PIX >addr speedUP
ZXNet echo conference «code.zx»
Hello, All
There is a procedure
PIX_ADDR DE,HL (call pix_addr)
- YXpix DE > $address HL, A - bit offset
LD A,E ;B DE YX
AND A ;B HL $addr
RRA
SCF
RRA
AND A
RRA
XOR E
AND #F8
XOR E
LD H,A ;H
LD A,D
RLCA
RLCA
RLCA
XOR E
AND #C7
XOR E
RLCA
RLCA
LD L,A
LD A,D
AND 7
RET
In short, everything is as usual.
But now there are offers for acceleration and/or use of only 1 reg. those. in
in this example
- YXpix DE > $address HL
but it is necessary that
- YXpix DE > $addr DE
Only preferably with specifications. examples and measurements.
Hello, key-jee
Yeah. It’s clear, but it’s not a matter of ex de,hl, but rather a matter of working only with 1 reg. i.e. on
input to de -> manipulation with the same de and the result in the same de.
Table... hmm... how then will it all look as a whole and how much
will it give an increase?
Hello, Witch Doctor
You can create a plate for 2 sectors, where the addresses of all extreme points are stored, and
then use this simple procedure:
┌─- code ───
ld a,l
and 7
ld(end-1),a ; save offset
xor l
rrca
rrca
rrca
ld l,h
ld h,table
or(hl)
inc h
ld h,(hl)
ld l,a
ld a,0 ; restore offset
end;
└──; code ───
I haven’t checked in practice that I just came up with the idea and wrote it, but it seems like it should
work. in fact, the procedure can be optimized for 4 clock cycles and 1 byte, if
the input parameters will not be of the form H = y, L = x (as is usually customary in spec), but
vice versa, that is, H = x and L = y.
Well, if you need to work with de, then ex de,hl at the beginning and at the end of the procedure.
Hello, Witch Doctor
> Yep. Clear, but it’s not a matter of ex de, hl, but rather a matter of working with only 1 reg
> i.e. at the input to de -> manipulation with the same de and the result in the same
> de.
I don't understand what the problem is. :smile;
EX; DE,HL can be put as the first command in the procedure, and then before RET on
then the output will not spoil (although it is used)..
The speed increase is approximately 20 cycles per call..
The table generator is primitive, you can even use your procedure
:smile;
Yes; and elfh has a good option, only hl spoils it and the plate takes up kilobytes
memory (4 sectors) And if you don’t even need to draw a point (but only coordinates
find out), then the routine is faster by another 14 cycles, but you will also need a generator
cheat tables.. :wink;
Actually; It's a really good procedure.
Hello, key-jee
http://zx.pk.ru/showthread.php?t=2823
Similar topic.
Hello, Wlodek
Ehhhh... Ie. since the good old days of "dots in 73 cycles" nothing
changed/accelerated?
Hello, Witch Doctor
Well, why, in the same good old days, I used this procedure:
┌─- CODE ───
LD H,#80 ;7
LD A,(HL) ;7
INC H ;4
LD B,(HL) ;7
INC H ;4
LD L,C ;4
OR (HL) ;7
LD C,A ;4
INC H ;4
LD A,(BC) ;7
OR (HL) ;7
LD (BC),A ;7
└── CODE ───
Total 69 cycles.
If the table address is kept in register E, then the first line is replaced with: LD H,E
and we get an acceleration of 3 cycles
Hello, key-jee
key> EX DE,HL can be put as the first command in the procedure, and then again
key> before RET at the output then hl will not be damaged (although it is used)..
key> Speed increase is approximately 20 cycles per call..
Not really, after LD H, A ; H
Goes LD A, D
What will it be like to work with one register, even with D, even with H, if one
do not change the pair.
Here is the table!