Patent Application
20110167415
This application is based upon and claims the benefit of priority from the prior Japanese Patent Application No. 2010-001424, filed on Jan. 6, 2010; the entire contents of which are incorporated herein by reference.
1. Field of the Invention
The present invention relates to a language processing apparatus, a language processing method, and a computer program product.
2. Description of the Related Art
A compiler that generates an assembler code (assembler program) for a target processor from a source program written in a high-level language allocates a variable written in a source code to a register to be an operand of an instruction written in an assembly language. The register allocated to the variable by the compiler includes two types when a function call is made, that is, a register (non-volatile register) that ensures that the value does not change before and after the function call and a register in which the value may change before and after the function call. An application binary interface (ABI) defines a register to be the non-volatile register for each target processor.
The compiler includes one that has a function of generating a code (hereinafter, save-restore code) for saving/restoring register content at an entry-exit point of a function on the called side to ensure that the value of the register does not change before and after the function call when the non-volatile register is used in the function. The function on the side that calls the function is defined as a caller, and the function on the called side is defined as a callee.
On the other hand, in the target processor, because the execution of the save-restore code involves a memory access, the execution speed becomes fast as the number of executions of the save register is small. Therefore, the compiler is desired to optimize the assembler code so that the number of executions of the save register becomes small.
As a technology for optimizing the save-restore code, Japanese Patent Application Laid-open No. H11-272473 discloses a technology for deleting a corresponding save-restore code from the callee when it is possible to determine that the operation of the whole program as a target does not change even if the value of the non-volatile register is destroyed in the callee.
A language processing apparatus according to an embodiment of the present invention comprises:
a first assembler file generating unit that allocates a variable included in a source program written in a single module to a register, generates an assembler code for each function, inserts a save-restore code for the register into an entry-exit point of a function that uses the register, and generates a first assembler program; and
a second assembler file generating unit that, when the register used in the function is not used in a caller, migrates the save-restore code for the register written in the first assembler file to an entry-exit point of the caller, and generates a second assembler program.
A language processing method according to an embodiment of the present invention comprises:
performing a lexical and syntax analysis on a source program written in a single module;
generating function call information in which a call relation between functions included in the source program is written based on an execution result of the lexical and syntax analysis;
allocating a register to a variable included in the source program based on the execution result of the lexical and syntax analysis;
generating non-volatile register information on a non-volatile register in which a value does not change in the function included in the source program for each function based on an allocation result of the variable;
generating a first assembler program in which a save-restore code for protecting content of the non-volatile register is inserted into an entry-exit point of a function that uses the non-volatile register from the source program based on the allocation result of the variable;
determining whether the save-restore code is capable of being migrated from the function into which the save-restore code is inserted to a function as a call source of the function based on the function call information and the non-volatile register information and generating save-restore code information that indicates a function as a migration destination of the save-restore code; and
migrating the save-restore code written in the first assembler program based on the non-volatile register information and the save-restore code information and generating a second assembler program.
A computer program product according to an embodiment of the present invention includes a plurality of instructions executable on a computer, wherein the instructions, when executed by the computer, cause the computer to perform:
performing a lexical and syntax analysis on a source program written in a single module;
generating function call information in which a call relation between functions included in the source program is written based on an execution result of the lexical and syntax analysis;
allocating a register to a variable included in the source program based on the execution result of the lexical and syntax analysis;
generating non-volatile register information on a non-volatile register in which a value does not change in the function included in the source program for each function based on an allocation result of the variable;
generating a first assembler program in which a save-restore code for protecting content of the non-volatile register is inserted into an entry-exit point of a function that uses the non-volatile register from the source program based on the allocation result of the variable;
determining whether the save-restore code is capable of being migrated from the function into which the save-restore code is inserted to a function as a call source of the function based on the function call information and the non-volatile register information and generating save-restore code information that indicates a function as a migration destination of the save-restore code; and
migrating the save-restore code written in the first assembler program based on the non-volatile register information and the save-restore code information and generating a second assembler program.
Exemplary embodiments of a language processing apparatus, a language processing method, and a computer program product according to the present invention will be explained below in detail with reference to the accompanying drawings. The present invention is not limited to the following embodiments.
A subroutine that returns a value is called a function and a subroutine that does not return a value is called a procedure depending on a programming language. In the embodiments of the present invention, the concept of the function also includes the procedure.
A first embodiment of the present invention is applied to a compiling apparatus (compiler) that converts a source program (source code) into an assembly language for each module to generate an assembler code (assembler program). The compiling apparatus in the first embodiment has a function (save-restore code generating function) of generating a save-restore code with respect to content of a non-volatile register at an entry-exit point of a callee when the non-volatile register is used in the callee. First, the characteristics of the first embodiment are explained with reference to
The compiler in the first embodiment first generates an assembler code (first assembler code) from the source code 101 by using the save-restore code generating function. Then, the compiler corrects the first assembler code so that the number of executions of the save-restore code becomes small and generates an assembler code (second assembler code) as a final product in the first embodiment.
An assembler code 102 is an example of the first assembler code generated from the source code 101. The compiler performs a register allocation so that $8 is used in the function sub and $8 is not used in the function test. As shown in the assembler code 102, an instruction “push $8” for saving content of $8 in a stack is written at the entry point of the function sub and an instruction “pop $8” for restoring the saved content to $8 is written at the exit point of the function sub for preventing the content from being destroyed by the use of $8. In the followings, such a pair of the push and the pop is called a save-restore code. Because $8 is not used in the function test, the save-restore code for $8 is not written in the function test. An instruction “call_sub” for calling the function sub is written in a block corresponding to the function test.
In the source code 101, the function sub is called in a for loop included in the function test. In other words, the “call_sub” and the function sub are repeatedly executed for the number of times satisfying the condition of the for loop. In the assembler code 102, the target processor executes saving/restoring of the content of $8 every time the function sub is called.
An assembler code 103 is an example of the second assembler code generated from the assembler code 102. As shown in
With the assembler code 103, the target processor saves the content of $8 at the entry point in the function test, then executes the loop, and restores the content of $8 at the exit point of the function. In this case, the number of executions of the save-restore code is once, different from the case of the assembler code 102 in which the number of executions of the save-restore code is the number of times of the loop in the function test.
In this manner, the first embodiment of the present invention is mainly characterized in that the assembler code in which the number of executions of the save-restore code is reduced can be generated by migrating the save-restore code for the non-volatile register written at the entry-exist point of the callee to the entry-exist point of the caller. The processing of migrating the save-restore code included in the first assembler code is called save-restore code migration processing.
The compiling apparatus 200 includes a lexical and syntax analyzer 201, a register allocating unit 202, and an assembler-code generating unit 203 as a configuration (first-assembler-file generating unit) for generating the first assembler file D2.
The functions of the lexical and syntax analyzer 201, the register allocating unit 202, and the assembler-code generating unit 203 are the same as the functions of the respective components included in a generally-available compiling apparatus. In the lexical analysis by the lexical and syntax analyzer 201, each character included in the input source file D1 is loaded into, for example, a RAM. Then, in the syntax analysis by the lexical and syntax analyzer 201, the meaning of each loaded character and the meaning (such as a function, a variable, and an assignment statement relation) on a program written in the source file D1 caused by the combination of characters are recognized.
The register allocating unit 202 allocates the register to each variable included in the source file D1 based on an analysis result by the lexical and syntax analyzer 201. The assembler-code generating unit 203 generates the assembler code based on a register allocation result by the register allocating unit 202 to generate the first assembler file D2. When the non-volatile register is allocated to the variable, the assembler-code generating unit 203 inserts the save-restore code for this non-volatile register into the entry-exit point of this function.
Furthermore, the compiling apparatus 200 includes a function-call extracting unit 204, a non-volatile register extracting unit 205, a save-restore code information generating unit 206, and a save-restore code migration processing unit 207 as a configuration (second-assembler-file generating unit) for generating the second assembler file D3 from the first assembler file D2.
The function-call extracting unit 204 extracts information on a call relation between functions included in the source file D1, information on whether each function is a function called from other modules, and information on whether there is a possibility of a dynamic call, based on the analysis result by the lexical and syntax analyzer 201. The function-call extracting unit 204 generates the extracted call relation as function call information D4.
The non-volatile register extracting unit 205 generates non-volatile register information D5 that is a list of the non-volatile register used in the function for each function based on the register allocation result by the register allocating unit 202.
The save-restore code information generating unit 206 determines whether the save-restore code for the non-volatile register used in the function can be migrated for each function based on the function call information D4 and the non-volatile register information D5. More specifically, when a target function uses the non-volatile register and there is no function that uses this non-volatile register among the callers that call this target function, the save-restore code information generating unit 206 determines that the save-restore code for this non-volatile register can be migrated. Moreover, the save-restore code information generating unit 206 generates save-restore code information D6 in which the position (i.e., the position into which the save-restore code is inserted by the assembler-code generating unit 203) of the save-restore code before the save-restore code migration processing and the position of the save-restore code after the save-restore code migration processing are written.
The save-restore code migration processing unit 207 performs the save-restore code migration processing on the save-restore code written in the first assembler file D2 based on the save-restore code information D6 to generate the second assembler file D3. The first assembler file D2 is not the final assembler code, so that the first assembler file D2 can be placed in a state of being stored in a memory as internal information without being output as a file.
The CPU 1 executes a compile program 6 that is a computer program product that executes a compiling method in the first embodiment. The display unit 5 is a display device such as a liquid crystal monitor and displays output information for a user such as an operation screen based on an instruction from the CPU 1. The input unit 4 is configured to include a mouse, a keyboard, and the like, and inputs an operation from a user to the compiling apparatus 200. The operation information input from the input unit 4 is sent to the CPU 1 to be processed.
The compile program 6 has a module configuration including the lexical and syntax analyzer 201, the register allocating unit 202, the assembler-code generating unit 203, the function-call extracting unit 204, the non-volatile register extracting unit 205, the save-restore code information generating unit 206, and the save-restore code migration processing unit 207. These configuration units are generated in the RAM 3 by loading the compile program 6 into the RAM 3.
The compile program 6 is stored in the ROM 2 and is loaded into the RAM 3 via a bus line.
The compile program 6 can be provided or distributed in such a way that the compile program 6 is stored in a computer connected to a network such as the Internet and is downloaded via the network. Alternatively, the compile program 6 can be incorporated in the ROM 2 or the like in advance and provided to the compiling apparatus 200
Next, the compiling method in the first embodiment of the present invention realized by using the compiling apparatus 200 is explained. The processor in which registers $8 to $11 are defined as the non-volatile register is set as the target processor. The source code (source file D1) shown in
The analysis by the lexical and syntax analyzer 201 is performed by the static analysis, so that the call in which a jump destination dynamically changes by the function call using a pointer of the function or the like cannot be analyzed. Therefore, in the first embodiment of the present invention, the save-restore code included in the function that may be called dynamically is not subjected to the save-restore code migration processing. Moreover, if the save-restore code migration processing is to be performed on the save-restore code included in the function that may be called from a module other than the input source file D1, all of the source modules constituting the program need to be analyzed, so that the processing of determining whether the save-restore code migration processing can be performed becomes complicated, and therefore the compile time becomes long. Thus, in the first embodiment of the present invention, the save-restore code included in the function that may be called from other modules is also not subjected to the save-restore code migration processing.
The function call information D4 includes a field in which whether the function is a function called from other modules different from the input source file D1 and whether the function may be called by the dynamic call are written for each function. Whether the function is a function called from other modules can be determined, for example, by determining whether the function is a function declared as static in the case of C. Moreover, in the case of C, an address of the function needs to be set to a pointer-type variable of the function by an initial value or an assignment statement for the dynamic call of the function. Therefore, when there is no reference to the callee other than the function call for the callee in the source file D1, it is possible to determine that the dynamic function call for the callee does not occur in the source file D1. In an example shown in
Next, the register allocating unit 202 allocates the register to the variable written in the source file D1 based on the analysis result by the lexical and syntax analyzer 201 (S3). In an example shown in
Then, the non-volatile register extracting unit 205 generates the non-volatile register information D5 based on the register allocation result by the register allocating unit 202 (S4).
In some cases, the target processor includes a plurality of types of registers such as a core register and a coprocessor register and the register allocation is performed on these registers by the register allocating unit 202. The non-volatile register extracting unit 205 generates the non-volatile register information D5 on all of the non-volatile registers regardless of the type of the register.
Next, the assembler-code generating unit 203 generates the first assembler file D2 based on the register allocation result by the register allocating unit 202 (S5). In this example, the first assembler file D2 shown in
The save-restore code information generating unit 206 generates the save-restore code information D6 based on the function call information D4 and the non-volatile register information D5 (S6).
As shown in
After S11, the save-restore code information generating unit 206 selects one function from the save-restore code information D6 (S12). Then, the save-restore code information generating unit 206 refers to the function call information D4 and determines whether the selected function is a function that can recognize all of the callers that call the function (S13). The function that can recognize all of the callers indicates that all of the callers that call the function are not a function called from other modules and are a function that is not called by the dynamic call.
When the function is a function that can recognize all of the callers (Yes at S13), the save-restore code information generating unit 206 refers to the post-migration information in the save-restore code information D6 and determines whether the register name is written in the post-migration information on the selected function (S14).
When the register name is written in the post-migration information on the selected function (Yes at S14), the save-restore code information generating unit 206 determines whether there is a register of which register name is not written in the pre-migration information on any caller among the callers that call the selected function, among the registers written in the post-migration information on the selected function (S15).
When there is a register of which register name is not written in the pre-migration information on any caller (Yes at S15), the save-restore code information generating unit 206 determines the save-restore code for this register as the migration target (S16), and updates the save-restore code information D6 as the temporary information (S17). Specifically, the save-restore code information generating unit 206 deletes the register name that is determined as the migration target from the post-migration information on the selected function and writes the register name in the post-migration information on all of the callers that call the selected function. When there is already a corresponding register name written in the write destination, the save-restore code information generating unit 206 does not write the register name in this write destination.
When the selected function is not a function that can recognize all of the callers (No at S13), the save-restore code information generating unit 206 determines that there is no save-restore code to be the migration target in the selected function (S18).
In the similar manner, when the register name is not written in the post-migration information (No at S14) or when there is no register of which register name is not written in the pre-migration information on any caller (No at S15), the system control proceeds to S18.
After S17 or S18, the save-restore code information generating unit 206 also determines whether all of the functions have been selected (S19). When there is an unselected function (No at S19), the system control proceeds to S12 and repeats S12 to S18 until there is no unselected function.
When all of the functions have already been selected (Yes at S19), the save-restore code information generating unit 206 determines whether the save-restore code information D6 as the temporary information is updated by the processing from S12 to S19 (S20). When the save-restore code information D6 is updated (Yes at S20), the save-restore code information generating unit 206 sets all of the functions to an unselected state (S21) and the system control proceeds to S12. When the save-restore code information D6 is not updated (No at S20), there is no save-restore code to be a new migration target any longer, so that the operation of the save-restore code information generating unit 206 is returned. In the first determination processing at S20, the save-restore code information generating unit 206 determines whether the save-restore code information D6 is updated from the initial state. In the second and subsequent determination processing at S20, the save-restore code information generating unit 206 determines whether the save-restore code information D6 as the temporary information is updated after the last determination processing at S20.
After S6, the save-restore code migration processing unit 207 performs the save-restore code migration processing on the first assembler file D2 based on the generated save-restore code information D6 to generate the second assembler file D3 (S7). In the save-restore code migration processing, the non-volatile register that is written in the pre-migration information and is not written in the post-migration information is extracted for each function and the save-restore code for each function for protecting the content of the extracted non-volatile register is deleted from the entry-exit point of the function that is written in the pre-migration information, and the non-volatile register that is written in the post-migration information and is not written in the pre-migration information is extracted for each function and the save-restore code for protecting the content of the extracted non-volatile register is inserted into the entry-exit point of the function that is written in the post-migration information. For example, in the save-restore code information D6 shown in
Explanation is given for the case where the save-restore codes for the same non-volatile register migrate to the same caller from a plurality of the callees.
When the function sub1 is selected at S12, the save-restore code information D6 shown in
In this manner, when the save-restore codes for the same non-volatile register migrate to the same caller from a plurality of the callees, the save-restore codes are controlled not to overlap in the migration destination, so that the code size of the assembler code can be reduced.
According to the technology disclosed in Japanese Patent Application Laid-open No. H11-272473, the number of executions of the save-restore code is reduced by deleting the save-restore code. However, for deleing the save-restore code, it is needed that the deletion of the save-restore code does not cause a problem on the operation of the program. In other words, the condition is required for deleting the save-restore code that the register is not used between the initial function of an application program and the function in which the save-restore code is deleted. On the contrary, in the first embodiment of the present invention, the save-restore code is only migrated from the callee to the caller, so that the value of the non-volatile register is saved as viewed from the upper function of the caller, and therefore the operation of the program is not affected. In other words, according to the first embodiment of the present invention, the condition for migrating the save-restore code is only that the register is not used in the caller, so that the assembler code in which the number of executions of the save-restore code is further reduced than the technology disclosed in Japanese Patent Application Laid-open No. H11-272473 can be generated by migrating the save-restore code even in the case where the save-restore code cannot be deleted by the technology disclosed in Japanese Patent Application Laid-open No. H11-272473.
In the first embodiment of the present invention, the non-volatile register information D5 on all of the registers is generated without distinguishing between the core register and the coprocessor register. For the processor in which the core register and the coprocessor register are distinguished, the processing using the coprocessor register more is performed in local functions and the processing using the core register without using the coprocessor register is performed in an upper function that calls the local functions in some cases. In such a case, the save-restore code for the coprocessor register in the local function can be often migrated to the upper function, so that the number of executions of the save-restore code can be reduced significantly.
As described above, according to the first embodiment of the present invention, the first assembler file D2 in which the save-restore code for protecting the non-volatile register content used is inserted into the entry-exit point of the function that uses the non-volatile register is generated from the source file D1, the function call information in which the call relation between the functions included in the source file D1 is written is generated based on the syntax analysis result output in the process of generating the first assembler file D2, the non-volatile register information D5 that is a list of the non-volatile register used in the function included in the source code for each function is generated based on the register allocation result output in the process of generating the first assembler file D2, the save-restore code information D6 that indicates the function as the insertion destination of the save-restore code after migration for each save-restore code is generated by performing the determination processing of determining whether the save-restore code can be migrated from the function into which the save-restore code is inserted to the function that is the call source of the function for each save-restore code based on the function call information D4 and the non-volatile register information D5, and the second assembler file D3 is generated by performing the save-restore code migration processing on the save-restore code written in the first assembler file D2 based on the save-restore code information. Therefore, the save-restore code inserted into the entry-exit point of the callee can be migrated to the entry-exit point of the caller, so that the assembler code in which the number of executions of saving the register content is reduced can be generated.
Moreover, the save-restore code information D6 as the temporary information that indicates the function as the temporary migration destination for each save-restore code is generated, the save-restore code information D6 as the temporary information is updated every time the determination processing is performed, and the save-restore code information D6 as the temporary information is set to the save-restore code information D6 for performing the save-restore code migration processing when the save-restore code information D6 as the temporary information becomes unchanged before and after the update, so that the assembler code in which the number of executions of the save-restore code is reduced as much as possible can be generated.
In the above explanation, explanation is given for the case where the push and pop instructions involving the reserving and releasing instructions of the stack area are set as the save-restore code; however, the save-restore code can be any instruction so long as the instruction is for saving/restoring the register content. For example, a pair of a load and store instruction and an add-subtract instruction of a stack pointer that does not involve the reserving and releasing instructions of the stack area can also be used as the save-restore code other than the push and pop instructions. Moreover, the program written in C is raised as an example of the source file D1; however, the description language of the source file D1 has a concept of the function other than C, so that the description language can be the high-level language that is converted into the assembler code by a compiler.
A user converts the second assembler file D3 into a relocatable file by using the assembler program for obtaining an execution file that is finally operated in the target processor. Then, the user integrates all of the relocatable files generated for respective files and performs an address resolution by using a link program thereby obtaining the execution file.
As shown in
A second embodiment of the present invention is applied to a compiling and assembling apparatus that converts the source code into the relocatable file for each file.
A compiling and assembling apparatus 300 includes a compiling unit 310 that generates the first assembler file D2, the function call information D4, and the non-volatile register information D5 from the input source file D1 and an assembling unit 320 that generates the second assembler file D3 based on the first assembler file D2, the function call information D4, and the non-volatile register information D5 and generates a relocatable file D7 from the generated second assembler file D3.
The compiling unit 310 includes the lexical and syntax analyzer 201 that performs the lexical syntax analysis on the source file D1, the function-call extracting unit 204 that generates the function call information D4 based on the analysis result by the lexical and syntax analyzer 201, the register allocating unit 202 that allocates the variable to the register based on the analysis result by the lexical and syntax analyzer 201, the non-volatile register extracting unit 205 that generates the non-volatile register information D5 based on the register allocation result by the register allocating unit 202, and the assembler-code generating unit 203 that generates the first assembler file D2 based on the register allocation result by the register allocating unit 202.
The assembling unit 320 includes the save-restore code information generating unit 206 that generates the save-restore code information D6 based on the function call information D4 and the non-volatile register information D5, the save-restore code migration processing unit 207 that performs the save-restore code migration processing on the first assembler file D2 based on the save-restore code information D6 and generates the second assembler file D3, and an assembler file converting unit 321 that generates the relocatable file D7 from the second assembler file D3.
The compiling and assembling apparatus 300 in the second embodiment is realized by a hardware configuration same as that shown in
In this manner, the first assembler file D2 is generated from the source file D1, and the save-restore code migration processing is performed on the generated first assembler file D2 to generate the second assembler file D3 that is the assembler file in which the number of executions of the save-restore code is reduced, so that the relocatable file D7 in which the number of executions of saving/restoring the non-volatile register content is reduced can be generated.
Additional advantages and modifications will readily occur to those skilled in the art. Therefore, the invention in its broader aspects is not limited to the specific details and representative embodiments shown and described herein. Accordingly, various modifications may be made without departing from the spirit or scope of the general inventive concept as defined by the appended claims and their equivalents.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2010-001424 | Jan 2010 | JP | national |
