1 |
On Wed, 22 Sep 2004 21:40:39 -0400 John Richard Moser |
2 |
<nigelenki@×××××××.net> wrote: |
3 |
| | Personally, I would be *very* wary about giving our x86 users a 5% |
4 |
| | performance hit |
5 |
| |
6 |
| 1. Where are you getting 5% from? |
7 |
|
8 |
I did some quick measurements on vim's test suite with and without |
9 |
-fstack-protector. The value varies per app, of course, but vim's regex |
10 |
stuff was taking a ~10% hit (not surprising) and the file ops were |
11 |
affected considerably less. *shrug* depends upon the app, of course. |
12 |
|
13 |
| 2. What context is this "Performance" hit in? gcc would take a |
14 |
| "performance" hit because it eats 100% CPU; most nothing else would |
15 |
| take a "performance" hit unless the *overhead* pulled CPU usage up to |
16 |
| 100% for a time. |
17 |
|
18 |
Well, performance when a box is at low load is of no interest... |
19 |
|
20 |
| I guess this is the point where I have to ebuild unpack nbyte and |
21 |
| generate SSP benchmarks. Why oh why didn't I bench ssp when I was |
22 |
| doing PIC? |
23 |
|
24 |
Note that some kinds of benchmarks, such as integer op tests, won't |
25 |
be affected at all by SSP. Certain kinds of string handling, on the |
26 |
other hand, will be hit really badly. Better to pick an app and |
27 |
benchmark it than running dedicated benchmarkers. |
28 |
|
29 |
-- |
30 |
Ciaran McCreesh : Gentoo Developer (Sparc, MIPS, Vim, Fluxbox) |
31 |
Mail : ciaranm at gentoo.org |
32 |
Web : http://dev.gentoo.org/~ciaranm |