IA-64 FP possibly broken
libgcc for IA-64 uses floating-point instructions to implement 64b integer division and modulo. The unsigned modulo (see source) however gives incorrect results for large dividends.
Superficially, it seems as if the operand is interpreted as signed, returning the signed (negative) modulo result instead of the correct unsigned one, however I haven't confirmed this.
While this might be problem in the libgcc assembly, I'm more inclined to suspect that the problem lies in our FP support.