Dear Aaron, thanks for your report and for caring about the non-x86 archs!
> Builds for libssw failed on the vast majority architectures due to lack > of support for x86 SSE instructions. AFAICT, libssw specifically makes a > point of using these instructions, so support for non-x86 architectures > is presumably a lost cause. Indeed, since regular SW implementations are pretty trivial; this library explicitly uses SSE. > However, you might still be able to save > *i386 by explicitly building with -msse or -msse2. I normally recommend > *against* using these flags, but the alternative is losing support for > these architectures altogether. I would be curious to hear why setting these flags on i386 might salvage the build. I just tried them on a Jessie i386 Vagrant box and the build indeed succeeded for -msse2, but the resulting binaries obviously(?) segfaulted when trying to run them. When you suggested these flags, did you expect me to do anything else than that — I can’t see why code containing SSE instructions would run on i386? > (All the same, if you do go this route, > please declare a dependency on sse-support [i386] once that package is > available, as discussed in #823672.) Sure, will do that. > Please update the Architecture: field in debian/control to reflect > whatever you wind up doing, remembering to account for non-Linux kernels: > > Architecture: any-amd64 any-i386 any-x32 > > or > > Architecture: any-amd64 any-x32 AFAICS at the moment, I would have to go with the latter. For my purposes (libssw as a dependency for spades) this wouldn’t make much of a difference, as the current SPAdes also only runs on SSE enabled archs (maybe due to the code copy of libssw inside?). I would be glad if you could briefly comment on why -msse would be able to make i386 work. If you can confirm there’s nothing more I can do (short of implementing a non-SIMD fallback) I would just change the architectures and re-upload. Thanks Sascha
