[Bug tree-optimization/81633] [7/8 Regression] Incorrect floating point result with tree vectoriser

["jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81633

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
To answer myself, child_index doesn't need to be equal to i, e.g. if some
operand is constant in all the statements, then there is no SLP child for it.
If there are no NULL oprnd, then we can as well just start from one above the
previous child_index, what we before this change.
The question for NULL oprnd is whether the omitted operand will always be a
SSA_NAME which always has (or always does not have) a corresponding
SLP_TREE_CHILDREN child node.  If it is not unconditional, then perhaps we need
to say add an index integer into struct _slp_tree and use those numbers to find
the slp child or lack thereof based on the position of the argument, not its
value.