Bandwidth-aware multi-frequency performance estimation mechanism
Systems, apparatuses, and methods for implementing performance estimation mechanisms are disclosed. In one embodiment, a computing system includes at least one processor and a memory subsystem. During a characterization phase, the system utilizes a memory intensive workload to detect when the memory subsystem reaches its saturation point. Then, the system collects performance counter values during a sampling phase of a target application to determine the memory bandwidth. If the memory bandwidth is greater than the saturation point, then the system generates a prediction of the memory time which is based on a ratio of the memory bandwidth over the saturation point. Otherwise, if the memory bandwidth is less than the saturation point, the system assumes memory time is constant versus processor frequency. Then, the system uses the memory time and an estimate of the compute time to estimate a phase time for the target application at different processor frequencies.
1. A system comprising:
a memory subsystem; and
one or more processors;
wherein the system is configured to:
determine if memory bandwidth of the memory subsystem during a first interval is greater than or equal to a threshold;
generate, based on a ratio of the memory bandwidth divided by the threshold, a first estimate of memory time for the one or more processors responsive to determining the memory bandwidth of the memory subsystem during the first interval is greater than or equal to the threshold; and
use a second estimate of memory time based on a value other than the ratio of the memory bandwidth divided by the threshold, responsive to determining the memory bandwidth during the first interval is not greater than or equal to the threshold.
2. The system as recited in claim 1 , wherein the system is further configured to:
predict a phase time of the one or more processors at a given processor frequency based on the first estimate of memory time; and
operate the one or more processors at the given processor frequency responsive to determining the phase time meets a specified performance target.
3. The system as recited in claim 2 , wherein the memory time is an amount of time the one or more processors spend waiting on memory requests to be processed by the memory subsystem.
4. The system as recited in claim 2 , wherein the memory bandwidth is calculated during the first interval while the one or more processors are operating at a first frequency, wherein the given processor frequency is a second frequency, and wherein the second frequency is different from the first frequency.
5. The system as recited in claim 1 , wherein the system is configured to use a fixed time as the second estimate of memory time responsive to determining the memory bandwidth during the first interval is not greater than or equal to the threshold.
6. The system as recited in claim 2 , wherein the performance target is specified by a user, and wherein the performance target is specified as a percentage of maximum performance.
7. The system as recited in claim 2 , wherein a frequency provided to the memory subsystem is independent of a frequency provided to the one or more processors, and wherein the threshold is set based on a point at which bandwidth saturates for the memory subsystem.
8. A method comprising:
determining if memory bandwidth of a memory subsystem during a first interval is greater than or equal to a threshold;
generating, based on a ratio of the memory bandwidth divided by the threshold, a first estimate of memory time responsive to determining the memory bandwidth during the first interval is greater than or equal to the threshold; and
using a second estimate of memory time based on a value other than the ratio of the memory bandwidth divided by the threshold, responsive to determining the memory bandwidth during the first interval is not greater than or equal to the threshold.
9. The method as recited in claim 8 , further comprising:
predicting a phase time of one or more processors at a given processor frequency based on the first estimate of memory time; and
operating the one or more processors at the given processor frequency responsive to determining the phase time meets a specified performance target.
10. The method as recited in claim 9 , wherein the memory time is an amount of time the one or more processors spend waiting on memory requests to be processed by the memory subsystem.
11. The method as recited in claim 9 , wherein the performance target is specified by a user, and wherein the performance target is specified as a percentage of maximum performance.
12. The method as recited in claim 9 , wherein a frequency provided to a memory subsystem is independent of a frequency provided to the one or more processors, and wherein the threshold is set based on a point at which bandwidth saturates for the memory subsystem.
13. The method as recited in claim 8 , wherein the memory bandwidth is calculated during the first interval while the one or more processors are operating at a first frequency, wherein the given processor frequency is a second frequency, and wherein the second frequency is different from the first frequency.
14. The method as recited in claim 8 , comprising using a fixed time as the second estimate of memory time responsive to determining the memory bandwidth during the first interval is not greater than or equal to the threshold.
15. A processor comprising:
one or more execution units; and
a power management unit;
wherein the processor is configured to:
determine if memory bandwidth of a memory subsystem during a first interval is greater than or equal to a threshold; and
generate, based on a ratio of the memory bandwidth divided by the threshold, a first estimate of memory time responsive to determining the memory bandwidth during the first interval is greater than or equal to the threshold; and
use a second estimate of memory time based on a value other than the ratio of the memory bandwidth divided by the threshold, responsive to determining the memory bandwidth during the first interval is not greater than or equal to the threshold.
16. The processor as recited in claim 15 , wherein the processor is further configured to:
predict a phase time of the processor at a given processor frequency based on the first estimate of memory time; and
operate at the given processor frequency responsive to determining the phase time meets a specified performance target.
17. The processor as recited in claim 16 , wherein the memory time is an amount of time spent waiting on memory requests to be processed by the memory subsystem.
18. The processor as recited in claim 16 , wherein the performance target is specified by a user, and wherein the performance target is specified as a percentage of maximum performance.
19. The processor as recited in claim 15 , wherein the memory bandwidth is calculated during the first interval while the processor is operating at a first frequency, wherein the given processor frequency is a second frequency, and wherein the second frequency is different from the first frequency.
20. The processor as recited in claim 15 , wherein the processor is configured to use a fixed time as the second estimate of memory time responsive to determining the memory bandwidth during the first interval is not greater than or equal to the threshold.