Sparc64 and -mno-fpu

pridkett · n00b Joined: 30 Jan 2003 Posts: 1

So I was compiling the kernel on the sunblade 100 that I just put gentoo on and I happened to notice something odd in the compile. Apparently the -mno-fpu flag is passed to the system for the architecture.

I did some looking around and saw that some people say it has nothing to do with math emulation, but reading the docs for GCC says that it's the same thing as -msoft-float which informs the compiler to use library calls for the floating point math. I also looked around in the tree and saw that the files in arch/sparc64/math-emu had been compiled indicating that it used the math emulation. I find this odd because I know that the UltraSparcIIe in a SunBlade 100 has a floating point unit.

Any help on this one? Should I be okay just to remove the -mno-fpu flag?

Kumba · Posted: Fri Jan 31, 2003 3:48 am Post subject:

I'm not too sure about this one. From I've heard, it's generally a wise idea to not bother flags the kernel uses because the kernel devs usually know what they're doing when they set them in the makefile. You could try and change it, but I'd keep a backup kernel just in case something went wrong. You might even try searching kernel mailing lists on sparc linux, see if anything is mentioned specifically about this flag.

Also, have you checked the EGCS docs regarding this flag? They may treat it differently than GCC 3.2.x, since all sparc64 kernels have to be built with egcs64-sparc.

--Kumba
_________________
"The past tempts us, the present confuses us, the future frightens us. And our lives slip away, moment by moment, lost in that vast, terrible in-between."
--Emperor Turhan, Centauri Republic