Computer with Hybrid Von-Neumann/Dataflow Execution Architecture

• US 20170031866A1
• Filed: 07/30/2015
• Published: 02/02/2017
• Est. Priority Date: 07/30/2015
• Status: Active Grant

• Current Assignee: Wisconsin Alumni Research Foundation (University of Wisconsin System)
• Original Assignee: Wisconsin Alumni Research Foundation (University of Wisconsin System)
• Inventors: Gangadhar, Vinay, Nowatzki, Anthony, Sankaralingam, Karthikeyan
• Granted Patent: US 10,216,693 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 1/324 by lowering clock frequency
• CPC Class Codes: G06F 1/3287 by switching off individual functional units in the computer system
• CPC Class Codes: G06F 1/3296 by lowering the supply or operating voltage
• CPC Class Codes: G06F 15/825 Dataflow computers
• CPC Class Codes: G06F 8/41 Compilation
• CPC Class Codes: G06F 9/30076 to perform miscellaneous control operations, e.g. NOP
• CPC Class Codes: G06F 9/3836 Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution
• CPC Class Codes: G06F 9/3877 using a slave processor, e.g. coprocessor peripheral processor G06F13/12; vector processor G06F15/8053
• CPC Class Codes: G06F 9/3897 with adaptable data path
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A dataflow computer processor is teamed with a general computer processor so that program portions of an application program particularly suited to dataflow execution may be transferred to the dataflow processor during portions of the execution of the application program by the general computer processor. During this time the general computer processor may be placed in partial shutdown for energy conservation.

Claims

1. A computer with improved function comprising:

a general computer processor providing;

(a) a memory interface for exchanging data and instructions with an electronic memory;

(b) an arithmetic logic unit receiving input data and instructions from the memory interface to process the same and to provide output data to the memory interface; and

(c) a program counter identifying instructions for execution by the arithmetic logic unit;

a dataflow computer processor providing;

(a) a memory interface for exchanging data and instructions with electronic memory;

(b) multiple functional units interconnected to receive input data from the memory interface or other functional units and provide output data to the memory interface or other functional units; and

(c) an interconnection control circuit controlling the interconnection of the multiple functional units to exchange data according to the dataflow description; and

a transfer interface operating to transfer the execution of an application program between the general purpose computer processor and the dataflow computer processor;

(a) at a beginning of a set of instructions of the application program identified as executable on the dataflow computer processor, switching execution from the general computer processor to the dataflow computer processor and providing to the dataflow computer processor a dataflow description of the set of instructions; and

(b) at a completion of execution of the set of instructions by the dataflow computer processor, returning execution to the general computer processor.

2. The computer of claim 1 wherein the general computer processor includes a low-power operating mode and a high-power operating mode where the low-power operating mode consumes less power than the high-power operating mode and wherein the computer moves the general computer processor to the low-power mode at the beginning of the set of instructions and to the high-power mode at the completion of execution of the set of instructions.

3. The computer of claim 1 further including a prediction table tracking execution on the dataflow computer processor of the set of instructions of the application as linked to the set of instructions, wherein the computer switches execution from the general computer processor to the dataflow computer processor for a given set of instructions only when the prediction table tracking for previous execution of a given set of instructions indicates likelihood of a predetermined benefit in execution of the transfer.

4. The computer of claim 3 wherein the tracking tracks at least one of execution time and number of executed instructions.

5. The computer of claim 1 wherein the functional elements of the dataflow processor execute in multiple sequential time steps in between which configuration of functional units of the functional elements and interconnection of the functional units is changed.

6. The computer of claim 5 wherein the dataflow computer processor includes registers for storage of data between time steps.

7. The computer of claim 1 wherein the computer identifies in-line instruction loops providing one or more loops of control flow without input-output operations as the set of instructions for transfer to the dataflow processor.

8. The computer of claim 1 wherein the general computer processor executes a transfer program identifying a beginning of the set of instructions and enabling operation of the transfer interface in switching execution from the general computer processor to the dataflow computer processor and wherein the dataflow computer processor executes the dataflow description to identify a completion of execution of the set of instructions to enable operation of the transfer interface and returns execution to the general computer processor.

9. The computer of claim 1 wherein the general computer processor is capable of executing the entire application program including the set of instructions.

10. The computer of claim 1 wherein the general computer processor adjusts its program counter upon completion of the execution of the instruction set to an instruction after the set of instructions.

11. The computer of claim 1 wherein the transfer circuit at the beginning of the set of instructions further transfers initial data values for the functional units.

12. The computer of claim 1 wherein the beginning of the set of instructions is identified from special instructions embedded in the application program.

13. The computer of claim 1 wherein the dataflow description is embedded in the application program executed by the general computer processor.

14. The computer of claim 1 wherein the set of instructions is limited to a predefined maximum number of static instructions.

15. The computer of claim 1 wherein the multiple functional units of the dataflow computer are interconnected by a bus structure that may interconnect only a subset less than the full set of functional units at a given time.

16. The computer of claim 1 wherein the general computer processor and dataflow computer processor both communicate with a common cache through the memory interfaces.

17. A method of executing an application program on a hybrid general-purpose dataflow computer having:

a general computer processor providing;

(a) a memory interface for exchanging data and instructions with an electronic memory;

(b) an arithmetic logic unit receiving input data and instructions from the memory interface to process the same and to provide output data to the memory interface; and

(c) a program counter identifying instructions for execution by the arithmetic logic unit;

a dataflow computer processor providing;

(a) a memory interface for exchanging data and instructions with electronic memory;

(b) multiple functional units interconnected to receive input data from the memory interface or other functional units and provide output data to the memory interface or other functional units; and

(c) an interconnection control circuit controlling the interconnection of the multiple functional units to exchange data according to the dataflow description; and

a transfer interface operating to transfer the execution of an application program between the general purpose computer processor and the dataflow computer processor;

(a) at a beginning of a set of instructions of the application program executable on the dataflow computer processor, switching execution from the general computer processor to the dataflow computer processor and providing to the dataflow computer processor a dataflow description of the set of instructions; and

(b) at a completion of execution of the set of instructions by the dataflow computer processor, returning execution to the general computer processor;

the method comprising the steps of;

(1) executing a first portion of the application program on the general computer processor while suspending operation of the dataflow processor; and

(2) executing a second portion of the application program identified for execution by the dataflow processor on the dataflow processor while suspending operation of the general computer processor.

18. A computer executing a compiler program to provide an application program for execution on a hybrid general-purpose dataflow computer having:

a general computer processor comprising;

(a) a memory interface for exchanging data and instructions with an electronic memory;

(b) an arithmetic logic unit receiving input data and instructions from the memory interface to process the same and to provide output data to the memory interface; and

(c) a program counter identifying instructions for execution by the arithmetic logic unit;

a dataflow computer processor providing;

(a) a memory interface for exchanging data and instructions with electronic memory;

(b) multiple functional units interconnected to receive input data from the memory interface or other functional units and provide output data to the memory interface or other functional units; and

(c) an interconnection control circuit controlling the interconnection of the multiple functional units to exchange data according to the dataflow description; and

a transfer interface operating to transfer the execution of an application program between the general purpose computer processor and the dataflow computer processor;

(a) at a beginning of a set of instructions of the application program previously identified as executable on the dataflow computer processor, switching execution from the general computer processor to the dataflow computer processor and providing to the dataflow computer processor a dataflow description of the set of instructions; and

(b) at a completion of execution of the set of instructions by the dataflow computer processor, returning execution to the general computer processor,

the computer executing the compiler program to;

(1) identify a program portion of an application program executable on the computer program for execution on the dataflow computer processor;

(2) convert instructions of the program portion to dataflow descriptions for receipt by the dataflow computer processor to execute the program portions; and

(3) place at least one marker in the application program signaling the location of the program portions.

19. The computer of claim 18 wherein the computer further executes the compiler to insert the dataflow descriptions into the application program while retaining the program portion in the application program.

20. The computer of claim 18 wherein the identified program portion is a loop and the computer further executes the compiler to convert the loop into an in-line nested loop without calls to other program portions.