JPH0417031A - System for optimizing propagation of constant value - Google Patents

Abstract

PURPOSE:To reduce the size of an object program and to increase the executing speed of it by propagating the constant value of an operand to the operands of the following instructions against an instruction within a range in which the control is linearly shifted. CONSTITUTION:A constant value propagating means 3 propagates a constant value among the operands of transferring instructions by analyzing instructions and operands at every instruction block divided by an instruction block dividing means and replaces an transferring instruction or calculating instruction from a register and memory having the same value as the constant value with the transferring instruction of the constant value. An unnecessary instruction deleting means 3 deletes an unnecessary instruction in an instruction block for which instruction replacement is made by means of the means 3 and an instruction output means 5 outputs the instruction train for which constant value propagation and unnecessary instruction deletion optimization are respectively made by the means 3 and 4 to an instruction storing medium 10. Therefore, the reduction of size and increase of executing speed of an object program can be performed.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a constant value propagation optimization method, and particularly to - optimization at the machine language instruction level on an architecture with an instruction set in which all instructions have the same size. This paper concerns a constant value propagation optimization method that performs optimization.

[Conventional technology]

Optimization performed on intermediate text within conventional compilers largely depends on the source program description. Furthermore, even with optimization at the machine language instruction level that does not depend on the description of the source program, there has been no optimization that takes into account the values of constants themselves. In particular, there has been no optimization method that propagates constant values to machine language instructions after the machine language instructions have been generated.

[Problem to be solved by the invention]

In the conventional optimization method described above, there was no optimization that was conscious of the constant value itself at the machine language instruction level, so the generated machine language instructions are not necessarily the optimal instruction sequence, but are based on the object program. It has the disadvantage of increasing size and decreasing execution speed.

Another drawback is that multiple compilers implement similar optimization processing, resulting in waste.

In view of the above-mentioned points, an object of the present invention is to make constant calculation possible at compile time by propagating constant values of operands to operands of subsequent instructions using instruction codes for instructions within the range of linear control transfer. and replace it with a faster instruction, or replace an unnecessary instruction that becomes unnecessary due to instruction replacement (the former setting instruction when the operand set by that instruction is set again by another instruction before it is referenced later). An object of the present invention is to provide a constant value propagation optimization method that reduces the size of an object program and shortens the execution speed by deleting the object program.

[Means to solve the problem]

The constant value propagation optimization method of the present invention provides an instruction input means for inputting a sequence of instructions from an instruction storage medium and expanding them onto memory on an architecture having an instruction set in which all instructions have the same size. , an instruction block dividing means for creating an instruction block by analyzing the instructions expanded on the memory by the instruction input means and dividing the sequence of instructions immediately after the branch instruction and at the branch destination label position, and this instruction block dividing means. Analyzes instructions and their operands for each instruction block divided by means, propagates constant values between operands of transfer instructions, replaces transfer instructions from registers and memory with the same value as the constant value, and performs operations such as replacing transfer instructions with transfer instructions with constant values. A constant value propagation means that performs constant operation and replaces the operation instruction with a transfer instruction of the constant value when all reference operands of an instruction are constant values and registers and memories that are equivalent to the constant values, and this constant value propagation means An unnecessary instruction deletion means for deleting unnecessary instructions in the instruction block in which the instructions have been replaced, and an instruction sequence that has been optimized for constant value propagation by the constant value propagation means and unnecessary instruction deletion by the unnecessary instruction deletion means. and command output means for outputting to a storage medium.

[Effect]

In the constant value propagation optimization method of the present invention, the instruction input means inputs a sequence of instructions from the instruction storage medium and expands it onto the memory, and the instruction block division means analyzes the instructions expanded on the memory by the instruction input means. The instruction block is created by dividing the instruction sequence immediately after the branch instruction and at the branch destination label position, and the constant value propagation means analyzes the instructions and their operands for each instruction block divided by the instruction block division means. propagate constant values between the operands of transfer instructions, and replace transfer instructions from registers and memory with constant values with transfer instructions with constant values; In the case of memory, a constant operation is performed to replace the operation instruction with a constant value transfer instruction, and the unnecessary instruction deletion means deletes the unnecessary instruction in the instruction block in which the instruction has been replaced by the constant value propagation means, and the instruction The output means outputs to the instruction storage medium an instruction sequence that has been optimized for constant value propagation by the constant value propagation means and for unnecessary instruction deletion by the unnecessary instruction deletion means.

〔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 constant value propagation optimization method according to an embodiment of the present invention. The constant value propagation optimization method of this embodiment includes an instruction input means 1, an instruction block division means 2, a constant value propagation means 3, an unnecessary instruction deletion means 4, an instruction output means 5, and an instruction expansion memory 6. , a plurality of prosock tables 7, a register table 8, a memory table 9, and an instruction storage medium IO consisting of files, memories, etc., and storing instructions at the machine language instruction level.

Referring to FIG. 2, each block table 7 is composed of an address sequential chain and an instruction pointer.
The addresses are chained in sequential order from the block table starting point pointer. The instruction pointer of each block table 7 is the address of the instruction on the instruction expansion memory 6 or the relative offset within the expanded instruction on the instruction expansion memory 6, and is the address of the instruction on the instruction expansion memory 6, or the relative offset within the expanded instruction on the instruction expansion memory 6. It points to the first address.

Referring to FIG. 3, the register table 8 is composed of a plurality of entries consisting of register numbers, setting flags, and constant values.

Referring to FIG. 4, the memory table 9 is composed of a plurality of entries consisting of memory information consisting of addresses and lengths and constant values.

Next, the operation of the constant value propagation optimization method of this embodiment configured as described above will be explained.

First, the instruction input means 1 inputs instructions one by one from any instruction storage medium 10 storing instructions, expands them onto the instruction expansion memory 6, and then calls the instruction block division means 2.

The instruction block dividing means 2 analyzes the instruction code and performs instruction block dividing processing only when the instruction is a branch instruction. An instruction block is a unit of instructions that has one control entrance, one exit, and no branches other than the exit.The divided instruction block is shown in Figure 2. As shown, the block table 7 is designated by pointing the first instruction of the instruction block in the instruction expansion memory 6 with an instruction pointer. In the instruction block division process, the instruction block is divided immediately after the branch instruction and at the branch destination label position by the following processes 2.1 and 2.2.

Processing↓Upper: Searches the block table 7 where the instruction pointer points to the instruction immediately after the branch instruction, and if it already exists, does nothing. If not, create one block table 7, set the instruction pointer to the instruction immediately after the branch instruction as the first instruction of the instruction block, and use address sequential chaining to chain the block table 7 in the order of the instructions' addresses. Chains with block table 7 immediately after.

Process 1 - Search the block table 7 pointing to the branch destination instruction from the branch relative offset of the branch instruction, and chain with the immediately preceding and succeeding block tables 7 in the order of the address of the instruction, as in process 2.1.

When the processing by the instruction input means 1 and the instruction block division means 2 is repeated and all the instructions in the instruction storage medium 10 are processed, the instruction sequence divided into blocks by the plurality of block tables 7 is stored on the instruction expansion memory 6. Be expanded.

Next, the constant value propagation means 3 follows the block table 7 according to the address sequential chain, and uses the register table 8 and the memory table 9 to perform constant value propagation processing on instructions in the instruction block that linearly transfer control. .

The constant value propagation means 3 performs the following process 3.1 and process 3.
Repeat either 2 or 3.3.

113.1 The block table 7 is traced according to the near address sequential chain to determine the instruction block to be analyzed.

When there are no more instruction blocks to be analyzed, the processing of the constant value propagation means 3 ends.

Once the instruction block to be analyzed is determined, the setting flags of all entries in the register table 8 are turned off, and all entries in the memory table 9 are deleted.

Then, for the instruction block to be analyzed, the instructions in the instruction block are traced in the order of control transfer to determine the instruction to be analyzed, and processing 3.2 or 3.3 is repeatedly performed. If even one instruction is replaced during the processing of the instruction block to be analyzed, after completing the repetition of processing 3.2 or 3.3 for all instructions in the instruction block, the unnecessary instruction deletion means 4 is executed. call.

Process 1 Yu↓: When the instruction to be analyzed is a transfer instruction, if the source operand has constant value information, the constant value is propagated to the destination operand. If the source operand does not have constant value information, the constant value information held by the destination operand is cleared. Specifically, the register table 8 and memory table 9 are used to perform the following processing.

■ Classify the transfer source operands and find the constant value C to be propagated for each as follows.

In the case of a constant: Find the value C of the constant.

In the case of a register, if the setting flag of the corresponding entry in register table 8 is on, the constant value C of the entry is determined.

In the case of memory: If there is an entry with the same memory information in the memory table 9, the constant value C set in that entry is determined.

(2) When the constant value C is obtained in (2), the destination operands are classified and the constant value C is propagated to each as follows.

In the case of a register, a constant value C is set in the constant value field of the corresponding entry in the second register table 8, and the setting flag is turned on.

In the case of memory: Among the entries in the memory table 9, memory entries that may access part or all of the same area as the memory are deleted. Furthermore, if there is an entry with the same memory information in the memory table 9, a constant value C is set in the constant value field of that entry, and if there is not, a new entry is added, and the memory information and constant value C are set.
and set.

■ If a constant value C is obtained in ■ and the source operand is not a constant, an instruction (called a constant transfer instruction) exists in the architecture that transfers the constant value C to the destination operand, and the source operand is not a constant. Comparing the execution speeds of the transfer instruction and the constant transfer instruction, if the latter is faster, the original transfer instruction in the instruction expansion memory 6 is replaced with the constant transfer instruction.

(2) If constant value C cannot be obtained in (2), the destination operands are classified and the constant information for each is cleared as shown below.

In the case of a register, the setting flag of the corresponding entry in register table 8 is turned off.

In the case of memory: If there is an entry with the same memory information in the memory table 9, that entry is deleted. Also,
Memory entries that may access part or all of the same area as the memory are also deleted.

Processing ↓ - Main: If the instruction to be analyzed is an arithmetic instruction, and all operand operands have constant value information, perform a constant operation and convert the arithmetic instruction to be analyzed to the operation result from the constant that is the result of the constant operation. Replace with a transfer instruction to the storage operand. If any of the operated operands does not have constant value information, the constant value information held by the operation result storage operand is cleared. Specifically, the register table 8 and memory table 9 are used to perform the following processing.

■ The operand of the operation instruction is Qpi (i=1
．． 2. ..., nun≧1). When all of the operands OPi satisfy any of the following conditions 1a) to tc+, the constant value Ci (i=1.2...n
) is subjected to a constant operation by the operation indicated by the operation instruction, and the result is set as a constant value R.

fa) OPi is a constant (constant value is Ci) [blOP
i is a register, and the setting flag of the corresponding entry in register table 8 is on (the constant value of the entry is set to Ci
) (clOPi is a memory, and there is an entry with the same memory information in the memory table 9 (the constant value of this entry is Ci) If the constant value R is found by satisfying the conditions of ■ ■, then
The arithmetic instruction to be analyzed in the instruction expansion memory 6 is set to a constant value R.
Replace it with a constant transfer instruction that transfers it to the operation result storage operand.

■ If the constant value R is found by satisfying the conditions of ■, then
Depending on the type of operation result storage operand, process 3.
Processing is performed in which the constant value C in 2.2 is replaced with the constant value R.

(2) If the condition (2) is not satisfied, the process (2) in Process 3.2 is performed.

The unnecessary instruction deletion means 4 performs processing 4 and (2) on the instruction block to be analyzed, and further repeats processing 4.2 to 4.4, and deletes the operand again before the operand set in the instruction block to be analyzed is referenced. Find the case where it is set,
The previous setting command is considered unnecessary and deleted.

On the m unit, the setting flags of all entries in the register table 8 are turned off, and all entries in the memory table 9 are deleted.

11↓-1 = Follow the instructions in the instruction block to be analyzed in the reverse order of control transition, decide the instruction to be analyzed, and process 4.3
and 4.4.

m-engineering □: Classify the setting operands of the instruction to be analyzed, and if the operands have been set before being referenced later, as shown below, consider them unnecessary instructions and delete them.
Otherwise, information indicating that a setting has been made for that operand is left.

In the case of a register, if the setting flag of the corresponding entry in the second register table 8 is off, it is turned on.

If the setting flag is on, the instruction to be analyzed in the instruction expansion memory 6 is replaced with an N0P (No 0 Peratjon) instruction.

In the case of memory: If there is no entry with the same memory information in the memory table 9, a new entry is added and the memory information is set. If so, the instruction to be analyzed in the instruction expansion memory 6 is replaced with a NOP instruction.

U↓U soil: Classify the reference operands of the instruction to be analyzed, and leave information indicating that there is a reference to each operand as shown below.

In the case of a register, the setting flag of the corresponding entry in register table 8 is turned off.

In case of failure: If there is no entry with the same memory information in the memory table 9, nothing is done. If so, delete that entry. Also, memory entries that may access part or all of the same area as the memory are also deleted.

Finally, the instruction output means 5 transfers the instructions in the instruction expansion memory 6 that have been optimized by the constant value propagation means 3 and the unnecessary instruction deletion means 4 to any instruction storage medium 1 that has stored the instructions.
Output to.

〔Effect of the invention〕

As explained above, the present invention performs constant calculation possible at compile time by propagating constant values of operands for instructions within the range of linear control transfer to the operands of subsequent instructions by instruction code, By replacing instructions with faster instructions or deleting unnecessary instructions that become unnecessary due to instruction replacement, it is possible to reduce the size of the object program and shorten the execution speed.

In addition, by applying this method to machine language instructions, it becomes possible to apply it universally to compilers on the same architecture, and furthermore, it becomes possible to apply it to object programs output by compilers. There is.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of a constant value propagation optimization method according to an embodiment of the present invention; FIG. 2 is a diagram showing the contents of the instruction expansion memory and block table in FIG. 1; FIG. 3 is a diagram showing the contents of the instruction expansion memory and block table in FIG. FIG. 4 is a diagram showing the contents of the register table in FIG. 1, and FIG. 4 is a diagram showing the contents of the memory table in FIG. In the figure, 1... Instruction input means, 2... Instruction block division means, 3... Constant value propagation means, 4... Unnecessary instruction deletion means, 5... Instruction output means, 6... Instruction expansion memory, 7゛...Block table, 8...Register table, 9...Memory table, 10...Instruction storage medium.

Claims (1)

[Claims] In an architecture having an instruction set in which each instruction has the same size, an instruction input means for inputting a sequence of instructions from an instruction storage medium and expanding it on a memory; an instruction block dividing means for creating an instruction block by analyzing the instructions expanded on the memory by dividing the sequence of instructions immediately after the branch instruction and at the branch destination label position; It analyzes instructions and their operands for each instruction block, propagates constant values between operands of transfer instructions, replaces transfer instructions from registers and memory with the same value as a constant value, and replaces reference operands of arithmetic instructions with transfer instructions with constant values. A constant value propagation means that performs a constant operation and replaces an operation instruction with a transfer instruction of a constant value in the case of all constant values and registers and memories that have the same value as the constant value; and a constant value propagation means that replaces instructions. unnecessary instruction deletion means for deleting unnecessary instructions in the instruction block; and outputting an instruction string to an instruction storage medium after optimization of constant value propagation by the constant value propagation means and unnecessary instruction deletion by the unnecessary instruction deletion means. A constant value propagation optimization method comprising: instruction output means.