This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH/RFC v2] PR68212: Improve Accounting of Block Frequencies During Loop Unrolling
- From: Jeff Law <law at redhat dot com>
- To: Bernd Schmidt <bschmidt at redhat dot com>, Kelvin Nilsen <kdnilsen at linux dot vnet dot ibm dot com>, gcc-patches at gcc dot gnu dot org
- Date: Wed, 2 Dec 2015 12:21:02 -0700
- Subject: Re: [PATCH/RFC v2] PR68212: Improve Accounting of Block Frequencies During Loop Unrolling
- Authentication-results: sourceware.org; auth=none
- References: <564FC6F2 dot 7050608 at linux dot vnet dot ibm dot com> <5655DF73 dot 7080004 at redhat dot com>
On 11/25/2015 09:18 AM, Bernd Schmidt wrote:
I'm reading up on this stuff, but I'm probably still not the best person
to review the actual frequency manipulation parts in this. There are a
few things I can comment on, however.
The first question would be, have you looked at the rebuild_frequencies
code in predict.c, and whether you can reuse that entirely or in part?
There's similar code in the tree-ssa-threadupdate.c which tries to
update the frequencies/probabilities in response to all the block
copying and CFG reorganization that occurs as a result of jump
threading. It's a bit of a nasty problem.
I can try to help with the frequency manipulation bits. It's not my
strong suit, but I did have to dig into it at more than the surface
level when the google folks submitted their rewrite of the frequency
adjustments in tree-ssa-threadupdate.c
Kelvin, if you could post your next iteration, it'd be helpful.
jeff