SoundTrack: SECTOR / SERIOUS SPECCY GROUP'98
__________________________________________
(C) Kolotov Sergei, SerzhSoft, may, 1998.
__________________________________________
Again OPTIMIZATION
Great and Terrible
The leader of the Samara
Hacker M.M.A aka
UnBeliever, with a smile.
The fourth issue Deja Vu readers
could familiarize themselves with Article MMA, devoted
gap optimizatsii.Bes
k controversial, Maxim raised
J k is very good topic and
k quite interested in all
F k opisal.Napomnyu that it
k was to create protseduB k algebra fast scrolling
[Total screen up. K
l [article includes research
[Hodnik program in the form
D [a new tough material ac
[Sembler STORM.Deystvi
[Indeed, it is quite
And [the place, and most
Ji [important - _ochen_ bys
`[Try ASM. But ... Personally
to me he is not attracted.
Although this is a purely subjective opinion. Just
I'm not used to the "cobbled together" mnemonic, when the
command and operands are separated by only a single space ...
And so, the guys from X-TRADE just rock! I advise everyone to
try in kodit on STORM'e, it's cool! If
DARK and LD still would have done the normal mnemonic, I did
not hesitate for a long time to storm'il!
Because This assembly has not yet received a fairly wide
distribution, it certainly is not all managed to find the
source of the procedure MMASledovatelno, it would be good to
publish it in "normal" text that we now do ...
;-------------------------------------; ABSOLUTLY COOL LDI
SCROLL ; COMPLITLY CODED BY M.M.A
;-------------------------------------
ORG # 8000
CALL INSTAL; Building a table
; Display addresses
LD HL, # C000, DE, # 4000, BC, 6144
LDIR; The transfer to the screen
; Loaded at address # C000
; Pictures
;-------------------------------------; 192 times call the
procedure shifts ; The screen up one line
;------------------------------------;
LD B, 192
LOO PUSH BC
DI
CALL ONELINE
EI: HALT; can be removed to obtain uniform "twitches"
of strings.
POP BC
DJNZ LOO
RET; /
;------------------------------------
INSTAL LD HL, # 4000; This installer
LD B, 192; addresses can be
LOOP2 CALL DOWNHL; make faster
PUSH HL; but why?
METK3 LD (TABL), HL;
LD HL, (METK3 +1)
INC HL
INC HL
LD (METK3 +1), HL
POP HL
DJNZ LOOP2
RET
;-------------------------------------; Procedure itself upward
shift on the line ;-------------------------------------
ONELINE
LD (METK2 +1), SP; 20
LD BC, 6144 - # 20, 10
LD HL, # 4000; 10
LD (METK1 +1), HL; 16
LD HL, TABL; 10
LD (METK4 +1), HL; 16
METK4 LD SP, 0, 10 (B)
POP HL; 11 (C)
LD (METK4 +1), SP; 20 (E)
METK1 LD DE, 0, 10 (F)
LD (METK1 +1), HL; 16 (O)
.32 LDI; 16 * 32 ()
JP PE, METK4; 10 (578)
LD SP, # 5800, 10
.16 PUSH BC; 11 * 16
METK2 LD SP, 0, 10
RET; TOTAL 110,682 CTTA
; ++++++++++++++++++++++++++++++++++++
; + +
; + "DOWN HL" SUBROUTEN FROM MASM1.1 +
; + +
; ++++++++++++++++++++++++++++++++++++
;
; For a quick all transitions JR on the team
; RET replaced with actual RET NZ, RET C
DOWNHL INC H; 4
LD A, H; 4
AND 7, 7
RET NZ; 5 / 11
LD A, L; 4
ADD A, 32; 7
LD L, A; 4
RET C; 5 / 11
LD A, H; 4
SUB; 8 7
LD H, A; 4
RET; 10 TOTAL 55
TABL
And now some comments ... Carefully looking closer, you will
notice that bars here and there counted wrong ... First, the
team POP HL is not 11 but 10 tick! And secondly, this command
does not seem at all taken into account! Indeed, in the cycle
(where else it says "TOTAL - 578") total instruction execution
time is not 578 cycles, as indicated, and all 588! A total
run time is much
more marked: 82 +112308 +196 = 112586
tick! As much as much as 2,000 policy makers at the top!
So, check out a program, let's try to optimize it, or better
yet - write at all from scratch! Will not you long tormented by
their rhetoric see and understand:
;--------------------------------------;
; FULL SCREEN SCROLL UP;
; Coded by Kolotov Sergey;
; (C) SerzhSoft, Shadrinsk, may, 1998;
;--------------------------------------;
_NULL EQU 0
;--------------------------------------;
_DATA EQU # 6000
SCR_TBL EQU _DATA;, # 0300
DATAEND EQU SCR_TBL + # 0300
;--------------------------------------;
ORG # 8000
;--------------------------------------;
MAINPRG
EI
;
CALL MK_STBL
; CALL MKSRLUP
;
LD HL, # 0000
LD DE, # 4000
LD BC, # 1800
LDIR
;
LD B, # C0
LP_MAIN PUSH BC
HALT
DI
CALL SRL_UP
EI
POP BC
DJNZ LP_MAIN
;
RET
;--------------------------------------;
MK_STBL
LD HL, SCR_TBL
LD DE, # 4000
LD B, # C0
LP_MSTB LD (HL), E
INC HL
LD (HL), D
INC HL
;
INC D
LD A, D
AND # 07
JR NZ, $ +12
LD A, E
ADD A, # 20
LD E, A
JR C, $ +6
LD A, D
SUB # 08
LD D, A
;
LD (HL), E
INC HL
LD (HL), D
INC HL
DJNZ LP_MSTB
; RET
;--------------------------------------;
MKSRLUP
LD HL, SRL_UP
LD (HL), # ED; ld (...), sp
INC HL
LD (HL), # 73
INC HL
PUSH HL
INC HL
INC HL
LD (HL), # 01; ld bc, ...
INC HL
LD (HL), # E0
INC HL
LD (HL), # 17; ld bc, # 17E0
INC HL
LD (HL), # 31; ld sp, ...
INC HL
LD DE, SCR_TBL
LD (HL), E
INC HL
LD (HL), D; ld sp, SCR_TBL
INC HL
PUSH HL; lp_srup
LD (HL), # D1; pop de
INC HL
LD (HL), # E1; pop hl
INC HL
LD B, # 20
LP_MSU1 LD (HL), # ED; ldi
INC HL
LD (HL), # A0
INC HL
DJNZ LP_MSU1
LD (HL), # EA; jp pe, ...
INC HL
POP DE
LD (HL), E
INC HL
LD (HL), D; jp pe, lp_srup
INC HL
LD (HL), # 31; ld sp, ...
INC HL
LD (HL), B; # 00
INC HL
LD (HL), # 58; ld sp, # 5800
INC HL
LD B, # 10
LP_MSU2 LD (HL), # C5; push bc
INC HL
DJNZ LP_MSU2
LD (HL), # 31; ld sp, _NULL
INC HL
EX DE, HL
POP HL; ld (...), sp
LD (HL), E; ^ ^ ^
INC HL
LD (HL), D
EX DE, HL
INC HL
INC HL
LD (HL), # C9; ret
RET
;--------------------------------------;
_CODE EQU $
SRL_UP EQU _CODE;, # 0066
CODEEND EQU SRL_UP + # 0066
;--------------------------------------;
A couple of words about the program ... As you can see all
written in traditional style "realtime -
programming ", a feature which
is to create a real-time fast running, large-scale procedures.
In this case - is generated by high-speed
procedure shift the entire screen to the line
up (SRL_UP). For this procedure is called MKSRLUP, the
transmission control is in the segment of additional code, "we
get something like:
SRL_UP
LD (SP_SRUP +1), SP; 20
LD BC, # 17E0; 10
LD SP, SCR_TBL; 10
;
LP_SRUP POP DE; 10 (ALL)
POP HL; 10 (542)
.32 LDI; 16 * 32 (*)
JP PE, LP_SRUP; 10 (191)
;
LD SP, # 5800, 10
.16 PUSH BC; 11 * 16
;
SP_SRUP LD SP, _NULL; 10
RET
;
TOTAL: 40 + 103 522 + 196 = 103 758 cycles!
MK_STBL procedure creates a segment
data plate screen addresses SCR_TBL,
occupies 768 bytes.
Since the load of a screen is not
very convenient (and, still need to locate a suitable! :-)), it
was decided copy to screen the standard present in all
computers picture placed in ROM at address # 0000. :-) It's
called - "CHAOS", although some authorities insist on a
different title, namely "Broken TV! Well nevazhno.Glavnoe that
the effect of scrolling is very good visible and thus seen
quite well - exactly where the beam scanning the screen
"surpasses" our procedure ...
The program takes only 147 bytes!
If you are ready to donate more
amounts of memory, the speed can still improve ... Then the
program will change a bit:
...
;--------------------------------------;
LNS_NUM EQU # 40
;--------------------------------------;
...
;--------------------------------------;
MKSRLUP
LD HL, SRL_UP
LD (HL), # ED; ld (...), sp
INC HL
LD (HL), # 73
INC HL
PUSH HL
INC HL
INC HL
LD (HL), # 01; ld bc, ...
INC HL
LD (HL), # E0
INC HL
LD (HL), # 17; ld bc, # 17E0
INC HL
LD (HL), # 31; ld sp, ...
INC HL
LD DE, SCR_TBL
LD (HL), E
INC HL
LD (HL), D; ld sp, SCR_TBL
INC HL
LD (HL), # C3; jp ...
INC HL
LD D, H
LD E, L
LD BC, # 0042 2
ADD HL, BC
EX DE, HL
LD (HL), E
INC HL
LD (HL), D; jp lp_srup + # 0042
INC HL
PUSH HL; lp_srup
LD C, LNS_NUM
LP_MSU0 LD (HL), # D1; pop de
INC HL
LD (HL), # E1; pop hl
INC HL
LD B, # 20
LP_MSU1 LD (HL), # ED; ldi
INC HL
LD (HL), # A0
INC HL
DJNZ LP_MSU1
DEC C
JR NZ, LP_MSU0
LD (HL), # EA; jp pe, ...
INC HL
POP DE
LD (HL), E
INC HL
LD (HL), D; jp pe, lp_srup
INC HL
LD (HL), # 31; ld sp, ...
INC HL
LD (HL), B; # 00
INC HL
LD (HL), # 58; ld sp, # 5800
INC HL
LD B, # 10
LP_MSU2 LD (HL), # C5; push bc
INC HL
DJNZ LP_MSU2
LD (HL), # 31; ld sp, _NULL
INC HL
EX DE, HL
POP HL; ld (...), sp
LD (HL), E; ^ ^ ^
INC HL
LD (HL), D
EX DE, HL
INC HL
INC HL
LD (HL), # C9; ret
RET
;--------------------------------------;
LN_SRUP EQU LNS_NUM * # 0042 + # 0027
;--------------------------------------;
_CODE EQU $
SRL_UP EQU _CODE;, LN_SRUP
CODEEND EQU SRL_UP + LN_SRUP
;--------------------------------------;
As you see, has changed the procedure MKSRLUP,
which generates the now much more
volumetric procedure SRL_UP:
SRL_UP
LD (SP_SRUP +1), SP; 20
LD BC, # 17E0; 10
LD SP, SCR_TBL; 10
JP LP_SRUP + # 0042, 10
;
LP_SRUP POP DE; 10
POP HL; 10
.32 LDI; 16 * 32
...
POP DE; 10 TOTAL
POP HL; 10 LNS_NUM
.32 LDI; 16 * 32 times
... /
POP DE; 10 /
POP HL; 10 /
.32 LDI; 16 * 32 /
;
JP PE, LP_SRUP; 10
;
LD SP, # 5800, 10
.16 PUSH BC; 11 * 16
;
SP_SRUP LD SP, _NULL; 10
RET
Constant LNS_NUM determines the number of
lines of the screen, which are copied in one
iteration cycle. Thus, when LNS_NUM = 64 (# 40)
total time of the procedure is calculated as follows:
50 + ((532 * 64 +10) * (192/64) -532) + 196 =
= 246 + 101642 = 101888 cycles.
Thus, the win was in 1870
cycles, which, of course, not so much.
Rather, from this you have to understand themselves
principles of optimization of the program on time
Implementation and use them where necessary and
where there is no need!
Finally, I want to still see that
In any case, for the transfer of large amounts of data faster
than most satisfied program, working with the stack. Perhaps
The next article is really about and
talk. In the meantime, try to understand
yourself ... Good luck!
With best wishes
Serzh.
