[Mesa-dev] [PATCH 05/18] mesa: android has no log2f nor ffs

[Chia-I Wu]

On Thu, Aug 18, 2011 at 3:25 PM, Jose Fonseca <jfonseca at vmware.com> wrote:
> ----- Original Message -----
>> On Wed, Aug 17, 2011 at 11:33 PM, Ian Romanick <idr at freedesktop.org>
>> wrote:
>> > -----BEGIN PGP SIGNED MESSAGE-----
>> > Hash: SHA1
>> >
>> > On 08/16/2011 11:28 PM, Chia-I Wu wrote:
>> >> Define log2f(v) to be log(v) / M_LN2 and ffs to __builtin_ffs.
>> >> ---
>> >>  src/mesa/main/imports.h |    8 +++++++-
>> >>  1 files changed, 7 insertions(+), 1 deletions(-)
>> >>
>> >> diff --git a/src/mesa/main/imports.h b/src/mesa/main/imports.h
>> >> index 3fa1db0..e404975 100644
>> >> --- a/src/mesa/main/imports.h
>> >> +++ b/src/mesa/main/imports.h
>> >> @@ -134,7 +134,13 @@ typedef union { GLfloat f; GLint i; }
>> >> fi_type;
>> >>  #define exp2f(f) ((float) exp2(f))
>> >>  #define floorf(f) ((float) floor(f))
>> >>  #define logf(f) ((float) log(f))
>> >> +
>> >> +#ifdef ANDROID
>> >> +#define log2f(f) ((float) (log(f) / M_LN2))
>
> Wouldn't (logf(f) * (float)(1.0 / M_LN2)) be more efficient (floating point only, no division) and equally precise?
That's true.  Thanks.
>> >> +#else
>> >>  #define log2f(f) ((float) log2(f))
>> >> +#endif
>> >> +
>> >
>> > This should use a HAVE_LOG2 and / or HAVE_LO2F feature macros
>> > instead.
>> Do you suggest we check for log2 in configure.ac and define
>> HAVE_LOG2?
>>  Looking at the code block and that follows, it looks like we are
>> doing platform or compiler workarounds (so that it works for
>> non-autoconf users?).
>
> I suppose we could have HAVE_LOG2 etc macros and define/undefine them inside #ifdef ANDRIOD/_WIN32/etc ... #endif, but I'n not sure it would simplify the code greatly for this particular case.
>
> configure.ac's defines would not automatically work for andriod/scons -- all the detection logic would have to be duplicated there too.
>
> Jose
>



-- 
olv at LunarG.com