Automated conversion of GPGPU workloads to 3D pipeline workloads
Systems, apparatuses and methods may provide for receiving a general purpose graphics processing unit (GPGPU) workload and converting the GPGPU workload to a three-dimensional (3D) workload. Additionally, the 3D workload may be dispatched to a 3D pipeline. In one example, converting the GPGPU workload to the 3D workload includes identifying a plurality of thread groups in the GPGPU workload and mapping the plurality of thread groups to a 3D matrix of cubes.
1. A computing device comprising:
a host processor including a workload converter to convert a general purpose graphics processing unit (GPGPU) workload to a three-dimensional (3D) workload by:
an identification of a plurality of thread groups in the GPGPU workload; and
an association of the plurality of thread groups to a 3D matrix of cubes;
a graphics processor including:
a command streamer to dispatch the 3D workload; and
a 3D pipeline separate from the workload converter and including 3D-specific fixed function units, the 3D pipeline to receive the 3D workload from the command streamer and process the 3D workload with the 3D-specific fixed function units;
a display interface to receive data associated with an execution of the 3D workload; and
a display device to generate a display output based on the data in the display interface.
2. The computing device of claim 1 , wherein each cube is to include an array of two-dimensional (2D) primitives.
3. The computing device of claim 2 , wherein each primitive is to be a quad.
4. The computing device of claim 2 , wherein the 3D pipeline includes:
a pixel shader interface; and
a vertex shader to identify, for each of the 2D primitives associated with the 3D workload, a plurality of vertices and generate system values for the plurality of vertices, wherein the system values are to have a one-to-one relationship with the pixel shader interface.
5. The computing device of claim 1 , wherein the command streamer is to call a plurality of non-GPGPU shader threads to dispatch the 3D workload.
6. The computing device of claim 1 , wherein the 3D-specific fixed function units are not available to compute kernels.
7. A computing architecture comprising:
a workload converter to receive a general purpose graphics processing unit (GPGPU) workload and convert the GPGPU workload to a three-dimensional (3D) workload by:
an identification of a plurality of thread groups in the GPGPU workload; and
an association of the plurality of thread groups to a 3D matrix of cubes;
a command streamer to dispatch the 3D workload; and
a 3D pipeline separate from the workload converter and including 3D-specific fixed function units, the 3D pipeline to receive the 3D workload from the command streamer and process the 3D workload with the 3D-specific fixed function units.
8. The computing architecture of claim 7 , wherein each cube is to include an array of two-dimensional (2D) primitives.
9. The computing architecture of claim 8 , wherein each primitive is to be a quad.
10. The computing architecture of claim 8 , wherein the 3D pipeline includes:
a pixel shader interface; and
a vertex shader to identify, for each of the 2D primitives associated with the 3D workload, a plurality of vertices and generate system values for the plurality of vertices, wherein the system values are to have a one-to-one relationship with the pixel shader interface.
11. The computing architecture of claim 7 , wherein the command streamer is to call a plurality of non-GPGPU shader threads to dispatch the 3D workload.
12. The computing architecture of claim 7 , wherein the 3D-specific fixed function units are not available to compute kernels.
13. A method comprising:
receiving a general purpose graphics processing unit (GPGPU) workload;
converting, with a workload converter, the GPGPU workload to a three-dimensional (3D) workload by:
identifying a plurality of thread groups in the GPGPU workload; and
associating the plurality of thread groups to a 3D matrix of cubes;
dispatching the 3D workload, that is converted from the GPGPU workload, to a 3D pipeline separate from the workload converter; and
processing the 3D workload with 3D-specific fixed function units of the 3D pipeline.
14. The method of claim 13 , wherein each cube includes an array of two-dimensional (2D) primitives.
15. The method of claim 14 , wherein each 2D primitive is a quad.
16. The method of claim 14 , further including:
identifying, for each of the 2D primitives associated with the 3D workload, a plurality of vertices; and
generating system values for the plurality of vertices, wherein the system values have a one-to-one relationship with a pixel shader interface associated with the 3D pipeline.
17. The method of claim 13 , wherein dispatching the 3D workload to the 3D pipeline includes calling a plurality of non-GPGPU shader threads.
18. The method of claim 13 , wherein the 3D-specific fixed function units are not available to compute kernels.
19. At least one non-transitory computer readable storage medium comprising a set of instructions, which when executed by a computing device, cause the computing device to:
receive a general purpose graphics processing unit (GPGPU) workload;
convert, with a workload converter, the GPGPU workload to a three-dimensional (3D) workload to be processed by 3D-specific fixed function units by:
an identification of a plurality of thread groups in the GPGPU workload; and
an association of the plurality of thread groups to a 3D matrix of cubes; and
dispatch the 3D workload, that is converted from the GPGPU workload, to a 3D pipeline, that includes the 3D-specific fixed function units and that is separate from the workload converter, so that the 3D workload is to be processed by the 3D-specific fixed function units.
20. The at least one non-transitory computer readable storage medium of claim 19 , wherein each cube is to include an array of two-dimensional (2D) primitives.
21. The at least one non-transitory computer readable storage medium of claim 20 , wherein each 2D primitive is to be a quad.
22. The at least one non-transitory computer readable storage medium of claim 20 , wherein the instructions, when executed, cause the computing device to:
identify, for each of a plurality of the 2D primitives associated with the 3D workload, a plurality of vertices; and
generate system values for the plurality of vertices, wherein the system values are to have a one-to-one relationship with a pixel shader interface associated with the 3D pipeline.
23. The at least one non-transitory computer readable storage medium of claim 19 , wherein the instructions, when executed, cause the computing device to call a plurality of non-GPGPU shader threads to dispatch the 3D workload.
24. The at least one non-transitory computer readable storage medium of claim 19 , wherein the 3D-specific fixed function units are not available to compute kernels.