cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

nibal
Challenger

Latest catalyst 15.9 coredumps when compiling ocl program

gdb Output (same with CodeXL's):

(gdb) where
#0 0x00007ffff53aaf30 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#1 0x00007ffff41ff0d5 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#2 0x00007ffff41fd568 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#3 0x00007ffff3fe13f8 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#4 0x00007ffff3fe2e00 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#5 0x00007ffff3fe3504 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#6 0x00007ffff3fe38aa in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#7 0x00007ffff4408286 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#8 0x00007ffff4440a07 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#9 0x00007ffff444073d in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#10 0x00007ffff444022b in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#11 0x00007ffff443c69d in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#12 0x00007ffff443b843 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#13 0x00007ffff43fcf44 in aclCompile ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#14 0x00007ffff3c0faf0 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#15 0x00007ffff3c0fc1e in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#16 0x00007ffff3c19954 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#17 0x00007ffff3c2141a in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#18 0x00007ffff3bb4ae5 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#19 0x00007ffff3bc95a6 in ?? ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#20 0x00007ffff3ba9cdd in clBuildProgram ()
from /opt/AMDAPPSDK-2.9-1/lib/x86_64/libamdocl64.so
#21 0x0000000000410d67 in createCLProgram (ctx=0xdef550, dev=0x69a0c0,
kernel=0x414faf "fft.cl", cflags=0x414fa8 "-g -I.") at clutils.c:497
#22 0x000000000040e522 in setupCL () at dsp.c:227
#23 0x000000000040ebfd in init_dsp (fc=2000, sr=300000) at dsp.c:311
#24 0x000000000040ab6e in main (argc=0, argv=0x7fffffffdcf0) at sdmp.l:2293

Core created when input kernel arguments change from __global to __local

0 Likes
1 Solution

Yes, I did. Please find the attached output file.

View solution in original post

11 Replies
dipak
Big Boss

Hi,

Could you please share the reproducible kernel code?

0 Likes

Sure, there you go.

To reproduce core, you need to compile as C file, with cflags "-g -I.". As c++ file errs OK.

It is a modification of your FFT sample, so it might be easier to just plug it in there.

Compiled in Linux (Ubuntu 14.04 x64)

0 Likes

When I tried to build the kernel file using CodeXL, I didn't get any such error. Instead, CodeXL reported a valid compilation error (please check the attached snapshot).

The error is obvious. Local data pointer is passed to the "Init()" function when it expects a global data pointer.

My setup: Ubuntu 14.04 64bit, latest 15.9-15.201.1151, Latest CodeXL 1.8, APP SDK 3.0

Regards,

0 Likes

I don't disagree that the error is obvious. i disagree that it core dumps instead of printing that error. Did you try to run it on its own, with "-g -I." compilation flags? If you compile it as c++ it will err correctly, as already stated.

0 Likes

Yes, I did. Please find the attached output file.

I don't how you did it, but you did it. In my case it reproducibly core dumps:-(

0 Likes

It wil crash on older drivers, it's a well known bug and has been reported in this section a few times. Here's one.

It has been fixed in 15.12 I think.

0 Likes

Where do you find catalyst 15.12? The most recent i could find in the public site is 15.9 😞

0 Likes

My control panel currently reports 15.20 FYI.

But hey, that's linux for you (I guess). They have been promising an unified display driver model more than a decade now, apparently it's still not a priority for them.

0 Likes

ty. I will check next time in the AMD site for the different versions of OS. And yes, I use Ubuntu 14.04 x64, which is Linux;-)

0 Likes

nibal wrote:

I don't how you did it, but you did it. In my case it reproducibly core dumps:-(

Hmmm. Rather I do know how. You used SDK 3.0, while i use SDK 2.9.1. Seems it is an SDK 2.9.1 coredump after all, as indicated in the trace, not a catalyst one as assumed...

0 Likes