Energy-focused re-compilation of executables and hardware mechanisms based on compiler-architecture interaction and compiler-inserted control

• US 10,248,395 B2
• Filed: 01/19/2017
• Issued: 04/02/2019
• Est. Priority Date: 10/29/2003
• Status: Active Grant

• Current Assignee: III Holdings 2, LLC (Intellectual Ventures LLC)
• Original Assignee: III Holdings 2, LLC (Intellectual Ventures LLC)
• Inventors: Chheda, Saurabh, Carver, Kristopher, Ashok, Raksit
• Primary Examiner(s): Do, Chat C
• Assistant Examiner(s): Slachta, Douglas M
• Application Number:
  US15/410,567
• Publication Number:
  US 20170131986A1
• Time in Patent Office: 803 Days
• CPC Class Codes: G06F 8/4432 Reducing the energy consumption
• CPC Class Codes: G06F 9/30083 Power or thermal control instructions
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A method comprising of analyzing and transforming a program executable at compile-time such that a processor design objective is optimized. A method including analyzing an executable to estimate energy consumption of an application component in a processor. A method including transforming an executable to reduce energy consumption in a processor. A processor framework controlled by compiler inserted control that statically exposes parallelism in an instruction sequence. A processor framework to reduce energy consumption in an instruction memory system with compiler inserted control.

Claims

1. A method, comprising:

identifying a first binary executable generated by a source-level complier from a source file;

generating an intermediate object from the first binary executable;

running one or more general analyses on the intermediate object to obtain input information for one or more power-specific analyses, wherein the input information includes disambiguated symbolic information;

updating the intermediate object with the input information for the one or more power-specific analyses;

identifying information relating to micro-operations of components of a microprocessor by running the one or more power-specific analyses on the updated intermediate object;

wherein the one or more power-specific analyses predict inactive periods of resources in the microprocessor using static information produced through a compilation of a computer program;

generating control instructions based on the information relating to the micro-operations of the components of the microprocessor, the control instructions comprising instructions that contain control data that renders unnecessary at least one of the micro-operations, wherein the control instructions are configured to control the resources during the inactive periods so as to reduce energy consumption of the resources during the inactive periods; and

combining the generated control instructions with at least one of instructions of the intermediate object to generate a second binary executable to consume less energy at runtime than the first binary executable.

2. The method of claim 1, further comprising rearranging the instructions of the intermediate object to generate the second binary executable.

3. The method of claim 1, further comprising removing a branched instruction from the instructions of the intermediate object.

4. The method of claim 2, wherein rearranging the instructions of the intermediate object further comprises reordering the instructions of the intermediate object to increase an amount of parallel instructions.

5. The method of claim 1, further comprising:

generating the intermediate object by replacing a first instruction sequence corresponding to the first binary executable with a second instruction sequence that is different than the first instruction sequence; and

generating the second binary executable from the intermediate object.

6. The method of claim 5, wherein the intermediate object comprises program blocks, wherein an analysis of the one or more general analyses is configured to identify a subset of the program blocks associated with a specific criticality, and wherein the input information for the one or more power-specific analyses includes one or more values to identify the subset of the program blocks differently than remaining program blocks.

7. The method of claim 5, wherein:

the first binary executable is a first type of executable that corresponds to a first format,

the intermediate object comprises a second format that is different than the first format,

and

the second binary executable comprises is a second type of executable that corresponds to a third format that is different than the second format.

8. The method of claim 1, further comprising:

converting the first binary executable into the intermediate object, wherein the first binary executable is a first type of executable that corresponds to a first format, and wherein the intermediate object comprises a second format that is different than the first format;

wherein the second binary executable is a second type of executable that corresponds to a third format.

9. The method of claim 8, wherein the third format is the same as the first format.

10. The method of claim 8, wherein the second type of executable corresponds to an instruction set that is different than an instruction set to which the first type of executable corresponds.

11. The method of claim 8, wherein the first format comprises at least one of an executable file format (ELF), a portable executable (PE) format, or a common object file format (COFF).

12. The method of claim 8, wherein the second format comprises a binary intermediate format (BIF).

13. The method of claim 8, wherein the third format is different than the first format.

14. The method of claim 1, wherein the control data identifies instructions in the second binary executable that may be executed in parallel.

15. The method of claim 14, wherein identifying instructions that may be executed in parallel further comprises identifying parallel instructions in blocks of instructions associated with the first binary executable.

16. The method of claim 14, wherein the control data comprises a control bit preceding the instructions in the second binary executable that may be executed in parallel.

17. The method of claim 16, further comprising identifying instructions that may be executed in parallel exposing parallel instructions in critical loops in the first binary executable.

18. The method of claim 16, wherein the control data comprises data to be removed from an instruction sequence of the second binary executable before a corresponding portion of the instruction sequence enters a pipeline of the microprocessor.

19. The method of claim 16, wherein the control data is usable during executing of the second binary executable to improve operation of the microprocessor relative to execution of the first binary executable.

20. The method of claim 19, wherein the control data is usable to reduce energy consumption of the microprocessor during execution of the second binary executable.

21. A memory having instructions stored thereon that, in response to execution by a processing, cause the processing to perform operations comprising:

generating an intermediate object from a first binary executable;

running one or more general analyses on the intermediate object to obtain input information for one or more power-specific analyses, wherein the input information includes disambiguated symbolic information;

updating the intermediate object with the input information for the one or more power-specific analyses;

identifying information relating to micro-operations of components of a microprocessor by running the one or more power-specific analyses on the updated intermediate object, wherein the one or more power-specific analyses predict inactive periods of resources in the microprocessor using static information produced through a compilation of a computer program;

generating control instructions based on the information relating to the micro-operations of the components of the microprocessor, the control instructions comprising instructions that contain control data that renders unnecessary at least one of the micro-operations, wherein the control instructions are configured to control the resources during the inactive periods so as to reduce energy consumption of the resources during the inactive periods; and

combining the generated control instructions with at least one of instructions of the intermediate object to generate a second binary executable to consume less energy at runtime than the first binary executable.

22. The memory of claim 21, wherein the operations further comprise rearranging the instructions of the intermediate object.

23. The memory of claim 21, wherein the operations further comprise removing a branched instruction from the instructions of the intermediate object.

24. The memory of claim 21, wherein the control instructions are configured to reduce an amount of voltage supplied to the resources during the inactive periods relative to an amount of voltage supplied to the resources during active periods of the resources.

25. The memory of claim 21, wherein the operations further comprise generating the intermediate object by replacing a first instruction sequence corresponding to the first binary executable with a second instruction sequence that is different than the first instruction sequence.

26. The memory of claim 25, wherein the intermediate object comprises program blocks, wherein an analysis of the one or more general analyses is configured to identify a subset of the program blocks associated with a specific criticality, and wherein the input information for the one or more power-specific analyses includes one or more values to identify the subset of the program blocks differently than remaining program blocks.