This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: PATCH - [tree-ssa] regrouping of expression tree for single multiply add.



On Mar 24, 2004, at 11:17 AM, law@redhat.com wrote:


In message <1080008847.31835.196.camel@localhost.localdomain>, Diego Novillo wr
ites:
On Mon, 2004-03-22 at 21:25, Fariborz Jahanian wrote:

tree-ssa because it is an 'improvement' patch and I believe mainline is
frozen for improvement patches.


On the contrary. It is tree-ssa the frozen one. mainline is in
stage1. If this goes in mainline, tree-ssa will inherit it at the next
merge.


OK for tree-ssa?

No. Try mainline.
And I think there's an excellent argument against this kind of transformation
for the mainline as well, at least for FP types.

What is the argument? If issue is correctness for corner cases, is there additional gcc flag; besides -ffast-math, which need be checked for?

- Fariborz


This kind of transformation makes sense for integers though :-) THere's
some mips targets that have integer madd and several targets now have
1 integer multiply pipeline. Thus stuff like

a * b * c * d which is implemented as (((a * b) * c) * d) can be profitably turned into (a * b) * (c * d).

jeff










Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]