Damien:
Thanks for catching this. The setting of CFLAGS was cruft and removing
the setting of CFLAGS completely from the spec file also works just
fine. I updated the spec-file in SVN head so we no longer set CFLAGS
in the spec-file.
Brian
> I don't understand this change:
> -export CFLAGS="%optflags -I%{_includedir}"
> +export CFLAGS="$CFLAGS -I%{_includedir}"
>
> It's not inside any %ifarch/%endif block so %optflags aren't used for either
> architecture.
> If some of the flags in %optflags are not compatible with gcc I could
> understand something like:
>
> %ifarch amd64 i386
> export CFLAGS="$CFLAGS -I%{_includedir}"
> %else
> export CFLAGS="%optflags -I%{_includedir}"
> %endif
>
> Is "-I%{_includedir}" necessary? Isn't it searched by default? (I know that
> it is not your addition)
>
> Damien
>
> ----- Original Message -----
> From: Brian Cameron <Brian.Cameron at Sun.COM>
> Date: Friday, February 9, 2007 5:25 am
> Subject: [jds-review] liboil improvements
> To: JDS-Review at opensolaris.org
>
>> The attached change to liboil causes it to be built with GCC when
>> building on x86 platforms, so that hardware acceleration support is
>> enabled. The new patch uses the getisax(2) kernel function so that
>> hardware acceleration support is determined using the correct Solaris
>> kernel interface.
>>
>> I have verified that liboil functions are 4-6 times faster using
>> GCC so that the GCC-style assembly code for hardware acceleration
>> gets compiled, and we don't just use the default C-implementations.
>>
>> Brian
>>
>>
>>
