A storage and computing architecture FPGA supporting instruction broadcast

By designing a storage and computing architecture that supports instruction broadcasting in FPGA, using broadcast network path transmission instructions and global read and write channels for data transmission, the problem of low data transmission efficiency between storage and computing units in FPGA is solved, and efficient data processing is achieved and computing speed, suitable for high computing power applications.

CN116775554BActive Publication Date: 2024-11-19WUXI ESIONTECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310771719.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-27
Publication Date
2024-11-19
Estimated Expiration
2043-06-27

AI Technical Summary

Technical Problem

When performing computing tasks, existing FPGAs need to frequently transmit large amounts of data between the storage unit and the computing unit, resulting in large consumption of wiring resources and interconnection resources, high latency and energy losses, limiting data processing efficiency.

Method used

A memory and computing architecture supporting instruction broadcast is designed. By setting up several resource modules, interconnect resources, global read and write channels and broadcast network paths inside the FPGA, a memory and computing unit is formed. Each memory and computing unit includes a processor and local. Data storage, instruction storage is independent of the computing unit. Instructions are transmitted through the broadcast network path. Each computing unit executes the same computing commands synchronously in parallel to reduce the demand for BRAM modules.

Benefits of technology

It achieves higher data processing efficiency and computing speed, reduces data transmission delay and energy loss, supports parallel computing with more cores, and is suitable for high computing power applications such as artificial intelligence and machine learning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116775554B_ABST
    Figure CN116775554B_ABST
Patent Text Reader

Abstract

The present application discloses a storage and computing architecture FPGA supporting instruction broadcast, which relates to the field of FPGA technology. In the FPGA, a resource module located in the same sub-area implements a storage and computing unit. For application scenarios in which each storage and computing unit performs the same computing operation, each storage and computing unit includes a processor and local data storage but does not need to set up an instruction storage separately. An instruction storage shared by multiple storage and computing units is separately formed in the FPGA. The storage and computing instructions in the instruction storage are transmitted to the processors of each storage and computing unit in a broadcast form, so that the processors of each storage and computing unit can execute the storage and computing instructions and perform operations according to their own data. Multiple storage and computing units can complete a storage and computing operation with the same storage and computing instruction in parallel and synchronously. Moreover, since each storage and computing unit reuses the same instruction storage, a larger number of storage and computing units can be formed to realize parallel operations with more cores, so as to achieve higher data processing efficiency and operation speed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of FPGA technology, and in particular to a storage-computing architecture FPGA that supports instruction broadcasting. Background Art

[0002] With the development of high-tech technologies such as the Internet of Things, cloud computing, and artificial intelligence, various data-intensive applications based on hardware platforms have also grown dramatically. FPGA (Field Programmable Gate Array) is a commonly used hardware platform due to its advantages of parallelism and reconfigurability.

[0003] The existing FPGA, like most traditional hardware platforms, adopts the von Neumann architecture. The FPGA contains a large number of resource modules such as CLB, DSP and BRAM. These resource modules are connected by highly configurable interconnects to form computing units and storage units in the FPGA. BRAM is an important on-chip storage unit. On-chip data is mainly stored in BRAM, and on-chip data must be transferred to the computing unit during calculation.

[0004] Therefore, existing FPGAs, like other hardware platforms that use the von Neumann architecture, need to repeatedly transfer large amounts of data between storage units and computing units when performing various computing tasks. Frequent and large-scale data movement not only consumes a large amount of wiring resources and interconnection resources, but also causes huge delays and energy losses, thereby limiting the data processing efficiency of FPGAs. Summary of the invention

[0005] In response to the above problems and technical requirements, the applicant has proposed a storage-computing architecture FPGA that supports instruction broadcasting. The technical solution of this application is as follows:

[0006] A storage-computing architecture FPGA supporting instruction broadcasting, wherein the storage-computing architecture FPGA internally includes a plurality of resource modules arranged according to a predetermined structure, interconnection resources arranged around the resource modules, a global read-write channel realized by additional hardware resources, and a broadcast network path realized by additional hardware resources;

[0007] Resource modules located in the same sub-area are connected via interconnection resources to realize a storage and computing unit, each storage and computing unit includes a processor and local data storage, and the processor in the storage and computing unit is connected to the local data storage via the interconnection resources;

[0008] The instruction storage independent of each storage and computing unit is connected to the off-chip of the storage and computing architecture FPGA through the instruction read and write interface of the storage and computing architecture FPGA for instruction transmission. The instruction storage connects the processors of multiple storage and computing units through a broadcast network path and transmits the stored storage and computing instructions in a broadcast form; the local data storage in each storage and computing unit is connected to the data read and write interface of the storage and computing architecture FPGA through a global read and write channel, and data is transmitted with the off-chip of the storage and computing architecture FPGA through the data read and write interface.

[0009] A further technical solution is that a method for a storage-computing architecture FPGA to complete a storage-computing operation by using multiple storage-computing units to work in parallel and synchronously includes:

[0010] The data to be processed of each storage and computing unit is written into the local data storage of each storage and computing unit in turn through the global read and write channel through the data read and write interface, and the storage and computing instructions are written into the instruction storage through the instruction read and write interface. The storage and computing instructions in the instruction storage are transmitted to each storage and computing unit in the form of broadcast. Each storage and computing unit executes the received storage and computing instructions, processes the data to be processed in its own local data storage, and stores the calculation results in its own local data storage.

[0011] A further technical solution is that when any storage and calculation unit detects that there is no data to be processed in the local data storage, it suspends calculation and does not respond to the received storage and calculation instructions.

[0012] Its further technical solution is that when any storage and computing unit suspends calculation, the local data storage of the storage and computing unit is transmitted to the outside of the storage and computing architecture FPGA through the global read and write channel through the data read and write interface to obtain new data to be processed input from outside the chip, or to output the stored calculation results to the outside of the chip.

[0013] A further technical solution is that the storage and computing instructions in the instruction storage are transmitted to the connected storage and computing units in the same main clock cycle in a broadcast form, or the storage and computing instructions in the instruction storage are transmitted to the connected storage and computing units in a plurality of different main clock cycles in a broadcast form.

[0014] Its further technical solution is that when the storage-computing architecture FPGA executes the same storage-computing instruction through multiple storage-computing units working in parallel and synchronously to complete a storage-computing operation, the order in which the multiple storage-computing units obtain the storage-computing instructions is the same as the data flow order of the data processed by the multiple storage-computing units.

[0015] Its further technical solution is that the storage and computing instructions in the instruction storage are transmitted to the various storage and computing units of the connected i-th storage and computing group at any i-th main clock cycle, and are transmitted to the various storage and computing units of the connected i+1-th storage and computing group at any i+1-th main clock cycle, and the data flow sequence flows from the various storage and computing units of the i-th storage and computing group to the various storage and computing units of the i+1-th storage and computing group, so that the various storage and computing units of the i+1-th storage and computing group obtain the intermediate results generated by the various storage and computing units of the i-th storage and computing group and perform calculations, where i is a parameter.

[0016] A further technical solution is that the storage and computing units implemented in each sub-area are arranged to form a two-dimensional row and column structure, the storage and computing units in the same column belong to the same storage and computing group, and the i-th storage and computing group is adjacent to the i+1-th storage and computing group.

[0017] A further technical solution is that the local data storage in each storage and computing unit is implemented based on a BRAM module, and is used to implement a sub-area of ​​a storage and computing unit including at least one BRAM module.

[0018] The beneficial technical effects of this application are:

[0019] The present application discloses a storage and computing architecture FPGA supporting instruction broadcasting, in which resource modules located in the same sub-area of ​​the FPGA are connected via interconnection resources inside the FPGA to realize a storage and computing unit. For application scenarios in which each storage and computing unit performs the same operation, each storage and computing unit includes a processor and local data storage but does not need to set up an instruction storage separately. An instruction storage shared by multiple storage and computing units is separately formed inside the FPGA, and the storage and computing instructions in the instruction storage are transmitted to the processors of each storage and computing unit in a broadcast form, so that the processors of each storage and computing unit can execute the storage and computing instructions and perform operations according to their own data, and multiple storage and computing units are used to complete a storage and computing operation with the same storage and computing instruction in parallel and synchronously. Moreover, since each storage and computing unit reuses the same instruction storage, the number of BRAM modules required for each storage and computing unit is reduced, so that a larger number of storage and computing units can be formed inside the FPGA to realize parallel operations of more cores, so that the storage and computing architecture FPGA has higher data processing efficiency and operation speed.

[0020] The instruction storage can broadcast the storage and computing instructions to each storage and computing unit synchronously, or it can broadcast the storage and computing instructions to each storage and computing unit in sequence in a pipeline form, which can meet the application requirements that the storage and computing units calculate in sequence according to the data flow order.

[0021] Each storage-computing unit also has a pause control signal with a higher priority than the storage-computing instruction, which can force the execution of the storage-computing instruction to be paused, which is very suitable for the storage-computing units at the edge of the array to handle boundary conditions.

[0022] The processor and local data storage in each storage and computing unit formed inside the FPGA are in the same sub-area. When executing, the processor only needs to read and write data in the adjacent local data storage, which reduces the wiring resources and interconnection resources consumed by transmitting data between the processor and the local data storage, reduces data transmission delay and energy loss, and improves the data processing efficiency of the FPGA. In addition, the FPGA can realize multi-core parallel storage and computing operations through multiple storage and computing units. The processors of each storage and computing unit will not cause conflicts with each other. Each storage and computing unit can perform operations at full speed, which is not affected by the storage bandwidth, and can further improve the overall data processing efficiency and computing speed of the FPGA. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of the arrangement of resource modules inside the storage-computing architecture FPGA and a schematic diagram of the sub-areas obtained by division in an example of the present application.

[0024] Figure 2 is based on Figure 1 A schematic diagram of the implemented storage and computing unit and instruction storage and the circuit structure formed.

[0025] Figure 3 is based on Figure 1 Schematic diagram of another circuit structure formed by the implemented storage and computing unit and instruction storage.

[0026] Figure 4 is based on Figure 1 Schematic diagram of another circuit structure formed by the implemented storage and computing unit and instruction storage. DETAILED DESCRIPTION

[0027] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.

[0028] This application discloses a storage and computing architecture FPGA that supports instruction broadcasting. Please refer to Figure 1The FPGA of the present application is the same as a conventional FPGA, and includes several resource modules arranged in a predetermined structure and interconnection resources arranged around the resource modules. The predetermined structure can be arranged according to actual conditions. For example, taking the currently mainstream Column-Based structure as an example, the resource modules are arranged in a row and column structure to form a two-dimensional array architecture. The same type of resource modules are arranged in each column, and interconnection resources are arranged around each resource module. The types of resource modules included in the FPGA include configurable logic modules CLB, configurable storage modules BRAM, and configurable multiplication and addition modules DSP. The configurable logic module CLB contains units such as lookup tables and registers, which can be configured to implement various logical functions. The configurable multiplication and addition module DSP contains arithmetic units that can perform multiple operations such as multiplication, addition, and multi-bit logic gates. The configurable storage module BRAM contains block storage units that can implement single-port and dual-port storage and FIFO storage of various widths and depths. Please refer to Figure 1 The schematic diagram of the resource module inside the FPGA shown is a schematic diagram of the resource module inside the FPGA. This part of the structure follows the structure inside the existing FPGA and will not be elaborated in this application.

[0029] The resource modules arranged in a predetermined structure in the FPGA of the present application are divided into several sub-areas, each of which contains a certain amount of resource modules, such as Figure 1 The area within each dotted box represents a sub-area. The resource modules located in the same sub-area are connected through the interconnection resources inside the FPGA to realize a storage and computing unit, and the resource modules in different sub-areas correspondingly realize multiple storage and computing units. Each storage and computing unit includes its own processor and local data storage. The processor in each storage and computing unit is connected to the local data storage in the storage and computing unit via the interconnection resources. The local data storage in each storage and computing unit is used to store the data of the storage and computing unit. The local data storage in each storage and computing unit is implemented based on the BRAM module in the corresponding sub-area, and the processor is implemented based on the DSP module in the corresponding sub-area, and a certain number of CLB modules are also required.

[0030] In one embodiment, the number and type of resource modules contained in any two sub-areas obtained by division are the same, or in another embodiment, the number and type of resource modules contained in any two sub-areas obtained by division are at least one different, that is, the multiple sub-areas obtained by division may be completely the same or not necessarily the same. This application does not limit the number and type of resource modules contained in each sub-area, but the number and type of resource modules contained in each sub-area must be sufficient to implement a processor and local data storage in a storage and computing unit.

[0031] In one embodiment, the instruction sets used by the two storage and computing units implemented by the resource modules in any two sub-areas are the same or different, and the bit widths of the two storage and computing units using the same instruction set are the same or different. In the present application, the storage and computing units formed inside the FPGA can form various typical general-purpose 8-bit processors including RISC, 8051 and 6802.

[0032] In addition to using resource modules in each sub-region to implement storage and computing units, the FPGA also implements instruction storage through resource modules independent of each sub-region. The implemented instruction storage is independent of each storage and computing unit and is used to store the storage and computing instructions of each storage and computing unit. Figure 1 The resource modules outside the dashed boxes in the figure implement instruction storage, which is mainly implemented based on the BRAM module in the FPGA. The instruction storage is connected to the off-chip of the storage and computing architecture FPGA through the instruction read and write interface of the storage and computing architecture FPGA for instruction transmission, so that the storage and computing instructions in the instruction storage can be read and written from the off-chip of the storage and computing architecture FPGA. Please refer to Figure 2 is corresponding to Figure 1 Schematic diagram of the implemented storage and computation unit and instruction storage.

[0033] In addition to resource modules and interconnection resources, the FPGA of the present application also adds global read-write channels implemented by additional hardware resources and broadcast network paths implemented by additional hardware resources compared to conventional FPGAs. The instruction storage is connected to the processors of multiple storage and computing units through a broadcast network path, and the local data storage in each storage and computing unit is connected to the data read-write interface of the storage and computing architecture FPGA through a global read-write channel. The local data storage in each storage and computing unit also transmits data to the outside of the storage and computing architecture FPGA through the data read-write interface, so that the data in each local data storage can be read and written from outside the storage and computing architecture FPGA.

[0034] Based on this structure, the method for the storage-computing architecture FPGA to complete a storage-computing operation through multiple storage-computing units working in parallel and synchronously includes: writing the to-be-processed data of each storage-computing unit into the local data storage of each storage-computing unit in turn through the global read-write channel through the data read-write interface, and writing the storage-computing instruction into the instruction storage through the instruction read-write interface. The storage-computing instruction in the instruction storage is transmitted to each storage-computing unit in the form of broadcast, and multiple storage-computing units are started. Each storage-computing unit executes the received storage-computing instruction, processes the to-be-processed data in its own local data storage, and stores the operation result in its own local data storage. Subsequently, the operation result in the local data storage of each storage-computing unit can also be read out to the outside of the FPGA chip through the global read-write channel and the data read-write interface. Each storage-computing unit supports independent storage-computing operations in the form of soft cores. The multiple storage-computing units formed inside the FPGA perform operations on their own data according to the same storage-computing instruction to jointly complete a multi-core storage-computing operation. This architecture is very suitable for applications with high computing power such as artificial intelligence, machine learning, and specific signal processing, such as convolution operations and various operations in neural networks.

[0035] In one embodiment, Figure 3 As shown, a storage and computing unit implemented by the resource module in each sub-area also includes an instruction register IR and a data register DR. The processor of a storage and computing unit reads and writes the connected local data storage through the data register DR. The instruction storage is connected to the instruction register IR of each storage and computing unit through a broadcast network path, and then connected to the processor. The storage and computing instructions in the instruction storage are first read out to the instruction register IR of each storage and computing unit, and then decoded into corresponding control signals and sent to the processor. Correspondingly, the data in the local data storage is first read out to the data register DR and then input into the processor, or the processor writes the data into the data register DR and then writes it into the local data storage.

[0036] This application aims at the scenario where all storage and computing units execute the same storage and computing instructions to perform the same operations on their respective data, and sets up an instruction storage independent of each storage and computing unit. Each storage and computing unit uniformly obtains the storage and computing instructions from the instruction storage. Therefore, there is no need to set up local instruction storage in each storage and computing unit separately, so it is sufficient to implement at least one BRAM module in a sub-area of ​​a storage and computing unit, that is, at least one BRAM module is required in each sub-area for local data storage. If each storage and computing unit is to set up its own local instruction storage, at least two BRAM modules are required in each sub-area, one BRAM module for local data storage, and one BRAM module for local instruction storage. Therefore, this approach of the present application can reduce the number of BRAM modules required for each storage and computing unit, and realize the sharing of instruction storage by multiple storage and computing units. Under the premise that the number of BRAM modules included on the chip is fixed, this structure can form a larger number of storage and computing units in the FPGA, thereby realizing storage and computing operations of more cores. When multiple storage and computing units work in parallel and synchronously to complete a storage and computing operation, the processor of each storage and computing unit only needs to read and write the data in the adjacent local data storage, thereby reducing the wiring resources and interconnection resources consumed by transmitting data between the processor and the local data storage, reducing data transmission delays and energy losses, and improving the data processing efficiency of the FPGA. In addition, in a multi-core scenario, if multiple computing cores read and write the same main memory at the same time, the read and write bandwidth of the main memory will form a bottleneck, restricting the data processing efficiency. This problem will be particularly obvious in high-performance computing. For FPGAs that support artificial intelligence or machine learning applications, the FPGA generally has a large number of relatively simple computing cores, and the number of computing cores will exceed 100. The demand for the read and write bandwidth of the main memory is greater, and the problem is more prominent. The FPGA of this application is also a multi-core scenario. Since the processors in each storage and computing unit read and write local data storage respectively, instead of all processors uniformly reading and writing the same data main memory, the processors of each storage and computing unit will not cause conflicts with each other, and each storage and computing unit can perform operations at full speed, which is not affected by the storage bandwidth, which is conducive to improving the overall data processing efficiency and computing speed. This makes the FPGA of the present application suitable for scenarios involving a large number of storage and computing units. Even if the number of storage and computing units exceeds 100, it can still have a high computing rate, thereby better supporting scenarios of artificial intelligence and machine learning.

[0037] On this basis, according to actual needs, FPGA can sequentially implement multiple storage and calculation operations through the internal storage and calculation units until all operations are completed. And when FPGA implements any storage and calculation operation, the FPGA chip can write data to the local data storage in all or part of the storage and calculation units, can start all or part of the storage and calculation units inside the FPGA, and can also read the calculation results in the local data storage of all or part of the storage and calculation units to the outside of the chip. That is, when implementing multiple storage and calculation operations, each storage and calculation operation can use all the storage and calculation units formed inside the FPGA, or only use part of the storage and calculation units and skip other storage and calculation units. When using part of the storage and calculation units, the same or different storage and calculation units can also be used multiple times. That is, the process of executing multiple storage and calculation operations is basically similar, but each storage and calculation operation can flexibly adjust the data reading and writing of each storage and calculation unit, and there is no need to perform the same data reading and writing operations on the local data storage in all storage and calculation units every time.

[0038] In one embodiment, when the FPGA completes a storage-calculation operation in parallel and synchronously through multiple storage-calculation units, any storage-calculation unit suspends the operation and does not respond to the received storage-calculation instruction when it detects that there is no data to be processed in the local data storage. That is, the storage-calculation unit has a pause control signal with a higher priority than the storage-calculation instruction, and the pause control signal is used to force the storage-calculation unit to execute the pause instruction and suspend the operation. This is more common for the storage-calculation units at the edge of the array and can be used to handle boundary conditions.

[0039] In addition, when any storage and computing unit is suspended from computing, the local data storage of the storage and computing unit is transmitted to the off-chip of the storage and computing architecture FPGA via the global read and write channel through the data read and write interface to obtain new data to be processed input from the off-chip, or to output the stored computing results to the off-chip. That is, the pause control signal of the above embodiment can be used, and when a storage and computing unit is suspended, the local data storage of the storage and computing unit can exchange data with the off-chip.

[0040] When the storage and computing instructions in the instruction storage are transmitted to each storage and computing unit in the form of broadcast, in one embodiment, the storage and computing instructions in the instruction storage are transmitted to each connected storage and computing unit in the same main clock cycle in the form of broadcast, that is, the storage and computing instructions are synchronously broadcast to each storage and computing unit for execution. Or in another embodiment, the storage and computing instructions in the instruction storage are transmitted to each connected storage and computing unit in the form of broadcast in multiple different main clock cycles, that is, the storage and computing instructions are broadcast to each storage and computing unit in a pipeline form for execution in sequence, and there is a time delay between the actions of broadcasting the storage and computing instructions to different storage and computing units. The pipeline broadcast method provided in this embodiment is very suitable for computing scenarios such as convolution operations and neural networks.

[0041] When the storage and computing instructions are broadcasted to each storage and computing unit in sequence, and the storage and computing architecture FPGA completes a storage and computing operation by executing the same storage and computing instructions through multiple storage and computing units working in parallel and synchronously, the order in which the multiple storage and computing units obtain the storage and computing instructions is the same as the data flow order of the data processed by the multiple storage and computing units. In the application of storage and computing operations such as convolution operations and neural networks, data needs to be transferred between multiple storage and computing units that complete a storage and computing operation to form a data flow. For example, the operation result obtained by a storage and computing unit will be transmitted to another storage and computing unit as intermediate data for use. The storage and computing unit with a later data flow order of the processed data may need to rely on the data generated by the previous storage and computing unit, so it needs to start later. Therefore, this embodiment broadcasts the storage and computing instructions to each storage and computing unit in sequence according to the data flow order of the processed data.

[0042] More commonly, the delay between the actions of broadcasting storage and computing instructions to different storage and computing units is one main clock cycle, and the delay can be implemented using a beat register. Then one implementation method is that the storage and computing instructions in the instruction storage are transmitted to each storage and computing unit of the connected i-th storage and computing group at any i-th main clock cycle, and are transmitted to each storage and computing unit of the connected i+1-th storage and computing group at any i+1-th main clock cycle, and the data flow sequence flows from each storage and computing unit of the i-th storage and computing group to each storage and computing unit of the i+1-th storage and computing group, so that each storage and computing unit of the i+1-th storage and computing group obtains the intermediate results generated by each storage and computing unit of the i-th storage and computing group and performs calculations, and i is an integer parameter.

[0043] More commonly, multiple storage and computing units in the same column of the BRAM module used to form local data storage belong to the same storage and computing group, that is, the data flow sequence is transmitted in sequence between the columns along the row direction, such as Figure 4 In the first main clock cycle, the instruction storage broadcasts the storage and calculation instruction to the storage and calculation unit 1 to the storage and calculation unit 4. After the instruction storage is delayed by a beat register for one main clock cycle, the storage and calculation instruction is broadcast to the storage and calculation unit 5 to the storage and calculation unit 8 in the second main clock cycle.

[0044] In one embodiment, when the predetermined structure adopted by the resource module arrangement inside the FPGA is a Column-Based structure, the BRAM module occupies one or more columns, and the local data storage in each storage and computing unit is implemented based on the BRAM module. Therefore, the BRAM module used to implement the local data storage in each storage and computing unit is located in one or more columns, and the BRAM modules used to implement the local data storage in the storage and computing unit and located in the same column transmit data with the FPGA chip via the same global read-write channel. The FPGA chip can exchange data with any local data storage connected to the global read-write channel via each global read-write channel. Based on this, the present application provides the following two methods to implement the local data storage in each storage and computing unit and the data transmission outside the FPGA chip:

[0045] In one embodiment, several BRAM modules located in the same column are used to implement local data storage in the storage and computing unit, and are connected to a data read / write interface of the FPGA through the same global read / write channel to perform data transmission with the outside of the FPGA, and different global read / write channels are connected to different data read / write interfaces of the FPGA respectively. Figure 2 and Figure 3 In the example shown, 8 storage and computing units are formed inside the FPGA. The local data storage in storage and computing units 1 to 4 is realized by the BRAM modules located in the same column. The local data storage of these 4 storage and computing units is connected to the data read and write interface 1 of the FPGA through the global read and write channel 1 to realize data transmission with the outside of the FPGA chip. The local data storage in storage and computing units 5 to 8 is realized by the BRAM modules located in the same column. The local data storage of these 4 storage and computing units is connected to the data read and write interface 2 of the FPGA through the global read and write channel 2 to realize data transmission with the outside of the FPGA chip.

[0046] The method provided in the above embodiment often needs to occupy multiple data read and write interfaces of FPGA, and the interface resources of FPGA are relatively precious. Therefore, in another embodiment, only one data read and write interface of FPGA is needed to realize local data storage in soft-core processor, and several BRAM modules located in the same column are connected to an active end of multiplexer MUX through the same global read and write channel, and the fixed end of multiplexer MUX is connected to the data read and write interface of FPGA. By selecting each active end of multiplexer MUX, each global read and write channel can be selected, so that data can be exchanged with any local data storage connected to the global read and write channel outside FPGA chip through the selected global read and write channel. For example, in Figure 4 In the structure diagram shown, it corresponds to Figure 2 and Figure 3, the local data storage in storage and calculation unit 1 to storage and calculation unit 4 is connected to an active end of the multiplexer MUX through the global read-write channel 1, the local data storage in storage and calculation unit 5 to storage and calculation unit 8 is connected to the other active end of the multiplexer MUX through the global read-write channel 2, and the fixed end of the multiplexer MUX is connected to the data read-write interface of the FPGA to realize data transmission with the outside of the chip, compared with Figure 2 and Figure 3 In the embodiment, the method of this embodiment only needs to occupy one data read and write interface of the FPGA.

[0047] In another embodiment, the resource module in each sub-area is also used to implement the control signal and status signal of a corresponding storage and computing unit, the control signal is used to control the operating status of the storage and computing unit, and the status signal is used to display the operating status of the storage and computing unit. The control signal and status signal of the storage and computing unit are connected to other circuit structures inside the FPGA, or the control signal and status signal of the storage and computing unit are connected to the outside of the FPGA. That is, the control signal and status signal required by each storage and computing unit when performing storage and computing are implemented by the resource module in the corresponding sub-area.

[0048] The control signal of each storage and computing unit includes at least one of an enable signal, a clock signal, a reset signal and an interrupt signal. The enable signal is used to control the corresponding storage and computing unit to execute instructions or stop executing instructions. The reset signal is used to reset the storage and computing unit. The interrupt signal provides external interrupt capability. The control signals of any two storage and computing units are the same or different.

[0049] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.

Claims

1. A storage-computing architecture FPGA supporting instruction broadcast, characterized in that: The storage-computing architecture FPGA includes a plurality of resource modules arranged in a predetermined structure, interconnection resources arranged around the resource modules, a global read-write channel implemented by additional hardware resources, and a broadcast network path implemented by additional hardware resources; The resource modules located in the same sub-area are connected via interconnection resources to realize a storage and computing unit, each storage and computing unit includes a processor and local data storage, and the processor in the storage and computing unit is connected to the local data storage via the interconnection resources; The instruction storage independent of each storage and computing unit is connected to the off-chip of the storage and computing architecture FPGA through the instruction read and write interface of the storage and computing architecture FPGA for instruction transmission. The instruction storage is connected to the processors of multiple storage and computing units through a broadcast network path and transmits the stored storage and computing instructions in a broadcast form; the local data storage in each storage and computing unit is connected to the data read and write interface of the storage and computing architecture FPGA through a global read and write channel, and data is transmitted with the off-chip of the storage and computing architecture FPGA through the data read and write interface; The method for the storage-computing architecture FPGA to complete a storage-computing operation by having multiple storage-computing units working in parallel and synchronously includes: writing the to-be-processed data of each storage-computing unit into the local data storage of each storage-computing unit in turn through the data read-write interface via the global read-write channel, writing the storage-computing instructions into the instruction storage through the instruction read-write interface, transmitting the storage-computing instructions in the instruction storage to each storage-computing unit in the form of broadcast, and each storage-computing unit executing the received storage-computing instructions, processing the to-be-processed data in its own local data storage, and storing the operation results in its own local data storage.

2. The storage-computing architecture FPGA according to claim 1, characterized in that: When any storage and calculation unit detects that there is no data to be processed in the local data storage, it suspends calculation and does not respond to the received storage and calculation instructions.

3. The storage-computing architecture FPGA according to claim 2, characterized in that: When any storage and computing unit is in the process of suspending calculation, the local data storage of the storage and computing unit is transmitted to the outside of the storage and computing architecture FPGA via the global read and write channel through the data read and write interface to obtain new data to be processed input from outside the chip, or to output the stored calculation results to the outside of the chip.

4. The storage-computing architecture FPGA according to claim 1, characterized in that: The storage and computing instructions in the instruction storage are transmitted to each connected storage and computing unit in a broadcast form in the same main clock cycle, or the storage and computing instructions in the instruction storage are transmitted to each connected storage and computing unit in a broadcast form in multiple different main clock cycles.

5. The storage-computing architecture FPGA according to claim 4, characterized in that: When the storage-computing architecture FPGA completes a storage-computing operation by executing the same storage-computing instruction through multiple storage-computing units working in parallel and synchronously, the order in which the multiple storage-computing units obtain the storage-computing instructions is the same as the data flow order of the data processed by the multiple storage-computing units.

6. The storage-computing architecture FPGA according to claim 5, characterized in that: The storage and computing instructions in the instruction storage are transmitted to each storage and computing unit of the connected i-th storage and computing group in any i-th main clock cycle, and are transmitted to each storage and computing unit of the connected i+1-th storage and computing group in any i+1-th main clock cycle. The data flow sequence flows from each storage and computing unit of the i-th storage and computing group to each storage and computing unit of the i+1-th storage and computing group, so that each storage and computing unit of the i+1-th storage and computing group obtains the intermediate results generated by each storage and computing unit of the i-th storage and computing group and performs calculations, where i is a parameter.

7. The storage-computing architecture FPGA according to claim 6, characterized in that: The storage and computing units implemented in each sub-area are arranged to form a two-dimensional row and column structure. The storage and computing units in the same column belong to the same storage and computing group, and the i-th storage and computing group is adjacent to the i+1-th storage and computing group.

8. The storage-computing architecture FPGA according to claim 1, characterized in that: The local data storage in each storage and computing unit is implemented based on a BRAM module, which is used to implement a sub-area of ​​a storage and computing unit including at least one BRAM module.