With some gauge fields generated by Chroma, calling QDP_string_ptr on the XML from the gauge field record (as opposed to the overall file) leads to a segfault. This, in particular, happens on line ~112 of lib/qio_field.cpp.
For now there's a WAR in place for gauge fields since I've never hit this issue with spinor fields (and MILC has also successfully read spinor fields generated by QUDA), which suggests this is a "narrow" issue.
So far as I can tell this is an issue with the way these gauge fields were written by QIO in the first place (via QDP++), but I haven't nailed it down yet---I haven't hit this issue with any files written by QUDA, and there's been success loading files written by QUDA into MILC. Anyway, further debugging abounds.
With some gauge fields generated by Chroma, calling
QDP_string_ptr
on the XML from the gauge field record (as opposed to the overall file) leads to a segfault. This, in particular, happens on line ~112 oflib/qio_field.cpp
.For now there's a WAR in place for gauge fields since I've never hit this issue with spinor fields (and MILC has also successfully read spinor fields generated by QUDA), which suggests this is a "narrow" issue.
So far as I can tell this is an issue with the way these gauge fields were written by QIO in the first place (via QDP++), but I haven't nailed it down yet---I haven't hit this issue with any files written by QUDA, and there's been success loading files written by QUDA into MILC. Anyway, further debugging abounds.