Author Topic: ASM Optimized routines (Read 108272 times)

Galandros · « **on:** February 28, 2010, 07:27:53 am »

There are some cools optimized routines around. Calcmaniac is the recordist in z80, probably. At least in calculators z80 forums is.

On to the code:

Code: [Select]

;calcmaniac84
cpHLDE:
 or a
 sbc hl,de
 add hl,de
 ret
;Important note: because the code is 3 bytes and a call is 3 bytes, just macro in:
;SPASM, TASM and BRASS compatible, I guess
#define cp_HLDE  or a \ sbc hl,de \ add hl,de

;- Reverse a
;input:	Byte in A
;output:	Reversed byte in A
;destroys	B
;Clock cycles: 66
;Bytes: 18
;author: calcmaniac84
reversea:
	ld b,a
	rrca
	rrca
	xor b
	and %10101010
	xor b
	ld b,a
	rrca
	rrca
	rrca
	rrca
	xor b
	and %01100110
	xor b
	rrca
	ret

;reverse hl
;curiosity: a easy port of a common reverse A register is more efficient than tricky stuff
;calcmaniac84
;28 bytes and 104 cycles
ld a,l
rla
rr h
rla
rr h
rla
rr h
rla
rr h
rla
rr h
rla
rr h
rla
rr h
rla
rr h
rla
rrca
ld l,a
ret

;calc84maniac
;in: a = ABCDEFGH
;out: hl= AABBCCDDEEFFGGHH
rrca
rra
rra
ld l,a
rra
sra l
rla
rr l
sra l
rra
rr l
sra l

rrca
rra
rra
ld h,a
rra
sra h
rla
rr h
sra h
rra
rr h
sra h
ret

Code: [Select]

;Galandros optimized routines
;try to beat me... maybe is possible...

;Displays A register content on screen in decimal ASCII number, using no addition memory
DispA:
	ld	c,-100
	call	Na1
	ld	c,-10
	call	Na1
	ld	c,-1
Na1:	ld	b,'0'-1
Na2:	inc	b
	add	a,c
	jr	c,Na2
	sub	c		;works as add 100/10/1
	push af			;safer than ld c,a
	ld	a,b		;char is in b
	CALL	PUTCHAR		;plot a char. Replace with bcall(_PutC) or similar.
	pop af			;safer than ld a,c
	ret


;Note the following one is optimized for RPGs menus and the such, it is quite flexible. I am going to use in Lost Legends I ^^
;I started with one which used addition RAM for temporary storage (made by me, too), and optimized for size, speed and no extra memory use! ^.^
;the inc's and dec's were trick to debug -.-", the registers b and c are like counters and flags

;DispHL for games
;input: hl=num, d=row,e=col, c=number of algarisms to skip
;number of numbers' characters to display: 5 ; example: 65000
;output: hl displayed, with algarisms skiped and spaces for initial zeros
DispHL_games:
	inc c
	ld b,1		;skip 0 flag
	ld (CurRow),de
;Number in hl to decimal ASCII
;Thanks to z80 Bits
;inputs:	hl = number to ASCII
;example: hl=300 outputs '  300'
;destroys: af, hl, de used
	ld	de,-10000
	call	Num1
	ld	de,-1000
	call	Num1
	ld	de,-100
	call	Num1
	ld	e,-10
	call	Num1
	ld	e,-1
Num1:
	ld	a,'0'-1
Num2:	inc	a
	add	hl,de
	jr	c,Num2
	sbc	hl,de
	dec c			;c is skipping
	jr nz,skipnum
	inc c
	djnz notcharnumzero
	cp '0'
	jr nz,notcharnumzero
leadingzero:
	inc b
skipnum:
	ld a,' '
notcharnumzero:
	push bc
	call PUTCHAR  ;bcall(_PutC) works, not sure if it preserves bc
	pop bc
	ret

PUTCHAR:
	bcall(_PutC)
	ret

;Example usage of DispHL_games to understand what I mean
Test2:
	ld hl,60003
	ld de,$0101
	ld c,0
	call DispHL_games
	ld hl,60003
	ld de,$0102
	ld c,1
	call DispHL_games
	ret

Well, don't try to understand or optimize calcmaniac84 ones. j/k, trying to understand can be harsh (tip: have a good instruction set summary) but teaches some inner details of the z80 asm.
About mine, do your best.

Quigibo · « **Reply #1 on:** February 28, 2010, 05:21:57 pm »

Here is a little optimization I use but haven't really seen around. When you need a direct key press, you have to wait about 7 clock cycles between setting the port and reading it. Most people just fill in the extra space with a waste instruction like this:

Code: [Select]

ld	a,xx
out	(1),a
ld	a,(de)
in	a,(1)
and	yy

9 Bytes, 43 T-States.

You can actually use the waste instruction to do something useful. It gives a slight speed increase.

Code: [Select]

ld	a,xx
out	(1),a
ld	b,yy
in	a,(1)
and	b

9 Bytes, 40 T-States.

calc84maniac · « **Reply #2 on:** February 28, 2010, 08:12:27 pm »

Small and quick setup for IM 2 (this example sets up vector table at $9900 and interrupt jump at $9a9a, but values can be changed)

Code: [Select]

di
ld a,$99
ld bc,$0100
ld h,a
ld d,a
ld l,c
ld e,b
ld i,a
inc a
ld (hl),a
ldir
ld l,a
ld (hl),$c3
inc l
ld (hl),intvec & $ff
inc l
ld (hl),intvec >> 8
im 2
ei

Galandros · « **Reply #3 on:** April 24, 2010, 12:12:44 pm »

I found this optimized routine around. It is as far optimized as z80 string copy can get.

Code: [Select]

;author: calcmaniac84, I think
;Copy zero terminated string at HL to DE.
StrCopy:
	xor a
docopystr:
	cp (hl)
	ldi
	jr nz,docopystr
	ret

These are quite optimized. But may be is possible to optimize further. (speed and size) But it is not needed...
They shift a graphics buffer (optimized to 96x64) up or down by pixels passed in A register.

Code: [Select]

scroll_up:
#ifdef DEBUG
	cp 64+1
	call c,ErrorOverFlow
#endif
	add a,a
	add a,a
	ld l,a
	ld e,a
	ld h,0
	ld d,h
	add hl,hl
	add hl,de	; hl=a*12
	
	push hl
	ld de,768
	ex de,hl
; carry is never set here if input is correct
;	or a
	sbc hl,de
	ld b,h
	ld c,l		; bc=768-12*a
	ex de,hl
	ld de,plotsscreen
	add hl,de
	ldir
;blank remaining area
	ld h,d
	ld l,e
	inc de
	ld (hl),$00
	pop bc
	dec bc		; bc=12*a-1
	ldir
	ret
;PSEUDO CODE
;	ld hl,plotsscreen+12*a
;	ld de,plotsscreen
;	ld bc,768-12*a
;	ldir
;	ld h,d
;	ld l,e
;	ld (hl),$00
;	inc de
;	ld bc,12*a
;	dec bc
;	ldir
;	ret



scroll_down:
#ifdef DEBUG
	cp 64+1
	call c,ErrorOverFlow
#endif
; a can be from 1 to 63
; a can be multiplied by 4
	add a,a
	add a,a		; a*4
	ld l,a		; hl = a*4
	ld e,a
	xor a
	ld h,a
	ld d,a
	add hl,hl	; hl = a*8
	add hl,de	; hl = a*12
	ld e,a		; de = 0

	push hl		; a*12 will needed later
	push hl		; 2 times
	ex de,hl
;carry is never set here
;	or a
	sbc hl,de	; hl= -a*12, de=a*12
	ld de,plotsscreen+767
	add hl,de	; hl=plotsscreen+767-12*a
	pop bc
	push hl
	ld hl,768+1
;carry always set
;	or a
	sbc hl,bc
	ld b,h
	ld c,l
	pop hl
	lddr
;blank remaining area
	ld h,d
	ld l,e
	ld (hl),$00
	dec de
	pop bc
	dec bc
	lddr
	ret

;	ld hl,plotsscreen+767-12*a
;	ld de,plotsscreen+767
;	ld bc,768-12*a
;	lddr
;					or
;	ld (hl),$00		;;	ld hl,plotsscreen
;	ld h,d			;;	ld (hl),$00
;	ld l,e			;;	ld de,hl+1
;	dec de			;;	ld bc,12*a-1
;	ld bc,12*a-1	;;	ldir
;	lddr			;;	ret
;	ret

mapar007 · « **Reply #4 on:** April 25, 2010, 03:58:56 am »

Very nice! I'll add these to my utils.z80 file that is included in all my app builds.

Anyone wanting to compile a stdlib.c and revive the tisdcc project? j/k

Galandros · « **Reply #5 on:** April 25, 2010, 05:04:47 am »

Quote from: mapar007 on April 25, 2010, 03:58:56 am

Very nice! I'll add these to my utils.z80 file that is included in all my app builds.

Anyone wanting to compile a stdlib.c and revive the tisdcc project? j/k

Actually I am working on something like that. I am hand writing C functions in z80 assembly just for fun.

I will share them when I finish.
After seeing Axe Parser, it seems that is possible doing a good C compiler for z80. And we have documentation on how to optimize z80 assembly to do a optimizer, check the WikiTI topic: http://wikiti.brandonw.net/index.php?title=Z80_Optimization.

DJ Omnimaga · « **Reply #6 on:** April 25, 2010, 12:19:53 pm »

Quote from: mapar007 on April 25, 2010, 03:58:56 am

Very nice! I'll add these to my utils.z80 file that is included in all my app builds.

Anyone wanting to compile a stdlib.c and revive the tisdcc project? j/k

I think I remember this, it was Halifax from the old Omnimaga forums who worked on it, right? There was a thread about it somewhere

Quigibo · « **Reply #7 on:** April 29, 2010, 05:59:58 pm »

Quigibo's Challenge!

Can any of the following be done in 6 or fewer bytes? The input and output must be HL.

Multiply by 128?
Signed division by any nontrivial constant, other than 2, including negative numbers?
Modulus with any constant that is not a power of 2?

I'm rewriting my math engine almost from scratch so I decided I would just optimize everything I could possibly conceive of at the same time. These are the ones I'm having trouble finding.

calc84maniac · « **Reply #8 on:** April 29, 2010, 06:31:16 pm »

Seems pretty impossible to me.

Quigibo · « **Reply #9 on:** April 29, 2010, 06:58:39 pm »

Okay, that's good. I spent hours trying to optimize some of these using all the tricks I know. That reassures me it was a wild goose chase.

DJ Omnimaga · « **Reply #10 on:** April 29, 2010, 07:01:08 pm »

Quote from: calc84maniac on April 29, 2010, 06:31:16 pm

Seems pretty impossible to me.

No way!

You're calc84god, you can do everything, even the impossible! (see TI-Boy SE/Project M/F-Zero)

j/k I can't wait to see what kind of optimizations there will be in the next versions of Axe

Quigibo · « **Reply #11 on:** April 29, 2010, 07:34:45 pm »

It's nothing big. Mostly it just extend multiplication, modulus, and addition to higher powers of 2. The big optimizations won't come for a long time unfortunately. Functionality is more important right now.

By the way, is there a better way to display hl at the coordinates (xx,yy) than this?

Code: [Select]

B_CALL(_SetXXXXOP2)
B_CALL(_Op2ToOP1)
ld hl,$yyxx
ld (PenCol),hl
ld a,5
B_CALL(_DispOP1A)

Its seems really roundabout to me. Is there a bcall I don't know about that does this automatically?

calcdude84se · « **Reply #12 on:** April 29, 2010, 07:57:10 pm »

yeah, there's _DispHL
so you're code would be:

Code: [Select]

push hl
ld hl,$yyxx
ld (PenCol),hl
pop hl
B_CALL(_DispHL)

Just be aware it's right-justified in 5 spaces. (Since $ffff is 5 decimal digits, 65535)
EDIT: oh, wait, that's pencol? so this code doesn't work then. Oops...

calc84maniac · « **Reply #13 on:** April 29, 2010, 10:27:56 pm »

He's talking about graph screen display.

Galandros · « **Reply #14 on:** April 30, 2010, 09:21:30 am »

Quote from: Quigibo on April 29, 2010, 05:59:58 pm

Quigibo's Challenge!

Can any of the following be done in 6 or fewer bytes? The input and output must be HL.

Multiply by 128?
Signed division by any nontrivial constant, other than 2, including negative numbers?
Modulus with any constant that is not a power of 2?

Challenge accepted.

Answer to the multiplication by 128 in 6 bytes:

I started coding a routine that multiply A by 128:

Spoiler For Spoiler:

After that, I very easily modified to (hl*128)%((2^16)-1). Unsigned version:

Spoiler For Spoiler:

I am very sure this routines works but I have not tested.
EDIT4: tested with a few values, it works.

EDIT3:
Multiply hl by 128, now signed. If I am right, to do signed, you only need to preserve the bit 7? If that's so:

Spoiler For Spoiler:

Now I will think about the others when I have more free time. Fun, fun, fun.
Give me some time, please.

EDIT: I am thinking in putting some of this challenges in WikiTI when we end the challenge. And maybe Axe's routines. If you have other routines/challenges of optimization share to see what I can do.
EDIT2: fixed a bug/typo and commented even more the code

Author Topic: ASM Optimized routines (Read 108272 times)

Galandros

ASM Optimized routines

Quigibo

Re: ASM Optimized routines

calc84maniac

Re: ASM Optimized routines

Galandros

Re: ASM Optimized routines

mapar007

Re: ASM Optimized routines

Galandros

Re: ASM Optimized routines

DJ Omnimaga

Re: ASM Optimized routines

Quigibo

Re: ASM Optimized routines

calc84maniac

Re: ASM Optimized routines

Quigibo

Re: ASM Optimized routines

DJ Omnimaga

Re: ASM Optimized routines

Quigibo

Re: ASM Optimized routines

calcdude84se

Re: ASM Optimized routines

calc84maniac

Re: ASM Optimized routines

Galandros

Re: ASM Optimized routines