Original Post
Hi!
I've made my own software gourand shading 3d renderer with zbuffering.
Now I should optimize it, but I'm not very good at it.
So here's the code:
struct FillStruct
{
DWORD *ScanLinePtr; // pointer to scanline
float *ZBufferPtr; // pointer to zbuffer
long x[2]; // fixedpoint values for
// left and right x
long c[4]; // color of the first pixel of
// the left side
long dc[4]; // dcolor/dx of color
float z; // zvalue of the left pixel
float dz; // dz/dx
};
long start_x, end_x;
long slc[4];
float slz;
int lpitch, scr_width,scr_height;
FillStruct xfs;
#define SLPTR 0
#define ZBPTR 4
#define X 8
#define C 16
#define DC 32
#define Z 48
#define DZ 52
void _declspec(naked) DrawScanline2_asm()
{
_asm
{
push esi
push edi
push ebx
mov eax, 08000h
mov edx, 10000h
mov ebx, [xfs + X]
cmp bx, ax
jb DSL2_NOINC1
add ebx, edx // eax ebx ecx edx
DSL2_NOINC1: // na startx na na
mov ecx, [xfs + X + 4]
cmp cx, ax
jb DSL2_NOINC2
add ecx, edx
// na startx endx na
DSL2_NOINC2:
sar ebx, 16
sar ecx, 16
mov edx, dword ptr [scr_width]
test ecx, ecx
jle DSL2_EXIT
cmp ebx, edx
jge DSL2_EXIT
cmp ecx, edx // na startx endx screen_width
jle DSL2_OVER1
mov ecx, edx
DSL2_OVER1:
test ebx, ebx
jge DSL2_OVER2
xor ebx, ebx
DSL2_OVER2: // na startx endx screen_width
sub ecx, ebx // na startx width sw
test ecx, ecx
jle DSL2_EXIT
fld [xfs + DZ]// eax ebx ecx edx
shl ebx, 2 // na startx*4 width sw
mov edi, [xfs + ZBPTR]
mov esi, [xfs + SLPTR]
//eax ebx ecx edx edi esi
//na startx*4 width sw zbptr slptr
add edi, ebx
add esi, ebx
cmp ecx, 32
jle DSL2_NOPREFETCH
/* prefetc screen and zbuffer */
mov eax, ecx
shr eax, 5
shl eax, 5
DSL2_PRELOOP:
mov ebx, [edi + eax]
mov ebx, [esi + eax]
sub eax, 32
test eax, eax
jge DSL2_PRELOOP
DSL2_NOPREFETCH:
fld [xfs + Z]
mov eax, [xfs + C]
mov ebx, [xfs + C + 4]
mov edx, [xfs + C + 8]
mov [slc], eax
mov [slc + 4], ebx
mov [slc + 8], edx
fld dword ptr [edi] //na na width na zbptr slcptr
st0: z st1: dz
DSL2_MAINLOOP:
fcomp st(1)
fnstsw ax
test ah,5
jp DSL2_END
fst dword ptr [edi]
mov eax, dword ptr [slc]
mov al, byte ptr [slc + 10]
mov ah, byte ptr [slc + 6]
mov dword ptr [esi], eax
DSL2_END:
dec ecx
jz DSL2_EXIT2
fadd st(0), st(1)
mov eax, [xfs + DC]
add edi, 4
mov ebx, [xfs + DC + 4]
add [slc], eax
fld dword ptr [edi]
add esi, 4
mov edx, [xfs + DC + 8]
add [slc + 4], ebx
add [slc + 8], edx
jmp DSL2_MAINLOOP
DSL2_EXIT2:
fstp st(0)
fstp st(0)
DSL2_EXIT:
pop ebx
pop edi
pop esi
ret
}
}
Next step would be to use MMX, but I don't know how I should do the z comparison. I know I can't mix MMX and floating point code. How should I do the z comparison with MMX?
edit: Tried to make the code look more readable.
edit: retry
[edited by - Atm97fin on March 3, 2004 6:18:18 AM]
[edited by - Atm97fin on March 3, 2004 7:00:50 AM]
