Re: Question for Herbert


Frank answerd Wannabee :

... before he fucked up his own thread... :)
:)

I have a question for you Herbert.

Frankly, I'm not Herbert... and as you know, I don't know too much about
"gofast"....

I'm not either Herbert...

Why is the following code slower to draw a vertical line
than the Window API,

'Cause the card manufacturer tells Uncle Bill how to do hardware
accelleration, and won't tell you?

That's it! ...even not the only cause
The cards features are only known by the (dX)drivers.

... (A memory backbuffer) (on a Athlon XP 1800+)

In opposition to windoze, I can direct access the video RAM and have
my back-buffer within the non cached non paged graphic cards memory.
This is almost as fast as the AGP BUS allowes.
ie: ~35 cycles per dot for (rare used) vertical image lines,
8 cycles per dot for horizontal lines including clip check
and 'skip if colour' (on my old 500MHz K7 with 66 MHz AGP).

But if I had info about the accellerator functions for line draw,
this could speed it up by a certain factor.

[code ...]

"sub ecx, 1" might be faster on some CPUs - not much, I don't think.

Not on AMDs as far as I could check this yet.

"The compiler can't win! If the compiler's code is faster, we use the
compiler's code. If our code is faster, we use our code. The best the
compiler can do is break even." Might apply to the API, too? I dunno...

I'm quite sure that windoze could be improved in all performance aspects.
But right without the needed info we are bound to what we've got.

__
wolfgang


.