Re: armv5b-softfloat with VFP

[Dimitry Andric <dimitry at andric dot com>]

On 2005-04-11 at 01:33:14 J. Scott Merritt wrote:

> I applied the arm-softfloat.patch from your website and performed a
> crosstool build for gcc-3.4.3-glibc-2.3.4.  I believe the VFP format
> is being used, because an fprintf of a floating number produces the
> *huge* result shown elsewhere,

Yes, make sure you also patch glibc with either glibc-vfp.patch from
my site, or if you have a recent crosstool, it's also included as
glibc-fp-byteorder.patch.


> however, even with the softfloat patch it looks as though the VFP
> format flag is not getting set in the ELF header (should be 0x600
> and its still at 0x200).

> - Should the softfloat patch update the ELF header to report
> VFP format or is this still missing ?

The softfloat patch modifies the flags that are passed by gcc to the
assembler (SUBTARGET_ASM_FLOAT_SPEC) to:

1) -mfpu=softvfp, if you give gcc no floating point options
2) -mfpu=softfpa, if you give gcc the -msoft-float option
3) -mfpu=fpa,     if you give gcc the -mhard-float option

Can you show us output of arm-linux-gcc -v -c, with just a random .c
file?


> - I assume that I need to also copy the glibc-vfp.patch forward
> to glibc-2.3.4 to solve the fprintf 1.0 problem ?

Yes.

Attachment: pgp00000.pgp
Description: PGP signature