These are chat archives for bluescarni/piranha

28th
Jan 2016
Isuru Fernando
@isuruf
Jan 28 2016 01:11
For series_sin instead of dividing by i!, I tried multiplying by prod(j!)/i! and then dividing the whole result by prod(j!)
Timings are 5570ms vs 1589ms for piranha vs flint
Francesco Biscani
@bluescarni
Jan 28 2016 01:22
is this with the same N?
Isuru Fernando
@isuruf
Jan 28 2016 01:22
yes
1000
Francesco Biscani
@bluescarni
Jan 28 2016 01:23
nice :)
and this is with the PR and nothing else? or also @rwst's patch?
Isuru Fernando
@isuruf
Jan 28 2016 01:23
This is only useful when the input to sin has all rationals with den=1
with @rwst 's patch
Francesco Biscani
@bluescarni
Jan 28 2016 01:24
ok, I am curious to see what happens in conjunction with the other modifications
Isuru Fernando
@isuruf
Jan 28 2016 01:25
For something like sin((x+x^2)/123)*cos((x+x^2)/123) timings are 245,203 ms vs 3593ms
Francesco Biscani
@bluescarni
Jan 28 2016 01:26
good to know, I hope the situation for this case improves when the pre-normalisation of the fractions is disabled
need to go get some sleep, I'll be back in a few hours
I'll merge the PR tomorrow
Isuru Fernando
@isuruf
Jan 28 2016 01:29
Ok. good night
Francesco Biscani
@bluescarni
Jan 28 2016 01:29
thanks!
Ralf Stephan
@rwst
Jan 28 2016 07:39
My patch has grown and also is a PR: symengine/symengine#784
Francesco Biscani
@bluescarni
Jan 28 2016 11:24
@isuruf I merged the PR here https://github.com/bluescarni/piranha/tree/small_series I will be pushing and testing other small series optimisations here as well
@rwst thanks, I will be testing against it
Francesco Biscani
@bluescarni
Jan 28 2016 13:12
@isuruf I changed slightly the logic of your original commit, extended to sub and operations with integers, and added testing bits
bluescarni/piranha@e7008c7