Re: [PATCH] cpufreq: intel_pstate: Change the calculation of next pstate
On 04/29/2014 02:52 PM, Rafael J. Wysocki wrote: On Tuesday, April 29, 2014 07:34:46 PM Stratos Karafotis wrote: On 29/04/2014 07:58 πμ, Viresh Kumar wrote: Cc'd Dirk, On 28 April 2014 03:42, Stratos Karafotis wrote: Currently the driver calculates the next pstate proportional to core_busy factor and reverse proportional to current pstate. Change the above method and calculate the next pstate independently of current pstate. We must mention why the change is required. Hi Viresh, Actually, I can't say that it's required. :) I just believe that calculation of next p-state should be independent from current one. In my opinion we can't scale the load across different p-states, because it's not always equivalent. For example suppose a load of 100% because of a tight for loop in the current p-state. It will be also a 100% load in any other p-state. It will be wrong if we scale the load in the calculation formula according to the current p-state. I included the test results in the change log to point out an improvement because of this patch. I will enrich more the change log as you suggested. Please do so. Also, we need to take your patch to our power lab and see if we can reproduce your results in other workloads. And I'm waiting for the intel_pstate developer Dirk Brandewie to comment. Sorry I just returned from dealing with a family emergency and am digging out of my inbox. I will run this patch through some tests. Thanks! -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] cpufreq: intel_pstate: Change the calculation of next pstate
On Tuesday, April 29, 2014 07:34:46 PM Stratos Karafotis wrote: > On 29/04/2014 07:58 πμ, Viresh Kumar wrote: > > Cc'd Dirk, > > > > On 28 April 2014 03:42, Stratos Karafotis wrote: > >> Currently the driver calculates the next pstate proportional to > >> core_busy factor and reverse proportional to current pstate. > >> > >> Change the above method and calculate the next pstate independently > >> of current pstate. > > > > We must mention why the change is required. > > > > Hi Viresh, > > Actually, I can't say that it's required. :) > I just believe that calculation of next p-state should be independent > from current one. In my opinion we can't scale the load across different > p-states, because it's not always equivalent. > > For example suppose a load of 100% because of a tight for loop in the > current p-state. It will be also a 100% load in any other p-state. > It will be wrong if we scale the load in the calculation formula > according to the current p-state. > > I included the test results in the change log to point out an improvement > because of this patch. > > I will enrich more the change log as you suggested. Please do so. Also, we need to take your patch to our power lab and see if we can reproduce your results in other workloads. And I'm waiting for the intel_pstate developer Dirk Brandewie to comment. Thanks! -- I speak only for myself. Rafael J. Wysocki, Intel Open Source Technology Center. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] cpufreq: intel_pstate: Change the calculation of next pstate
On 29/04/2014 07:58 πμ, Viresh Kumar wrote: > Cc'd Dirk, > > On 28 April 2014 03:42, Stratos Karafotis wrote: >> Currently the driver calculates the next pstate proportional to >> core_busy factor and reverse proportional to current pstate. >> >> Change the above method and calculate the next pstate independently >> of current pstate. > > We must mention why the change is required. > Hi Viresh, Actually, I can't say that it's required. :) I just believe that calculation of next p-state should be independent from current one. In my opinion we can't scale the load across different p-states, because it's not always equivalent. For example suppose a load of 100% because of a tight for loop in the current p-state. It will be also a 100% load in any other p-state. It will be wrong if we scale the load in the calculation formula according to the current p-state. I included the test results in the change log to point out an improvement because of this patch. I will enrich more the change log as you suggested. Thanks, Stratos Karafotis -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] cpufreq: intel_pstate: Change the calculation of next pstate
Cc'd Dirk, On 28 April 2014 03:42, Stratos Karafotis wrote: > Currently the driver calculates the next pstate proportional to > core_busy factor and reverse proportional to current pstate. > > Change the above method and calculate the next pstate independently > of current pstate. We must mention why the change is required. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
