Re: [ft-devel] Speeding up PNG loading

[Behdad Esfahbod]

Check the branch now:
https://github.com/behdad/freetype/commit/40112725e9041a5e3f2878e71bdd08249beaa743

I think this is good enough to go in.  What do you think?

On Mon, Aug 14, 2017 at 11:56 AM, Behdad Esfahbod <address@hidden> wrote:

On Mon, Aug 14, 2017 at 12:32 AM, Werner LEMBERG <address@hidden> wrote:

> I thought I see if I can speed up PNG loading by vectorizing alpha
> premultiplication, and it actually does give a nice speedup: [...]

Nice!

> The code is rather terse but readable.  I can add comments.

Not necessary, I think.

> Needs some GCC/clang checks, ...

Indeed, probably controlled by a configuration macro.

Or just:

#if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE__)

 

> ... as well as implementing the big-endian case (or disable it for
> big-endian).  I couldn't find any endianness macros in FreeType.

We don't have such macros; it should be easy to add, though.  On the
other hand: Do big-endian processors exist at all that support gcc's
(or clang's) `vector_size' extension?

Not that I know!

Looks like gcc defines this:
#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__

 

Are you going to work further on your patch towards upstream FreeType
integration?

I'll do the above and send an updated patch.

 

Thanks,

--

behdad
http://behdad.org/

--

behdad
http://behdad.org/