Architecture and instructions for accessing multi-dimensional formatted surface memory

• US 9,519,947 B2
• Filed: 09/24/2010
• Issued: 12/13/2016
• Est. Priority Date: 09/25/2009
• Status: Active Grant

• Current Assignee: NVIDIA Corporation
• Original Assignee: NVIDIA Corporation
• Inventors: Nickolls, John R., Fahs, Brian, Nyland, Lars, Lindholm, John Erik, Johnson, Richard Craig
• Primary Examiner(s): BEUTEL, WILLIAM A
• Application Number:
  US12/890,171
• Publication Number:
  US 20110074802A1
• Time in Patent Office: 2,272 Days
• US Class Current: 1/1
• CPC Class Codes: G06T 1/60 Memory management

Abstract

One embodiment of the present invention sets forth a technique for a program to access multi-dimensional formatted graphics surface memory. Multi-dimensional memory objects called “surfaces” stored in a user-specified data or pixel format and arranged in a graphics optimized layout are accessed by programs using surface instructions. A set of memory access instructions e.g., load, store, reduce, and atomic, referred to as surface instructions, may be used to access the surfaces. Coordinate bounds checking is performed with configurable clamping. Caching behavior may also be specified by the surface instructions. Data format conversion and packing to a specified storage format is supported for store, reduction, and atomic surface instructions. Data format conversion and unpacking from a specified storage format is supported for loads and atomic surface instructions.

Claims

1. A method for accessing multi-dimensional formatted surface memory by a program, comprising:

receiving a surface instruction that specifies an x-coordinate of a sample and a y-coordinate of the sample, wherein the sample is associated with a multi-dimensional formatted surface;

computing an address of a location in the multi-dimensional formatted surface memory based on the x-coordinate, the y-coordinate, and attributes of the multi-dimensional formatted surface, by computing a plurality of address fields included in the address, wherein two or more address fields in the plurality of address fields comprise one or more bits corresponding to the x-coordinate and one or more bits corresponding to the y-coordinate; and

accessing the sample via the address;

wherein;

a first address field in the plurality of address fields identifies a block within the multi-dimensional formatted surface, and

a second address field in the plurality of address fields identifies a group of blocks (GOB) within the block.

2. The method of claim 1, wherein the surface instruction specifies the coordinates in bytes.

3. The method of claim 1, wherein the multi-dimensional formatted surface is configured by the program to access the samples in a pattern using multi-dimensional tiled addressing.

4. The method of claim 1, further comprising:

determining that the location is out-of-bounds of the multi-dimensional formatted surface; and

returning a default value for the surface instruction based on a mode specified by the surface instruction.

5. The method of claim 1, further comprising:

determining that the location is out-of-bounds of the multi-dimensional formatted surface; and

discarding the surface instruction based on a mode specified by the surface instruction.

6. The method of claim 1, further comprising:

determining that the location is out-of-bounds of the multi-dimensional formatted surface; and

reading data stored at a nearest location that is nearest to the location within the multi-dimensional formatted surface based on a mode specified by the surface instruction.

7. The method of claim 1, further comprising:

determining that the location is out-of-bounds of the multi-dimensional formatted surface; and

writing data to a nearest location that is nearest to the location within the multi-dimensional formatted surface based on a mode specified by the surface instruction.

8. The method of claim 1, further comprising caching data read from or written to the location based on a mode specified by the surface instruction.

9. The method of claim 1, further comprising:

receiving a second surface instruction that is configured to query an attribute of the multi-dimensional formatted surface; and

returning the attribute of the multi-dimensional formatted surface.

10. The method of claim 1, further comprising:

determining that the surface instruction is configured to store data to the multi-dimensional formatted surface; and

configuring a parallel thread processor that is coupled to the multi-dimensional formatted surface memory to convert the data to a format specified for the multi-dimensional formatted surface by the program before storing the data.

11. The method of claim 1, further comprising:

determining that the surface instruction is configured to load data from the multi-dimensional formatted surface; and

configuring a parallel thread processor that is coupled to the multi-dimensional formatted surface memory to convert the data from a format specified for the multi-dimensional formatted surface by the program before returning the data.

12. The method of claim 1, further comprising:

determining that the surface instruction is configured to perform an atomic operation on data from the multi-dimensional formatted surface; and

configuring a parallel thread processor that is coupled to the multi-dimensional formatted surface memory to convert the data from a format specified for the multi-dimensional formatted surface by the program before performing the atomic operation on the data.

13. The method of claim 1, further comprising:

determining that the surface instruction is configured to perform an atomic reduction operation on data from the multi-dimensional formatted surface; and

configuring a parallel thread processor that is coupled to the multi-dimensional formatted surface memory to convert the data from a format specified for the multi-dimensional formatted surface by the program before performing the atomic reduction operation on the data.

14. The method of claim 1, wherein the surface instruction is configured to load data from the multi-dimensional formatted surface, store data to the multi-dimensional formatted surface, query an attribute of the multi-dimensional formatted surface, perform an atomic operation on data from the multi-dimensional formatted surface, or perform an atomic reduction operation on data from the multi-dimensional formatted surface.

15. The method of claim 1, wherein the surface instruction comprises a query to obtain the attributes of the multi-dimensional formatted surface.

16. The method of claim 1, wherein a third address field in the plurality of address fields identifies a cache line in the GOB.

17. The method of claim 16, wherein a fourth address field in the plurality of address fields identifies a sector in the cache line.

18. The method of claim 17, wherein a fifth address field in the plurality of address fields identifies a byte in the sector.

19. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor executing a program to access a multi-dimensional formatted surface memory, by performing the steps of:

receiving a surface instruction that specifies an x-coordinate of a sample and a y-coordinate of the sample, wherein the sample is associated with a multi-dimensional formatted surface;

computing an address of a location in the multi-dimensional formatted surface memory based on the x-coordinate, the y-coordinate, and attributes of the multi-dimensional formatted surface, by computing a plurality of address fields included in the address, wherein two or more address fields in the plurality of address fields comprise one or more bits corresponding to the x-coordinate and one or more bits corresponding to the y-coordinate; and

accessing the sample via the address;

wherein;

a first address field in the plurality of address fields identifies a block within the multi-dimensional formatted surface, and

a second address field in the plurality of address fields identifies a group of blocks (GOB) within the block.

20. A computer-readable storage medium of claim 19, wherein the surface instruction specifies the coordinates in bytes.

21. A computer-readable storage medium of claim 19, wherein the multi-dimensional formatted surface is configured by the program to access the samples in a pattern using multi-dimensional tiled addressing.

22. A computer-readable storage medium of claim 19, further comprising:

determining that the location is out-of-bounds of the multi-dimensional formatted surface; and

returning a default value for the instruction based on a mode specified by the surface instruction.

23. A computer-readable storage medium of claim 19, further comprising caching data read from or written to the location based on a mode specified by the surface instruction.

24. The non-transitory computer-readable storage medium of claim 19, wherein the surface instruction comprises a query to obtain the attributes of the multi-dimensional formatted surface.

25. A system comprising:

a multi-dimensional formatted surface memory; and

a parallel thread processor coupled to the multi-dimensional formatted surface memory and configured to;

receive a surface instruction that specifies an x-coordinate of a sample and a y-coordinate of the sample, wherein the sample is associated with a multi-dimensional formatted surface;

compute an address of a location in the multi-dimensional formatted surface memory based on the x-coordinate, the y-coordinate, and attributes of the multi-dimensional formatted surface, by computing a plurality of address fields included in the address, wherein two or more address fields in the plurality of address fields comprise one or more bits corresponding to the x-coordinate and one or more bits corresponding to the y-coordinate; and

access the sample via the address;

wherein;

a first address field in the plurality of address fields identifies a block within the multi-dimensional formatted surface, and

a second address field in the plurality of address fields identifies a group of blocks (GOB) within the block.

26. The system of claim 25, wherein the surface instruction comprises a query to obtain the attributes of the multi-dimensional formatted surface.

27. A method for accessing multi-dimensional formatted surface memory by a program, comprising:

receiving a surface instruction that specifies an x-coordinate of a sample and a y-coordinate of the sample, wherein the sample is associated with a multidimensional formatted surface;

computing an address of a location in the multi-dimensional formatted surface memory based on the x-coordinate, the y-coordinate, and attributes of the multi-dimensional formatted surface, by computing a plurality of address fields included in the address, wherein two or more address fields in the plurality of address fields comprise one or more bits corresponding to the x-coordinate and one or more bits corresponding to the y-coordinate; and

accessing the sample via the address;

wherein;

a first address field in the plurality of address fields identifies a block within the multi-dimensional formatted surface and is computed based on a first bit corresponding to the x-coordinate and a first bit corresponding to the y-coordinate,

a second address field in the plurality of address fields identifies a group of blocks (GOB) within the block and is computed based on a second bit corresponding to the x-coordinate and a second bit corresponding to the y-coordinate,

a third address field in the plurality of address fields identifies a cache line in the GOB and is computed based on a third bit corresponding to the x-coordinate and a third bit corresponding to the y-coordinate,

a fourth address field in the plurality of address fields identifies a sector in the cache line and is computed based on a fourth bit corresponding to the x-coordinate and a fourth bit corresponding to the y-coordinate, and

a fifth address field in the plurality of address fields identifies a byte in the sector and is computed based on a fifth bit corresponding to the x-coordinate and a fifth bit corresponding to the y-coordinate.