Dynamic Clear Screen v2, completed....

Originally posted by shreyas_potnis
whenever I try to go to devart, internet explorer crashes.

Originally said by UnconeD
HOWEVER, and this is VERY important... you should realize by now that
if you
do this for every pixel, that the result won't be very fast. Doing 8
ANDs, 3
ORs, 3 additions and 3 divisions per color is not acceptable for lots
of
operations. That's why you should use MMX (multimedia extensions).
These are
special instructions in the CPU that allow you to do one operation on
several variables at once (perfect for working with the 3 color
channels
simultaneously). To use them, you must use assembly programming: this
is
when you use literal CPU instructions rather than a higher-level
language
(like C++ or Pascal) and is much harder.
With most C++ compilers, you can include pieces of assembly code inside
an
__asm-block, so you can leave your entire program in C++ except for a
few
small parts.

So here are some pieces of code for the additive and 50/50 blend mode
that
you can use. If you want to understand how they work, you will need to
learn
about assembly programming, and this can take a while (a good book is
'The
Art of Assembly Language Programming', but this doesn't cover MMX). You
can
use Google to find more info if you want (look for stuff related to
'mmx
instructions' and 'alpha-blending').

In this code, we assume that 'color1' and 'color2' are 2 32-bit
integers
representing the two pixels. The result is put back into color1.

/* Additive:*/
__asm {
movd mm0, color1
paddusb mm0, color2
movd color1, mm0
}


/* 50/50 */
__asm {
pxor mm7, mm7
movd mm0, color1
movd mm1, color2
punpcklbw mm0, mm7
punpcklbw mm1, mm7
paddusw mm0, mm1
psrlw mm0, 1
packuswb mm0, mm0
movd color1, mm0
}

Finally, at the end of any function that uses these MMX codes, you need
to
put the following code (before the return statement of course):
__asm {
emms
}

This is due to technical reasons and signals the end of MMX-enabled
code.
What's also important is that you cannot use any floating-point code
(this
means using real-numbers like 0.12345 or 3.1415, or functions like
sine/cosine/square-root) in combination with MMX. Between any piece of
MMX
code and floating point code, you must put an 'emms' block (see above).