Re: [Minios-devel] [UNIKRAFT/NEWLIB PATCH 2/2] Automatically select FLOAT_POINT on arm64 for newlib

[Felipe Huici <Felipe.Huici@xxxxxxxxx>]

Hi Julien,

On 23.12.19, 19:07, "Julien Grall" <julien@xxxxxxx> wrote:

    Hi,
    
    On 20/12/2019 05:54, Jia He wrote:
    > There are a few reasons that we can't support newlibc + nofp:
    > 1. -mgeneral-regs-only is not a compilable gcc option to newlib.
    > Even with -DNO_FLOAT_POINT, some source files of newlib will
    > generated float point instructions and then cause compilation errors
    > 
    > 2. with -mcpu=XXX,+nofp newlib can be built, but from the disassembly
    > result, newlib will still use q0-q31.
    > 
    > In a word, if you want to link with newlib on arm, float point should
    > be enabled in Unikraft.
    > 
    > Signed-off-by: Jia He <justin.he@xxxxxxx>
    > ---
    >   Config.uk | 1 +
    >   1 file changed, 1 insertion(+)
    > 
    > diff --git a/Config.uk b/Config.uk
    > index ab127f0..2110029 100644
    > --- a/Config.uk
    > +++ b/Config.uk
    > @@ -13,6 +13,7 @@ menuconfig LIBNEWLIBC
    >           select LIBVFSCORE
    >           select LIBPOSIX_PROCESS
    >           select LIBPOSIX_USER
    > + select FLOAT_POINT if ARCH_ARM_64
    
    I guess this depends on your floating point series? If so, then I would 
    highly recommend to write it down in the cover letter or after "---".

Yes, the plan is not to upstream this until the FP series is upstreamed.

-- Felipe
    
    Cheers,
    
    -- 
    Julien Grall
    

_______________________________________________
Minios-devel mailing list
Minios-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/minios-devel