mirror of
https://github.com/KolibriOS/kolibrios.git
synced 2024-12-24 15:46:49 +03:00
4b8e0542e3
1. 32bit vertices indexes and ability to load whole RAM limited objects. (Above 65535 vertices and triangles). 2. I switch off painters algotithm mode (depth sorting). In app impelementetion it has limited vertices count and produce less quality image than Z buffer Catmull algo. In addition this switch off reduces app size. git-svn-id: svn://kolibrios.org@8014 a494cfbc-eb01-0410-851d-a64ba20cac60
566 lines
12 KiB
Plaintext
566 lines
12 KiB
Plaintext
x3d equ 0
|
|
y3d equ 2
|
|
z3d equ 4
|
|
vec_x equ 0
|
|
vec_y equ 4
|
|
vec_z equ 8
|
|
; 3d point - triple integer word coordinate
|
|
; vector - triple float dword coordinate
|
|
;----------------------in: --------------------------------
|
|
;------------------------ esi - pointer to 1st 3d point ---
|
|
;------------------------ edi - pointer to 2nd 3d point ---
|
|
;------------------------ ebx - pointer to result vector --
|
|
;---------------------- out : none ------------------------
|
|
if 0
|
|
make_vector:
|
|
fninit
|
|
fild word[edi+x3d] ;edi+x3d
|
|
fisub word[esi+x3d] ;esi+x3d
|
|
fstp dword[ebx+vec_x]
|
|
|
|
fild word[edi+y3d]
|
|
fisub word[esi+y3d]
|
|
fstp dword[ebx+vec_y]
|
|
|
|
fild word[edi+z3d]
|
|
fisub word[esi+z3d]
|
|
fstp dword[ebx+vec_z]
|
|
|
|
ret
|
|
end if
|
|
reverse_mx_3x3:
|
|
; esi - source matrix
|
|
; edi - desired reversed matrix
|
|
|
|
push ebp
|
|
mov ebp,esp
|
|
sub esp,4
|
|
.det equ ebp-4
|
|
|
|
fninit
|
|
fld dword[esi]
|
|
fmul dword[esi+16]
|
|
fmul dword[esi+32]
|
|
fld dword[esi+12]
|
|
fmul dword[esi+28]
|
|
fmul dword[esi+8]
|
|
faddp
|
|
fld dword[esi+24]
|
|
fmul dword[esi+4]
|
|
fmul dword[esi+20]
|
|
faddp
|
|
fld dword[esi]
|
|
fmul dword[esi+28]
|
|
fmul dword[esi+20]
|
|
fchs
|
|
faddp
|
|
fld dword[esi+24]
|
|
fmul dword[esi+16]
|
|
fmul dword[esi+8]
|
|
fchs
|
|
faddp
|
|
fld dword[esi+12]
|
|
fmul dword[esi+4]
|
|
fmul dword[esi+32]
|
|
fchs
|
|
faddp
|
|
fstp dword[.det]
|
|
cmp dword[.det],0
|
|
jne @f
|
|
int3
|
|
@@:
|
|
; fld1
|
|
; fdiv dword[.det]
|
|
; fstp dword[.det]
|
|
|
|
fld dword[esi+16]
|
|
fmul dword[esi+32]
|
|
fld dword[esi+20]
|
|
fmul dword[esi+28]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi]
|
|
|
|
fld dword[esi+8]
|
|
fmul dword[esi+28]
|
|
fld dword[esi+4]
|
|
fmul dword[esi+32]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+4]
|
|
|
|
fld dword[esi+4]
|
|
fmul dword[esi+20]
|
|
fld dword[esi+8]
|
|
fmul dword[esi+16]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+8]
|
|
|
|
fld dword[esi+20]
|
|
fmul dword[esi+24]
|
|
fld dword[esi+12]
|
|
fmul dword[esi+32]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+12]
|
|
|
|
fld dword[esi]
|
|
fmul dword[esi+32]
|
|
fld dword[esi+8]
|
|
fmul dword[esi+24]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+16]
|
|
|
|
fld dword[esi+8]
|
|
fmul dword[esi+12]
|
|
fld dword[esi]
|
|
fmul dword[esi+20]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+20]
|
|
|
|
fld dword[esi+12]
|
|
fmul dword[esi+28]
|
|
fld dword[esi+16]
|
|
fmul dword[esi+24]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+24]
|
|
|
|
fld dword[esi+4]
|
|
fmul dword[esi+24]
|
|
fld dword[esi]
|
|
fmul dword[esi+28]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+28]
|
|
|
|
fld dword[esi]
|
|
fmul dword[esi+16]
|
|
fld dword[esi+4]
|
|
fmul dword[esi+12]
|
|
fchs
|
|
faddp
|
|
fdiv dword[.det]
|
|
fstp dword[edi+32]
|
|
|
|
|
|
mov esp,ebp
|
|
pop ebp
|
|
ret
|
|
|
|
make_vector_r:
|
|
if Ext < SSE2
|
|
fninit
|
|
fld dword[edi] ;edi+x3d
|
|
fsub dword[esi] ;esi+x3d
|
|
fstp dword[ebx+vec_x]
|
|
|
|
fld dword[edi+4]
|
|
fsub dword[esi+4]
|
|
fstp dword[ebx+vec_y]
|
|
|
|
fld dword[edi+8]
|
|
fsub dword[esi+8]
|
|
fstp dword[ebx+vec_z]
|
|
else
|
|
movups xmm0,[esi]
|
|
movups xmm1,[edi]
|
|
subps xmm1,xmm0
|
|
movlps [ebx],xmm1
|
|
movhlps xmm1,xmm1
|
|
movss [ebx+8],xmm1
|
|
end if
|
|
|
|
ret
|
|
;---------------------- in: -------------------------------
|
|
;--------------------------- esi - pointer to 1st vector --
|
|
;--------------------------- edi - pointer to 2nd vector --
|
|
;--------------------------- ebx - pointer to result vector
|
|
;---------------------- out : none
|
|
cross_product:
|
|
fninit
|
|
fld dword [esi+vec_y]
|
|
fmul dword [edi+vec_z]
|
|
fld dword [esi+vec_z]
|
|
fmul dword [edi+vec_y]
|
|
fsubp ;st1 ,st
|
|
fstp dword [ebx+vec_x]
|
|
|
|
fld dword [esi+vec_z]
|
|
fmul dword [edi+vec_x]
|
|
fld dword [esi+vec_x]
|
|
fmul dword [edi+vec_z]
|
|
fsubp ;st1 ,st
|
|
fstp dword [ebx+vec_y]
|
|
|
|
fld dword [esi+vec_x]
|
|
fmul dword [edi+vec_y]
|
|
fld dword [esi+vec_y]
|
|
fmul dword [edi+vec_x]
|
|
fsubp ;st1 ,st
|
|
fstp dword [ebx+vec_z]
|
|
ret
|
|
;----------------------- in: ------------------------------
|
|
;---------------------------- edi - pointer to vector -----
|
|
;----------------------- out : none
|
|
normalize_vector:
|
|
fninit
|
|
fld dword [edi+vec_x]
|
|
fmul st, st
|
|
fld dword [edi+vec_y]
|
|
fmul st, st
|
|
fld dword [edi+vec_z]
|
|
fmul st, st
|
|
faddp st1, st
|
|
faddp st1, st
|
|
fsqrt
|
|
|
|
ftst
|
|
fstsw ax
|
|
sahf
|
|
jnz @f
|
|
|
|
fst dword [edi+vec_x]
|
|
fst dword [edi+vec_y]
|
|
fstp dword [edi+vec_z]
|
|
ret
|
|
@@:
|
|
fld st
|
|
fld st
|
|
fdivr dword [edi+vec_x]
|
|
fstp dword [edi+vec_x]
|
|
fdivr dword [edi+vec_y]
|
|
fstp dword [edi+vec_y]
|
|
fdivr dword [edi+vec_z]
|
|
fstp dword [edi+vec_z]
|
|
ret
|
|
;------------------in: -------------------------
|
|
;------------------ esi - pointer to 1st vector
|
|
;------------------ edi - pointer to 2nd vector
|
|
;------------------out: ------------------------
|
|
;------------------ st0 - dot-product
|
|
dot_product:
|
|
fninit
|
|
fld dword [esi+vec_x]
|
|
fmul dword [edi+vec_x]
|
|
fld dword [esi+vec_y]
|
|
fmul dword [edi+vec_y]
|
|
fld dword [esi+vec_z]
|
|
fmul dword [edi+vec_z]
|
|
faddp
|
|
faddp
|
|
ret
|
|
|
|
; DOS version Coded by Mikolaj Felix aka Majuma
|
|
; mfelix@polbox.com
|
|
; www.majuma.xt.pl
|
|
; into FASM translation by Macgub
|
|
init_sincos_tab:
|
|
.counter equ dword [ebp-4] ; cur angle
|
|
|
|
push ebp
|
|
mov ebp,esp
|
|
|
|
xor eax,eax
|
|
push eax ; init .counter
|
|
mov edi,cos_tab
|
|
mov esi,sin_tab
|
|
mov ecx,256
|
|
fninit
|
|
|
|
fld .counter
|
|
@@:
|
|
fld st
|
|
fsincos
|
|
fstp dword [edi]
|
|
fstp dword [esi]
|
|
; fadd [piD180]
|
|
fadd [piD128]
|
|
add esi,4
|
|
add edi,4
|
|
loop @b
|
|
ffree st
|
|
|
|
mov esp,ebp
|
|
pop ebp
|
|
ret
|
|
;------
|
|
; esi - offset (pointer) to angles, edi offset to 3x3 matrix
|
|
make_rotation_matrix:
|
|
.sinx equ dword[ebp-4]
|
|
.cosx equ dword[ebp-8]
|
|
.siny equ dword[ebp-12]
|
|
.cosy equ dword[ebp-16]
|
|
.sinz equ dword[ebp-20]
|
|
.cosz equ dword[ebp-24]
|
|
push ebp
|
|
mov ebp,esp
|
|
sub esp,24
|
|
|
|
movzx ebx,word[esi]
|
|
shl ebx,2
|
|
mov eax,dword[sin_tab+ebx]
|
|
mov .sinx,eax
|
|
mov edx,dword[cos_tab+ebx]
|
|
mov .cosx,edx
|
|
|
|
movzx ebx,word[esi+2]
|
|
shl ebx,2
|
|
mov eax,dword[sin_tab+ebx]
|
|
mov .siny,eax
|
|
mov edx,dword[cos_tab+ebx]
|
|
mov .cosy,edx
|
|
|
|
movzx ebx,word[esi+4]
|
|
shl ebx,2
|
|
mov eax,dword[sin_tab+ebx]
|
|
mov .sinz,eax
|
|
mov edx,dword[cos_tab+ebx]
|
|
mov .cosz,edx
|
|
|
|
fninit
|
|
fld .cosy
|
|
fmul .cosz
|
|
fstp dword[edi]
|
|
|
|
fld .sinx
|
|
fmul .siny
|
|
fmul .cosz
|
|
fld .cosx
|
|
fmul .sinz
|
|
fchs
|
|
faddp
|
|
fstp dword[edi+12]
|
|
|
|
fld .cosx
|
|
fmul .siny
|
|
fmul .cosz
|
|
fld .sinx
|
|
fmul .sinz
|
|
faddp
|
|
fstp dword[edi+24]
|
|
|
|
fld .cosy
|
|
fmul .sinz
|
|
fstp dword[edi+4]
|
|
|
|
fld .sinx
|
|
fmul .siny
|
|
fmul .sinz
|
|
fld .cosx
|
|
fmul .cosz
|
|
faddp
|
|
fstp dword[edi+16]
|
|
|
|
fld .cosx
|
|
fmul .siny
|
|
fmul .sinz
|
|
fld .sinx
|
|
fchs
|
|
fmul .cosz
|
|
faddp
|
|
fstp dword[edi+28]
|
|
|
|
fld .siny
|
|
fchs
|
|
fstp dword[edi+8]
|
|
|
|
fld .cosy
|
|
fmul .sinx
|
|
fstp dword[edi+20]
|
|
|
|
fld .cosx
|
|
fmul .cosy
|
|
fstp dword[edi+32]
|
|
|
|
mov esp,ebp
|
|
pop ebp
|
|
ret
|
|
;---------------------
|
|
; in: esi - ptr to points(normals], each point(normal) coeficient as dword
|
|
; edi - ptr to rotated points(normals)
|
|
; ebx - ptr to 3x3 (9 dwords, 36 bytes) rotation matrix
|
|
; ecx - number of points(normals)
|
|
rotary:
|
|
if Ext<SSE
|
|
fninit
|
|
.again:
|
|
|
|
fld dword[esi]
|
|
fmul dword[ebx]
|
|
fld dword[esi+4]
|
|
fmul dword[ebx+12]
|
|
faddp
|
|
fld dword[esi+8]
|
|
fmul dword[ebx+24]
|
|
faddp
|
|
fstp dword[edi]
|
|
|
|
|
|
fld dword[esi+4]
|
|
fmul dword[ebx+16]
|
|
fld dword[esi]
|
|
fmul dword[ebx+4]
|
|
faddp
|
|
fld dword[esi+8]
|
|
fmul dword[ebx+28]
|
|
faddp
|
|
fstp dword[edi+4]
|
|
|
|
|
|
fld dword[esi+8]
|
|
fmul dword[ebx+32]
|
|
fld dword[esi]
|
|
fmul dword[ebx+8]
|
|
fld dword[esi+4]
|
|
fmul dword[ebx+20]
|
|
faddp
|
|
faddp
|
|
fstp dword[edi+8]
|
|
|
|
|
|
add esi,12
|
|
add edi,12
|
|
loop .again
|
|
mov [edi],dword -1
|
|
else
|
|
; Copyright (C) 1999-2001 Brian Paul
|
|
; Copyright (C) Maciej Guba
|
|
;---------------------
|
|
; in: esi - ptr to points(normals], each point(normal) coeficient as dword
|
|
; edi - ptr to rotated points(normals)
|
|
; ebx - ptr to 3x3 (9 dwords, 36 bytes) rotation matrix
|
|
; ecx - number of points(normals)
|
|
;align 32
|
|
movups xmm4,[ebx]
|
|
; lddqu xmm4,[ebx] ; I tried sse3 :D
|
|
movups xmm5,[ebx+12]
|
|
movups xmm6,[ebx+24]
|
|
;align 32
|
|
.again:
|
|
movss xmm0,dword[esi]
|
|
shufps xmm0,xmm0,0
|
|
mulps xmm0,xmm4
|
|
|
|
movss xmm1,dword[esi+4]
|
|
shufps xmm1,xmm1,0
|
|
mulps xmm1,xmm5
|
|
|
|
movss xmm2,dword[esi+8]
|
|
shufps xmm2,xmm2,0
|
|
mulps xmm2,xmm6
|
|
|
|
addps xmm0,xmm1
|
|
addps xmm0,xmm2
|
|
|
|
movups [edi],xmm0
|
|
|
|
add esi,12
|
|
add edi,12
|
|
dec ecx
|
|
jne .again
|
|
mov [edi],dword -1
|
|
end if
|
|
ret
|
|
;----------------------------------------------
|
|
; esi - pointer to 3x3 matrix
|
|
add_scale_to_matrix:
|
|
fninit
|
|
fld [rsscale]
|
|
fld dword[esi] ;-----
|
|
fmul st,st1
|
|
fstp dword[esi]
|
|
fld dword[esi+12] ; x scale
|
|
fmul st,st1
|
|
fstp dword[esi+12]
|
|
fld dword[esi+24]
|
|
fmul st,st1
|
|
fstp dword[esi+24] ;------
|
|
|
|
fld dword[esi+4] ;-----
|
|
fmul st,st1
|
|
fstp dword[esi+4]
|
|
fld dword[esi+16] ; y scale
|
|
fmul st,st1
|
|
fstp dword[esi+16]
|
|
fld dword[esi+28]
|
|
fmul st,st1
|
|
fstp dword[esi+28] ;------
|
|
|
|
|
|
fld dword[esi+8] ;-----
|
|
fmul st,st1
|
|
fstp dword[esi+8]
|
|
fld dword[esi+20] ; z scale
|
|
fmul st,st1
|
|
fstp dword[esi+20]
|
|
fld dword[esi+32]
|
|
fmulp st1,st
|
|
fstp dword[esi+32] ;------
|
|
|
|
ret
|
|
|
|
;in esi - offset to 3d points (point as 3 dwords float)
|
|
; edi - offset to 2d points ( as 3 words integer)
|
|
; ecx - number of points
|
|
translate_points: ; just convert into integer; z coord still needed
|
|
fninit
|
|
.again:
|
|
if 0
|
|
fld dword[esi+8]
|
|
; fmul [rsscale]
|
|
fist word[edi+4]
|
|
|
|
fisub [zobs]
|
|
fchs
|
|
|
|
fld dword[esi]
|
|
; fmul [rsscale]
|
|
fisub [xobs]
|
|
fimul [zobs]
|
|
fdiv st0,st1
|
|
|
|
fiadd [xobs]
|
|
fiadd [vect_x]
|
|
fistp word[edi]
|
|
|
|
fld dword[esi+4]
|
|
; fmul [rsscale]
|
|
fisub [yobs]
|
|
fimul [zobs]
|
|
fdivrp ; st0,st1
|
|
|
|
fiadd [yobs]
|
|
fiadd [vect_y]
|
|
fistp word[edi+2]
|
|
end if
|
|
; movups xmm0,[esi]
|
|
; cvtps2dq xmm0,xmm0
|
|
; packsdw xmm0,xmm0
|
|
; movq [edi]
|
|
fld dword[esi]
|
|
fiadd [vect_x]
|
|
fistp word[edi]
|
|
fld dword[esi+4]
|
|
fiadd [vect_y]
|
|
fistp word[edi+2]
|
|
fld dword[esi+8]
|
|
fistp word[edi+4]
|
|
add esi,12
|
|
add edi,6
|
|
dec ecx
|
|
jnz .again
|
|
|
|
ret
|