|
From: | Dmitry Antipov |
Subject: | Re: Optimize glyph row clearing and copying routines |
Date: | Tue, 24 Sep 2013 16:42:57 +0400 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.0 |
On 09/24/2013 03:50 PM, Eli Zaretskii wrote:
I think -O3 is not interesting in this case. For -O2 with GCC, which is the most important case (as Emacs is normally built like that), you get net loss, because AFAIR copy_row_except_pointers is called much more often than clear_glyph_row. The important comparison is, of course, as part of some redisplay scenario, not just cycle comparison.
1) I do not believe in 5 CPU cycles difference too much. This is just ~5%, which may be explained by some irregular inaccuracy. 2) I'm talking about GCC 4.8.1, but people from this list still uses 4.2.x. or even older. Unfortunately GCC is known to have some serious regressions from time to time. 3) As it's proven by icc/clang, good compiler _really_ gets more optimization opportunities from new code rather than from old. IMHO we should provide workaround if it's known that gcc-X.Y.Z has serious bug; but we should not reject generally better code if we have just an unconvincing suspicion that gcc-P.Q.R introduces some miserable regression. Dmitry
[Prev in Thread] | Current Thread | [Next in Thread] |