Closed letalvoj closed 8 years ago
So did you get it to work now? What OS, backend and libraries are you using? From what the problem looks like, it's OS X? Probably an incomatible libstdc++/clang++.
No I did not, but I thought that according to the thread the build was broken.
I have a Macbook Pro (late 2014) with Intel Iris Pro GPU. I do not use anything unusual:
I am able to compile the original caffe, after setting the proper paths to python, without any problem. But as you can see then I get weird semantic errors on this branch.
If a try to compile the latest master as it is then I get this error.
edit: libstdc++ ?? let's see...
libstdc++ is being set only if the version of CUDA is lower than 7. Forcing it manually just causes another error.
I do not think that it matters, but my Makefile.config is: http://pastebin.com/99Zk7m7b
Well, I recently switched my branch to use 64 bit indexing on everything. That might cause issue for OS X, which I personally do not have and can not test. Would be nice if we could resolve that somehow; meanwhile, try using a commit that is a bit older. For example 80d045263f26c41a1e886906a30d649a5c812038 could work for you.
Oh thanks! :) Btw. some of the casting errors were: ‘Could not cast long long to long’. If it is it, are not the longs long enough? Already 64bits…
On 02 Nov 2015, at 15:30, Fabian Tschopp notifications@github.com wrote:
Well, I recently switched my branch to use 64 bit indexing on everything. That might cause issue for OS X, which I personally do not have and can not test. Would be nice if we could resolve that somehow; meanwhile, try using a commit that is a bit older. For example 80d0452 https://github.com/naibaf7/caffe/commit/80d045263f26c41a1e886906a30d649a5c812038 could work for you.
— Reply to this email directly or view it on GitHub https://github.com/naibaf7/caffe/issues/13#issuecomment-153031494.
Ok, I give up.
I've already spend too much time trying to compile this stuff. After compiling the version you proposed it crashes during runtime. It seems to be the same behaviour as if I try to use a GPU with CPU only compiled binaries. Also the older Makefile does not allow disabling compilation of the CUDA backend, so it takes a looooooong time to compile.
@letalvoj Ok, the older Makefile from the commit version I noted above does allow to disable the backends the same way actually. As I said before, since I do not have a computer with OS X I can't really support it. My test systems use Ubuntu 14.04 and Fedora 22. If you still consider to get it working anyways, the only thing I can suggest is contacting one of those here who got it working on OS X, for example @lavania @paulbroyles @JacksonGariety
@naibaf7 Hi, I resolved OS X support issue at https://github.com/BVLC/caffe/pull/3720 Do you want to make a pull request in your repository? Please let me know :)
Thanks.
@cepiross I'll regression test this and merge into both branches, thanks. No need for another pull request.
Fix is merged.
Hi,
I tried to compile this code but it does not seem to work. Is there any commit from which I can start and compile the stuff, or is the code unusable yet?
While trying to compile the code I got mostly easy to fix casting errors:
but then I got this one: