Systems and methods for power optimization using throughput feedback

• US 9,436,263 B2
• Filed: 02/21/2014
• Issued: 09/06/2016
• Est. Priority Date: 02/21/2014
• Status: Active Grant

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: Venumuddala, Vinay Reddy, Duvvuru, Nagarjuna, Srivastava, Gur Prasad
• Primary Examiner(s): Abbaszadeh, Jaweed A
• Assistant Examiner(s): Pandey, Keshab
• Application Number:
  US14/186,745
• Publication Number:
  US 20150241942A1
• Time in Patent Office: 928 Days
• Field of Search: 713/300, 713/310, 713/320, 713/321, 713/322, 713/323, 713/324, 713/330, 713/340, 713/375
• US Class Current: 1/1
• CPC Class Codes: G06F 1/32 Means for saving power
• CPC Class Codes: G06F 1/3206 Monitoring of events, devices or parameters that trigger a change in power modality
• CPC Class Codes: G06F 1/3234 Power saving characterised by the action undertaken
• CPC Class Codes: G06F 1/324 by lowering clock frequency
• CPC Class Codes: G06F 1/3296 by lowering the supply or operating voltage
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A voltage and frequency scaling system for a processor is provided that may be implemented in dedicated logic or in software. The various voltage and frequency settings for the processor comprise a set of performance settings. The system includes a profiler module that maps each performance setting to a workload range for the processor. The profiler module also maps each workload range to a profiled throughput for the processor. Using a predicated average throughput from the mapping, the voltage and frequency scaling system advantageously selects from the performance settings and commands the processor to operate according to the selected performance setting.

Claims

1. A voltage and frequency system for an integrated circuit;

comprising;

a processor operable according to a selection from a set of power supply voltage and clock frequency performance settings;

a throughput tracker logic circuit configured to calculate a current average throughput in a current frame in a series of frames from a current throughput for the processor and from at least one previous average throughput for a preceding frame;

a profiler logic circuit configured to determine a mapping relating each performance setting to a corresponding workload range for the processor and relating each performance setting to a corresponding profiled throughput for the processor, wherein each workload range relates to a range of processor-executing-instruction clock cycles over each frame; and

an optimizer logic circuit configured to retrieve a first profiled throughput from the mapping responsive to a determination that the current average throughput is less than a desired average throughput, wherein the first profiled throughput corresponds to an increased performance setting, and to determine a first predicted throughput for a subsequent frame from the first profiled throughput and to command the processor to operate according to the increased performance setting in the subsequent frame responsive to a determination that the first predicted throughput is greater than the desired average throughput.

2. The voltage and frequency system of claim 1, further comprising:

a workload translator logic circuit configured to receive a workload signal from the processor during each of the frames and to calculate a current workload for the processor in each frame from the received workload signal; and

a throughput translator logic circuit configured to receive a throughput signal from the processor during each of the frames and to calculate the current throughput for the processor in each frame from the received throughput signal.

3. The voltage and frequency system of claim 2, further comprising:

a workload tracker logic circuit configured to calculate an average workload in the current frame from the current workload for the current frame and from at least one average workload for a preceding frame, wherein the optimizer logic circuit is further configured to determine a predicted workload for the subsequent frame based upon the average workload for the current frame and the current workload, and wherein the optimizer logic circuit is further configured to determine the workload range corresponding to the predicted workload and to use the mapping to determine a second predicted average throughput based upon the profiled throughput corresponding to the determined workload range and to command the processor to operate according to a decreased performance setting in the subsequent frame responsive to a determination that both the first predicated throughput and the second predicted throughput are greater than the desired average throughput.

4. The voltage and frequency system of claim 1, wherein the increased performance setting is a subsequent performance setting to the selection.

5. The voltage and frequency system of claim 1, wherein the increased performance setting is the maximum performance setting.

6. The voltage and frequency system of claim 1, further comprising an interface to the processor, wherein the interface is configured to store the desired average throughput as received from the processor.

7. A method, comprising:

with regard to a current frame in a series of frames, operating a processor in the current frame according to a current performance setting selected from a set of performance settings, wherein each performance setting comprises a corresponding power supply voltage and clock frequency setting for the processor;

determining a current average throughput for the processor responsive to the processor'"'"'s operation during the current frame;

storing a lookup table that relates each performance setting to a workload range for the processor and each performance setting to a profiled throughput for the processor, wherein each workload range relates to a range of processor-executing-instruction clock cycles over each frame;

responsive to a determination that the average throughput for the current frame is less than a desired average throughput, retrieving a first profiled throughput from the lookup table corresponding to an increased performance setting and determining a first predicted throughput for a subsequent frame from the first profiled throughput; and

commanding the processor to operate according to the increased performance setting in a subsequent frame responsive to a determination that the first predicted throughput is greater than the desired average throughput.

8. The method of claim 7, wherein the performance settings are arranged sequentially from a lowest performance setting to a highest performance setting, and wherein the increased performance setting corresponds to the subsequent performance setting to the current performance setting in the sequence.

9. The method of claim 7, further comprising:

commanding the processor to operate in the subsequent frame according to a maximum one of the performance setting responsive to the first predicted throughput being less than the desired average throughput.

10. The method of claim 7, further comprising:

determining a predicted workload for the subsequent frame;

determining which workload range includes the predicted workload; and

responsive to a determination that the average throughput for the current frame is greater than the desired average throughput, retrieving a second profiled throughput from the lookup table corresponding to the determined workload range and determining a second predicted throughput for the subsequent frame based upon the second profiled throughput.

11. The method of claim 10, further comprising:

commanding the processor to operate in the subsequent frame according to the current performance setting responsive to a determination that the second predicted throughput is less than the desired average throughput.

12. The method of claim 10, wherein the performance settings are arranged sequentially from a lowest performance setting to a highest performance setting, and wherein a decreased performance setting corresponds to a preceding performance setting to the current performance setting in the sequence, the method further comprising:

commanding the processor to operate in the subsequent frame according to the decreased performance setting responsive to a determination that the second predicated throughput is greater than the desired average throughput.