compile error

tgm

unread,

May 20, 2008, 10:49:37 PM5/20/08

to CUDPP

How to fix it?
Thanks!

nvcc -o obj/release/segmented_scan_app.cu_o -c src/app/
segmented_scan_app.cu --host-compilation=C --compiler-options -fno-
strict-aliasing -I./ -I./include/ -Isrc/ -Isrc/app/ -Isrc/kernel/ -
Isrc/cta/ -I. -I/usr/local/cuda/include -I./../common/inc -DUNIX -O3
nvcc error : 'ptxas' died due to signal 11 (Invalid memory reference)
make: *** [obj/release/segmented_scan_app.cu_o] Error 255

Shubho Sengupta

unread,

Jun 2, 2008, 7:07:21 PM6/2/08

to cu...@googlegroups.com

Which OS? compiler version?

Shubho

oxuser

unread,

Jun 5, 2008, 11:27:39 AM6/5/08

to CUDPP

I also have a similar problem:

make -f Makefile
*** stack smashing detected ***: ptxas terminated
Aborted (core dumped)
make: *** [obj/release/scan_app.cu_o] Error 255

I am using Ubuntu, gcc 4.1.3 and nvcc 1.1

Shubho Sengupta

unread,

Jun 5, 2008, 9:36:06 PM6/5/08

to cu...@googlegroups.com

Which version of Ubuntu?

Shubho

Mark Harris

unread,

Jun 6, 2008, 4:43:40 AM6/6/08

to cu...@googlegroups.com

Yes, I believe the build failure is a problem only with CUDA 1.1. CUDA 2.0 will be released soon, and the beta is already available. Are you planning to upgrade?

Thanks,

Mark

Kristleifur Daðason

unread,

Jun 6, 2008, 6:31:22 AM6/6/08

to cu...@googlegroups.com

Just to chime in:
I've seen the compiler fail like that when
[CUDA kernel name] + [variable names in CUDA kernel] = too long!

Mark Harris

unread,

Jun 6, 2008, 6:49:56 AM6/6/08

to cu...@googlegroups.com

True, but this is just a ptxas *crash*, which can happen for a number of reasons. No matter what, it's a bug in ptxas. It's possible that all the template parameters make the mangled kernel names too long for CUDA 1.1.

Mark

Reply all

Reply to author

Forward