Index
> Main > edfedham line algorythm for fool. |
| Author |
|
|
vid 14 Jan 2010, 14:27
Quote: to test this function, it is very easy: I suppose much more is needed... |
|||
14 Jan 2010, 14:27 |
|
|
edfed 14 Jan 2010, 15:34
yep.
setup mode 13h set fs as frame buffer. Code: org 100h mov ax,13h int 10h mov ax,0a000h mov fs,ax jmp start include 'line.inc' testline dd line,0,0,320,200,4 start: mov esi,testline call near[esi] jmp start |
|||
|
14 Jan 2010, 15:34 |
|
|
bitshifter 14 Jan 2010, 19:51
LOL, I love the name, edfedham
A few tips for speedup... 1) Bresenham was fast but now we have FPU which can eliminate the jumping withing the main loop and also we can mix INT and FPU operations to run in same cycle. 2) Clip the line with outcode testing instead of per pixel. Even Barsky clipping would be better for that matter... 3) Handle all 8 octants seperately (with code size penalty of course) 4) Use a mask that may render multiple pixels at once... And then we have run slice lines, which ill save for another day... I have some fast C code for this (not yet converted ASM) If you would like to see my ideas i can send you some code... |
|||
|
14 Jan 2010, 19:51 |
|
|
edfed 14 Jan 2010, 22:20
8 FPU register???
hum... and what about a multithred appliction? how can i efficienttly save FPU registers? |
|||
|
14 Jan 2010, 22:20 |
|
|
bitshifter 15 Jan 2010, 10:55
Only a couple FPU registers are needed to handle minor delta in loop.
Most everything is done with integer all except adding the fractional part. Here is some C-code for DDA line drawing... Please forgive me but i do ALL my prototyping in C first. Code: /////////////////////////////////////////////////////////////////////////////// // // 315 0 45 // \ | / // \ 8 | 1 / // \ | / // \ | / // \ | / // 7 \ | / 2 // \|/ // 270 ------(*)------ 90 // /|\ // 6 / | \ 3 // / | \ // / | \ // / | \ // / 5 | 4 \ // / | \ // 225 180 135 // /////////////////////////////////////////////////////////////////////////////// void DrawLine( long x1, long y1, long x2, long y2, long color) { // if going right -> possible sectors are [1,2,3,4] if(x1 < x2) { long delta_x = x2 - x1; // if going down -> possible sectors are [3,4] if(y1 < y2) { long delta_y = y2 - y1; // if more vertical (SECTOR 4) if(delta_x < delta_y) { printf("TRACE: SECTOR(4)\n"); float m = float(delta_x) / float(delta_y); float x = float(x1); for(long y = y1; y <= y2; y++) { SetPixel(long(x),y,color); x += m; } } // else more horizontal (SECTOR 3) else { printf("TRACE: SECTOR(3)\n"); float m = float(delta_y) / float(delta_x); float y = float(y1); for(long x = x1; x <= x2; x++) { SetPixel(x,long(y),color); y += m; } } } // else going up -> possible sectors are [1,2] else { long delta_y = y1 - y2; // if more vertical (SECTOR 1) if(delta_x < delta_y) { printf("TRACE: SECTOR(1)\n"); g_elapse = -1; float m = float(delta_x) / float(delta_y); float x = float(x1); for(long y = y1; y >= y2; y--) { SetPixel(long(x),y,color); x += m; } } // else more horizontal (SECTOR 2) else { printf("TRACE: SECTOR(2)\n"); float m = float(delta_y) / float(delta_x); float y = float(y2); for(long x = x2; x >= x1; x--) { SetPixel(x,long(y),color); y += m; } } } } // else going left -> possible sectors are [5,6,7,8] else { long delta_x = x1 - x2; // if going down -> possible sectors are [5,6] if(y1 < y2) { long delta_y = y2 - y1; // if more vertical (SECTOR 5) if(delta_x < delta_y) { printf("TRACE: SECTOR(5)\n"); float m = float(delta_x) / float(delta_y); float x = float(x2); for(long y = y2; y >= y1; y--) { SetPixel(long(x),y,color); x += m; } } // else more horizontal (SECTOR 6) else { printf("TRACE: SECTOR(6)\n"); float m = float(delta_y) / float(delta_x); float y = float(y1); for(long x = x1; x >= x2; x--) { SetPixel(x,long(y),color); y += m; } } } // else going up -> possible sectors are [7,8] else { long delta_y = y1 - y2; // if more vertical (SECTOR 8) if(delta_x < delta_y) { printf("TRACE: SECTOR(8)\n"); float m = float(delta_x) / float(delta_y); float x = float(x2); for(long y = y2; y <= y1; y++) { SetPixel(long(x),y,color); x += m; } } // else if more horizontal (SECTOR 7) else { printf("TRACE: SECTOR(7)\n"); float m = float(delta_y) / float(delta_x); float y = float(y2); for(long x = x2; x <= x1; x++) { SetPixel(x,long(y),color); y += m; } } } } } Notice that i adjusted the loop so the minor (fractional part) always increase. This is to have consistent float->int conversion within every octants :) I had a few goals when i wrote this code... 1) Refresh my memory on linear algebra slope formula. 2) Do very little precalculation as possable. 3) Render all octants seperately and watch them work. Code size was not a problem so thats why its big but straight forward. I am happy with the results, but some INT-FPU mixing could make it sing. Just i have many projects going and it is not very high priority... If you need help with 2D line clipping just let me know, i have much gfx code... |
|||
|
15 Jan 2010, 10:55 |
|
|
edfed 16 Jan 2010, 00:58
i need to clip lines because i want no tests inside put pixel function.
then, putpixel will only take color, X, and Y and hoplà, it will be in frame buffer. i still does it for rectangle drawing, i clip its coordinates (x,y,xl,yl too) with the screen, and then, the drawing loop is faster. the screen itself is a rectangle. with X= Y = 0 xl=320 for the moment YL=200 for the moment... then, it will be a general function . clip item1 in item 2 but a line doens't clip the same way as a rectangle. about C? it means Chinese?  |
|||
|
16 Jan 2010, 00:58 |
|
|
bitshifter 16 Jan 2010, 01:05
edfed said:
but a line doens't clip the same way as a rectangle. Sure it does, think about points, which they both have in common... The goal is to classify a point to a bounding volume. And C is way easier to understand than chinese |
|||
|
16 Jan 2010, 01:05 |
|
|
edfed 16 Jan 2010, 01:36
Code:
mov edi,[desk]
mov eax,[esi+.x]
mov ebx,[esi+.xl]
mov ecx,[edi+.x]
mov edx,[edi+.xl]
add edx,ecx
cmp eax,edx
jge .end
cmp eax,ecx
jnl @f
sub eax,ecx
add ebx,eax
jle .end
mov eax,ecx
@@:
lea ecx,[eax+ebx]
cmp ecx,edx
jl @f
sub edx,eax
mov ebx,edx
@@:
mov [esi+.x],eax
mov [esi+.xl],ebx
mov eax,[esi+.y]
mov ebx,[esi+.yl]
mov ecx,[edi+.y]
mov edx,[edi+.yl]
add edx,ecx
cmp eax,edx
jge .end
cmp eax,ecx
jnl @f
sub eax,ecx
add ebx,eax
jle .end
mov eax,ecx
@@:
lea ecx,[eax+ebx]
cmp ecx,edx
jl @f
sub edx,eax
mov ebx,edx
@@:
mov [esi+.y],eax
mov [esi+.yl],ebx
if i use this code (it clips a rectangle with another rectangle (the screen)) on a line, it will not give me the intersects of the line with the screen. |
|||
|
16 Jan 2010, 01:36 |
|
|
bitshifter 16 Jan 2010, 03:24
When i get to my other PC (downstairs) i will get
some code for slope->edge intersection formula. I have a bit more free time now, i was chasing a stupid bug in one of my new gfx demo's for two days and finally found it! So stupid was i to overlook something as simple as this... Code: mov ds,[g_segment_a] mov es,[g_segment_b] |
|||
|
16 Jan 2010, 03:24 |
|
< Last Thread | Next Thread > |
Forum Rules:
|