Frequency scaling responding to a performance change method and apparatus and computer-readable storage medium

1. A frequency scaling method, wherein the method comprises:

predicting an energy efficiency parameter for processing a current frame of an image by at least one module, wherein the at least one module comprises at least one of a central processing unit (CPU), a graphics processing unit (GPU), a memory configured to store the current frame, or a neural network processing unit (NPU), and the energy efficiency parameter comprises at least one of the following: a quantity of instructions for processing the current frame by the CPU, a cache miss generated when the CPU processes the current frame, a quantity of draw calls for processing the current frame by the GPU, a bandwidth for storing the current frame by the memory, or a calculation amount for processing the current frame by the NPU;

selecting a first frequency set from a plurality of frequency sets based on the predicted energy efficiency parameter, wherein the first frequency set comprises a preset frequency corresponding to each of the at least one module; and

scaling a working frequency of each of the at least one module for processing the current frame to the preset frequency corresponding to each of the at least one module.

2. The method according to claim 1 , wherein the first frequency set meets an energy efficiency requirement, and the energy efficiency requirement comprises at least one of the following: a power consumption requirement or a performance requirement.

3. The method according to claim 2 , wherein the power consumption requirement is a minimum power consumption requirement, and the performance requirement is that a performance characteristic meets a preset threshold.

4. The method according to claim 2 , wherein the selecting a first frequency set from a plurality of frequency sets based on the predicted energy efficiency parameter comprises:

predicting, based on the predicted energy efficiency parameter, a performance characteristic corresponding to each frequency set; and

selecting, from the plurality of frequency sets, the first frequency set corresponding to the performance characteristic meeting a preset threshold.

5. The method according to claim 4 , wherein the selecting, from the plurality of frequency sets, the first frequency set corresponding to the performance characteristic meeting the preset threshold comprises:

predicting, based on the predicted energy efficiency parameter, power consumption corresponding to each frequency set;

selecting, from the plurality of frequency sets, a plurality of second frequency sets corresponding to a plurality of performance characteristics meeting the preset threshold; and

selecting, from the plurality of second frequency sets, the first frequency set corresponding to lowest power consumption.

6. The method according to claim 1 , wherein the predicting an energy efficiency parameter for processing a current frame of an image by at least one module comprises:

predicting, based on a historical energy efficiency parameter, the energy efficiency parameter for processing the current frame of the image by the at least one module, wherein the historical energy efficiency parameter is obtained based on an energy efficiency parameter of at least one frame before the current frame.

7. The method according to claim 6 , wherein the predicting, based on a historical energy efficiency parameter, the energy efficiency parameter for processing the current frame of the image by the at least one module comprises:

searching a load prediction table based on the historical energy efficiency parameter, to predict the energy efficiency parameter for processing the current frame of the image by the at least one module.

8. A frequency scaling apparatus, comprising:

a non-transitory memory storage comprising instructions; and

one or more hardware processors in communication with the non-transitory memory storage, wherein the one or more hardware processors execute the instructions to:

predict an energy efficiency parameter for processing a current frame of an image by at least one module, wherein the at least one module comprises at least one of a central processing unit (CPU), a graphics processing unit (GPU), a memory configured to store the current frame, or a neural network processing unit (NPU), and the energy efficiency parameter comprises at least one of the following: a quantity of instructions for processing the current frame by the CPU, a cache miss generated when the CPU processes the current frame, a quantity of draw calls for processing the current frame by the GPU, a bandwidth for storing the current frame by the memory, or a calculation amount for processing the current frame by the NPU;

select a first frequency set from a plurality of frequency sets based on the predicted energy efficiency parameter, wherein the first frequency set comprises a preset frequency corresponding to each of the at least one module; and

scale a working frequency of each of the at least one module for processing the current frame to the preset frequency corresponding to each of the at least one module.

9. The apparatus according to claim 8 , wherein the first frequency set meets an energy efficiency requirement, and the energy efficiency requirement comprises at least one of the following: a power consumption requirement or a performance requirement.

10. The apparatus according to claim 9 , wherein the power consumption requirement is a minimum power consumption requirement, and the performance requirement is that a performance characteristic meets a preset threshold.

11. The apparatus according to claim 9 , wherein the one or more hardware processors further execute the instructions to:

predict, based on the predicted energy efficiency parameter, a performance characteristic corresponding to each frequency set; and

select, from the plurality of frequency sets, the first frequency set corresponding to the performance characteristic meeting a preset threshold.

12. The apparatus according to claim 11 , wherein the one or more hardware processors further execute the instructions to:

predict, based on the predicted energy efficiency parameter, power consumption corresponding to each frequency set;

select, from the plurality of frequency sets, a plurality of second frequency sets corresponding to a plurality of performance characteristics meeting the preset threshold; and

select, from the plurality of second frequency sets, the first frequency set corresponding to lowest power consumption.

13. The apparatus according to claim 8 , wherein the one or more hardware processors further execute the instructions to:

predict, based on a historical energy efficiency parameter, the energy efficiency parameter for processing the current frame of the image by the at least one module, wherein the historical energy efficiency parameter is obtained based on an energy efficiency parameter of at least one frame before the current frame.

14. The apparatus according to claim 13 , wherein the one or more hardware processors further execute the instructions to:

search a load prediction table based on the historical energy efficiency parameter, to predict the energy efficiency parameter for processing the current frame of the image by the at least one module.

15. A non-transitory computer-readable storage medium, comprising a computer program, wherein when running on a computer or a processor, the computer program enables the computer or the processor to execute a frequency scaling operation that comprises:

predicting an energy efficiency parameter for processing a current frame of an image by at least one module, wherein the at least one module comprises at least one of a central processing unit (CPU), a graphics processing unit (GPU), a memory configured to store the current frame, or a neural network processing unit (NPU), and the energy efficiency parameter comprises at least one of the following: a quantity of instructions for processing the current frame by the CPU, a cache miss generated when the CPU processes the current frame, a quantity of draw calls for processing the current frame by the GPU, a bandwidth for storing the current frame by the memory, or a calculation amount for processing the current frame by the NPU;

selecting a first frequency set from a plurality of frequency sets based on the predicted energy efficiency parameter, wherein the first frequency set comprises a preset frequency corresponding to each of the at least one module; and

scaling a working frequency of each of the at least one module for processing the current frame to the preset frequency corresponding to each of the at least one module.

16. The computer-readable storage medium according to claim 15 , wherein the first frequency set meets an energy efficiency requirement, and the energy efficiency requirement comprises at least one of the following: a power consumption requirement or a performance requirement.

17. The computer-readable storage medium according to claim 15 , wherein the predicting an energy efficiency parameter for processing a current frame of an image by at least one module comprises:

predicting, based on a historical energy efficiency parameter, the energy efficiency parameter for processing the current frame of the image by the at least one module, wherein the historical energy efficiency parameter is obtained based on an energy efficiency parameter of at least one frame before the current frame.