Kazu Hirata wrote:
I am planning to reorganize fold as suggested by Roger Sayle on IRC.
good for you! reorganizing fold is an excellent idea.
The shortest way to describe this mini project would be to develop the tree equivalent of simplify_gen_ARITY and simplify_ARITY in the RTL world. Doing so should reduce the number of scratch tree nodes created when idioms like fold (buildN (...)) are used. Hopefully, we should be able to pick up some compile-time improvement along the way.
By the way, the past proposals from Roger Sayle are found at:
http://gcc.gnu.org/ml/gcc-patches/2003-10/msg01514.html http://gcc.gnu.org/ml/gcc/2004-01/msg00560.html
Both of these are along the same lines as above.
Any comments?
I question if it is better to fold early. As I've said before, I think the optimizations that fold performs should be turned into a proper SSA optimization phase% -- that can be repeatedly applied as necessary. In the front end, folding should not generally be done. I see two needs for early folding,
1) evaluation of constant expressions, where the language requires a compile time constant value. Such values should be calculated lazily by some language specific function (possibly using common backend functions). This would produce an error or an appropriate _CST node.
2) A common occurrence is such things as fold (build (PLUS_EXPR, a, b)), where it is known that a and b are _CST nodes. This kind of thing happens in array initialization and vtable construction, and probably several other bookkeeping tasks. Here the operation is virtually always one of ADD, SUB, MUL and the types are integral types. Here we should have an interface that ONLY deals with those three operations and ONLY deals with already-coerced integral types. Whether we have separate eval_integral_OP (a, b) functions or eval_integral (OP, a, b), I don't know.
nathan
%-I guess gimplifying unfolded expressions might lead to an explosion of statements. It might be prudent to apply folding during the gimplification phase as well. If the SSA folding pass has the correct interface, I'd hope it can be applied to gimple statements and pre-gimplified trees.
-- Nathan Sidwell :: http://www.codesourcery.com :: CodeSourcery LLC [EMAIL PROTECTED] :: http://www.planetfall.pwp.blueyonder.co.uk
