IRC channel logs

<damo22>youpi: what is the fastest algorithm for GPU accelerated DNA sequence alignment? is it some kind of Smith Waterman? i am looking for something fun to implement that i can use at work

<youpi>damo22: I have no idea

<youpi>I just schedule tasks, my research is not about what's inside the tasks :)

<damo22>ok

<jpoiret>damo22: janneke is running Guix/Hurd on bare metal iirc

<damo22>fair enough

<damo22>youpi: do you have ideas for a new scheduler in mach?

<youpi>no

<youpi>I don't think it is the most urgent task to achieve

<damo22>:(

<youpi>possibly it needs some fixing

<youpi>but that's probably just side corners

<youpi>as in: missing smp plumbing

<damo22>ok

<youpi>as a reminder: mach was used in research project on smp support

<janneke>jpoiret: that's right, patiently yet eagerly anticipating some of the new goodies such as rumpnet, >4GB on i386, and whatnotall:)

<youpi>so it's probably just a mere problem of porting the right things on the right x86 bits

<damo22>i dont know what its missing, but i think ext2fs has a strange threading issue

<damo22>causing early hang

<janneke>meanwhile, iwbn if us over at #guix got pull and system reconfigure to actually work :)

<damo22>cool

<azert>damo22 : for dna alignment there is the new minimap2 by the absolute genius Heng Li.

<azert>It’s good for long sequences that are becoming particularly relevant nowadays

<azert>It’s fast on cpu but I’ve seen gpu implementations browsing Google so i wonder if it is faster on gpu

<damo22>azert: why do you keep quitting just when i am replying

<damo22>my old NV Quadro doesnt work on latest cuda, i need 10.2 d'oh

<azert>It’s the problem of phone-chatting but I read the logs

<damo22>azert: i know of minimap2, but i want to implement something simpler that loads genome into gpu ram and runs a full smith waterman alignment algorithm

<damo22>dna is base 4, you can fit 16 dna bases in a uint32_t

<azert>That should be possible

<damo22>yes

<damo22>and fast

<azert>I wonder how you deal with N or other missing/ambiguous, probably trimming or discarding?

<damo22>if a sequence contains an N, you can pretend its A

<damo22>it will not affect the alignment that much

<azert>I’d consider the encoding where you reserve one bit per nucleotide

<azert>So N would be F in exacimal

<damo22>hmm too expensive

<azert>Then you could use the AND bitwise operation instead of equality

<damo22>you only need 2 bits per nucleotide, since there are 2^2 states

<gnucode>hello hurd people!

<gnucode>also apparently "sudo apt install elpa-git-commit" is causing an apt error.

<gnucode>git-commit.el: 126:1: Error : Cannot open load file : No such file or directory, transient

IRC channel logs

2023-09-01.log