Bug 10606 – optimization flags and advice

• Product: [OLD] Docs-user
• Status: RESOLVED FIXED

[Description Spider (RETIRED) 2002-11-11 18:09:00 UTC]

Could we get a notice that if something breaks with customized CFLAGS (in the
CFLAGS section), that the user is to try  "-march= -O2"  before reporting a bug?

I've had bugreports from people who use :
-march=pentium3 -O3  -pipe -fomit-frame-pointer -fexpensive-optimizations
-fprefetch-loop-arrays -ffast-math -funroll-loops -foptimize-sibling-calls
-freorder-blocks 

and it was only because of overoptimization that things refused to go through
configure stage even.

[Comment 1 Stoyan Zhekov (RETIRED) 2002-11-13 00:56:07 UTC]

In witch document to add this? Install guide? FAQs? somewhere else?

[Comment 2 Spider (RETIRED) 2002-11-13 11:50:09 UTC]

in the FAQ, preferrably directly after this section:
I'm finding things to be really unstable and I'm using "-O9 -ffast-math
-fomit-frame-pointer" optimizations. What gives? 

When you use any optimizations beyond -O3, you're really taking the risk of
having broken packages. Very aggressive optimizations sometimes cause the
compiler to streamline the assembly code to the point where it doesn't quite do
the same thing anymore. Loc-Dog (on IRC) uses -O3 -mcpu=i686 -march=i686
-fforce-addr -fomit-frame-pointer -funroll-loops -frerun-cse-after-loop
-frerun-loop-opt -malign-functions=4, which is about as much as I'd want to push
global optimization settings. Beyond this, it's best to use ultra-high
optimizations only with specific packages where you really need that extra 2%,
(eg graphics and various multimedia programs), and where you can easily test the
package to ensure that it hasn't been optimized into oblivion.

[Comment 3 Stoyan Zhekov (RETIRED) 2002-11-18 00:13:29 UTC]

fixed in the CVS