Deja Vu #05 31 мая 1998

Программирование

CODING - Kodit want - Inference Procedure sprites without attributes, quick and versatile procedure for the withdrawal of sprites, the procedure for multiplication and calculate the square root.

SoundTrack: DEVOUR OF BRAIN BY DX-1969 1998
__________________________________________


(C) Card! Nal / PGC / BD
__________________________________________



   Hello, dear readers of our magazine!
Daniel then asked me to write an article
by coding'u, and I'm here for a text editor. This article has 
no definite purpose, I just talk about the optimization and

all that I have accumulated with the release of
number four Deja Vu.



   To begin with talk about the withdrawal of sprites
without attributes. That's standard procedure:


        LD HL, adr; address sprite

        LD DE, # 4000; address on the screen

        LD B, 192; height in pixels

        LD C, 32; width of familiarity
LOOP2 PUSH DE

        PUSH BC

        LD B, 0

        LDIR

        POP BC

        POP DE

        INC D

        LD A, D

        AND 7

        JP NZ, METKA1

        LD A, E

        ADD A, 32

        LD E, A

        JP C, METKA1

        LD A, D

        SUB 8

        LD D, A
METKA1 DJNZ LOOP2

        RET



   Now let's count how many
cycles it works. Without going into details, as I thought, I 
will say that It employs about 146,000 strokes in the 
derivation of the sprite-sized screen. On average, spends 23.7 
cycles per byte. Generally, it is good procedure for deriving a 
static picture, well, if you need a fast, generic procedure, 
then see the bottom listing. The word I'm Universal implied 
that one can arbitrarily set the height, width, sprite and its 
coordinates on the screen. 


        CALL DECRUN; this program makes

                    , Address table

                    ; Screen


        LD HL, adr; address sprite

        LD D, 3; coord. X in the familiarity

        LD E, 8; coord. Y in pixels

        LD B, 100; height in pixels

        LD C, 20; width of familiarity

WIWOD DI

        LD A, D

        LD (WIW1 +1), A

        LD A, C

        ADD A, A

        SUB 64

        NEG

        LD C, A

        LD A, B

        LD B, 0

        LD IX, WIW2

        ADD IX, BC

        EX DE, HL

        LD H, B

        ADD HL, HL

        LD BC, BUFER

        ADD HL, BC

        LD (STACK +1), SP

        LD SP, HL

        EX DE, HL

        LD B, A
LOOP1 POP DE

        LD A, E
WIW1 ADD A, 0

        LD E, A

        LD C, D

        JP (IX)
WIW2

        DUP 1932

        LDI; team LDI repeated

        EDUP; 32 times


        DJNZ LOOP1
STACK LD SP, 0

        EI

        RET


DECRUN LD HL, BUFER

        LD DE, # 4000

        LD B, 192
LOOP LD (HL), E

        INC HL

        LD (HL), D

        INC HL

        INC D

        LD A, D

        AND 7

        JR NZ, METKA

        LD A, E

        ADD A, 32

        LD E, A

        JR C, METKA

        LD A, D

        SUB 8

        LD D, A
METKA DJNZ LOOP

        RET
BUFER DEFS 384



   It works for 108,076 cycles
withdrawal of sprite the size of a screen that
1.35 fold faster than the previous one. On average,
spends 17.6 cycles per byte. Subroutine
DECRUN runs only once, and then you can use subroutine WIWOD
indefinitely. I can only say that it is necessary to prohibit 
the interruption, as yuzaetsya stack. Try yourself to find out 
how everything works. I just want what will prompt the command 
LD C, D before the team JP (IX). As known, LDI at work reduces 
the register-pair BC, and so, in order not to spoil Register B, 
we need to keep C register number is greater than the maximum

width of the sprite that is larger than 32, and register
D always contains a number greater than 32.


   Well all sorted out with the conclusion of sprites,
talk about turning a sprite at any angle
(0 - 360). In the fourth Deja Vu was the source Kolotova "TURN 
SPRITES". Honestly, I saw this program from earlier and I will 
say that the quality of the program me not very happy :-(. I 
tried turn-drenched square 5 by 5 familiarity

45 degrees, and saw a nasty picture, the square was already 
looks like not a square, and on the square lattice, and use this

program was difficult. And the thing
in the algorithm of rotation and accounting structure
screen. I implemented a slightly different algorithm and the 
result is simply stunning. The picture is almost (considering 
lower resolution screen) is not distorted. If, for example,

turn filled circle radius of 50 to 45
degrees around its center, then get
the same circle. True, this algorithm is not me
came up, and took it out of the magazine ZX NEWS 3.
There was a little article about turning sprites, and
was given to program in BASIC. Programm
This has worked for over an hour, turning the whole
screen. Analog in the code of this program is not
was, as the author confessed that he does not make friends with
BASIC built-in calculator. Honestly
speaking, I am also not friendly with built-in
calculator, and therefore implement this
algorithm in the code without using a calculator. The first 
version of my program worked 3.5 minutes, turning the whole 
screen is not it is important to any angle. I thought it was a 
full SUXX and slightly to optimize it. In finally got around 40 
seconds, which is about 5 times faster than the original 
version, and 100 times faster than BASIC, and this, mind you, 
without compromising image quality. Further optimize was lazy, 
though the prospect of time to bring up to 10-15 seconds, and 
then faster, but this joy I I can give you, SERZH, you 
understand me :-). The annex will find a source in format ALASM 
3.8c. Source contains comments, so it will be dismantled. Do 
not look to poor code, no time to lick it.

The origin is lower left
corner of the screen, if I'm not mistaken. Rotating a sprite 
(do not let yourself wither!) Should already be on your screen. 
The program uses a lot of memory (6144 + approx 2kg) bytes for 
the buffer, see there ... I almost forgot to tell the algorithm 
of rotation. Take the point with coordinates (x0, y0), rotated 
at an angle and the coordinates of (x1, y1), from coordinates 
(x1, y1) point is taken (if any) and transferred in the 
coordinate (x0, y0), then is taken next. (X0, y0), etc. Check 
out beyond screen, OF COZZZ, are made so that all nama :-).



   So what? Let's go on! I was there a
CRANK asked for help to understand the "arcane xor'kah", and 
therefore I will say this. Learn to understand the xor'kah you 
must own (Through STS v6.2, for example). Try

understand that it rasxorivaet, trace, followed closely by the 
registers, the resident STS'a endure, if necessary, to a safe 
place. The simplest xor'ka looks as follows:



        LD HL, adr

        LD BC, lenght
LOOP LD A, (HL)

        XOR 23

        LD (HL), A

        INC HL

        DEC BC

        LD A, B

        OR C

        JR NZ, LOOP

        RET



   If you have a pC, where you can dial
program in assembly language or in hiew (if you
perverse :-)).


        mov dx, adr

        mov bx, lenght
loop mov si, dx

        mov al, [si]

        xor al, 1923

        mov [si], al

        inc dx

        dec bx

        jnz loop

        retn



   Of course I do not care where kodit, but
SPECCY, OF COZZZ, RULEZ FOREVER!



   With xor'kami like figured out how to
Spectrum, and on PISI, we have decided
more :-). I would like to say a few
words about switching pages. For example, what I saw that 
program from SERZH'a in one of his ruleznyh (as always) stateek:



        LD BC, # 7FFD

        OUT (C), A

        LD (BANK), A

        RET
BANK DEFB 0



   Or something like that. What do we see? It seems everything 
is correct ... But no! Consider a situation that interrupts the 
teams came out (c), a and ld (bank), a. Interrupts will work 
out and come back. But on exit to restore the old value

BANKi and eventually colored squares on the screen, and 
therefore need to preserve the value of front out (c), a.




   Because Deja Vu will now go out without
protection, I would like to talk about classroom
method of copy protection invented
I (drive can not be copied even in pC
and Amiga), and even wanted to throw in the source
application (formatter and reader format), but
then I thought that would be too bold! AND
abandoned the idea. Although, if I
well be asked, then maybe tell you.
Now we have gone further.


   I want to say a few words about the game Black
Raven, cool I must say igruha. But some levels are simply 
suicidal (I because of the almost impenetrable), but since I do 
not gamer toys are not fond helluva lot, so make your life 
easier, put a couple of POKES and saw FINAL CUT! If zero

three bytes from the address # D5CE in RAM0, then you
will no longer run out of money. Well, a number
current level is located at # 79D8.
For example, enter the next room there
level, do the "retry" and you
on another level. If there Record
# 10, see FINAL CUT. "Everything's fine," you might say - "But 
what about protection?" Here You're right, even the shadow 
businesses will not help. And so the twist yourself, break your 
own. I have no moral right to teach you breaking commercial 
products. I will only say, I spent three days to insert POKES! 


   Next, let's talk about the multiplication algorithm.
The simplest procedure for multiplication SPECCY
looks like this: HL = A * E



        LD H, A

        LD L, 0

        LD D, L

        LD B, 8
LOOP ADD HL, HL

        JR NC, $ +4

        ADD HL, DE

        DJNZ LOOP

        RET



   Not difficult to count the number of cycles:
at least 310, a maximum of 358 without RET.
Of course, if you expand the loop, we obtain
at least 206, and a maximum of 254 bar. A possible
Does faster? Yes, of cozzz! But for this
to remember the initial algebra. A more precise formulas of 
abridged multiplication: 


    The square sum of: (a + b) ^ 2 = a ^ 2 +2 ab + b ^ 2

    Square difference: (ab) ^ 2 = a ^ 2-2ab + b ^ 2



   Take these formulas into service. For
In order to write the procedure of multiplication,
need to form a plate of squares of numbers from 0 to 255, 
inclusive. Table squares will be the main part of the program.

Address table must be a multiple of 256, ie
Low byte = 0. This makes the following
programm.

SET_TBL LD DE, TABL

        LD H, E

        LD L, E

        LD B, E

        LD C, E
SET_T1 LD A, L

        LD (DE), A

        INC D

        LD A, H

        LD (DE), A

        DEC D

        ADD HL, BC

        INC C

        ADD HL, BC

        INC E

        JR NZ, SET_T1

        RET
TABL DEFS 512



   And here is the procedure for multiplication. Numbers
from 0 to 255 are set in the registers L and E.
Response in HL.


MULS LD H, 'TABL

        LD C, (HL)

        INC H

        LD B, (HL)

        LD A, L

        ADD A, E

        JP NC, MUL4

        JP M, MUL1

        SUB E

        SUB E

        JP NC, MUL2

        LD A, E

        SUB L
MUL2 LD L, E

        LD D, (HL)

        DEC H

        LD E, (HL)

        LD L, A

        LD A, (HL)

        INC H

        LD H, (HL)

        LD L, A

        SBC HL, BC

        OR A

        SBC HL, DE

        LD A, L

        CPL

        LD L, A

        LD A, H

        CPL

        LD H, A

        INC HL

        SRL H

        RR L

        RET
MUL4 LD L, E

        LD D, (HL)

        DEC H

        LD E, (HL)

        LD L, A

        LD A, (HL)

        INC H

        LD H, (HL)

        LD L, A

        SBC HL, BC

        OR A

        SBC HL, DE

        SRL H

        RR L

        RET
MUL1 LD A, L

        SUB E

        JP NC, MUL3

        LD A, E

        SUB L
MUL3 LD L, E

        LD D, (HL)

        DEC H

        LD E, (HL)

        LD L, A

        LD A, (HL)

        INC H

        LD H, (HL)

        LD L, A

        SBC HL, BC

        OR A

        SBC HL, DE

        SRL H

        RR L

        LD A, L

        CPL

        LD L, A

        LD A, H

        CPL

        LD H, A

        INC HL

        RET



   As a result, the number of cycles at least 141,
a maximum of 207, which will agree, a little bit.
This algorithm was borrowed from the game
BATTLE COMMAND 128. Also write about this in the electronic 
journal of Spectrum Expert 1. However, there are a few other 
programs that are marks when multiplied, and there was limiting 
the numbers of -128 To +127, which is not always convenient.


   Now let's talk about the calculation
square root. In ZX FORMATE 7 was some kind of programm for 
calculating the roots, but I just say that not very much. And 
the thing speed. I poprobyval computed using

of the programs from the root of 1024. Potrassiroval STS'om and 
found that program 5 (Five) times does the division, and in 
calculating the square root of 65,535 makes 10 (desyat!) 
divisions. Moreover, the procedure division holds about 1000 
cycles, and even more. I then thought for a moment, and wrote 
his own. Its listing is given below. My program does only one 
division, if the number falls in interval from 2 to 16383, and 
holds two division, if the number in the range of 16,384 -

-65535. When the numbers 0 and 1 division are not met. The 
program runs on a modified Newton algorithm is me. Try 
themselves to understand how everything works. Dam hint: in the 
table are so-called secondary roots of numbers from 1 to 2 ^ 2 
^ 15. Calculated by the formula:


(Sqr (2 ^ n) + sqr (2 ^ (n +1))) / 2


SQRT LD A, H; HL = SQR (HL)

        OR L

        RET Z

        LD D, H

        LD E, L

        LD B, 15

        ADD HL, HL

        JR C, SQRT2

        DEC B

        ADD HL, HL

        JR C, SQRT2

        DEC B
SQRTL ADD HL, HL

        JR C, SQRT1

        DJNZ SQRTL

        LD HL, 1

        RET
SQRT1 LD HL, TABL

        LD C, B

        LD B, 0

        ADD HL, BC

        LD C, (HL)
SQRT_ PUSH BC

        CALL DIV

        POP BC

        ADD HL, BC

        SRL H

        RR L

        RET
SQRT2 PUSH DE

        CALL SQRT1

        POP DE

        LD B, H

        LD C, L

        JR SQRT_


DIV PUSH BC; HL = DE / BC to rounding

        LD A, B; response

        CPL

        LD B, A

        LD A, C

        CPL

        LD C, A

        INC BC

        LD HL, 0

        LD A, E

        ADD A, A

        RL D
DUP 1916

        ADC HL, HL

        ADD HL, BC

        JR C, $ +4

        SBC HL, BC

        RLA

        RL D
EDUP

        LD E, A

        POP BC

        SRL B

        RR C

        OR A

        SBC HL, BC

        EX DE, HL

        RET C

        INC HL

        RET
TABL DEFB 1,2,2,3,5,7,10,14,19,27

        DEFB 39,55,77,109,155,219



   Run time, I did not think, but it
amounts to approximately 1500 - 2500 cycles.
The program can be accelerated if a more rapid procedure 
division. But You can do without division, if

use a different algorithm. I have
to accept this, but there is not enough
free time to investigate it thoroughly and write a program :-(


   There my story ends. I can only say that the listings used 
assembler directives ALASM. For example:


DUP 1916

      body program
EDUP
means that the body of the program assembled 16 times.


   Finally I would like to inform you that I want to prepare 
for the second number, a long article on 3D-graphics on SPECCY. 
It will be considered: the formula turning points in space, the 
algorithm of drawing the cast of the triangle decomposition, 
the algorithm to hide invisible faces, the algorithm for 
constructing complex nonconvex object known, algorithms fill 
the faces depending on the location of the light source and 
something else. Naturally we give examples in assembler. Was 
have the desire and free time. On this until everything. BYE.