Molecular electrostatic force determination system, method, electronic device, storage medium and product

By synchronously performing Fourier transform and data transpose when calculating electrostatic force, the delay caused by data loading and result back-passing and Fourier transform process separation is solved, and the electrostatic force calculation efficiency is improved.

CN118675632BActive Publication Date: 2024-11-22INSPUR SUZHOU INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411154470.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-11-22
Estimated Expiration
2044-08-21

AI Technical Summary

Technical Problem

In the existing electrostatic power calculation methods, the data loading and result back-passing processes are completely separated from the Fourier transform process, resulting in low electrostatic power calculation efficiency.

Method used

The multi-dimensional Fourier transform and data transpose are synchronized during the process of loading the electrostatic potential energy matrix, and the results are synchronized back to the global memory when performing the multi-dimensional Fourier inverse transformation, to avoid data transmission delay.

Benefits of technology

The calculation time of electrostatic force is shortened, the calculation efficiency is improved, and the problem of low electrostatic force calculation efficiency is solved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118675632B_ABST
    Figure CN118675632B_ABST
Patent Text Reader

Abstract

The embodiment of the present application provides a molecular electrostatic force determination system, method, electronic device, storage medium and product, the system comprising: a direct memory access engine for storing an electrostatic potential energy matrix and a coefficient matrix in a global memory, and sending a start calculation instruction to a computing unit; the computing unit is used to synchronously perform a multidimensional Fourier transform and data transposition on the electrostatic potential energy matrix during the process of loading the electrostatic potential energy matrix, perform data convolution, data transposition and multidimensional Fourier inverse transform on the calculation result of the multidimensional Fourier transform, obtain a target electrostatic force result, and synchronously transmit the target electrostatic force result back to the global memory during the process of performing the multidimensional inverse Fourier transform. The present application solves the problem of low electrostatic force calculation efficiency by synchronously performing a multidimensional Fourier transform and data transposition on the electrostatic potential energy matrix when loading the electrostatic potential energy matrix, and synchronously transmitting the target electrostatic force result back to the global memory when performing the multidimensional inverse Fourier transform.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a system, method, electronic device, storage medium and product for determining molecular electrostatic force. Background Art

[0002] Molecular dynamics simulation is a key tool for studying the behavior of molecular systems. It can reveal the chemical and physical properties of molecular systems. The core of molecular dynamics simulation is to accurately calculate the forces between molecules. Among them, electrostatic forces, as an important component of non-bonded interactions, are crucial to the accuracy of simulation results. However, as the number of molecules in the simulation system increases, the resources required to calculate electrostatic forces grow exponentially, becoming a bottleneck for computational efficiency. Therefore, the development of algorithms for efficiently calculating electrostatic forces has become a hot topic in molecular dynamics simulation research.

[0003] However, in existing methods for calculating electrostatic force, the data loading and result feedback processes are completely separated from the Fourier transform process, resulting in additional data transmission delays and, in turn, causing the problem of low efficiency in electrostatic force calculation. Summary of the invention

[0004] In view of this, the present application aims to propose a molecular electrostatic force determination system, method, electronic device, storage medium and product to solve the problem of low electrostatic force calculation efficiency caused by the complete separation of the data loading and result return process from the Fourier transform process.

[0005] According to a first aspect of the present application, a molecular electrostatic force determination system is provided, characterized in that the molecular electrostatic force determination system comprises a direct memory access engine, a global memory and a computing unit, wherein the direct memory access engine establishes a communication connection with the global memory and the computing unit respectively, and the global memory establishes a communication connection with the computing unit;

[0006] The direct memory access engine is used to obtain an electrostatic potential energy matrix and a coefficient matrix of a preset number of grid points, store the electrostatic potential energy matrix and the coefficient matrix in the global memory, and send a start calculation instruction to the calculation unit after completing storing the electrostatic potential energy matrix and the coefficient matrix in the global memory;

[0007] The calculation unit is used to load the electrostatic potential energy matrix upon receiving the start calculation instruction, and in the process of loading the electrostatic potential energy matrix, synchronously perform multidimensional Fourier transform and data transposition on the electrostatic potential energy matrix, and after completing the multidimensional Fourier transform and the data transposition, perform data convolution, data transposition and multidimensional Fourier inverse transform on the calculation result of the multidimensional Fourier transform to obtain a target electrostatic force result, and in the process of performing the multidimensional inverse Fourier transform, synchronously transmit the target electrostatic force result back to the global memory.

[0008] Optionally, the computing unit includes a data loading module and a Fourier transform group, wherein the data loading module establishes a communication connection with the Fourier transform group, the data loading module includes a loading counter, and the Fourier transform group includes an electrostatic potential energy local memory, a transformation counter, and a dimension counter;

[0009] The data loading module is used to obtain the electrostatic potential energy matrix from the global memory, and when the electrostatic potential energy matrix is ​​obtained, store the electrostatic potential energy matrix in the electrostatic potential energy local memory, and when the electrostatic potential energy matrix is ​​stored in the electrostatic potential energy local memory, update the value of the loading counter to obtain the value of the target loading counter, and send the value of the target loading counter to the Fourier transform group;

[0010] The Fourier transform group is used to perform X-dimensional Fourier transform, data transposition, Y-dimensional Fourier transform and Z-dimensional Fourier transform on the electrostatic potential energy matrix in sequence if the value of the target loading counter is received, the value of the dimension counter obtained in advance is a first preset value and the value of the transformation counter obtained in advance is less than the value of the target loading counter.

[0011] Optionally, the Fourier transform group includes a Fourier transform kernel, and the data loading module includes a return counter;

[0012] The Fourier transform core is used for, in the process of sequentially performing a Z-dimensional inverse Fourier transform, a Y-dimensional inverse Fourier transform, and an X-dimensional inverse Fourier transform, if the inverse Fourier transform of any dimension is completed, then the value of the dimension counter is increased by 1, and the value of the transform counter is cleared to 0; if the value of the dimension counter is a second preset value, then the X-dimensional inverse Fourier transform is performed; in the process of performing the X-dimensional inverse Fourier transform, each time the X-dimensional inverse Fourier transform is completed to obtain a target electrostatic force result, the value of the transform counter is increased by 1, the target electrostatic force result is returned to the global memory, the value of the return counter is increased by 1, and if the value of the return counter is less than the value of the transform counter, the calculation of the new target electrostatic force result is completed.

[0013] Optionally, the computing unit comprises a data transposition module, wherein the data transposition module establishes a communication connection with the Fourier transform core;

[0014] The data transposition module is used to obtain a calculation result of an X-dimensional Fourier transform through the Fourier transform kernel, and perform data transposition on the calculation result of the X-dimensional Fourier transform to obtain a first transposition result;

[0015] The Fourier transform kernel is used to perform Y-dimensional Fourier transform and Z-dimensional Fourier transform on the first transposed result in sequence to obtain a calculation result of the Z-dimensional Fourier transform.

[0016] Optionally, the computing unit includes a coefficient storage tuple and a convolution computing module, wherein the coefficient storage tuple includes a coefficient matrix local memory, the data loading module establishes communication connections with the global memory and the coefficient matrix local memory respectively, and the convolution computing module establishes communication connections with the coefficient matrix local memory and the electrostatic potential energy local memory respectively;

[0017] The data loading module is used to obtain the coefficient matrix from the global memory, and when the coefficient matrix is ​​obtained, store the coefficient matrix in the coefficient matrix local memory;

[0018] The convolution calculation module is used to obtain the coefficient matrix from the coefficient matrix local memory, obtain the calculation result of the Z-dimensional Fourier transform from the electrostatic potential energy local memory, and when the coefficient matrix and the calculation result of the Z-dimensional Fourier transform are obtained, perform data convolution on the calculation result of the Z-dimensional Fourier transform and the coefficient matrix to obtain the convolution calculation result.

[0019] Optionally, the data transposition module is used to obtain the calculation result of the Y-dimensional inverse Fourier transform through the Fourier transform kernel, and perform data transposition on the calculation result of the Y-dimensional inverse Fourier transform to obtain a second transposition result;

[0020] The Fourier transform kernel is used to perform the X-dimensional inverse Fourier transform on the second transposed result to obtain a target electrostatic force result.

[0021] Optionally, the data transposition module includes an input data selection module, a first-in first-out memory, a transposition processing module, a random access memory, and an output data selection module, wherein the input data selection module establishes a communication connection with the first-in first-out memory, the first-in first-out memory establishes a communication connection with the transposition processing module, the transposition processing module establishes a communication connection with the random access memory, and the random access memory establishes a communication connection with the output data selection module;

[0022] The input data selection module is used to determine the first data to be transposed and the second data to be transposed in sequence for the Fourier transform or the inverse Fourier transform of any dimension, and store the first data to be transposed and the second data to be transposed in the first-in-first-out memory;

[0023] The transposition processing module is used to obtain the first data to be transposed from the first-in-first-out memory, perform transposition processing on the first data to be transposed to obtain first transposed data, and store the first transposed data in the random access memory, and to obtain the second data to be transposed from the first-in-first-out memory, perform transposition processing on the second data to be transposed to obtain second transposed data, and store the second transposed data in the random access memory;

[0024] The output data selection module is used to determine the calculation result of data transposition from the random access memory according to a preset address.

[0025] Optionally, the Fourier transform kernel comprises a butterfly computing unit;

[0026] The butterfly calculation unit is used to determine the calculation result of the Fourier transform for any one Fourier transform in any dimensional Fourier transform, upon receiving the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor, according to the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor.

[0027] Optionally, the Fourier transform core includes an input address selection unit, a sequence real random access memory, a sequence imaginary random access memory, a rotation factor real read-only memory, and a rotation factor imaginary read-only memory, wherein the input address selection unit establishes a communication connection with the sequence real random access memory, the sequence imaginary random access memory, the rotation factor real read-only memory, and the rotation factor imaginary read-only memory respectively;

[0028] The input address selection unit is used to send the real part address of the target electrostatic potential energy matrix to the sequence real random access memory, send the imaginary part address of the target electrostatic potential energy matrix to the sequence imaginary random access memory, send the real part address of the target rotation factor to the real part read-only memory of the rotation factor, and send the imaginary part address of the target rotation factor to the imaginary part read-only memory of the rotation factor;

[0029] The sequence real random access memory is used to obtain the real part of the target electrostatic potential energy matrix and send the real part of the target electrostatic potential energy matrix to the butterfly computing unit;

[0030] The sequence imaginary part random access memory is used to obtain the imaginary part of the target electrostatic potential energy matrix, and send the imaginary part of the target electrostatic potential energy matrix to the butterfly calculation unit;

[0031] The real part read-only memory of the rotation factor is used to obtain the real part of the target rotation factor and send the real part of the target rotation factor to the butterfly computing unit;

[0032] The imaginary part read-only memory of the rotation factor is used to obtain the imaginary part of the target rotation factor and send the imaginary part of the target rotation factor to the butterfly calculation unit.

[0033] Optionally, the sequence real part random access memory is used to obtain the real part of the target electrostatic potential energy matrix according to the real part address of the target electrostatic potential energy matrix when receiving the real part address of the target electrostatic potential energy matrix;

[0034] The sequence imaginary part random access memory is used to obtain the imaginary part of the target electrostatic potential energy matrix according to the imaginary part address of the target electrostatic potential energy matrix when receiving the imaginary part address of the target electrostatic potential energy matrix;

[0035] The real part read-only memory of the rotation factor is used to obtain the real part of the target rotation factor according to the real part address of the target rotation factor when receiving the real part address of the target rotation factor;

[0036] The twiddle factor imaginary part read-only memory is used to obtain the target twiddle factor imaginary part according to the target twiddle factor imaginary part address when receiving the target twiddle factor imaginary part address.

[0037] Optionally, the Fourier transform kernel includes a data preprocessing unit;

[0038] The data preprocessing unit is used to split the electrostatic potential energy matrix to obtain the real part of the electrostatic potential energy matrix and the imaginary part of the electrostatic potential energy matrix, sort the real part of the electrostatic potential energy matrix and the imaginary part of the electrostatic potential energy matrix respectively to obtain the sorted real part of the electrostatic potential energy matrix and the sorted imaginary part of the electrostatic potential energy matrix, store the sorted real part of the electrostatic potential energy matrix in the sequence real part random access memory, and store the sorted imaginary part of the electrostatic potential energy matrix in the sequence imaginary part random access memory.

[0039] According to a second aspect of the present application, a method for determining molecular electrostatic force is provided, characterized in that the method comprises:

[0040] Obtaining an electrostatic potential energy matrix and a coefficient matrix of a preset number of grid points through a direct memory access engine, and storing the electrostatic potential energy matrix and the coefficient matrix in a global memory;

[0041] After the electrostatic potential energy matrix and the coefficient matrix are stored in the global memory, a start calculation instruction is sent to a calculation unit through the direct memory access engine;

[0042] When the calculation unit receives the start calculation instruction, loading the electrostatic potential energy matrix;

[0043] In the process of loading the electrostatic potential energy matrix, a multi-dimensional Fourier transform and a data transposition are synchronously performed on the electrostatic potential energy matrix;

[0044] When the multidimensional Fourier transform and the data transposition are completed, data convolution, data transposition and multidimensional inverse Fourier transform are performed on the calculation result of the multidimensional Fourier transform to obtain a target electrostatic force result;

[0045] During the process of performing the multi-dimensional inverse Fourier transform, the target electrostatic force result is synchronously transmitted back to the global memory.

[0046] According to a third aspect of the present application, a molecular electrostatic force determination device is provided, the molecular electrostatic force determination device comprising:

[0047] A data storage module, used for acquiring an electrostatic potential energy matrix and a coefficient matrix of a preset number of grid points through a direct memory access engine, and storing the electrostatic potential energy matrix and the coefficient matrix in a global memory;

[0048] A start calculation instruction sending module, used for sending a start calculation instruction to a calculation unit through the direct memory access engine when the electrostatic potential energy matrix and the coefficient matrix are stored in the global memory;

[0049] An electrostatic potential energy matrix loading module, used for loading the electrostatic potential energy matrix when the calculation unit receives the start calculation instruction;

[0050] A first synchronization module is used to synchronously perform multi-dimensional Fourier transform and data transposition on the electrostatic potential energy matrix during the process of loading the electrostatic potential energy matrix;

[0051] a target electrostatic force result determination module, configured to perform data convolution, data transposition and multidimensional Fourier inverse transformation on the calculation result of the multidimensional Fourier transform after completing the multidimensional Fourier transform and the data transposition, so as to obtain a target electrostatic force result;

[0052] The second synchronization module is used to synchronously transmit the target electrostatic force result back to the global memory during the process of performing the multi-dimensional inverse Fourier transform.

[0053] According to another aspect of the present application, there is also provided an electronic device, including:

[0054] processor;

[0055] a memory for storing instructions executable by the processor;

[0056] The processor is configured to execute the instructions to implement the steps of the molecular electrostatic force determination method as described above.

[0057] According to another aspect of the present application, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for determining the molecular electrostatic force as described above are implemented.

[0058] According to another aspect of the present application, a computer program product is also provided, comprising a computer program / instruction, wherein the computer program / instruction implements the steps of the above-mentioned method for determining molecular electrostatic force when executed by a processor.

[0059] The molecular electrostatic force determination system provided in the embodiment of the present application includes a direct memory access engine, a global memory and a computing unit, wherein the direct memory access engine establishes communication connections with the global memory and the computing unit respectively, and the global memory establishes a communication connection with the computing unit; the direct memory access engine is used to obtain the electrostatic potential energy matrix and the coefficient matrix of a preset grid point number, store the electrostatic potential energy matrix and the coefficient matrix in the global memory, and send a start calculation instruction to the computing unit after completing the storage of the electrostatic potential energy matrix and the coefficient matrix in the global memory; the computing unit is used to load the electrostatic potential energy matrix when receiving the start calculation instruction, and in the process of loading the electrostatic potential energy matrix, synchronously perform multi-dimensional Fourier transform and data transposition on the electrostatic potential energy matrix, and in the case of completing the multi-dimensional Fourier transform and data transposition, perform data convolution, data transposition and multi-dimensional Fourier inverse transform on the calculation result of the multi-dimensional Fourier transform to obtain the target electrostatic force result, and in the process of performing the multi-dimensional inverse Fourier transform, synchronously transmit the target electrostatic force result back to the global memory. The present application avoids additional data transmission delays caused by the complete separation of the data loading and result return processes from the Fourier transform process, thereby shortening the calculation time of the molecular electrostatic force, improving the calculation efficiency of the molecular electrostatic force, and solving the problem of low calculation efficiency of the electrostatic force.

[0060] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0062] Figure 1 One of the schematic diagrams of a molecular electrostatic force determination system provided in an embodiment of the present application;

[0063] Figure 2 A second schematic diagram of a molecular electrostatic force determination system provided in an embodiment of the present application;

[0064] Figure 3 A schematic diagram of a molecular electrostatic force calculation process provided in an embodiment of the present application;

[0065] Figure 4 A schematic diagram of synchronous processing of a data loading module provided in an embodiment of the present application;

[0066] Figure 5 A schematic diagram of the architecture of a data transposition module provided in an embodiment of the present application;

[0067] Figure 6 A schematic diagram of a plane data provided in an embodiment of the present application;

[0068] Figure 7 One of the schematic diagrams of a data transposition process provided in an embodiment of the present application;

[0069] Figure 8 A second schematic diagram of a data transposition process provided in an embodiment of the present application;

[0070] Fig. 9 One of the schematic diagrams of calculation results of data transposition provided in an embodiment of the present application;

[0071] Fig.10 A second schematic diagram of calculation results of data transposition provided in an embodiment of the present application;

[0072] Fig.11 A schematic diagram of a butterfly computing network provided in an embodiment of the present application;

[0073] Fig.12 A schematic diagram of the architecture of a Fourier transform kernel provided in an embodiment of the present application;

[0074] Fig.13 is a flow chart of the steps of a method for determining molecular electrostatic force provided in an embodiment of the present application;

[0075] Fig.14 is a device block diagram of a molecular electrostatic force determination device provided in an embodiment of the present application;

[0076] Fig.15 It is a schematic diagram of a communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0077] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, each embodiment of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in each embodiment of the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not constitute any limitation, and the various embodiments can be combined with each other and referenced to each other without contradiction.

[0078] It should be noted that in the embodiments of the present application, the PME (Particle Mesh Ewald) method is a widely accepted and efficient algorithm for calculating electrostatic forces. It decomposes the electrostatic potential energy into three parts: real space, Fourier space, and energy correction term. The real space part is quickly converged by adjusting the Ewald coefficient, allowing it to be truncated. The energy correction term itself does not require much computation. Therefore, calculating the electrostatic force in Fourier space becomes the most time-consuming part of the PME algorithm. In order to speed up this process, heterogeneous computing systems, that is, computing systems that combine a central processing unit, a graphics processing unit, or a field programmable gate array, are introduced to improve computing efficiency. In a heterogeneous computing system, the central processing unit is responsible for task management and scheduling, while the graphics processing unit or field programmable gate array undertakes specific computing tasks, especially the calculation of electrostatic force in Fourier space, thereby significantly improving the computing speed.

[0079] However, in the existing method of calculating electrostatic force using the PME algorithm, the data loading and result feedback process is completely separated from the Fourier transform process, resulting in additional data transmission delays, which in turn causes the problem of low efficiency in electrostatic force calculation.

[0080] Therefore, in order to solve the problem of low electrostatic force calculation efficiency caused by the complete separation of the data loading and result return process from the Fourier transform process, the present application synchronously performs multi-dimensional Fourier transform and data transposition on the electrostatic potential energy matrix during the process of loading the electrostatic potential energy matrix, and synchronously returns the target electrostatic force result to the global memory during the process of performing the multi-dimensional inverse Fourier transform, thereby avoiding additional data transmission delays caused by the complete separation of the data loading and result return process from the Fourier transform process, thereby shortening the calculation time of the molecular electrostatic force, improving the calculation efficiency of the molecular electrostatic force, and solving the problem of low electrostatic force calculation efficiency.

[0081] Reference Figure 1 , showing a schematic diagram of a molecular electrostatic force determination system provided in an embodiment of the present application, the molecular electrostatic force determination system includes a direct memory access engine, a global memory, and a computing unit, wherein the direct memory access engine establishes communication connections with the global memory and the computing unit respectively, and the global memory establishes a communication connection with the computing unit.

[0082] The direct memory access engine is used to obtain the electrostatic potential energy matrix and the coefficient matrix of the preset grid points, store the electrostatic potential energy matrix and the coefficient matrix in the global memory, and send a start calculation instruction to the computing unit after completing the storage of the electrostatic potential energy matrix and the coefficient matrix in the global memory.

[0083] It should be noted that in the embodiments of the present application, Figure 2 As shown, the molecular electrostatic force determination system also includes a host. The direct memory access engine, the global memory and the computing unit together constitute a field programmable gate array accelerator. The field programmable gate array accelerator is connected to the host through a high-speed serial computer expansion bus interface.

[0084] The host stores the electrostatic potential energy matrix and coefficient matrix of the preset grid points acquired in advance into the global memory through the direct memory access engine.

[0085] After storing the electrostatic potential energy matrix and the coefficient matrix of the preset grid points in the global memory, the host sends a start calculation instruction to the computing unit through the direct memory access engine.

[0086] The calculation unit is used to load the electrostatic potential energy matrix upon receiving a start calculation instruction. In the process of loading the electrostatic potential energy matrix, the electrostatic potential energy matrix is ​​synchronously subjected to multi-dimensional Fourier transform and data transposition. After completing the multi-dimensional Fourier transform and data transposition, the calculation result of the multi-dimensional Fourier transform is subjected to data convolution, data transposition and multi-dimensional inverse Fourier transform to obtain a target electrostatic force result. In the process of performing the multi-dimensional inverse Fourier transform, the target electrostatic force result is synchronously transmitted back to the global memory.

[0087] It should be noted that in the embodiments of the present application, the entire calculation process of molecular electrostatic force includes multidimensional Fourier transform, data transposition, data convolution and multidimensional inverse Fourier transform.

[0088] When the computing unit receives a start computing instruction sent by the direct memory access engine, the electrostatic potential energy matrix is ​​loaded.

[0089] In order to reduce the time of data loading and result transmission, and thus shorten the calculation time of molecular electrostatic forces, such as Figure 3 As shown, the present application performs loading of the electrostatic potential energy matrix and X-dimensional Fourier transform simultaneously, and performs returning of the calculation result and X-dimensional inverse Fourier transform simultaneously.

[0090] Specifically, in the process of loading the electrostatic potential energy matrix, the electrostatic potential energy matrix is ​​synchronously subjected to multidimensional Fourier transform and data transposition. When the multidimensional Fourier transform and data transposition are completed, the calculation result of the multidimensional Fourier transform is subjected to data convolution, data transposition and multidimensional Fourier inverse transform to obtain the target electrostatic force result.

[0091] After obtaining the target electrostatic force result, the target electrostatic force result is stored in the global memory. The host can obtain the target electrostatic force result from the global memory through the direct memory access engine.

[0092] The present application avoids additional data transmission delays caused by the complete separation of the data loading and result return processes from the Fourier transform process, thereby shortening the calculation time of the molecular electrostatic force, improving the calculation efficiency of the molecular electrostatic force, and solving the problem of low calculation efficiency of the electrostatic force.

[0093] Further, in the embodiments of the present application, Figure 2 As shown, the computing unit further includes a data loading module and a Fourier transform group, and the data loading module establishes a communication connection with the Fourier transform group, wherein, Figure 4 As shown, the data loading module includes a loading counter, and the Fourier transform group also includes an electrostatic potential energy local memory, a transformation counter and a dimension counter.

[0094] The data loading module is used to obtain the electrostatic potential energy matrix from the global memory. When the electrostatic potential energy matrix is ​​obtained, the electrostatic potential energy matrix is ​​stored in the electrostatic potential energy local memory. When the electrostatic potential energy matrix is ​​stored in the electrostatic potential energy local memory, the value of the loading counter is updated to obtain the value of the target loading counter, and the value of the target loading counter is sent to the Fourier transform group.

[0095] The Fourier transform group is used to perform X-dimensional Fourier transform, data transposition, Y-dimensional Fourier transform and Z-dimensional Fourier transform on the electrostatic potential energy matrix in sequence if the value of the target loading counter is received, the value of the pre-acquired dimension counter is a first preset value and the value of the pre-acquired transformation counter is less than the value of the target loading counter.

[0096] It should be noted that in the embodiment of the present application, the data interface bit width of the global memory is 256 bits, the bit width of each data is 64 bits, and 4 sequences of data can be transmitted simultaneously, so the 4 Fourier transform cores and 4 electrostatic potential energy local memories are regarded as a Fourier transform group.

[0097] like Figure 4 As shown, the data loading module includes several groups of loading counters and return counters, and each group of loading counters and return counters corresponds to a Fourier transform group.

[0098] The data of one Fourier transform group is loaded at a time through the data loading module. Specifically, the electrostatic potential energy matrix of one Fourier transform group is obtained from the global memory, and the electrostatic potential energy matrix of one Fourier transform group is stored in the electrostatic potential energy local memory. When the electrostatic potential energy matrix of one Fourier transform group is stored in the electrostatic potential energy local memory, the value of the load counter is updated, that is, the value of the load counter is increased by 1, the value of the target load counter is obtained, and the value of the target load counter is sent to the Fourier transform group.

[0099] like Figure 4 As shown, a Fourier transform group includes an electrostatic potential energy local memory, a transform counter, and a dimension counter. If the value of the dimension counter is 0, it means that a Fourier transform of dimension X is performed; if the value of the dimension counter is 5, it means that an inverse Fourier transform of dimension X is performed. Each time a Fourier transform or inverse Fourier transform of a dimension is completed, the value of the dimension counter is increased by 1, and the value of the transform counter is cleared to 0. For a Fourier transform or inverse Fourier transform of any dimension, each time a Fourier transform or inverse Fourier transform is completed, the value of the transform counter is increased by 1.

[0100] If the Fourier transform group receives the value of the target load counter sent by the data loading module, the value of the pre-acquired dimension counter is the first preset value (0) and the value of the pre-acquired transformation counter is less than the value of the target load counter, then the Fourier transform core in the Fourier transform group can obtain the electrostatic potential energy matrix from the electrostatic potential energy local memory, and sequentially perform X-dimensional Fourier transform, data transposition, Y-dimensional Fourier transform and Z-dimensional Fourier transform on the electrostatic potential energy matrix. After completing any one Fourier transform in any dimension, the value of the transformation counter is increased by 1, so as to realize the synchronization of data loading and Fourier transform.

[0101] The present application avoids the additional data transmission delay caused by the complete separation of the data loading process and the Fourier transform process by synchronously performing Fourier transform on the electrostatic matrix during the process of loading the electrostatic potential energy matrix, thereby shortening the calculation time of the molecular electrostatic force, thereby shortening the calculation time of the molecular electrostatic force, improving the calculation efficiency of the molecular electrostatic force, and solving the problem of low calculation efficiency of the electrostatic force.

[0102] Further, in the embodiments of the present application, Figure 2 As shown, the Fourier transform group also includes a Fourier transform kernel. Figure 4 As shown, the data loading module also includes a return counter.

[0103] The Fourier transform core is used to, in the process of sequentially performing a Z-dimensional inverse Fourier transform, a Y-dimensional inverse Fourier transform, and an X-dimensional inverse Fourier transform, if the Fourier transform of any dimension is completed, then the value of the dimension counter is increased by 1, and the value of the transform counter is cleared to 0; if the value of the dimension counter is a second preset value, then an X-dimensional inverse Fourier transform is performed; in the process of performing an X-dimensional inverse Fourier transform, each time an X-dimensional inverse Fourier transform is completed to obtain a target electrostatic force result, the value of the transform counter is increased by 1, the target electrostatic force result is returned to the global memory, the value of the return counter is increased by 1, and if the value of the return counter is less than the value of the transform counter, the calculation of the new target electrostatic force result is completed.

[0104] It should be noted that in the embodiment of the present application, in the process of sequentially performing the Z-dimensional Fourier inverse transform, the Y-dimensional Fourier inverse transform, and the X-dimensional Fourier inverse transform through the Fourier transform kernel, if the Fourier inverse transform of any dimension is completed, the value of the dimension counter is increased by 1, and the value of the transform counter is cleared to 0. If the value of the dimension counter is the second preset value (5), the X-dimensional Fourier inverse transform is performed.

[0105] In the process of performing an X-dimensional inverse Fourier transform, each time an X-dimensional inverse Fourier transform is completed to obtain a target electrostatic force result, the value of the transformation counter is increased by 1. At the same time, the Fourier transform group sends the target electrostatic force result to the data loading module, and the data loading module returns the target electrostatic force result to the global memory, and increases the value of the return counter by 1. If the value of the return counter is less than the value of the transform counter, the new target electrostatic force result is calculated. Each time an X-dimensional inverse Fourier transform is completed to obtain a target electrostatic force result, the target electrostatic force result is returned to the global memory, and the value of the return counter is increased by 1, so as to achieve the synchronization of the inverse Fourier transform and the result return. After the calculation is completed, the values ​​of all counters are cleared.

[0106] The present application avoids the additional data transmission delay caused by the complete separation of the result return and the inverse Fourier transform process by synchronously returning the calculation result of the inverse Fourier transform to the global memory during the process of performing the X-dimensional inverse Fourier transform, thereby shortening the calculation time of the molecular electrostatic force, thereby shortening the calculation time of the molecular electrostatic force, improving the calculation efficiency of the molecular electrostatic force, and solving the problem of low calculation efficiency of the electrostatic force.

[0107] Further, in the embodiments of the present application, Figure 2 As shown, the computing unit includes a data transposition module, wherein the data transposition module establishes a communication connection with the Fourier transform core.

[0108] The data transposition module is used to obtain the calculation result of the X-dimensional Fourier transform through the Fourier transform kernel, and perform data transposition on the calculation result of the X-dimensional Fourier transform to obtain a first transposition result.

[0109] The Fourier transform kernel is used to perform Y-dimensional Fourier transform and Z-dimensional Fourier transform on the first transposed result in sequence to obtain a calculation result of the Z-dimensional Fourier transform.

[0110] It should be noted that, in the embodiment of the present application, the electrostatic potential energy matrix is ​​sequentially subjected to an X-dimensional Fourier transform, data transposition, a Y-dimensional Fourier transform, and a Z-dimensional Fourier transform. Specifically, the electrostatic potential energy matrix is ​​subjected to an X-dimensional Fourier transform by a Fourier transform core, and the calculation result of the X-dimensional Fourier transform is obtained, and the calculation result of the X-dimensional Fourier transform is stored in the local memory of the electrostatic potential energy. The calculation result of the X-dimensional Fourier transform is obtained from the local memory of the electrostatic potential energy by a data transposition module, and the calculation result of the X-dimensional Fourier transform is subjected to data transposition to obtain a first transposition result, and the first transposition result is stored in the local memory of the electrostatic potential energy. The first transposition result is obtained from the local memory of the electrostatic potential energy by a Fourier transform core, and a Y-dimensional Fourier transform is performed on the first transposition result to obtain the calculation result of the Y-dimensional Fourier transform. The calculation result of the Y-dimensional Fourier transform is subjected to a Z-dimensional Fourier transform by a Fourier transform core, and the calculation result of the Z-dimensional Fourier transform is obtained.

[0111] The present application avoids the additional data transmission delay caused by the complete separation of the data loading process and the Fourier transform process by synchronously and sequentially performing X-dimensional Fourier transform, data transposition, Y-dimensional Fourier transform and Z-dimensional Fourier transform on the electrostatic potential energy matrix during the data loading process, thereby shortening the calculation time of the molecular electrostatic force, thereby shortening the calculation time of the molecular electrostatic force, improving the calculation efficiency of the molecular electrostatic force, and solving the problem of low calculation efficiency of the electrostatic force.

[0112] Further, in the embodiments of the present application, Figure 2 As shown, the computing unit includes a coefficient storage tuple and a convolution computing module, wherein the coefficient storage tuple includes a coefficient matrix local memory, the data loading module establishes communication connections with the global memory and the coefficient matrix local memory respectively, and the convolution computing module establishes communication connections with the coefficient matrix local memory and the electrostatic potential energy local memory respectively.

[0113] The data loading module is used to obtain the coefficient matrix from the global memory, and when the coefficient matrix is ​​obtained, store the coefficient matrix in the coefficient matrix local memory.

[0114] The convolution calculation module is used to obtain the coefficient matrix from the coefficient matrix local memory, obtain the calculation result of the Z-dimensional Fourier transform from the electrostatic potential energy local memory, and when the coefficient matrix and the calculation result of the Z-dimensional Fourier transform are obtained, perform data convolution on the calculation result of the Z-dimensional Fourier transform and the coefficient matrix to obtain the convolution calculation result.

[0115] It should be noted that in the embodiment of the present application, the coefficient matrix is ​​loaded, specifically, the coefficient matrix is ​​obtained from the global memory through the data loading module, and when the coefficient matrix is ​​obtained, the coefficient matrix is ​​stored in the coefficient matrix local memory through the data loading module.

[0116] The coefficient matrix corresponding to the Fourier transform group is obtained from the coefficient matrix local memory through the convolution calculation module, and the calculation result of the Z-dimensional Fourier transform is obtained from the electrostatic potential energy local memory through the convolution calculation module. When the convolution calculation module obtains the calculation result of the coefficient matrix and the Z-dimensional Fourier transform, the calculation result of the Z-dimensional Fourier transform and the coefficient matrix are convolved to obtain the convolution calculation result.

[0117] This application performs data convolution on the calculation results and coefficient matrix of the Z-dimensional Fourier transform through a convolution calculation module to obtain the convolution calculation results, which provides a data basis for subsequent multi-dimensional inverse Fourier transform and data transposition, thereby improving the efficiency and accuracy of molecular electrostatic force calculations.

[0118] Furthermore, in an embodiment of the present application, the data transposition module is used to obtain the calculation result of the Y-dimensional inverse Fourier transform through the Fourier transform kernel, and perform data transposition on the calculation result of the Y-dimensional inverse Fourier transform to obtain a second transposed result.

[0119] The Fourier transform kernel is used to perform an X-dimensional inverse Fourier transform on the second transposed result to obtain a target electrostatic force result.

[0120] It should be noted that in an embodiment of the present application, the convolution calculation result is subjected to a Z-dimensional inverse Fourier transform by a Fourier transform core to obtain the calculation result of the Z-dimensional inverse Fourier transform. The calculation result of the Z-dimensional inverse Fourier transform is subjected to a Y-dimensional inverse Fourier transform by a Fourier transform core to obtain the calculation result of the Y-dimensional inverse Fourier transform, and the calculation result of the Y-dimensional inverse Fourier transform is stored in the local memory of the electrostatic potential energy. The calculation result of the Y-dimensional inverse Fourier transform is obtained from the local memory of the electrostatic potential energy by a data transposition module, and the calculation result of the Y-dimensional inverse Fourier transform is subjected to a data transposition to obtain a second transposed result. The second transposed result is subjected to an X-dimensional inverse Fourier transform by a Fourier transform core to obtain a target electrostatic force result.

[0121] This application uses Fourier transform to check the convolution calculation results and then performs Z-dimensional inverse Fourier transform, Y-dimensional inverse Fourier transform and data transposition in sequence to obtain the target electrostatic force result. It can not only utilize the high efficiency of the fast Fourier transform algorithm to significantly reduce the calculation complexity and time, but also allow flexible conversion of data between different dimensions, which helps to optimize the data processing process and make the processing of data in different dimensions more efficient and orderly.

[0122] Further, in the embodiments of the present application, Figure 5As shown, the data transposition module includes an input data selection module, a first-in first-out memory, a transposition processing module, a random access memory and an output data selection module, wherein the input data selection module establishes a communication connection with the first-in first-out memory, the first-in first-out memory establishes a communication connection with the transposition processing module, the transposition processing module establishes a communication connection with the random access memory, and the random access memory establishes a communication connection with the output data selection module.

[0123] The input data selection module is used to determine the first data to be transposed and the second data to be transposed in sequence for Fourier transform or inverse Fourier transform of any dimension, and store the first data to be transposed and the second data to be transposed in a first-in-first-out memory.

[0124] The transposition processing module is used to obtain the first data to be transposed from the first-in first-out memory, perform transposition processing on the first data to be transposed to obtain the first transposed data, and store the first transposed data in the random access memory, and to obtain the second data to be transposed from the first-in first-out memory, perform transposition processing on the second data to be transposed to obtain the second transposed data, and store the second transposed data in the random access memory.

[0125] The output data selection module is used to determine the calculation result of data transposition from the random access memory according to the preset address.

[0126] It should be noted that in the embodiment of the present application, the three-dimensional Fourier transform includes an X-dimensional Fourier transform, a Y-dimensional Fourier transform, and a Z-dimensional Fourier transform, and the three-dimensional inverse Fourier transform includes an X-dimensional inverse Fourier transform, a Y-dimensional inverse Fourier transform, and a Z-dimensional inverse Fourier transform. The data transposition module needs to transpose the data between the X-dimensional Fourier transform and the Y-dimensional Fourier transform, and between the Y-dimensional Fourier transform and the X-dimensional Fourier transform. The data transposition module can greatly reduce the consumption of hardware resources in the calculation process, thereby solving the congestion problem of layout and routing during the compilation process due to the increase in the number of grid points, and the problem of incorrect calculation results due to poor timing after the compilation is completed.

[0127] For data transposition between X-dimensional Fourier transform and Y-dimensional Fourier transform, or data transposition between Y-dimensional inverse Fourier transform and X-dimensional inverse Fourier transform, the specific implementation process of data transposition is as follows: Figure 5As shown, the first data to be transposed and the second data to be transposed are determined in sequence by the input data selection module, and the first data to be transposed and the second data to be transposed are stored in the first-in-first-out memory. The first data to be transposed is obtained from the first-in-first-out memory by the transposition processing module, and the first data to be transposed is transposed to obtain the first transposed data, and the first transposed data is stored in the random access memory. The second data to be transposed is obtained from the first-in-first-out memory by the transposition processing module, and the second data to be transposed is transposed to obtain the second transposed data, and the second transposed data is stored in the random access memory. The calculation result of the data transposition is determined from the random access memory according to the preset address by the output data selection module.

[0128] Take the data transposition between X-dimensional Fourier transform and Y-dimensional Fourier transform as an example. If the number of grid points is 8, the three-dimensional grid is 8×8×8, and a plane is 8×8, it is necessary to perform data transposition on the calculation results of the X-dimensional Fourier transform of the two Fourier transform groups respectively, and then complete the data transformation of a plane based on the calculation results of the data transposition of the two Fourier transform groups. Figure 6 As shown, for X-dimensional Fourier transform, each Fourier transform kernel processes a horizontal sequence each time, i.e., 1-1, 1-2, 1-3, ..., 1-8, while for Y-dimensional Fourier transform, each Fourier transform kernel processes a vertical sequence each time, i.e., 1-1, 2-1, 3-1, ..., 8-1. The function of the data transposition module is to reorganize the result after the X-dimensional Fourier transform, combine the horizontal sequence into a vertical sequence and store it in the electrostatic potential energy local memory as the initial data of the Y-dimensional Fourier transform.

[0129] For the X-dimensional Fourier transform, each Fourier transform group performs the Fourier transform asynchronously, so the output of the calculation results of the X-dimensional Fourier transform of the Fourier transform group is serial. Due to this characteristic, the data transposition module can perform transposition processing on the calculation results of the X-dimensional Fourier transform of each Fourier transform group in turn. In the embodiment of the present application, the calculation results of the X-dimensional Fourier transform of the first Fourier transform group are transposed as shown in 7, each line is a calculation result of an X-dimensional Fourier transform, and the calculation result of an X-dimensional Fourier transform includes 8 data, and the calculation results of 4 X-dimensional Fourier transforms are shown in FIG. Figure 7 As shown above the arrow. The calculation results of the four X-dimensional Fourier transforms are stored in four first-in first-out memories, and then the calculation results of the four X-dimensional Fourier transforms are read out from the four first-in first-out memories at intervals of one clock cycle, and then the calculation results of the four X-dimensional Fourier transforms are recombined and stored in four random access memories. The calculation results of the recombined four X-dimensional Fourier transforms are shown in Figure 7As shown below the arrow, the addresses of the eight data in each reassembled X-dimensional Fourier transform calculation result in the random access memory are 0, 1, 2, ..., 7.

[0130] The calculation result of the X-dimensional Fourier transform of the second Fourier transform group is transposed as shown in 8. The specific implementation process of data transposition refers to the process of transposing the calculation result of the X-dimensional Fourier transform of the first Fourier transform group. The addresses of the eight data in each reassembled calculation result of the X-dimensional Fourier transform in the random access memory are 8, 9, 10, ..., 15.

[0131] After completing the data transposition of the calculation results of the X-dimensional Fourier transform of the two Fourier transform groups, the Y-dimensional Fourier transform initial data of the two Fourier transform groups are taken out from the 4 random access memories in turn and stored in the electrostatic potential energy local memory. For the first Fourier transform group, the data of addresses 0, 1, 2, 3, 8, 9, 10, and 11 are taken out in turn to form Fig. 9 The four 8-point sequences shown are stored in the four electrostatic potential energy local memories of the first Fourier transform group. For the second Fourier transform group, the data of addresses 4, 5, 6, 7, 12, 13, 14, and 15 are taken out in turn to form the following: Fig.10 The four 8-point sequences shown are respectively stored in the four electrostatic potential energy local memories of the second Fourier transform group. At this time, a planar data transformation is completed based on the calculation results of the data transposition of the two Fourier transform groups.

[0132] When the number of grid points increases, such as increasing to 64 points, the number of Fourier transform groups also increases to 16 groups synchronously, and the architecture of the data transposition module can still remain unchanged, completing the data transposition of 16 groups of 64 sequences in series, and the hardware resource usage of the data transposition module will not increase sharply with the increase in the number of grid points, thereby avoiding the layout and routing congestion, timing violation, calculation error and other problems caused by excessive consumption of hardware resources of the data transposition module. However, if the hardware resources are sufficient, the parallelism of the data transposition processing can be increased by increasing the number of transposition processing modules, thereby shortening the data transposition processing time. In short, under the architecture of the data transposition module, the hardware resource usage and data transposition time can be flexibly balanced to meet the needs of different hardware resources.

[0133] The present application uses a data transposition module, including an input data selection module, a first-in first-out memory, a transposition processing module, and a random access memory, to perform data transposition between an X-dimensional Fourier transform and a Y-dimensional Fourier transform, or between a Y-dimensional Fourier inverse transform and an X-dimensional Fourier inverse transform, thereby avoiding problems such as layout and routing congestion, timing violations, and calculation errors caused by excessive consumption of hardware resources of the data transposition module. In addition, the present application can increase the parallelism of the data transposition processing by increasing the number of transposition processing modules, thereby shortening the time of the data transposition processing.

[0134] Furthermore, in an embodiment of the present application, the Fourier transform kernel includes a butterfly computing unit.

[0135] The butterfly calculation unit is used to determine the calculation result of the Fourier transform for any one Fourier transform in any dimensional Fourier transform, based on the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor, upon receiving the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor.

[0136] It should be noted that, in the embodiment of the present application, for a sequence of length 8, the butterfly computing network of the fast Fourier transform architecture of the base 2 algorithm is as follows: Fig.11 As shown in the figure, the entire fast Fourier transform calculation process requires log28=3 levels, each level requires 8 / 2=4 butterfly computing units, and a total of 12 butterfly computing units are required. Each butterfly computing unit completes the calculation of two data. As the sequence length increases, the number of butterfly computing units required increases rapidly, thus consuming a large amount of hardware resources.

[0137] In order to reduce the consumption of hardware resources, this application proposes a serial fast Fourier transform architecture as the Fourier transform core. There is only one butterfly computing unit under this architecture, which completes the calculation of each group of data at each level in sequence. The architecture diagram is shown in Fig.12As shown. The serial fast Fourier transform architecture includes two random access memories, namely, a sequence real random access memory and a sequence imaginary random access memory, including two read-only memories, namely, a rotation factor real read-only memory and a rotation factor imaginary read-only memory, a butterfly calculation unit, a data preprocessing unit, two data selection units, an input address selection unit, an output address selection unit and a data post-processing unit. Among them, the data selection module is used to select whether to store the input data in the random access memory or to store the result calculated by the butterfly calculation unit in the random access memory. After completing a Fourier transform through the serial fast Fourier transform architecture, the data post-processing unit takes out the real part of the Fourier transform calculation result from the sequence real random access memory, takes out the imaginary part of the Fourier transform calculation result from the sequence imaginary random access memory, combines the real part of the Fourier transform calculation result and the imaginary part of the Fourier transform calculation result, and obtains the calculation result of this Fourier transform.

[0138] For any one Fourier transform in any dimensional Fourier transform, when the butterfly computing unit receives the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor, the calculation result of the Fourier transform is determined according to the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor.

[0139] The present application uses a serial fast Fourier transform architecture as the Fourier transform core and performs Fourier transform on the electrostatic potential energy matrix, thereby reducing the consumption of hardware resources, thereby solving problems such as layout and routing congestion, timing violations, and calculation errors caused by excessive consumption of hardware resources.

[0140] Further, in the embodiments of the present application, Fig.12 As shown, the Fourier transform core includes an input address selection unit, a sequence real random access memory, a sequence imaginary random access memory, a rotation factor real read-only memory, and a rotation factor imaginary read-only memory, wherein the input address selection unit establishes communication connections with the sequence real random access memory, the sequence imaginary random access memory, the rotation factor real read-only memory, and the rotation factor imaginary read-only memory, respectively.

[0141] The input address selection unit is used to send the real part address of the target electrostatic potential energy matrix to the sequence real part random access memory, send the imaginary part address of the target electrostatic potential energy matrix to the sequence imaginary part random access memory, send the real part address of the target rotation factor to the real part read-only memory of the rotation factor, and send the imaginary part address of the target rotation factor to the imaginary part read-only memory of the rotation factor.

[0142] The sequence real part random access memory is used to obtain the real part of the target electrostatic potential energy matrix and send the real part of the target electrostatic potential energy matrix to the butterfly computing unit.

[0143] The sequence imaginary part random access memory is used to obtain the imaginary part of the target electrostatic potential energy matrix and send the imaginary part of the target electrostatic potential energy matrix to the butterfly calculation unit.

[0144] The real part read-only memory of the rotation factor is used to obtain the real part of the target rotation factor and send the real part of the target rotation factor to the butterfly computing unit.

[0145] The imaginary part read-only memory of the rotation factor is used to obtain the imaginary part of the target rotation factor and send the imaginary part of the target rotation factor to the butterfly calculation unit.

[0146] It should be noted that in an embodiment of the present application, the input address selection unit sends the real part address of the target electrostatic potential energy matrix to the sequence real random access memory, sends the imaginary part address of the target electrostatic potential energy matrix to the sequence imaginary random access memory, sends the real part address of the target rotation factor to the real part read-only memory of the rotation factor, and sends the imaginary part address of the target rotation factor to the imaginary part read-only memory of the rotation factor. The real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor, and the imaginary part of the target rotation factor are obtained through the sequence real random access memory, the sequence imaginary random access memory, the real part of the rotation factor read-only memory, and the imaginary part of the rotation factor, respectively, and the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor, and the imaginary part of the target rotation factor are sent to the butterfly calculation unit.

[0147] Among them, if the input data is a complex sequence of length 8, in the first, second and third level calculations, the address transformations of the sequence real random access memory and the sequence imaginary random access memory are [0,1,2,3,4,5,6, 7], [0,2,4,6,1,3,5,7], [0,4,1,5,2,6,3,7] respectively. And the address transformations of the rotation factor real read-only memory and the rotation factor imaginary read-only memory are [0,0,0,0,0,0,0,0], [0,0,0,0,2,2,2,2], [0,0,1,1,2,2,3,3] respectively.

[0148] The present application accurately sends addresses to each memory through an input address selection unit. When each memory receives the address, it can obtain data according to the address, thereby reducing the delay in data access, improving the efficiency of data acquisition, and further improving the calculation efficiency of Fourier transform.

[0149] Further, in an embodiment of the present application, the sequence real part random access memory is used to obtain the real part of the target electrostatic potential energy matrix according to the real part address of the target electrostatic potential energy matrix when the real part address of the target electrostatic potential energy matrix is ​​received.

[0150] The serial imaginary part random access memory is used to obtain the imaginary part of the target electrostatic potential energy matrix according to the imaginary part address of the target electrostatic potential energy matrix when the imaginary part address of the target electrostatic potential energy matrix is ​​received.

[0151] The real part read-only memory of the twiddle factor is used to obtain the real part of the target twiddle factor according to the real part address of the target twiddle factor when the real part address of the target twiddle factor is received.

[0152] The twiddle factor imaginary part read-only memory is used to obtain the target twiddle factor imaginary part according to the target twiddle factor imaginary part address when receiving the target twiddle factor imaginary part address.

[0153] It should be noted that in the embodiment of the present application, the sequence real random access memory receives the real address of the target electrostatic potential energy matrix, and obtains the real part of the target electrostatic potential energy matrix according to the real address of the target electrostatic potential energy matrix. The sequence imaginary random access memory receives the imaginary address of the target electrostatic potential energy matrix, and obtains the imaginary part of the target electrostatic potential energy matrix according to the imaginary address of the target electrostatic potential energy matrix. The real part read-only memory of the rotation factor receives the real part address of the target rotation factor, and obtains the real part of the target rotation factor according to the real part address of the target rotation factor. The imaginary part read-only memory of the rotation factor receives the imaginary part address of the target rotation factor, and obtains the imaginary part of the target rotation factor according to the imaginary part address of the target rotation factor.

[0154] The present application can reduce the delay of data access and improve the efficiency of data access for Fourier transform calculations that require frequent data access through precise address transfer and memory access. In addition, the present application can effectively manage memory resources, reduce memory fragmentation, and improve memory usage efficiency by using a sequence real random access memory and a sequence imaginary random access memory to store the real part and imaginary part of the data respectively.

[0155] Further, in the embodiments of the present application, Fig.12 As shown, the Fourier transform kernel includes a data preprocessing unit.

[0156] The data preprocessing unit is used to split the electrostatic potential energy matrix to obtain the real part of the electrostatic potential energy matrix and the imaginary part of the electrostatic potential energy matrix, sort the real part of the electrostatic potential energy matrix and the imaginary part of the electrostatic potential energy matrix respectively to obtain the sorted real part of the electrostatic potential energy matrix and the sorted imaginary part of the electrostatic potential energy matrix, store the sorted real part of the electrostatic potential energy matrix in a sequence real part random access memory, and store the sorted imaginary part of the electrostatic potential energy matrix in a sequence imaginary part random access memory.

[0157] It should be noted that, in the embodiment of the present application, the electrostatic potential energy matrix is ​​a complex number.

[0158] The electrostatic potential energy matrix is ​​split into a real part of the electrostatic potential energy matrix and an imaginary part of the electrostatic potential energy matrix by a data preprocessing unit. The real part of the electrostatic potential energy matrix and the imaginary part of the electrostatic potential energy matrix are sorted respectively to obtain a sorted real part of the electrostatic potential energy matrix and a sorted imaginary part of the electrostatic potential energy matrix. Then, the sorted real part of the electrostatic potential energy matrix is ​​stored in a sequence real part random access memory, and the sorted imaginary part of the electrostatic potential energy matrix is ​​stored in a sequence imaginary part random access memory.

[0159] The value of the rotation factor is related to the length of the electrostatic potential matrix, and the value of the rotation factor is the same under the same length of the electrostatic potential matrix. Therefore, after the length of the electrostatic potential matrix is ​​determined, the value of the rotation factor can be calculated in advance according to the length of the electrostatic potential matrix. Since the value of the rotation factor is a complex number, the real part of the rotation factor can be stored in the rotation factor real part read-only memory, and the imaginary part of the rotation factor can be stored in the rotation factor imaginary part read-only memory.

[0160] The present application splits and sorts the electrostatic potential energy matrix through a data preprocessing unit, ensuring that the electrostatic potential energy matrix has undergone necessary preprocessing before Fourier transform calculation, which can reduce the data processing time in the subsequent Fourier transform calculation process and improve the calculation efficiency and accuracy of the Fourier transform.

[0161] Further, in the embodiments of the present application, Fig.12 As shown, the Fourier transform core also includes an output address selection unit, wherein the output address selection unit establishes communication connections with the sequence real random access memory and the sequence imaginary random access memory respectively.

[0162] The output address selection unit is used to store the real part of the Fourier transform calculation result into the sequence real random access memory according to the real part address of the target electrostatic potential energy matrix, and to store the imaginary part of the Fourier transform calculation result into the sequence imaginary random access memory according to the imaginary part address of the target electrostatic potential energy matrix.

[0163] It should be noted that, in an embodiment of the present application, after a Fourier transform is completed through a serial fast Fourier transform architecture, the calculation result of the Fourier transform is stored back in the sequence real random access memory and the sequence imaginary random access memory. Specifically, the output address selection unit stores the real part of the calculation result of the Fourier transform to the sequence real random access memory according to the real part address of the target electrostatic potential energy matrix, and stores the imaginary part of the calculation result of the Fourier transform to the sequence imaginary random access memory according to the imaginary part address of the target electrostatic potential energy matrix. The address of the output address selection unit is consistent with the address of the input address selection unit.

[0164] The present application stores the real part of the Fourier transform calculation result in a sequence real random access memory and stores the imaginary part of the Fourier transform calculation result in a sequence imaginary random access memory through an output address selection unit according to the same address as the input address selection unit, which helps to maintain the consistency and predictability of data processing, is very important for high-precision Fourier transform, and provides an accurate data basis for the subsequent next Fourier transform of the data.

[0165] According to an embodiment of the present application, a method for determining molecular electrostatic force is also provided, referring to Fig.13 , Fig.13 : is a flow chart of the steps of a method for determining molecular electrostatic force provided in an embodiment of the present application, the method comprising:

[0166] Step 1301, obtaining the electrostatic potential energy matrix and coefficient matrix of a preset number of grid points through a direct memory access engine, and storing the electrostatic potential energy matrix and coefficient matrix in a global memory.

[0167] Step 1302 , after the electrostatic potential energy matrix and the coefficient matrix are stored in the global memory, a start calculation instruction is sent to the computing unit through the direct memory access engine.

[0168] Step 1303: When the computing unit receives a start computing instruction, the electrostatic potential energy matrix is ​​loaded.

[0169] Step 1304 , during the process of loading the electrostatic potential energy matrix, a multi-dimensional Fourier transform and data transposition are synchronously performed on the electrostatic potential energy matrix.

[0170] Step 1305, after completing the multi-dimensional Fourier transform and data transposition, perform data convolution, data transposition and multi-dimensional Fourier inverse transform on the calculation result of the multi-dimensional Fourier transform to obtain the target electrostatic force result.

[0171] Step 1306, during the process of performing the multi-dimensional inverse Fourier transform, the target electrostatic force result is synchronously transmitted back to the global memory.

[0172] It should be noted that in an embodiment of the present application, the host stores the electrostatic potential matrix and coefficient matrix of the preset grid points acquired in advance to the global memory through the direct memory access engine. After storing the electrostatic potential matrix and coefficient matrix of the preset grid points to the global memory, the host sends a start calculation instruction to the computing unit through the direct memory access engine. When the computing unit receives the start calculation instruction, the electrostatic potential matrix is ​​loaded. The entire calculation process of molecular electrostatic force includes multidimensional Fourier transform, data transposition, data convolution and multidimensional inverse Fourier transform.

[0173] In order to reduce the time of data loading and result transmission, and thus shorten the calculation time of molecular electrostatic force, the present application synchronizes the loading of electrostatic potential matrix with X-dimensional Fourier transform, and synchronizes the transmission of calculation results with X-dimensional inverse Fourier transform.

[0174] Specifically, in the process of loading the electrostatic potential energy matrix, the electrostatic potential energy matrix is ​​synchronously subjected to multidimensional Fourier transform and data transposition. When the multidimensional Fourier transform and data transposition are completed, the calculation result of the multidimensional Fourier transform is subjected to data convolution, data transposition and multidimensional Fourier inverse transform to obtain the target electrostatic force result.

[0175] After obtaining the target electrostatic force result, the target electrostatic force result is stored in the global memory. The host can obtain the target electrostatic force result from the global memory through the direct memory access engine.

[0176] The present application avoids additional data transmission delays caused by the complete separation of the data loading and result return processes from the Fourier transform process, thereby shortening the calculation time of the molecular electrostatic force, improving the calculation efficiency of the molecular electrostatic force, and solving the problem of low calculation efficiency of the electrostatic force.

[0177] According to an embodiment of the present application, a molecular electrostatic force determination device is also provided. Fig.14 , Fig.14 : is a device block diagram of a molecular electrostatic force determination device provided in an embodiment of the present application, the molecular electrostatic force determination device comprising:

[0178] A data storage module 1401 is used to obtain an electrostatic potential matrix and a coefficient matrix of a preset number of grid points through a direct memory access engine, and store the electrostatic potential matrix and the coefficient matrix in a global memory;

[0179] A start calculation instruction sending module 1402 is used to send a start calculation instruction to a calculation unit through a direct memory access engine when the electrostatic potential energy matrix and the coefficient matrix are stored in the global memory;

[0180] The electrostatic potential energy matrix loading module 1403 is used to load the electrostatic potential energy matrix when the calculation unit receives the start calculation instruction;

[0181] A first synchronization module 1404 is used to synchronously perform multi-dimensional Fourier transform and data transposition on the electrostatic potential energy matrix during the process of loading the electrostatic potential energy matrix;

[0182] The target electrostatic force result determination module 1405 is used to perform data convolution, data transposition and multidimensional Fourier inverse transformation on the calculation result of the multidimensional Fourier transform after completing the multidimensional Fourier transform and data transposition, so as to obtain the target electrostatic force result;

[0183] The second synchronization module 1406 is used to synchronously transmit the target electrostatic force result back to the global memory during the process of performing the multi-dimensional inverse Fourier transform.

[0184] The present application also provides an electronic device, such as Fig.15 As shown, it includes a processor 1501, a communication interface 1502, a memory 1503 and a communication bus 1504, wherein the processor 1501, the communication interface 1502, and the memory 1503 communicate with each other through the communication bus 1504.

[0185] Memory 1503, used for storing computer programs;

[0186] The processor 1501 is used to implement the following steps when executing the program stored in the memory 1503:

[0187] The electrostatic potential energy matrix and the coefficient matrix of the preset grid points are obtained through the direct memory access engine, and the electrostatic potential energy matrix and the coefficient matrix are stored in the global memory.

[0188] After the electrostatic potential energy matrix and the coefficient matrix are stored in the global memory, a start calculation instruction is sent to the calculation unit through the direct memory access engine.

[0189] When the calculation unit receives the instruction to start calculation, the electrostatic potential energy matrix is ​​loaded.

[0190] In the process of loading the electrostatic potential energy matrix, a multi-dimensional Fourier transform and data transposition are performed on the electrostatic potential energy matrix simultaneously.

[0191] After completing the multi-dimensional Fourier transform and data transposition, the calculation result of the multi-dimensional Fourier transform is subjected to data convolution, data transposition and multi-dimensional inverse Fourier transform to obtain the target electrostatic force result.

[0192] During the multi-dimensional inverse Fourier transform process, the target electrostatic force results are synchronously sent back to the global memory.

[0193] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0194] The communication interface is used for communication between the above terminal and other devices.

[0195] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0196] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0197] In another embodiment provided in the present application, a computer-readable storage medium is provided, which stores instructions. When the computer-readable storage medium is run on a computer, the computer executes the steps of the molecular electrostatic force determination method described in any of the above embodiments.

[0198] In another embodiment provided in the present application, a computer program product is also provided, including a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the method for determining the molecular electrostatic force described in any of the above embodiments are implemented.

[0199] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Sol logo State Disk (SSD)), etc.

[0200] 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.

[0201] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system 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.

[0202] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the protection scope of the present application.

Claims

1. A system for determining molecular electrostatic forces, characterized in that: The molecular electrostatic force determination system comprises a direct memory access engine, a global memory and a computing unit, wherein the direct memory access engine establishes a communication connection with the global memory and the computing unit respectively, and the global memory establishes a communication connection with the computing unit; The direct memory access engine is used to obtain an electrostatic potential energy matrix and a coefficient matrix of a preset number of grid points, store the electrostatic potential energy matrix and the coefficient matrix in the global memory, and send a start calculation instruction to the calculation unit after completing storing the electrostatic potential energy matrix and the coefficient matrix in the global memory; The calculation unit is used to load the electrostatic potential energy matrix upon receiving the start calculation instruction, and in the process of loading the electrostatic potential energy matrix, synchronously perform multidimensional Fourier transform and data transposition on the electrostatic potential energy matrix, and in the case of completing the multidimensional Fourier transform and the data transposition, perform data convolution, data transposition and multidimensional Fourier inverse transform on the calculation result of the multidimensional Fourier transform to obtain a target electrostatic force result, and in the process of performing the multidimensional inverse Fourier transform, synchronously transmit the target electrostatic force result back to the global memory; The computing unit includes a data loading module and a Fourier transform group, wherein the data loading module establishes a communication connection with the Fourier transform group, the data loading module includes a loading counter, and the Fourier transform group includes an electrostatic potential energy local memory, a transformation counter, and a dimension counter; The data loading module is used to obtain the electrostatic potential energy matrix from the global memory, and when the electrostatic potential energy matrix is ​​obtained, store the electrostatic potential energy matrix in the electrostatic potential energy local memory, and when the electrostatic potential energy matrix is ​​stored in the electrostatic potential energy local memory, update the value of the loading counter to obtain the value of the target loading counter, and send the value of the target loading counter to the Fourier transform group; The Fourier transform group is used to perform X-dimensional Fourier transform, data transposition, Y-dimensional Fourier transform and Z-dimensional Fourier transform on the electrostatic potential energy matrix in sequence if the value of the target loading counter is received, the value of the dimension counter obtained in advance is a first preset value and the value of the transformation counter obtained in advance is less than the value of the target loading counter.

2. The molecular electrostatic force determination system according to claim 1, characterized in that: The Fourier transform group includes a Fourier transform kernel, and the data loading module includes a return counter; The Fourier transform core is used for, in the process of sequentially performing a Z-dimensional inverse Fourier transform, a Y-dimensional inverse Fourier transform, and an X-dimensional inverse Fourier transform, if the inverse Fourier transform of any dimension is completed, then the value of the dimension counter is increased by 1, and the value of the transform counter is cleared to 0; if the value of the dimension counter is a second preset value, then the X-dimensional inverse Fourier transform is performed; in the process of performing the X-dimensional inverse Fourier transform, each time the X-dimensional inverse Fourier transform is completed to obtain a target electrostatic force result, the value of the transform counter is increased by 1, the target electrostatic force result is returned to the global memory, the value of the return counter is increased by 1, and if the value of the return counter is less than the value of the transform counter, the calculation of the new target electrostatic force result is completed.

3. The molecular electrostatic force determination system according to claim 2, characterized in that: The computing unit includes a data transposition module, wherein the data transposition module establishes a communication connection with the Fourier transform core; The data transposition module is used to obtain a calculation result of an X-dimensional Fourier transform through the Fourier transform kernel, and perform data transposition on the calculation result of the X-dimensional Fourier transform to obtain a first transposition result; The Fourier transform kernel is used to perform Y-dimensional Fourier transform and Z-dimensional Fourier transform on the first transposed result in sequence to obtain a calculation result of the Z-dimensional Fourier transform.

4. The molecular electrostatic force determination system according to claim 3, characterized in that: The calculation unit includes a coefficient storage tuple and a convolution calculation module, wherein the coefficient storage tuple includes a coefficient matrix local memory, the data loading module establishes communication connections with the global memory and the coefficient matrix local memory respectively, and the convolution calculation module establishes communication connections with the coefficient matrix local memory and the electrostatic potential energy local memory respectively; The data loading module is used to obtain the coefficient matrix from the global memory, and when the coefficient matrix is ​​obtained, store the coefficient matrix in the coefficient matrix local memory; The convolution calculation module is used to obtain the coefficient matrix from the coefficient matrix local memory, obtain the calculation result of the Z-dimensional Fourier transform from the electrostatic potential energy local memory, and when the coefficient matrix and the calculation result of the Z-dimensional Fourier transform are obtained, perform data convolution on the calculation result of the Z-dimensional Fourier transform and the coefficient matrix to obtain the convolution calculation result.

5. The molecular electrostatic force determination system according to claim 3, characterized in that: The data transposition module is used to obtain the calculation result of the Y-dimensional inverse Fourier transform through the Fourier transform kernel, and perform data transposition on the calculation result of the Y-dimensional inverse Fourier transform to obtain a second transposition result; The Fourier transform kernel is used to perform the X-dimensional inverse Fourier transform on the second transposed result to obtain a target electrostatic force result.

6. The molecular electrostatic force determination system according to claim 3 or claim 5, characterized in that: The data transposition module includes an input data selection module, a first-in first-out memory, a transposition processing module, a random access memory, and an output data selection module, wherein the input data selection module establishes a communication connection with the first-in first-out memory, the first-in first-out memory establishes a communication connection with the transposition processing module, the transposition processing module establishes a communication connection with the random access memory, and the random access memory establishes a communication connection with the output data selection module; The input data selection module is used to determine the first data to be transposed and the second data to be transposed in sequence for the Fourier transform or the inverse Fourier transform of any dimension, and store the first data to be transposed and the second data to be transposed in the first-in-first-out memory; The transposition processing module is used to obtain the first data to be transposed from the first-in-first-out memory, perform transposition processing on the first data to be transposed to obtain first transposed data, and store the first transposed data in the random access memory, and to obtain the second data to be transposed from the first-in-first-out memory, perform transposition processing on the second data to be transposed to obtain second transposed data, and store the second transposed data in the random access memory; The output data selection module is used to determine the calculation result of data transposition from the random access memory according to a preset address.

7. The molecular electrostatic force determination system according to claim 2, characterized in that: The Fourier transform core includes a butterfly computing unit; The butterfly calculation unit is used to determine the calculation result of the Fourier transform for any one Fourier transform in any dimensional Fourier transform, upon receiving the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor, according to the real part of the target electrostatic potential energy matrix, the imaginary part of the target electrostatic potential energy matrix, the real part of the target rotation factor and the imaginary part of the target rotation factor.

8. The molecular electrostatic force determination system according to claim 7, characterized in that: The Fourier transform core includes an input address selection unit, a sequence real random access memory, a sequence imaginary random access memory, a rotation factor real read-only memory, and a rotation factor imaginary read-only memory, wherein the input address selection unit establishes communication connections with the sequence real random access memory, the sequence imaginary random access memory, the rotation factor real read-only memory, and the rotation factor imaginary read-only memory respectively; The input address selection unit is used to send the real part address of the target electrostatic potential energy matrix to the sequence real random access memory, send the imaginary part address of the target electrostatic potential energy matrix to the sequence imaginary random access memory, send the real part address of the target rotation factor to the real part read-only memory of the rotation factor, and send the imaginary part address of the target rotation factor to the imaginary part read-only memory of the rotation factor; The sequence real random access memory is used to obtain the real part of the target electrostatic potential energy matrix and send the real part of the target electrostatic potential energy matrix to the butterfly computing unit; The sequence imaginary part random access memory is used to obtain the imaginary part of the target electrostatic potential energy matrix, and send the imaginary part of the target electrostatic potential energy matrix to the butterfly calculation unit; The real part read-only memory of the rotation factor is used to obtain the real part of the target rotation factor and send the real part of the target rotation factor to the butterfly computing unit; The imaginary part read-only memory of the rotation factor is used to obtain the imaginary part of the target rotation factor and send the imaginary part of the target rotation factor to the butterfly calculation unit.

9. The molecular electrostatic force determination system according to claim 8, characterized in that: The sequence real part random access memory is used to obtain the real part of the target electrostatic potential energy matrix according to the real part address of the target electrostatic potential energy matrix when receiving the real part address of the target electrostatic potential energy matrix; The sequence imaginary part random access memory is used to obtain the imaginary part of the target electrostatic potential energy matrix according to the imaginary part address of the target electrostatic potential energy matrix when receiving the imaginary part address of the target electrostatic potential energy matrix; The real part read-only memory of the rotation factor is used to obtain the real part of the target rotation factor according to the real part address of the target rotation factor when the real part address of the target rotation factor is received; The twiddle factor imaginary part read-only memory is used for obtaining the target twiddle factor imaginary part according to the target twiddle factor imaginary part address when receiving the target twiddle factor imaginary part address.

10. The molecular electrostatic force determination system according to claim 8, characterized in that: The Fourier transform core includes a data preprocessing unit; The data preprocessing unit is used to split the electrostatic potential energy matrix to obtain the real part of the electrostatic potential energy matrix and the imaginary part of the electrostatic potential energy matrix, sort the real part of the electrostatic potential energy matrix and the imaginary part of the electrostatic potential energy matrix respectively to obtain the sorted real part of the electrostatic potential energy matrix and the sorted imaginary part of the electrostatic potential energy matrix, store the sorted real part of the electrostatic potential energy matrix in the sequence real part random access memory, and store the sorted imaginary part of the electrostatic potential energy matrix in the sequence imaginary part random access memory.

11. A method for determining molecular electrostatic force, characterized in that: The method comprises: Obtaining an electrostatic potential energy matrix and a coefficient matrix of a preset number of grid points through a direct memory access engine, and storing the electrostatic potential energy matrix and the coefficient matrix in a global memory; After the electrostatic potential energy matrix and the coefficient matrix are stored in the global memory, a start calculation instruction is sent to a calculation unit through the direct memory access engine; When the calculation unit receives the start calculation instruction, loading the electrostatic potential energy matrix; In the process of loading the electrostatic potential energy matrix, a multi-dimensional Fourier transform and a data transposition are synchronously performed on the electrostatic potential energy matrix; When the multidimensional Fourier transform and the data transposition are completed, data convolution, data transposition and multidimensional inverse Fourier transform are performed on the calculation result of the multidimensional Fourier transform to obtain a target electrostatic force result; In the process of performing the multi-dimensional inverse Fourier transform, synchronously transmitting the target electrostatic force result back to the global memory; Wherein, in the process of loading the electrostatic potential energy matrix, synchronously performing a multi-dimensional Fourier transform and data transposition on the electrostatic potential energy matrix includes: Obtaining the electrostatic potential energy matrix from the global memory; When the electrostatic potential energy matrix is ​​obtained, the electrostatic potential energy matrix is ​​stored in the electrostatic potential energy local memory; When the electrostatic potential energy matrix is ​​stored in the electrostatic potential energy local memory, the value of the load counter is updated to obtain the value of the target load counter; Sending the value of the target load counter to the Fourier transform group; If the Fourier transform group receives the value of the target loading counter, the value of the pre-acquired dimension counter is a first preset value and the value of the pre-acquired transformation counter is less than the value of the target loading counter, then the electrostatic potential energy matrix is ​​sequentially subjected to X-dimensional Fourier transform, data transposition, Y-dimensional Fourier transform and Z-dimensional Fourier transform.

12. An electronic device, characterized in that: include: a memory for storing processor-executable instructions; Wherein, the processor is configured to execute the instructions to implement the steps of the molecular electrostatic force determination method as described in claim 11.

13. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for determining the molecular electrostatic force as claimed in claim 11 are implemented.

14. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method for determining molecular electrostatic forces as claimed in claim 11 are implemented.