This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [RFC] Matrix Flattening optimization
- From: Falk Hueffner <falk at debian dot org>
- To: "Richard Guenther" <richard dot guenther at gmail dot com>
- Cc: "Ayal Zaks" <ZAKS at il dot ibm dot com>, "Shantonu Sen" <ssen at opendarwin dot org>, "Razya Ladelsky" <RAZYA at il dot ibm dot com>, gcc-patches at gcc dot gnu dot org, "Daniel Berlin" <dberlin at dberlin dot org>, jh at suse dot cz
- Date: Fri, 03 Mar 2006 15:53:48 +0100
- Subject: Re: [RFC] Matrix Flattening optimization
- References: <OF6E18706D.C44D80FC-ONC2257125.007ABE9A-C2257125.007AC898@LocalDomain> <OFD86ED807.00B76AD8-ONC2257125.007AD22D-C2257125.007CC426@il.ibm.com> <84fc9c000603030310p5945ad83idec475b4590787ac@mail.gmail.com>
"Richard Guenther" <richard.guenther@gmail.com> writes:
> Even padding to either power-of-two row size (for cheap
> multiplication) or cache-line aligned row-size if cheap enough may
> be worth in general.
Power-of-two rows can increase the probability of cache line thrashing
due to mapping conflicts (depending on the cache architecture), so
actually you might want to avoid them.
BTW, what about targets where a multiplication is slower than an
indirect access? Unless strength reduction can be done, this
transformation might actually lead to a slowdown there...
--
Falk