Automated graphics and compute tile interleave

1. A method to reduce external memory accesses in a graphics system having an on-chip memory and an external memory, comprising:

processing a combination of tiled graphics rendering operations and tiled compute shader operations of interdependent render targets, each of the interdependent render targets corresponding to a different image frame; and

interleaving the combination of the tiled graphics rendering operations and tiled compute shader operations automatically based on the processing,

wherein the tiled graphics rendering operations comprise raster operations, pixel processing operations, vertex processing operations, patch processing operations, or primitive processing operations, and

wherein the tiled graphics rendering operations are separate from the tiled compute shader operations.

2. The method of claim 1 , wherein the interleaving comprises determining a schedule of a sequence of interleaved tiled operations selected such that at least one intermediate data result of a first operation of the sequence is directly consumed from the on-chip memory by a second operation of the sequence.

3. The method of claim 1 , wherein the interleaving is scheduled to have a sequence of interleaved tiled operations selected to reduce traffic to the external memory associated with intermediate data results.

4. The method of claim 1 , wherein a data access pattern of graphics rendering operations and compute shader operations is analyzed to determine whether a global memory barrier can be safely removed as a condition for interleaving and removing memory barriers.

5. The method of claim 1 , wherein the interleaving is performed on a tile basis and a compute shader configured to perform the compute shader operations is configured for a tiled access pattern.

6. The method of claim 5 , wherein a workgroup dimension of the compute shader is redefined to be an integer divisor of a width and a height of a tile.

7. The method of claim 6 , wherein the processing of the combination comprises replacing, for the compute shader, image load instructions with tile buffer load instructions.

8. The method of claim 1 , wherein the processing of the combination comprises analyzing API calls and grouping application programming interface (API) calls to build a sequence of interleaved execution of graphics rendering operations and compute shader operations.

9. The method of claim 1 , wherein the processing of the combination comprises generating a directed acyclic graph (DAG) of interdependent render targets and interdependent tiles and utilize the DAG to schedule interleaving of graphics rendering operations and compute shader operations.

10. The method of claim 1 , wherein the interleaving comprises grouping API calls to form a sequence of interleaved execution of graphics rendering operations and compute shader operations.

11. The method of claim 1 , wherein the processing of the combination comprises analyzing a data access pattern of load and store operation to determine candidates for interleaving.

12. The method of claim 11 , wherein the processing of the combination comprises identifying a data access pattern of a statically predetermined strided pattern as a candidate for interleaving.

13. In a graphics processing system, a method comprising:

recompiling a compute shader to have a tiled access pattern; and

interleaving processing of a graphics renderer and the recompiled compute shader for a set of interdependent images, each of the interdependent images corresponding to a different image frame,

wherein the interleaving is performed on a tile-by-tile basis for the interdependent images,

wherein the processing of the graphics renderer comprises raster operations, pixel processing operations, vertex processing operations, patch processing operations, or primitive processing operations, and

wherein the processing of the graphics renderer is separate from the processing of the compute shader.

14. The method of claim 13 , wherein the interleaving processing has a sequence of interleaved operations selected to maintain at least some results of tile processing computations of dependent render targets in an on-chip memory of the graphics processing system.

15. The method of claim 13 , further comprising generating a directed acyclic graph (DAG) of interdependent render targets and interdependent tiles and utilizing the DAG to schedule interleaving of the processing of the graphics renderer and the recompiled compute shader.

16. The method of claim 13 , wherein the recompiling comprises:

recompiling the compute shader to access and output data in a tiled pattern compatible with the tiled access pattern of the graphics renderer.

17. The method of claim 16 , wherein recompiling the compute shader to access and output data in a tiled pattern comprising redefining a workgroup dimension of the compute shader to be an integer divisor of a width and a height of a tile.

18. The method of claim 17 , further comprising replacing, for the compute shader, image load instructions with tile buffer load instructions.

19. The method of claim 13 , further comprising removing a memory barrier between the graphics renderer and the compute shader.

20. The method of claim 13 , wherein the interleaving is scheduled to have a sequence of interleaved tiled operations selected to maintain at least one intermediate data result, of an interleaved combination resulting from the interleaving, in an on-chip memory.

21. The method of claim 13 , wherein the interleaving is scheduled to have a sequence of interleaved tiled operations selected to reduce traffic to an external memory associated with intermediate data results.

22. A non-transitory computer readable medium having computer code, which when executed on a processor implements a method to:

determine dependencies of graphics rendering operations and compute shader operations of a set of interdependent render target operations, each of the interdependent render target operations corresponding to a different image frame; and

schedule an interleaved order of tile processing of interleaved graphics rendering and compute shader operations to reduce traffic to an external memory of a graphics system by maintaining at least a subset of intermediate tile processing computations of the interleaved graphics rendering and computer shader operations in an on-chip memory of a graphics processing unit,

wherein the graphics rendering operations comprise raster operations, pixel processing operations, vertex processing operations, patch processing operations, or primitive processing operations, and

wherein the graphics rendering operations are separate from the compute shader operations.