nvcc trouble nvcc bus error

Accelerated Computing CUDA CUDA Programming and Performance
1

So, where should I report errors like this:

; /usr/local/cuda/bin/nvcc -I/usr/local/cuda/include -I/Users/robertvanliere/include -I/sw/include -I/Users/robertvanliere/include/glew -I/Users/robertvanliere/src/boost_1_34_1 -c main.cu -o main.o
Signal: Bus error in Global Optimization – Compute alias classification phase.
Error: Signal Bus error in phase Global Optimization – Compute alias classification – processing aborted
nvopencc INTERNAL ERROR: /usr/local/cuda/bin/…/open64/lib//be died due to signal 4

; uname -a
Darwin rvl.lan 9.2.0 Darwin Kernel Version 9.2.0: Tue Feb 5 16:13:22 PST 2008; root:xnu-1228.3.13~1/RELEASE_I386 i386

This code compiles and runs correctly on vista (emu mode), XP, and fedora8 boxes.

Regards,
– Robert van Liere

2

This is the right place.

Can you post a small repro? Or contact me via PM.

Edit: It is a bug

3

Is there a work around for this error, or any idea what might cause it? I’m gettting the same thing:

habakuk:meshless_vis acorriga$ make

nvcc  -o obj/release/meshless_vis.cu_o -c meshless_vis.cu -I/usr/local/cuda/include -I/Users/acorriga/NVIDIA_CUDA_SDK_MACOSX/common/inc -I. -I../include -I/Users/acorriga/glew/include -DUNIX -O3

Signal: Bus error in Global Optimization -- Compute alias classification phase.

Error: Signal Bus error in phase Global Optimization -- Compute alias classification -- processing aborted

nvopencc INTERNAL ERROR: /usr/local/cuda/bin/../open64/lib//be died due to signal 4

make: *** [obj/release/meshless_vis.cu_o] Error 255

trying to compile libmeshlessvislibmeshlessvis, uname -a gives:

Darwin habakuk.local 9.2.2 Darwin Kernel Version 9.2.2: Tue Mar  4 21:17:34 PST 2008; root:xnu-1228.4.31~1/RELEASE_I386 i386

I’ve compiled and run the code in 64-bit Ubuntu and 32-bit Windows XP.

4

Not sure if this is a workaround… but if you do ‘nvcc -v’ you’ll see the steps that are taken to produce a .o object from the .cu source file. In my case, the step ‘nvopencc’ which generates the .ptx file was dumping core. So, I do these steps on a linux machine, and only do the steps from .ptx till .o on the mac.

The whole procedure can be automated by packing all this in a few shell scripts, and put the dependencies in the makefile. So, all I do is type make.

Tedious? Yes, but in this way I can still generate a Mach.o object of my cuda source.

Good luck.

– Robert van Liere

PS: alternative is to wait till the next ‘nvcc’ update…