Assembler in DBPro - GameCreators Forum

Author

Message

Sagan

23

Years of Service

User Offline

Joined: 2nd Oct 2002

Location:

Posted: 25th Apr 2003 22:16

Link

Is any way of including directly assembler code in a DB Program, like in C/C++:

asm

or

asm {
//
// assembler code
//
}

If not that will be a good thing to include in a new patch, and maybe a easy one, because will be great to program the core of the calculations, like AI and movements.

What the best way to include now assembler in the program, maybe a dll, in C++, or a dll directly in assembler? Anyway I don’t like dll, in other words I don’t know how to make them… Is easy to code assembler that make a dll… :-s

Back to top

Profile PM

Dazzy

23

Years of Service

User Offline

Joined: 26th Aug 2002

Location: United Kingdom

Posted: 25th Apr 2003 22:26

Link

I have to disagree with the fact that this would be a good addition, a nice one yes but considering the amount of people who would be capable of making use of it, DBS are better putting their energy into making DBpro stable.

Theres no way currently that I know off to include ASM directly in a DB program, however I believe it can be done via a dll!

Dazzy
Athlon 2600, 1gb pc2700 ram, 160gb HD, 128mb GeF FX, DVD RW, Audigy 2.

Back to top

Profile PM Email Website

Shadow Robert

23

Years of Service

User Offline

Joined: 22nd Sep 2002

Location: Hertfordshire, England

Posted: 25th Apr 2003 22:49

Link

yes you'd have to achieve this through a DLL, and it'd preferably need to be precompiled - however if you really ache for ASM coding use the Pixel & Vertex Shaders to achieve what you want.

Tsu'va Oni Ni Jyuuko Fiori Sei Tau!
One block follows the suit ... the whole suit of blocks is the path ... what have you found?

Back to top

Profile PM Email Website

Richard Davey

Retired Moderator

24

Years of Service

User Offline

Joined: 30th Apr 2002

Location: On the Jupiter Probe

Posted: 25th Apr 2003 22:54

Link

Create a C++ DLL with in-line assembler.

On a modern PC assembler is well over-rated IMHO, there are many C compilers which will compile to almost identical execution speeds. This ain't the 16-bit home computer world any more.

Cheers,

Rich

"Gentlemen, we are about to short-circuit the Universe!"

Back to top

Profile PM Email Website

mhack

23

Years of Service

User Offline

Joined: 24th Apr 2003

Location:

Posted: 25th Apr 2003 23:04

Link

I have to disagree

It ain't 16 bit assembler anymore

Assembler is faster because all other languages build of it

But I don't think it's needed in DB, no body wants to spend time on assembler if there is already tons of stuff done in DB!

DB team - Great Job!

Back to top

Profile PM

Shadow Robert

23

Years of Service

User Offline

Joined: 22nd Sep 2002

Location: Hertfordshire, England

Posted: 25th Apr 2003 23:23

Link

ASM for general tasks is a waste of time within C/C++ as it's already setup to do the exact same operations ... however there are still somethings where it is alot faster for execution and use, suchas using the Processor Extension directly and accessing data in a way that bypasses the system.

though i'd have to say there isn't really any call for this in DB/DBpro

Tsu'va Oni Ni Jyuuko Fiori Sei Tau!
One block follows the suit ... the whole suit of blocks is the path ... what have you found?

Back to top

Profile PM Email Website

Richard Davey

Retired Moderator

24

Years of Service

User Offline

Joined: 30th Apr 2002

Location: On the Jupiter Probe

Posted: 26th Apr 2003 00:07

Link

"Assembler is faster because all other languages build of it"

I still think this is a bit of an urban myth. Once upon a time I would agree, but there are some very advanced C compilers out there that will compile code to an equal speed as native asm.

Let me put it another way - how many of the games you like playing are written in assembler? Probably none. Do they suffer from speed problems?

Cheers,

Rich

"Gentlemen, we are about to short-circuit the Universe!"

Back to top

Profile PM Email Website

IanM

Retired Moderator

23

Years of Service

User Offline

Joined: 11th Sep 2002

Location: In my moon base

Posted: 26th Apr 2003 00:39

Link

There is almost no need to revert to assembly language if you use modern compilers - it takes an expert asm coder to outperform even a reasonable optimiser.

I sometimes switch on assembly language output when I'm compiling, but only to see what it's doing behind the scenes - sometimes you have to be careful with C++ to ensure that unnecessary constructions/copies are not being made.

Back to top

Profile PM Email Website

Sagan

23

Years of Service

User Offline

Joined: 2nd Oct 2002

Location:

Posted: 26th Apr 2003 20:32

Link

I agree that you must really have a good reason to write assembly language, but part of code that need speed or size smashed can be usefully code in asm.
If you try to do most things in assembler you end up with unfinished projects.

Assembly is fast that any compiler of any language could ever produce, and lot closer to machine level, all commands are mapped 1-1 to machine instructions.
Assembly code is smaller that any compiler of any language could ever produce.

Yap most programs are done in C++, but small parts of it can be done in asm, like the matrix calculations, encrypting…

I have done a small program in asm and DBPro to test speed:
The program just makes 3Billion of cycles with basic arithmetic

In asm it take in my computer ( P4 2400 ) 8 seconds, in DBPro it takes 50 seconds, that is about 6 times slow…

I show here the program in asm, with some other extra stuff, just to show some of you how the asm look like:

.MODEL TINY			; both code and data fit in the same 64K segment
    .STACK 200h			; space for stack

.DATA			; allows you to create a data segment

day     DB 20		; day=20, variable as byte ( 1 byte )
	money   DW 2000		; money=2000, variable as word ( 2 byte )
	count1  DW 50000	; count1=50000, variable as word
	count2  DW 60000	; count2=60000, variable as word
	score   DD 5000		; score=5000, variable as double  ( 4 bytes )
	my_arr  DW 50,23,11,3	; my_arr as word

.CODE			; allows you to create a code segment

START:
    MOV   AX, [money]		; AX=2000
    ADD   AX, 5			; AX=AX+5
    SUB   AX, 2			; AX=AX-2
    MOV	  BX, [my_arr+2] 	; BX=11

myloop1:
    MOV   CX, [count2]		; CX = 60000
myloop2:
    	ADD   DX, [money]	; DX=DX+money
    	ADD   DX, 1000		; DX=DX+1000
    	ADD   DX, 100		; DX=DX+100
    	SUB   DX, 1050		; DX=DX-1050
    LOOP  myloop2		; CX=CX-1 , if CX<>0 then jump myloop2

SUB   [count1], 1		; count1=count1-1
    CMP   [count1], 0		; compare count1 with 0
    JNE   myloop1		; jump if not equal

MOV   AX, 4C00h		;AX=4C00h (hexa format)
    INT   21h			;interrupt 21h, with AH=4C00h, exit program, back to DOS

END START

+ Code Snippet

    .MODEL TINY			; both code and data fit in the same 64K segment
    .STACK 200h			; space for stack

    .DATA			; allows you to create a data segment

	day     DB 20		; day=20, variable as byte ( 1 byte )
	money   DW 2000		; money=2000, variable as word ( 2 byte )
	count1  DW 50000	; count1=50000, variable as word
	count2  DW 60000	; count2=60000, variable as word
	score   DD 5000		; score=5000, variable as double  ( 4 bytes )
	my_arr  DW 50,23,11,3	; my_arr as word

    .CODE			; allows you to create a code segment

START:
    MOV   AX, [money]		; AX=2000
    ADD   AX, 5			; AX=AX+5
    SUB   AX, 2			; AX=AX-2
    MOV	  BX, [my_arr+2] 	; BX=11

myloop1:
    MOV   CX, [count2]		; CX = 60000
myloop2:
    	ADD   DX, [money]	; DX=DX+money
    	ADD   DX, 1000		; DX=DX+1000
    	ADD   DX, 100		; DX=DX+100
    	SUB   DX, 1050		; DX=DX-1050
    LOOP  myloop2		; CX=CX-1 , if CX<>0 then jump myloop2

    SUB   [count1], 1		; count1=count1-1
    CMP   [count1], 0		; compare count1 with 0
    JNE   myloop1		; jump if not equal

    MOV   AX, 4C00h		;AX=4C00h (hexa format)
    INT   21h			;interrupt 21h, with AH=4C00h, exit program, back to DOS

END START

A program that do the same thing in DBPro, I fist try with goto, and other cycles but it get to slow, so I have done with for/next.

+ Code Snippet


sync on; sync rate 0

For x=1 to 50000
For y=1 to 60000
      INC DX,money+1000+100-1050
Next y
Next x

print "END"
do
sync
loop

Back to top

Profile PM

Sagan

23

Years of Service

User Offline

Joined: 2nd Oct 2002

Location:

Posted: 26th Apr 2003 21:13

Link

Have done a small change in the ASM core code, now it runs in 4 sec.
12x faster that DBPro code

+ Code Snippet


    .MODEL TINY			; both code and data fit in the same 64K segment
    .STACK 200h			; space for stack
    .DATA			; allows you to create a data segment
    .CODE			; allows you to create a code segment

START:

    MOV   DX, 10000		; DX=10000
    MOV   CX, 60000		; BX=50000

loop1:

    XCHG   CX, BX			; exchange CX with BX
    MOV   CX, 50000			; CX = 50000
    loop2:
        ADD   AX, DX
	ADD   AX, 1000
	ADD   AX, 100
	SUB   AX, 1050
    LOOP  loop2				; CX=CX-1 , if CX<>0 then jump myloop

    XCHG   CX, BX			; exchange CX with BX
LOOP loop1		; CX=CX-1 , if CX<>0 then jump myloop

    MOV   AX, 4C00h		;AX=4C00h (hexa format)
    INT   21h			;interrupt 21h, with AH=4C00h, exit to system

END START

Back to top

Profile PM

MrTAToad

23

Years of Service

User Offline

Joined: 26th Aug 2002

Location: United Kingdom

Posted: 26th Apr 2003 21:49

Link

Its not that surprising really - DBPro is still using the debug libraries, which means there are all sorts of checks everywhere - parameter checks, stack checks etc...

Good news everyone! I really am THAT good...
http://www.nickk.nildram.co.uk/ for great plug-ins - oh my, yes!

Back to top

Profile PM Email Website

IanM

Retired Moderator

23

Years of Service

User Offline

Joined: 11th Sep 2002

Location: In my moon base

Posted: 26th Apr 2003 22:38 Edited at: 26th Apr 2003 22:40

Link

But you already have a way to add additional assembly language routines to DBPro if you choose to use it - you can create a DLL.

In addition, adding native assembly language to DBPro would:
1) Break existing code (eg I might have a variable called 'add')
2) Break the existing language (END and LOOP are already keywords)
3) Be of minimal use in speeding up games programs - because most of the time the program is using the DBPro libraries, and is not running your code, and the compilers *can* do it better.

I put the equivalent of your code through VisualC++7

+ Code Snippet

int FuncTest()
{

  int Count1=50000;
  int Count2=60000;
  int Money=2000;
  int Result=0;

  for (int x=0; x<Count1; ++x)
    for (int y=0; y<Count2; ++y)
      Result+=Money+1000+100-1050;

  return Result;
}

The result I got was fairly interesting :

+ Code Snippet

	mov	eax, -393167872				; e890bc00H
	ret	0

Basically, it detected that everything could be precalculated, and did so.

When I removed the 'return Result;' command, the function was optimised to the single machine instruction 'ret 0' - it worked out that there were no side effects in the code and optimised the whole thing away.

Back to top

Profile PM Email Website

Sagan

23

Years of Service

User Offline

Joined: 2nd Oct 2002

Location:

Posted: 26th Apr 2003 22:56

Link

Quite smart that VisualC++

Back to top

Profile PM

Sagan

23

Years of Service

User Offline

Joined: 2nd Oct 2002

Location:

Posted: 26th Apr 2003 23:09

Link

By the way anyone know how to build a dll, to use with DBPro in C++Builder 5, Enterprise Suite, Borland.
Or any good site with that information?

Back to top

Profile PM

Sorry your browser is not supported!

DarkBASIC Professional Discussion / Assembler in DBPro