The alignment is determined by compiling a small test program. Most likely, this fails. The real error is then the inability to compile even small programs.
-erik On 2013-07-19, at 20:41 , Chuck Zhao <[email protected]> wrote: > I found the issue. > > The root problem is that, inside ${OBJ}/config.h, > /* The normal alignment of `double16', in bytes. */ > #define ALIGNOF_DOUBLE16 0 > > /* The normal alignment of `float16', in bytes. */ > #define ALIGNOF_FLOAT16 0 > > > Notice the values for both of the alignof macros are 0. > > This caused the posix_memalign() call inside pthread.c to fail, because it is > not aligned on any power-of-2 boundary. > > Once I put respective values of 128, 64 in, they are both fine. > > All 3 examples apps run smoothly now. > > I looked into ${POCL_SRC_0.8RC7}/configure, there is nothing in particular > wrong with the configuration script. > the default of ac_cv_alighof_double16 is 0, but the same script works fine on > my Ubuntu/x64 box, with the proper alignment generated. > > Did anybody ever seems to have seen such a behavior before? > Not sure if it is a compiler issue, or autoconf/automake issue, or is an > issue with my particular pandaboard (I have only 1 panda board). > > I changed a few versions of gcc, the issue remains. > > > Thank you > > Chuck > > On 7/19/2013 4:32 PM, Chuck Zhao wrote: >> Omm, it doesn't even reach the kernel compilation stage. >> >> The failure is inside clCreateBuffer() call; >> Inside clCreateBuffer(), it failed on device_ptr = device->malloc(...), >> This is the malloc() defined inside pthread.c:252-pocl_pthread_malloc(); >> >> continue debugging... >> >> Chuck >> >> >> On 7/19/2013 3:05 PM, Kalle Raiskila wrote: >> >>> On Fri, 19 Jul 2013 14:23:26 -0700 >>> Chuck Zhao >>> <[email protected]> >>> wrote: >>> >>> >>>> Kalle, >>>> >>>> Thank you for the hints. >>>> >>>> I noticed that when building the lib/kernel package, both >>>> clang-3.3/3.2 complained heavily on "-target armv7", saying >>>> "unrecognized flag, use -mfloat=soft" instead. >>>> >>>> I hacked the Makefile a little bit, and use "-target >>>> armv7l-unknown-linux-gnueabi" instead. >>>> The original warning all go away. >>>> >>>> However, the produced tests still fail. >>>> E.g. >>>> >>>> czhao@panda0:~/ResearchTools/Compiler/POCL/pocl-0.8/robj32-rc7/examples/example1/.libs$ >>>> >>>> ./lt-example1 >>>> ERROR >>>> (0.000000, 0.000000, 0.000000, 0.000000) . (0.000000, 0.000000, >>>> 0.000000, 0.000000) = -0.000007 >>>> FAIL >>>> >>>> czhao@panda0:~/ResearchTools/Compiler/POCL/pocl-0.8/robj32-rc7/examples/example1/.libs$ >>>> >>>> >>>> >>>> >>>> I am debugging now. >>>> >>>> Any hints/suggestions are highly appreciated. >>>> >>> Do save the intermediate files the kernel compiler creates (i.e. set >>> evnironment variable POCL_LEAVE_TEMP_DIRS, see doc/envs.txt), and have >>> a look at the target triple in the generated .bc files. Is it still the >>> original "armv7l-..."? At least with clang 3.2 I observed it was not, >>> causing errors as above. >>> >>> HTH :) >>> >>> kalle >>> >>> ------------------------------------------------------------------------------ >>> See everything from the browser to the database with AppDynamics >>> Get end-to-end visibility with application monitoring from AppDynamics >>> Isolate bottlenecks and diagnose root cause in seconds. >>> Start your free trial of AppDynamics Pro today! >>> >>> http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk >>> >>> _______________________________________________ >>> pocl-devel mailing list >>> >>> [email protected] >>> https://lists.sourceforge.net/lists/listinfo/pocl-devel >>> >>> >>> >> >> ------------------------------------------------------------------------------ >> See everything from the browser to the database with AppDynamics >> Get end-to-end visibility with application monitoring from AppDynamics >> Isolate bottlenecks and diagnose root cause in seconds. >> Start your free trial of AppDynamics Pro today! >> >> http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk >> >> _______________________________________________ >> pocl-devel mailing list >> >> [email protected] >> https://lists.sourceforge.net/lists/listinfo/pocl-devel >> >> >> > > ------------------------------------------------------------------------------ > See everything from the browser to the database with AppDynamics > Get end-to-end visibility with application monitoring from AppDynamics > Isolate bottlenecks and diagnose root cause in seconds. > Start your free trial of AppDynamics Pro today! > http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk_______________________________________________ > pocl-devel mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/pocl-devel -- Erik Schnetter <[email protected]> http://www.perimeterinstitute.ca/personal/eschnetter/ My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://keys.gnupg.net. ------------------------------------------------------------------------------ See everything from the browser to the database with AppDynamics Get end-to-end visibility with application monitoring from AppDynamics Isolate bottlenecks and diagnose root cause in seconds. Start your free trial of AppDynamics Pro today! http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk _______________________________________________ pocl-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/pocl-devel
