Culling in a vertex processing unit

• US 7,468,726 B1
• Filed: 12/01/2005
• Issued: 12/23/2008
• Est. Priority Date: 12/01/2005
• Status: Active Grant

• Current Assignee: NVIDIA Corporation
• Original Assignee: NVIDIA Corporation
• Inventors: Wloka, Matthias M., Dietrich, Douglas Sim Jr.
• Primary Examiner(s): Chauhan; Ulka
• Assistant Examiner(s): Hajnik; Daniel F
• Application Number:
  US11/293,876
• Time in Patent Office: 1,118 Days
• Field of Search: None
• US Class Current: 345/421
• CPC Class Codes: G06T 15/40 Hidden part removal

Abstract

A graphics processor performs culling of invisible primitives in a vertex processing unit that includes a geometry shader or other processing engine that performs per-primitive operations. Primitives can be culled after clip space coordinates for the vertices have been computed and in some instances before at least one other vertex attribute has been computed. To the extent that this early culling reduces the number of vertices for which the full set of attributes is computed or reduces the number of primitives or vertices delivered to downstream units, throughput of the processor is increased.

Claims

1. A graphics processor comprising:

a vertex processing unit coupled to receive data for a plurality of vertices, each vertex being associated with at least one of a plurality of primitives in a scene to be rendered,

the vertex processing unit being configured to apply a visibility test to each of the plurality of primitives and to cull primitives that fail the visibility test from a list of primitives to be processed, the vertex processing unit comprising;

a programmable vertex shader configured to compute one or more per-vertex attribute values for a vertex; and

a primitive processing engine configured to apply the visibility test and to cull each of the primitives that fails the visibility test from the list of primitives to be processed, wherein the visibility test is defined by reference to clip space coordinates of vertices of a primitive, wherein the visibility test is a diamond visibility test, wherein each pixel has an associated internal region having a diamond shape that covers a center of the pixel, and wherein if a primitive intersects the internal region of more than one pixel, the primitive is visible and is not culled;

wherein the vertex shader is programmed to compute clip space coordinates for the vertices during a first pass executed prior to applying the visibility test and to compute at least one other per-vertex attribute value during a second pass executed subsequently to culling the primitives that fail the visibility test; and

a downstream processing unit coupled to receive the primitives that are not culled from the vertex processing unit and configured to generate pixel data from the received primitives,

wherein vertices of primitives that fail the visibility test are not received by the downstream processing unit.

2. The graphics processor of claim 1 wherein the vertex processing unit is further configured to compute at least one per-vertex attribute value for each vertex that is associated with at least one primitive that is not culled.

3. The graphics processor of claim 1 wherein the primitive processing engine is a programmable geometry shader.

4. The graphics processor of claim 3 wherein the geometry shader is programmed to read vertices of primitives from a vertex buffer and to write vertices of primitives that are not culled to a primitive buffer.

5. The graphics processor of claim 4 wherein the vertex shader is further configured to read the vertices from the primitive buffer and to compute at least one per-vertex attribute value for each vertex read from the primitive buffer.

6. The graphics processor of claim 3 wherein the geometry shader is further configured to read vertices of primitives from a vertex buffer and to set a visibility flag in the vertex buffer for each of the vertices of each primitive that is not culled.

7. The graphics processor of claim 6 wherein the vertex shader is further configured to compute at least one of the per-vertex attribute values for a vertex only if the visibility flag for that vertex has been set.

8. The graphics processor of claim 3 wherein the geometry shader is further configured to read the vertices of the primitives from a vertex buffer and to generate a list of identifiers of the vertices of each primitive that is not culled.

9. The graphics processor of claim 8 wherein the vertex shader is further configured to use the list of identifiers to select vertices from the vertex buffer for which a per-vertex attribute is to be computed.

10. A method of processing vertex data, the method comprising:

receiving, at a vertex processing unit of a graphics processor, vertex data for a plurality of vertices, each vertex being associated with at least one of a plurality of primitives in a scene to be rendered;

applying a visibility test to each of the plurality of primitives, wherein the visibility test is a diamond visibility test, wherein each pixel has an associated internal region having a diamond shape that covers a center of the pixel, and wherein if a primitive intersects the internal region of more than one pixel, the primitive is visible and is not culled;

culling each primitive that fails the visibility test;

subsequently to the act of culling, computing at least one per-vertex attribute value for each vertex that is associated with at least one primitive that is not culled; and

subsequently to the act of computing, propagating the vertices of each primitive that is not culled during the act of culling from the vertex processing unit to a downstream processing unit of the graphics processor.

11. The method of claim 10 wherein the acts of computing the clip-space coordinates and computing the attribute value are performed by a programmable vertex shader in the vertex processing unit.

12. The method of claim 11 wherein the acts of applying the visibility test and culling each primitive that fails the visibility test are performed by a programmable geometry shader in the vertex processing unit.

13. The method of claim 10 wherein the act of computing at least one attribute value is performed once per vertex regardless of the number of primitives with which the vertex is associated.

14. The method of claim 10 wherein the act of computing the at least one attribute is performed by a programmable vertex shader in the vertex processing unit.

15. The method of claim 10 further comprising:

prior to the act of applying the visibility test, computing clip-space coordinates for each vertex,

wherein the visibility test is based at least in part on the clip-space coordinates.

16. The method of claim 15 wherein the visibility test is based at least in part on a view frustum defined in the clip space.

17. The method of claim 15 wherein the visibility test is based at least in part on an orientation of the primitive in the clip space.

18. The method of claim 15 wherein the visibility test is based at least in part on a signed area of the primitive in the clip-space coordinates.

19. The method of claim 10 further comprising:

in the downstream processing unit, performing a further culling operation on each primitive propagated from the vertex processing unit.