RU2786347C1 - Method for automatic creation of a parallel program with temporal parameterization of multiprocessor computing systems with the same memory access - Google Patents

Abstract

FIELD: digital data processing.

SUBSTANCE: invention relates to the field of digital data processing. A method is claimed for automatically creating a parallel program with time parameterization of multiprocessor computing systems with the same memory access, in which the control device, the instruction fetch unit, the arithmetic logic unit perform operations on which an ordered choice of tokens of the original sequential program is obtained and descriptors are formed for the tokens, receive orderly selection of token descriptors from the set of descriptors formed at the previous stage, develop new structures of the program specification with details to operations / functions and to fragments, check the equivalence of the text specification of the task program and its representation by new specification structures, calculate priority values for operators of the new specification, form sets operators-candidates for the start of execution at a point in time, assign an operator to be implemented on a free processor, develop textual specifications for parallel threads Programs with temporal parameterization evaluate the correctness of the results of developing parallel programs with temporal parameterization, compile the created parallel code with temporal parameterization.

EFFECT: reducing the time for solving a problem by a computer system by optimizing the uniformity of processor load in the process of parallel solving a problem.

1 cl, 2 dwg

Description

The invention relates to the field of digital data processing using electronic devices, and in particular to methods for creating program code designed for multiprocessor computing systems with the same memory access (Symmetric Multiprocessors (SMP)) and can be used in programming to reduce the processing time of digital data by computing SMP systems.

A known method of automatic parallelization of programs, which consists in the fact that in the algorithmic part of the program, a control flow graph, a dominator tree, a cycle tree, a data flow graph are preliminarily obtained; perform substitutions of the intermediate representation of procedures in place calls; perform interprocedural data flow analysis; to detect equivalent operations perform data flow analysis, preferably in the way of enumeration of values; perform analysis of loop variables for invariance and inductance; perform analysis of access operations to arrays, build indexes of access to arrays in the form of canonical forms of sums of products; perform merge cycles; carry out removal of invariant conditions; change the order of traversal of the iteration space of cycles; perform parallel loop analysis [1].

There is a known method for constructing a program, which consists in defining marked cycles in the source code of the program in assembler and classifying them into several predefined types, aligning the addresses of the beginning of marked cycles, if this is required for a cycle of this type, by adding assembler instructions and, saving the source code in assembler in memory , build by compiling and linking the modified assembler code for the destination device [2].

The disadvantages of these methods are the impossibility of using for statements/functions and fragments of the source code of the program that do not contain cycles, the lack of consideration of the specific architecture for connecting processors to RAM, and the absence of consideration of the time parameter.

One of the possible ways to increase the efficiency of information processing is to organize the distribution of program data over the shared memory. However, when distributing program data over a shared memory, problems arise in the time synchronization of processors, which leads to a decrease in the efficiency of digital information processing, which must be taken into account when choosing the start time for the execution of operators during parallel execution of the original program.

The purpose of the invention is to improve the efficiency of digital information processing (reducing the time for solving the problem) by optimizing the uniformity of processor load in the process of parallel solving the problem.

This goal is achieved by the method of automatic creation of a parallel program with time parameterization of multiprocessor computing systems with the same memory access, which consists in performing the following procedures:

1. Ordered selection of lexemes of the source sequential program, determination of their belonging to one or another class of lexemes of the high-level programming language, verification of the belonging of the lexemes to the formed (at the considered point in time) set of lexemes, inclusion of the lexemes in the formed set of lexemes of the corresponding type (in case of its absence in composition of the set), forming for the lexeme its descriptor, which specifies the numerical coding of the necessary attributes.

2. An ordered selection of token descriptors from the set of descriptors formed at the previous stage, determining the high-level programming language construct corresponding to the descriptor under consideration, forming a postfix specification for this construct based on the reverse Polish notation formation method in accordance with the Dijkstra algorithm based on the use of the "stack mechanism" with priorities, which allows you to change the order of the symbols of operands and operations.

3. Development of new program specification structures that describe the source program with details to operations/functions:

- selection from the set of elements of the postfix specification of the program of a subset of operators of operations / functions of the program;

- end-to-end numbering of operators;

- end-to-end numbering of inputs for each operator and outputs;

- input of numerical coding of types of operators on the basis of postfix representation of each operation/function;

- formation for each operator of the developed numerical specification of the set of numbers of its operands and setting its power;

- formation for each operator of a set of its external operators (using the results of the operator's execution) and setting its power;

- formation, based on the postfix representation of operations/functions, for each operator of the corresponding labels.

4. Development of new program specification structures that describe the source program with details to fragments:

- selection from the set of operators of the main structure of subsets of operators that have the same value of the fragment number:

- definition for each subset of numbers and types of input and output operators, fixing their control links and corresponding labels of control transfers;

- development in a numerical format of the main, connected and temporary structure, specifying the types and scheme of control links of fragments.

The development results are the following new program specification structures at this level:

- main structure of fragments: fragment number; fragment marks; fragment types; pointers to the fragment number; cardinality of the conjugate set for the fragment; a pointer to the beginning of the sequence of fragment numbers that form the outer set of the fragment; cardinality of the outer set for the fragment; labels of fragments of an unconditional jump and a conditional jump by the value "true"; labels of fragments of a conditional transition by the value "false" of the program.

- connected structure of fragments: number of lines of the structure of links; a pointer to the continuation of the sequence of fragment numbers that form the conjugate set of the fragment for the fragment in question; the conjugate set of the fragment for the fragment in question; a pointer to the continuation of the sequence of fragment numbers that form the external fragment set for the fragment in question; the outer set of the fragment.

- temporal structure of fragments: number of vertices; graph vertex number; the point in time at which the execution of the instruction of the parallel algorithm begins, interpreted by the vertex of the corresponding temporary parallel circuit.

5. Evaluation of the correctness of the developed new specification structures of the original program in order to check the equivalence of the textual specification of the task program and its representation by new specification structures, i.e. evaluation of data types, types of operations/functions on data, connections of operations on data and control, correctness of units of measurement of physical quantities, correctness of the moments of the beginning and duration of computational operations/functions and operators of control transfers and synchronization of temporal parallel processes.

6. Calculation for the operators of the new specification of the priority value, which determines the importance of the operator in relation to other operators, from the set and, thus, the order of its consideration when solving the problem of selecting equipment for the purpose of its implementation.

In the general case, the set of operators may include operators belonging to different tasks for which the problem of parallel implementation is considered. In such cases, the overall priority of the operator is calculated, the priorities of all algorithms that have a higher priority compared to this algorithm, and the private priorities of its operators are adjusted for each of these algorithms.

For various "private" task algorithms, operator priorities can be determined in accordance with various strategies (by the maximum execution time of the branches of the algorithm containing the given operator as the initial operator of the branch; by the execution time of the operator; assigning absolute priorities to some operators; "dynamic priorities" (priority operator may change over time and be set at certain points in time, depending on the fulfillment of certain conditions)).

7. Formation of a set of operators-candidates for the start of execution at a moment of time, which consists of a set of operators, the implementation of which has not been started earlier due to the absence of this one or more operands, and a set of operators, the execution of which can be started at the current moment of time due to with the availability of all the data necessary for them, taking into account previously executed operators and information and control links between operators.

8. Selection of the candidate operator with the highest priority from the generated set, fixing the type of the corresponding operation, checking for the presence of a free processor at the time, and assigning an operator for implementation to a free processor. In the absence of a free processor necessary for the execution of an operator, such an operator is included in the list of contenders for the beginning of implementation at the next moment of time, by including it in the set formed by the procedure.

Simultaneously with the assignment of an operator for implementation with the help of the corresponding processor, this processor is transferred to the "busy" state and the moment of completion of the execution of the operator by it and the transition of the processor to the "free" state is calculated.

This sequence of actions is repeated for each operator-applicant of the set with the next highest priority. The process of assigning operators for execution ends either when the consideration of all operators of the set is completed, or when there are no free processors.

The procedure ends when the consideration and arrangement of all the task operators in the new format at the time levels is completed.

9. Development of text specifications for parallel program threads with time parameterization includes the development of the following objects:

- text specifications of time threads of a parallel program with built-in means (sleep statements) of time synchronization;

- structures of the time specification of the parallel program in the form of individual texts of the threads of the programs of the aircraft processors with built-in means (sleep statements) of time synchronization;

- estimates of the dependence of the parallel program execution time on the number of processors and the dependence of the processor load factor on the number of processors.

Initial data of the stage of development of text specifications of temporary threads of a parallel program: new structures (main, connected, temporary) of a parallel process with a temporary parameterization that meets the specified requirements (time of implementation of a parallel program on a given resource, processor load factor from the number of processors); assigning operators to threads or processors; distribution of data in the common memory of the aircraft.

The main stages in the development of textual specifications of temporary threads of programs in a parallel program include:

- cycle by numbers of operators of the main structure;

- determination of the number of the processor that implements the operator;

- formation of a text specification of an operator/operation, including:

1) determining the type of processor instruction using the main structure;

2) for the formation of the program - a selection from the connected structure of the names of conjugate operators and a record of the text specification of the operator;

3) during the formation of processor threads of programs - selection of the names of conjugate operators from the connected structure, replacement of the names of operators with their actual addresses and recording of the text specification;

4) temporal parameterization of operators of processor threads of programs;

5) presentation of text specifications of program threads with time parameterization of each of the processors for programs in the form of a set of lines of the following form: the number of commands of the processor program threads; attribute of the operation class; operation name; addresses of the first and second operands; the value of the current discrete time corresponding to the beginning of the implementation of the task operators.

10. Evaluation of the correctness of the results of the development of parallel programs with temporal parameterization, i.e. assessment of the correctness of data types, types of operations/functions on data, relations of operations on data and control, correctness of units of measurement of physical quantities, correctness of the moments of the beginning and duration of computational operations/functions and operators of control transfers and synchronization of temporary parallel processes.

11. Compiling the created parallel code with temporary parameterization.

Thus, in order to increase the efficiency of digital information processing (reduce the time for solving the problem), it is necessary to develop time threads of the program, taking into account the requirement to optimize the uniformity of the processor load in the process of parallel solving the problem, thereby ensuring the necessary time synchronization of the processors.

New features with significant differences are:

1. Parallelization of operations/functions and fragments of the source code of the program that do not contain cycles.

2. Accounting for the architecture of multiprocessor computing systems with the same memory access.

3. Accounting for the time parameter of the start of execution of instructions of the parallel algorithm.

These signs have significant differences, since they were not found in known methods.

The use of new features, in conjunction with the known ones, will improve the efficiency of digital information processing by optimizing the uniformity of processor load in the process of parallel problem solving.

A method for automatically creating a parallel program with time parameterization of multiprocessor computing systems with the same memory access is implemented as follows.

In FIG. 1 shows a diagram of the main components of a multiprocessor computing system with the same memory access consisting of two processors, which implements the proposed method. The computing system contains a main system 1, configured to optimize the load of processors, which is connected to the destination device 2 (a device designed to create a parallel code with time parameterization) via an interface 3. The process of performing operations in the main system is controlled via interface 3 using a control device 21, which controls the operation of program creation tools (compilers, linkers, etc.), which are also stored in memory 14. Information about the program creation process is shown on the display 15. The parameters for creating the program are entered into the system using the input device 13. The exchange of information in the main system is carried out through the data bus 16. At the output of the proposed method, a work program for the processors 11, 12 is obtained; it is transferred to the memory 14 of the main system 1 through the interface 3. During the operation of the main system 1, the processors 11, 12 execute a parallel program with time parameterization. Destination device 2 includes a control device 21, an instruction fetch unit 22, an arithmetic logic unit 23. Optimization according to the proposed method is aimed at more efficient loading of processors due to the time synchronization of program threads. The exchange of information in the destination device 2 is carried out through the bus 24.

Consider the step-by-step implementation of the proposed method for creating a program in the system described above (Fig. 1). The source sequential code of the program is built for the processors 11, 12 of the main system 1, through the interface 3 using the instruction fetch unit 22 of the destination device 2, an ordered selection of tokens of the source sequential program is obtained and descriptors of the tokens are formed (step 1). The instruction fetch unit 22 receives the ordered selection of token descriptors, generates a postfix specification for this construct, and stores it in memory 25 (step 2). Develops new program specification structures, down to operations/functions (step 3). Develops new program specification structures down to fragments (step 4). The control device 21 checks the equivalence of the textual specification of the task program and its representation by the new specification structures (step 5.) The arithmetic logic unit 23 calculates priority values for the operators of the new specification (step 6). The control device 21 generates a plurality of operators-candidates for the start of execution at a point in time (step 7). Assigns the operators with the highest priority for implementation to free processor 11 or 12 (step 8). The instruction fetch block 22 develops text specifications for threads of a parallel program with temporal parameterization; adds built-in means (sleep statements) of time synchronization to the code for each processor 11, 12 (step 9). The control device 21 evaluates the correctness of the results of the development of parallel programs with temporal parameterization (step 10). The control device 21 through the interface 3 commands the processors 11, 12 to compile the created parallel code with time parameterization (step 11). Below is an example of the created parallel code with temporal parameterization (Fig. 2).

Thus, the proposed method will reduce the problem solving time on two processors up to 19% when automatically creating a parallel program with time parameterization from a sequential program for multiprocessor computing systems with the same memory access, that is, to increase the efficiency of digital information processing by optimizing the uniformity of processor load in the process of parallel problem solving.

Sources of information 1. Drozdov A.Yu., Novikov S.V. Method for automatic parallelization of programs. Patent for invention No. 2411569, bul. No. 4, 2011 (analogue).

2. Yakovlev S.V., Safonov I.V., Bykova T.V. A way to build a program. Patent for invention No. 2406112, bul. No. 34, 2010 (prototype).

Claims (1)

A method for automatically creating a parallel program with temporal parameterization of multiprocessor computing systems with the same memory access, which consists in the fact that to implement the method, the control device, the instruction fetch unit and the arithmetic logic unit perform the following operations: they receive an ordered choice of tokens of the original sequential program and form for lexemes are its descriptors; receive an ordered choice of token descriptors from the set of descriptors formed at the previous stage; develop new structures of the program specification with details up to operations/functions; develop new structures of the program specification with detailing to fragments; checking the equivalence of the textual specification of the task program and its representation by new specification structures; calculate priority values for operators of the new specification; form a set of operators-candidates for the start of execution at a point in time; assign an operator for implementation to a free processor; develop text specifications of parallel program threads with temporal parameterization; evaluate the correctness of the results of the development of parallel programs with temporal parameterization; compile the created parallel code with temporary parameterization.

Images