Please do not reply directly to this email. All additional comments should be made in the comments box of this bug. https://bugzilla.redhat.com/show_bug.cgi?id=456353 --- Comment #40 from Jeff Garzik <jgarzik@xxxxxxxxxx> 2010-07-14 01:17:03 EDT --- (In reply to comment #39) > Yeah, I have one more question. ffado wants to use the flags > -fomit-frame-pointer, -ffast-math, -funroll-loops for optimization. Will these > cause any harm? Is it okay if I add them to the build flags? -ffast-math is generally fine. It's a program choice which enables additional optimizations by deviating slightly from the IEEE standard, in ways most programs won't notice. It cannot be enabled by default build-system-wide because it affects strict correctness, but it is fine if knowledgeable persons OK it on a per-pkg basis. -fomit-frame-pointer should IMO be avoided in standard Fedora packages. It is typically an over-optimization left over from i386 days, where registers are such a precious resource. Although it varies from platform to platform, sometimes this can decrease ability to debug, and in Fedora, we want to generate valid debug info (stored in -debuginfo package). -funroll-loops unrolls loops whose iteration count can be determined at compile time. The idea is to increase pipelining at the expense of larger generated code size. If the developers really insist it's necessary, go ahead and keep it. But my default recommendation is to remove -funroll-loops, and let the compiler figure out what's best on its own. Sometimes, it makes more sense to keep the code compact and code size small, using fewer cachelines. So, unless there are vigorous objections, I would only add -ffast-math to the standard RPM build flags. -- Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug. _______________________________________________ package-review mailing list package-review@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/package-review