[tree-ssa] tree-ssa vs. fold

[Roger Sayle]

Hi Jeff,
> d) Fix fold() not to save_expr.
>
> Is d) the only real choice?  Am I missing something easy?

Maybe.  A very large number of optimizations in fold() use
save_expr (I committed one just this morning) and it would
probably be a significant performance hit to disable them all.
Even if this was just for java, there are a large number of
optimizations that you'd have to conditionalize.

However, one solution may be to provide a save_expr language
hook or equivalent.  For tree-ssa/gimple, this would generate an
assignment to a temporary and insert the assignment immediately
before the current statement (recorded in a global perhaps), and
then return the temporary.  For non-SSA, the hook would do as it
does at the moment.

There are a number of hooks or callbacks it would be nice for
fold to have, so that language front-ends such as java and ada
have some form of control over the optimizations/transformations
that it performs.  Avoiding bit-field accesses when generating
.class files comes to mind.

Does a save_expr hook seem a reasonable compromise/workable?

Roger
--
Roger Sayle,                         E-mail: roger@eyesopen.com
OpenEye Scientific Software,         WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road,     Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507.         Fax: (+1) 505-473-0833