Thanks for the solution AndreasI will try that.
On Apr 15, 2015 6:29 PM, Andreas Hansson andreas.hans...@arm.com wrote:
Hi all,
The nice and clean option would be to use add an architecturally visible
performance counter read from the DVFS governor (in software).
The hackish option
Hi Andreas,
The nice and clean option would be to use add an architecturally visible
performance counter read from the DVFS governor (in software).
Is there any example of any such counter which I can look at to get an idea
how this will be done??
On Apr 16, 2015 10:43 AM, Nimish Girdhar
Hello,
I have been following the building kernel turotial in
http://www.gem5.org/BBench-gem5. I intend to apply m5struct.diff patch
to it so I can integrate gem5 with streamline.
However, even without patching, the freshly compiled kernel using make
ARCH=arm CROSS_COMPILE=arm-none-linux-gnueabi-
This controls timer frequency in Linux ?
I've played with generic timer frequency in freebsd decreasing
the frequency from 100MHz (default) to 10MHz - it gives some speed up
but not too much.
Ruslan
On Tue, Apr 14, 2015 at 11:00:50PM -0700, Ali Saidi wrote:
For linux we pass in a jiffies= on
Not just about controlling the frequency, but *avoiding* the costly
calibration loop involved. Some other tests that come to mind are
timeouts waiting for devices (prune the compiled modules / support for
devices) and other calibration loops (RAID throughput, encryption
throughput,...)
--
Hi Nimish,
as Andreas pointed out, there are two possible ways for this.
If you want to simulate a smart *software* governor, then you can access
the PMU (see code in src/arch/arm/{pmu.cc,pmu.hh,ArmPMU.py}). The
interface to that should be through the normal PMU functions, and the
counters
Thanks, Andreas. This makes sense.
On Wed, Apr 15, 2015 at 5:26 PM, Andreas Hansson andreas.hans...@arm.com
wrote:
Hi Patrick,
When it comes to the stores you are looking at a rather small number of
operations, and my guess is that they are still in the DRAM write queues.
These queues are
Hi Lokesh,
gem5 does indeed currently only capture the delay in the transition, and
still “clocks” the components that are affected. A more detailed model of
the PMIC and voltage regulators, battery etc can definitely be added, the
transitions refined etc. The question is how much it matters, and
Hi
Is Android supported only with RealView_PBX machine type. The discussion at
https://www.mail-archive.com/gem5-users@gem5.org/msg11185.html suggests so.
If yes then how to use 2GB configuration for Android. Any help will he
highly appreciated.
--
Have a great day!
Thanks and Warm Regards
Hello Andreas
Thanks for prompt reply. Can you please let me know what you mean by You
should only ever have to do this in the L1 (or the LSQ) ?
My implementation is for L2 cache line i.e. when cache line request comes
from L1 cache corresponding to Master Id of 12.
I know the implementation is
Hello everyone
In DVFS in real systems, there are 2 parameters that come into picture
during voltage-frequency transition.
1. Core dead time during which core does not execute anything and is
the time taken to actually change the frequency.
2. voltage change time during which the core is
11 matches
Mail list logo