CN109683877B - SystemC-based GPU software and hardware interaction TLM system - Google Patents

Abstract

The invention relates to the technical field of computer hardware modeling, in particular to a GPU software and hardware interaction TLM model based on SystemC. The GPU software and hardware interaction TLM model based on SystemC comprises a host agent layer model and a host interface layer model, wherein the host agent layer model provides an interaction path between a software driver and the host interface layer model, the host interface layer model provides an interaction path between the host agent layer model and the GPU hardware TLM model, and the interaction content comprises issuing of an OpenGL CM (Command Packet) Command Packet, reading and writing of a system structure register, interrupt processing and loading or obtaining of image and texture data. The invention realizes the TLM model-based GPU software and hardware model interaction method, solves the problem of software drive and hardware model interaction in GPU modeling, and effectively helps RTL design development and verification.

Description

SystemC-based GPU software and hardware interaction TLM system

Technical Field

The invention relates to the technical field of computer hardware modeling, in particular to a GPU software and hardware interaction TLM system based on SystemC.

Background

At present, China autonomously develops a domestic Graphic Processing Unit (GPU), but since a domestic commercial or military GPU chip is always adopted at home, the GPU is lack of design experience in the aspect of autonomous development of the GPU, and belongs to the stage of exploring and developing a core technology, meanwhile, the GPU chip has a huge hardware logic scale, and is realized from system architecture document compiling to rtl design, so that the GPU chip is a long project, and even if the rtl design is finished, the verification of functional performance and the like also needs a large amount of time, so that in order to improve the progress and reliability of autonomously developing Graphic processor projects, a college method is urgently needed to shorten the whole design period and assist the design process of a Graphic processor, and the method is (Transaction Level Modeling, TLM) Transaction Level Modeling. TLM (object-level modeling method) is an advanced digital system modeling method that can quickly build system models at early stages of a project, separating the communication details between models from the details of functional units or communication architectures. The communication mechanism (e.g., bus or FIFO) is modeled as a channel and presented to the module in the form of a SystemC interface class. Transaction requests typically occur when interface functions of these channel models are invoked, and the interface functions encapsulate the underlying details of the information exchange. At a business level, TLM emphasizes the functionality of data transfer itself-the content of the data and the start and end points of the transfer, and involves as few implementations as possible.

Disclosure of Invention

Based on the problems in the background art, the GPU software and hardware interaction TLM model design based on the SystemC can solve the interaction problem of the software driver and the GPU hardware model, and opens up software and hardware interaction channels for function verification on the hardware microstructure TLM model in advance.

The invention provides a Graphics Processing Unit (GPU) hardware and software interaction TLM system based on SystemC, which is characterized by comprising a host agent layer model and a host interface layer model, wherein the host agent layer model and the host interface layer model are used for realizing the hardware and software interaction of driving software and a GPU hardware model.

Optionally, the host agent layer model is configured to implement data interaction between the C language-based driver software and the host interface layer model based on SystemC, and specifically includes: the device comprises a Command issuing module, a read-write module, an interrupt processing module and a main memory reading module, wherein the Command issuing module is used for sending an OpenGL CM (Command Package) Command packet, and the Command packet carries OpenGL functional codes and related parameter information; the read-write module is used for reading and writing the architecture register, and the architecture register is used for representing the working state and the mode of the GPU.

Optionally, the host interface layer model is configured to implement interaction between a GPU host interface layer model based on a SystemC transaction level interface and a GPU PCIe hardware model, and the host interface layer model includes: the system comprises a host memory read-write transaction level module, a system structure register read-write transaction level module, an interrupt state read-write transaction level module and a direct memory read DMA configuration transaction level module, wherein the host memory read-write transaction level module provides a host memory read-write transaction level interface, the system structure register read-write transaction level module provides a system structure register read-write transaction level interface, the interrupt state read-write transaction level module provides an interrupt state read-write transaction level interface, and the direct memory read DMA configuration transaction level module provides a DMA configuration transaction level interface for calling a host agent layer model and a GPU hardware model.

Optionally, the command issuing module is connected to the DMA configuration transaction level module through the DMA configuration transaction level interface, and issues the CM command packet to the host interface layer model in the form of a DMA descriptor, where the DMA descriptor includes a DMA destination address, a DMA source address, a DMA path, and a DMA length.

The read-write module is connected with the architecture register read-write transaction level module through the architecture register read-write transaction level interface and is used for configuring a GPU working mode and acquiring a GPU working state.

The interrupt processing module is connected with the interrupt state read-write transaction level module through the interrupt state read-write transaction level interface and is used for acquiring and clearing host interrupt.

The main memory reading module is connected with the host memory reading and writing transaction level module through the host memory reading and writing transaction level interface and is used for reading and writing main memory data.

The GPU software and hardware interaction TLM system based on SystemC comprises a host agent layer model and a host interface layer model, wherein the host agent layer model provides an interaction path of a software driver and the host interface layer model, the host interface layer model provides an interaction path of the host agent layer model and the GPU hardware TLM model, and the interaction content comprises issuing of an OpenGL CM (Command Packet) Command Packet, reading and writing of a system structure register, interrupt processing and loading or obtaining of image and texture data. The method can solve the interaction problem of the software driver and the GPU hardware model, and can open a software and hardware interaction channel for performing functional verification on the hardware microstructure TLM model in advance.

Drawings

FIG. 1 is a schematic structural diagram of a GPU software and hardware interaction TLM model based on SystemC provided by the invention;

Detailed Description

A Graphics Processing Unit (GPU) hardware and software interaction TLM system based on SystemC is shown in figure 1 and comprises a host agent layer model and a host interface layer model, wherein the host agent layer model and the host interface layer model are used for realizing the hardware and software interaction between driving software and a GPU hardware model.

Optionally, the host agent layer model is configured to implement data interaction between the C language-based driver software and the host interface layer model based on SystemC, and specifically includes: the device comprises a Command issuing module, a read-write module, an interrupt processing module and a main memory reading module, wherein the Command issuing module is used for sending an OpenGL CM (Command Package) Command packet, and the Command packet carries OpenGL functional codes and related parameter information; the read-write module is used for reading and writing the architecture register, and the architecture register is used for representing the working state and the mode of the GPU.

Optionally, the host interface layer model is configured to implement interaction between a GPU host interface layer model based on a SystemC transaction level interface and a GPU PCIe hardware model, and the host interface layer model includes: the system comprises a host memory read-write transaction level module, a system structure register read-write transaction level module, an interrupt state read-write transaction level module and a direct memory read DMA configuration transaction level module, wherein the host memory read-write transaction level module provides a host memory read-write transaction level interface, the system structure register read-write transaction level module provides a system structure register read-write transaction level interface, the interrupt state read-write transaction level module provides an interrupt state read-write transaction level interface, and the direct memory read DMA configuration transaction level module provides a DMA configuration transaction level interface for calling a host agent layer model and a GPU hardware model.

Optionally, the method specifically includes:

the command issuing module is connected with the direct memory access DMA configuration transaction level module through the DMA configuration transaction level interface, and issues the CM command packet to the host interface layer model in a DMA descriptor mode, wherein the DMA descriptor comprises a DMA destination address, a DMA source address, a DMA path and a DMA length.

The read-write module is connected with the architecture register read-write transaction level module through the architecture register read-write transaction level interface and is used for configuring a GPU working mode and acquiring a GPU working state.

The interrupt processing module is connected with the interrupt state read-write transaction level module through the interrupt state read-write transaction level interface and is used for acquiring and clearing host interrupt.

The main memory reading module is connected with the host memory reading and writing transaction level module through the host memory reading and writing transaction level interface and is used for reading and writing main memory data.

The GPU software and hardware interaction TLM system based on SystemC comprises a host agent layer model and a host interface layer model, wherein the host agent layer model provides an interaction path of a software driver and the host interface layer model, the host interface layer model provides an interaction path of the host agent layer model and the GPU hardware TLM model, and the interaction content comprises issuing of an OpenGL CM (Command Packet) Command Packet, reading and writing of a system structure register, interrupt processing and loading or obtaining of image and texture data. The method can solve the interaction problem of the software driver and the GPU hardware model, and can open a software and hardware interaction channel for performing functional verification on the hardware microstructure TLM model in advance.

Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art; the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (1)

1. A Graphics Processing Unit (GPU) hardware and software interaction TLM system based on SystemC is characterized by comprising a host agent layer model and a host interface layer model, wherein the host agent layer model and the host interface layer model are used for realizing the hardware and software interaction between driving software and a GPU hardware model;

the host agent layer model is used for realizing data interaction between the driving software based on the C language and the host interface layer model based on the SystemC, and specifically comprises the following steps: the device comprises a Command issuing module, a read-write module, an interrupt processing module and a main memory reading module, wherein the Command issuing module is used for sending an OpenGL CM (Command Package) Command packet, and the Command packet carries OpenGL functional codes and related parameter information; the read-write module is used for reading and writing an architecture register, and the architecture register is used for representing the working state and the mode of the GPU;

the host interface layer model is used for realizing interaction between a GPU host interface layer model based on a SystemC transaction-level interface and a GPU PCIe hardware model, and comprises: the system comprises a host memory read-write transaction level module, an architecture register read-write transaction level module, an interrupt state read-write transaction level module and a direct memory read DMA configuration transaction level module, wherein the host memory read-write transaction level module provides a host memory read-write transaction level interface, the architecture register read-write transaction level module provides an architecture register read-write transaction level interface, the interrupt state read-write transaction level module provides an interrupt state read-write transaction level interface, and the direct memory read DMA configuration transaction level module provides a DMA configuration transaction level interface for calling a host agent layer model and a GPU hardware model;

the command issuing module is connected with the direct memory access DMA configuration transaction level module through the DMA configuration transaction level interface, and issues the CM command packet to the host interface layer model in a DMA descriptor mode, wherein the DMA descriptor comprises a DMA destination address, a DMA source address, a DMA path and a DMA length;

the read-write module is connected with the read-write transaction level module of the architecture register through the read-write transaction level interface of the architecture register and is used for configuring a GPU working mode and acquiring a GPU working state;

the interrupt processing module is connected with the interrupt state read-write transaction level module through the interrupt state read-write transaction level interface and is used for acquiring and clearing host interrupt;

the main memory reading module is connected with the host memory reading and writing transaction level module through the host memory reading and writing transaction level interface and is used for reading and writing main memory data.

Images