On-die tessellation distribution
The tessellation processing rate of a graphics processor may be increased using of local tessellation work redistribution. The redistribution mechanism may avoid the need for large on-die buffers and, as the distribution is local, the performance and power penalty incurred by use of off-chip memory accesses may also avoided in some embodiments.
1. A method comprising: re-distributing, in an integrated circuit including a graphics processor, patches among parallel interconnected geometry processing fixed-function units without accessing memory external to said integrated circuit, said fixed function units comprising separate front and back ends;
storing tessellation factors in a memory on said integrated circuit;
reading said tessellation factors by said front end from said memory;
performing patch culling and transmission by said front end;
performing patch reception and patch tessellation by said back end; and
communicating over a communication bus between said front end and said back end.
2. The method of claim 1 further comprising creating an ordered list of parallel identified geometry processing fixed-function pipelines.
3. The method of claim 2 including sharing said list among said interconnected geometry processing fixed-function pipelines.
4. The method of claim 3 including sharing status of patch handles being processed across said geometry processing fixed function pipelines.
5. The method of claim 4 including identifying, in each of said geometry processing fixed-function pipelines, which geometry processing fixed-function pipelines is to process a next patch handle using said shared list and said shared status.
6. The method of claim 1 including broadcasting to all back ends an identifier for a back end designated to receive a next patch handle.
7. One or more non-transitory computer readable media storing instructions executed to perform a sequence comprising
re-distributing, in an integrated circuit including a graphics processor, patches among parallel interconnected geometry processing fixed-function units without accessing memory external to said integrated circuit, said fixed function units comprising separate front and back ends;
storing tessellation factors in a memory on said integrated circuit;
reading said tessellation factors by said front end from said memory;
performing patch culling and transmission by said front end;
performing patch reception and patch tessellation by said back end; and
communicating over a communication bus between said front end and said back end.
8. The media of claim 7 , further storing instructions to perform a sequence further comprising creating an ordered list of parallel identified geometry processing fixed-function pipelines.
9. The media of claim 8 , further storing instructions to perform a sequence including sharing said list among said interconnected geometry processing fixed-function pipelines.
10. The media of claim 9 , further storing instructions to perform a sequence including sharing status of patch handles being processed across said geometry processing fixed function pipelines.
11. The media of claim 10 , further storing instructions to perform a sequence including identifying, in each of said geometry processing fixed-function pipelines, which geometry processing fixed-function pipelines is to process a next patch handle using said shared list and said shared status.
12. The media of claim 11 , further storing instructions to perform a sequence including breaking said pipelines into separate front and back ends.
13. The media of claim 12 , further storing instructions to perform a sequence including providing a communication path from each front and to each back end.
14. The method of claim 13 , further storing instructions to perform a sequence including broadcasting to all back ends an identifier for a back end designated to receive a next patch handle.
15. An integrated circuit comprising:
a memory;
parallel interconnected geometry processing fixed function units; and
a graphics processor to redistribute patches among parallel interconnected geometry processing fixed-function units without accessing memory external to said integrated circuit said fixed function units having separate front and back ends, said front ends to perform patch culling and transmission and said back ends to perform patch reception from a front end and patch tessellation, said front ends to read tessellation factors from said memory; and
a communication bus between the front and back ends.
16. The circuit of claim 15 , said processor to further create an ordered list of parallel identified geometry processing fixed-function pipelines.
17. The circuit of claim 16 , said processor to share said list among said interconnected geometry processing fixed-function pipelines.
18. The circuit of claim 17 , said processor to share status of patch handles being processed across said geometry processing fixed function pipelines.
19. The circuit of claim 18 , said processor to identify, in each of said geometry processing fixed-function pipelines, which geometry processing fixed-function pipelines is to process a next patch handle using said shared list and said shared status.
20. The circuit of claim 19 , said processor to break said pipelines into separate front and back ends.
21. The circuit of claim 19 , said processor to provide a communication path from each front and to each back end.
22. The circuit of claim 21 , said processor to broadcast to all back ends an identifier for a back end designated to receive a next patch handle.