ARM Architecture

[Michael Talyansky]

Apologies if the question has been asked already (did search, could not find):

Has anyone tried to compile folly for the ARM architecture, for instance, on Raspberry Pi?

Thanks in advance!

[Blake Matheny]

It won't all work. At least base should. Contributions welcome :)

---
Sent from mobile device

--

---
You received this message because you are subscribed to the Google Groups "Folly: the Facebook Open-source LibrarY" group.
To unsubscribe from this group and stop receiving emails from it, send an email to facebook-foll...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Michael Talyansky]

I will try to give it a whirl. I have seen some commits in regards to making FBString work for big-endian, I am sure there will be more of these. BTW, is there a way to build just the core?

You received this message because you are subscribed to a topic in the Google Groups "Folly: the Facebook Open-source LibrarY" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/facebook-folly/gmBlqfliUNE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to facebook-foll...@googlegroups.com.

[Blake Matheny]

Yes we run a subset of folly on ARM devices. It doesn't look like the autoconf build has a good way of building the core unfortunately.

Blake Matheny
bmat...@mobocracy.net
http://mobocracy.net

[Michael Talyansky]

But at least it would compile for you, with all the gtest and other libraries it pulls in? Or have you excluded the tests?

[Michael Talyansky]

Could you tell me which version of folly are you building? There are things in futures directory that don’t build for me even on 32-bit Intel, let alone ARM…

Thanks in advance!

[Blake Matheny]

Yes, futures and a couple of other things (the elf parser?) will not yet work on ARM.

[Michael Talyansky]

Yes, but do you still compile them? In version 57 futures doesn’t even compile for me on 32-bit Intel: 

libtool: compile:  g++ -DHAVE_CONFIG_H -I./.. -pthread -I/usr/include -std=gnu++0x -g -O2 -c futures/ManualExecutor.cpp  -fPIC -DPIC -o futures/.libs/ManualExecutor.o

In file included from ./../folly/futures/ManualExecutor.h:18:0,

                 from futures/ManualExecutor.cpp:17:

./../folly/LifoSem.h: In instantiation of 'struct folly::detail::LifoSemNode<folly::Baton<std::atomic>, std::atomic>':

./../folly/LifoSem.h:326:7:   required from 'void folly::detail::LifoSemBase<Handoff, Atom>::post() [with Handoff = folly::Baton<std::atomic>; Atom = std::atomic]'

./../folly/futures/ManualExecutor.h:79:17:   required from here

./../folly/LifoSem.h:152:3: error: static assertion failed: Handoff too big for small-object optimization, use indirection

   static_assert(sizeof(Handoff) <= sizeof(LifoSemRawNode<Atom>::raw),

   ^

make[2]: *** [futures/ManualExecutor.lo] Error 1

Thanks for the help!

[Blake Matheny]

We don't compile them for ARM, we do for x86-64. I doubt we have 32 bit machines building folly, but it's possible? The only thing I know for sure works with folly on 32 bit devices is the core functionality. Added Hans to CC who might know.

[Hans Fugal]

We don't build 32-bit internally and I'm not aware of any explicit work done to make it work with 32-bit, so I would be a little surprised if it works out of the box. I also wouldn't be too surprised if it works without *too* much effort, mostly in the realm of usage of other folly features like Baton. Some of the ifdefs for apple might be a good place to start - I think the alternatives we use for darwin might just work on linux i386.

I'm not sure if you're asking about building core futures, or core folly, or what exactly. I doubt our current (autoconf) public build system splits things out very conveniently, but as we migrate to BUCK it would also make sense to be modular about what we build. I don't know if anyone's working on that at the moment, though.