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] Add tree rest of compilation timevar


On Dec 29, 2004, at 11:36 AM, Andrew Pinski wrote:


On Dec 29, 2004, at 2:28 PM, Jeffrey A Law wrote:


On Wed, 2004-12-29 at 00:28 -0500, Andrew Pinski wrote:
As Jeff Law and others have discussed before, we attribute some non
expand related code to expand. This patch add a new time var for
tree rest of compilation, just like rest of compilation for RTL level.


OK? Bootstrapped and tested on powerpc-darwin.

Thanks,
Andrew Pinski

ChangeLog:

* timevar.def (TV_TREE_REST): New time var.
* tree-optimize.c (tree_rest_of_compilation): Use TV_TREE_REST instead
of TV_EXPAND.
All this does is move the problem from one place TV_EXPAND to
TV_TREE_REST -- it does nothing to actually resolve the issues at hand.
[ ie, TV_TREE_REST gets all the random tree-optimizer stuff plus the
tree->rtl expansion time. ]

No it does not get the expansion time because pass_expand has TV_EXPAND as its timevar.

Actually, I've got a more general question: do people find this timevar stuff useful in understanding compilation performance? If so, what do they find it useful for? If we have a clearer idea of the purpose of this feature, we might get some better guidance on how to change it.


For myself, I don't find this feature very useful.. I've been working pretty seriously on compiler performance for a while, and I usually don't get much out of -ftime-report. Most of the categories are either too coarse or too fine, so it's hard to take the -ftime-report results and translate them into which parts in the source are taking the most time. Besides, turning on -ftime-report changes the compilation time so much that I'm not sure I should believe it anyway. Timing measurements are extremely important in getting better performance, but I've found other kinds of measurements more useful.

--Matt


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