Error: Signal Illegal instruction in phase Writing WHIRL file

Hey all,

A couple weeks ago I upgraded to Ubutu 10.04, with CUDA 3.0. I Got FindCUDA.cmake up and running and everything. However, I haven’t done much development in CUDA for several weeks until today. When I try and compile my project with Cmake, or even if I compile a project under the CUDA sdk, I get the following type of erros:

[codebox]nere@red5:~/cuda/current/NVIDIA_GPU_Computing_SDK/C/src/matrixMul$ make

Signal: Illegal instruction in Writing WHIRL file phase.

(0): Error: Signal Illegal instruction in phase Writing WHIRL file – processing aborted

*** Internal stack backtrace:

/home/nere/cuda/current/cuda/open64/lib//gfec() [0x7489af]

/home/nere/cuda/current/cuda/open64/lib//gfec() [0x7495f9]

/home/nere/cuda/current/cuda/open64/lib//gfec(ErrMsgLine+0x8d) [0x748d4d]

/home/nere/cuda/current/cuda/open64/lib//gfec() [0x749f96]

/lib/libc.so.6(+0x33af0) [0x2ba1c385faf0]

/home/nere/cuda/current/cuda/open64/lib//gfec(put_var_into_stack+0) [0x56e500]

/home/nere/cuda/current/cuda/open64/lib//gfec(c_mark_addressable+0x74) [0x511644]

/home/nere/cuda/current/cuda/open64/lib//gfec(build_array_ref+0xa5) [0x515675]

/home/nere/cuda/current/cuda/open64/lib//gfec(yyparse+0x3131) [0x6d48e1]

/home/nere/cuda/current/cuda/open64/lib//gfec(c_common_parse_file+0xf) [0x4ee8cf]

/home/nere/cuda/current/cuda/open64/lib//gfec(compile_file+0x55) [0x6a2c25]

/home/nere/cuda/current/cuda/open64/lib//gfec(main+0x48) [0x4d96b8]

/lib/libc.so.6(__libc_start_main+0xfd) [0x2ba1c384ac4d]

/home/nere/cuda/current/cuda/open64/lib//gfec(calloc+0x23a) [0x455b0a]

make: *** [obj/x86_64/release/matrixMul.cu.o] Error 1

[/codebox]

Never seen this before, and I don’t think I did anything to mess up CUDA, though it is likely I have run updates since moving to Ubuntu 10.04. Any clues?

Thanks

Andy

If the CUDA SDK build is also having problems, I don’t believe your problem is related to CMake or FindCUDA.cmake.

The only thing that I can think of that you should try is to make sure you are using GCC 4.3 instead of 4.4 (default in newer Ubuntu installations). I heard CUDA 3.0 doesn’t have support for GCC 4.0 yet.

See documentation for ‘nvcc --compiler-bin=/usr/bin/gcc-4.3/’. There were also forum posts about the GCC 4.4 issue, so search around.