Method for efficient handling of texture cache misses by recirculation

• US 6,259,460 B1
• Filed: 03/26/1998
• Issued: 07/10/2001
• Est. Priority Date: 03/26/1998
• Status: Expired due to Term

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Silicon Graphics Incorporated (Graphics Properties Holdings, Inc.)
• Inventors: Gossett, Carroll Philip, Goudy, Mark, Bentz, Ole
• Primary Examiner(s): Chauhan, Ulka J.
• Application Number:
  US09/048,099
• Time in Patent Office: 1,202 Days
• Field of Search: 345/501-506, 345/520, 345/521, 345/514, 711/118, 711/144, 711/145, 711/169
• US Class Current: 345/552
• CPC Class Codes: G06T 1/60 Memory management

Abstract

A method of a computer graphics system recirculates texture cache misses into a graphics pipeline without stalling the graphics pipeline, increasing the processing speed of the computer graphics system. The method reads data from a texture cache memory by a read request placed in the graphics pipeline sequence, then reads the data from the texture cache memory if the data is stored in the texture cache memory and places the data in the pipeline sequence. If the data is not stored in the texture cache memory, the method recirculates the read request in the pipeline sequence by indicating in the pipeline sequence that the data is not stored in the texture cache memory, placing the read request at a subsequent, determined place in the pipeline sequence, reading the data into the texture cache memory from a main memory, and executing the read request from the subsequent, determined place and after the data has been read into the texture cache memory.

Claims

1. A method for continuous operation of a graphics pipeline, comprising:

continuing partially-formed pixels in the graphics pipeline upon a texture cache fault; and

recirculating in the graphics pipeline the partially-formed pixels to be again serviced by the texture cache.

2. The method according to claim 1, wherein the recirculating is repeated until the texture cache service results in a texture cache hit.

3. The method according to claim 1, wherein the partially-formed pixels are designated initially as sleeping and remain designated as sleeping upon the texture cache fault.

4. The method according to claim 3, wherein the recirculating is executed based upon whether the partially-formed pixels are designated as sleeping.

5. The method according to claim 3, wherein the partially-formed pixels are designated as sleeping by setting sleep bits, and the sleep bits are reset upon a texture cache hit corresponding to the partially-formed pixels.

6. The method according to claim 1, wherein the recirculating places the partially-formed pixels into the graphics pipeline at an earlier, predetermined slot.

7. The method according to claim 1, wherein the partially-formed pixels are grouped according to primitives, and the partially-formed pixels are recirculated in order based upon the primitives.

8. The method according to claim 1, wherein the partially-formed pixels are both recirculated in and continue along the graphics pipeline.

9. The method according to claim 1, wherein while the partially-formed pixels are being recirculated, other pixels are serviced by the texture cache and data corresponding to the partially-formed pixels is read from a memory to the texture cache.

10. The method according to claim 1, wherein the graphics pipeline is not stalled upon recirculation of the partially-formed pixels.

11. An apparatus including a texture cache and for continuous operation of a graphics pipeline, said apparatus comprising:

a scheduler determining whether to place one of a quad of pixels and an open slot into the graphics pipeline based upon whether pixels are recirculated in the graphics pipeline; and

a controller, coupled to the scheduler through the graphics pipeline, passing one of the quad of pixels and recirculated pixels into the graphics pipeline.

12. The apparatus according to claim 11, further comprising a recirculation bus coupled to the controller through the graphics pipeline and transmitting the recirculated pixels to the controller based upon a texture cache miss.

13. The apparatus according to claim 11, wherein the controller places the recirculated pixels into the open slot.

14. The apparatus according to claim 11, further comprising a recirculation point at which the quad of pixels are recirculated to the controller upon a texture cache miss.

15. The apparatus according to claim 11, further comprising a recirculation point at which the quad of pixels are recirculated to the controller if at least one sleep bit in the quad of pixels is set.

16. The apparatus according to claim 15, wherein sleep bits corresponding to the pixels are reset upon a texture cache hit.

17. A method of a graphics pipeline, comprising:

determining 2×

2 regions for each sleeping pixel in a quad of pixels in the graphics pipeline in both fine and coarse levels of detail;

computing a footprint in each level of detail that covers as many of the 2×

2 regions as possible, with a maximum footprint size of 4×

4 texels;

if texels corresponding to each of the footprints do not reside in a texture cache, issuing a texture cache fault to retrieve required texels from a memory, opening by a raster unit an empty slot in the graphics pipeline, and recirculating the quad of pixels in the graphics pipeline;

if texels corresponding to each of the footprints reside in the texture cache, reading the texels from the texture cache, selecting up to 4 2×

2 regions from each of the footprints based on an offset of each of the 2×

2 regions within each of the footprints, and waking up the sleeping pixels whose 2×

2 regions were covered by the footprints.

18. The method according to claim 17, wherein the quad of pixels continuously recirculates until texels corresponding to the footprints are read from the texture cache.

19. The method according to claim 18, wherein attribute tokens, tags, and barycentric coordinates are synchronized with the empty slot.

20. An apparatus including a continuously-operating graphics pipeline, said apparatus comprising:

a raster unit comprising a scheduler determining whether to place one of a quad of pixels and an open slot into the graphics pipeline based upon whether pixels are recirculated in the graphics pipeline; and

a texture unit comprising a controller, coupled to the scheduler through the graphics pipeline, passing one of the quad of pixels and recirculated pixels into the graphics pipeline.

21. The apparatus according to claim 20, further comprising a shader unit coupled to the texture unit and receiving one of the quad of pixels and the open slot.

22. The apparatus according to claim 20, wherein the texture unit further comprises a texture cache and wherein the scheduler determines whether to place the one of the quad of pixels and the open slot into the graphics pipeline based upon whether a texture cache miss has occurred corresponding to a sleeping quad of pixels.

23. The apparatus according to claim 22, further comprising a memory coupled to the texture cache and from which texels corresponding to the sleeping quad of pixels are read if the texture cache miss has occurred.

24. The apparatus according to claim 23, wherein the sleeping quad of pixels is passed into the empty slot by the controller and the graphics pipeline is configured such that the texels corresponding to the sleeping quad of pixels are available in the texture cache during recirculation of the sleeping quad of pixels in the graphics pipeline.

25. A method of a computer graphics system that recirculates texture cache misses into a graphics pipeline sequence without stalling the graphics pipeline sequence, said method comprising:

reading data from a texture cache memory by a read request placed in the graphics pipeline sequence;

if the data is stored in the texture cache memory, reading the data from the texture cache memory and placing the data in the graphics pipeline sequence; and

if the data is not stored in the texture cache memory, recirculating the read request in the graphics pipeline sequence by indicating in the graphics pipeline sequence that the data is not stored in the texture cache memory, placing the read request at a subsequent, determined place in the graphics pipeline sequence, reading the data into the texture cache memory from a main memory, and executing the read request from the subsequent, determined place and after the data has been read into the texture cache memory.