Some FYI:
So I think I may have confused my terminology a bit. In any case, what I am
trying to do was recompile the Linux kernel to use the THUMB instruction
subset (ARM v4T), since the ARM website boasts a potential for 30% reduction
in code size. I thought this was what you meant by "half-word data-handling
instructions", but I think I was mistaken here.
So, using my gcc compiler (which was compiled using the
--with-cpu=strongarm110 option) with the following flag combinations, I came
up with these code sizes:
-mcpu=arm7 : 712,016
-mcpu=arm7 -march=armv3 : 712,016
-mcpu=arm7tdmi -march=armv4t : 703,824 (1.15% savings, 8192 bytes)
So I don't see the 30% savings here. Is it possible that I need to
recompile my GCC with different settings (does the strongarm110 option not
do THUMB)? Or do you think the 1.15% savings is all I'm going to get out of
it?
Thanks again!
-- Dan
> >Unfortunately, I didn't have the presense of mind to record
> the exact size
> >of the original kernel, so I'll need to estimate. I believe I saved
> >approximately 5K-15K on my "arch/arm/boot/Image" file, which
> is currently
> >712,016 bytes. I guess I expected more significant savings.
>
> Sounds about right to me.
>
> >Does gcc have a compile-time option to tell it not to use
> the Thumb (ARM v4)
> >instructions? Maybe that way I could get you some more
> accurate numbers
> >(without rebuilding my compiler again).
>
> Yup, you can use -march=armv3 to control this. This isn't very well
> documented but I think "gcc --help" does mention it.
>
> (Incidentally, LDRH isn't actually a Thumb instruction;
> StrongARM supports
> halfword operations, which are in v4 but not Thumb, which is
> v3T or v4T.)
>
> p.
>
unsubscribe: body of `unsubscribe linux-arm' to [EMAIL PROTECTED]
