JPS64725B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for optimizing a target program using a COBOL PERFORM instruction.

There are several ways to write the COBOL PERFORM command, but the following explanation does not depend on the different ways of writing, so only the basic way of writing will be shown.

PERFORM procedure name-1 [THRU-procedure name-
2] The function of the PERFORM command is to transfer control to procedure name-1 and execute procedure name-1 (or procedure name-2).
control after executing the last instruction belonging to
It is to return after the PERFORM instruction. IBM,
For OS F ₂ /F ₄ /X ₃ , 1 is used as shown in Figure 1.
While the procedure specified by one PERFORM instruction is executing, other PERFORM instructions that share an exit with it are allowed to execute. In the figure, ,,... indicate the control transfer order.

The conventional method of compiling a PERFORM instruction is to basically generate a branch instruction that can change the destination at the exit, and in PERFORM, you just set the return address at the exit destination and jump, but that alone is not enough.
In the second PERFORM instruction in the diagram, the first
Since this will destroy the return address set by PERFORM, the actual task is to first save the exit destination, set the return address, fly, and restore the exit destination after returning.
In Figure 2, PSV ₁ and PSV ₂ are save areas, VN is the branch destination address, and GN ₁ and GN ₂ are the return destination addresses.
GO VN each indicates a branch instruction. Second
As shown in the figure, PERFORM A THRU C is converted into the instruction sequence PSV ₁ ←VN VN←GN ₁ GO A GN ₁ VN←PSV ₁ .

When PSV ₁ ← VN, the exit branch destination address is saved to the save area PSV ₁ , when VN ← GN ₁ , the return point address is stored in the memory location specified by the branch destination address, and when GO A, jumps to address A and GN ₁
VN←PSV ₁ indicates that the contents of the save area PSV ₁ are set to the storage location specified by the branch destination address VN.

PERFORM B THRU C is also converted to the instruction sequence PSV ₂ ←VN VN ←GN ₂ GO B GN ₂ VN←PSV ₂ .

In the prior art, it was not possible to detect at the time of translation whether PERFORM instructions were nested, so an expanded form like the one described above was always generated.
Therefore, in a simple case as shown in FIG. 3, redundant objects are generated, and there is room for improvement. In FIG. 3, the cross mark indicates an unnecessary command.

SUMMARY OF THE INVENTION An object of the present invention is to provide an optimization method for a target program that can be expanded to a target program without including unnecessary instructions when a program including PERFORM instructions is expanded. Therefore, the optimization method of the target program of the present invention expands the PERFORM instruction PERFORM A THRU C into the intermediate language instruction sequence PSV←VN VN ←GN GO A GN VN←PSV, and
Placing an automatic return mechanism GO VN at the exit of the intermediate language program routine corresponding to the program routine specified in the PERFORM instruction,
In a target program optimization method that expands a COBOL program containing a PERFORM instruction into an intermediate language program, optimizes the intermediate language program, and converts the optimized intermediate language program into a machine language target program, ( b) VN← on all paths leading to the automatic return mechanism
GN or VN←PSV exists; (b) The last execution on that path is VN←
It is characterized by deleting the intermediate language commands PSV←VN and VN←PSV in the expanded form of PERFORM that uses the automatic return mechanism when the two conditions of GN are satisfied. be.

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

The command marked with a cross in FIG. 3 is unnecessary when both of the following two conditions are satisfied, and even if it is deleted, it will not have any effect on the logic of the program.

(b) VN←GNx on all paths leading to the automatic return mechanism
Or VN←PSVn exists.

(b) The last thing to be executed on that path is VN←
It is GNx.

Note that the automatic return mechanism refers to the GO VN shown in Figure 3.

Next, a method for determining whether the above two conditions are satisfied will be described.

(a) Divide a program into a collection of instructions (hereinafter referred to as blocks) that have one input and one output, and within which control flows in the order in which they are written.

(b) For each block, define a bit vector 〓b corresponding to 1 bit for each VN as follows. However, Vb is a component of 〓b.

Vb = 0: At the exit of b, VN has the result of VN←GNx. 1: At the exit of b, VN has the initial value or the result of VN←PSVx. (c) For block b, set the function fb as follows. Define.

fb (V) = 0: There is an assignment of VN in b, and the end is VN←GNx 1: There is an assignment of VN, and the end is VN←PSVx V: There is no assignment of VN in b (d) The blocks immediately before control is passed to block b (called Immediate Predecessor) are b ₁ and b ₂
…, bn, Vb=fb (Vb ₁ ∨Vb ₂ ∨…∨Vbn) … Vb is found from Vb ₁ …, Vbn. However, at the entrance block of the program, Vb=fb (1).

(e) By applying the formula in order from the program entrance, Vb can be found for all blocks.
In a block where the automatic return mechanism GO VN exists, if Vb=0, conditions (1) and (2) are satisfied, and if Vb=1, they are not satisfied.

FIG. 4 is a block diagram of a general compiler.
The source program (original program) is transformed into an intermediate language (VN←
PSV or GO VN), the optimization phase described below is performed, and the code generation phase is expanded to machine instructions (load, store), etc., and becomes the target program.

FIG. 5 is a flow chart showing the processing of the optimization phase. In the block division stage, the program is divided into blocks, and each
The format of the function fb(V) is determined by checking whether there is an assignment to VN. At the flow graph creation stage, the control transition within the block is analyzed and a flow graph like the one on the left side of FIG. 6 is created. Bit Vector〓b
In the calculation step, the value of Vb is obtained by applying the function fb(V) for each block in order from the entrance.

Once the above is completed, for each VN, check the value of Vb in a certain block of GO VN, and delete all of them as there is no need to save the VN to PSV (PSV←VN) or restore (VN←PSV) Optimization is performed by

Figure 6 is a flow graph, the value of fb (V) and Vb
An example of the value of is shown.

FIG. 7 is a block diagram of a device for detecting whether or not Vb is zero in a block where an automatic return mechanism is present. In FIG. 7, 1 to 6 are unit calculation circuits, 7 is a logic "0" constant generation circuit, and 8
is a logic “1” constant generation circuit, 9 is an OR circuit, 1
0 indicates a changeover switch, and 11 indicates an output terminal. The value of the function fb corresponds to the position of this changeover switch 10. Block b ₁ of the flow graph,
Unit calculation circuits 1, ₂ ,...6 for b 2,...b _6, respectively
The position of the changeover switch 10 is set according to the value of the function fb. Then, connect these unit calculation circuits 1, 2, ..., 6 according to the flow graph and set the constant "1" to the unit calculation circuit 1 at the entrance.
Enter. Optimization is possible if the output of the unit calculation circuit 3 corresponding to the block in which GO VN exists is logic "0". This optimization process is actually performed by software, but its function is equivalent to that of the apparatus shown in FIG.

FIG. 8 shows the result of optimizing the intermediate language program of FIG.

As is clear from the above description, according to the present invention, a source program including PERFORM instructions can be developed into an efficient machine language instruction program with a small number of instructions.

[Brief explanation of the drawing]

Figure 1 shows an example of a COBOL program in which PERFORM instructions are nested.
The figure shows a conventional example of expansion of the PERFORM instruction into an intermediate language, Figure 3 shows a case where there is no need to save or recover the exit address in a program that includes the PERFORM instruction, and Figure 4 shows a general example. A configuration diagram of the compiler; Figure 5 is a flowchart showing the processing of the optimization phase in Figure 4;
The figure shows an example of a flow graph and the values of fb and Vb. Figure 7 is a block diagram of a device that processes the optimization phase. Figure 8 optimizes the intermediate language program in Figure 6. It is a figure showing the result. 1 to 6...Unit calculation circuit, 7...Logic "0" constant generation circuit, 8...Logic "1" constant generation circuit,
9...OR circuit, 10...changeover switch, 11...output terminal.

Claims (1)

[Claims] 1 PERFORM A THRU C PERFORM
In addition to expanding the instruction into the intermediate language instruction sequence PSV←VN VN ←GN GO A GN VN←PSV, the above
Placing an automatic return mechanism GO VN at the exit of the intermediate language program routine corresponding to the program routine specified in the PERFORM instruction,
In a target program optimization method that expands a COBOL program including a PERFORM instruction into an intermediate language program, optimizes the intermediate language program, and converts the optimized intermediate language program into a machine language target program, (b) VN← on all paths leading to the automatic return mechanism
GN or VN←PSV exists; (b) The last execution on that path is VN←
PSV in the expanded form of the PERFORM instruction that uses the automatic return mechanism when the two conditions of being GN and are satisfied are ←
An optimization method for the target program in the COBOL PERFORM instruction, which is characterized by deleting the intermediate language instructions VN and VN←PSV, where VN is a variable indicating the return destination address of the automatic return mechanism, and GN is the PERFORM instruction Address indicating the return point from the automatic return mechanism defined in the expanded form, PSV is the save area for the return destination address secured for each PERFORM instruction, PSV←VN is the variable VN
VN←GN is an instruction to assign address GN to variable VN, GO A is an instruction to jump to address A, VN←PSV is an instruction to assign the contents of save area PSV to variable VN. It shows.