I just spotted that we're not depending on SQR_BASECASE_THRESHOLD in any of the recursive toomN_sqr calls. I think we should, at least from toom2_sqr. -- Torbjörn