On Fri, 3 Jan 2014 22:29:20 +0200
Kalle Raiskila <[email protected]> wrote:

> > > > 
> > > > Indeed, but seems X and Y mean different things on GNU and
> > > > BSD :) I fixed the pocl.icd contents in 
> > > > https://github.com/pocl/pocl/commit/9f0f106dd9db48819501e13796553248cfef6253
> > > > to conform to what seemed to be the consensus on the 'net. The
> > > > best explanation (and I hope the correct one) I found was this:
> > > > http://en.opensuse.org/openSUSE%3aShared_library_packaging_policy#Versioning_schemes
> > > > 
> > > > So libpocl.so.3 seems to be correct, on BSD.
> > > 
> > > All right, I didn't digg that deep. I can live with version .3, if
> > > this is correct. So I think we need not to change much, do we?
> 
> I guess not, if the .so.3 does not look "wrong". After all, pocl
> 0.9 is going to be version 3 of the library, ABI compatible with
> version 1 (0.7).  
> 
> Seems *someone* thinks this is not correct behaviour, but the
> "bug" [sic?] was inserted some 15 years ago, according to:
> http://stackoverflow.com/questions/15215898/why-is-libtools-current-used-as-soversion-on-bsd-rather-than-major
> 
> But like I said, I have now half a day's experience of using
> FreeBSD, so I might be in the wrong :)
> 
> 
> > > > Unfortunately, I didn't get pocl to work :/ It blocks in
> > > > pthread_create on a mutex. Since Oliver already reported pocl to
> > > > work on BSD, I blame this failure on my BSD usage skills: I must
> > > > have goofed somewhere (this was my first time setting up a
> > > > FreeBSD). 
> > > 
> > > Wait, wait, wait ....
> > > 
> > > I was able to compile the port, but I wasn't able to perform even
> > > a single "OpenCL book benchmark/test run".
> 
> Uh, ok. What I did was clone the repo & build normally (without using
> any ports), then run "gmake check". the first check hangs on the
> mutex. I didn't even bother trying any other OCL applications.
> 
> > > ViennaCL is broken, it
> > > doens't even build with 0.9-sources with the ports Makefile I have
> > > created.
> 
> Oh? What sort of error are you seeing?

I will report detailed when I have installed the LLVM/CLANG 3.4 ports
again, see below.

> 
> 
> > > I should take a look at this on FBSD 11-CURRENT as well. How did
> > > you check? Did you also try to run the OpenCL book tests on your
> > > FBSD box?
> 
> Just a ../pocl/configure && make && make check.
> Then, when it stuck, I ran ./example/example1/.libs/example1 in the
> debugger. (remember to 'source ../pocl/tools/scripts/devel-envs.sh')

I use that port's Makefile I created. When running the OpenCL book's
examples, example1 gets stuck and using "procstat" shows something like

root@gate [~] procstat -t 56909
  PID    TID COMM             TDNAME           CPU  PRI STATE
WCHAN 56909 100602 example1         -                  1  152 sleep
umtxn root@gate [~] procstat -k 56909
  PID    TID COMM             TDNAME
KSTACK 56909 100602 example1         -                mi_switch
sleepq_catch_signals sleepq_wait_sig _sleep umtxq_sleep do_lock_umutex
__umtx_op_wait_umutex amd64_syscall Xfast_syscall

-t: Thread information
-k: Display the stacks of kernel threads in the process


I googled for [umtxn] since this seems to be the "state", but I do not
get much information, except that with the boost library, this "stuck
in some request to the kernel" seems to occur often.

> 
> > I compiled POCL 0.9 with LLVM/CLANG 3.3 since 3.4 seems to break.
> 
> Argh. I used 3.3 too, didn't even try 3.4. What is the issue there?
> 
> 
> kalle

I can't reacall exactly at the moment, there was a weird error about
missing header or similar, something really strange. I deleted the
ports from my systems, but I'll install them again. It seems that in 11
CURRENT LLVM 3.4/CLANG 3.4 is about to be the standard soon ... 


Well, I regret that my development abilities are very limited.

Oliver

Attachment: signature.asc
Description: PGP signature

------------------------------------------------------------------------------
Rapidly troubleshoot problems before they affect your business. Most IT 
organizations don't have a clear picture of how application performance 
affects their revenue. With AppDynamics, you get 100% visibility into your 
Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro!
http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
_______________________________________________
pocl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/pocl-devel

Reply via email to