Optimal shifting for set pixel?

tstih · May 2014

Hi,

I'm optimizing my set pixel code. For speed. But without using partitioned memory (i.e. a table on 256 byte boundary) I have optimal memory address position code and am now trying to optimize the shifting part.

Here's what I came up with so far.

		ld	a,#x		; symbolic. a is x (0->255)
		and	#7		; remainder of division by 8 (0->7)
		inc	a		; avoid special condition without shift / i.e. jr z
		ld	b,a		; our shift counter
		xor	a		; a=0
		scf			; our shift bit
loop:		rra			; to a
		djnz	loop		; until apropriate position

Is there a more optimal approach to creating a mask for shifting then this?

T.

Kayamon · May 2014

Using a lookup table is probably best, but if you really wanted to do it without, you could use self-modifying code.

Have a SET 0,(HL) instruction, and modify the last byte from C6 to be (C6 | (x&7)<<3), then execute that.

(disclaimer: haven't tried it)

catmeows · May 2014

What kayamon said + table can be at any place ( ld hl, tab \ ld d, 0 \ ld e, entry \ add hl, de \ ld a, (hl))
Also, this could work:
ld (jump+1), a \ ld a, 1 \ jump: jr 0 \ rla \ rla \ rla \ rla \ rla \ rla \ rla

jamorski · May 2014

Something like this?

Pixel Set
---------

	LD A,#x	(#00-FF)
	AND #07
	INC A
	RLA
	RLA
	RLA
	ADD A,#C6
	LD (NN+1),A
NN	SET N,(HL)

Dr BEEP · May 2014

        ld de,table
        and 7
        add a,e
        ld e,a
        ld a,(de)     
        ld (setn+1),a
setn    set 0,(hl)
        ret
table   defb #c6,#ce,#d6,#de,#e6,#ed,#d6,#de

Edit: looking previous code, tablevalues might be incorrect

Edit2:
As mentioned above C6 and further are correct values.
When table at #nn00:

        ld d,tablehighbyte
        and 7
        ld e,a
        ld a,(de)     
        ld (setn+1),a
setn    set 0,(hl)
        ret
table   defb #c6,#ce,#d6,#de,#e6,#ed,#d6,#de

Dr BEEP · May 2014

jamorski wrote: »

Something like this?

Pixel Set
---------
 
    LD A,#x    (#00-FF)
    AND #07
;    INC A
    RLA
    RLA
    RLA
    ADD A,#C6+8
    LD (NN+1),A
NN    SET N,(HL)

Like this?

catmeows · May 2014

Dr BEEP wrote: »

Like this?

Intetesting, computed version beats LUT.

joefish · May 2014

Surely if you're using the last 3 bits of the X-value to set the bit in a byte then aren't you counting them in reverse order to how they appear on the screen?

Jimmy Falcon · May 2014

LD A,#x;
AND #7;
LD H, table/256; # locate table in such manner that L=0
LD L,A;
LD A,(HL);

table defb #80,#40,#20,#10,#08,#04,#02,#01;

It's untested, but you should get the idea.

g0blinish · May 2014

Jimmy Falcon wrote: »

LD A,#x;
AND #7;
LD H, table/256; # locate table in such manner that L=0
LD L,A;
LD A,(HL);

table defb #80,#40,#20,#10,#08,#04,#02,#01;

It's untested, but you should get the idea.

easy way is fill 256 bytes with the same sequence, so code looks like:

ld h,bw/256
ld l,X; X is value
ld a,(hl)
bw:db #80,#40,#20,#10,#08,#04,#02,#01 ;32 times

Jimmy Falcon · May 2014

g0blinish, X is a passed value, not a constant, because it's a pixel set routine and the rest of X is used to calculate display buffer memory offset, it's going to be used anyway. Repeating byte sequence is a waste of memory, IMO...

If it weren't a passed value, we could do:

LD A,(bw+X);
bw:db #80,#40,#20,#10,#08,#04,#02,#01 ;32 times

g0blinish · May 2014

Jimmy Falcon wrote: »

g0blinish, X is a passed value, not a constant, because it's a pixel set routine and the rest of X is used to calculate display buffer memory offset, it's going to be used anyway. Repeating byte sequence is a waste of memory, IMO...

If it weren't a passed value, we could do:

LD A,(bw+X);
bw:db #80,#40,#20,#10,#08,#04,#02,#01 ;32 times

routine by Busy:

;DE=Y.X
PLOT    PUSH HL,BC:LD H,HIGH PLOTTBL,L,D,B,(HL):INC H
        LD A,(HL),L,E:INC H:OR (HL)
        INC H:LD C,A,A,(BC)
        OR (HL):LD (BC),A
        POP BC,HL:RET

of course, routine ain't optimized.

to use plot some data must be prepared(1024 bytes):

PLOTTBL EQU #A000

  call FORMER

FORMER  LD DE,#4000,BC,#8000,L,E
FLP1    LD H,high PLOTTBL
        LD (HL),D:INC H:LD (HL),E:INC H
        LD (HL),C:INC H:LD (HL),B
        RRC B
        LD A,C:ADC A,0:LD C,A
FBR1    INC D:LD A,D:AND 7
        JR NZ,FNXT:LD A,E:ADD A,32
        LD E,A:JR C,FNXT
        LD A,D:SUB 8:LD D,A
FNXT    INC L:JR NZ,FLP1
        LD HL,PLOTTBL+#C0,BC,#3F
        LD DE,HL:INC E
        LD (HL),0:LDIR
        RET

tstih · May 2014

Self modifying code is bad karma /not ROMable/. But the jump can still be made (if placing shift table on 256-byte boundary) by using JP (HL) I guess.

LD H,$HIGH(SHIFTS)
LD L,#nrshifts
JP (HL)
SHIFTS:
...
8 shift instructions in the other direction
...

Optimal shifting for set pixel?

Comments