Index
> Main > Bresenham's linedrawing algorithm in ASM [SOLVED]Goto page Previous 1, 2 |
| Author |
|
|
Madis731 03 Aug 2006, 11:26
AAAAAAAAAAAAAAAAHHHHHHHHHHH
 so sweeet http://enos.itcollege.ee/~mkalme/PAHN/MenuetOS/untitled.PNG
every SINGLE blue line is in its place (A) nothing's off... and I think you want to know what was wrong: Code:
lea rdx,[rix+rdi]
test rdx,rdx
jne @f
neg rdi
neg rix
@@:
This was missing from the part where dx>dy I had to put a name to the jump because the new code introduced a jump Look why @@s are bad
Code:
cmp rex,rfx
jc .ver2
xor rsi,rsi ; xinc1 = 0
xor rjx,rjx ; yinc2 = 0
;=>RIGHT HERE
mov rkx,rex ; den = deltax
mov rlx,rex
shr rlx,1 ; num = deltax/2
mov rbp,rfx ; numadd = deltay
mov rdx,rex ; numpixels = deltax
jmp .endloc
.ver2:
THANKs for eberybody who contributed! You got me closer to the answer and finally its here...now I'll try to make this code fragment more transparent and fit it somewhere in the code...meaning - I will optimize a bit  Last edited by Madis731 on 03 Aug 2006, 11:37; edited 2 times in total |
|||
03 Aug 2006, 11:26 |
|
|
shoorick 03 Aug 2006, 11:33
 |
|||
|
03 Aug 2006, 11:33 |
|
|
Madis731 03 Aug 2006, 11:59
ARGH you won't believe how angry I am with MYSELF !!!!
I wrote down RSI, RDI for Xs and RIX, RJX for Ys, but EVERYWHERE else I used them as RSI, RIX for Xs and RDI, RJX for Ys. I wondered why this fragment of code fitted everywhere and then I finally just optimized it to: Code:
xchg rdi,rix
C'mon - everyone can only DREAM about this kind of optimization. THEN I remembered what Shoorick said and Voila! there it was - rsi and rdi standing as if they where Xs 
Okay - the WORKING unoptimized, but small and fast code
Code:
mov rex,rcx ;rbx=x2
sub rex,rax ;rax=x1
sbb rhx,rhx
xor rex,rhx
sub rex,rhx ;rex=abs(rcx-rax)
mov rfx,rdx ;rdx=y2
sub rfx,rbx ;rcx=y1
sbb rhx,rhx
xor rfx,rhx
sub rfx,rhx ;rfx=abs(rdx-rbx)
or rsi,-1
or rix,-1
cmp rcx,rax
jc .X2notlessthanX1
neg rsi
neg rix
.X2notlessthanX1:
or rdi,-1
or rjx,-1
cmp rdx,rbx
jc .Y2notlessthanY1
neg rdi
neg rjx
.Y2notlessthanY1:
cmp rex,rfx
jc .moreYthanX
xor rsi,rsi ; xinc1 = 0
xor rjx,rjx ; yinc2 = 0
mov rkx,rex ; den = deltax
mov rlx,rex
shr rlx,1 ; num = deltax/2
mov rbp,rfx ; numadd = deltay
mov rdx,rex ; numpixels = deltax
jmp .endadjust
.moreYthanX:
xor rdi,rdi ; xinc2 = 0
xor rix,rix ; yinc1 = 0
mov rkx,rfx ; den=deltay
mov rlx,rfx
shr rlx,1 ; num=deltay/2
mov rbp,rex ; numadd = deltax
mov rdx,rfx ; numpixels = deltay
.endadjust:
.linedraw:
push rax rbx rcx rdx rex rfx rsi rdi rix rjx
mov rex,rbx
add rex,1
mov rdx,rax
mov rcx,rbx
mov rbx,rax
mov rax,38
mov rfx,255
int 60h
pop rjx rix rdi rsi rfx rex rdx rcx rbx rax
add rlx,rbp ; num += numadd
cmp rlx,rkx ; num >= den
jc @f
sub rlx,rkx ; num -= den
add rax,rsi
add rbx,rdi
@@:
add rax,rix
add rbx,rjx
sub rdx,1
jnc .linedraw
Talk about Optimized to death (http://board.flatassembler.net/topic.php?t=5584) |
|||
|
03 Aug 2006, 11:59 |
|
|
r22 03 Aug 2006, 21:51
Slightly faster absolute value using compare and move instruction
Code: mov reg2, reg1 neg reg1 cmovl reg1, reg2 ;;;; cmovb if using UNsigned integers |
|||
|
03 Aug 2006, 21:51 |
|
|
Tomasz Grysztar 04 Aug 2006, 04:04
r22 wrote:
But the absolute value doesn't make much sense with the UNsinged integers, does it? |
|||
|
04 Aug 2006, 04:04 |
|
|
Madis731 04 Aug 2006, 07:20
Not the UNsigned integers themself, but the difference of two unsigned ones...
@r22: Did you test it on AMD/Intel? I've heard that on AMD the conditional move is slow and cmp/jcc is faster :S |
|||
|
04 Aug 2006, 07:20 |
|
|
Tomasz Grysztar 04 Aug 2006, 08:54
Madis731 wrote:
For difference it doesn't really matter whether it is of signed or unsigned values. |
|||
|
04 Aug 2006, 08:54 |
|
|
r22 04 Aug 2006, 12:10
Very true, BUT the comment would be valid IF it was used for a min/max function. That's exactly why I don't comment code, half the time I just type nonsense after ;'s.
The AMD64 optimization guide shows the cmov reg,reg as having a 1 clock latency and I've tested it on my AMD64 x2 3800+ and it's shown to be slightly faster than the carry method. |
|||
|
04 Aug 2006, 12:10 |
|
|
Madis731 05 Aug 2006, 14:48
@Tomasz: Of course - you are right - I wasn't thinking clearly. It does NOT matter! And more importantly you can even use it on signed values (i.e. if one or both of the source operands are negative in this sence).
|
|||
|
05 Aug 2006, 14:48 |
|
| Goto page Previous 1, 2 < Last Thread | Next Thread > |
Forum Rules:
|