Performance parameter prediction method and device of test program and electronic equipment

Abstract

The embodiment of the invention discloses a performance parameter prediction method and device of a test program, electronic equipment and a storage medium, relates to the technical field of computers, and can effectively improve the speed and accuracy of performance parameter prediction of the test program. The method comprises the steps that platform specification parameters of a first platform are obtained to serve as a first parameter set, and the platform specification parameters comprise at least one of specification parameters of a processor in the platform and specification parameters of peripheral equipment connected with the processor in the platform; inputting the first parameter group into a performance prediction model, and obtaining an output result, the performance prediction model being a pre-trained artificial intelligence model; and according to the output result of the performance prediction model, predicting performance parameters of a test program running on the first platform. The method can be used for predicting the performance parameters of the electronic product test program.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for predicting performance parameters of a test program, electronic equipment, and a storage medium. Background Art

[0002] During the processor design process, designers usually need to run benchmark programs (or benchmark loads) on a simulation platform to evaluate the design. However, since processor simulation is very time-consuming, in order to obtain evaluation results within an acceptable time, only a small number of test programs can be used to replace a large number of real applications to test the processor, which greatly reduces the accuracy of the performance parameter prediction of the test program. In addition, during the post-silicon debugging of the processor, it is also necessary to run some benchmark programs on the hardware platform to further evaluate the processor. However, since the performance standards of most benchmark programs are derived from experience, it is impossible to accurately determine whether the performance parameters of the actual test program meet expectations. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides a method and device for predicting performance parameters of a test program, an electronic device, and a storage medium, which can effectively improve the speed and accuracy of predicting the performance parameters of the test program.

[0004] In a first aspect, an embodiment of the present invention provides a method for predicting performance parameters of a test program, comprising: obtaining platform specification parameters of a first platform as a first parameter group, wherein the platform specification parameters include at least one of the following: specification parameters of a processor in the platform, specification parameters of peripheral devices connected to the processor in the platform; inputting the first parameter group into a performance prediction model and obtaining an output result, wherein the performance prediction model is a pre-trained artificial intelligence model; and predicting the performance parameters of the test program running on the first platform based on the output result of the performance prediction model.

[0005] In one embodiment, the number of the test programs is at least one, and the types of the test programs are the same or different.

[0006] In one embodiment, inputting the first parameter group into a performance prediction model and obtaining an output result includes: constructing a first input vector using the first parameter group, the first input vector being compatible with an input layer of the performance prediction model; inputting the first input vector into the input layer of the performance prediction model, and obtaining a first output result from the output layer of the performance prediction model.

[0007] In one embodiment, each of the platform specification parameters in the first parameter group corresponds to a dimension of the first input vector, and the number of dimensions of the first input vector is greater than or equal to the number of the platform specification parameters in the first parameter group.

[0008] In one embodiment, after predicting the performance parameters of the test program running on the first platform based on the output result of the performance prediction model, the method further includes: adjusting the values ​​corresponding to a part of the dimensions in the first input vector to obtain at least one second input vector; inputting each of the second input vectors into the performance prediction model respectively to obtain a corresponding second output result; and determining the impact of the platform specification parameters corresponding to the adjusted dimensions on the performance parameters of the test program running on the first platform based on the difference between each of the second input vectors and the first input vector, and the difference between each of the second output results and the first output result.

[0009] In one embodiment, before inputting the first parameter group into the performance prediction model and obtaining the output result, the method also includes: obtaining operating status data of at least one second platform, wherein each second platform runs the test program, and the operating status data includes at least one platform specification parameter of the second platform and performance parameters of the test program running on the second platform; using the operating status data to perform model training to obtain the performance prediction model.

[0010] In one embodiment, the use of the operating status data to perform model training to obtain the performance prediction model includes: generating a first data set based on the operating status data, the first data set including at least one specification-performance data pair, each of the specification-performance data pair including a second parameter group formed by one of the platform specification parameters of the second platform, and performance parameters of the test program running on the second platform corresponding to the second parameter group; performing model training based on the first data set to obtain the performance prediction model.

[0011] In one embodiment, the model training based on the first data set to obtain the performance prediction model includes: using principal component analysis to reduce the dimension of the first data set to obtain a second data set; and using the second data set obtained after the dimensionality reduction of the first data set to perform model training to obtain the performance prediction model.

[0012] In one embodiment, the dimensionality reduction processing of the first data set using the principal component analysis method includes: taking the same platform specification parameter in each of the second parameter groups as a statistical item, calculating the variance contribution rate of each statistical item, wherein the variance contribution rate is equal to the ratio of the variance of each statistical item to the sum of the variances of all statistical items; determining the cumulative variance contribution rate based on the variance contribution rate of each statistical item, wherein the cumulative variance contribution rate is equal to the sum of the variance contribution rates of the first k statistical items after arranging the variance contribution rates of the statistical items in order from large to small, wherein k is a positive integer; determining the principal component item from each of the statistical items based on the relationship between the cumulative variance contribution rate and a preset threshold, and the platform specification parameter corresponding to the principal component item is the principal component parameter; using the principal component parameter in each of the specification-performance data pairs and the performance parameters of the test program in the specification-performance data pair running on the second platform, generating a dimensionality reduction data pair, and the dimensionality reduction data pairs corresponding to all the specification-performance data pairs form the second data set.

[0013] In one implementation, the second platform includes at least one of the following: a simulation platform, a hardware platform.

[0014] In one embodiment, the platform specification parameter includes at least one of the following: core frequency, number of cores, bus frequency, bus bandwidth, memory capacity, number of memory channels, and memory speed.

[0015] In one embodiment, the performance prediction model includes at least one of the following: a long short-term memory network model, a support vector regression model, a random forest model, and a linear regression model.

[0016] In a second aspect, an embodiment of the present invention also provides a performance parameter prediction device for a test program, comprising: an acquisition unit, used to acquire platform specification parameters of a first platform as a first parameter group, wherein the platform specification parameters include at least one of the following: specification parameters of a processor in the platform, specification parameters of peripheral devices connected to the processor in the platform; an input unit, used to input the first parameter group into a performance prediction model and obtain an output result, wherein the performance prediction model is a pre-trained artificial intelligence model; a prediction unit, used to predict the performance parameters of the test program running on the first platform based on the output result of the performance prediction model.

[0017] In one embodiment, the number of the test programs is at least one, and the types of the test programs are the same or different.

[0018] In one embodiment, the input unit is specifically used to: construct a first input vector using the first parameter group, and the first input vector is compatible with the input layer of the performance prediction model; input the first input vector into the input layer of the performance prediction model, and obtain a first output result from the output layer of the performance prediction model.

[0019] In one embodiment, each of the specification parameters is a dimension of the first input vector, and the number of dimensions of the first input vector is greater than or equal to the number of the platform specification parameters in the first parameter group.

[0020] In one embodiment, it also includes: an adjustment unit, which is used to adjust the values ​​corresponding to a part of the dimensions in the first input vector after predicting the performance parameters of the test program running on the first platform according to the output result of the performance prediction model, so as to obtain at least one second input vector; the input unit is also used to input each of the second input vectors into the performance prediction model to obtain the corresponding second output result; the device also includes a determination unit, which is used to determine the impact of the platform specification parameters corresponding to the adjusted dimensions on the performance parameters of the test program running on the first platform based on the difference between each of the second input vectors and the first input vector, and the difference between each of the second output results and the first output result.

[0021] In one embodiment, the acquisition unit is also used to acquire the operating status data of at least one second platform before inputting the first parameter group into the performance prediction model and obtaining the output result, wherein the test program is running in each of the second platforms respectively; the operating status data includes at least one platform specification parameter of the second platform and the performance parameters of the test program running in the second platform; the device also includes a model training unit, which is used to perform model training using the operating status data to obtain the performance prediction model.

[0022] In one embodiment, the model training unit includes: a generation module for generating a first data set based on the operating status data, the first data set including at least one specification-performance data pair, each of the specification-performance data pair including a second parameter group formed by one of the platform specification parameters of the second platform, and performance parameters of the test program running on the second platform corresponding to the second parameter group; a training module for performing model training based on the first data set to obtain the performance prediction model.

[0023] In one embodiment, the training module includes: a dimensionality reduction submodule, which is used to use principal component analysis to reduce the dimensionality of the first data set to obtain a second data set; a training submodule, which is specifically used to perform model training on the second data set obtained after dimensionality reduction of the first data set to obtain the performance prediction model.

[0024] In one embodiment, the dimension reduction submodule is specifically used to: take the same platform specification parameter in each of the second parameter groups as a statistical item, calculate the variance contribution rate of each statistical item, wherein the variance contribution rate is equal to the ratio of the variance of each statistical item to the sum of the variances of all statistical items; determine the cumulative variance contribution rate according to the variance contribution rate of each statistical item, wherein the cumulative variance contribution rate is equal to the sum of the variance contribution rates of the first k statistical items after arranging the variance contribution rates of the statistical items in order from large to small, wherein k is a positive integer; determine the principal component item from each of the statistical items according to the relationship between the cumulative variance contribution rate and a preset threshold, and the platform specification parameter corresponding to the principal component item is the principal component parameter; use the principal component parameter in each of the specification-performance data pairs and the performance parameters of the test program in the specification-performance data pair running on the second platform to generate a dimension reduction data pair, and the dimension reduction data pairs corresponding to all the specification-performance data pairs form the second data set.

[0025] In one implementation, the second platform includes at least one of the following: a simulation platform, a hardware platform.

[0026] In one embodiment, the platform specification parameter includes at least one of the following: core frequency, number of cores, bus frequency, bus bandwidth, memory capacity, number of memory channels, and memory speed.

[0027] In one embodiment, the performance prediction model includes at least one of the following: a long short-term memory network model, a support vector regression model, a random forest model, and a linear regression model.

[0028] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising: a memory and a processor, wherein the memory is used to store computer-readable instructions; the processor is used to read the computer-readable instructions and implement the performance parameter prediction method of the test program provided in any embodiment of the present invention.

[0029] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the performance parameter prediction method of the test program provided by any embodiment of the present invention is implemented.

[0030] The performance parameter prediction method and device, electronic device, and storage medium of the test program provided in the embodiment of the present invention can obtain the platform specification parameters of the first platform as the first parameter group input into the performance prediction model, and predict the performance parameters of the test program running on the first platform according to the output result of the performance prediction model. Since the first parameter group can include the specification parameters of the processor in the platform and / or the specification parameters of the peripheral devices connected to the processor, the various operating environments of the test program can be conveniently and flexibly defined through the first parameter group. After the specification parameters are input into the performance prediction model, the performance of the test program in the corresponding operating environment can be quickly and accurately predicted with the help of the performance prediction model, without the need for large-scale simulation and without relying on the subjective experience of engineers. Therefore, the speed and accuracy of the performance parameter prediction of the test program can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 A flow chart of a method for predicting performance parameters of a test program provided by an embodiment of the present invention; Figure 2 A flow chart of the performance parameter impact analysis of the test program in the embodiment of the present invention; Figure 3 A logic flow chart of the performance parameter impact analysis of the test program in the embodiment of the present invention; Figure 4 LSTM prediction model framework diagram provided for an embodiment of the present invention; Figure 5 A detailed flow chart of a method for predicting performance parameters of a test program provided by an embodiment of the present invention; Figure 6 A schematic diagram of a structure of a performance parameter prediction device for a test program provided by an embodiment of the present invention; Figure 7 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0034] It should be clear that the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] In a first aspect, an embodiment of the present invention provides a method for predicting performance parameters of a test program, which can effectively improve the speed and accuracy of predicting the performance parameters of the test program.

[0036] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting performance parameters of a test program, comprising: S11, obtaining platform specification parameters of a first platform as a first parameter group, wherein the platform specification parameters include at least one of the following: specification parameters of a processor in the platform, and specification parameters of a peripheral device in the platform connected to the processor; In this step, one or more platform specification parameters of the first platform may be obtained, and the platform specification parameters together form a first parameter group. The first platform may be a platform that has been built, or a conceptual platform that has not been actually built but is formed by various design parameters. The first parameter group may refer to a parameter group formed by platform specification parameters corresponding to the first platform. In an embodiment of the present invention, the platform specification parameters may include specification parameters of a processor in the platform and / or specification parameters of peripheral devices connected to the processor in the platform.

[0037] For example, in an embodiment of the present invention, the specification parameters of the processor in the platform may include, for example, core frequency, number of cores, bus frequency, bus bandwidth, etc. The specification parameters of the peripheral devices connected to the processor in the platform may include, for example, memory capacity, number of memory channels, memory speed, etc. For example, the core frequency is 3.0Ghz and the number of cores is 8.

[0038] Exemplarily, in one embodiment of the present invention, the platform specification parameters of the first platform obtained may include: the core frequency is 3.0Ghz, the number of cores is 32, the data bus clock is 1200M, the interconnect bus clock is 800M, the interconnect bus width is X16, the first-level cache capacity is 32kb, and the second-level cache capacity is 512kb.

[0039] It should be noted that for different platforms, the processors provided in the platforms may be the same or different, and the peripheral devices connected to the processors may also be the same or different, and therefore, the platform specification parameters corresponding to different platforms may also be the same or different. The embodiments of the present invention do not limit this.

[0040] S12, inputting the first parameter group into a performance prediction model and obtaining an output result, wherein the performance prediction model is a pre-trained artificial intelligence model; In an embodiment of the present invention, the performance prediction model may be a pre-trained artificial intelligence model. The performance prediction model may be obtained by machine learning, for example, a support vector regression model, a random forest model, an LSTM (Long Short-Term Memory) model, etc.

[0041] S13: predicting the performance parameters of the test program running on the first platform according to the output result of the performance prediction model.

[0042] In an embodiment of the present invention, a test program may refer to a program that can test the performance of one or more aspects of the platform on which it is located, and the performance of different aspects of the same platform can be tested by different test programs. In some embodiments, the test program may include, for example, a benchmark load. The benchmark load is a test program that can be used as a benchmark. In one example, the benchmark load can be used to evaluate the processor computing performance, such as for evaluating the computing power of integers and floating points. In another example, the benchmark load can also be used to evaluate the processor IO performance, such as for evaluating the performance of storage devices and network card devices connected to the same processor. In another example, the benchmark load can also be used to evaluate the processor performance under certain specific application scenarios, such as for evaluating the processor performance under application scenarios such as databases, java services, video encoding and decoding, scientific computing, website processing, artificial intelligence, etc. Different test programs have different specific test functions or application scenarios, and the degree of influence of each platform specification parameter on the performance parameters of the test program is also different.

[0043] The output result of the performance prediction model is the performance parameters predicted by the performance prediction model when the test program is running on the first platform, where the first platform is the platform defined by the first parameter group. For example, in one example, the test program is a benchmark load Linpack (Linear system package), and the performance prediction model can be used to predict the computing performance parameters of the test program when it is running on the first platform.

[0044] Specifically, in this embodiment, the first parameter group may include: core frequency = 3.0Ghz, number of cores = 32, data bus clock = 1200M, interconnect bus clock = 800M, interconnect bus width = X16, level 1 cache capacity = 32kb, level 2 cache capacity = 512kb, level 3 cache capacity = 64M, data bus bandwidth = 50GB / s, pipeline concurrency = 4, floating point register bit width = 64bit, number of memory channels = 12, memory rate = 3200Mhz, vector register bit width = 128bit. The first parameter group is input into a pre-trained performance prediction model to predict the floating point performance of the Linpack benchmark load. According to the model output, the floating point performance parameter shown when the test program is running is 2000 FLOPS (floating point operations per second).

[0045] The performance parameter prediction method of the test program provided by the embodiment of the present invention can obtain the platform specification parameters of the first platform as the first parameter group, input the first parameter group into the performance prediction model, and obtain the output result, and predict the performance parameters of the test program running on the first platform according to the output result of the performance prediction model. Since the first parameter group can include the specification parameters of the processor in the platform and / or the specification parameters of the peripheral devices connected to the processor, the various operating environments of the test program can be conveniently and flexibly defined through the first parameter group. After the first parameter group is input into the performance prediction model, the performance of the test program in the corresponding operating environment can be quickly and accurately predicted with the help of the performance prediction model, without the need for large-scale simulation and without relying on the subjective experience of engineers. Therefore, the speed and accuracy of the performance parameter prediction of the test program can be effectively improved.

[0046] In addition, since the performance parameter prediction model is used to predict the performance parameters of the test program, it is not necessary for the test program to run on the actual platform. That is, the performance parameter prediction method for the test program provided by the embodiment of the present invention can be independent of the processor's operating platform. Therefore, it can effectively improve the flexibility of the performance parameter evaluation of the test program during the processor development and debugging process, and can also greatly reduce the time consumption of the performance parameter prediction of the test program on the hardware platform and the simulation platform.

[0047] Specifically, in one embodiment of the present invention, in step S11, platform specification parameters of the first platform may be obtained as the first parameter group. The first parameter group is the platform specification parameters corresponding to the first platform. The platform specification parameters may include specification parameters of a processor in the platform and / or specification parameters of a peripheral device connected to the processor in the platform.

[0048] After obtaining the first parameter group, the first parameter group can be input into the performance prediction model in step S12. The performance prediction model can be an artificial intelligence model implemented according to various algorithms. The performance prediction model can have an input layer, a hidden layer, an output layer, etc. The input layer is used to receive input data, the hidden layer is used to process data, and the output layer is used to output the result.

[0049] When each first parameter group is input into the performance prediction model, a vector can be constructed using the first parameter group according to the structure of the input layer, and the constructed vector can be input into the performance prediction model to obtain a model output result. Specifically, in one embodiment of the present invention, step S12 of inputting the first parameter group into the performance prediction model and obtaining an output result can include: constructing a first input vector using the first parameter group, the first input vector being compatible with the input layer of the performance prediction model; inputting the first input vector into the input layer of the performance prediction model, and obtaining a first output result from the output layer of the performance prediction model.

[0050] When constructing the first input vector using the first parameter group, each of the platform specification parameters in the first parameter group may correspond to a dimension of the first input vector. The number of dimensions of the first input vector may be greater than or equal to the number of platform specification parameters in the first parameter group. That is, in an embodiment of the present invention, each platform specification parameter in the first parameter group may be mapped to a dimension of the first input vector, but each dimension in the first input vector may not necessarily have a corresponding platform specification parameter.

[0051] Specifically, in an embodiment of the present invention, the first input vector is adapted to the input layer of the performance prediction model, and the adaptation here may refer to the number of dimensions of the first input vector and the attributes of each dimension being the same as the number of dimensions of the input variables of the input layer of the performance prediction model and the attributes of each dimension. When the first input vector is constructed using the first parameter group, if the number and attributes of the platform specification parameters contained in the first parameter group are the same as the number of dimensions and attributes of the input variables of the input layer, then each platform specification parameter in the first parameter group can be directly mapped to each corresponding component of the first input vector. If the number of platform specification parameters contained in the first parameter group is less than the number of dimensions of the input variables of the input layer, then each platform specification parameter contained in the first parameter group can be mapped to the corresponding component of the first input vector, and the components that are not mapped in the first input vector are assigned default values. In this way, in some cases, when predicting the performance parameters of some test programs, it is not necessary to obtain all platform specification parameters (for example, it is not necessary to obtain platform specification parameters that have little effect on the performance parameters of the test program), thereby effectively improving the convenience of predicting the performance parameters of the test program. For example, in one embodiment of the present invention, the first parameter group includes three platform specification parameters X1, X2, and X3, and the first input vector is (a1, a2, a3, a4, a5). Then, X1, X2, and X3 can be mapped to a1, a2, and a3, respectively, and a4 and a5 can be assigned default values, respectively.

[0052] By inputting the first parameter group into the performance prediction model, a corresponding output result can be obtained. Based on this, in step S13, the performance parameters of the test program running on the first platform can be predicted according to the output result. Among them, the number of test programs can be at least one, and the types of each of the test programs can be the same or different. For example, in one example, the test program may include test program A and test program B, wherein test program A is used to test the floating-point computing capability of the processor, and test program B is used to test the network speed. The output result of the performance prediction model may include, for example, 200,000 floating-point operations, a network speed of 120G bytes / second, etc.

[0053] The above embodiment describes in detail how to predict the performance parameters of the test program, but the embodiments of the present invention are not limited to this. In another embodiment of the present invention, after predicting the performance parameters of the test program running on the first platform based on the output results of the performance prediction model, it is also possible to analyze the impact of one or more platform specification parameters on the performance parameters of the test program running on the first platform.

[0054] Since the platform specification parameters that affect the performance parameters of the test program are numerous and their relationships are complex, the impact of a certain platform specification parameter on the performance parameters of the test program is difficult to quantify through manual calculation. The performance parameter prediction method of the test program provided in the embodiment of the present invention can analyze the degree of impact of changes in one or some platform specification parameters on the performance parameters of the test program through a performance prediction model to perform quantitative analysis.

[0055] In a specific implementation, the influence of the platform specification parameters (or multiple platform specification parameters) on the performance parameters of the test program can be determined based on the changes in the performance parameters of the test program with the platform specification parameters. For example, in one embodiment, if the performance parameters of the test program do not change with the changes in the platform specification parameters, it means that the current changes in the platform specification parameters have no effect (or a small effect) on the performance parameters of the test program or the current changes in the platform specification parameters are not the current performance bottleneck of the test program. If the performance parameters of the test program change, it means that the current changes in the platform specification parameters have a greater impact on the performance parameters of the test program, and the performance parameters of the test program before and after the change can be recorded.

[0056] It should be noted that in the process of analyzing the impact of platform specification parameters on the performance parameters of the test program, one platform specification parameter can be adjusted to different degrees and the corresponding output results of the performance prediction model can be recorded, or multiple platform specification parameters can be adjusted to different degrees and the corresponding output results of the performance prediction model can be recorded. The embodiments of the present invention are not limited to this.

[0057] It should also be noted that in the embodiments of the present invention, different platform specification parameters or combinations of different platform specification parameters have different effects on the performance parameters of the test program. These effects can be linear or nonlinear. In one embodiment of the present invention, the relationship between the platform specification parameters (or parameter combinations) and the performance parameters of the test program can be plotted into a curve, so that the analysis of the impact of the platform specification parameters on the performance parameters of the test program is more intuitive.

[0058] like Figure 2 As shown, in one embodiment of the present invention, when the first parameter group is used to construct the first input vector, the adjustment of a certain platform specification parameter in the first parameter group can be converted into the adjustment of the value of a certain dimension of the first input vector. Exemplarily, the logic flow of the analysis of the impact of the platform specification parameter on the performance parameters of the test program can be as follows: Figure 3 Specifically, after predicting the performance parameters of the test program running on the first platform according to the output result of the performance prediction model, the performance prediction method of the test program provided by the embodiment of the present invention may further include: S21, adjusting values ​​corresponding to a portion of dimensions in the first input vector to obtain at least one second input vector; In this step, the platform specification parameters to be analyzed can be determined, and the components corresponding to the platform specification parameters in the first input vector can be adjusted. For example, to analyze the impact of the core frequency on the performance parameters of the current test program under the current test program, the component corresponding to the core frequency in the first input vector can be adjusted from 2.4 GHz to 3.0 GHz, and the values ​​of other parameters remain unchanged, to obtain a second input vector, and the component corresponding to the core frequency in the first input vector is adjusted from 2.4G to 2.0G, and the values ​​of other parameters remain unchanged, to obtain another second input vector. For another example, to analyze the impact of the core frequency and bus bandwidth on the performance parameters of the current test program under the current test program, the core frequency can be adjusted from 2.4 GHz to 3.0 GHz, and the bus bandwidth can be adjusted from 25GB / s to 50GB / s in the first input vector, and the values ​​of other parameters remain unchanged, to obtain a second input vector.

[0059] S22, inputting each of the second input vectors into the performance prediction model respectively to obtain a corresponding second output result; In this step, each of the second input vectors may be input into the performance prediction model respectively, and the performance prediction model may return a second output result for each of the second input vectors.

[0060] S23, based on the difference between each of the second input vectors and the first input vector, and the difference between each of the second output results and the first output result, determine the impact of the platform specification parameters corresponding to the adjusted dimension on the performance parameters of the test program running on the first platform.

[0061] Since the values ​​of some dimensions between each second input vector and the first input vector are the same, while the values ​​of other dimensions have been adjusted to different degrees, the output results of the performance prediction model may also have different degrees of difference. When the degree of adjustment of the dimension values ​​is constant, the greater the difference between the first output result and the second output result, the greater the impact of the current change of the dimension on the output result, and the smaller the difference between the first output result and the second output result, the smaller the impact of the current change of the dimension on the output result.

[0062] For example, in one embodiment of the present invention, to evaluate the impact of the data bus clock on the performance parameters of the test program, the value of the dimension corresponding to the data bus clock in the first input vector can be adjusted, and the other platform specification parameters in the first input vector except the data bus clock remain unchanged to obtain a second input vector. The second input vector is input into the performance prediction model to predict the performance parameters of the test program, and then the change in the performance parameters of the test program before and after the adjustment of the data bus clock is known. If the performance parameters have not changed, it means that the current range of the data bus clock does not affect the performance parameters of the test program or is not the bottleneck of the processor; if the performance has changed, it means that the current range of the data bus clock has a greater impact on the performance parameters of the test program, and the actual impact of adjusting the data bus clock on the performance parameters can be further analyzed.

[0063] The above embodiment describes in detail the use of a performance prediction model to predict the performance parameters of a test program. In order to obtain a performance prediction model with high prediction accuracy, in one embodiment of the present invention, before inputting the first parameter group into the performance prediction model in step S12, the performance parameter prediction method of the test program provided by the embodiment of the present invention may further include: Obtaining the operating status data of at least one second platform, wherein each second platform runs the test program respectively; the operating status data includes at least one platform specification parameter of the second platform and the performance parameter of the test program running on the second platform; using the operating status data to perform model training to obtain the performance prediction model.

[0064] That is to say, in an embodiment of the present invention, before using a performance prediction model to predict the performance parameters of a test program running on a first platform, running status data of the test program running on at least one second platform can be collected, and model training can be performed based on these running status data to obtain a performance prediction model. The second platform can include at least one of the following: a simulation platform, a hardware platform. For example, in the processor design stage, the simulated running status data of the test program on the simulation platform can be collected; in the post-silicon debugging stage, the real running status data of the test program when it is running on the actual processor hardware platform can be collected.

[0065] In an embodiment of the present invention, the running status data may include various status data of the hardware platform when the test program is running in the hardware platform. For example, in one example, the running status data may include various platform specification parameters in the second platform, and may also include performance parameters of the test program running on the second platform. Here, the types of platform specification parameters of the second platform may be the same as the types of platform specification parameters of the first platform, corresponding one to one, or the types of platform specification parameters of the second platform may be more than the types of platform specification parameters of the first platform, that is, the types of specification parameters of the first platform may be a subset of the types of specification parameters of the second platform.

[0066] Based on this, in one embodiment of the present invention, using the operating status data to perform model training to obtain the performance prediction model may specifically include: generating a first data set based on the operating status data, the first data set including at least one specification-performance data pair, each of the specification-performance data pairs including a second parameter group formed by one of the platform specification parameters of the second platform, and performance parameters of the test program running on the second platform corresponding to the second parameter group; performing model training based on the first data set to obtain the performance prediction model.

[0067] In this embodiment, after obtaining the running status data of the second platform, the platform specification parameters of the same second platform can be combined together to form a second parameter group, and the second parameter group and the performance parameters of the test program running on the second platform corresponding to the second parameter group are combined together to form a specification-performance data pair, and all specification-performance data pairs together form the first data set. Exemplarily, the platform specification parameters included in the second parameter group can be as shown in Table 1.

[0068]

[0069] During the development of the processor, various platform specification parameters can be debugged repeatedly. In different second parameter groups, the platform specification parameters are different. Each second parameter group and the performance parameters of the corresponding test program running on the second platform can form a specification-performance data pair. Exemplarily, a specification-performance data pair can be: (X--Y), where X can be expressed as (X1, X2, X3...X n ). Here, X is the second parameter group, (X1, X2, X3...X n) are the specification parameters of each platform in the second parameter group, such as core frequency, number of cores, data bus clock, interconnect bus clock, interconnect bus width, first-level cache capacity, second-level cache capacity, third-level cache capacity, data bus bandwidth, pipeline concurrency, floating-point register bit width, number of memory channels, memory rate, vector register bit width, instruction parallelism, thermal power consumption, etc. Y is the performance parameter of the test program corresponding to the second parameter group running on the second platform, for example, 200,000 floating-point calculations.

[0070] When using the first data set for model training, the first data set can be divided into a training data set and a test data set; for example, 80% of the data is used as a training set and 20% of the data is used as a test set. The training data set is used to train the LSTM model; the test data set is used to test and evaluate the trained LSTM model to obtain the performance prediction model.

[0071] For example, the LSTM prediction model framework can be as follows Figure 4 As shown in the figure, it includes: normalizing multi-dimensional feature variables; setting the number of samples used in each iteration of the model by configuring batch_size; building an LSTM model, including the input dimension, hidden layer dimension, number of layers, optimizer, and loss function settings; mapping the hidden state of the LSTM to the final output through the fully connected layer; and finally using the test data set to evaluate the performance and prediction accuracy of the model, etc., which will not be discussed in detail here.

[0072] In an embodiment of the present invention, a LSTM neural network model is trained using a data set, which can more effectively learn data hidden features from the historical data of performance parameters of numerous test programs, discover the hidden rules of the impact of platform specification parameters such as processor specification parameters and specification parameters of peripheral devices connected to the processor in the platform on the performance parameters of the test program, improve the accuracy of the prediction method, and achieve the purpose of quickly and accurately predicting the performance parameters of the test program.

[0073] In the above embodiment, the performance prediction model is obtained by using the first data set for model training, but the embodiments of the present invention are not limited thereto. In other embodiments of the present invention, the first data set may be processed first, and the processed first data set may be used for model training.

[0074] Specifically, in another embodiment of the present invention, performing model training based on the first data set to obtain the performance prediction model may include: performing dimensionality reduction processing on the first data set using the principal component analysis method to obtain a second data set; performing model training on the second data set obtained after the dimensionality reduction processing of the first data set to obtain the performance prediction model. In this way, the associated specification parameters in the first data set that have little or no effect on the performance parameters of the test program can be removed, which can reduce the amount of calculation on the one hand, and avoid model overfitting caused by too many irrelevant features on the other hand. For example, in one embodiment of the present invention, when predicting the floating-point computing performance of the processor, the floating-point computing performance has a greater correlation with the core-related parameters and the memory parameters, and has a smaller correlation with the IO-related parameters, such as the PCIe (peripheral component interconnect express, peripheral component interconnect express) bus bandwidth and other parameters. Therefore, the IO-related parameters can be removed by dimensionality reduction processing.

[0075] In a specific implementation, performing dimensionality reduction processing on the first data set using the principal component analysis method may include: Taking the same platform specification parameter in each of the second parameter groups as a statistical item, calculating the variance contribution rate of each statistical item, wherein the variance contribution rate is equal to the ratio of the variance of each statistical item to the sum of the variances of all statistical items; Determine the cumulative variance contribution rate according to the variance contribution rate of each statistical item, wherein the cumulative variance contribution rate is equal to the sum of the variance contribution rates of the first k statistical items after arranging the variance contribution rates of the statistical items in descending order, wherein k is a positive integer; According to the relationship between the cumulative variance contribution rate and the preset threshold, the principal component item is determined from each of the statistical items, and the platform specification parameters corresponding to the principal component item are the principal component parameters; illustratively, the value range of the preset threshold can be 70%-95%, for example, it can be 70%, 85%, 90%, 95%, etc.

[0076] A dimensionality reduction data pair is generated using the principal component parameters in each specification-performance data pair and the performance parameters of the test program in the specification-performance data pair running on the second platform, and the dimensionality reduction data pairs corresponding to all the specification-performance data pairs form the second data set.

[0077] For example, in one embodiment of the present invention, the second parameter group includes core frequency, core number, data bus clock, interconnect bus clock, and interconnect bus width, wherein the variance contribution rate of the core frequency is 10%, the variance contribution rate of the core number is 8%, the variance contribution rate of the data bus clock is 35%, the variance contribution rate of the interconnect bus clock is 27%, and the variance contribution rate of the interconnect bus width is 20%. The variance contribution rates are sorted from large to small: 35%, 27%, 20%, 10%, 8%, and the preset threshold is 80%. After calculation, it is found that the sum of the first three variance contribution rates is 35%+27%+20%=82%, which is greater than the preset threshold of 80%, and the sum of the first two variance contribution rates is 35%+27%=62%, which is less than the preset threshold of 80%. Therefore, it can be determined that the platform specification parameters corresponding to the first three variances are the principal component parameters in the second parameter group, that is, the principal component parameters are: data bus clock, interconnect bus clock, and interconnect bus width.

[0078] In the embodiment of the present invention, the principal component analysis method is used to extract the main components of the performance parameters of the test program in the processor parameters, and the original n-dimensional feature can be reduced to a k-dimensional principal component feature variable, where n is a positive integer, k is a positive integer, and k≤n. In this way, the principal component analysis method is used to reduce the dimension of the data set, extract the main parameters, and remove irrelevant or redundant parameters to improve the convergence and accuracy of the model, thereby improving the accuracy of the performance prediction model.

[0079] The performance parameter prediction method of the test program provided by the embodiment of the present invention can construct a first data set from the measured data of the test program running on the processor platform and the simulation data of the simulation platform, and use the data set or the second data set after dimensionality reduction processing to train machine learning models such as LSTM, so as to more effectively learn the hidden features of the data, discover the hidden rules of the influence of the internal parameters of the processor and the interconnection hardware parameters on the performance parameters of the test program, and achieve fast and accurate prediction of the performance parameters of the test program without predicting the processor hardware platform and the simulation platform.

[0080] The performance parameter prediction method of the test program provided by the embodiment of the present invention is described in detail below through a specific embodiment.

[0081] like Figure 5 As shown, the performance parameter prediction method of the test program provided by the embodiment of the present invention may include: S401, acquiring operation status data of at least one second platform and forming a first data set according to the operation status data; The first data set includes at least one specification-performance data pair, each of which includes a second parameter group formed by one of the platform specification parameters of the second platform, and a performance parameter of the test program running on the second platform corresponding to the second parameter group; Optionally, the second platform may include a simulation platform and / or a hardware platform.

[0082] S402, performing dimensionality reduction processing on the first data set using a principal component analysis method to obtain a second data set; S403, performing model training using the second data set obtained after dimensionality reduction processing of the first data set to obtain a performance prediction model; S404, obtaining platform specification parameters of the first platform as a first parameter group; S405: construct a first input vector using the first parameter group, where the first input vector is compatible with an input layer of the performance prediction model; S406, inputting the first input vector into the input layer of the performance prediction model, and obtaining a first output result from the output layer of the performance prediction model; S407: Adjust the values ​​of a portion of dimensions in the first input vector to obtain at least one second input vector; S408, inputting each of the second input vectors into the performance prediction model respectively to obtain a corresponding second output result; S409. Determine, based on the difference between each of the second input vectors and the first input vector, and the difference between each of the second output results and the first output result, the impact of the platform specification parameters corresponding to the adjusted dimension on the performance parameters of the test program running on the first platform.

[0083] In a second aspect, an embodiment of the present invention provides a performance parameter prediction device for a test program, which can effectively improve the speed and accuracy of the performance parameter prediction of the test program.

[0084] like Figure 6 As shown, an embodiment of the present invention further provides a performance parameter prediction device for a test program, comprising: An acquiring unit 31 is used to acquire platform specification parameters of a first platform as a first parameter group; An input unit 32, used to input the first parameter group into a performance prediction model and obtain an output result, wherein the performance prediction model is a pre-trained artificial intelligence model; The prediction unit 33 is used to predict the performance parameters of the test program running on the first platform according to the output result of the performance prediction model.

[0085] The performance parameter prediction device of the test program provided in the embodiment of the present invention can obtain the platform specification parameters of the first platform as the first parameter group, input the first parameter group into the performance prediction model, obtain the output result, and predict the performance parameters of the test program running on the first platform according to the output result of the performance prediction model. Since the first parameter group can include the specification parameters of the processor in the platform and / or the specification parameters of the peripheral devices connected to the processor, the various operating environments of the test program can be conveniently and flexibly defined through the first parameter group. After the first parameter group is input into the performance prediction model, the performance parameters of the test program in the corresponding operating environment can be quickly and accurately predicted with the help of the performance prediction model, without the need for large-scale simulation and without relying on the subjective experience of engineers. Therefore, the speed and accuracy of the performance parameter prediction of the test program can be effectively improved.

[0086] In addition, since the performance parameter prediction model is used to predict the performance parameters of the test program, it is not necessary for the test program to run on the actual platform. That is, the performance parameter prediction method for the test program provided by the embodiment of the present invention can be independent of the processor's operating platform. Therefore, it can effectively improve the flexibility of the performance parameter evaluation of the test program during the processor development and debugging process, and can also greatly reduce the time consumption of the performance parameter prediction of the test program on the hardware platform and the simulation platform.

[0087] In one embodiment, the number of the test programs is at least one, and the types of the test programs are the same or different.

[0088] In one embodiment, the input unit 32 is specifically used to: construct a first input vector using the first parameter group, and the first input vector is compatible with the input layer of the performance prediction model; input the first input vector into the input layer of the performance prediction model, and obtain a first output result from the output layer of the performance prediction model.

[0089] In one embodiment, each of the specification parameters is a dimension of the first input vector, and the number of dimensions of the first input vector is greater than or equal to the number of the platform specification parameters in the first parameter group.

[0090] In one embodiment, it also includes: an adjustment unit, which is used to adjust the values ​​corresponding to a part of the dimensions in the first input vector after predicting the performance parameters of the test program running on the first platform according to the output result of the performance prediction model, so as to obtain at least one second input vector, and input each second input vector into the performance prediction model through the input unit to obtain a corresponding second output result; the device also includes a determination unit, which is used to determine the impact of the platform specification parameters corresponding to the adjusted dimensions on the performance parameters of the test program running on the first platform according to the difference between each second input vector and the first input vector, and the difference between each second output result and the first output result.

[0091] In one embodiment, the acquisition unit 31 is also used to obtain the operating status data of at least one second platform before inputting the first parameter group into the performance prediction model and obtaining the output result, wherein each second platform runs the test program respectively; the operating status data includes at least one platform specification parameter of the second platform and the performance parameters of the test program running on the second platform; the device also includes a model training unit, which is used to use the operating status data to perform model training to obtain the performance prediction model.

[0092] In one embodiment, the device includes: a generation module, used to generate a first data set based on the operating status data, the first data set includes at least one specification-performance data pair, each of the specification-performance data pair includes a second parameter group formed by one of the platform specification parameters of the second platform, and performance parameters of the test program running on the second platform corresponding to the second parameter group; model training is performed based on the first data set to obtain the performance prediction model; a training module is used to perform model training based on the first data set to obtain the performance prediction model.

[0093] In one embodiment, the training module includes: a dimensionality reduction submodule, which is used to use principal component analysis to reduce the dimensionality of the first data set to obtain a second data set; the training submodule is specifically used to perform model training on the second data set obtained after dimensionality reduction of the first data set to obtain the performance prediction model.

[0094] In one embodiment, the dimension reduction submodule is specifically used to: take the same platform specification parameter in each of the second parameter groups as a statistical item, calculate the variance contribution rate of each statistical item, wherein the variance contribution rate is equal to the ratio of the variance of each statistical item to the sum of the variances of all statistical items; determine the cumulative variance contribution rate based on the variance contribution rate of each statistical item, wherein the cumulative variance contribution rate is equal to the sum of the variance contribution rates of the first k statistical items after arranging the variance contribution rates of the statistical items in order from large to small, wherein k is a positive integer; determine the principal component item from each of the statistical items based on the relationship between the cumulative variance contribution rate and a preset threshold, and the platform specification parameter corresponding to the principal component item is the principal component parameter; use the principal component parameter in each of the specification-performance data pairs and the performance parameters of the test program in the specification-performance data pair running on the second platform to generate a dimension reduction data pair, and the dimension reduction data pairs corresponding to all the specification-performance data pairs form the second data set.

[0095] In one implementation, the second platform includes at least one of the following: a simulation platform, a hardware platform.

[0096] In one embodiment, the platform specification parameter includes at least one of the following: core frequency, number of cores, bus frequency, bus bandwidth, memory capacity, number of memory channels, and memory speed.

[0097] In one embodiment, the performance prediction model includes at least one of the following: an LSTM model, a support vector regression model, a random forest model, and a linear regression model.

[0098] In a third aspect, an embodiment of the present invention further provides an electronic device that can effectively improve the speed and accuracy of predicting the performance parameters of a test program.

[0099] like Figure 7 As shown, the electronic device provided by an embodiment of the present invention may include: a shell 51, a processor 52, a memory 53, a circuit board 54 and a power supply circuit 55, wherein the circuit board 54 is arranged inside the space enclosed by the shell 51, and the processor 52 and the memory 53 are arranged on the circuit board 54; the power supply circuit 55 is used to supply power to various circuits or devices of the above-mentioned electronic device; the memory 53 is used to store executable program codes; the processor 52 runs the program corresponding to the executable program code by reading the executable program code stored in the memory 53, so as to execute the performance parameter prediction method of the test program provided in any of the aforementioned embodiments.

[0100] The specific execution process of the above steps by the processor 52 and the steps further executed by the processor 52 by running the executable program code can be found in the description of the previous embodiment, which will not be repeated here.

[0101] In the fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the performance parameter prediction method of any test program provided in the aforementioned embodiments, thereby also being able to achieve the corresponding technical effects, which have been described in detail above and will not be repeated here.

[0102] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0103] The processor used in the performance parameter prediction of the test program described in this specification is not limited to the CPU (central processing unit), but can also be a GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), NPU (Neural Network Processing Unit) and DSP (Digital Signal Processing) and the like.

[0104] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0105] In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0106] For the convenience of description, the above device is described by dividing the functions into various units / modules. Of course, when implementing the present invention, the functions of each unit / module can be implemented in the same or more than two software and / or hardware.

[0107] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0108] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

[0109] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for predicting performance parameters of a test program, characterized in that: include: Acquire platform specification parameters of the first platform as a first parameter group, wherein the platform specification parameters include at least one of the following: specification parameters of a processor in the platform, and specification parameters of a peripheral device in the platform connected to the processor; Inputting the first parameter group into a performance prediction model and obtaining an output result, wherein the performance prediction model is a pre-trained artificial intelligence model; The performance parameters of the test program running on the first platform are predicted based on the output results of the performance prediction model.

2. The method according to claim 1, characterized in that The number of the test programs is at least one, and the types of the test programs are the same or different.

3. The method according to claim 1, characterized in that The step of inputting the first parameter group into the performance prediction model and obtaining an output result includes: constructing a first input vector using the first parameter group, the first input vector being adapted to an input layer of the performance prediction model; The first input vector is input into the input layer of the performance prediction model, and a first output result is obtained from the output layer of the performance prediction model.

4. The method according to claim 3, characterized in that Each of the platform-specification parameters in the first parameter group corresponds to a dimension of the first input vector, and the number of dimensions of the first input vector is greater than or equal to the number of the platform-specification parameters in the first parameter group.

5. The method according to claim 3, characterized in that: After predicting the performance parameters of the test program running on the first platform according to the output result of the performance prediction model, the method further includes: Adjusting values ​​of a portion of dimensions in the first input vector to obtain at least one second input vector; Inputting each of the second input vectors into the performance prediction model respectively to obtain a corresponding second output result; Based on the difference between each of the second input vectors and the first input vector, and the difference between each of the second output results and the first output result, determine the impact of the platform specification parameters corresponding to the adjusted dimension on the performance parameters of the test program running on the first platform.

6. The method according to any one of claims 1 to 5, characterized in that Before inputting the first parameter group into the performance prediction model and obtaining an output result, the method further includes: Acquire running status data of at least one second platform, wherein each second platform runs the test program respectively, and the running status data includes at least one platform specification parameter of the second platform and a performance parameter of the test program running on the second platform; The operating status data is used to perform model training to obtain the performance prediction model.

7. The method according to claim 6, characterized in that The using the operating status data to perform model training to obtain the performance prediction model comprises: Generate a first data set based on the running status data, the first data set including at least one specification-performance data pair, each of the specification-performance data pairs including a second parameter group formed by one of the platform specification parameters of the second platform, and a performance parameter of the test program running on the second platform corresponding to the second parameter group; Model training is performed based on the first data set to obtain the performance prediction model.

8. The method according to claim 7, characterized in that The performing model training based on the first data set to obtain the performance prediction model comprises: Using principal component analysis to perform dimensionality reduction processing on the first data set to obtain a second data set; The second data set obtained after dimensionality reduction processing of the first data set is used to perform model training to obtain the performance prediction model.

9. The method according to claim 8, characterized in that The using principal component analysis method to perform dimensionality reduction processing on the first data set includes: Taking the same platform specification parameter in each of the second parameter groups as a statistical item, calculating the variance contribution rate of each statistical item, wherein the variance contribution rate is equal to the ratio of the variance of each statistical item to the sum of the variances of all statistical items; Determine the cumulative variance contribution rate according to the variance contribution rate of each statistical item, wherein the cumulative variance contribution rate is equal to the sum of the variance contribution rates of the first k statistical items after arranging the variance contribution rates of the statistical items in descending order, wherein k is a positive integer; According to the relationship between the cumulative variance contribution rate and the preset threshold, a principal component item is determined from each of the statistical items, and the platform specification parameter corresponding to the principal component item is the principal component parameter; A dimensionality reduction data pair is generated using the principal component parameters in each specification-performance data pair and the performance parameters of the test program in the specification-performance data pair running on the second platform, and the dimensionality reduction data pairs corresponding to all the specification-performance data pairs form the second data set.

10. The method according to claim 6, characterized in that The second platform includes at least one of the following: a simulation platform and a hardware platform.

11. The method according to any one of claims 1 to 5, characterized in that The platform specification parameters include at least one of the following: core frequency, number of cores, bus frequency, bus bandwidth, memory capacity, number of memory channels, and memory speed.

12. The method according to any one of claims 1 to 5, characterized in that The performance prediction model includes at least one of the following: a long short-term memory network model, a support vector regression model, a random forest model, and a linear regression model.

13. A performance parameter prediction device for a test program, characterized in that: include: An acquiring unit, configured to acquire platform specification parameters of a first platform as a first parameter group, wherein the platform specification parameters include at least one of the following: specification parameters of a processor in the platform, and specification parameters of a peripheral device in the platform connected to the processor; An input unit, used to input the first parameter group into a performance prediction model and obtain an output result, wherein the performance prediction model is a pre-trained artificial intelligence model; A prediction unit is used to predict the performance parameters of the test program running on the first platform according to the output result of the performance prediction model.

14. The device according to claim 13, characterized in that The number of the test programs is at least one, and the types of the test programs are the same or different.

15. The device according to claim 13, characterized in that The input unit is specifically used for: constructing a first input vector using the first parameter group, the first input vector being adapted to an input layer of the performance prediction model; The first input vector is input into the input layer of the performance prediction model, and a first output result is obtained from the output layer of the performance prediction model.

16. The device according to claim 15, characterized in that Each of the specification parameters is a dimension of the first input vector, and the number of dimensions of the first input vector is greater than or equal to the number of the platform specification parameters in the first parameter group.

17. The device according to claim 15, characterized in that The device also includes: an adjusting unit, configured to adjust values ​​corresponding to a portion of dimensions in the first input vector after predicting the performance parameters of the test program running on the first platform according to the output result of the performance prediction model, so as to obtain at least one second input vector; The input unit is further used to input each of the second input vectors into the performance prediction model to obtain a corresponding second output result; The device also includes a determination unit for determining the impact of the platform specification parameters corresponding to the adjusted dimension on the performance parameters of the test program running on the first platform based on the differences between each of the second input vectors and the first input vector, and the differences between each of the second output results and the first output result.

18. The device according to any one of claims 13 to 17, characterized in that: The acquisition unit is further configured to input the first parameter group into the performance prediction model, and before obtaining an output result, acquire operation status data of at least one second platform, wherein each second platform runs the test program respectively, and the operation status data includes at least one platform specification parameter of the second platform and a performance parameter of the test program running on the second platform; The device also includes a model training unit, which is used to perform model training using the operating status data to obtain the performance prediction model.

19. The device according to claim 18, characterized in that The model training unit comprises: a generating module, configured to generate a first data set based on the running status data, the first data set comprising at least one specification-performance data pair, each of the specification-performance data pairs comprising a second parameter group formed by one of the platform specification parameters of the second platform, and a performance parameter of the test program running on the second platform corresponding to the second parameter group; A training module is used to perform model training based on the first data set to obtain the performance prediction model.

20. The device according to claim 19, characterized in that The training module includes: A dimensionality reduction submodule, used to perform dimensionality reduction processing on the first data set by using a principal component analysis method to obtain a second data set; The training submodule is specifically used to perform model training using the second data set obtained after dimensionality reduction processing of the first data set to obtain the performance prediction model.

21. The device according to claim 20, characterized in that: The dimension reduction submodule is specifically used for: Taking the same platform specification parameter in each of the second parameter groups as a statistical item, calculating the variance contribution rate of each statistical item, wherein the variance contribution rate is equal to the ratio of the variance of each statistical item to the sum of the variances of all statistical items; Determine the cumulative variance contribution rate according to the variance contribution rate of each statistical item, wherein the cumulative variance contribution rate is equal to the sum of the variance contribution rates of the first k statistical items after arranging the variance contribution rates of the statistical items in descending order, wherein k is a positive integer; According to the relationship between the cumulative variance contribution rate and the preset threshold, a principal component item is determined from each of the statistical items, and the platform specification parameter corresponding to the principal component item is the principal component parameter; A dimensionality reduction data pair is generated using the principal component parameters in each specification-performance data pair and the performance parameters of the test program in the specification-performance data pair running on the second platform, and the dimensionality reduction data pairs corresponding to all the specification-performance data pairs form the second data set.

22. The device according to claim 18, characterized in that The second platform includes at least one of the following: a simulation platform and a hardware platform.

23. The device according to any one of claims 13 to 17, characterized in that The platform specification parameters include at least one of the following: core frequency, number of cores, bus frequency, bus bandwidth, memory capacity, number of memory channels, and memory speed.

24. The device according to any one of claims 13 to 17, characterized in that The performance prediction model includes at least one of the following: a long short-term memory network model, a support vector regression model, a random forest model, and a linear regression model.

25. An electronic device, characterized in that: The electronic device comprises: a memory and a processor, wherein the memory is used to store computer-readable instructions; the processor is used to read the computer-readable instructions and implement the performance parameter prediction method of the test program as described in any one of claims 1 to 12.

26. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the performance parameter prediction method of the test program according to any one of claims 1 to 12 is implemented.

Images