Processor containing data path units with forwarding paths between two data path units and a unique configuration or register blocks

• US 6,301,653 B1
• Filed: 10/14/1998
• Issued: 10/09/2001
• Est. Priority Date: 10/14/1998
• Status: Expired due to Term

• Current Assignee: MACOM Technology Solutions Holdings, Inc.
• Original Assignee: Conexant Systems Incorporated (Synaptics Incorporated)
• Inventors: Malich, Kenneth W., Spence, John R., Mohamed, Moataz A.
• Primary Examiner(s): Kim, Kenneth S.
• Application Number:
  US09/173,257
• Time in Patent Office: 1,091 Days
• Field of Search: 712/24, 712/214
• US Class Current: 712/214
• CPC Class Codes: G06F 9/3836 Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution
• CPC Class Codes: G06F 9/3838 Dependency mechanisms, e.g. register scoreboarding
• CPC Class Codes: G06F 9/3853 of compound instructions
• CPC Class Codes: G06F 9/3885 using a plurality of independent parallel functional units
• CPC Class Codes: G06F 9/3891 organised in groups of units sharing resources, e.g. clusters

Abstract

The present invention provides an efficient method of forwarding and sharing information between functional units and register files in an effort to execute instructions. A digital signal processor includes a plurality of register blocks for storing data operands coupled to a plurality of data path units for executing instructions. Preferably, each register block is coupled to at least two data path units. In addition, the processor preferably has a plurality of forwarding paths which forward information from one data path unit to another. A scheduler efficiently forwards instructions to data path units based on information regarding the configuration of the processor and any restrictions which might be imposed on the scheduler.

Claims

1. A digital signal processor for use in executing a plurality of instructions contained in an instruction packet, said processor comprising:

a plurality of register blocks, each containing operands;

a plurality of data path units, wherein each of said data path units comprises a plurality of functional units, for executing said instructions using said operands and yielding results;

a coupling fabric configured to facilitate communication between said register blocks and said data path units, wherein said coupling fabric comprises a plurality of forwarding paths wherein each of said forwarding paths is configured to couple two of said data path units together, said forwarding paths being unidirectional; and

a scheduler which contains information regarding said coupling fabric, said scheduler configured to forward each of said plurality of instructions to one of said data path units based on the configuration of said coupling fabric; and

wherein

said plurality of data path units equals M data path units, wherein said coupling fabric is further configured to couple each of said register blocks to L of said M data path units, where 1<

L<

M.

2. The processor of claim 1, wherein said coupling fabric further comprises a reverse forwarding path between one of said two coupled data path units, said reverse forwarding path being unidirectional.

3. The processor of claim 1, wherein said scheduler includes one or more restrictions which limit the number of said instructions which can utilize said operands and said results.

4. The processor of claim 3, wherein one of said restrictions governs information available to said scheduler to include first information of a first cycle and second information of a second cycle, and wherein said scheduler forwards said instructions to said data path units based on said coupling fabric, said first information and said second information.

5. The processor of claim 1, wherein L=2.

6. The processor of claim 1, wherein said register blocks comprise multi-banked vector register files and multi-banked scalar register files.

7. A method of efficiently scheduling instructions in a very long instruction word digital signal processor having a plurality of data path units, each data path unit comprising a plurality of functional units, a plurality of register blocks having data operands and a coupling fabric configured to facilitate communication between said register blocks and said data path units, wherein said coupling fabric comprises a forwarding path configured to couple two of said data path units together and register blocks being coupled to fewer than all of said data path units, said method comprising the steps of:

receiving an instruction referencing said data operands that, upon execution, yields a result;

determining in a scheduler which of said data path units are capable of executing said instruction based on said coupling fabric and further based on one or more restrictions imposed by said scheduler;

selecting which of said capable data path units would be efficient for executing said instruction; and

forwarding said instruction to said selected data path unit for execution by said selected data path unit using said data operands.

8. The method of claim 7, wherein said register blocks comprise multi-banked vector register files and multi-banked scalar register files.

9. The method of claim 7, wherein one of said restrictions governs information available to said scheduler to include first information of a first cycle and second information of a second cycle.

10. The method of claim 7, wherein said plurality of data path units equals M data path units, wherein said coupling fabric is further configured to couple each of said register blocks to L of said M data path units, where 1<

L<

M.

11. The method of claim 10, wherein L=2.

12. A signal processing system for use in executing the instructions of a very long instruction word architecture, said system comprising:

register blocks configured to store data;

data path units comprising functional units which execute said instructions using said data;

a fabric configured to route data and instructions, said fabric linking said register blocks and said data path units, wherein each of said register blocks is linked to at least two of said data path units, but fewer than all of said data path units, and wherein at least one of said data path units is linked to another of said data path units;

a scheduler for scheduling instructions to said data path units, said scheduler configured to maintain information about said fabric, said register blocks, and said data path units; and

a selector for selecting which of said data path units is to execute an instruction.

13. The system of claim 12, wherein said register blocks comprise multi-banked vector register files and multi-banked scalar register files.

14. The system of claim 12, said fabric being further configured such that a first of said data path units may only forward data to a second of said data path units.

15. The system of claim 12, said fabric being further configured such that a first of said data path units and second of said data path units may only forward data to each other.

16. The system of claim 12, said scheduler comprising a restriction controlling when instructions and other data may be forwarded.

17. A digital signal processor for use in executing instructions of a very long instruction word model, said processor comprising:

N register blocks configured to store data;

M data path units, each of said M data path units comprising functional units for executing instructions using said data; and

a fabric configured to couple each of said N register blocks to L of said M data path units, where 1<

L<

M, to enable said data to be exchanged between said N register blocks and said M data path units, said fabric further configured to couple two of said M data path units together to enable said data to be exchanged between said two data path units; and

a scheduling mechanism to schedule instructions to said data path units, said mechanism including information about said N register blocks, said M data path units and said fabric.

18. The processor of claim 17, wherein each of said M data path units are associated with a numerical indexing scheme, and wherein said coupling fabric is further configured to couple each of a first of said M data path units and a last of said M data path units with only one of said N storage elements.

19. The processor of claim 18, wherein said first data path unit and said second data path unit are coupled to different ones of said storage elements.

20. The processor of claim 17, wherein said coupling fabric comprises a plurality of forwarding paths, each of said forwarding paths configured to couple two of said data path units together.

21. The processor of claim 20, wherein said coupling fabric further comprises a reverse forwarding path between one of said two coupled data path units.

22. The processor of claim 17, wherein L=2.

23. The processor of claim 17, wherein said N register blocks each comprise multi-banked vector register files and multi-banked scalar register files.

24. A method of executing instructions in a digital signal processor using a very long instruction word architecture, wherein said processor comprises a memory unit, data path units, register blocks and a coupling fabric configured to facilitate communication between and said register blocks and said data path units, and between said data path units, each of said register blocks being coupled to fewer than all of said data path units, said method comprising the steps of:

receiving at a scheduler, a first instruction from said memory unit;

determining, in said scheduler, which of said data path units are capable of executing said first instruction based on said coupling fabric;

selecting one of said data path units for executing said first instruction based on results of said determining step;

forwarding said first instruction from said scheduler to said one data path unit;

executing said first instruction in said one data path unit, producing a result; and

storing said result from said executing step in one of said register blocks coupled to said one data path unit.

25. The method of claim 24, wherein said register blocks comprise multi-banked vector register files and multi-banked scalar register files.

26. The method of claim 24, in which said executing step yields a result which is forwarded from said one data path unit to a second of said data path units.

27. The method of claim 24, wherein said determining step is responsive to restrictions imposed upon said scheduler.

28. The method of claim 27, in which said executing step yields a result which is forwarded from said one data path unit to a second of said data path units at a time determined by said restrictions.