More ARM binutils fuckage

Nicolas Pitre
Thu Dec 7 18:25:00 GMT 2006

On Thu, 7 Dec 2006, Michael K. Edwards wrote:

> Judging from the code in gcc/config/arm/arm.c (prefaced with "This has
> now turned into a maze"), completely specifying the code generation
> parameters involves:
>    -march
>    -mcpu (overrides -march, with a warning if they conflict)
>    -mtune (overrides tunings implied by -march and -mcpu,
>            but does not alter the __ARM_ARCH_xxx__ macro)
> or if none of the above are specified,
>    interworking, thumb, and mode26/32 selectors, which can result in
> selection of a different CPU tuning from the compile-time default;
> followed by some more Thumb, PIC, and frame register related flags, and then:
>    -mabi
>    -mfpe
>    -mfpu
>    -mfloat-abi
>    -mtp
> followed by structure alignment and the choice of PIC register.
> Endianness seems to be handled elsewhere, and aliases like -macps and
> -msoft-float appear to be mapped into the above.
> The OABI build of 2.6.19 that I just did has (among others):
>    -macps -mabi=apcs-gnu -march=armv5te -mtune=xscale
> -Wa,-mcpu=xscale  -msoft-float
> It does not have FPE, FPU, and TP selectors.  Does this mean that I
> get the compiler's default float format (FPA vs. VFP)?

Probably.  But floating point in the kernel is forbiden anyway and the 
-msoft-float will ensure that no FP instructions are issued at all (and 
give you a link error) ifyou slip some floats in the code by mistake.

> Are the FPE and TP selectors irrelevant to kernel compilation?



For unsubscribe information see

More information about the crossgcc mailing list