[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [Request for help] Bug: #976479 scrappie: FTBFS: scrappie_matrix.h:5:14: fatal error: immintrin.h: No such file or directory

Hi o/

First of all, thank you for the fixes and letting me know about cowbuilder-dist :-)

On Mon, 7 Dec 2020 at 17:51, Michael Crusoe <michael.crusoe@gmail.com> wrote:

I'd be really grateful if you could take a look at my changes and let me know if they look fine.

I'll leave comments on Salsa

Thanks a lot, cherry-picked + pulled the changes mentioned.

Have you been sending PRs and otherwise forwarding your patches upstream?

Yeah, I almost always do the needed forwards. I did 2 simde fixes before this one - for plast and ngmlr.
I forwarded the patch for plast upstream[1] and I'll forward for ngmlr in some more days - since this seems to need more changes in the upstream codebase (I don't exactly remember what exactly, but this is in my TODO)

[1]: https://github.com/PLAST-software/plast-library/pull/8

Also, another question: This package has a MPL-2.0 license and AFAIK, MPL is a restrictive Free software license, so does this qualify for adding in a "Built-Using" field?

"if the source package requires the full source code be available"

Does the MPL have this requirement?

I skimmed through /usr/share/common-licenses/MPL-2.0 but admittedly, it is still not very clear to me. From what I read, it looks like the answer is yes for this question, but I'm not completely confident about it.
Do you think we should ask debian-legal? - This seems like the last blocker for an upload now

On Mon, 7 Dec 2020 at 18:06, Michael Crusoe <michael.crusoe@gmail.com> wrote:
Don't forget step 8 of https://wiki.debian.org/SIMDEverywhere#Approach ; I've updated the sample scripts for multi-compiling x86 at different SIMD levels
Done and pushed. it took me quite a while to figure this out. Two questions though:
1) We are fundamentally doing this because we want to build this for all possible compilation flags right? - and in the dispatch script we are trying to figure out the most efficient implementation and use the binary for the same?
2) Could you please review the changes once? Works fine everywhere IMO

BTW I noticed that the rules file which was initially just around 10 lines increased by approximately an order of magnitude because I had to almost handcraft the makefile to do various stuff, and I observed a similar phenomenon for kalign and mmseqs2 where you made similar changes.
My intent here is to say - I (vaguely)wonder can we make a tool/helper that can _automate_ the entire process while using simde to reduce handcrafted stuff?
Because fundamentally we seem to be making similar changes everywhere.
And ofcourse, apologies if this doesn't sound good.

I'll definitely update this in the next couple of days.

Kind Regards,

Reply to: