OpenACC bug?


I dont know if I missed something, but when I try to compile this piece of code:

#pragma acc region
    for (j = 0; j < Ly ; j++)
	#pragma acc for private ( s[0: ] )
	for (i = 0; i < Lx ; i=i+2)	// Just for Lx odd
	    s[i + Lx * j] = s_tmp[i + Lx * j];

I get:
call to cuMemFree returned error 700: Launch failed
CUDA driver version: 4010

If I trade i=i+2 for i++ the code runs correctly.

Hi maicon,

A “700” error typically means that your kernel abnormally aborted so some reason. The most common case being a out-of-bounds array access. Is “s” being accessed out-of-bounds?

If not, can you post reproducing example?