JP2002099424A - Compile method to use register in storage area - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a compile method to reduce memory references which occur when there is no register to be assigned to a single precision variable. SOLUTION: In a process to assign a register to a variable, when there is no register to be assigned to the single precision variable, it is checked whether a register usable only for a double precision is allowed to be assigned in a storage area of a single precision register, if allowed, the register usable only for the double precision is assigned.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compiling method for reducing the number of references to a main memory (memory), and more particularly to a register allocating method.

[0002]

2. Description of the Related Art The performance of microprocessors has been improving year by year due to improvements in instruction level parallelism and operating frequencies. On the other hand, the performance of the main storage device (memory) cannot keep up with the improvement of the processor performance. Therefore, when the number of references to the memory increases, the memory performance may become a bottleneck and the execution performance may decrease. Therefore, in a compiler that converts a source program into an object program, it is important to reduce the number of times of referring to a memory in order to improve execution performance.

[0003] In a compiler for converting a source program into an object program, a number of optimization methods for improving the execution performance of the object program have been developed so far, but one of the optimization methods is one of them. Register allocation processing to reduce the number of memory references and improve the execution performance of object programs by executing the variables referenced in the source program while holding the data in registers, which are limited resources Exists.

[0004] This technology is described in "AV Eho JD.
By Ullman / Translated by Norihisa Doi: Compiler, 1986, Inc.
Baifukan, pages 485 to 488 ". In this method, if there is no register to be allocated to a certain type of variable in the register allocation process, a storage area is reserved in memory to hold the value of the variable,
A memory reference instruction for retrieval was generated.

[0005]

According to the first prior art, since only registers that can be used for a certain type are allocated to variables of a certain type, single-precision variables can be used with single precision. Registers are assigned to registers, and registers that can be used in double precision are assigned to double precision variables. Therefore, when a register that can be used for single precision is not allocated when allocating a register to a single precision variable, a memory reference instruction is generated even if there is a free space in the double precision register.

An object of the present invention is to use a register that can be used only in double precision as a storage destination of a register that can be used in single precision when there is a register that can be used only in double precision. An object of the present invention is to provide a compiling method for reducing the number of memory reference instructions generated when no assignment is made and improving the execution performance of an object program.

[0007]

SUMMARY OF THE INVENTION An object of the present invention is to determine whether a register that can be used only for double-precision is allocated when there is no register to be allocated to a single-precision variable. This is achieved by allocating registers that can only be used.

[0008]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a configuration diagram of a computer system on which a compiler according to the present invention operates. The computer system is
CPU 101, display device 102, keyboard 1
03, a main storage device 104, and an external storage device 107. The main storage device 104 has a compiler 105
And the intermediate code 106 required in the compilation process
Is held. The external storage device 107 stores a source program 108 and an object program 109. Compile processing is performed by the compiler 105
01.

FIG. 2 is a flowchart showing an example of the compiling process performed by the compiler 105.

The compiler 105 includes a syntax analysis 201, a register allocation 202, and a code generator 203. In the syntax analysis 201, the source program 108 is read, and the intermediate code 106 that can be processed inside the compiler is created. In the register allocation 202, the intermediate code 10
6 is read, reference information of variables appearing in the intermediate language is collected, registers are allocated, and then each reference is converted into a register reference or a memory reference.
Generate In the code generation 203, the intermediate code 106
Is read, and an object program 109 is generated.

FIG. 6 shows a part of a FORTRAN program in which variables A, B, C, and D are declared in a single-precision floating-point type. At a certain point in the program, the result of adding the variables B and C is referred to as a variable A. This indicates that the result of addition of the variable D and the variable A is stored in the variable C after that.

FIG. 3 is a flowchart showing an example of the processing of the register allocation 202.

First, in step 301, the intermediate code 10
Register the variables to be assigned to registers starting from 6.
Next, in step 302, live sections of all registered variables are calculated. The live section refers to a section from when a variable is defined to when it is used. For example, in the live section of the variable A in FIG.
And is used in (704), so (702)
To (704). Next, in step 303, it is checked whether or not there is a register that is not used in the live section of the variable in that section, and if there is a register that is not used, the register is allocated. If there is no unused register, a storage area for storing the value of the variable is secured. Finally, in step 304, the variable reference
When a register is assigned to a variable, the register is referred to. When a register is not assigned to the variable and a storage area is reserved, the storage area is referred to. This reference to the storage area is a reference to the memory.

Here, register allocation in the conventional compiler will be described with reference to FIG. The process of examining a register assigned to the variable A is achieved by examining a register that is not used in the live section of the variable. In this case, there are three registers fr1, fr2, and fr3 that can be used for single precision, where fr1 is (701) to (705), and fr2 is (701) to (7).
03), (704) to (705), and fr3 is (70)
This indicates that the register dr1 already used in 1) to (704) and available in double precision is not used in the section from (701) to (705). In such a case, the variable A
Since there is no register that can be used for single precision in the live section of the above, the compiler of the conventional method generates an instruction for storing the variable from the register to the memory at the definition point and extracting the variable from the memory to the register at the point of use. At this time, a register used for communication with the memory is referred to as a work register. A work register is a register used when storing or retrieving the value of a variable for which a register could not be allocated to or from a memory.

FIG. 8 shows an example of an object program when the source program of FIG. 6 is compiled by a conventional compiler. The object program in FIG. 8 is generated when there is no register to be assigned to the variable A.

When there is no register to be assigned to a variable, the value of the variable is stored from the work register to the memory at the definition point, and fetched from the memory to the work register at the point of use. At the definition point 602, in order to store the defined value in the memory, the value of the variable A defined as a result of the addition is assigned to the work register (fr0) (801). Then, an instruction for storing the contents of the work register in the memory is generated (802). At use point 603, an instruction to retrieve a value from memory to a work register is generated to retrieve the value from memory (803). Then, an operation is performed using the work register (804).

FIG. 4 is a flowchart showing an example of the process 303 for allocating registers to variables according to the method of the present invention.

First, at step 401, it is checked whether or not a register corresponding to a type of a certain variable is allocated. If there is a register to be allocated, at step 405, the register is allocated and the processing is terminated. If there is no register to be allocated, step 402 checks whether the type of the variable is single precision. If the variable is not a single-precision variable, a storage area for storing the value of the variable is secured in step 404, and the process ends. In the case of a single-precision variable, it is checked in step 403 whether there is a register that can be used only for double-precision and that can be used as a variable storage area. This is achieved by checking in the live section of the variable if there is a register available only for double precision that is not being used. If there are registers available, step 405
If there are no available registers, a storage area is reserved in step 404 and the processing is terminated.

FIG. 5 is a flowchart showing an example of the process 304 for converting a reference to a variable into a reference to a register or a storage area.

First, in step 501, it is checked whether or not a register corresponding to a certain variable type is allocated. If such a register has been allocated, the reference of the variable is converted into the reference of the register in step 502, and the processing is terminated. If a register corresponding to the type of the variable has not been assigned, it is checked in step 503 whether a single-precision variable is assigned a register that can be used only in double precision.
If a register has not been allocated, the reference of the variable is converted into the reference of the storage area in step 505, and the processing is terminated. If a register that can be used only in double precision is assigned to the single precision variable, the process of step 504 is performed. Step 5
In step 04, first, the reference point of the variable is converted into the reference of the work register, and then at the definition point, a transfer instruction from the work register to the allocated register is created and inserted immediately after the definition point. At the point of use, a transfer instruction from the assigned register to the work register is created and inserted immediately before the point of use.

FIG. 9 is an example of an object program generated when the compiler of the present invention is applied to the source program of FIG.

The live section of variable A in FIG.
In (2) to (704), there are no available registers available for single precision, but registers (dr) available only for double precision
It can be seen that 1) has a vacancy. Therefore, it is assumed that a register (dr1) that can be used only for double precision is assigned to the variable A. Then, the memory reference instructions (802) and (803), which have been generated in the past, become the transfer instructions (902) and (904) between the registers, and the memory reference is eliminated.

[0024]

According to the present invention, in a register allocation process in a system having a register which can be used in single precision and a register which can be used only in double precision, a register which can be used in single precision only in a register which can be used in double precision only. When there is no register to be assigned to a single-precision variable, the number of times of referring to the memory can be reduced, which leads to an improvement in the execution performance of the object code.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a system on which a compiler operates.

FIG. 2 is a diagram showing a processing procedure of a compiler.

FIG. 3 is a diagram showing a processing procedure of register allocation.

FIG. 4 is a diagram showing a processing procedure for allocating registers to variables.

FIG. 5 is a diagram showing a processing procedure for converting a reference of a variable into a reference of a register or a storage area.

FIG. 6 is a diagram showing an example of a source program.

FIG. 7 is a diagram showing a live section of a variable.

FIG. 8 is a diagram showing an object code according to a conventional method.

FIG. 9 is a diagram showing an object code when the present invention is applied.

[Explanation of symbols]

401: a process for checking whether there is a register assigned to a variable; 402: a process for checking whether the type of a variable is single-precision; 403: a register that can be used for a storage area of a register that can be used for single-precision and that can be used only for double-precision 404... Processing for reserving a variable storage area, 405... Processing for allocating registers to variables.

Claims (4)

[Claims] 1. A first register group in which operation instructions for a first type are prepared, a second register group in which operation instructions for a second type are prepared, and a first register group and a second register group. A compiler for a processor having a transfer instruction to and from a second group of registers, wherein the first group of registers is used as a storage area for a variable of a second type.
Compilation method. 2. The compiling method according to claim 1, wherein the first type is a double-precision floating-point type, and the second type is a single-precision floating-point type. 3. A compiler using the compiling method according to claim 1. 4. A storage medium storing the compiler according to claim 3.