> On June 14, 2012, 7:21 a.m., Ali Saidi wrote:
> > looks fine to me, but it only seems to impact performance slightly. Did you 
> > see bigger gains than a couple %?
> 
> Nathanael Premillieu wrote:
>     I have re-run the experiment with the arm_detailed model. There is 
> nearlly no gain (there is even loses for some runs). Now, I wonder why there 
> is so much RAs mispredictions. Is it relalted to the ARM ISA? Because, with 
> the alpha ISA (in an other simulator), I have a lot less RAS mispredictions.

I found a bug tagging a pop operation correctly. I couldn't  find it used in 
some spec2k binaries that I looked at, but apparently some version of compiles 
do emit them a lot. I'll be pushing it soon.


- Ali


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://reviews.gem5.org/r/1268/#review2947
-----------------------------------------------------------


On June 11, 2012, 7:49 a.m., Nathanael Premillieu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://reviews.gem5.org/r/1268/
> -----------------------------------------------------------
> 
> (Updated June 11, 2012, 7:49 a.m.)
> 
> 
> Review request for Default.
> 
> 
> Description
> -------
> 
> I have added a new flag (named pushedRAS) in the PredictorHistory structure. 
> This flag tracks whether the RAS has been pushed or not during a prediction. 
> Then, in the squash function it is used to pop the RAS if necessary.
> 
> 
> Diffs
> -----
> 
>   src/cpu/o3/bpred_unit.hh d0d9d10b3930 
>   src/cpu/o3/bpred_unit_impl.hh d0d9d10b3930 
> 
> Diff: http://reviews.gem5.org/r/1268/diff/
> 
> 
> Testing
> -------
> 
> It has been tested on several of the Spec2006 benchmarks compiled for ARM. 
> For most of the cases, there are less RAS mispredictions.
> 
> 
> Thanks,
> 
> Nathanael Premillieu
> 
>

_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev

Reply via email to