System and methods for efficient execution of a collaborative task in a shader system
Methods and systems are disclosed for executing a collaborative task in a shader system. Techniques disclosed include receiving, by the system, input data and computing instructions associated with the collaborative task, as well as a configuration setting, causing the system to operate in a takeover mode. The system then launches, exclusively in one workgroup processor, a workgroup including wavefronts configured to execute the collaborative task.
1. A method for executing a collaborative task in a system, comprising:
receiving, by the system, input data and computing instructions associated with the collaborative task;
receiving, by the system, a configuration setting, causing the system to operate in a takeover mode; and
launching, by the system, exclusively in one workgroup processor (WGP), a workgroup including wavefronts configured to execute the collaborative task.
2. The method of claim 1 , further comprising:
executing, by the system, the wavefronts in a first stage of operation, wherein each wavefront employs the computing instructions on a respective portion of the input data, stored in vector general purpose registers (VGPRs) associated with the wavefront.
3. The method of claim 2 , further comprising:
when the executing of the wavefronts is completed, receiving, by the system, further computing instructions associated with the collaborative task;
launching, by the system, exclusively in the one WGP, the workgroup including the wavefronts; and
executing, by the system, the wavefronts in a second stage of operation, wherein each wavefront employs the further computing instructions on a respective portion of data, including data generated in the first stage of operation by other wavefronts in the workgroup that are stored in VGPRs associated with the other wavefronts.
4. The method of claim 1 , further comprising:
providing a wavefront of the workgroup access to memory resources of the WGP that are associated with another wavefront of the workgroup.
5. The method of claim 4 , wherein the memory resources of the WGP are vector general purpose registers (VGPRs).
6. The method of claim 4 , wherein:
the memory resources of the WGP are local data share (LDS) memory, and
access to memory space of the LDS memory is symmetric for all the wavefronts of the workgroup.
7. The method of claim 1 , wherein the WGP comprises:
one or more single instruction multiple data (SIMD) units, each unit is used to execute a subset of the wavefronts, and each unit including an arithmetic logic unit (ALU) and vector general purpose registers (VGPRs) associated with the subset of the wavefronts.
8. The method of claim 1 , further comprising:
activating a cache management policy for the WGP, the cache management policy is related to a workload pattern of the collaborative task.
9. A system for executing a collaborative task, comprising:
at least one processor; and
a memory storing instructions that, when executed by the at least one processor, cause the system to:
receive input data and computing instructions associated with the collaborative task,
receive a configuration setting, causing the system to operate in a takeover mode, and
launch, exclusively in one workgroup processor (WGP), a workgroup including wavefronts configured to execute the collaborative task.
10. The system of claim 9 , wherein the instructions further cause the system to:
execute the wavefronts in a first stage of operation, wherein each wavefront employs the computing instructions on a respective portion of the input data, stored in vector general purpose registers (VGPRs) associated with the wavefront.
11. The system of claim 10 , wherein the instructions further cause the system to:
when the executing of the wavefronts is completed, receive further computing instructions associated with the collaborative task;
launch, exclusively in the one WGP, the workgroup including the wavefronts; and
execute the wavefronts in a second stage of operation, wherein each wavefront employs the further computing instructions on a respective portion of data, including data generated in the first stage of operation by other wavefronts in the workgroup that are stored in VGPRs associated with the other wavefronts.
12. The system of claim 9 , wherein the instructions further cause the system to:
provide a wavefront of the workgroup access to memory resources of the WGP that are associated with another wavefront of the workgroup.
13. The system of claim 12 , wherein the memory resources of the WGP are vector general purpose registers (VGPRs).
14. The system of claim 12 , wherein:
the memory resources of the WGP are local data share (LDS) memory, and
access to memory space of the LDS memory is symmetric for all the wavefronts of the workgroup.
15. The system of claim 9 , wherein the WGP comprises:
one or more single instruction multiple data (SIMD) units, each unit is used to execute a subset of the wavefronts, and each unit including an arithmetic logic unit (ALU) and vector general purpose registers (VGPRs) associated with the subset of the wavefronts.
16. The system of claim 9 , wherein the instructions further cause the system to:
activate a cache management policy for the WGP, the cache management policy is related to a workload pattern of the collaborative task.
17. A non-transitory computer-readable medium comprising instructions executable by at least one processor to perform a method for executing a collaborative task in a system, the method comprising:
receiving, by the system, input data and computing instructions associated with the collaborative task;
receiving, by the system, a configuration setting, causing the system to operate in a takeover mode; and
launching, by the system, exclusively in one workgroup processor (WGP), a workgroup including wavefronts configured to execute the collaborative task.
18. The non-transitory computer-readable medium of claim 17 , further comprising:
executing, by the system, the wavefronts in a first stage of operation, wherein each wavefront employs the computing instructions on a respective portion of the input data, stored in vector general purpose registers (VGPRs) associated with the wavefront.
19. The non-transitory computer-readable medium of claim 18 , further comprising:
when the executing of the wavefronts is completed, receiving, by the system, further computing instructions associated with the collaborative task;
launching, by the system, exclusively in the one WGP, the workgroup including the wavefronts; and
executing, by the system, the wavefronts in a second stage of operation, wherein each wavefront employs the further computing instructions on a respective portion of data, including data generated in the first stage of operation by other wavefronts in the workgroup that are stored in VGPRs associated with the other wavefronts.
20. The non-transitory computer-readable medium of claim 17 , further comprising:
providing a wavefront of the workgroup access to memory resources of the WGP that are associated with another wavefront of the workgroup.