption corresponding to the operation by performing a simulation at the transistor level, the conventional power estimation method requires a huge calculation time. Further, the power estimation method in a short time cannot estimate the power corresponding to the operation of the LSI.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a power estimation method and apparatus capable of estimating power consumption corresponding to LSI operation in a short time.

  In order to achieve the above object, according to the first aspect of the present invention, as a preliminary preparation, a function model described in a source code at a behavior level or a register transfer level is analyzed to affect power consumption. Functional model parameters (signals, variables, etc.) are extracted, netlist models (terminal names, register names, etc.) corresponding to the functional model parameters are analyzed by analyzing the netlist model generated from the functional model and described in the netlist Are extracted and the correspondence between them is maintained. Furthermore, as a preliminary preparation, operation of the netlist model is simulated to measure power consumption. Then, from the measurement result, a power model including a combination of function model parameters that affect power and the corresponding power consumption is created. The above is the preliminary preparation (preliminary modeling) process.

  Then, an operation simulation is performed on the function model for the application program targeted for power estimation, and a log of the function model parameters when the application program is executed is acquired. Then, by referring to the power model generated in the pre-modeling process, the power corresponding to the combination of the log function model parameters is extracted, and the power consumption of the function model in the operation of the power estimation target application is estimated. The above is the power estimation process of the function model in the power estimation target application.

  In the pre-modeling process, for the function model, an operation simulation is executed with a netlist model, and power consumption is obtained by an operation simulation at the transistor level or the register level. Although this process requires enormous calculation time, relatively accurate power consumption can be obtained. If the power model having the correspondence between the combination of the function model parameters obtained by the operation simulation and the power consumption is stored, the operation simulation is executed for the function model at the time of executing the application program to be estimated, From the obtained function model parameter log, the power value can be estimated with reference to the power model described above. The operation simulation with this functional model does not require a long calculation time as with the netlist model. Therefore, the power consumption of the function model depending on the operation of the application program can be estimated in a short time.

In order to achieve the above object, according to the second aspect of the present invention, a power estimation method for estimating power consumption of an LSI includes:
(A) Analyzing the function model described in the source code to extract function model parameters that affect power consumption, analyzing the netlist model described in the netlist corresponding to the function model, Extracting a netlist parameter corresponding to the functional model parameter, and maintaining a correspondence between the functional model parameter and the netlist parameter;
(B) measuring the value of the functional model parameter and the power consumption by simulating the operation of the netlist model;
(C) creating a power model composed of the relationship between the combination of the values of the function model parameters obtained in step (b) and the corresponding power consumption;
(D) Execute an operation simulation of the function model in the operation of the application program to be estimated, extract the power consumption corresponding to the combination of the values of the function model parameters at the time of execution with reference to the power model, and And estimating the power consumption of the functional model in the operation of the target application.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the technical scope of the present invention is not limited to these embodiments, but extends to the matters described in the claims and equivalents thereof.

  Since the power consumption of an LSI is generally related to the internal operation of the LSI, the power consumption is also determined when the internal operation is determined. On the other hand, internal operations are reflected in parameter values such as internal signal lines and register values. However, power consumption does not depend on all parameter values, but depends on specific parameter values that affect power consumption.

  For example, the power consumption of a macro (IP) having a DMA (direct memory access) function is determined depending on the presence / absence of a DMA operation and an internal parameter value at the time of reading or writing, for example, a burst size. Alternatively, in the case of a macro having an MPEG decoder function, the power is determined depending on internal parameter values such as the image size and image type.

  Therefore, in the present embodiment, the operation simulation of the netlist model is executed for the function model, and the relationship between the combination of the function model parameters and the power consumption is obtained in advance as the power model. Then, the operation of the function model when the estimation target application program is executed is simulated, and a log of the function model parameter value at that time is acquired. With respect to the combination of the function model parameters of this log, the corresponding power consumption is obtained with reference to the power model described above. Just by simulating the functional model for the application program to be estimated, the power consumption can be estimated with the accuracy of simulation of the netlist model.

  The above functional model is a behavior level description level that expresses the operation and behavior of the circuit, and is described in a language such as SystemC. Alternatively, the functional model may be a description level of a register transfer level (RTL), which is a level representing an operation between registers.

  FIG. 1 is a diagram showing an outline of a power estimation method in the present embodiment. This power estimation method includes pre-modeling (S1, S2, S3) as pre-preparation and power estimation (S4) for the estimation target application. The first step is step S1 for extracting all parameters of the model. Here, in the system 100 having the model 10 to be estimated as a peripheral macro, parameters of the model 10 are extracted (S1).

  Specifically, the description of the function model 10F described in the behavior level source code (for example, language SystemC) corresponding to the model 10 is analyzed, and the function model parameter 12F included in the function model 10F is extracted (S11). The description of the net list model 10N described in the net list corresponding to the model 10 is analyzed, and the net list parameter 12N included in the net list model 10N is extracted (S12). It is desirable that both parameters are associated one-to-one.

  Next, for the netlist model 10N, the simulation system 200 performs an operation simulation and measures power (S2). At this time, a log of parameter values and power consumption for them is acquired. Then, a power model 208 is generated that indicates the correlation between the combination of power model-related function model parameter values and power consumption (S3). At this time, it is desirable to narrow down the power model to function model parameters 210 related to (influencing) the power value. This completes the pre-modeling.

  Then, power estimation is performed by function model simulation when the system 300 executes the application program to be estimated (S4). Specifically, an operation simulation of the function model 10F when the system 300 executes the application program to be estimated is executed, and a log of the value of the function model parameter at that time is acquired (S41). From the log of function model parameter values (simulation results), the above-described power model is referred to, the power consumption corresponding to the same value combination is extracted, and the power of the function model 10F is estimated (S42).

  Hereinafter, each process will be described with reference to specific examples.

  FIG. 2 is a detailed flowchart of step S1. The process S1 for extracting all parameters of the model includes the process S11 for extracting the function model parameter 12F by analyzing the source code of the function model 10F as described above, and the process for extracting the netlist parameter 12N by analyzing the netlist model. S12.

  3 and 4 are diagrams for explaining a model having a DMA (direct memory access) function as an example of a function model. FIG. 3 shows the configuration of the system 110. This system includes a CPU, two memories RAMa and RAMb, a system bus BUS, and a function macro 10 having a DMA function. The DMA macro 10 includes a transfer unit 112 having a read unit Read, a write unit Write, and a transfer buffer Buf, a controller 114, and a register group 116. The register group 116 includes a source address register SrcAdr that stores a DMA transfer source address, a destination address register DstAdr that stores a DMA transfer destination address, a size register Size that stores a transfer size indicating the amount of DMA transfer data, It has a burst size register BurstSize that stores the number of transfers (burst size) for one DMA request, and an active register Active that stores a DMA transfer flag.

  In the DMA macro 10, for example, the CPU transfers a source address, a destination address, a destination address register DstAdr, a size register Size, and a burst size register BurstSize in the register group 116, respectively. After setting the transfer size and burst size, the CPU sets the active register to the active state “1” and asserts the DMA request. In response to the setting of “1” in the active register, the controller 114 of the DMA macro 10 causes the read unit Read to execute a read operation to the transfer source address, stores the read data in the transfer buffer Buf, The write unit Write is caused to execute the write operation to the transfer destination address. Then, the controller 114 updates the transfer source address, transfer destination address, and transfer size in the register. The controller 114 repeats the above read and write operations until the transfer size reaches “0”. When the transfer size reaches “0”, the inactive state is set to “0” in the active register and the DMA transfer is performed. Exit.

  FIG. 4 shows an example of a function model described in behavior level source code corresponding to the DMA macro. However, in FIG. 4, it is not described in the SystemC language or the RTL language, but is described in a more advanced pseudo code.

  The function model 10F (1) on the left side of FIG. 4 declares a model name “DMA”, a control register, and an external signal line. Further, the right-side function model 10F (2) describes the DMA start sequence behavior and the DMA operation behavior. The DMA start sequence and the DMA operation sequence described in the pseudo code match the DMA transfer operation sequence of the DMA macro 10 described above. However, the actual function model 10F is described by source code in a high-level language such as the SystemC language or the RTL language.

  The description of the source code includes a signal line that serves as an interface with the outside and a variable. Signal lines and variables included in the description of the source code are extracted as function model parameters by source code analysis (for example, syntax analysis) of the function model in step S11 of FIG.

  FIG. 5 is a diagram showing an example of function model parameters and net list parameters in the DMA function model. Analyzing the description in the source code of the function model 10F shown in FIG. 4, the signals “address”, “data”, “ReadEnable”, “WriteEnable” and variables “SrcAdr”, “DstAdr”, “size”, “burstsize”, “ "Active" is extracted as the function model parameter 12F.

  As shown in FIG. 2, the function model 10F is converted into a corresponding netlist 10N by a conversion tool 14 such as a logic synthesis tool. The conversion tool 14 converts the netlist parameter of the netlist to the same name as the function model parameter or to a corresponding but different name corresponding to the function model parameter of the function model. Therefore, the function model parameter 12F and the network parameter 12N have a one-to-one relationship.

  Therefore, the netlist parameter 12N is extracted by performing the netlist analysis S12 on the netlist model 10N. By referring to the parameter correspondence table created by the conversion tool 14 at the time of conversion, the function model parameters 12F and the netlist parameters 12N can be associated one-to-one.

  An example of the net list parameter 12N is shown on the right side of FIG. In this example, the netlist parameter 12N is extracted as “Taddress”, “Tdata”, “TReadEnable”, and “TWriteEnable” as terminal names, and “RSrcAdr”, “RDstAdr”, “Rsize”, “Rburstsize”, and “RActive” as register names, respectively. Yes. In this example, the terminal name and register name of the netlist parameter are changed to the signal name and variable name of the function model parameter with “T” and “R” added to the initials. Thus, as shown in FIG. 5, the function model parameter 12F and the netlist parameter 12N have a one-to-one correspondence.

  FIG. 6 is a diagram for explaining the power measurement step S2 by netlist simulation. Tools for measuring power by performing an operation simulation of a netlist are generally popular. Therefore, the simulation system 200, which is the power measurement tool, gives a plurality of input patterns to the netlist 10N and executes an operation simulation. The input pattern 202 is an input data pattern composed of a number of combinations of parameter values of the netlist parameter 12N, and an operation simulation is performed on the number of input parameters 202 (1) to 202 (x), and parameter values 206 at the time of execution are executed. And a log of the power value 204 are acquired.

  FIG. 7 is a diagram illustrating an example of parameter values and power values acquired by netlist simulation. FIG. 7 shows a change in the power value with respect to the horizontal axis time and a change in the parameter value corresponding to the function model parameter. In the netlist simulation, the power value at the sampling timing indicated by a plurality of vertical lines in FIG. 7 and the values of all parameters (values at the positions of the circles in the figure) are recorded as a log. As shown in the example of FIG. 7, it can be understood that the power value sequentially changes in response to the change of the parameter value. In particular, when the variable Active of the active register becomes “1”, it is recognized that the DMA macro executes DMA transfer and the power value increases. The power value fluctuates under the influence of other parameter values such as read enable and write enable.

  FIG. 8 is a diagram showing a log of power values corresponding to combinations of parameter values obtained by executing the network simulation of FIG. Since operation simulation is performed for a large number of input patterns, the amount of log data obtained is considerable. In FIG. 8, only a small part is shown as an example, and the power values for the values of nine parameters (shown by function model parameters) are shown.

  FIG. 9 is a specific flowchart of step S3 in FIG. In step S3, the function model parameters not related to the power value are deleted from the power and parameter value data 204 and 206 (log data in FIG. 8) obtained by the netlist simulation (S32), and the data amount of the log data is set. To reduce. Then, a correspondence table between the function model parameters related to the power value and the power value is created to generate the power model 208 (S34). That is, the power model 208 is a correspondence table between function model parameters related to power values and power values.

  FIG. 10 is a diagram showing pseudo code of the processing program in step S32 in FIG. First, the power values W (X1 to Xn) for the combinations of the function model parameters X1 to Xn and the function model parameters X1 to Xn are defined (P1). Then, the program P2 executes a subroutine SUB1 for checking whether or not the designated parameter is related to power for all parameters, and deletes the parameter if there is no relationship.

  In the subroutine SUB1, the average value of the power values when the values of the parameters Xk (Xk = X0 to Xn) are Nk0 to Nkn is obtained for all the measurement data in FIG. 8, and the variance of the average value of the power values is calculated. To do. If the variance of the average value of the power values is equal to or less than the specified value S, it is determined that the parameter Xk is a parameter that does not depend on power, that is, is not related to (does not influence) power. If the variance of the average power value exceeds the specified value S, the parameter is determined as a parameter on which power depends. This is because when the variance exceeds the specified value S, the degree of change in the power value in relation to the value of the parameter Xk is large.

  According to the processing program of FIG. 10, it is possible to detect a parameter that is not related to the power value, delete the parameter from the measurement data log, and reduce the data amount of the power model.

  FIG. 11 is another specific flowchart of step S3 of FIG. According to this, in step S3, the function model parameters not related to the power value are determined in two stages by two algorithms from the power and parameter value data 204 and 206 (log data of FIG. 8) obtained by the netlist simulation. It deletes (S32, S32-1, S32-2), and reduces the data amount of log data. The difference from FIG. 9 is that the parameters are reduced in two stages. Then, a correspondence table between the function model parameters related to the power value and the power value is created to generate the power model 208 (S34).

  FIG. 12 is a diagram showing a pseudo code of the processing program in the above steps S32-1 and S32-2 in FIG. As in FIG. 10, first, the power values W (X1 to Xn) for the combinations of the function model parameters X1 to Xn and the function model parameters X1 to Xn are defined (P1). Then, the program P2 executes a subroutine SUB2 for checking whether or not the designated parameter is related to power for all parameters, and deletes the parameter if there is no relationship.

  In the subroutine SUB2, the average value of the power value and the variance of the average value when the value of the parameter Xk is Nk0 to Nkn are obtained for all the measurement data in FIG. 8 (P3). If the variance of the power value when the value of the parameter Xk is Nk0 to Nkn is equal to or less than the specified value S, it is determined that the parameter value Xk = Nk0 to Nkn is not dependent on other parameters ( S4). This determination P4 is based on the algorithm AL1 of FIG. 11, and if it can be determined that other parameters do not affect the power value when a certain parameter Xk is a specific value Nkk, the number of parameter combinations from the power model is determined. Can be reduced.

  Further, as the algorithm AL2, when the variance of the average value of the power values is equal to or less than the specified value S, it is determined that the parameter Xk is a parameter on which power does not depend. If the variance of the average power value exceeds the specified value S, the parameter is determined as a parameter on which power depends (P5). This algorithm AL2 is the same as the subroutine SUB1 in FIG.

  FIG. 13 is a diagram illustrating an example of measurement data acquired by netlist simulation. In FIG. 13A, the horizontal axis is the burst size BurstSize, the vertical axis is the power value, and when the active active = 0, the power value when the burst size is changed to 1, 2, 4, 8, 16 is plotted. Has been. White circles are measured values, and black circles are average values. According to this, according to the determination processing by the algorithm AL1, when active Active = 0, there is almost no dispersion of the power value, so it can be determined that the burst size which is another parameter does not affect the power. That is, when active Active = 0, the burst size becomes “knot care (does not depend on)”.

  FIG. 13B also shows the burst size BurstSize on the horizontal axis and the power value on the vertical axis. However, the power value when the burst size is changed to 1, 2, 4, 8, and 16 when active Active = 1. Is plotted. According to this, when active Active = 1, a power value depending on the burst size is detected. According to this, it is determined by the determination processing by the algorithm AL1 that when active Active = 1, the dispersion of the power value is large and the burst size which is another parameter affects the power, and active Active = 1. In this case, the parameter called burst size cannot be reduced.

  In addition, since the active values of the active active are different depending on the case of “0” and “1”, the active active of the parameter is not deleted by the determination processing by the algorithm AL2.

  FIG. 14 is a diagram showing an example of a power model generated by the processing program in steps S32-1 and S32-2 as an example of measurement data in FIG. In the parameters of the power model 208, parameters that do not depend on the power value are deleted from the measurement data 204/206 in FIG. 8, and parameters that do not depend on the power value when a certain parameter is a specific value are also deleted.

  For example, M1 indicates that the power value does not depend on the values of the other three parameters when active Active = 0. M2 indicates that even when Active Active = 1, the read enable ReadEnable = 0 and the write enable WriteEnable = 0 do not depend on the burst size BurstSize. The other M3 to M8 are associated with power values corresponding to combinations of four parameter values.

  In order to determine the model M2, the power value when the burst size BurstSize is variable in the case of active Active = 1, read enable ReadEnable = 0, and write enable WriteEnable = 0 is not distributed. Need to be detected. Therefore, in the programs P3 and P4 in FIG. 12, it is necessary to calculate the variance of the average value of the power values for all the combinations that the parameter Xk subset can take.

  Therefore, in FIG. 12, the processing of the program P5 is first performed to delete some parameters, and thereafter, for the programs P3 and P4, the dispersion value of the power value when a single parameter takes a specific value, a plurality of parameters It is desirable to obtain a dispersion value of the power value when taking a combination of specific values and check the relationship between the dispersion value and the specified value S to compare the power value.

  As shown in FIG. 14, in the power model 208, the number of parameters is reduced more than the measurement data, and the number of parameter value combinations is also reduced.

  FIG. 15 is a diagram illustrating a specific flowchart of step S4. The functional model system 300 includes an IP macro (functional model) 10F that is a power measurement target in addition to the CPU and other IP (macro). The IP macro 10F is a functional model in which the DMA macro 10 in FIG. However, the functional model system 300 executes an estimation target application program 212 different from the functional model system 100 in FIG. The estimation target application program 212 is, for example, an MPEG decoding program. By referring to the power model 208, the power consumption of the DMA macro 10F that is an IP macro to be measured when the application program 212 is executed. Can be estimated.

  As shown in FIG. 15, the estimation target application program 212 and the function model parameters 210 that affect the power value among the function model parameters of the function model 10F corresponding to the target IP macro 10 are input to the function model simulator. , Functional model simulation S41 is executed. This simulation S41 uses existing function simulator tools such as a Verilog function simulator, which is an example of a description language, a SystemC function simulator, which is another example of a description language, and an RTL function simulator. Can be done. By executing the function model simulation S41, it is possible to log changes in the function model parameters when the estimation target application program 212 is executed for the function model 10F described in the source code.

  As a result, the value of the function model parameter for each fixed time interval is acquired as a simulation log LOG.

  FIG. 16 is a diagram illustrating an example of a log of function model parameters and an electric power value estimated from the log. FIG. 16A is a diagram showing the log LOG obtained in FIG. This log LOG has values for each sampling timing of the values of the four functional model parameters burstSize, Active, ReadEnable, and WriteEnable that remain in the power model. When an application program to be estimated, for example, an MPEG decoder processing program is executed, the CPU issues a DMA request as necessary, and the change of the function model parameter when the DMA macro 10 executes DMA in response to this is expressed as LOG To be acquired.

  Returning to FIG. 15, with respect to the log LOG acquired in the function model simulation S <b> 41, the power model 208 generated in advance is referred to, and the power value corresponding to the combination of the corresponding parameter values is extracted, and FIG. An estimated power value 214 can be generated.

  As described above, according to the present embodiment, as shown in FIG. 1, the power value is measured in the netlist simulation S2 for the IP macro that is the target of power estimation in the pre-modeling, and the power corresponding to the combination of the function macro parameters. A power model 208 having a value is generated. Then, the function macro parameter log of the function model 10F of the IP macro subject to power estimation when the application program targeted for power estimation is executed is acquired by function model simulation (S41), and the log is referred to the power model 208. The power value corresponding to the combination of the values of the function macro parameters included in can be extracted and the power can be estimated (S42). Therefore, it is possible to estimate a more accurate power value simply by executing a simulation of the power value in various operations for the functional model.

  The above embodiment is summarized as follows.

(Appendix 1) In a power estimation method for estimating the power consumption of an LSI,
(A) Analyzing the function model described in the source code to extract function model parameters that affect power consumption, analyzing the netlist model described in the netlist corresponding to the function model, Extracting a netlist parameter corresponding to the functional model parameter, and maintaining a correspondence between the functional model parameter and the netlist parameter;
(B) measuring the value of the functional model parameter and the power consumption by simulating the operation of the netlist model;
(C) creating a power model composed of the relationship between the combination of the values of the function model parameters obtained in step (b) and the corresponding power consumption;
(D) Execute an operation simulation of the function model in the operation of the application program to be estimated, extract the power consumption corresponding to the combination of the values of the function model parameters at the time of execution with reference to the power model, and A method for estimating power consumption of a functional model in operation of a target application.

(Appendix 2) In Appendix 1,
The functional model parameters include signals and variables included in the functional model description,
The power estimation method, wherein the netlist parameter includes a terminal name and a register name included in the description of the netlist.

(Appendix 3) In Appendix 1,
The step (c) is a power estimation method including a reduction step of reducing the number of combinations of function model parameters in the power model by deleting function model parameters that do not affect the power consumption.

(Appendix 4) In Appendix 3,
In the reduction step, when the variance of power consumption corresponding to different values of the function model parameter is equal to or less than a reference value, the power estimation method is to delete the function model parameter.

(Appendix 5) In Appendix 4,
In the reduction step, when the variance of the power consumption corresponding to the first value of the function model parameter is equal to or less than the first reference value, the average value of the power consumption corresponding to the first value of the function model parameter is A power estimation method in which power consumption when the function model parameter is a first value and other function model parameters are arbitrary values is used.

(Appendix 6) In Appendix 1,
In the operation simulation of the netlist model in the step (b), the operation simulation is executed for a plurality of types of input patterns, and the netlist parameter value at each sampling timing and the log of the power consumption at that time are recorded. Characteristic power estimation method.

(Appendix 7) In Appendix 1,
In the step (d), in the operation simulation of the function model, a log of the value of the function model parameter at each sampling timing is recorded.

(Supplementary note 8) In a power estimation device for estimating the power consumption of an LSI,
(A) Analyzing the function model described in the source code to extract function model parameters that affect power consumption, analyzing the netlist model described in the netlist corresponding to the function model, Means for extracting a netlist parameter corresponding to the functional model parameter and holding a correspondence relationship between the functional model parameter and the netlist parameter;
(B) means for simulating the operation of the netlist model to measure the value of the functional model parameter and power consumption;
(C) means for creating a power model composed of the relationship between the combination of the values of the function model parameters obtained in step (b) and the corresponding power consumption;
(D) Execute an operation simulation of the function model in the operation of the estimation target application, extract the power consumption corresponding to the combination of the value of the function model parameter when the function model is executed with reference to the power model, and A power estimation device having means for estimating power consumption of a functional model in operation of an application.

It is a figure which shows the outline of the electric power estimation method in this Embodiment. It is a detailed flowchart figure of process S1. It is a figure explaining the model which has the function of DMA (direct memory access) as an example of a functional model. It is a figure explaining the model which has the function of DMA (direct memory access) as an example of a functional model. It is a figure which shows the example of the function model parameter in a DMA function model, and a net list parameter. It is a figure explaining electric power measurement process S2 by a net list simulation. It is a figure which shows an example of the parameter value and electric power value which were acquired by the netlist simulation. It is a figure which shows the log of the electric power value corresponding to the combination of the parameter value obtained by performing the network simulation of FIG. It is a specific flowchart figure of process S3 of FIG. It is a figure which shows the pseudo code of the processing program of said process S32 of FIG. It is another specific flowchart figure of process S3 of FIG. It is a figure which shows the pseudo code of the processing program of said process S32-1, S32-2 of FIG. It is a figure which shows an example of the measurement data acquired by the net list simulation. It is a figure which shows the example of the electric power model produced | generated by the processing program of process S32-1, S32-2 in the measurement data example of FIG. It is a figure which shows the specific flowchart of process S4. It is a figure which shows an example of the log of a function model parameter, and the electric power value estimated from it.

Explanation of symbols

10F: Function model 10N: Net list 12N: Net list parameter 12F: Function model parameter 208: Power model

Claims (5)

In the power estimation method for estimating the power consumption of LSI,
(A) Analyzing the function model described in the source code to extract function model parameters that affect power consumption, analyzing the netlist model described in the netlist corresponding to the function model, Extracting a netlist parameter corresponding to the functional model parameter, and maintaining a correspondence between the functional model parameter and the netlist parameter;
(B) measuring the value of the functional model parameter and the power consumption by simulating the operation of the netlist model;
(C) creating a power model composed of the relationship between the combination of the values of the function model parameters obtained in step (b) and the corresponding power consumption;
(D) Execute an operation simulation of the function model in the operation of the application program to be estimated, extract the power consumption corresponding to the combination of the values of the function model parameters at the time of execution with reference to the power model, and A method for estimating power consumption of a functional model in operation of a target application. In claim 1,
The functional model parameters include signals and variables included in the functional model description,
The power estimation method, wherein the netlist parameter includes a terminal name and a register name included in the description of the netlist. In claim 1,
The step (c) is a power estimation method including a reduction step of reducing the number of combinations of function model parameters in the power model by deleting function model parameters that do not affect the power consumption. In claim 3,
In the reduction step, when the variance of power consumption corresponding to different values of the function model parameter is equal to or less than a reference value, the power estimation method is to delete the function model parameter. In a power estimation device that estimates the power consumption of an LSI,
(A) Analyzing the function model described in the source code to extract function model parameters that affect power consumption, analyzing the netlist model described in the netlist corresponding to the function model, Means for extracting a netlist parameter corresponding to the functional model parameter and holding a correspondence relationship between the functional model parameter and the netlist parameter;
(B) means for simulating the operation of the netlist model to measure the value of the functional model parameter and power consumption;
(C) means for creating a power model composed of the relationship between the combination of the values of the function model parameters obtained in step (b) and the corresponding power consumption;
(D) Execute an operation simulation of the function model in the operation of the estimation target application, extract the power consumption corresponding to the combination of the value of the function model parameter when the function model is executed with reference to the power model, and A power estimation device having means for estimating power consumption of a functional model in operation of an application.

Images