Claims
- 1. A compilation processing method for performing a translation process for a program including at least one branch statement, using initial branch probability information for the at least one branch statement, the at least one branch statement having a plurality of branch conditions, said compilation processing method comprising the steps of:
- providing nodes, with regard to a branch statement, each node being one of a branch node representative of a corresponding branch condition in the branch statement, a first basic block representative of code to be executed when a decision of the branch statement is true, and a second basic block representative of code to be executed when the decision of the branch statement is false;
- creating a flow graph by determining a path from each node to each corresponding node, each path being input to one of the nodes and output from another one of the nodes; and
- determining final branch probability information for each path from each branch condition in the branch statement based on the flow graph, each path having corresponding branch probability information, said corresponding branch probability information indicative of a frequency that a specified path is executed such that the branch probability information for the plurality of branch conditions in the branch statement initially corresponds to the initial branch probability information for the branch statement.
- 2. The compilation processing method according to claim 1, further comprising:
- decreasing an optimization level of the path corresponding to a selected branch condition of the plurality of branch conditions, said selected branch condition being estimated as having a low execution frequency based on the determined branch probability information determined for said corresponding path.
- 3. A compilation processing method for performing a translation process for a program including at least one branch statement, using initial branch probability information for the at least one branch statement, the at least one branch statement having branch conditions, said compilation processing method comprising the steps of:
- providing nodes, with regard to a branch statement, each node being one of a branch node representative of a corresponding branch condition in the branch statement, a first basic block representative of code to be executed when a decision of the branch statement is true, and a second basic block representative of code to be executed when the decision of the branch statement is false;
- creating a flow graph by determining paths from each node to each corresponding node, each path being input to one of the nodes and output from another one of the nodes; and
- calculating final branch probabilities for the nodes, the final branch probability being calculated for each node based on a number of paths incident to the node in the flow graph, the initial branch probability of each path incident to a selected and corresponding node, each node having a corresponding branch probability for the selected and corresponding node, such that the branch probabilities of the incident paths of the nodes of said branch statement correlate to designated branch probability information for the branch statement.
- 4. The compilation processing method according to claim 3, further comprising:
- decreasing an optimization level of a path corresponding to a selected branch condition of the branch conditions, said selected branch condition being estimated as having a low execution frequency based on the calculated branch probability information.
- 5. A compilation processing method for performing a translation process for a program including at least one branch statement, using initial branch probability information for the at least one branch statement, the at least one branch statement having branch conditions, said compilation processing method comprising the steps of:
- providing nodes, with regard to a branch statement, each node being one of a branch node representative of a corresponding branch condition in the branch statement, a truth condition block representative of code to be executed when a decision of the branch statement is true, and a false condition block representative of code to be executed when the decision of the branch statement is false;
- creating a flow graph by determining a path from each node to each corresponding node, each path being input to one of the nodes and output from another one of the nodes; and
- allocating respective values corresponding to numbers of executions for the node corresponding to said truth condition block and for the node corresponding to said false condition block, respectively, each of the respective values calculated by a ratio based on the initial branch probability information designated for the branch statement;
- dividing, for each respective node, the value corresponding to the number of executions allocated for the respective node by the respective number of paths input to the respective node and producing a result;
- allocating the results of the respective division as the respective values corresponding to the numbers of executions of the respective paths input to the respective nodes, for the node which is the truth condition block and the node which is the false condition block;
- determining a sum of the values corresponding to the numbers of executions for all of the paths incident to the respective node and allocating the sum as a corresponding value for the number of executions of the respective node;
- dividing the respective values corresponding to the numbers of executions allocated for the respective node by the number of paths input to the respective node, and allocating a result of the division as the respective values corresponding to the numbers of executions of respective paths input to the node, for each of the nodes of the branch statement; and
- calculating a final branch probability of each respective paths of the flow graph as the final branch probability of the branch condition corresponding to the node from which the path is output, by dividing the respective values corresponding to the number of executions allocated for the respective paths by the corresponding value for the number of executions allocated for the respective node from which each respective path is output.
- 6. The compilation processing method according to claim 5, further comprising:
- decreasing an optimization level of the path corresponding to a selected branch condition of the branch conditions estimated as having a low execution frequency based on the determined branch probability information for said corresponding path.
- 7. A compilation processing apparatus for performing a translation process for a program including at least one branch statement, using initial branch probability information for the at least one branch statement, the at least one branch statement defining conditional decisions, said compilation processing apparatus comprising:
- information collecting phase processing means for extracting nodes, each node being one of a branch node representing respective branch conditions in a branch statement, a first basic block executed when a corresponding conditional decision in the branch statement is true and a second basic block executed when the corresponding conditional decision in the branch statement is false, each node executed in a determined order, for determining a path from each branch node to each corresponding node, and for collecting information concerning a preceding node and a succeeding node;
- block connection relation data forming means for forming, based on the information collected by said information collecting phase processing means, block connection relation data having a plurality of elements, each element indicative of a corresponding connection relation between respective nodes; and
- block connection relation data calculating means for calculating, based on the block connection relation data, a final branch probability for each of the paths corresponding to the respective elements, as final branch probabilities of the branch conditions included by a corresponding branch node from which the respective paths are output such that the final branch probabilities correspond to designated branch probability information for the branch statement.
- 8. A compilation processing apparatus for performing a translation process for a program upon execution of the program, using initial branch probability information for at least one branch statement defining conditional decisions in the program, said compilation processing apparatus comprising:
- information collecting phase processing means for extracting program code blocks, each program code block being one of a branch block including respective branch conditions in a branch statement, a first basic block executed when a corresponding conditional decision in the branch statement is true and a second basic block executed when the corresponding conditional decision in the branch statement is false, the program code blocks executed in a determined order, for determining paths, a path determined from each branch block to each corresponding program code block, and for collecting information concerning a preceding program code block of respective program code blocks and information concerning a succeeding block of the respective program code blocks;
- block connection relation data forming means for forming, based on the information collected by said information collecting phase processing means, block connection relation data having a plurality of elements, each element indicative of a corresponding connection relation between respective program code blocks; and
- block connection relation data calculating means for calculating, based on the block connection relation data and the paths between the respective program code blocks, a branch probability for each of the paths corresponding to each element, as final branch probabilities of the branch conditions included by the respective program code blocks from which the paths are output such that the branch probabilities for the plurality of paths correspond to designated branch probability information for the branch statement.
- 9. A compilation processing apparatus for performing a translation process for a program, the program including at least one branch statement and branch probability information for the at least one branch statement, the at least one branch statement defining conditional decisions, said compilation processing apparatus comprise:
- information collecting phase processing means for extracting basic blocks, each basic block being one of a first basic block including respective branch conditions in a branch statement, a second basic block executed when decision of the branch statement is true and a third basic block executed when decision of the branch statement is false, each said basic block executed in a determined order, for determining a path from each basic block to each corresponding basic block, and for collecting concerning a preceding basic block of the respective first, second and third basic blocks and information concerning a succeeding basic block of the respective first, second and third basic blocks;
- block connection relation data forming means for forming, based on the information collected by said information collecting phase processing means, block connection relation data having a plurality of elements, each element indicative of a corresponding connection relation between respective basic blocks;
- first block connection calculating means for allocating, in the block connection relation data, a first value corresponding to a first number of executions, by dividing a second value corresponding to a second number of executions allocated for the truth condition block, calculated based on a branch probability designated for the branch statement, by the number of elements corresponding to all the paths input to the truth condition block, each element designating one of the paths which is input to the truth condition block;
- second block connection calculating means for allocating, in the block connection relation data, a third value corresponding to a third number of executions, by dividing a fourth value corresponding to a fourth number of executions allocated for the false condition block, based on the branch probability designated for the branch statement, by the number of elements corresponding to all the paths input to the false condition block, each element designating one of the paths which is input to the false condition block;
- third block connection calculating means for allocating, in the block connection relation data, a fifth value corresponding to a fifth number of executions, by dividing a sum of sixth values corresponding to sixth numbers of the executions allocated for each of the respective elements corresponding to all the paths output from a selected one of the basic blocks by the number of elements representing the paths input to the selected basic block, for each element designating the path input to the selected basic block, for each of the basic blocks of the branch statement; and
- fourth block connection calculating means for calculating, a branch probability for each respective element of the block connection relation data, as the branch probability of the corresponding branch condition included by the basic block from which the path corresponding to each element is output by dividing a seventh value corresponding to a seventh number of executions allocated for each element by a sum of eighth values corresponding to eighth numbers of executions allocated for the elements corresponding to all of the paths output from the basic block from which the paths corresponding to the respective element are output.
- 10. A compilation processing method for performing a translation process for a program including at least one branch statement, using initial branch probability information for the at least one branch statement, the at least one branch statement defining a plurality of conditional decisions, said compilation processing method comprising the steps of:
- creating a flow graph by resolving the plurality of conditional decisions defined by the at least one branch statement into respective conditional decisions; and
- determining branch probability information for each respective conditional decision in the flow graph based on the flow graph and the initial branch probability information to to obtain final branch probability information for the at least one branch statement.
- 11. A compilation processing method for performing a translation process for a program including at least one branch statement, using initial branch probability information for the branch statement, the branch statement defining a plurality of conditional decisions, said compilation processing method comprising the steps of:
- creating a flow graph by resolving the at least one branch statement into respective conditional decisions; and
- allocating branch probability for each respective conditional decision in the flow graph, based on the flow graph and the initial branch probability information to to obtain final branch probability information of of the branch statement.
- 12. The compilation processing method according to claim 11 wherein:
- said creating of a flow graph further determines paths for the conditional decisions; and
- said allocating of branch probability allocates branch probabilities for each path; and
- further comprising the steps of decreasing an optimization level of a selected path of the paths which is determined to have a low execution frequency based on a respective branch probability for the path.
- 13. A compilation processing apparatus for performing a translation process for a program including at least one branch statement, using initial branch probability information for the branch statement, the branch statement defining a plurality of conditional decisions, said compilation processing apparatus comprising:
- information collecting phase processing means for resolving the at least one branch statement, into respective conditional decisions allocating, for each of the respective condition decisions, a succeeding basic block, and collecting information for creating a flow graph;
- connection matrix creating phase processing means for creating a connection matrix representing the flow graph designating connection relations relating to branch nodes and succeeding basic blocks, based on the collected information; and
- connection matrix calculating phase processing means for calculating a ratio of a number of executions in accordance with the number of paths from the succeeding basic block to one of the branch nodes, based on the initial branch probability information by referring to the created connection matrix, and allocating respective branch probability information for each of the respective conditional decisions.
Priority Claims (1)
Number |
Date |
Country |
Kind |
5-045609 |
Mar 1993 |
JPX |
|
Parent Case Info
This application is a continuation, of application Ser. No. 08/396,590, filed Mar. 1, 1995, now abandoned, which is a continuation of Ser. No. 08/113,907, filed Aug. 31, 1993, now abandoned.
US Referenced Citations (11)
Non-Patent Literature Citations (2)
Entry |
Superscalar Microprocessor design by Mike Johnson, 1991 Prentice Hall publication, pp. 181-192, 205-212. |
A VLIW Architecture for a Trace Scheduling Compiler by Colwell et al., IEEE 1988 publication. |
Continuations (2)
|
Number |
Date |
Country |
Parent |
396590 |
Mar 1995 |
|
Parent |
113907 |
Aug 1993 |
|