Unified virtual addressed register file

• US 8,766,996 B2
• Filed: 06/21/2006
• Issued: 07/01/2014
• Est. Priority Date: 06/21/2006
• Status: Active Grant

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: Du, Yun, Jiao, Guofang, Yu, Chun, Hsu, De Dzwo
• Primary Examiner(s): Wu, Xiao
• Assistant Examiner(s): Tseng, Charles
• Application Number:
  US11/472,701
• Publication Number:
  US 20070296729A1
• Time in Patent Office: 2,932 Days
• Field of Search: 345501-506, 345530-574
• US Class Current: 345/559
• CPC Class Codes: G06F 9/3012 Organisation of register space, e.g. banked or distributed register file
• CPC Class Codes: G06F 9/30123 according to context, e.g. thread buffers
• CPC Class Codes: G06F 9/30138 Extension of register space, e.g. register cache
• CPC Class Codes: G06F 9/384 Register renaming
• CPC Class Codes: G06F 9/3851 from multiple instruction streams, e.g. multistreaming
• CPC Class Codes: G06T 15/005 General purpose rendering architectures

Abstract

A multi-threaded processor is provided, such as a shader processor, having an internal unified memory space that is shared by a plurality of threads and is dynamically assigned to threads as needed. A mapping table that maps virtual registers to available internal addresses in the unified memory space so that thread registers can be stored in contiguous or non-contiguous memory addresses. Dynamic sizing of the virtual registers allows flexible allocation of the unified memory space depending on the type and size of data in a thread register. Yet another feature provides an efficient method for storing graphics data in the unified memory space to improve fetch and store operations from the memory space. In particular, pixel data for four pixels in a thread are stored across four memory devices having independent input/output ports that permit the four pixels to be read in a single clock cycle for processing.

Claims

1. A multi-threaded processor comprising:

a thread scheduler configured to receive multiple threads, each of the threads having one or more thread registers, determine a dynamic size of each of the thread registers of each of the multiple threads, create, in a unified register file, a virtual register mapping that defines virtual registers having contiguous addresses and that maps the contiguous virtual registers to internal addresses in a unified register file'"'"'s unified memory space mapped by the virtual registers, wherein the virtual register mapping maps at least one of the contiguous virtual registers to a non-contiguous internal address in the unified memory space in accordance with the determined size of each of the thread registers, wherein the non-contiguous internal address addresses a portion of the unified memory space, wherein a size of the portion of the unified memory space that is addressed by the non-contiguous internal address is based on the determined size of each of the thread registers of each of the multiple threads, wherein the thread scheduler is further configured to allocate one or more of the contiguous virtual registers to the one or more thread registers of each of the multiple threads, a size of at least one of the allocated virtual registers being different from a size of another of the allocated virtual registers, and the size of the at least one of the allocated virtual registers being based on the determined size of each of the thread registers of each of the multiple threads, and store content of each of the thread registers of each of the multiple threads at the internal addresses in the unified memory space in accordance with the virtual register mapping that also indicates which of the one or more allocated virtual registers is allocated to which thread of the multiple threads;

the unified register file coupled to the thread scheduler;

and a processing unit coupled to the unified register file and configured to retrieve content of the thread registers of each of the multiple threads from the internal addresses in the unified memory space in accordance with the virtual register mapping that maps the one or more allocated virtual registers to internal addresses in the unified memory space.

2. The multi-threaded processor of claim 1 wherein the processing unit is further configured to

process the content of the thread registers of the multiple threads to obtain a result, and

store the result to one or more other virtual registers mapped to one or more other internal addresses in the unified memory space.

3. The multi-threaded processor of claim 1 wherein the thread scheduler is configured to de-allocate the one or more virtual registers allocated to one thread of the multiple threads once the processing unit has processed that thread.

4. The multi-threaded processor of claim 1 wherein the mapped internal addresses in the unified memory space corresponding to the virtual registers allocated for the received multiple threads comprise the at least one non-contiguous internal address in the unified memory space.

5. The multi-threaded processor of claim 1 wherein the thread scheduler maintains the virtual register mapping as a mapping table of allocated virtual registers mapped to internal addresses in the unified memory space.

6. The multi-threaded processor of claim 1 wherein at least one thread of the multiple threads includes pixel data for a plurality of pixels.

7. The multi-threaded processor of claim 6 wherein the unified register file is divided into a plurality of memory banks and the pixel data for two or more pixels of the plurality of pixels is stored across two or more of the plurality of memory banks.

8. The multi-threaded processor of claim 7 wherein the memory banks include a plurality of simultaneous read and write ports that permit stored content for the thread registers to be read from the memory banks while content for new thread registers is stored to the same memory banks.

9. The multi-threaded processor of claim 7 wherein the processing unit is configured to retrieve pixel data for the two or more pixels stored across two or more memory banks in a single clock cycle.

10. A multi-threaded processor comprising:

means for receiving multiple threads, each of the threads having one or more thread registers;

means for determining a dynamic size of each of the thread registers of each of the multiple threads;

means for creating, in a unified register file, a virtual register mapping that defines virtual registers having contiguous addresses and that maps the contiguous virtual registers to internal addresses in the unified register file'"'"'s unified memory space mapped by the virtual registers, wherein the virtual register mapping maps at least one of the contiguous virtual registers to a non-contiguous internal address in the unified memory space in accordance with the determined size of each of the thread registers, wherein the non-contiguous internal address addresses a portion of the unified memory space, and wherein a size of the portion of the unified memory space that is addressed by the non-contiguous internal address is based on the determined size of each of the thread registers of each of the multiple threads; and

means for allocating one or more of the contiguous virtual registers to the one or more thread registers of each of the multiple threads, a size of at least one of the allocated virtual registers being different from a size of another of the allocated virtual registers, and the size of the at least one of the allocated virtual registers being based on the determined size of each of the thread registers of each of the multiple threads; and

means for storing content of each of the thread registers of each of the multiple threads at the internal addresses in the unified memory space in accordance with the virtual register mapping that also indicates which of the one or more allocated virtual registers is allocated to which thread of the multiple threads.

11. The multi-threaded processor of claim 10 further comprising:

means for retrieving content for the thread registers of each of the multiple threads from the unified memory space in accordance with the one or more allocated virtual registers and the virtual register mapping.

12. The multi-threaded processor of claim 10 further comprising:

means for processing content of the thread registers of the multiple threads stored in the unified memory space in accordance with the one or more allocated virtual registers and the virtual register mapping to obtain a result; and

means for storing the result to one or more other virtual registers mapped to one or more other internal addresses in the unified memory space.

13. The multi-threaded processor of claim 10 wherein the unified register file is divided into a plurality of memory banks, and pixel data for a pixel is stored across the plurality of memory banks.

14. The multi-threaded processor of claim 13 wherein the pixel data comprises position, color and texture data that is stored across the plurality of memory banks.

15. A method comprising:

receiving multiple threads at a multi-threaded processor, each of the multiple threads including one or more thread registers;

determining a dynamic size of each of the thread registers of each of the multiple threads;

creating, in the multi-threaded processor'"'"'s unified register file, a virtual register mapping that defines virtual registers having contiguous addresses and that maps the contiguous virtual registers to internal addresses in the unified register file'"'"'s unified memory space mapped by the virtual registers, wherein the virtual register mapping maps at least one of the contiguous virtual registers to a non-contiguous internal address in the unified memory space in accordance with the determined size of each of the thread registers, wherein the non-contiguous internal address addresses a portion of the unified memory space, and wherein a size of the portion of the unified memory space that is addressed by the non-contiguous internal address is based on the determined size of each of the thread registers of each of the multiple threads; and

allocating one or more of the contiguous virtual registers to the one or more thread registers of each of the multiple threads, a size of at least one of the allocated virtual registers being different from a size of another of the allocated virtual registers, and the size of the at least one of the allocated virtual registers being based on the determined size of each of the thread registers of each of the multiple threads; and

storing content of the each of the thread registers of each of the multiple threads at the internal addresses in the unified memory space in accordance with the virtual register mapping that also indicates which of the one or more allocated virtual registers is allocated to which thread of the multiple.

16. The method of claim 15 wherein the one or more virtual registers are allocated to correspond to a size of each of the thread registers of each thread of the multiple threads.

17. The method of claim 15 further comprising

retrieving content for the thread registers of each of the multiple threads stored in the unified memory space in accordance with the one or more allocated virtual registers and the virtual register mapping;

and

de-allocating the one or more virtual registers allocated to one thread of the multiple threads once that thread has been processed.

18. The method of claim 15 further comprising:

re-allocating the one or more virtual registers to a new thread.

19. The method of claim 15 wherein at least one thread of the multiple threads includes pixel data for a plurality of pixels.

20. The method of claim 19 wherein the pixel data for a pixel of the plurality of the pixels is stored across two or more memory storage devices.

21. The method of claim 20 further comprising:

retrieving the pixel data stored in the two or more memory storage devices while simultaneously storing data to the two or more memory storage devices.

22. A non-transitory machine-readable medium as a memory storage device tangibly storing one or more instructions for processing multiple threads in a multi-threaded processor, which when executed by the multi-threaded processor causes the multi-threaded processor to:

receive multiple threads, each of the multiple threads including one or more thread registers;

determine a dynamic size of each of the thread registers of each of the multiple threads;

create, in a multi-threaded processor'"'"'s unified register file, a virtual register mapping that defines virtual registers having contiguous addresses and that maps the contiguous virtual registers to internal addresses in a unified register file'"'"'s unified memory space mapped by the virtual registers, wherein the virtual register mapping maps at least one of the contiguous virtual registers to a non-contiguous internal address in the unified memory space in accordance with the determined size of each of the thread registers, wherein the non-contiguous internal address addresses a portion of the unified memory space, and wherein a size of the portion of the unified memory space that is addressed by the non-contiguous internal address is based on the determined size of each of the thread registers of each of the multiple threads;

allocate one or more of the contiguous virtual registers to the one or more thread registers of each of the multiple threads, a size of at least one of the allocated virtual registers being different from a size of another of the allocated virtual registers, and the size of the at least one of the allocated virtual registers being based on the determined size of each of the thread registers of each of the multiple threads; and

store content of each of the thread registers of each of the multiple threads at the internal address in the unified memory space in accordance with the virtual register mapping that also indicates which of the one or more allocated virtual registers is allocated to which thread of the multiple threads.

23. The machine-readable medium of claim 22 further having one or more instructions which when executed cause the multi-threaded processor to:

retrieve content of the thread registers of each of the multiple threads stored in the unified memory space in accordance with the one or more allocated virtual registers and the virtual register mapping; and

de-allocate the one or more virtual registers allocated to one thread of the multiple threads once that thread has been processed.

24. The machine-readable medium of claim 22 further having one or more instructions which when executed cause the multi-threaded processor to:

re-allocate the one or more virtual registers to a new thread.

25. A graphics processor comprising:

a cache memory for receiving external instructions;

a texture engine for storing graphics texture data;

a multi-threaded processor coupled to the cache memory and texture engine, the multi-threaded processor configured to

receive multiple threads each having one or more thread registers;

determine a dynamic size of each of the thread registers of each thread of the multiple threads;

create, in the multi-threaded processor'"'"'s unified register file, a virtual register mapping that defines virtual registers having contiguous addresses and that maps the contiguous virtual registers to internal addresses in the unified register file'"'"'s unified memory space mapped by the virtual registers, wherein the virtual register mapping maps at least one of the contiguous virtual registers to a non-contiguous internal address in the unified memory space in accordance with the determined size of each of the thread registers of each thread of the multiple threads, wherein the non-contiguous internal address addresses a portion of the unified memory space, and wherein a size of the portion of the unified memory space that is addressed by the non-contiguous internal address is based on the determined size of each of the thread registers of each of the multiple threads;

allocate one or more of the contiguous virtual registers to the one or more thread registers of each of the multiple threads, a size of at least one of the allocated virtual registers being different from a size of another of the allocated virtual registers, and the size of the at least one of the allocated virtual registers being based on the determined size of each of the thread registers of each of the multiple threads; and

store content of each of the thread registers of each of the multiple threads at the internal addresses in the unified memory space in accordance with the virtual register mapping that also indicates which of the one or more allocated virtual registers is allocated to which thread of the multiple threads.

26. The graphics processor of claim 25 wherein the multi-threaded processor is further configured to

obtain one or more instructions associated with the received multiple threads from the cache memory;

and

obtain texture data associated with the received multiple threads from the texture engine.

27. A wireless communication device comprising:

a communication interface to wirelessly communicate with other devices;

a graphics processor coupled to provide graphics data to the display unit, the graphics processor configured to

receive multiple threads each including one or more thread registers;

determining a dynamic size of each of the thread registers of each thread of the multiple threads;

create a unified register file, a virtual register mapping that defines virtual registers having contiguous addresses and that maps the contiguous virtual registers to internal addresses in the unified register file'"'"'s unified memory space mapped by the virtual registers, wherein the virtual register mapping maps at least one of the contiguous virtual registers to a non-contiguous internal address in the unified memory space in accordance with the determined size of each of the thread registers of each thread of the multiple threads, wherein the non-contiguous internal address addresses a portion of the unified memory space, and wherein a size of the portion of the unified memory space that is addressed by the non-contiguous internal address is based on the determined size of each of the thread registers of each of the multiple threads;

allocate one or more of the contiguous virtual registers to the one or more thread registers of each of the multiple threads, a size of at least one of the allocated virtual registers being different from a size of another of the allocated virtual registers, and the size of the at least one of the allocated virtual registers being based on the determined size of each of the thread registers of each of the multiple threads; and

store content of each of the thread registers of each of the multiple threads at the internal address in the unified memory space in accordance with the virtual mapping that also indicates which of the one or more allocated virtual registers is allocated to which thread of the multiple threads.