JPS6182241A - Dividing system of procedure block - Google Patents

Abstract

PURPOSE:To decrease the number of loops over a boundary of a procedure block, and to decrease a load instruction by dividing into procedure blocks immediately before a loop containing directly a point for division, when a point to be divided to the procedure blocks is in the loop. CONSTITUTION:When executing a procedure block division at a compile time, if the present point is a dividing point, whether it is a loop inlet or not is checked, and if it is the loop inlet, a procedure block (PD) dividing procedure size is cleared by an inlet of a loop containing directly the dividing point, and if it is not the inlet, the PD dividing procedure size is cleared. If the present point is not the dividing point, whether it is the loop inlet or not is checked, and if it is the loop inlet, the inlet name is stacked in a loop state stacker, whether it is a branch instruction or not is checked, and if it is the branch instruction, a branch instruction procedure size is added to the PD dividing procedure size, and if not so, an instruction generating procedure size is added to the PD dividing procedure size. If said dividing point is a loop outlet, the loop state stacker is updated. Said work is continued until the remaining instruction is gone.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention provides a compiler, in particular, when creating an object module, a loop is created as a procedure section.
This relates to a method of dividing procedure sections that is set so as not to straddle the boundaries of e.g.

[Conventional technology and problems]

FIG. 5 explains the conventional division method of procedure partitions, where 1 is the data division, 2 is the procedure division, OM is the object module, PI)T is the procedure division table, and PDI
PD3 to PD3 respectively indicate procedure sections. The object module OM is divided into a data section 1 and a procedure section 2.

The procedure section 2 has a plurality of procedure sections PDI, PD2, PD3.
It is divided into... In the prior art, the procedure partitions PDI, PD2, PD3, . . . all have the same size, which is 4K (4096) bytes. In the data section 1, there is a procedure partition table PDT. 4. The procedure section table PDT records the starting address of the procedure section. a. When branching to procedural section PDi, L Br = A (PDi) ■B
An instruction such as Cm, D (0, Br) (2) may be required. ■The instruction in procedural section P'Di
There is an instruction to load the starting address of the base register Br into the base register Br, and an instruction (2) to branch to the address obtained by adding displacement D to the contents of the base register Br when the branch condition m is satisfied. The instruction to load the first address of the procedure section in (2) only needs to be executed once at the beginning while branches to the same procedure section continue. However, near the boundary of a procedure partition, branches into another procedure partition and branches into the current procedure partition are intermingled, and the number of pages requiring the load command (■) becomes high. In particular, if the boundary of a procedure section exists within a loop, this has a large effect because it is a part that is executed many times.

If the loop is placed in one procedure section, the number of rotary instructions in (1) can be reduced considerably, and an efficient object module can be created.

[Purpose of the invention]

The present invention is based on the above consideration, and an object of the present invention is to provide a procedure partition division method that defines procedure partitions so as to reduce the number of executions of an instruction that loads the start address of a procedure partition into a base register. It is said that

[Configuration for Achieving the Purpose] And for this reason, the method of dividing procedure sections of the present invention allows objects 1-...
In the compiler that generates the module, a procedure partition division device that divides procedure partitions and a procedure partition division device in the loop are installed.The procedure partition division device divides procedure partitions. If the current point is within a loop, control is passed to the procedural partition controller in the loop, and when control is passed, the procedural partition controller in the loop divides the procedural partition without being aware of the loop. It is characterized by performing the following.

[Embodiments of the invention]

To summarize, the present invention is such that when a point to be divided into a procedural section is inside a loop (J, the point is immediately divided into a procedural section immediately after the loop that directly contains that point. If you need to split again after splitting at that point, split at that point.

FIG. 1 is a diagram showing an overview of the compiler. The compiler 3 inputs the source program, parses it, gives it meaning,
Performs optimization, register allocation, and object generation. By the time the optimization process is completed, a loop table, etc. that stores information about loops is created. The present invention relates to the object generation process shown in FIG.

FIG. 2 is a functional block diagram of an apparatus for implementing the present invention. In FIG. 2, 3 is an object generation device;
Reference numeral 4 indicates a PD (procedure partition) dividing device, 5 a PD dividing device in a loop, and 6 a branch instruction generating device.

The process of dividing the procedure section 2 into procedure sections is performed by the PD division device 4, but if the point at which the procedure section 2 is divided into procedure sections is within a loop, control is passed from the PD division device 4 to the PD division device 5 in the loop. The PD dividing device 5 in the loop divides the procedure sections while being aware of the loop.

FIG. 3 is a diagram showing the flow of processing performed in the system of FIG. 2.

■ Check whether it is a division point of the procedure section.

When Yes, process [phase] is performed, and when No, process ■ is performed.

■ Check whether it is the entrance of a loop. If Yes, process ``■'' is performed, and if N'o, process ``■'' is performed.

■ Stack the entrance floor (Rahel) in the loop status stacker.

■ Check whether it is a branch instruction. If Yes, process (2) is performed, and if No, process (2) is performed.

■ Add the branch instruction procedure size to the PD split procedure size. This processing is performed by the branch instruction generation device 6.

■ Add the instruction generation procedure size to the PD division procedure size.

■ Check whether it is the exit of the loop. If Yes, process (2) is performed, and if No, process (2) is performed.

■ Update the loop status stacker.

■ Check to see if there are any remaining instructions. If the answer is No, it is the end, and if the answer is Yes, go back to ■.

[Phase] Check whether it is in a loop. If YES, process 0 is performed, and if NO, process 2 is performed.

Clear OPD split procedure size. The process @120 is performed by the PD dividing device 4.

@ Clear the PD split procedure size at the entrance of the loop that directly eats the split point.

0 Re-generate objects from the division point to the initial division point. The processing of @ and 0 is performed by the dividing device 5 in a loop.

FIG. 4 explains changes in the contents of the loop status stacker. Consider a double loop as shown. When processing the instruction (2), the start position (2) is set in the loop status stacker 7. This state is shown in (b). When processing the instruction (2), the start position (2) is set in the loop status stacker 7. This state is shown in (c). When processing the instruction ``■'', ``■'' is deleted from the loop status stacker 7. When a procedure section must be divided at ■, this point is included in a loop and the entrance of that loop is ■
You can find out the details from the loop status star 7ka7. If you want, you can divide it by ■ instead of using ■. ■ and ■ are loop inputs ``1'' and ■ and ■ are loop exits ``1''.
1 and 4, as shown in the loop table.

〔Effect of the invention〕

As is clear from the explanation of the above, according to the present invention, it is possible to set the boundaries of procedure partitions without reducing the number of loops compared to the conventional method. It is possible to create off-series modules with good execution performance.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an overview of a compiler, FIG. 2 is a functional block diagram showing an overview of a device for implementing the present invention, and FIG.
The figure shows the flow of processing performed in the system in Figure 2.
FIG. 4 is a diagram for explaining changes in the contents of the loop status stacker, and FIGS. 5 and 5 are diagrams for explaining the division of conventional procedure sections. DESCRIPTION OF SYMBOLS 1... Take unit, 2... Procedure unit, 3... Original L command generation device, 4... PD division device, 5... PD division device in loop, 6... Branch instruction generation device.

Claims (1)

[Claims] Objects whose procedure division is divided into multiple procedure sections
A compiler that generates a module is provided with a procedure partition division device that divides procedure partitions and a procedure partition division device in a loop, and the procedure partition division device has a procedure partition division device that divides a procedure partition when the point at which the procedure partition is divided is within the loop. If so, control is passed to a procedure partition control device in the loop, and the procedure partition control device in the loop divides the procedure partitions while being aware of the loop when control is passed. Division division method.