JPH01217632A - System for allocating register to data - Google Patents

Abstract

PURPOSE:To improve the availability of registers by allocating these registers among blocks related with each other like the brethren in a shared form. CONSTITUTION:An information collecting process part 1 checks and stores the application frequency of each data on a source program at every block. Then a register allocation processing part 2 allocates the registers to the data having higher application frequencies while sharing those data allocated to data among the blocks related with each other like the brethren based on the application frequency of each data stored in the part 1. A prolog/epilog processing part 3 saves the contents of the shared register at the prolog of the block between the blocks related with each other like the brethren based on the allocation carried out by the part 2. Then an object is produced so that the contents of the shared register are recovered at the epilog between the blocks related with each other like the brethren based on the allocation of the part 2. Thus the availability of registers can be improved.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a register allocation method for data, and in particular object generation (generation of a target program) of a compiler that handles a source program written in a language with a block structure. The register allocation to data in is related to the method.

[Conventional technology]

Conventionally, the method for register allocation for this type of data was to check the frequency of use (number of references) of each data in the source program, regardless of the block structure, and to assign registers to frequently used data based on the frequency of use of each data. Registers were allocated preferentially (data for which registers were not allocated will be handled in memory during execution).

[Problem to be solved by the invention]

In the conventional register allocation method for data described above, the frequency of use of each data is checked regardless of the block structure, and registers are allocated preferentially to frequently used data based on the frequency of use of each data. Therefore, when a register is allocated to data in a certain block, the allocated register is used only within that block (the register is not used in other blocks that are siblings of the block), and the data Inefficient use of registers allocated to The disadvantage is that the lack of such a system means that the usage efficiency is poor.

In view of the above-mentioned points, an object of the present invention is to provide a method for allocating registers to data in which registers allocated to data are shared between blocks having a sibling relationship, and register usage efficiency is improved.

[Means to solve the problem]

The register allocation method for data in the present invention consists of an information collection processing section that investigates and stores the frequency of use of each data in the source program in block units, and an information collection processing section that investigates and stores the frequency of use of each data stored in the source program. Register allocation, which allocates registers for frequently used data while sharing registers allocated to data between blocks that have a sibling relationship, is a process between the processing unit and the block that is sibling to the block in the prologue of the block. In the epilogue of the block, the contents of the registers that are shared based on the allocation of the processing unit are saved, and the register allocation is shared between the block and the blocks that have a sibling relationship based on the allocation of the processing unit. It has a prologue/epilogue processing unit that generates objects so as to restore the contents of the registers.

[Effect]

In the method of register allocation to data of the present invention, the information collection processing section investigates and stores the frequency of use of each data in the source program in block units, and the processing section performs register allocation for each data stored by the information collection processing section. The prologue/epilogue processing unit allocates registers to frequently used data while sharing the registers allocated to data between blocks that have a sibling relationship based on the frequency of use of the data, and the prologue/epilogue processing unit has a sibling relationship with the block in the prologue of the block. The contents of the shared registers are saved based on the processing unit's allocation with the block in question, and in the epilogue of the block, the register allocation between the block and the sibling block is based on the processing unit’s allocation. Objects are generated to recover the contents of the shared registers based on the information.

〔Example〕

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

FIG. 1 is a block diagram showing the configuration of an embodiment of the register allocation method for data according to the present invention.

Register allocation to data in this embodiment is performed by an information collection processing unit 1 that investigates and stores the frequency of use of each data in the source program block by block (sets it in the information table 4); Registers are allocated to frequently used data while sharing registers allocated to data between blocks in a sibling relationship using the usage frequency of each stored data (block structure of the language used to write source programs and Registers are allocated to data based on the frequency of use of each data in information table 4) Register allocation is done by processing unit 2, the contents of registers allocated to data are saved in the prologue of the block, and the contents of the registers allocated to the data are saved in the epilogue of the block. The program includes a prologue/epilogue processing unit 3 that generates objects so as to recover the contents of registers allocated to the source program, and an information table 4 that stores block information and frequency of use regarding each data of the source program. .

Referring to FIG. 2, the processing of the information collection processing section 1 consists of a data extraction block information setting step 11 and a usage frequency setting step 12.

Referring to FIG. 3, the processing of the register allocation processing section 2 consists of step 21 for allocation of frequently used data registers, and step 22 for reallocation of sibling block data registers.

Referring to FIG. 4, the prologue/epilogue processing unit 3
The process includes a register content save recovery object generation step 31.

Referring to FIG. 5, the register allocation is set in the information collection processing unit 1, the register allocation is referenced and set in the processing unit 2, and the register allocation is referenced in the prologue/epilogue processing unit 3. There are entries for the number of possible pieces of data (entries such as "information for data A" and "information for data B" in FIG. 5). The entry for each data includes a name for identifying the data, the frequency of use of the data set by the information collection processing unit 1, and the block in which the data is defined (the data is included). Block information including the block number of the block (lock) and the block number of the parent block of that block, the presence or absence of register allocation indicating whether a register has been allocated to the data, and the register number to identify the allocated register. It consists of the register information that it has.

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

When an object is generated by a compiler based on a source program written in a language with a block structure, the information collection processing unit 1 collects data to be allocated to registers in blocks in the source program data. Extract the block number of that block and the block number of the parent block of that block (parent block number)
is set in the block information in a predetermined entry (for each data) in the information table 4 (step 11).

Further, the information collection processing unit 1 counts the frequency of use of the data extracted in step 11 in the source program and sets it in a predetermined entry in the information table 4 (step 12).

The register allocation processing unit 2, to which control is passed from the information collection processing unit 1, collects each data included in one of the direct child blocks of the highest parent block and creates an information table. The usage frequency and block information of each data set in 4 ("usage frequency" and "block information" in a predetermined entry of information table 4 corresponding to each data (identified by name tag)) are displayed in the information table. 4 and allocates registers to frequently used data (registers are allocated to "register information" in a predetermined entry in information table 4 corresponding to each data (register allocation is (set to "Presence/Presence") and identification information of the allocated register (set to "Register Number") (step 21).

Next, the register allocation processing unit 2 calculates the frequency of use of each data set in the information table 4 and the block information (this block The presence or absence of a sibling relationship can be checked based on the information) is obtained from the information table 4, and the registers allocated in step 21 are allocated again to frequently used data (step 22).

The processes of steps 21 and 22 described above are repeated up to the lowest child block with the child block as the parent block.

The prologue/epilogue processing unit 3, to which control is passed from the processing unit 2 for register allocation, saves the contents of the registers shared between the block and sibling blocks in the prologue of each block, and Generate a target program that, in the epilogue, restores the contents saved in the prologue to the registers shared by the relevant block and sibling blocks (this saving and recovery will cause data contained in other blocks to be restored). (preventing the value from being destroyed) (step 31).

In this way, the registers allocated to the data included in each block can be shared between the blocks having a sibling relationship, and the efficiency of register use can be improved.

〔Effect of the invention〕

As explained above, in the present invention, by sharing and allocating registers between blocks that have a sibling relationship, the range of use of registers is expanded and one register can be allocated for multiple data), registers This has the effect of increasing the amount of data that can be allocated and improving register usage efficiency.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention, FIG. 2 is a flowchart showing the processing of the information collection processing section in FIG. 1, and FIG. 3 is a register allocation shown in FIG. 4 is a flowchart showing the processing of the prologue/epilogue processing section in FIG. 1; FIG. 5 is a diagram showing the configuration of the information table in FIG. 1. In the figure, 1... information collection processing section, 2... register allocation processing section, 3... prologue/epilogue processing section, 4... information table.

Claims (1)

[Scope of Claims] An information collection processing unit that investigates and stores the usage frequency of each data block in the source program, and an information collection processing unit that investigates and stores the usage frequency of each data block in the source program, and an information collection processing unit that allocates data using the usage frequency of each data item stored by the information collection processing unit. a register allocation processing unit that allocates registers for frequently used data while sharing the registers between blocks that are in a sibling relationship; and a register allocation processing unit that allocates registers to frequently used data while sharing the registers between the blocks that are in a sibling relationship; save the contents of the registers that are shared based on the allocation of the block, and restore the contents of the registers that are shared based on the allocation of the register allocation processing section between the block and the sibling blocks in the epilogue of the block. 1. A register allocation method for data, comprising: a prologue/epilogue processing unit that generates an object at a time; and a register allocation method for data.