Displaying 3 results from an estimated 3 matches for "takenbranchfrequ".
2011 Oct 22
0
[LLVMdev] Question regarding basic-block placement optimization
...t you may want to think about it and
come with your own design first.
Finally, we need some way to validate the design and verify the
implementation. Weighted statistics will be very useful here, similar
to those that Jakob used in the register allocator. For example:
For each nonadjacent edge:
TakenBranchFrequency += frequency(Edge)
This metric will prefer the most "extreme" layout because it assumes
perfect accuracy of block frequencies. It would be nice to look at the
same metric computed from skewed branch probabilities to see how
sensitive it is to small shifts in branch behavior.
Taken b...
2011 Oct 21
3
[LLVMdev] Question regarding basic-block placement optimization
On Thu, Oct 20, 2011 at 10:53 AM, Jakob Stoklund Olesen <stoklund at 2pi.dk>wrote:
>
> On Oct 20, 2011, at 9:56 AM, Chandler Carruth wrote:
>
> > A new patch is attached that is *much* less of a rough draft. Sorry for
> any confusion due to the early state of the patch.
>
> Thanks, Chandler. This is great stuff.
>
> > Still, I never intended this to be
2011 Oct 23
4
[LLVMdev] Question regarding basic-block placement optimization
.... It looks pretty good to me though. =]
> Finally, we need some way to validate the design and verify the
> implementation. Weighted statistics will be very useful here, similar
> to those that Jakob used in the register allocator. For example:
>
> For each nonadjacent edge:
> TakenBranchFrequency += frequency(Edge)
>
> This metric will prefer the most "extreme" layout because it assumes
> perfect accuracy of block frequencies. It would be nice to look at the
> same metric computed from skewed branch probabilities to see how
> sensitive it is to small shifts in b...