Re: Potential FastCode speed increase?

Hi Avatar

> I'd prefer something like this:
>
> BHT Jcc // Branch Hit: Taken
> BHNT Jcc // Branch Hit: Not Taken

I agree

> So would i. I will add the request to QC and, when i have time, i could

Nice.

> also try the effect these have by modifying some existing FastCode
> functions in the B&V utilities. Which function do you think could
> benifit most? (that is, has lots of predictable branches, especially
> those that are statically mispredicted)

Pehaps Min/Max for Integer.

Perhaps you need to modify the benchmark. It is balanced such that the two
cases A>B and B<A appears approximately the same number of times. If you
tweak the balance such that the static prediction will be wrong most of the
time, then you will perhaps be able to see the prefixes work, but the same
effect could be to obtained in this simple eaxample by rearranging the code
to improve the static prediction. The prefixes will only be usefull in
situations where it is impossible to rearrange code to fit the static branch
predictor.

Regards
Dennis


.