Re: [pygame] Benchmarking alpha blits (was Re: How do I make a image transparent?)

[Casey Duncan <casey@xxxxxxxxxxx>]

One thing to note: I think RLEACCEL is highly dependent on the  
content of the image, and the name implies "run-length encoding"  
which implies that images with large areas that are the same color  
would benefit most.

I think it would be useful to run these tests "scientifically" with  
various test images such as:

- Image containing random color/alpha noise
- Image that is a block of one color, with no alpha
- Image that is a block of one color with a single alpha value
- Image that is a block of one color with random alpha values
- Image containing a circle (2 colors) with no alpha
- Image containing a circle with alpha blending at the edges
- Image containing a circle of random colors with alpha blending at  
the edges and a single color surrounding it with alpha 0

And test these images at various sizes.

-Casey


On Sep 16, 2007, at 10:12 AM, Ethan Glasser-Camp wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Ethan Glasser-Camp wrote:
> > Content-Type: image/svg+xml;
>
> I have no idea what happened here. I sent a follow-up email to try to
> rephrase, but it hasn't gone through yet. Maybe it needs to be
> moderated; it's 107 KB. I've decided to take the attachments and put
> them online instead of attaching them. They're at:
>
> http://betacantrips.com/attic/pygame-blits/pygame-blits.csv
>
> http://betacantrips.com/attic/pygame-blits/pygame-plot-graph1.png
>
> http://betacantrips.com/attic/pygame-blits/pygame-plot-graph2.png
>
> Here's the text:
>
>
> Yikes, what happened here?
>
> As I was trying to say:
>
> I took the liberty of converting your numbers into CSV and
> constructing a few plots. No hard-and-fast conclusions, especially
> without some idea of the alpha values in the images, but:
>
> 1) without RLE, colorkey can be one of the slowest things you can do
> - -- possibly even slower than alpha?
>
> 2) a colorkey that isn't in your image is worse than a colorkey that
> is -- presumably the same number of checks must be performed, but more
> pixels are blitted because none match the colorkey.
>
> 3) sometimes alpha with RLE is slower than alpha without RLE (the last
> image). I would guess this is due to a widely varying number of alpha
> values, for which RLE doesn't cut down on memory use but inflates it?
>
> 4) sometimes colorkey with RLE is much faster than alpha (a factor of
> 2-4x on some images).
>
> Thanks Casey for these numbers! And also thanks for the code, using
> "from __main__ import" is very clever and I'll have to use it in the
> future.
>
> Ethan
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.6 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
>
> iD8DBQFG7WP/hRlgoLPrRPwRAjrfAKDUbSvErLZscgfjnqEsWPimyGBOuQCfW66f
> VYFXKf2hEEHtZR4Hn5v4XHE=
> =XL52
> -----END PGP SIGNATURE-----