Chunks 2x2
Poisoned CyberJack / TBK
New trend in demomaking
Greets to everybody, who reckon
himself a coder or some kind of
such! Today i'm gonna amaze you:)
The modern demomaking (Spectrum
of course) is based mainly on
chunk engines. And that's right -
other way it would be very difficuilt to realize some
interesting things. I'm not going to explain
anything, it was made sometime ago
before me. Moreover, the choice
has been already done.
Also, I don't want to explain what
is a 'chunk'. Try to find some information in any other sourses,
cos i'm trying to tell something
more important and such
'Remembrance' is out of my plans
here.
So, at the beginning i'd say, that
there are different kinds of
chunks. On Spectrum it's usually
4x4 (pixel - Refresh, Dogma, Inbetween and multicolour - Eye
Ache 1 / 2, Rage, Stellar Contour)
cos it's the most optimal ratio
between the screen resolution and
its dimensions. The 8x8 kind of
chunks, attribute (any demo of
95-97) and pixel-attribute (Emergency, 63Bit3, Smile_4K), due to
low resolution look poorer, cos
it's harder to make something good
in 32x24. By the way, here I mean
3D coding, cos it's the most difficuilt (and therefore, the
most interesting) coder's abilities application.
All this has been tested long time
ago (in 1997) and is boring almost
for everybody. Maybe there is an
alternative?
And there is a kind of alternative
and it's well-known for a long time too. Remember Higher State /
3SC, Over The Top / Enigma, Matricks / Baze? Yes, i'm talking
about 2x2 chunks.
Many coders, who are far from amateurs now, can smile
distrustfully. Of course, at first
sight there is not so much use of
such dimensions of chunks, cos
screen enlarges four times comparing with 4x4 (and therefore, a
frame will be drawn 4 times
slower) - and it's not so easy to
put such screen in memory. The use
of windowed mode is not an exit.
Besides, the main feature of this
chunks is five colours, and it's not
enough for illumination realization in 3D effects (Phong, Goraud
and even Flat looks awfully).
Alongside with all this drawbacks,
c2p is slower, cos reading has to
be more often than writing (again
comparing with 4x4).
But! I shouldn't pay waste your
time (and mine, which is much valuable), if I haven't realize
2x2 chunks in our demo, which we couldn't finish for CAFe'2002
and still we can't finish it: (((So,
here we go ...
The most easily solving thing problem is the lack of colours.
Of course it's dithering. In 2000 we made a c2p routine for
chunks 2x2 with Baze/3SC which he used in
his Matricks with 16 colour dithering (but his personal c2p is
slower). c2p is following:
sp - chunk screen
de - 1st line on spectrum screen (# 401F)
bc - 2nd line on spectrum screen (# 4120)
pop hl
ld a, (hl)
pop hl
ld l, (hl)
ld h, a
ldd
ld a, (hl)
ld (bc), a
summary: 10 +7 +10 +7 +4 +16 +7 +7 = 68tct
The main feature of this procedure
is LDD command that turnes the picture horizontally, cos SP
increases and BC and DE decrease. Chunks of% --- xxxx0 type.
For dithering it's necessary to put the gfx tables in different
memory banks and to work with a single screen. I
think there's no use to explain
it's all well-known:) If you have
some questions - email me.
As you can see, such speed is high
enough - the whole screen can be
drawn in three frames. But the problem
of a too large screen remains - so
that's not a full solution.
Experimenting with different engines and filters, a wonderful
thought has visited me:) What if we will process not a whole
screen, but just a part of it, and the
rest of it will fill applying, for
example, linear interpolation?
Then there will be no use in a
slowly working filter (ala
Sair00s) that helps to hide the
drawbacks of the engine - so we
get the benefit in speed and
screen usage. There's some similarity with Amiga gfx mode HAM8
(as Isee), where's the colour of a
group of pixels depends on the colour of nodal pixel - and the
latest demos work not in 640x480 or 320x240 resolution but much
lesser. As i'm not an Amiga freak
(At all), Icould say a nonsense -
it's not the point:) We're talking about Spectrum.
What I call a 'linear interpolation'? I calculate every second
chunk in horizontal direction, and
the rest of them are average between two certain.
For example:
Usually made:
! C0! C1! C2! C3! C4! C5! C6! C7! C8!
I make like:
! C0! -! C2! -! C4! -! C6! -! C8!
(C0 + c2) / 2 (c4 + c6) / 2
As a result, a picture is a little
blured, and I like that:) Anyway,
it a bit risky to trace the chunked screen, cos due to the low
resolution and simplified shaders the screen is usually ugly:)
In the last demos Sair00s acts like
this with two screens - but I could
use only one.
There's no benefit in speed in
using a separate operation for interpolation, cos it's not too
fast (Get two numbers from ram, then addition, then according
to the table recieving the result and then put
it into ram - about 40 tacts per
chunk). That's why I tried to combine interpolation with c2p:
sp - chunk screen
de - spectrum screen
h - previous chunk
pop bc
ld a, (bc)
ld l, a
ld a, (hl)
ld (de), a
inc h
inc d
ldi
dec d
ld h, b
summary: 10 +7 +4 +7 +7 +4 +4 +16 +4 +4 = 67tct
Mmmm ... i'll try to exaplain. To
draw a plot (8x2 pixels) we have
to consider three chunks:
! ! !
! - C0 - c1! - C2 - c3!
! ! !
^ ^ ^
So we have 16 tables (cos chunk
can be of different colour in previous plot), and graphics
adress can be calculated as (for the second plot):
c1 * # 0100 + c2 * # 0010 + c3
Chunks of% 1100 kind, i.e.
# E0-# FE. In the end of the sector
there's a table of low graphic adresses, i.e. from # E0E0, #
E2E0, etc. And the gfx is placed within
the rest of 32 sectors.
As you can see - gfx and multiplication table are placed on the
same adresses. I avoided it by excluding of two colours:) So i
have 14 colours instead of 16. But it's more comfortable to
work with 16 colours, that's why I combined 0 with 1 and 14
with 15 (the multiplication table has the same values).
However, it is possible to make real 16 colours, but it will
slow down c2p, so I've picked out
this method.
Dithering I make with two memory
banks with different chunked gfx -
for even and uneven lines. It uses
a lot of memory and works with one
screen:)))
Finally, what do we have? Comparing with chunks 4x4 screen
increased twice (not four), the fullscreen is traced in three
frames approx., we don't need additional
filter, the quality of the screen
is much higher i.e. we loose in
speed, but win in quality. This engine is quite efficient, and
we tried to proove it in our demo.
In Attachment you can find the sources of chunked gfx generator
and c2p with Strom syntax: ((And a
small engine demonstration.
respect to all spectrum's coders
and Adventurer's readers!!!
(C) Triebkraft
And here is an optimized version
of this engine, made by JtN/4d.
Read and think:)
pop hl; 10
l, (hl); 7
ld a, h; 4
exa; 4
ld h, a; 4
ldd; 16
ld a, (hl); 7
ld (bc), a; 7
; = 59t !!!!!!!!!!!
