Grid point traversal method, device and equipment based on double-gradient enhanced regression device

Abstract

The invention relates to the technical field of spatial arrangement design of electronic devices, in particular to a lattice point traversal method, device and equipment based on a double-gradient enhanced regression device. The method comprises the steps of obtaining a target space design set; generating a target area according to the target space design set; selecting a to-be-traversed sub-region, and performing initialization traversal on each spatial design grid point to obtain at least one reference grid point; determining a next to-be-traversed sub-region according to each reference grid point, traversing the spatial design grid points in the next to-be-traversed sub-region, and performing performance parameter prediction on each candidate grid point; acquiring actual performance parameters of the target candidate grid points; updating the one or more reference grid points; and repeating the steps until all the space design grid points are traversed, so as to find out the reference grid point with the optimal performance parameter. According to the method, the number of times of comprehensive operation can be reduced, the calculation cost of space design exploration is reduced, and meanwhile, the finally updated reference grid points have optimal performance parameters.

Description

Background Art

[0002] The purpose of design space exploration (DSE) is to identify implementation plans with the best quality characteristics, that is, with the best design parameters, while satisfying all imposed design constraints. In the field of electronic devices, such as integrated circuits, chips and other devices, the most suitable design parameters are found through design space exploration. Existing methods use high-level synthesis methods to comprehensively traverse all design spaces to identify Pareto optimal designs. However, this existing approach requires running high-level synthesis tools a large number of times to identify performance parameters of different space designs. Due to the multi-dimensional characteristics of the design and the high cost of calculation during the operation of the synthesis tool, the existing space design exploration becomes a difficult and arduous task. Therefore, there are problems of low efficiency and high calculation and exploration costs in the existing technology. Summary of the invention

[0003] The present application provides a grid traversal method, device and equipment based on a dual gradient enhanced regressor to solve one or more technical problems existing in the prior art and at least provide a beneficial choice or create conditions.

[0004] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present application.

[0005] According to one aspect of an embodiment of the present application, a grid traversal method based on a dual gradient enhanced regressor is provided, the method comprising: S1, obtaining a target space design set, wherein the target space design set is composed of a plurality of space design instruction sets; S2, generating a target area according to the target space design set, wherein the target area includes space design grid points corresponding to each of the space design instruction sets; S3, selecting a sub-region to be traversed in the target region, and performing initial traversal on each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point, wherein the performance parameter of the reference grid point meets a preset performance condition; S4, determining the next sub-region to be traversed according to each of the reference grid points, and traversing each of the spatial design grid points in the next sub-region to be traversed to obtain candidate grid points adjacent to each of the reference grid points; S5, predicting the performance parameters of each of the candidate grid points by using a dual gradient enhancement regressor to obtain the performance prediction parameters of each of the candidate grid points; S6, comprehensively comparing the performance prediction parameters of each of the candidate grid points with the performance parameters of each of the reference grid points, and if a target candidate grid point with a performance prediction parameter that is better than or partially better than any of the performance parameters is obtained, obtaining the actual performance parameters of each of the target candidate grid points; S7, comprehensively comparing the actual performance parameters of each of the target candidate grid points with the performance parameters of each of the reference grid points, and if a target grid point whose actual performance parameter is better than or partially better than any of the performance parameters is obtained, updating one or more of the reference grid points according to the target grid point; S8, repeating steps S4-S7 until all spatial design grid points in the target area are traversed to find the reference grid point with the best performance parameters.

[0006] In one embodiment of the present application, based on the above solution, the initializing traversal of each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point includes: Acquire performance parameters of each spatial design grid point in the sub-region to be traversed by a high-level synthesis tool, wherein the performance parameters include a delay and an area for completing the spatial design; Comprehensively comparing the various performance parameters to determine one or more reference grid points with the lowest delay and / or the smallest area; The preset performance condition is that, among the traversed spatial design grid points, there exists a preset non-dominated relationship between any two reference grid points.

[0007] In one embodiment of the present application, based on the above solution, determining the next sub-region to be traversed according to each of the reference grid points includes: If the number of the reference grid point is one, constructing the next to-be-traversed sub-region of a first preset shape and a first preset size with the reference grid point as a starting point; If there are multiple reference grid points, a starting area is constructed according to the positions of the reference grid points, and the next sub-area to be traversed of a second preset shape and a second preset size is constructed according to the starting area.

[0008] In one embodiment of the present application, based on the above-mentioned solution, the dual gradient enhancement regressor includes a first regressor group related to time delay and a second regressor group related to area, and the performance parameter prediction of each candidate grid point is performed by the dual gradient enhancement regressor to obtain the performance prediction parameter of each candidate grid point, including: Performing delay prediction on each of the candidate grid points respectively by using the first regressor group to obtain a predicted delay of each of the candidate grid points; Using the second regressor group, each candidate grid point is predicted to have an area, thereby obtaining a predicted area of ​​each candidate grid point; The performance prediction parameters include the predicted delay and the predicted area.

[0009] In one embodiment of the present application, based on the above-mentioned solution, if the target candidate grid point whose performance prediction parameter is better than or partially better than any of the performance parameters is obtained, then the actual performance parameters of each of the target candidate grid points are obtained, including: If the target candidate grid point whose predicted delay is lower than the delay of any of the reference grid points is obtained, the actual delay of each of the target candidate grid points is obtained by the high-level synthesis tool; If the predicted area of ​​the target candidate grid point is smaller than the area of ​​any of the reference grid points, the actual area of ​​each of the target candidate grid points is obtained by the high-level synthesis tool; If the target candidate grid point is obtained whose predicted delay is lower than the delay of any of the reference grid points and whose predicted area is smaller than the area of ​​any of the reference grid points, the actual delay and actual area of ​​each of the target candidate grid points are obtained through the high-level synthesis tool.

[0010] In one embodiment of the present application, based on the above solution, if the actual performance parameter is better than or partially better than any target grid point of the performance parameter, then updating one or more reference grid points according to the target grid point includes: If the target grid point whose actual delay is lower than the delay of any of the reference grid points is obtained, then searching for the target reference grid point whose delay is higher than the actual delay, marking each of the target reference grid points as a failed grid point, and marking the target grid point as a reference grid point; If the target grid point whose actual area is smaller than the area of ​​any reference grid point is obtained, then a target reference grid point whose area is larger than the actual time delay is searched, and each of the target reference grid points is marked as a failed grid point, and the target grid point is marked as a reference grid point; If the target grid point whose actual delay and actual area are both smaller than the delay and area of ​​any reference grid point is obtained, then a target reference grid point whose delay is higher than the actual delay and whose area is larger than the actual area is searched, and each of the target reference grid points is marked as a failed grid point, and the target grid point is marked as a reference grid point.

[0011] In one embodiment of the present application, based on the above solution, after comprehensively comparing the performance prediction parameters of each candidate grid point with the performance parameters of each reference grid point, the method further includes: If the predicted delays of the candidate grid points are all higher than the delays of the reference grid points and the predicted areas of the candidate grid points are all larger than the areas of the reference grid points, the candidate grid points are marked as poor quality grid points.

[0012] In one embodiment of the present application, based on the above solution, after all spatial design grid points in the target area are traversed, the method further includes: Obtain abandoned grid points that are not predicted by the dual gradient boosting regressor for performance parameters, Obtaining the delay and area of ​​each of the inferior grid points and the abandoned grid points by the high-level synthesis tool; If the delay of any of the inferior grid points or the abandoned grid points is lower than the delay of the reference grid points and / or the area of ​​any of the inferior grid points or the abandoned grid points is smaller than the area of ​​the reference grid points, the inferior grid points or abandoned grid points whose delay is lower than the delay of the reference grid points and / or whose area is smaller than the area of ​​the reference grid points will be marked as reference grid points.

[0013] According to one aspect of an embodiment of the present application, a grid traversal device based on a dual gradient enhanced regressor is provided, the device comprising: An acquisition unit, used for acquiring a target space design set, wherein the target space design set is composed of a plurality of space design instruction sets; A generating unit, configured to generate a target area according to the target space design set, wherein the target area includes space design grid points corresponding to each of the space design instruction sets in a one-to-one manner; A selection unit, configured to select a sub-region to be traversed in the target region, and initialize and traverse each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point; A determination unit, configured to determine a next sub-region to be traversed according to each of the reference grid points, and traverse each of the spatial design grid points in the next sub-region to be traversed to obtain candidate grid points adjacent to each of the reference grid points; A prediction unit, used to predict the performance parameters of each candidate grid point by using a dual gradient enhancement regressor to obtain the performance prediction parameters of each candidate grid point; A first comprehensive comparison unit is used to comprehensively compare the performance prediction parameters of each of the candidate grid points with the performance parameters of each of the reference grid points, and if a target candidate grid point with a performance prediction parameter that is better than or partially better than any of the performance parameters is obtained, then the actual performance parameters of each of the target candidate grid points are obtained; A second comprehensive comparison unit is used to make a comprehensive comparison between the actual performance parameters of each of the target candidate grid points and the performance parameters of each of the reference grid points, and if a target grid point with an actual performance parameter that is better than or partially better than any of the performance parameters is obtained, one or more of the reference grid points are updated according to the target grid point; The iteration unit is used to repeat the above steps until all the spatial design grid points in the target area are traversed to find the reference grid point with the best performance parameters.

[0014] According to one aspect of an embodiment of the present application, an electronic device is provided, comprising: one or more processors; and a memory for storing executable instructions of the processors, wherein when the executable instructions are executed by the one or more processors, the one or more processors implement the methods described in the above embodiments.

[0015] The beneficial effect of the present application is: by obtaining a pre-set target space design set, the target space design set is composed of several space design instruction sets, and each space design instruction set corresponds to a space design. What the present application needs to do is to find the space design with the best performance among the various space design instruction sets in the target space design set, and then to manufacture electronic devices based on the space design with the most quality characteristics found.

[0016] Among them, the difference between the present application and the existing spatial design exploration method is that the performance parameters of the partial area are obtained by initializing the traversal, that is, by performing the first high-level synthesis, that is, the performance parameters of all spatial design grids in the sub-area to be traversed as described in the present application, so as to obtain one or more reference grids with the best performance. It should be noted that the performance parameters can be composed of multiple evaluation parameters. The reference grid with the best performance can represent that a certain evaluation parameter of the reference grid is the best, while other evaluation parameters are not better than the evaluation parameters of other reference grids, so there may be one or more reference grids with the best performance.

[0017] Furthermore, after the initial traversal is completed in the sub-region to be traversed, the next sub-region to be traversed is determined, and each spatial design grid point in the next traversal sub-region is traversed, that is, the candidate grid points adjacent to each of the reference grid points are found. Because the performance parameters of the adjacent candidate grid points are close to the reference grid points, the performance parameters are continuously compared with the reference grid points to find the spatial design grid points with better performance parameters, so as to facilitate the subsequent updating of the reference grid points.

[0018] Furthermore, the performance parameters of each of the candidate grid points are predicted by a dual gradient enhancement regressor, without the need to run a high-level synthesis tool to actually obtain the performance parameters, thereby reducing the computational cost during spatial design exploration, and the prediction method using the dual gradient enhancement regressor can effectively improve the efficiency of traversal, and quickly predict the performance prediction parameters of the candidate grid points, and then comprehensively compare the predicted performance prediction parameters with the performance parameters of the reference grid points. If there is a candidate grid point whose performance prediction parameters are better than one or more reference grid points, then the candidate grid point is marked as a target candidate grid point, and then the actual performance parameters of the target candidate grid point are obtained through a high-level synthesis tool for final comparison with the performance parameters of the reference grid points. If it is better than or partially better than the performance parameters of the reference grid points, then the reference grid point is updated, that is, the target candidate grid point can be marked as a reference grid point.

[0019] Furthermore, by continuously repeating steps S4-S7, the next sub-area to be traversed can be continuously determined, that is, the spatial design grid points in the target area are continuously traversed. If a spatial design grid point with better performance parameters is found during the traversal process, it is marked as a reference grid point and updated until all spatial design grid points in the target area are traversed. In this way, the marked or updated reference grid point is the reference grid point with the best performance parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief description of the drawings required for use in the description of the embodiments. Obviously, the drawings described are only part of the embodiments of the present application, not all of the embodiments, and those skilled in the art can also obtain other design solutions and drawings based on these drawings without creative work.

[0021] Figure 1 It is a flowchart of a grid traversal method based on a dual gradient enhanced regressor according to an embodiment of the present application; Figure 2 A logical schematic diagram of traversing a target area according to an embodiment of the present application; Figure 3 A logic flow chart showing a grid traversal method based on a dual gradient enhanced regressor according to an embodiment of the present application; Figure 4 It is a block diagram of a grid traversal device based on a dual gradient enhancement regressor according to an embodiment of the present application; Figure 5 It is a schematic diagram of the system structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Example embodiments are now described more fully in conjunction with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.

[0023] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the present application.

[0024] The block diagrams shown in the accompanying drawings are only functional entities and do not necessarily correspond to physically independent entities. These functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or micro-control node devices.

[0025] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0026] It should be noted that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0027] The following is a detailed introduction to the background technology of this application: Design Space Exploration (DSE) methods can be divided into two categories: model-based methods and synthesis-based methods. Model-based methods include supervised learning-based methods and graph analysis-based methods. Supervised learning-based methods rely on the knowledge of previously synthesized designs (the relationship between design configurations and features) to train offline supervised models. The trained model has rich pre-screening capabilities and can approximate the characteristics of new designs, thereby reducing the burden of synthesis. Graph analysis-based methods use prediction techniques on control data flow graphs (CDFGs) to simulate the behavior of HLS (High-level Synthesis) tools without the knowledge of previous synthesized designs. Then, techniques such as graph neural networks are used to infer the characteristics of the input design based on the CDFG. With the assistance of some intelligent sampling strategies, this method can quickly produce results for design space exploration, but it has shortcomings in terms of maintenance and accuracy.

[0028] The synthesis-based approach relies on the HLS (High-level Synthesis) tool to perform the synthesis process to obtain the characteristics of the design and use heuristic algorithms (dedicated heuristics or meta-heuristics) or prior knowledge to identify important areas. New candidate designs are continuously generated in these areas, and the most promising designs are selected for synthesis. As the number of synthesis increases, the quality of the DSE results is iteratively optimized until the exit criteria are met. The synthesis-based approach performs better in accuracy and flexibility because it relies on the synthesis process. However, the lack of learning ability also leads to a higher synthesis burden, which may be further aggravated if the candidate selection is inappropriate. In addition, the synthesis-based approach only uses the characteristic information of the synthesized design to identify important areas, but ignores the synthesis knowledge generated during the design space exploration process.

[0029] Therefore, based on the above-mentioned defects, this application proposes a grid traversal method (DGBR-Lattice) based on dual gradient boosting regressors, which has both learning and optimization capabilities and can effectively solve the design space exploration (DSE) problem of high-level synthesis. DGBR-Lattice uses the comprehensive knowledge learned by two gradient boosting regressors (GBR, Gradient boosting regression) based on ensemble learning technology, that is, corresponding to the first regressor group related to delay and the second regressor group related to area described in this application, to assist the grid traversal method and identify the most promising designs in the optimization process (traversal process). DGBR-Lattice can reduce the number of runs of the synthesis tool by at least 1.42 times and at most 9.62 times while obtaining a Pareto front close to the real one, greatly reducing the computational cost of space design exploration.

[0030] The implementation details of the technical solution of the embodiment of the present application are described in detail below: According to one aspect of an embodiment of the present application, a grid traversal method based on a dual gradient enhanced regressor is provided. Figure 1 This is a flowchart of a grid traversal method based on a dual gradient enhanced regressor according to an embodiment of the present application. The method includes at least steps S1 to S8, which are described in detail as follows: In step S1, a target space design set is obtained, where the target space design set consists of several space design instruction sets.

[0031] Specifically, in HLS (High-Level Synthesis), the spatial design instructions (i.e., #pragma instructions) and specific behavioral descriptions are mapped to RTL (Register Transfer Level) microarchitectures by calling a high-level synthesis tool to perform the synthesis process (synthesis run). Given a specific behavioral description, by selecting different combinations of #pragma instruction values ​​(i.e., spatial design), RTL outputs with different performance parameters (delay and area) can be obtained.

[0032] That is to say, the target space design set contains multiple space design instruction sets, and a single space design instruction set is specifically defined by the value of the space design instruction (i.e., the #pragma instruction), that is, the space design instruction set is composed of instructions under multiple different dimensions, which can be reflected as follows:

[0033] Among them, DS is the space design instruction set, , ,... Each of them is a space design directive, and the space design directive is composed of multiple #pragma directive values. A single space design directive can be expressed as follows:

[0034] in, is a specific instruction value. Since the instruction value of the space design instruction grows exponentially, the time required to call the synthesis tool to execute the synthesis process, that is, the number of synthesis runs, ranges from minutes to hours. Therefore, it is very necessary to propose an intelligent and efficient design space exploration (DSE) method to find a Pareto design (optimal performance / most ideal space design) with high performance and low cost while reducing the number of synthesis runs.

[0035] In step S2, a target area is generated according to the target space design set, and the target area includes space design grid points corresponding one-to-one to each of the space design instruction sets.

[0036] Specifically, Figure 2 As shown in a, Figure 2 a or b or c or d in the figure are exemplary target areas, wherein the target area has a plurality of spatial design grid points ( Figure 2 The design group in the above formula is the blue grid and the red reference grid). The area of ​​the horizontal axis and the delay of the vertical axis are the performance parameters of each spatial design grid, that is, the evaluation parameters. Each spatial design grid corresponds to a spatial design, which is the DS in the above formula.

[0037] First, after initializing the traversal, we get Figure 2 As shown in a in the figure, four reference grid points are obtained, and there is a non-dominated relationship between each reference grid point. It should be noted that in the embodiment of the present application, it is defined that there is a non-dominated relationship between each reference grid point, that is, the non-dominated relationship belongs to one of the preset performance conditions.

[0038] In step S3, a sub-region to be traversed is selected in the target region, and each spatial design grid point in the sub-region to be traversed is initialized and traversed to obtain at least one reference grid point, and the performance parameters of the reference grid point meet the preset performance conditions.

[0039] Specifically, the preset performance condition can be the non-dominated relationship mentioned above. As long as a certain spatial design grid point has a non-dominated relationship with any reference grid point, it means that the spatial design grid point meets the preset performance condition, and the spatial design grid point can be marked as a reference grid point.

[0040] The sub-region to be traversed can select any region in the target region at will. The area and shape of the sub-region to be traversed can be set as needed. The performance parameters of each spatial design grid point in the sub-region to be traversed, that is, the delay and area, can be obtained through high-level synthesis tools, and then the reference grid point can be traversed to obtain one or more grid points with the best performance in the sub-region to be traversed, such as Figure 2 As shown in a in , there are four reference grid points, and there is a non-dominated relationship between each reference grid point.

[0041] In one embodiment of the present application, the initializing traversal of each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point includes: Acquire performance parameters of each spatial design grid point in the sub-region to be traversed by a high-level synthesis tool, wherein the performance parameters include a delay and an area for completing the spatial design; Comprehensively comparing the various performance parameters to determine one or more reference grid points with the lowest delay and / or the smallest area; The preset performance condition is that, among the traversed spatial design grid points, there exists a preset non-dominated relationship between any two reference grid points.

[0042] Specifically, the non-dominated relationship may be specifically as follows: selecting any one of the reference grid points as the first reference grid point, and selecting any one of the reference grid points other than the first reference grid point as the second reference grid point; Then the delay of the first reference grid point is higher than the delay of the second reference grid point and the area of ​​the first reference grid point is smaller than the area of ​​the second reference grid point, or the delay of the first reference grid point is lower than the delay of the second reference grid point and the area of ​​the first reference grid point is larger than the area of ​​the second reference grid point, which is called a non-dominant relationship; On the contrary, the dominance relationship can be specifically as follows: the delay and area of ​​the first reference grid point are both smaller than those of the second reference grid point. It can be considered that the first reference grid point can dominate the second reference grid point. In this case, the performance parameters of the first reference grid point are all better than those of the second reference grid point. Alternatively, the delay and area of ​​the first reference grid point are both greater than those of the second reference grid point, and it can be considered that the second reference grid point can dominate the first reference grid point. In this case, the performance parameters of the second reference grid point are all better than those of the first reference grid point.

[0043] In step S4, the next sub-region to be traversed is determined according to each of the reference grid points, and each of the spatial design grid points in the next sub-region to be traversed is traversed to obtain candidate grid points adjacent to each of the reference grid points.

[0044] Specifically, the next sub-area to be traversed can also be set according to actual needs. After traversing the current next sub-area to be traversed, the next sub-area to be traversed in the next round will be determined. By continuously performing spatial exploration on the untraversed areas in the target area (finding reference grids / updating reference grids), the spatial design grid with the best performance parameters is found during the exploration process, and iterative updates are performed. Finally, after the target area is traversed, the spatial design grid with the best performance design can be obtained, that is, the reference grid that has been continuously updated and iterated.

[0045] In one embodiment of the present application, determining the next sub-region to be traversed according to each of the reference grid points includes: If the number of the reference grid point is one, constructing the next to-be-traversed sub-region of a first preset shape and a first preset size with the reference grid point as a starting point; If there are multiple reference grid points, a starting area is constructed according to the positions of the reference grid points, and the next sub-area to be traversed of a second preset shape and a second preset size is constructed according to the starting area.

[0046] Specifically, the next sub-area to be traversed can refer to Figure 2 The circular area corresponding to the search range shown in b in FIG. 1 is a circle area. It should be noted that the circular area is only an exemplary description, and the next sub-area to be traversed may also be set to a rectangular, polygonal or irregular area.

[0047] Specifically, if only one reference grid point is currently traversed, it means that the delay and area of ​​this reference grid point are smaller than the delay and area of ​​other spatial design grid points that have been traversed. If multiple reference grid points appear, it means that there is a non-dominant relationship between these reference grid points, that is, the performance parameters of each reference grid point are partially better than the performance parameters of other reference grid points, for example, the delay of a reference grid point is lower, but the area is larger, and the delay of another reference grid point is higher, but the area is smaller.

[0048] It should be noted that delay, as an evaluation parameter or a performance parameter, corresponds to the spatial design of electronic devices. Delay represents the time required to complete the spatial design, while area is easier to understand. Area represents the area required to complete the spatial design.

[0049] The starting area is described as an example, that is, the starting area is determined based on at least two reference grid points, and the midpoint of the starting area is used as the starting point to set the traversal area of ​​any step length as the next sub-area to be traversed. The first preset shape and the second preset shape can be set arbitrarily. Figure 2 The shape shown in is a circle, and the first preset size and the second preset size can also be set arbitrarily. Figure 2 The dimensions shown in are circular areas of radius r.

[0050] In step S5, a performance parameter prediction is performed on each of the candidate grid points using a dual gradient enhancement regressor to obtain a performance prediction parameter of each of the candidate grid points.

[0051] Specifically, Figure 3 As shown, Figure 3 This is the logic flow chart of the present application. After finding the candidate grid points, they are archived and then each candidate grid point is input into regressor group 1 and regressor group 2.

[0052] In one embodiment of the present application, the dual gradient enhancement regressor includes a first regressor group related to time delay and a second regressor group related to area, and the performance parameter prediction of each candidate grid point is performed by the dual gradient enhancement regressor to obtain the performance prediction parameter of each candidate grid point, including: Performing delay prediction on each of the candidate grid points respectively by using the first regressor group to obtain a predicted delay of each of the candidate grid points; Using the second regressor group, each candidate grid point is predicted to have an area, thereby obtaining a predicted area of ​​each candidate grid point; The performance prediction parameters include the predicted delay and the predicted area.

[0053] Specifically, Figure 3 The regressor group 1 in is the first regressor group related to the time delay, and the regressor group 2 is the second regressor group related to the area. Figure 3 The selected space design instruction values ​​d1, d2, ... d n Corresponding to the DS in the above formula , , that is, these command values ​​are respectively predicted by the dual gradient enhancement regressor for delay and area, and the predicted delay and predicted area of ​​each candidate grid point are obtained. Figure 2 As shown in c in the figure, the nearest design is found as the target candidate grid point. Figure 2The d in indicates that the integrated design (i.e., the target grid point obtained by the current round of traversal) has been found, and then the integrated design (target grid point) is updated to the reference grid point.

[0054] In step S6, the performance prediction parameters of each candidate grid point are comprehensively compared with the performance parameters of each reference grid point. If a target candidate grid point is obtained whose performance prediction parameters are better than or partially better than any of the performance parameters, the actual performance parameters of each target candidate grid point are obtained.

[0055] Specifically, the comprehensive comparison of the performance prediction parameters of each of the candidate grid points with the performance parameters of each of the reference grid points includes: The predicted time delay of each candidate grid point is compared with the time delay of each reference grid point, and the predicted area of ​​each candidate grid point is compared with the area of ​​each reference grid point.

[0056] In one embodiment of the present application, if the target candidate grid point whose performance prediction parameter is better than or partially better than any of the performance parameters is obtained, then obtaining the actual performance parameters of each of the target candidate grid points includes: If the target candidate grid point whose predicted delay is lower than the delay of any of the reference grid points is obtained, the actual delay of each of the target candidate grid points is obtained by the high-level synthesis tool; If the predicted area of ​​the target candidate grid point is smaller than the area of ​​any of the reference grid points, the actual area of ​​each of the target candidate grid points is obtained by the high-level synthesis tool; If the target candidate grid point is obtained whose predicted delay is lower than the delay of any of the reference grid points and whose predicted area is smaller than the area of ​​any of the reference grid points, the actual delay and actual area of ​​each of the target candidate grid points are obtained through the high-level synthesis tool.

[0057] Specifically, as long as a predicted delay or predicted area is found in a candidate grid point that is greater than the delay or area of ​​any reference grid point, then the candidate grid point is marked as the target candidate grid point, and the actual area and actual delay of the target candidate grid point are obtained through the high-level synthesis tool, and then compared with the delay and area of ​​the reference grid point. This approach can reduce the number of runs of the high-level synthesis tool, and make predictions through the dual gradient enhancement regressor, which is equivalent to pre-screening potential target candidate grid points, and then making the final actual comparison, which can improve efficiency and reduce computational costs while ensuring accuracy.

[0058] In step S7, a comprehensive comparison is made between the actual performance parameters of each of the target candidate grid points and the performance parameters of each of the reference grid points. If a target grid point is obtained whose actual performance parameters are better than or partially better than any of the performance parameters, one or more of the reference grid points are updated according to the target grid point.

[0059] Specifically, the comprehensive comparison of the actual performance parameters of each of the target candidate grid points with the performance parameters of each of the reference grid points includes: The actual time delay of each of the target candidate grid points is compared with the time delay of each of the reference grid points, and the actual area of ​​each of the target candidate grid points is compared with the area of ​​each of the reference grid points.

[0060] In one embodiment of the present application, if the actual performance parameter is better than or partially better than any target grid point of the performance parameter, updating one or more reference grid points according to the target grid point includes: If the target grid point whose actual delay is lower than the delay of any of the reference grid points is obtained, then searching for the target reference grid point whose delay is higher than the actual delay, marking each of the target reference grid points as a failed grid point, and marking the target grid point as a reference grid point; If the target grid point whose actual area is smaller than the area of ​​any reference grid point is obtained, then a target reference grid point whose area is larger than the actual time delay is searched, and each of the target reference grid points is marked as a failed grid point, and the target grid point is marked as a reference grid point; If the target grid point whose actual delay and actual area are both smaller than the delay and area of ​​any reference grid point is obtained, then a target reference grid point whose delay is higher than the actual delay and whose area is larger than the actual area is searched, and each of the target reference grid points is marked as a failed grid point, and the target grid point is marked as a reference grid point.

[0061] Specifically, if the delay and area of ​​a target grid point are both smaller than a target reference grid point, the target reference grid point will be marked as a failed grid point and the target grid point will be marked as a reference grid point, and the target grid point is obtained from the target candidate grid points.

[0062] In one embodiment of the present application, after comprehensively comparing the performance prediction parameters of each of the candidate grid points with the performance parameters of each of the reference grid points, the method further includes: If the predicted delays of the candidate grid points are all higher than the delays of the reference grid points and the predicted areas of the candidate grid points are all larger than the areas of the reference grid points, the candidate grid points are marked as poor quality grid points.

[0063] Specifically, if the predicted time delay and predicted area of ​​a candidate grid point are both greater than the areas of each of the reference grid points, the candidate grid point is marked as a poor-quality grid point.

[0064] In step S8, steps S4-S7 are repeated until all spatial design grid points in the target area are traversed to find the reference grid point with the best performance parameters.

[0065] Specifically, by continuously repeating steps S4-S7, the next sub-area to be traversed can be continuously determined, that is, the spatial design grid points in the target area are continuously traversed. If a spatial design grid point with better performance parameters is found during the traversal process, it is marked as a reference grid point and updated until all spatial design grid points in the target area are traversed. In this way, the marked or updated reference grid point is the reference grid point with the best performance parameters.

[0066] In one embodiment of the present application, after all spatial design grid points in the target area are traversed, the method further includes: Obtaining the delay and area of ​​each of the poor-quality grid points by using the high-level synthesis tool; If the delay of any of the inferior grid points is lower than the delay of the reference grid point and / or the area of ​​any of the inferior grid points is smaller than the area of ​​the reference grid point, the inferior grid point whose delay is lower than the delay of the reference grid point and / or whose area is smaller than the area of ​​the reference grid point will be marked as a reference grid point.

[0067] Abandoned grid points are spatial design grid points abandoned during the traversal process. Poor-quality grid points are grid points screened out by comprehensive comparison of the performance parameters of candidate grid points and reference grid points. In order to improve the guarantee rate of optimal performance of reference grid points, the actual performance parameters of all poor-quality grid points and abandoned grid points are obtained by running the synthesis tool for the last time, that is, the delay and area of ​​each poor-quality grid point and each abandoned grid point are obtained through the high-level synthesis tool, and then a comprehensive comparison is performed. If the delay of any poor-quality grid point or abandoned grid point is lower than the delay of the reference grid point and / or the area of ​​any poor-quality grid point or abandoned grid point is smaller than the area of ​​the reference grid point, the poor-quality grid point or abandoned grid point can be updated to the reference grid point. This can prevent some reference grid points with better performance parameters from being abandoned due to some misjudgments based on the dual gradient enhancement regressor.

[0068] The following is a detailed introduction to the synthesis operation proposed in this application, that is, the synthesis process performed by the synthesis tool: The overall framework of the embodiment of the present application is as follows Figure 3As shown, it mainly consists of two parts: optimization and learning. Compared with the original grid traversal method, the main innovation of this application is to train a learning machine (double gradient boosting regressor) online to predict whether the selected design can update the reference frontier (reference grid), instead of performing cumbersome synthesis operations of high-level synthesis tools.

[0069] The optimization part uses a grid traversal algorithm (i.e., the grid traversal method based on dual gradient boosting regressors proposed in this application) to explore the design space, while the learning part uses dual gradient boosting regressors (regressor group 1 and regressor group 2) to learn comprehensive knowledge based on ensemble learning technology to predict the area and latency of each spatial design instruction set. Figure 3 As shown, each regressor group builds N regression trees (RT) in a step-by-step manner. At each stage, a regression tree is fitted to optimize the loss function on the negative gradient. Therefore, the prediction accuracy of both regressor groups 1 and 2 can be improved by optimizing the loss function. In this application, we use mean square error (MSE) as the loss function and set N to the size of the space design.

[0070] In the overall architecture of this application, all integrated designs are stored in a file, which serves as a training Given a comprehensive design (i.e., the spatial design instruction set described in this application), the corresponding comprehensive knowledge can be expressed as ( , , ),in [ , ..., ,..., ], and The configuration, area, and latency of the synthesized design are defined respectively. During the learning phase, two independent regressors ( ) are respectively obtained by learning training samples ( , ) and , ) for training. As the exploration proceeds, regressor group 1 and regressor group 2 will undergo multiple training cycles. In each training, we incorporate the latest comprehensive design (i.e., the target candidate grid point in the next sub-region to be traversed in each round of this application) into the training data. When the grid traversal stagnates, a final update is performed. The necessity of the final update stems from two aspects. First, there may be promising areas that have not been explored by the grid traversal algorithm, especially when the search radius is small. Second, true non-dominated relations may be misjudged as dominant relations by the dual gradient enhancement regression, resulting in a poor reference front. Therefore, once the grid traversal algorithm exits, we will perform a final update (that is, the above content about obtaining the delay and area of ​​each of the poor quality grid points through the high-level synthesis tool) to further improve the quality of the DSE results.

[0071] In summary, the embodiment of the present application can perform grid traversal of the target area by continuously performing spatial grid traversal of the next sub-area to be traversed. The reference grid can be continuously updated through continuous traversal, that is, the reference frontier is continuously updated. The reference grid / reference frontier obtained after the target area is traversed has the best performance parameters. When applied to the manufacture of existing electronic devices, the quality and performance characteristics of the electronic devices can be greatly improved. Through continuous prediction and learning by the dual gradient enhanced regressor, the prediction accuracy of the predicted performance parameters such as delay and area is high. As an auxiliary reference means, it can greatly reduce the computational cost of spatial design exploration while improving the exploration efficiency. More importantly, further comprehensive comparison is made based on the predicted performance parameters to improve authenticity and accuracy. Based on this, the grid traversal method based on the dual gradient enhanced regressor proposed in this application has the technical effects of reducing computational costs, reducing comprehensive operating loads, improving spatial exploration efficiency, and improving spatial exploration accuracy.

[0072] Figure 4 The block diagram of a grid traversal device 300 based on a dual gradient enhancement regressor according to an embodiment of the present application is shown. According to a grid traversal device 300 based on a dual gradient enhancement regressor according to an embodiment of the present application, the device 300 includes: an acquisition unit 301, a generation unit 302, a selection unit 303, a determination unit 304, a prediction unit 305, a first comprehensive comparison unit 306, a second comprehensive comparison unit 307, and an iteration unit 308. An acquisition unit 301 is used to acquire a target space design set, where the target space design set consists of a plurality of space design instruction sets; A generating unit 302, configured to generate a target region according to the target space design set, wherein the target region includes space design grid points corresponding to each of the space design instruction sets in a one-to-one manner; A selection unit 303 is used to select a sub-region to be traversed in the target region, and initialize traversal of each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point; A determination unit 304 is used to determine the next sub-region to be traversed according to each of the reference grid points, and traverse each of the spatial design grid points in the next sub-region to be traversed to obtain candidate grid points adjacent to each of the reference grid points; A prediction unit 305 is used to predict the performance parameters of each candidate grid point by using a dual gradient enhancement regressor to obtain the performance prediction parameters of each candidate grid point; A first comprehensive comparison unit 306 is used to comprehensively compare the performance prediction parameters of each candidate grid point with the performance parameters of each reference grid point, and if a target candidate grid point with a performance prediction parameter that is better than or partially better than any of the performance parameters is obtained, then the actual performance parameters of each target candidate grid point are obtained; A second comprehensive comparison unit 307 is used to make a comprehensive comparison between the actual performance parameters of each of the target candidate grid points and the performance parameters of each of the reference grid points, and if a target grid point with an actual performance parameter that is better than or partially better than any of the performance parameters is obtained, one or more of the reference grid points are updated according to the target grid point; The iteration unit 308 is used to repeat the above steps until all the spatial design grid points in the target area are traversed to find the reference grid point with the best performance parameters.

[0073] As another aspect, the present application further provides a computer-readable storage medium on which a program product capable of implementing the method provided above in this specification is stored. In some possible implementations, various aspects of the present application may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary implementations of the present application described in the above "Embodiment Method" section of this specification.

[0074] According to the program product for implementing the above method in the embodiment of the present application, it can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present application is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0075] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0076] Computer readable signal media may include a data signal propagated in baseband as part of a carrier wave, wherein readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0077] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0078] Program code for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0079] As another aspect, the present application also provides an electronic device capable of implementing the above method.

[0080] Those skilled in the art will appreciate that various aspects of the present application may be implemented as a system, method or program product. Therefore, various aspects of the present application may be specifically implemented in the following forms, that is, a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to as "circuit", "module" or "system" herein.

[0081] Refer to the following Figure 5 The electronic device 400 according to this embodiment of the present application is described. Figure 5 The electronic device 400 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0082] like Figure 5 As shown, the electronic device 400 is in the form of a general computing device. The components of the electronic device 400 may include but are not limited to: at least one processing unit 410, at least one storage unit 420, and a bus 430 connecting different system components (including the storage unit 420 and the processing unit 410).

[0083] The storage unit stores program codes, which can be executed by the processing unit 410, so that the processing unit 410 executes the steps described in the above “Example Method” section of this specification according to various exemplary implementations of the present application.

[0084] The storage unit 420 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 421 and / or a cache memory unit 422 , and may further include a read-only memory unit (ROM) 423 .

[0085] The storage unit 420 may also include a program / utility 424 having a set (at least one) of program modules 425, such program modules 425 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0086] Bus 430 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller node, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0087] The electronic device 400 may also communicate with one or more external devices 1200 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 400, and / or communicate with any device that enables the electronic device 400 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 450. In addition, the electronic device 400 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 460. As shown, the network adapter 460 communicates with other modules of the electronic device 400 via a bus 430. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 400, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0088] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device or a network device, etc.) to execute the method according to the implementation methods of the present application.

[0089] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiment of the present application, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously in multiple modules.

[0090] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be performed without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A grid traversal method based on a dual gradient enhanced regressor, characterized in that: The method comprises: S1, obtaining a target space design set, wherein the target space design set is composed of a plurality of space design instruction sets; S2, generating a target area according to the target space design set, wherein the target area includes space design grid points corresponding to each of the space design instruction sets; S3, selecting a sub-region to be traversed in the target region, and performing initial traversal on each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point, wherein the performance parameter of the reference grid point meets a preset performance condition; S4, determining the next sub-region to be traversed according to each of the reference grid points, and traversing each of the spatial design grid points in the next sub-region to be traversed to obtain candidate grid points adjacent to each of the reference grid points; S5, predicting the performance parameters of each of the candidate grid points by using a dual gradient enhancement regressor to obtain the performance prediction parameters of each of the candidate grid points; S6, comprehensively comparing the performance prediction parameters of each of the candidate grid points with the performance parameters of each of the reference grid points, and if a target candidate grid point with a performance prediction parameter that is better than or partially better than any of the performance parameters is obtained, obtaining the actual performance parameters of each of the target candidate grid points; S7, comprehensively comparing the actual performance parameters of each of the target candidate grid points with the performance parameters of each of the reference grid points, and if a target grid point whose actual performance parameter is better than or partially better than any of the performance parameters is obtained, updating one or more of the reference grid points according to the target grid point; S8, repeating steps S4-S7 until all spatial design grid points in the target area are traversed to find the reference grid point with the best performance parameters.

2. The grid traversal method based on dual gradient enhanced regressor according to claim 1, characterized in that: The initializing and traversing each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point includes: Acquire performance parameters of each spatial design grid point in the sub-region to be traversed by a high-level synthesis tool, wherein the performance parameters include a delay and an area for completing the spatial design; Comprehensively comparing the various performance parameters to determine one or more reference grid points with the lowest delay and / or the smallest area; The preset performance condition is that, among the traversed spatial design grid points, there exists a preset non-dominated relationship between any two reference grid points.

3. The grid traversal method based on dual gradient enhanced regressor according to claim 2, characterized in that: The step of determining the next sub-region to be traversed according to each of the reference grid points comprises: If the number of the reference grid point is one, constructing the next to-be-traversed sub-region of a first preset shape and a first preset size with the reference grid point as a starting point; If there are multiple reference grid points, a starting area is constructed according to the positions of the reference grid points, and the next sub-area to be traversed of a second preset shape and a second preset size is constructed according to the starting area.

4. The grid traversal method based on dual gradient enhanced regressor according to claim 3 is characterized in that: The dual gradient enhancement regressor includes a first regressor group related to time delay and a second regressor group related to area. The dual gradient enhancement regressor is used to predict the performance parameters of each candidate grid point to obtain the performance prediction parameters of each candidate grid point, including: Performing delay prediction on each of the candidate grid points respectively by using the first regressor group to obtain a predicted delay of each of the candidate grid points; Using the second regressor group, each candidate grid point is predicted to have an area, thereby obtaining a predicted area of ​​each candidate grid point; The performance prediction parameters include the predicted delay and the predicted area.

5. The grid traversal method based on dual gradient enhanced regressor according to claim 4, characterized in that: If the target candidate grid point whose performance prediction parameter is better than or partially better than any of the performance parameters is obtained, then the actual performance parameters of each of the target candidate grid points are obtained, including: If the target candidate grid point whose predicted delay is lower than the delay of any of the reference grid points is obtained, the actual delay of each of the target candidate grid points is obtained by the high-level synthesis tool; If the predicted area of ​​the target candidate grid point is smaller than the area of ​​any of the reference grid points, the actual area of ​​each of the target candidate grid points is obtained by the high-level synthesis tool; If the target candidate grid point is obtained whose predicted delay is lower than the delay of any of the reference grid points and whose predicted area is smaller than the area of ​​any of the reference grid points, the actual delay and actual area of ​​each of the target candidate grid points are obtained through the high-level synthesis tool.

6. The grid traversal method based on dual gradient enhanced regressor according to claim 5, characterized in that: If the actual performance parameter is better than or partially better than any target grid point of the performance parameter, then one or more reference grid points are updated according to the target grid point, including: If the target grid point whose actual delay is lower than the delay of any of the reference grid points is obtained, then searching for the target reference grid point whose delay is higher than the actual delay, marking each of the target reference grid points as a failed grid point, and marking the target grid point as a reference grid point; If the target grid point whose actual area is smaller than the area of ​​any reference grid point is obtained, then a target reference grid point whose area is larger than the actual time delay is searched, and each of the target reference grid points is marked as a failed grid point, and the target grid point is marked as a reference grid point; If the target grid point whose actual delay and actual area are both smaller than the delay and area of ​​any reference grid point is obtained, then a target reference grid point whose delay is higher than the actual delay and whose area is larger than the actual area is searched, and each of the target reference grid points is marked as a failed grid point, and the target grid point is marked as a reference grid point.

7. The grid traversal method based on dual gradient enhanced regressor according to claim 6, characterized in that: After comprehensively comparing the performance prediction parameters of each of the candidate grid points with the performance parameters of each of the reference grid points, the method further includes: If the predicted delays of the candidate grid points are all higher than the delays of the reference grid points and the predicted areas of the candidate grid points are all larger than the areas of the reference grid points, the candidate grid points are marked as poor quality grid points.

8. The grid traversal method based on dual gradient enhanced regressor according to claim 7, characterized in that: After all spatial design grid points in the target area are traversed, the method further includes: Obtain abandoned grid points that are not predicted by the dual gradient boosting regressor for performance parameters, Obtaining the delay and area of ​​each of the inferior grid points and the abandoned grid points by the high-level synthesis tool; If the delay of any of the inferior grid points or the abandoned grid points is lower than the delay of the reference grid points and / or the area of ​​any of the inferior grid points or the abandoned grid points is smaller than the area of ​​the reference grid points, the inferior grid points or abandoned grid points whose delay is lower than the delay of the reference grid points and / or whose area is smaller than the area of ​​the reference grid points will be marked as reference grid points.

9. A grid traversal device based on a dual gradient enhanced regressor, characterized in that: The device comprises: An acquisition unit, used for acquiring a target space design set, wherein the target space design set is composed of a plurality of space design instruction sets; A generating unit, configured to generate a target area according to the target space design set, wherein the target area includes space design grid points corresponding to each of the space design instruction sets in a one-to-one manner; A selection unit, configured to select a sub-region to be traversed in the target region, and initialize and traverse each spatial design grid point in the sub-region to be traversed to obtain at least one reference grid point; A determination unit, configured to determine a next sub-region to be traversed according to each of the reference grid points, and traverse each of the spatial design grid points in the next sub-region to be traversed to obtain candidate grid points adjacent to each of the reference grid points; A prediction unit, used to predict the performance parameters of each candidate grid point by using a dual gradient enhancement regressor to obtain the performance prediction parameters of each candidate grid point; A first comprehensive comparison unit is used to comprehensively compare the performance prediction parameters of each of the candidate grid points with the performance parameters of each of the reference grid points, and if a target candidate grid point with a performance prediction parameter that is better than or partially better than any of the performance parameters is obtained, then the actual performance parameters of each of the target candidate grid points are obtained; A second comprehensive comparison unit is used to make a comprehensive comparison between the actual performance parameters of each of the target candidate grid points and the performance parameters of each of the reference grid points, and if a target grid point with an actual performance parameter that is better than or partially better than any of the performance parameters is obtained, one or more of the reference grid points are updated according to the target grid point; The iteration unit is used to repeat the above steps until all the spatial design grid points in the target area are traversed to find the reference grid point with the best performance parameters.

10. An electronic device, characterized in that: The electronic device includes one or more processors and one or more memories, wherein the one or more memories store at least one program code, and the at least one program code is loaded and executed by the one or more processors to implement the operations performed by the method according to any one of claims 1 to 8.

Images