Re: [PATCH] scripts: add softfp support

["Yann E. MORIN" <yann.morin.1998@anciens.enib.fr>]

Michael, All,

On Wednesday 19 October 2011 04:29:20 Michael Hope wrote:
> # HG changeset patch
> # User Michael Hope <michael.hope@linaro.org>
> # Date 1318991252 -46800
> # Node ID a31d097e28cd73d07a5484129929a500b4d58efa
> # Parent  a32156bd31c0d395e8d346431b123a7d2caa14cd
> scripts: add softfp support
> 
> ARM compilers can be built for soft float (software only, floats in
> core registers), hard float (uses floating point instructions, floats
> in FPU registers), or the half-way house softfp (uses floating point
> instructions, floats in core registers).

Feature definitely a nice addition, but too close to the release to
add it now (which reminds me I should document the release plan on
the website...). FYI, it's a release every three months, with about
a 15-day slack before, used to stabilise the stuff. Next release is
due by October the 31st, so we just entered the 15-day delay...

I was thinking about cutting the release branch ahead of time, but
handling both the relase and the devel at the same time is a bit
complicated in my head, and does colide a bit on the schedule...
I'd like to try, but this release is special: it also colides with
the Prague events. Probably I'll do it for the next release (Feb`12)...

> Add support for softfp cross compilers to the GCC and GLIBC
> configuration.  Needed for Ubuntu and other distros that are softfp.

What about uClibc? How will it cope with softfp?

> Signed-off-by: Michael Hope <michael.hope@linaro.org>
> 
> diff -r a32156bd31c0 -r a31d097e28cd config/target.in
> --- a/config/target.in	Sun Oct 16 17:51:42 2011 +0200
> +++ b/config/target.in	Wed Oct 19 15:27:32 2011 +1300
> @@ -271,6 +271,22 @@
>        If your processor has no FPU, then you most probably want this, as it
>        is faster than emulating the FPU in the kernel.
>  
> +config ARCH_FLOAT_SOFTFP

I'm a bit reluctant at adding an architecture-specific option to this
generic file. Currently, all arch options are in the related arch file.

However, I agreee that there is no easy way to nicely handle that with
the current infrastructure... :-/

*But* I recall a similar approach a few months back... It seemed that ARM
is not the only architecture that support softfp. Seems PPC also uses it.
So:

> +    bool
> +    prompt "softfp"
> +    depends on ARCH_arm

 - this arch-specific "depends on" should go away
 - and either we keep the option un-protected, or we hide it behind
   ARCH_SUPPORT_SOFTFP (or the like) which is set by archs that
   support it.

> +    help
> +      Emit hardware floating point opcodes but use the software
> +      floating point calling convention.
> +      
> +      Architectures such as ARM use different registers for passing
> +      floating point values depending on if they're in software mode
> +      or hardware mode.  softfp emits FPU instructions but uses the
> +      software FP calling convention allowing softfp code to
> +      interoperate with legacy software only code.
> +
> +      If in doubt, use 'software' or 'hardware' mode instead.
> +
>  endchoice
>  
>  config TARGET_CFLAGS
> diff -r a32156bd31c0 -r a31d097e28cd scripts/build/libc/glibc-eglibc.sh-common
> --- a/scripts/build/libc/glibc-eglibc.sh-common	Sun Oct 16 17:51:42 2011 +0200
> +++ b/scripts/build/libc/glibc-eglibc.sh-common	Wed Oct 19 15:27:32 2011 +1300
> @@ -132,9 +132,10 @@
>          *) extra_config+=("--disable-shared");;
>      esac
>  
> -    case "${CT_ARCH_FLOAT_HW},${CT_ARCH_FLOAT_SW}" in
> -        y,) extra_config+=("--with-fp");;
> -        ,y) extra_config+=("--without-fp");;
> +    case "${CT_ARCH_FLOAT_HW},${CT_ARCH_FLOAT_SW},${CT_ARCH_FLOAT_SOFTFP}" in
> +        y,,) extra_config+=("--with-fp");;
> +        ,y,) extra_config+=("--without-fp");;
> +        ,,y) extra_config+=("--with-fp");;
>      esac

Argh!... This is starting to be unreadable... :-/

config ARCH_FLOAT
    string
    default "hard"   if ARCH_FLOAT_HW
    default "soft"   if ARCH_FLOAT_SW
    default "softfp" if ARCH_FLOAT_SOFTFP

Then:
    case "${CT_ARCH_FLOAT}" in
        hard)   ...;;
        soft)   ...;;
        softfp) ...;;
    esac

I'll do it.

>      if [ "${CT_LIBC_DISABLE_VERSIONING}" = "y" ]; then
> diff -r a32156bd31c0 -r a31d097e28cd scripts/functions
> --- a/scripts/functions	Sun Oct 16 17:51:42 2011 +0200
> +++ b/scripts/functions	Wed Oct 19 15:27:32 2011 +1300
> @@ -984,6 +984,7 @@
>      [ "${CT_ARCH_TUNE}"     ] && { CT_ARCH_TUNE_CFLAG="-mtune=${CT_ARCH_TUNE}";  CT_ARCH_WITH_TUNE="--with-tune=${CT_ARCH_TUNE}"; }
>      [ "${CT_ARCH_FPU}"      ] && { CT_ARCH_FPU_CFLAG="-mfpu=${CT_ARCH_FPU}";     CT_ARCH_WITH_FPU="--with-fpu=${CT_ARCH_FPU}";    }
>      [ "${CT_ARCH_FLOAT_SW}" ] && { CT_ARCH_FLOAT_CFLAG="-msoft-float";           CT_ARCH_WITH_FLOAT="--with-float=soft";          }
> +    [ "${CT_ARCH_FLOAT_SOFTFP}" ] && { CT_ARCH_FLOAT_CFLAG="-mfloat-abi=softfp"; CT_ARCH_WITH_FLOAT="--with-float=softfp";        }

And the last time this came up, it was pointed that CT_ARCH_FLOAT_HW did
force neither -hard-float not --with-float=hard

I'll look at it...

>      # Build the default kernel tuple part
>      CT_TARGET_KERNEL="${CT_KERNEL}"

Can we sit on this for now, and revisit after the release?

Thank you!

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'

--
For unsubscribe information see http://sourceware.org/lists.html#faq