For non-X86 platforms, we perform integer division by converting the integer arguments to doubles, perform the division and then convert the result back to an Int32. The insertion of the conversion nodes is done by the fixup phase. Currently the code uses fixDoubleEdge() for the inputs, but we have already performed a check of the int-ness of the arguments and this is really a special case. Therefore it makes more sense to do the Int32ToDouble node insertion without further checks of the child nodes.
From <rdar://problem/12645632>
Committed r140203: <http://trac.webkit.org/changeset/140203>