This is the mail archive of the gcc-help@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: 32 bit float emulation on 16 bit machine


I tried but ...

> __ucmpsi2 is a function which does an unsigned comparison of two 32-bit
> integers.  You can get gcc to build that by compiling libgcc2.c with

GCC is able to compare unsigned 32-int in normal programs (out of libgcc).

> LIBGCC2_UNITS_PER_WORD defined as 2.  If your processor has a carry

If I define LIBGCC2_UNITS_PER_WORD as 2 then I get

/home/guest1/tam16/lib/gcc/tam16/4.3.3/libgcc.a(_addsub_df.o): In
function `_fpadd_parts':
/home/guest1/gcc/new_build/tam16/libgcc/../.././gcc/dp-bit.c:670:
undefined reference to `__ucmpdi2'
/home/guest1/gcc/new_build/tam16/libgcc/../.././gcc/dp-bit.c:675:
undefined reference to `__ucmpdi2'

Plus, in both case (LIBGCC2_UNITS_PER_WORD as 2 or as 4), __ucmpsi2 or
__ucmpdi2 is not called in _fpadd_parts. I checked by disassembling
libgcc.a.
And in the second case, float operation do not work. There are wrong
memory accesses.

Once, I managed to have float operations working well by changing the
macros definitions a lot and rebuilding each time. That is why I think
the problem is about macros settings.

> flag, it will often be more efficient to provide an insn which does it,
> these days typically via cbranchsi4.

Noted.

Thank you for your reply.

Regards,

Florent.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]