[GEM-dev] RGBA confusion...
IOhannes m zmoelnig
zmoelnig at iem.at
Wed May 28 18:54:38 CEST 2008
chris clepper wrote:
> The problem with the pix_ objects is the MMX code on is fixed to a
> different order than what OSX Intel uses. For some objects this makes
> no difference, but others require the components in a specific order.
> So to fix some bug complaints I swapped the byte order to match the MMX
> code, but this broke other things so it was switched back. Perhaps
> defining some MMX/SSE macros for the order would sort this out?
>
> I don't think there are any GL alpha problems at all.
the problem is most likely not with openGL.
however, the problem is not only with SIMD code, even after your changes.
a test of [pix_mask] revealed a lot of weirdnesses (but unfortunately i
don't have access to my test-patch right now, so i try to reconstruct it
by heart)
the tests have been done on an i686 (OSX-10.4.?)
i got different results when using [pix_image], [pix_film] and
[pix_video] as pix-source; i also get different results when using
[colorspace RGBA( vs [pix_rgba]
the results differ in both swapping color-channels _and_ having
different "types". iirc, images in RGBA space with type
GL_UNSIGNED_INT_8_8_8_8_REV where almost certain to yield blueish
results with alpha mask not working.
finally, [pix_mask] does _not_ involve any MMX/SSE2 code.
i am pretty sure that on macIntel images of type GL_BGRA_EXT are _not_
of type GL_UNSIGNED_INT_8_8_8_8_REV, is this correct?
anyhow, i thought before i start thinking about changing things and
making them worse, i would rather now what the target layout of the
colors is, that's why i started the wiki.
and yes, i also think that having some macros for the SSE/MMX code will
help us (but i am not that far yet...)
fgmadsr
IOhannes
More information about the GEM-dev
mailing list