JPH0371233A - Optimum data allocating system for compiler - Google Patents

Abstract

PURPOSE:To increase the translation speed of a source program by allocating data in the descending order in the number of times of dynamically referencing data. CONSTITUTION:An aligning means 4 in order of the number of times of dynami cal reference obtains pointer values in a code table for respective data and aligns the pointer values in the descending order in the number of times of dynamically referencing data to form allocation order information 9 storing the aligned pointer values and the allocating means 5 in order of the number of times of dynamic reference inputs the allocating order information formed by the means 4 and allocates the data successively from the data with the highest dynamic reference data. Consequently, the output or calculation of excess object program 10 for data approximation can be omitted and the transla tion speed of the source program 7 can be increased.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a compiler for a computer system, and in particular, the present invention relates to a compiler for a computer system. Optimal data allocation in a computer system compiler that approaches data by setting relative addresses in index registers and displacements is related to a method.

[Conventional technology]

Conventionally, the data allocation method in this type of compiler is that data allocation is performed in the order of data declaration in the source program, and the data declarations are made in the order of data D, data B, data C, data A, and data E. If so, as shown in FIG. 7, data was allocated to the data area 71 in the order of data D, data B, data C, data A, and data E. For this reason,
To access data A and data B, set address C to base register # for data access of data A.
2 and an instruction to load address b into base register #1 for data access of data B had to be output.

[Problem to be solved by the invention]

The data allocation method in the conventional compiler mentioned above is based on unreferenced data or dynamic reference count (the number of times the target program is actually referenced when the target program is executed, or static reference count, which is the number of times it appears in the source program). )
If data with a small number of dynamic references is declared in the source program first, there is a high possibility that data with a large number of dynamic references will be allocated to an area larger than the maximum displacement of the data area. If the data area is allocated to an area larger than the maximum displacement of the data area, processes such as outputting a load instruction to the base register to access the data and calculating the relative address from the base register must be performed during translation. This has the disadvantage that the translation speed of the source program decreases.

Furthermore, since the number of steps in the target program increases, there is a disadvantage that the efficiency of executing the target program decreases.

Furthermore, if the data is in a repeating structure in the source program and the compiler outputs a load instruction into the repeating structure to access the data, the execution speed of the target program will be significantly reduced. It has the disadvantage of being

In particular, the above-mentioned drawback becomes significant when data with a large number of dynamic references is scattered and declared among data with no references or data with a small number of dynamic references.

In view of the above-mentioned points, an object of the present invention is to allocate data to a data area in the order of data with the highest number of dynamic references, thereby eliminating the need for outputting or calculating an extra target program for data access, and translating source programs. The objective is to provide a method for optimal data allocation in a compiler that increases speed and also increases the execution speed of the target program.

[Means to solve the problem]

The optimal data allocation method in the compiler of the present invention is as follows:
In the compiler of a computer system, data is accessed by assigning base addresses at intervals of maximum displacement, setting the base address in the base register, and setting relative addresses from the base address in the index register or displacement. The allocation that obtains the pointer values to the symbol table, sorts the pointer values in descending order of the number of times the data is dynamically referenced, and stores the sorted pointer values is performed using a dynamic sorting means in order of the number of references that generates the original information and this dynamic The layout generated by the target reference count order sorting means inputs original information, and the dynamic reference count order data allocation means performs data allocation in descending order of the number of dynamic references based on the original information.

[Effect]

In the optimal data allocation method in the compiler of the present invention, a dynamic reference number sorting means obtains the pointer value to the symbol table of each data, sorts the pointer values in descending order of the number of dynamic references of the data, and sorts the pointer values in order of the number of dynamic references of the data. Allocations that save values generate the original information, and dynamic reference count order data allocation uses the dynamic reference count order sorting means, and the generated allocations input the original information, and the allocations are dynamically based on the original information. Allocate data starting from the data that is referenced the most.

〔Example〕

Next, the present invention will be explained in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a compiler 1 to which an optimal data allocation method according to an embodiment of the present invention is applied. This compiler 1 includes a dynamic reference count sorting means 4.
A front end section 2 having a procedure section analysis section 3 including a procedure section analysis section 3, a means 5 for dynamic reference count order data allocation, and a back end section 6.

The front end unit 2 inputs a source program 7 and outputs an intermediate language 8 and original layout information 9. There are parts in the front end part 2 that share various roles, but the procedure part analysis part 3 is mainly responsible for analyzing the procedure part of the source program 7 written by the user.
In the procedure division analysis section 3, there is provided a dynamic reference count sorting means 4 for arranging the layer information 9 in the order of the dynamic reference count.

Dynamic reference count order data allocation is done by means 5, and allocation is done by layer information 9.
Data allocation is actually a means of processing based on
It is located between the front end section 2 and the bank end section 6.

The bank end unit 6 receives the intermediate language 8 output from the front end unit 2 and mainly performs code generation and output of the target program 10.

Referring to FIG. 2, the processing in the procedure division analysis unit 3 including the dynamic reference count sorting means 4 includes a procedure division reading step 21, a procedure division end determination step 22, an intermediate language generation step 23, and a symbol table. Acquisition step 24, highest level data determination step 25, highest level data symbol table acquisition step 26, registration bit on determination step 27, address registration and registration bit on step 28, repetitive structure determination step 29, Dynamic reference number count 1 count amplifier step 30, dynamic reference number count dynamic reference predicted number count amplifier step 31, allocation is order information dynamic reference number order arrangement step 3
It consists of 2.

Referring to FIG. 3, the processing in the means 5 for dynamic reference count order data allocation and the symbol table data allocation consists of step 33.

Next, the operation of the optimal data allocation method in the compiler of this embodiment configured as described above will be explained.

The front end section 2 of the compiler 1 is the source program 7
, the procedure division analysis unit 3 inputs the source program 7
A unit of the procedure division of (e.g., a unit of assignment, comparison, etc.)
is read (step 21), and if the procedure division is not finished (step 22), intermediate language 8 for the read unit is generated (step 23).

Next, the procedure division analysis unit 3 obtains a symbol table of data related to the generated intermediate language 8 (step 24). For example, if the user writes “A-B;” in the procedure division of the source program 7, and this description is the language specification of the source program 7 that assigns the contents of data B to data A, the generated intermediate The information that it is a language 8/9 assignment is associated with a symbol table for data and a symbol table for data B.

Next, the procedure division analysis unit 3 checks whether the data is the data with the highest level number (top level data) from the obtained symbol table (step 25), and determines whether the data is the data of the subordinate level. (data other than the highest level data), obtain the symbol table of the data with the highest level number (step 26). This is only possible after assigning the data with the highest level number. This is because addresses are determined for data allocation at a level subordinate to that data.

The symbol table of the data with the highest level number obtained in this way has bits whose initial value is off (hereinafter referred to as registration bits), and the procedure division analysis unit 3 analyzes the registration focus. Check whether it is in the initial state Li (off) (step 27), and if it is in the initial state, the layout is layer information 9.
The address of the symbol table is registered as a pointer to the symbol table, the pointer (address) to the layout layer information is also saved in the symbol table, and the registration focus is turned on (step 28).

If the registration bit is on in step 27, indicating that the symbol table has been registered in the allocation layer information 9, step 2
Skip 8.

Next, the procedure division analysis unit 3 determines whether the data of the symbol table currently targeted is in a repeating structure (step 29), and if not, the layout is simply performed using the dynamic data in the layer information 9. The number of references is counted by one (step 30). If the data of the symbol table currently being targeted is in a repeating structure, the procedure division analysis unit 3 calculates the reference count by one (step 30). Dynamically, the number of references increases depending on the number of repetitions, so the number of repetitions is predicted from the repetition statement, etc., and the dynamic reference number count in layer information 9 is allocated by the predicted number of references, which is the prediction result. Count up (step 31).

For example, as shown in FIG. 4, if the reference order of data in the source program 7 written by the user is data D, data B, data C, data A and data E, data D, data B .

Symbol table 41 for data C and data A. 42. 43
When data E is next referenced from the state where 44 is registered in layer information 9, the address of the symbol table 45 of data E is registered as a pointer to the fifth symbol table in layer information 9. That will happen. And the allocation is layer information 9
The registration will also be in this order, but the allocation is Junjo! In addition to the dynamic reference count in 19, data D, data B, data C, and data E that are not in the repeating structure are simply incremented by 1 each time they are referenced, but for data that is in the repeating structure. A is the number of reference predictions based on the number of repetitions, and in this example, "lO" is added.

When the procedure division of the source program 7 is completed after repeating such a series of operations (step 22), the procedure division analysis unit 3 uses the dynamic reference count sorting means 4 to determine whether the dynamic reference count is 1 or more. The layout in which all the data is registered is layer information 9.
are arranged in descending order of the number of dynamic references using the dynamic reference count as a key (step 32).

For example, if the only data referenced in the procedure division of the source program 7 are data A, data B, data C, data D, and data E, the layout will be registered in the layer information 9 for these five data. There are only symbol tables 41 to 45, and when they are arranged in order of the number of dynamic references, layer information 9 is obtained with the layout shown in FIG. Here, since the layer information 9 is no longer referred to in the layout from the symbol tables 41 to 45, the pointer to the layer information is cleared to NULL at the same time in the layout of the symbol tables 41 to 45.

When the processing in the front end section 2 is completed, the data allocation means 5 in the order of dynamic reference times inputs the layer information 9 for the layouts sorted by the dynamic reference number order means 4, and the data allocation means 5 inputs the layer information 9 for the data allocation sorted by the dynamic reference number ordering means 4. Data is sequentially allocated based on the symbol table obtained from the pointer to the symbol table (step 33).
.

As a result, the data area 61 to which data is allocated is
For example, as shown in FIG.
The instruction to load address a, which indicates the beginning of There is no need to output a load instruction to the base register.

〔Effect of the invention〕

As explained above, the present invention allocates data in the order of data with the highest number of dynamic references, thereby eliminating the need for extra target program output and calculations for accessing data.
This has the effect of increasing the translation speed of the source program.

Furthermore, since the number of steps in the target program is reduced, the execution speed is also increased.

Furthermore, since only referenced data is allocated to the data area, there is no need to process the allocation of unnecessary data, which speeds up the translation of source programs. This has the effect of improving the execution efficiency of the target program.

Furthermore, by counting the number of references dynamically rather than statically, for example, data that is referenced 5 times outside a repeating structure is allocated before data that is referenced only once in a repeating structure that is repeated 10 times. This has the effect that translation of the source program and execution of the target program become more efficient.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a compiler to which the optimal data allocation method according to an embodiment of the present invention is applied. FIG. 2 is a procedure division analysis unit including the dynamic reference count ordering means in FIG. 1. Figure 3 is a flowchart showing the process of dynamic reference count order data allocation in Figure 1, and Figure 4 is an example of the source program in Figure 1, and the allocation is based on the layer information. FIG. 5 is a diagram showing an example of the arrangement of layer information in which the layout in FIG. FIG. 7 is a diagram illustrating an example of data allocation according to the data allocation method in a conventional compiler. In the figure, 1...Compiler, 2...Front end section, 3...Procedure section analysis section, 4...Dynamic reference count ordering means, 5...Data allocation in dynamic reference count order. means, 6... back-end section, 7... source program, 8... intermediate language, 9... layout is layer information, 10... objective program, 41-45... symbol table. Figure 3 Dynamic data allocation in order of number of references is a means

Claims (1)

[Claims] A computer system that accesses data by allocating base addresses as reference points at intervals of maximum displacement, setting base addresses in base registers, and setting relative addresses from the base addresses in index registers and displacements. In the compiler, the pointer value to the symbol table of each data is obtained, the pointer values are arranged in descending order of the number of times the data is dynamically referenced, and the sorted pointer values are saved to generate allocation order information.Dynamic sorting by number of references and dynamic reference count order data allocation means for inputting the allocation order information generated by the dynamic reference count order sorting means and allocating data in descending order of the number of dynamic references based on the allocation order information. An optimal data allocation method for a compiler characterized by: