CN102902569B - The information processing method that soft or hard calculates grain is distinguished under heterogeneous reconfigurable environment - Google Patents

Abstract

The invention relates to an information processing method for distinguishing soft and hard computing granules in a heterogeneous reconfigurable environment, comprising the following steps: 1) defining computing granules; 2) computer description of computing granules; 3) describing computing granules using XML When , distinguish the software and hardware characteristics of all computing particles to describe the software and hardware computing particles; 4) Combination and execution of computing particles. Compared with the prior art, the present invention has the advantages of improving programming efficiency and the like.

Description

Information processing method for distinguishing soft and hard computing particles in a heterogeneous reconfigurable environment

technical field

The invention relates to a programming method, in particular to an information processing method for distinguishing soft and hard computing particles in a heterogeneous reconfigurable environment.

Background technique

The programming model is a program abstraction, characterization and simplification of parallel programs. It shields the complex hardware details of parallel computers and provides programmers with a transparent diagram of computer hardware and software systems, thus providing sufficient support for parallel program design. The parallelism of the development hardware provides convenience. The programming model, operating environment, and debugging environment of parallel programs are much more complicated than those of serial programs. Providing a good high-performance computing development environment has always been the goal pursued by academia and industry.

In the parallel environment, existing programming models include shared memory model, message passing model, data parallel model, object-oriented model and function and logic model. The corresponding specific programming method is as follows:

(1) Shared memory model

Parallel commands, work area sharing, combined constructs for parallel work sharing, master commands, and synchronization constructs.

(2) Message passing model

In the message passing model, a parallel program consists of multiple parallel processes. Each parallel holds its own data and performs computational operations on it. Data exchange between tasks is accomplished through explicit message passing statements, and users must explicitly send and receive messages to achieve data exchange between processors.

(3) Data Parallel Model

Data Parallel Model is a programming model developed for computers like SIMD, SPMD. In this model, tasks are statically or semi-statically mapped to processes, and each task performs similar operations on different data.

(4) Object-oriented model

The natural parallelism of the object-oriented model, as well as its encapsulation, inheritance, polymorphism and other characteristics, can reduce the complexity of parallel program design and improve the readability, maintainability and portability of parallel programs .

(5) Function and logic model

The basic idea of the function and logic programming model is to combine parallel processing hardware and AI software for research. Including functional programming model and logical model.

However, each programming model has its own shortcomings. Take MPI in the message passing model as an example. MPI is a standard specification of a message passing function library, which provides a non-proprietary (public) and platform-independent message passing The library has nothing to do with the language form and provides bindings with Fortran and C, which has good portability. But like most extension languages derived from Fortran or C, it lacks either extensibility or heterogeneity at the programming model level.

Contents of the invention

The object of the present invention is to provide an information processing method for distinguishing between soft and hard computing particles in a heterogeneous reconfigurable environment, in order to overcome the above-mentioned defects in the prior art.

The purpose of the present invention can be achieved through the following technical solutions:

An information processing method for distinguishing soft and hard computing particles in a heterogeneous reconfigurable environment, characterized in that it includes the following steps:

1) Definition of counting particles;

2) Computer description of counting particles;

3) When using XML to describe computing granules, distinguish the software and hardware characteristics of all computing granules to describe software and hardware computing granules;

4) Combination and execution of computing particles.

Described step 1) in the definition of counting particle comprises:

Calculation, which is a series of calculation methods to realize this calculation particle;

Storage capacity, which is the data involved in the calculation;

Communication, which refers to all data transmissions involved in the calculation process, including input and output;

There is a possible rotation relationship between computing, storage, and communication;

Calculate the parallel relationship related to particles.

The parallel relationships related to computing particles include space parallelism, time parallelism, heterogeneous parallelism and reconstruction structure parallelism.

The computing granule computer is described by using C++ language to describe the computing granule.

The combination of computing granules includes computing granules serial, computing granules parallel and computing granules nesting.

Compared with the prior art, the present invention has the following advantages:

In the system description stage, for example, when using a high-level language description, it is necessary to distinguish the software and hardware characteristics of all computing particles, and describe the software and hardware computing particles, so that the program does not need to divide the software and hardware in the later stage, but only needs to describe the computing particles. Analysis or compiler compilation, directly carry out the process of software and hardware synthesis.

Description of drawings

Fig. 1 is the flowchart of existing software and hardware division;

Fig. 2 is the flowchart of software and hardware division of the present invention;

Fig. 3 is the imitation UML representation schematic diagram of computing grain of the present invention;

Fig. 4 is a schematic diagram of the serial combination of calculation particles of the present invention;

Fig. 5 is a schematic diagram of parallel combination of counting particles of the present invention;

Fig. 6 is a schematic diagram of the nested combination of calculation particles of the present invention;

Fig. 7 is a schematic diagram of the composition elements of the software computing granule description of the present invention;

FIG. 8 is a schematic diagram of elements describing components of a hardware computing granule in the present invention.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

Example

As shown in Figure 1, generally speaking, in a heterogeneous reconfigurable environment, when performing software-hardware co-design, a unified description of the system is first performed, or the system is described uniformly using a high-level software language, or using a hardware description The language is described uniformly, and then on this basis, the software and hardware of the application program are manually or automatically divided. As shown in Figure 2, the present invention proposes that in the system description stage, for example, when using a high-level language description, the programmer is allowed to distinguish the software and hardware characteristics of all computing particles and describe the software and hardware computing particles, so that the program does not need to perform software Hardware division, and only through the analysis of the calculation particle description or compiler compilation, the software and hardware synthesis process can be directly carried out.

The programming method of the present invention includes the definition of computing granules, the computer description of computing granules, the xml description of computing granules, the combination and execution of computing granules, etc., so as to better perform programming tasks at the computing granule level. In addition, for the heterogeneous reconfiguration environment, a programming method for distinguishing soft and hard computing particles is given.

Computing grain-oriented programming method refers to a new programming method designed for the purpose of high-performance parallel computing, combined with object-oriented programming ideas, adding parallel factors, and reconstructing heterogeneous factors. Guides the process of parallel program design. The so-called computing particle-oriented refers to the mechanism of analyzing and designing parallel applications with computing particles as the core and combining high-performance parallel thinking.

In a calculation granule, calculation refers to a series of calculation methods (similar functions) to realize this calculation granule, storage capacity refers to the data involved in the calculation, and communication refers to all data involved in the calculation process, including input and output. transmission. More importantly, there is a possible rotation relationship between computing, storage, and communication, as well as parallel factors related to computing particles. All in all, a computing granule includes calculation, storage, communication, the rotation relationship between the three, and the parallel nature of the three. Figure 3 shows the imitation UML graphical representation of computing granules. As can be seen from the above figure, computing particles include four parallelisms: space parallelism, time parallelism, heterogeneous parallelism (optimized matching), and reconstructed structure parallelism (communication).

In the present invention, computing particles are described as follows in computer language by extending the C++ language:

In addition, the present invention uses XML scripting language to describe computing grains, so as to describe the execution of combined computing grains later. In order to unify the specification, the description language is agreed to be Grain Description Language (GDL). The main structure of a GDL document is something like this:

Combination of computing granules: The combination, combination and nesting of several basic computing granules to form a composite computing granule to complete a complex computing function is called the combination of computing granules, and the combined computing granules are called composite computing granules .

As shown in Figure 4, the serialization of computing granules; as shown in Figure 5, the parallel computing of computing granules; as shown in Figure 6, the nesting of computing granules. In addition, the basic calculation grains are composed of serial, parallel and nested large calculation grains.

Aiming at the heterogeneous reconfigurable environment, a programming method that distinguishes software and hardware computing granules is proposed. In this method, all computing granules are centered on computing granules, and based on the premise of software and hardware division, computing granules are divided into software computing granules and hardware computing granules. . The description composition elements of soft and hard computing grains are shown in Figures 7 and 8.

Claims (3)

1. under heterogeneous reconfigurable environment, distinguish the information processing method that soft or hard calculates grain, it is characterized in that, comprise the following steps:

1) definition of grain is calculated;

2) computing machine calculating grain describes;

3) when using XML to describe calculation grain, the software and hardware characteristic distinguishing all calculation grains carries out the description of software and hardware calculation grain;

4) combination and the execution of grain is calculated;

Described step 1) in the definition of calculation grain comprise:

Calculate, this is calculated as and realizes the series of computation method that this calculates grain;

Store, this is stored as the data participating in calculating;

Communication, this communication is all data transmission comprising input, export involved in computation process;

Calculate, store, communicating exists possible wheel between this three and transfer the registration of Party membership, etc. from one unit to another;

Calculate the concurrency relation that grain is relevant;

The combination of described calculation grain comprises the serial of calculation grain, calculation grain is parallel nested with calculation grain.

2. a kind of information processing method distinguishing soft or hard calculation grain under heterogeneous reconfigurable environment according to claim 1, it is characterized in that, it is parallel that the concurrency relation that described calculation grain is relevant comprises spatial parallelism, time parallel, isomerism parallel and reconfigured geometry.

3. a kind of information processing method distinguishing soft or hard calculation grain under heterogeneous reconfigurable environment according to claim 1, it is characterized in that, the computing machine of described calculation grain is described as describing calculation grain by C Plus Plus.