Re: [Quantlib-users] Release candidate for version 1.32

[Tom A. <tw...@ur...>]

I can try. It will be easiest if i build from source myself, but if you 
would like to suggest a few commits between 1.31 and 1.32, i can try those 
rather than picking randomly. Sadly, the process is too manual to simply 
do a git bisect.

tom

On Fri, 6 Oct 2023, Luigi Ballabio wrote:

> Hmm.  Thanks for reporting this—would you mind trying to see when the
> performance got worse?  I can send you a few tarballs from different points
> between 1.31.1 and now.
>
> Luigi
>
>
> On Fri, Oct 6, 2023 at 7:09 PM Tom Anderson <tw...@ur...> wrote:
>
>> Hi Luigi,
>>
>> On Wed, 4 Oct 2023, Luigi Ballabio wrote:
>>
>>> Hello everybody,
>>>    I just published sources for a release candidate for 1.32 at <
>>> https://github.com/lballabio/QuantLib/releases/tag/1.32rc>.  If you have
>>> some time, please try them out this week or the next and report any
>>> problems here.  Thanks!
>>
>> I tried it. It builds, my app builds (after a few changes needed due to
>> the introduction of the FixedVsFloatingSwap), and then produces the same
>> results as with 1.31.
>>
>> However, it seems to be much slower in some places. I haven't dug into why
>> yet. But for example, the job which builds an ESTR curve and prices a load
>> of swaps off it went from 92 ms to 328 ms, and the job which does the same
>> for Euribor went from 210 ms to 631 ms. Pricing our swap inventory went
>> from 85 ms to 257 ms, and calculating risk metrics went from 15,270 ms to
>> 50,037 ms. This is all based on a single run with each version, so i have
>> no error bars on these numbers, but the random variance is much smaller
>> than these differences.
>>
>> The inventory and risk jobs do not build curves, but use the curves built
>> by the previous jobs, so this looks like it could be a big slowdown in
>> pricing swaps.
>>
>> I built QuantLib with GCC 7.2.0, and my app with GCC 13.1.0. I am building
>> QuantLib with GCC 7 because when i last tried to built it with GCC 13, it
>> was much slower!
>>
>> Regards,
>> tom
>>
>>
>> --
>> Subvert normality.
>>
>

-- 
Subvert normality.