Allocaholic - the 250 millions malloc killer - Githubissues

cbg-ethz / shorah

Repo for the software suite ShoRAH (Short Reads Assembly into Haplotypes)

GNU General Public License v3.0

39 stars 14 forks source link

Allocaholic - the 250 millions malloc killer #55

Closed DrYak closed 4 years ago

DrYak commented 5 years ago

A series of fix that speeds-up shorah :

by avoiding labyrinthine stacks of function calls for log, exp and pow
by avoiding way too many calls to malloc, mostly due to building look-up tables for the discrete distributions generator

(Even more useless in cases where the generator is called once. The dual look-up table that enable O(1) with two simple steps is over kill).

DrYak commented 4 years ago

xyplot comparing commits while processing data from 5 Virus Mix:

Blue: xyplot between two first commits of this branch (both using GSL for the random distribution engine)
Orange: xyplot against the tip of this branch (using Boost C++ random number generator, and table-less simple naive distribution)

xyplot

DrYak commented 4 years ago

Merging in.

Big thanks to the reviewers, David and now Monica. Thanks to Susie for the test visualisation suggestions.

SoapZA commented 4 years ago

hold off, I havent reviewed a second time

DrYak commented 4 years ago

@SoapZA okay. (BTW: was writting you an e-mail asking how your doing...)

DrYak commented 4 years ago

rebased, force-pushed. granpa's inlines stay in ;-)

DrYak commented 4 years ago

@SoapZA, do you have a preference:

shall we TAG it as ShoRAH 1.99.0 (and release the meson-ified as 2.0 final)
shall we TAG it as ShoRAH 2.0 (and release the meson-ified as 2.0.1)

SoapZA commented 4 years ago

@DrYak I have asked you to please remove them, given that you added new code. Given how we're already using C++11 features, trying to force GCC 2.95 to inline these functions is a pointless endeavor.

You can tag it as 1.99.0.