ptxas - invalid memory reference

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hello,

I was wondering what in general causes a ‘‘ptxas’ died due to signal 11 (Invalid memory reference)’ error. I have checked the kernel name and it does not seem to exceed the name limit of 1060 and I use CUDA 3.2.

For a little context , I get the error when I try to print out certain values in the inner loop :-

for( j = 0 ; j < info->nq ; j++ )

        {             

          Aterm[j] = 0.0;

          Bterm[j] = 0.0;

          Cterm[j] = 0.0;

printf("-----PTXAS--------------- %lf-----%lf-----%lf -------------------\n", Aterm[j] , Bterm[j] , Cterm[j] );  

for( k = 0 ; k < info->nq ; k++ )

            {                           

              Aterm[j] = Aterm[j] + Atau[j][k]*delF[k];

              Bterm[j] = Bterm[j] + Btau[j][k]*delF[k];

              Cterm[j] = Cterm[j] + Ctau[j][k]*delF[k];

printf("-------------------- %lf-----%lf-----%lf-------------------", Aterm[j] , Bterm[j] , Cterm[j] );                 

            }

        }

Now why would accessing any of these arrays, which are statically allocated, cause a memory error ?

Thanks

#2

This looks like an internal compiler error to me. If this issue persists with the CUDA 4.0 release candidate, it would be helpful if you could file a bug, attaching a small, self-contained repro case. Thanks!

#3

Thanks !