mirror of
https://github.com/isledecomp/isle.git
synced 2024-11-27 01:56:28 -05:00
1086 lines
36 KiB
C
1086 lines
36 KiB
C
|
/*
|
||
|
* vec.h -- Vector macros for 2,3, and 4 dimensions,
|
||
|
* for any combination of C scalar types.
|
||
|
*
|
||
|
* Author: Don Hatch (hatch@sgi.com)
|
||
|
* Last modified: Fri Dec 15 01:57:07 PST 1995
|
||
|
*
|
||
|
* General description:
|
||
|
*
|
||
|
* The macro name describes its arguments; e.g.
|
||
|
* MXS3 is "matrix times scalar in 3 dimensions";
|
||
|
* VMV2 is "vector minus vector in 2 dimensions".
|
||
|
*
|
||
|
* If the result of an operation is a scalar, then the macro "returns"
|
||
|
* the value; e.g.
|
||
|
* result = DOT3(v,w);
|
||
|
* result = DET4(m);
|
||
|
*
|
||
|
* If the result of an operation is a vector or matrix, then
|
||
|
* the first argument is the destination; e.g.
|
||
|
* SET2(tovec, fromvec);
|
||
|
* MXM3(result, m1, m2);
|
||
|
*
|
||
|
* WARNING: For the operations that are not done "componentwise"
|
||
|
* (e.g. vector cross products and matrix multiplies)
|
||
|
* the destination should not be either of the arguments,
|
||
|
* for obvious reasons. For example, the following is wrong:
|
||
|
* VXM2(v,v,m);
|
||
|
* For such "unsafe" macros, there are safe versions provided,
|
||
|
* but you have to specify a type for the temporary
|
||
|
* result vector or matrix. For example, the safe versions
|
||
|
* of VXM2 are:
|
||
|
* VXM2d(v,v,m) if v's scalar type is double or float
|
||
|
* VXM2i(v,v,m) if v's scalar type is int or char
|
||
|
* VXM2l(v,v,m) if v's scalar type is long
|
||
|
* VXM2r(v,v,m) if v's scalar type is real
|
||
|
* VXM2safe(type,v,v,m) for other scalar types.
|
||
|
*
|
||
|
* These "safe" macros and INVERTMAT do not evaluate to C expressions
|
||
|
* (so, for example, they can't be used inside the parentheses of
|
||
|
* a for(...)).
|
||
|
*
|
||
|
* Specific descriptions:
|
||
|
*
|
||
|
* The "?"'s in the following can be 2, 3, or 4.
|
||
|
*
|
||
|
* EXPAND?(v) comma-separated list of elements of v
|
||
|
*
|
||
|
* SET?(to,from) to = from
|
||
|
* SETMAT?(to,from) to = from
|
||
|
* ROUNDVEC?(to,from) to = from with entries rounded
|
||
|
* to nearest integer
|
||
|
* ROUNDMAT?(to,from) to = from with entries rounded
|
||
|
* to nearest integer
|
||
|
* FILLVEC?(v,s) set each entry of vector v to be s
|
||
|
* FILLMAT?(m,s) set each entry of matrix m to be s
|
||
|
* ZEROVEC?(v) v = 0
|
||
|
* ISZEROVEC?(v) v == 0
|
||
|
* EQVEC?(v,w) v == w
|
||
|
* EQMAT?(m1,m2) m1 == m2
|
||
|
* ZEROMAT?(m) m = 0
|
||
|
* IDENTMAT?(m) m = 1
|
||
|
* TRANSPOSE?(to,from) (matrix to) = (transpose of matrix from)
|
||
|
* ADJOINT?(to,from) (matrix to) = (adjoint of matrix from)
|
||
|
* i.e. its determinant times its inverse
|
||
|
* INVERTMAT?{d,i,l,r}(to,from) (matrix to) = (inverse of matrix from)
|
||
|
* with temp adjoint and determinant type
|
||
|
* double, int, long, or real respectively
|
||
|
*
|
||
|
* V{P,M}V?(to,v,w) to = v {+,-} w
|
||
|
* M{P,M}M?(to,m1,m2) to = m1 {+,-} m2
|
||
|
* SX{V,M}?(to,s,from) to = s * from
|
||
|
* VPSXV?(to,v,s,w) to = v + s*w
|
||
|
* VPVXS?(to,v,w,s) to = v + w*s
|
||
|
* M{V,M}?(to,from) to = -from
|
||
|
* {V,M}{X,D}S?(to,from,s) to = from {*,/} s
|
||
|
* MXM?(to,m1,m2) to = m1 * m2
|
||
|
* VXM?(to,v,m) (row vec to) = (row vec v) * m
|
||
|
* MXV?(to,m,v) (column vec to) = m * (column vec v)
|
||
|
* VMODS?(to,v,s) to = v mod s (always >= 0)
|
||
|
* VMODV?(to,v0,v1) to = v0 mod v1 componentwise
|
||
|
* VDIVS?(to,v,s) to = (v-(v mod s))/s
|
||
|
* VDIVV?(to,v0,v1) to = (v0-(v0 mod v1))/v1 componentwise
|
||
|
* V{MIN,MAX}S?(to,v,s) to = {MIN,MAX}(v, s)
|
||
|
* V{MIN,MAX}V?(to,v0,v1) to = {MIN,MAX}(v0, v1)
|
||
|
* LERP?(to,v0,v1,t) to = v0 + t*(v1-v0)
|
||
|
*
|
||
|
* DET?(m) determinant of m
|
||
|
* TRACE?(m) trace (sum of diagonal entries) of m
|
||
|
* DOT?(v,w) dot (scalar) product of v and w
|
||
|
* NORMSQRD?(v) square of |v|
|
||
|
* DISTSQRD?(v,w) square of |v-w|
|
||
|
*
|
||
|
* XV2(to,v) to = v rotated by 90 degrees
|
||
|
* VXV3(to,v1,v2) to = cross (vector) product of v1 and v2
|
||
|
* VXVXV4(to,v1,v2,v3) to = 4-dimensional vector cross product
|
||
|
* of v1,v2,v3 (a vector orthogonal to
|
||
|
* v1,v2,v3 whose length equals the
|
||
|
* volume of the spanned parallelotope)
|
||
|
* VXV2(v0,v1) determinant of matrix with rows v0,v1
|
||
|
* VXVXV3(v0,v1,v2) determinant of matrix with rows v0,v1,v2
|
||
|
* VXVXVXV4(v0,v1,v2,v3) determinant of matrix with rows v0,..,v3
|
||
|
*
|
||
|
* The following macros mix objects from different dimensions.
|
||
|
* For example, V3XM4 would be used to apply a composite
|
||
|
* 4x4 rotation-and-translation matrix to a 3d vector.
|
||
|
*
|
||
|
* SET3from2(to,from,pad) (3d vec to) = (2d vec from) with pad
|
||
|
* SET4from3(to,from,pad) (4d vec to) = (3d vec from) with pad
|
||
|
* SETMAT3from2(to,from,pad0,pad1) (3x3 mat to) = (2x2 mat from)
|
||
|
* padded with pad0 on the sides
|
||
|
* and pad1 in the corner
|
||
|
* SETMAT4from3(to,from,pad0,pad1) (4x4 mat to) = (3x3 mat from)
|
||
|
* padded with pad0 on the sides
|
||
|
* and pad1 in the corner
|
||
|
* V2XM3(to2,v2,m3) (2d row vec to2) = (2d row vec v2) * (3x3 mat m3)
|
||
|
* V3XM4(to3,v3,m4) (3d row vec to3) = (3d row vec v2) * (4x4 mat m4)
|
||
|
* M3XV2(to2,m3,v2) (2d col vec to2) = (3x3 mat m3) * (2d col vec v2)
|
||
|
* M4XV3(to3,m4,v3) (3d col vec to3) = (4x4 mat m4) * (3d col vec v3)
|
||
|
* M2XM3(to3,m2,m3) (3x3 mat to3) = (2x2 mat m2) * (3x3 mat m3)
|
||
|
* M3XM4(to4,m3,m4) (4x4 mat to4) = (3x3 mat m3) * (4x4 mat m4)
|
||
|
* M3XM2(to3,m3,m2) (3x3 mat to3) = (3x3 mat m3) * (2x2 mat m2)
|
||
|
* M4XM3(to4,m4,m3) (4x4 mat to4) = (4x4 mat m4) * (3x3 mat m3)
|
||
|
*
|
||
|
*
|
||
|
* This file is machine-generated and can be regenerated
|
||
|
* for any number of dimensions.
|
||
|
* The program that generated it is available upon request.
|
||
|
*/
|
||
|
|
||
|
#ifndef VEC_H
|
||
|
#define VEC_H 4
|
||
|
#include <math.h> /* for definition of floor() */
|
||
|
#define EXPAND2(v) (v)[0], (v)[1]
|
||
|
#define EXPAND3(v) (v)[0], (v)[1], (v)[2]
|
||
|
#define EXPAND4(v) (v)[0], (v)[1], (v)[2], (v)[3]
|
||
|
#define SET2(to,from) \
|
||
|
((to)[0] = (from)[0], \
|
||
|
(to)[1] = (from)[1])
|
||
|
#define SETMAT2(to,from) \
|
||
|
(SET2((to)[0], (from)[0]), \
|
||
|
SET2((to)[1], (from)[1]))
|
||
|
#define ROUNDVEC2(to,from) \
|
||
|
((to)[0] = floor((from)[0]+.5), \
|
||
|
(to)[1] = floor((from)[1]+.5))
|
||
|
#define ROUNDMAT2(to,from) \
|
||
|
(ROUNDVEC2((to)[0], (from)[0]), \
|
||
|
ROUNDVEC2((to)[1], (from)[1]))
|
||
|
#define FILLVEC2(v,s) \
|
||
|
((v)[0] = (s), \
|
||
|
(v)[1] = (s))
|
||
|
#define FILLMAT2(m,s) \
|
||
|
(FILLVEC2((m)[0], s), \
|
||
|
FILLVEC2((m)[1], s))
|
||
|
#define ZEROVEC2(v) \
|
||
|
((v)[0] = 0, \
|
||
|
(v)[1] = 0)
|
||
|
#define ISZEROVEC2(v) \
|
||
|
((v)[0] == 0 && \
|
||
|
(v)[1] == 0)
|
||
|
#define EQVEC2(v,w) \
|
||
|
((v)[0] == (w)[0] && \
|
||
|
(v)[1] == (w)[1])
|
||
|
#define EQMAT2(m1,m2) \
|
||
|
(EQVEC2((m1)[0], (m2)[0]) && \
|
||
|
EQVEC2((m1)[1], (m2)[1]))
|
||
|
#define ZEROMAT2(m) \
|
||
|
(ZEROVEC2((m)[0]), \
|
||
|
ZEROVEC2((m)[1]))
|
||
|
#define IDENTMAT2(m) \
|
||
|
(ZEROVEC2((m)[0]), (m)[0][0]=1, \
|
||
|
ZEROVEC2((m)[1]), (m)[1][1]=1)
|
||
|
#define TRANSPOSE2(to,from) \
|
||
|
(_SETcol2((to)[0], from, 0), \
|
||
|
_SETcol2((to)[1], from, 1))
|
||
|
#define VPSXV2(to,v,s,w) \
|
||
|
((to)[0] = (v)[0] + (s) * (w)[0], \
|
||
|
(to)[1] = (v)[1] + (s) * (w)[1])
|
||
|
#define VPVXS2(to,v,w,s) \
|
||
|
((to)[0] = (v)[0] + (w)[0] * (s), \
|
||
|
(to)[1] = (v)[1] + (w)[1] * (s))
|
||
|
#define VPV2(to,v,w) \
|
||
|
((to)[0] = (v)[0] + (w)[0], \
|
||
|
(to)[1] = (v)[1] + (w)[1])
|
||
|
#define VMV2(to,v,w) \
|
||
|
((to)[0] = (v)[0] - (w)[0], \
|
||
|
(to)[1] = (v)[1] - (w)[1])
|
||
|
#define MPM2(to,m1,m2) \
|
||
|
(VPV2((to)[0], (m1)[0], (m2)[0]), \
|
||
|
VPV2((to)[1], (m1)[1], (m2)[1]))
|
||
|
#define MMM2(to,m1,m2) \
|
||
|
(VMV2((to)[0], (m1)[0], (m2)[0]), \
|
||
|
VMV2((to)[1], (m1)[1], (m2)[1]))
|
||
|
#define SXV2(to,s,from) \
|
||
|
((to)[0] = (s) * (from)[0], \
|
||
|
(to)[1] = (s) * (from)[1])
|
||
|
#define SXM2(to,s,from) \
|
||
|
(SXV2((to)[0], s, (from)[0]), \
|
||
|
SXV2((to)[1], s, (from)[1]))
|
||
|
#define MV2(to,from) \
|
||
|
((to)[0] = -(from)[0], \
|
||
|
(to)[1] = -(from)[1])
|
||
|
#define MM2(to,from) \
|
||
|
(MV2((to)[0], (from)[0]), \
|
||
|
MV2((to)[1], (from)[1]))
|
||
|
#define VXS2(to,from,s) \
|
||
|
((to)[0] = (from)[0] * (s), \
|
||
|
(to)[1] = (from)[1] * (s))
|
||
|
#define VDS2(to,from,s) \
|
||
|
((to)[0] = (from)[0] / (s), \
|
||
|
(to)[1] = (from)[1] / (s))
|
||
|
#define MXS2(to,from,s) \
|
||
|
(VXS2((to)[0], (from)[0], s), \
|
||
|
VXS2((to)[1], (from)[1], s))
|
||
|
#define MDS2(to,from,s) \
|
||
|
(VDS2((to)[0], (from)[0], s), \
|
||
|
VDS2((to)[1], (from)[1], s))
|
||
|
#define MXM2(to,m1,m2) \
|
||
|
(VXM2((to)[0], (m1)[0], m2), \
|
||
|
VXM2((to)[1], (m1)[1], m2))
|
||
|
#define VXM2(to,v,m) \
|
||
|
((to)[0] = _DOTcol2(v, m, 0), \
|
||
|
(to)[1] = _DOTcol2(v, m, 1))
|
||
|
#define MXV2(to,m,v) \
|
||
|
((to)[0] = DOT2((m)[0], v), \
|
||
|
(to)[1] = DOT2((m)[1], v))
|
||
|
#define VMODS2(to,v,s) \
|
||
|
((to)[0] = SMODS1((v)[0], s), \
|
||
|
(to)[1] = SMODS1((v)[1], s))
|
||
|
#define VMODV2(to,v0,v1) \
|
||
|
((to)[0] = SMODS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMODS1((v0)[1], (v1)[1]))
|
||
|
#define VDIVS2(to,v,s) \
|
||
|
((to)[0] = SDIVS1((v)[0], s), \
|
||
|
(to)[1] = SDIVS1((v)[1], s))
|
||
|
#define VDIVV2(to,v0,v1) \
|
||
|
((to)[0] = SDIVS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SDIVS1((v0)[1], (v1)[1]))
|
||
|
#define VMINS2(to,v,s) \
|
||
|
((to)[0] = SMINS1((v)[0], s), \
|
||
|
(to)[1] = SMINS1((v)[1], s))
|
||
|
#define VMINV2(to,v0,v1) \
|
||
|
((to)[0] = SMINS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMINS1((v0)[1], (v1)[1]))
|
||
|
#define VMAXS2(to,v,s) \
|
||
|
((to)[0] = SMAXS1((v)[0], s), \
|
||
|
(to)[1] = SMAXS1((v)[1], s))
|
||
|
#define VMAXV2(to,v0,v1) \
|
||
|
((to)[0] = SMAXS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMAXS1((v0)[1], (v1)[1]))
|
||
|
#define LERP2(to,v0,v1,t) \
|
||
|
((to)[0]=(v0)[0]+(t)*((v1)[0]-(v0)[0]), \
|
||
|
(to)[1]=(v0)[1]+(t)*((v1)[1]-(v0)[1]))
|
||
|
#define TRACE2(m) \
|
||
|
((m)[0][0] + \
|
||
|
(m)[1][1])
|
||
|
#define DOT2(v,w) \
|
||
|
((v)[0] * (w)[0] + \
|
||
|
(v)[1] * (w)[1])
|
||
|
#define NORMSQRD2(v) \
|
||
|
((v)[0] * (v)[0] + \
|
||
|
(v)[1] * (v)[1])
|
||
|
#define DISTSQRD2(v,w) \
|
||
|
(((v)[0]-(w)[0])*((v)[0]-(w)[0]) + \
|
||
|
((v)[1]-(w)[1])*((v)[1]-(w)[1]))
|
||
|
#define _DOTcol2(v,m,j) \
|
||
|
((v)[0] * (m)[0][j] + \
|
||
|
(v)[1] * (m)[1][j])
|
||
|
#define _SETcol2(v,m,j) \
|
||
|
((v)[0] = (m)[0][j], \
|
||
|
(v)[1] = (m)[1][j])
|
||
|
#define _MXVcol2(to,m,M,j) \
|
||
|
((to)[0][j] = _DOTcol2((m)[0],M,j), \
|
||
|
(to)[1][j] = _DOTcol2((m)[1],M,j))
|
||
|
#define _DET2(v0,v1,i0,i1) \
|
||
|
((v0)[i0]* _DET1(v1,i1) + \
|
||
|
(v0)[i1]*-_DET1(v1,i0))
|
||
|
#define XV2(to,v1) \
|
||
|
((to)[0] = -_DET1(v1, 1), \
|
||
|
(to)[1] = _DET1(v1, 0))
|
||
|
#define V2XM3(to2,v2,m3) \
|
||
|
((to2)[0] = _DOTcol2(v2,m3,0) + (m3)[2][0], \
|
||
|
(to2)[1] = _DOTcol2(v2,m3,1) + (m3)[2][1])
|
||
|
#define M3XV2(to2,m3,v2) \
|
||
|
((to2)[0] = DOT2((m3)[0],v2) + (m3)[0][2], \
|
||
|
(to2)[1] = DOT2((m3)[1],v2) + (m3)[1][2])
|
||
|
#define _DET1(v0,i0) \
|
||
|
((v0)[i0])
|
||
|
#define VXV2(v0,v1) \
|
||
|
(_DET2(v0,v1,0,1))
|
||
|
#define DET2(m) \
|
||
|
(VXV2((m)[0],(m)[1]))
|
||
|
#define SMODS1(a,b) \
|
||
|
((((a)%(b)+(b))%(b)))
|
||
|
#define SDIVS1(a,b) \
|
||
|
((((a)-SMODS1(a,b))/(b)))
|
||
|
#define SMINS1(a,b) \
|
||
|
(((a) < (b) ? (a) : (b)))
|
||
|
#define SMAXS1(a,b) \
|
||
|
(((a) > (b) ? (a) : (b)))
|
||
|
#define ADJOINT2(to,m) \
|
||
|
( _ADJOINTcol2(to,0,m,1), \
|
||
|
__ADJOINTcol2(to,1,m,0))
|
||
|
#define _ADJOINTcol2(to,col,m,i1) \
|
||
|
((to)[0][col] = _DET1(m[i1], 1), \
|
||
|
(to)[1][col] = -_DET1(m[i1], 0))
|
||
|
#define __ADJOINTcol2(to,col,m,i1) \
|
||
|
((to)[0][col] = -_DET1(m[i1], 1), \
|
||
|
(to)[1][col] = _DET1(m[i1], 0))
|
||
|
#define SET3(to,from) \
|
||
|
((to)[0] = (from)[0], \
|
||
|
(to)[1] = (from)[1], \
|
||
|
(to)[2] = (from)[2])
|
||
|
#define SETMAT3(to,from) \
|
||
|
(SET3((to)[0], (from)[0]), \
|
||
|
SET3((to)[1], (from)[1]), \
|
||
|
SET3((to)[2], (from)[2]))
|
||
|
#define ROUNDVEC3(to,from) \
|
||
|
((to)[0] = floor((from)[0]+.5), \
|
||
|
(to)[1] = floor((from)[1]+.5), \
|
||
|
(to)[2] = floor((from)[2]+.5))
|
||
|
#define ROUNDMAT3(to,from) \
|
||
|
(ROUNDVEC3((to)[0], (from)[0]), \
|
||
|
ROUNDVEC3((to)[1], (from)[1]), \
|
||
|
ROUNDVEC3((to)[2], (from)[2]))
|
||
|
#define FILLVEC3(v,s) \
|
||
|
((v)[0] = (s), \
|
||
|
(v)[1] = (s), \
|
||
|
(v)[2] = (s))
|
||
|
#define FILLMAT3(m,s) \
|
||
|
(FILLVEC3((m)[0], s), \
|
||
|
FILLVEC3((m)[1], s), \
|
||
|
FILLVEC3((m)[2], s))
|
||
|
#define ZEROVEC3(v) \
|
||
|
((v)[0] = 0, \
|
||
|
(v)[1] = 0, \
|
||
|
(v)[2] = 0)
|
||
|
#define ISZEROVEC3(v) \
|
||
|
((v)[0] == 0 && \
|
||
|
(v)[1] == 0 && \
|
||
|
(v)[2] == 0)
|
||
|
#define EQVEC3(v,w) \
|
||
|
((v)[0] == (w)[0] && \
|
||
|
(v)[1] == (w)[1] && \
|
||
|
(v)[2] == (w)[2])
|
||
|
#define EQMAT3(m1,m2) \
|
||
|
(EQVEC3((m1)[0], (m2)[0]) && \
|
||
|
EQVEC3((m1)[1], (m2)[1]) && \
|
||
|
EQVEC3((m1)[2], (m2)[2]))
|
||
|
#define ZEROMAT3(m) \
|
||
|
(ZEROVEC3((m)[0]), \
|
||
|
ZEROVEC3((m)[1]), \
|
||
|
ZEROVEC3((m)[2]))
|
||
|
#define IDENTMAT3(m) \
|
||
|
(ZEROVEC3((m)[0]), (m)[0][0]=1, \
|
||
|
ZEROVEC3((m)[1]), (m)[1][1]=1, \
|
||
|
ZEROVEC3((m)[2]), (m)[2][2]=1)
|
||
|
#define TRANSPOSE3(to,from) \
|
||
|
(_SETcol3((to)[0], from, 0), \
|
||
|
_SETcol3((to)[1], from, 1), \
|
||
|
_SETcol3((to)[2], from, 2))
|
||
|
#define VPSXV3(to,v,s,w) \
|
||
|
((to)[0] = (v)[0] + (s) * (w)[0], \
|
||
|
(to)[1] = (v)[1] + (s) * (w)[1], \
|
||
|
(to)[2] = (v)[2] + (s) * (w)[2])
|
||
|
#define VPVXS3(to,v,w,s) \
|
||
|
((to)[0] = (v)[0] + (w)[0] * (s), \
|
||
|
(to)[1] = (v)[1] + (w)[1] * (s), \
|
||
|
(to)[2] = (v)[2] + (w)[2] * (s))
|
||
|
#define VPV3(to,v,w) \
|
||
|
((to)[0] = (v)[0] + (w)[0], \
|
||
|
(to)[1] = (v)[1] + (w)[1], \
|
||
|
(to)[2] = (v)[2] + (w)[2])
|
||
|
#define VMV3(to,v,w) \
|
||
|
((to)[0] = (v)[0] - (w)[0], \
|
||
|
(to)[1] = (v)[1] - (w)[1], \
|
||
|
(to)[2] = (v)[2] - (w)[2])
|
||
|
#define MPM3(to,m1,m2) \
|
||
|
(VPV3((to)[0], (m1)[0], (m2)[0]), \
|
||
|
VPV3((to)[1], (m1)[1], (m2)[1]), \
|
||
|
VPV3((to)[2], (m1)[2], (m2)[2]))
|
||
|
#define MMM3(to,m1,m2) \
|
||
|
(VMV3((to)[0], (m1)[0], (m2)[0]), \
|
||
|
VMV3((to)[1], (m1)[1], (m2)[1]), \
|
||
|
VMV3((to)[2], (m1)[2], (m2)[2]))
|
||
|
#define SXV3(to,s,from) \
|
||
|
((to)[0] = (s) * (from)[0], \
|
||
|
(to)[1] = (s) * (from)[1], \
|
||
|
(to)[2] = (s) * (from)[2])
|
||
|
#define SXM3(to,s,from) \
|
||
|
(SXV3((to)[0], s, (from)[0]), \
|
||
|
SXV3((to)[1], s, (from)[1]), \
|
||
|
SXV3((to)[2], s, (from)[2]))
|
||
|
#define MV3(to,from) \
|
||
|
((to)[0] = -(from)[0], \
|
||
|
(to)[1] = -(from)[1], \
|
||
|
(to)[2] = -(from)[2])
|
||
|
#define MM3(to,from) \
|
||
|
(MV3((to)[0], (from)[0]), \
|
||
|
MV3((to)[1], (from)[1]), \
|
||
|
MV3((to)[2], (from)[2]))
|
||
|
#define VXS3(to,from,s) \
|
||
|
((to)[0] = (from)[0] * (s), \
|
||
|
(to)[1] = (from)[1] * (s), \
|
||
|
(to)[2] = (from)[2] * (s))
|
||
|
#define VDS3(to,from,s) \
|
||
|
((to)[0] = (from)[0] / (s), \
|
||
|
(to)[1] = (from)[1] / (s), \
|
||
|
(to)[2] = (from)[2] / (s))
|
||
|
#define MXS3(to,from,s) \
|
||
|
(VXS3((to)[0], (from)[0], s), \
|
||
|
VXS3((to)[1], (from)[1], s), \
|
||
|
VXS3((to)[2], (from)[2], s))
|
||
|
#define MDS3(to,from,s) \
|
||
|
(VDS3((to)[0], (from)[0], s), \
|
||
|
VDS3((to)[1], (from)[1], s), \
|
||
|
VDS3((to)[2], (from)[2], s))
|
||
|
#define MXM3(to,m1,m2) \
|
||
|
(VXM3((to)[0], (m1)[0], m2), \
|
||
|
VXM3((to)[1], (m1)[1], m2), \
|
||
|
VXM3((to)[2], (m1)[2], m2))
|
||
|
#define VXM3(to,v,m) \
|
||
|
((to)[0] = _DOTcol3(v, m, 0), \
|
||
|
(to)[1] = _DOTcol3(v, m, 1), \
|
||
|
(to)[2] = _DOTcol3(v, m, 2))
|
||
|
#define MXV3(to,m,v) \
|
||
|
((to)[0] = DOT3((m)[0], v), \
|
||
|
(to)[1] = DOT3((m)[1], v), \
|
||
|
(to)[2] = DOT3((m)[2], v))
|
||
|
#define VMODS3(to,v,s) \
|
||
|
((to)[0] = SMODS1((v)[0], s), \
|
||
|
(to)[1] = SMODS1((v)[1], s), \
|
||
|
(to)[2] = SMODS1((v)[2], s))
|
||
|
#define VMODV3(to,v0,v1) \
|
||
|
((to)[0] = SMODS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMODS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SMODS1((v0)[2], (v1)[2]))
|
||
|
#define VDIVS3(to,v,s) \
|
||
|
((to)[0] = SDIVS1((v)[0], s), \
|
||
|
(to)[1] = SDIVS1((v)[1], s), \
|
||
|
(to)[2] = SDIVS1((v)[2], s))
|
||
|
#define VDIVV3(to,v0,v1) \
|
||
|
((to)[0] = SDIVS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SDIVS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SDIVS1((v0)[2], (v1)[2]))
|
||
|
#define VMINS3(to,v,s) \
|
||
|
((to)[0] = SMINS1((v)[0], s), \
|
||
|
(to)[1] = SMINS1((v)[1], s), \
|
||
|
(to)[2] = SMINS1((v)[2], s))
|
||
|
#define VMINV3(to,v0,v1) \
|
||
|
((to)[0] = SMINS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMINS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SMINS1((v0)[2], (v1)[2]))
|
||
|
#define VMAXS3(to,v,s) \
|
||
|
((to)[0] = SMAXS1((v)[0], s), \
|
||
|
(to)[1] = SMAXS1((v)[1], s), \
|
||
|
(to)[2] = SMAXS1((v)[2], s))
|
||
|
#define VMAXV3(to,v0,v1) \
|
||
|
((to)[0] = SMAXS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMAXS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SMAXS1((v0)[2], (v1)[2]))
|
||
|
#define LERP3(to,v0,v1,t) \
|
||
|
((to)[0]=(v0)[0]+(t)*((v1)[0]-(v0)[0]), \
|
||
|
(to)[1]=(v0)[1]+(t)*((v1)[1]-(v0)[1]), \
|
||
|
(to)[2]=(v0)[2]+(t)*((v1)[2]-(v0)[2]))
|
||
|
#define TRACE3(m) \
|
||
|
((m)[0][0] + \
|
||
|
(m)[1][1] + \
|
||
|
(m)[2][2])
|
||
|
#define DOT3(v,w) \
|
||
|
((v)[0] * (w)[0] + \
|
||
|
(v)[1] * (w)[1] + \
|
||
|
(v)[2] * (w)[2])
|
||
|
#define NORMSQRD3(v) \
|
||
|
((v)[0] * (v)[0] + \
|
||
|
(v)[1] * (v)[1] + \
|
||
|
(v)[2] * (v)[2])
|
||
|
#define DISTSQRD3(v,w) \
|
||
|
(((v)[0]-(w)[0])*((v)[0]-(w)[0]) + \
|
||
|
((v)[1]-(w)[1])*((v)[1]-(w)[1]) + \
|
||
|
((v)[2]-(w)[2])*((v)[2]-(w)[2]))
|
||
|
#define _DOTcol3(v,m,j) \
|
||
|
((v)[0] * (m)[0][j] + \
|
||
|
(v)[1] * (m)[1][j] + \
|
||
|
(v)[2] * (m)[2][j])
|
||
|
#define _SETcol3(v,m,j) \
|
||
|
((v)[0] = (m)[0][j], \
|
||
|
(v)[1] = (m)[1][j], \
|
||
|
(v)[2] = (m)[2][j])
|
||
|
#define _MXVcol3(to,m,M,j) \
|
||
|
((to)[0][j] = _DOTcol3((m)[0],M,j), \
|
||
|
(to)[1][j] = _DOTcol3((m)[1],M,j), \
|
||
|
(to)[2][j] = _DOTcol3((m)[2],M,j))
|
||
|
#define _DET3(v0,v1,v2,i0,i1,i2) \
|
||
|
((v0)[i0]* _DET2(v1,v2,i1,i2) + \
|
||
|
(v0)[i1]*-_DET2(v1,v2,i0,i2) + \
|
||
|
(v0)[i2]* _DET2(v1,v2,i0,i1))
|
||
|
#define VXV3(to,v1,v2) \
|
||
|
((to)[0] = _DET2(v1,v2, 1,2), \
|
||
|
(to)[1] = -_DET2(v1,v2, 0,2), \
|
||
|
(to)[2] = _DET2(v1,v2, 0,1))
|
||
|
#define SET3from2(to,from,pad) \
|
||
|
((to)[0] = (from)[0], \
|
||
|
(to)[1] = (from)[1], \
|
||
|
(to)[2] = (pad))
|
||
|
#define SETMAT3from2(to,from,pad0,pad1) \
|
||
|
(SET3from2((to)[0], (from)[0], pad0), \
|
||
|
SET3from2((to)[1], (from)[1], pad0), \
|
||
|
FILLVEC2((to)[2], (pad0)), (to)[2][2] = (pad1))
|
||
|
#define M2XM3(to3,m2,m3) \
|
||
|
(_MXVcol2(to3,m2,m3,0), (to3)[2][0]=(m3)[2][0], \
|
||
|
_MXVcol2(to3,m2,m3,1), (to3)[2][1]=(m3)[2][1], \
|
||
|
_MXVcol2(to3,m2,m3,2), (to3)[2][2]=(m3)[2][2])
|
||
|
#define M3XM2(to3,m3,m2) \
|
||
|
(VXM2((to3)[0],(m3)[0],m2), (to3)[0][2]=(m3)[0][2], \
|
||
|
VXM2((to3)[1],(m3)[1],m2), (to3)[1][2]=(m3)[1][2], \
|
||
|
VXM2((to3)[2],(m3)[2],m2), (to3)[2][2]=(m3)[2][2])
|
||
|
#define V3XM4(to3,v3,m4) \
|
||
|
((to3)[0] = _DOTcol3(v3,m4,0) + (m4)[3][0], \
|
||
|
(to3)[1] = _DOTcol3(v3,m4,1) + (m4)[3][1], \
|
||
|
(to3)[2] = _DOTcol3(v3,m4,2) + (m4)[3][2])
|
||
|
#define M4XV3(to3,m4,v3) \
|
||
|
((to3)[0] = DOT3((m4)[0],v3) + (m4)[0][3], \
|
||
|
(to3)[1] = DOT3((m4)[1],v3) + (m4)[1][3], \
|
||
|
(to3)[2] = DOT3((m4)[2],v3) + (m4)[2][3])
|
||
|
#define VXVXV3(v0,v1,v2) \
|
||
|
(_DET3(v0,v1,v2,0,1,2))
|
||
|
#define DET3(m) \
|
||
|
(VXVXV3((m)[0],(m)[1],(m)[2]))
|
||
|
#define ADJOINT3(to,m) \
|
||
|
( _ADJOINTcol3(to,0,m,1,2), \
|
||
|
__ADJOINTcol3(to,1,m,0,2), \
|
||
|
_ADJOINTcol3(to,2,m,0,1))
|
||
|
#define _ADJOINTcol3(to,col,m,i1,i2) \
|
||
|
((to)[0][col] = _DET2(m[i1],m[i2], 1,2), \
|
||
|
(to)[1][col] = -_DET2(m[i1],m[i2], 0,2), \
|
||
|
(to)[2][col] = _DET2(m[i1],m[i2], 0,1))
|
||
|
#define __ADJOINTcol3(to,col,m,i1,i2) \
|
||
|
((to)[0][col] = -_DET2(m[i1],m[i2], 1,2), \
|
||
|
(to)[1][col] = _DET2(m[i1],m[i2], 0,2), \
|
||
|
(to)[2][col] = -_DET2(m[i1],m[i2], 0,1))
|
||
|
#define SET4(to,from) \
|
||
|
((to)[0] = (from)[0], \
|
||
|
(to)[1] = (from)[1], \
|
||
|
(to)[2] = (from)[2], \
|
||
|
(to)[3] = (from)[3])
|
||
|
#define SETMAT4(to,from) \
|
||
|
(SET4((to)[0], (from)[0]), \
|
||
|
SET4((to)[1], (from)[1]), \
|
||
|
SET4((to)[2], (from)[2]), \
|
||
|
SET4((to)[3], (from)[3]))
|
||
|
#define ROUNDVEC4(to,from) \
|
||
|
((to)[0] = floor((from)[0]+.5), \
|
||
|
(to)[1] = floor((from)[1]+.5), \
|
||
|
(to)[2] = floor((from)[2]+.5), \
|
||
|
(to)[3] = floor((from)[3]+.5))
|
||
|
#define ROUNDMAT4(to,from) \
|
||
|
(ROUNDVEC4((to)[0], (from)[0]), \
|
||
|
ROUNDVEC4((to)[1], (from)[1]), \
|
||
|
ROUNDVEC4((to)[2], (from)[2]), \
|
||
|
ROUNDVEC4((to)[3], (from)[3]))
|
||
|
#define FILLVEC4(v,s) \
|
||
|
((v)[0] = (s), \
|
||
|
(v)[1] = (s), \
|
||
|
(v)[2] = (s), \
|
||
|
(v)[3] = (s))
|
||
|
#define FILLMAT4(m,s) \
|
||
|
(FILLVEC4((m)[0], s), \
|
||
|
FILLVEC4((m)[1], s), \
|
||
|
FILLVEC4((m)[2], s), \
|
||
|
FILLVEC4((m)[3], s))
|
||
|
#define ZEROVEC4(v) \
|
||
|
((v)[0] = 0, \
|
||
|
(v)[1] = 0, \
|
||
|
(v)[2] = 0, \
|
||
|
(v)[3] = 0)
|
||
|
#define ISZEROVEC4(v) \
|
||
|
((v)[0] == 0 && \
|
||
|
(v)[1] == 0 && \
|
||
|
(v)[2] == 0 && \
|
||
|
(v)[3] == 0)
|
||
|
#define EQVEC4(v,w) \
|
||
|
((v)[0] == (w)[0] && \
|
||
|
(v)[1] == (w)[1] && \
|
||
|
(v)[2] == (w)[2] && \
|
||
|
(v)[3] == (w)[3])
|
||
|
#define EQMAT4(m1,m2) \
|
||
|
(EQVEC4((m1)[0], (m2)[0]) && \
|
||
|
EQVEC4((m1)[1], (m2)[1]) && \
|
||
|
EQVEC4((m1)[2], (m2)[2]) && \
|
||
|
EQVEC4((m1)[3], (m2)[3]))
|
||
|
#define ZEROMAT4(m) \
|
||
|
(ZEROVEC4((m)[0]), \
|
||
|
ZEROVEC4((m)[1]), \
|
||
|
ZEROVEC4((m)[2]), \
|
||
|
ZEROVEC4((m)[3]))
|
||
|
#define IDENTMAT4(m) \
|
||
|
(ZEROVEC4((m)[0]), (m)[0][0]=1, \
|
||
|
ZEROVEC4((m)[1]), (m)[1][1]=1, \
|
||
|
ZEROVEC4((m)[2]), (m)[2][2]=1, \
|
||
|
ZEROVEC4((m)[3]), (m)[3][3]=1)
|
||
|
#define TRANSPOSE4(to,from) \
|
||
|
(_SETcol4((to)[0], from, 0), \
|
||
|
_SETcol4((to)[1], from, 1), \
|
||
|
_SETcol4((to)[2], from, 2), \
|
||
|
_SETcol4((to)[3], from, 3))
|
||
|
#define VPSXV4(to,v,s,w) \
|
||
|
((to)[0] = (v)[0] + (s) * (w)[0], \
|
||
|
(to)[1] = (v)[1] + (s) * (w)[1], \
|
||
|
(to)[2] = (v)[2] + (s) * (w)[2], \
|
||
|
(to)[3] = (v)[3] + (s) * (w)[3])
|
||
|
#define VPVXS4(to,v,w,s) \
|
||
|
((to)[0] = (v)[0] + (w)[0] * (s), \
|
||
|
(to)[1] = (v)[1] + (w)[1] * (s), \
|
||
|
(to)[2] = (v)[2] + (w)[2] * (s), \
|
||
|
(to)[3] = (v)[3] + (w)[3] * (s))
|
||
|
#define VPV4(to,v,w) \
|
||
|
((to)[0] = (v)[0] + (w)[0], \
|
||
|
(to)[1] = (v)[1] + (w)[1], \
|
||
|
(to)[2] = (v)[2] + (w)[2], \
|
||
|
(to)[3] = (v)[3] + (w)[3])
|
||
|
#define VMV4(to,v,w) \
|
||
|
((to)[0] = (v)[0] - (w)[0], \
|
||
|
(to)[1] = (v)[1] - (w)[1], \
|
||
|
(to)[2] = (v)[2] - (w)[2], \
|
||
|
(to)[3] = (v)[3] - (w)[3])
|
||
|
#define MPM4(to,m1,m2) \
|
||
|
(VPV4((to)[0], (m1)[0], (m2)[0]), \
|
||
|
VPV4((to)[1], (m1)[1], (m2)[1]), \
|
||
|
VPV4((to)[2], (m1)[2], (m2)[2]), \
|
||
|
VPV4((to)[3], (m1)[3], (m2)[3]))
|
||
|
#define MMM4(to,m1,m2) \
|
||
|
(VMV4((to)[0], (m1)[0], (m2)[0]), \
|
||
|
VMV4((to)[1], (m1)[1], (m2)[1]), \
|
||
|
VMV4((to)[2], (m1)[2], (m2)[2]), \
|
||
|
VMV4((to)[3], (m1)[3], (m2)[3]))
|
||
|
#define SXV4(to,s,from) \
|
||
|
((to)[0] = (s) * (from)[0], \
|
||
|
(to)[1] = (s) * (from)[1], \
|
||
|
(to)[2] = (s) * (from)[2], \
|
||
|
(to)[3] = (s) * (from)[3])
|
||
|
#define SXM4(to,s,from) \
|
||
|
(SXV4((to)[0], s, (from)[0]), \
|
||
|
SXV4((to)[1], s, (from)[1]), \
|
||
|
SXV4((to)[2], s, (from)[2]), \
|
||
|
SXV4((to)[3], s, (from)[3]))
|
||
|
#define MV4(to,from) \
|
||
|
((to)[0] = -(from)[0], \
|
||
|
(to)[1] = -(from)[1], \
|
||
|
(to)[2] = -(from)[2], \
|
||
|
(to)[3] = -(from)[3])
|
||
|
#define MM4(to,from) \
|
||
|
(MV4((to)[0], (from)[0]), \
|
||
|
MV4((to)[1], (from)[1]), \
|
||
|
MV4((to)[2], (from)[2]), \
|
||
|
MV4((to)[3], (from)[3]))
|
||
|
#define VXS4(to,from,s) \
|
||
|
((to)[0] = (from)[0] * (s), \
|
||
|
(to)[1] = (from)[1] * (s), \
|
||
|
(to)[2] = (from)[2] * (s), \
|
||
|
(to)[3] = (from)[3] * (s))
|
||
|
#define VDS4(to,from,s) \
|
||
|
((to)[0] = (from)[0] / (s), \
|
||
|
(to)[1] = (from)[1] / (s), \
|
||
|
(to)[2] = (from)[2] / (s), \
|
||
|
(to)[3] = (from)[3] / (s))
|
||
|
#define MXS4(to,from,s) \
|
||
|
(VXS4((to)[0], (from)[0], s), \
|
||
|
VXS4((to)[1], (from)[1], s), \
|
||
|
VXS4((to)[2], (from)[2], s), \
|
||
|
VXS4((to)[3], (from)[3], s))
|
||
|
#define MDS4(to,from,s) \
|
||
|
(VDS4((to)[0], (from)[0], s), \
|
||
|
VDS4((to)[1], (from)[1], s), \
|
||
|
VDS4((to)[2], (from)[2], s), \
|
||
|
VDS4((to)[3], (from)[3], s))
|
||
|
#define MXM4(to,m1,m2) \
|
||
|
(VXM4((to)[0], (m1)[0], m2), \
|
||
|
VXM4((to)[1], (m1)[1], m2), \
|
||
|
VXM4((to)[2], (m1)[2], m2), \
|
||
|
VXM4((to)[3], (m1)[3], m2))
|
||
|
#define VXM4(to,v,m) \
|
||
|
((to)[0] = _DOTcol4(v, m, 0), \
|
||
|
(to)[1] = _DOTcol4(v, m, 1), \
|
||
|
(to)[2] = _DOTcol4(v, m, 2), \
|
||
|
(to)[3] = _DOTcol4(v, m, 3))
|
||
|
#define MXV4(to,m,v) \
|
||
|
((to)[0] = DOT4((m)[0], v), \
|
||
|
(to)[1] = DOT4((m)[1], v), \
|
||
|
(to)[2] = DOT4((m)[2], v), \
|
||
|
(to)[3] = DOT4((m)[3], v))
|
||
|
#define VMODS4(to,v,s) \
|
||
|
((to)[0] = SMODS1((v)[0], s), \
|
||
|
(to)[1] = SMODS1((v)[1], s), \
|
||
|
(to)[2] = SMODS1((v)[2], s), \
|
||
|
(to)[3] = SMODS1((v)[3], s))
|
||
|
#define VMODV4(to,v0,v1) \
|
||
|
((to)[0] = SMODS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMODS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SMODS1((v0)[2], (v1)[2]), \
|
||
|
(to)[3] = SMODS1((v0)[3], (v1)[3]))
|
||
|
#define VDIVS4(to,v,s) \
|
||
|
((to)[0] = SDIVS1((v)[0], s), \
|
||
|
(to)[1] = SDIVS1((v)[1], s), \
|
||
|
(to)[2] = SDIVS1((v)[2], s), \
|
||
|
(to)[3] = SDIVS1((v)[3], s))
|
||
|
#define VDIVV4(to,v0,v1) \
|
||
|
((to)[0] = SDIVS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SDIVS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SDIVS1((v0)[2], (v1)[2]), \
|
||
|
(to)[3] = SDIVS1((v0)[3], (v1)[3]))
|
||
|
#define VMINS4(to,v,s) \
|
||
|
((to)[0] = SMINS1((v)[0], s), \
|
||
|
(to)[1] = SMINS1((v)[1], s), \
|
||
|
(to)[2] = SMINS1((v)[2], s), \
|
||
|
(to)[3] = SMINS1((v)[3], s))
|
||
|
#define VMINV4(to,v0,v1) \
|
||
|
((to)[0] = SMINS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMINS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SMINS1((v0)[2], (v1)[2]), \
|
||
|
(to)[3] = SMINS1((v0)[3], (v1)[3]))
|
||
|
#define VMAXS4(to,v,s) \
|
||
|
((to)[0] = SMAXS1((v)[0], s), \
|
||
|
(to)[1] = SMAXS1((v)[1], s), \
|
||
|
(to)[2] = SMAXS1((v)[2], s), \
|
||
|
(to)[3] = SMAXS1((v)[3], s))
|
||
|
#define VMAXV4(to,v0,v1) \
|
||
|
((to)[0] = SMAXS1((v0)[0], (v1)[0]), \
|
||
|
(to)[1] = SMAXS1((v0)[1], (v1)[1]), \
|
||
|
(to)[2] = SMAXS1((v0)[2], (v1)[2]), \
|
||
|
(to)[3] = SMAXS1((v0)[3], (v1)[3]))
|
||
|
#define LERP4(to,v0,v1,t) \
|
||
|
((to)[0]=(v0)[0]+(t)*((v1)[0]-(v0)[0]), \
|
||
|
(to)[1]=(v0)[1]+(t)*((v1)[1]-(v0)[1]), \
|
||
|
(to)[2]=(v0)[2]+(t)*((v1)[2]-(v0)[2]), \
|
||
|
(to)[3]=(v0)[3]+(t)*((v1)[3]-(v0)[3]))
|
||
|
#define TRACE4(m) \
|
||
|
((m)[0][0] + \
|
||
|
(m)[1][1] + \
|
||
|
(m)[2][2] + \
|
||
|
(m)[3][3])
|
||
|
#define DOT4(v,w) \
|
||
|
((v)[0] * (w)[0] + \
|
||
|
(v)[1] * (w)[1] + \
|
||
|
(v)[2] * (w)[2] + \
|
||
|
(v)[3] * (w)[3])
|
||
|
#define NORMSQRD4(v) \
|
||
|
((v)[0] * (v)[0] + \
|
||
|
(v)[1] * (v)[1] + \
|
||
|
(v)[2] * (v)[2] + \
|
||
|
(v)[3] * (v)[3])
|
||
|
#define DISTSQRD4(v,w) \
|
||
|
(((v)[0]-(w)[0])*((v)[0]-(w)[0]) + \
|
||
|
((v)[1]-(w)[1])*((v)[1]-(w)[1]) + \
|
||
|
((v)[2]-(w)[2])*((v)[2]-(w)[2]) + \
|
||
|
((v)[3]-(w)[3])*((v)[3]-(w)[3]))
|
||
|
#define _DOTcol4(v,m,j) \
|
||
|
((v)[0] * (m)[0][j] + \
|
||
|
(v)[1] * (m)[1][j] + \
|
||
|
(v)[2] * (m)[2][j] + \
|
||
|
(v)[3] * (m)[3][j])
|
||
|
#define _SETcol4(v,m,j) \
|
||
|
((v)[0] = (m)[0][j], \
|
||
|
(v)[1] = (m)[1][j], \
|
||
|
(v)[2] = (m)[2][j], \
|
||
|
(v)[3] = (m)[3][j])
|
||
|
#define _MXVcol4(to,m,M,j) \
|
||
|
((to)[0][j] = _DOTcol4((m)[0],M,j), \
|
||
|
(to)[1][j] = _DOTcol4((m)[1],M,j), \
|
||
|
(to)[2][j] = _DOTcol4((m)[2],M,j), \
|
||
|
(to)[3][j] = _DOTcol4((m)[3],M,j))
|
||
|
#define _DET4(v0,v1,v2,v3,i0,i1,i2,i3) \
|
||
|
((v0)[i0]* _DET3(v1,v2,v3,i1,i2,i3) + \
|
||
|
(v0)[i1]*-_DET3(v1,v2,v3,i0,i2,i3) + \
|
||
|
(v0)[i2]* _DET3(v1,v2,v3,i0,i1,i3) + \
|
||
|
(v0)[i3]*-_DET3(v1,v2,v3,i0,i1,i2))
|
||
|
#define VXVXV4(to,v1,v2,v3) \
|
||
|
((to)[0] = -_DET3(v1,v2,v3, 1,2,3), \
|
||
|
(to)[1] = _DET3(v1,v2,v3, 0,2,3), \
|
||
|
(to)[2] = -_DET3(v1,v2,v3, 0,1,3), \
|
||
|
(to)[3] = _DET3(v1,v2,v3, 0,1,2))
|
||
|
#define SET4from3(to,from,pad) \
|
||
|
((to)[0] = (from)[0], \
|
||
|
(to)[1] = (from)[1], \
|
||
|
(to)[2] = (from)[2], \
|
||
|
(to)[3] = (pad))
|
||
|
#define SETMAT4from3(to,from,pad0,pad1) \
|
||
|
(SET4from3((to)[0], (from)[0], pad0), \
|
||
|
SET4from3((to)[1], (from)[1], pad0), \
|
||
|
SET4from3((to)[2], (from)[2], pad0), \
|
||
|
FILLVEC3((to)[3], (pad0)), (to)[3][3] = (pad1))
|
||
|
#define M3XM4(to4,m3,m4) \
|
||
|
(_MXVcol3(to4,m3,m4,0), (to4)[3][0]=(m4)[3][0], \
|
||
|
_MXVcol3(to4,m3,m4,1), (to4)[3][1]=(m4)[3][1], \
|
||
|
_MXVcol3(to4,m3,m4,2), (to4)[3][2]=(m4)[3][2], \
|
||
|
_MXVcol3(to4,m3,m4,3), (to4)[3][3]=(m4)[3][3])
|
||
|
#define M4XM3(to4,m4,m3) \
|
||
|
(VXM3((to4)[0],(m4)[0],m3), (to4)[0][3]=(m4)[0][3], \
|
||
|
VXM3((to4)[1],(m4)[1],m3), (to4)[1][3]=(m4)[1][3], \
|
||
|
VXM3((to4)[2],(m4)[2],m3), (to4)[2][3]=(m4)[2][3], \
|
||
|
VXM3((to4)[3],(m4)[3],m3), (to4)[3][3]=(m4)[3][3])
|
||
|
#define VXVXVXV4(v0,v1,v2,v3) \
|
||
|
(_DET4(v0,v1,v2,v3,0,1,2,3))
|
||
|
#define DET4(m) \
|
||
|
(VXVXVXV4((m)[0],(m)[1],(m)[2],(m)[3]))
|
||
|
#define ADJOINT4(to,m) \
|
||
|
( _ADJOINTcol4(to,0,m,1,2,3), \
|
||
|
__ADJOINTcol4(to,1,m,0,2,3), \
|
||
|
_ADJOINTcol4(to,2,m,0,1,3), \
|
||
|
__ADJOINTcol4(to,3,m,0,1,2))
|
||
|
#define _ADJOINTcol4(to,col,m,i1,i2,i3) \
|
||
|
((to)[0][col] = _DET3(m[i1],m[i2],m[i3], 1,2,3), \
|
||
|
(to)[1][col] = -_DET3(m[i1],m[i2],m[i3], 0,2,3), \
|
||
|
(to)[2][col] = _DET3(m[i1],m[i2],m[i3], 0,1,3), \
|
||
|
(to)[3][col] = -_DET3(m[i1],m[i2],m[i3], 0,1,2))
|
||
|
#define __ADJOINTcol4(to,col,m,i1,i2,i3) \
|
||
|
((to)[0][col] = -_DET3(m[i1],m[i2],m[i3], 1,2,3), \
|
||
|
(to)[1][col] = _DET3(m[i1],m[i2],m[i3], 0,2,3), \
|
||
|
(to)[2][col] = -_DET3(m[i1],m[i2],m[i3], 0,1,3), \
|
||
|
(to)[3][col] = _DET3(m[i1],m[i2],m[i3], 0,1,2))
|
||
|
#define TRANSPOSE2safe(type,to,from) \
|
||
|
do {type _vec_h_temp_[2][2]; \
|
||
|
TRANSPOSE2(_vec_h_temp_,from); \
|
||
|
SETMAT2(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define TRANSPOSE2d(to,from) TRANSPOSE2safe(double,to,from)
|
||
|
#define TRANSPOSE2i(to,from) TRANSPOSE2safe(int,to,from)
|
||
|
#define TRANSPOSE2l(to,from) TRANSPOSE2safe(long,to,from)
|
||
|
#define TRANSPOSE2r(to,from) TRANSPOSE2safe(real,to,from)
|
||
|
#define MXM2safe(type,to,m1,m2) \
|
||
|
do {type _vec_h_temp_[2][2]; \
|
||
|
MXM2(_vec_h_temp_,m1,m2); \
|
||
|
SETMAT2(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define MXM2d(to,m1,m2) MXM2safe(double,to,m1,m2)
|
||
|
#define MXM2i(to,m1,m2) MXM2safe(int,to,m1,m2)
|
||
|
#define MXM2l(to,m1,m2) MXM2safe(long,to,m1,m2)
|
||
|
#define MXM2r(to,m1,m2) MXM2safe(real,to,m1,m2)
|
||
|
#define VXM2safe(type,to,v,m) \
|
||
|
do {type _vec_h_temp_[2]; \
|
||
|
VXM2(_vec_h_temp_,v,m); \
|
||
|
SET2(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define VXM2d(to,v,m) VXM2safe(double,to,v,m)
|
||
|
#define VXM2i(to,v,m) VXM2safe(int,to,v,m)
|
||
|
#define VXM2l(to,v,m) VXM2safe(long,to,v,m)
|
||
|
#define VXM2r(to,v,m) VXM2safe(real,to,v,m)
|
||
|
#define MXV2safe(type,to,m,v) \
|
||
|
do {type _vec_h_temp_[2]; \
|
||
|
MXV2(_vec_h_temp_,m,v); \
|
||
|
SET2(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define MXV2d(to,m,v) MXV2safe(double,to,m,v)
|
||
|
#define MXV2i(to,m,v) MXV2safe(int,to,m,v)
|
||
|
#define MXV2l(to,m,v) MXV2safe(long,to,m,v)
|
||
|
#define MXV2r(to,m,v) MXV2safe(real,to,m,v)
|
||
|
#define XV2safe(type,to,v1) \
|
||
|
do {type _vec_h_temp_[2]; \
|
||
|
XV2(_vec_h_temp_,v1); \
|
||
|
SET2(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define XV2d(to,v1) XV2safe(double,to,v1)
|
||
|
#define XV2i(to,v1) XV2safe(int,to,v1)
|
||
|
#define XV2l(to,v1) XV2safe(long,to,v1)
|
||
|
#define XV2r(to,v1) XV2safe(real,to,v1)
|
||
|
#define V2XM3safe(type,to2,v2,m3) \
|
||
|
do {type _vec_h_temp_[2]; \
|
||
|
V2XM3(_vec_h_temp_,v2,m3); \
|
||
|
SET2(to2, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define V2XM3d(to2,v2,m3) V2XM3safe(double,to2,v2,m3)
|
||
|
#define V2XM3i(to2,v2,m3) V2XM3safe(int,to2,v2,m3)
|
||
|
#define V2XM3l(to2,v2,m3) V2XM3safe(long,to2,v2,m3)
|
||
|
#define V2XM3r(to2,v2,m3) V2XM3safe(real,to2,v2,m3)
|
||
|
#define M3XV2safe(type,to2,m3,v2) \
|
||
|
do {type _vec_h_temp_[2]; \
|
||
|
M3XV2(_vec_h_temp_,m3,v2); \
|
||
|
SET2(to2, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define M3XV2d(to2,m3,v2) M3XV2safe(double,to2,m3,v2)
|
||
|
#define M3XV2i(to2,m3,v2) M3XV2safe(int,to2,m3,v2)
|
||
|
#define M3XV2l(to2,m3,v2) M3XV2safe(long,to2,m3,v2)
|
||
|
#define M3XV2r(to2,m3,v2) M3XV2safe(real,to2,m3,v2)
|
||
|
#define ADJOINT2safe(type,to,m) \
|
||
|
do {type _vec_h_temp_[2][2]; \
|
||
|
ADJOINT2(_vec_h_temp_,m); \
|
||
|
SETMAT2(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define ADJOINT2d(to,m) ADJOINT2safe(double,to,m)
|
||
|
#define ADJOINT2i(to,m) ADJOINT2safe(int,to,m)
|
||
|
#define ADJOINT2l(to,m) ADJOINT2safe(long,to,m)
|
||
|
#define ADJOINT2r(to,m) ADJOINT2safe(real,to,m)
|
||
|
#define INVERTMAT2safe(type,to,from) \
|
||
|
do {type _vec_h_temp_[2][2]; \
|
||
|
ADJOINT2(_vec_h_temp_, from); \
|
||
|
type _vec_h_temp_invdet_ = (type)1/(type)DET2(from); \
|
||
|
SXM2(to, _vec_h_temp_invdet_, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define INVERTMAT2d(to,from) INVERTMAT2safe(double,to,from)
|
||
|
#define INVERTMAT2i(to,from) INVERTMAT2safe(int,to,from)
|
||
|
#define INVERTMAT2l(to,from) INVERTMAT2safe(long,to,from)
|
||
|
#define INVERTMAT2r(to,from) INVERTMAT2safe(real,to,from)
|
||
|
#define TRANSPOSE3safe(type,to,from) \
|
||
|
do {type _vec_h_temp_[3][3]; \
|
||
|
TRANSPOSE3(_vec_h_temp_,from); \
|
||
|
SETMAT3(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define TRANSPOSE3d(to,from) TRANSPOSE3safe(double,to,from)
|
||
|
#define TRANSPOSE3i(to,from) TRANSPOSE3safe(int,to,from)
|
||
|
#define TRANSPOSE3l(to,from) TRANSPOSE3safe(long,to,from)
|
||
|
#define TRANSPOSE3r(to,from) TRANSPOSE3safe(real,to,from)
|
||
|
#define MXM3safe(type,to,m1,m2) \
|
||
|
do {type _vec_h_temp_[3][3]; \
|
||
|
MXM3(_vec_h_temp_,m1,m2); \
|
||
|
SETMAT3(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define MXM3d(to,m1,m2) MXM3safe(double,to,m1,m2)
|
||
|
#define MXM3i(to,m1,m2) MXM3safe(int,to,m1,m2)
|
||
|
#define MXM3l(to,m1,m2) MXM3safe(long,to,m1,m2)
|
||
|
#define MXM3r(to,m1,m2) MXM3safe(real,to,m1,m2)
|
||
|
#define VXM3safe(type,to,v,m) \
|
||
|
do {type _vec_h_temp_[3]; \
|
||
|
VXM3(_vec_h_temp_,v,m); \
|
||
|
SET3(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define VXM3d(to,v,m) VXM3safe(double,to,v,m)
|
||
|
#define VXM3i(to,v,m) VXM3safe(int,to,v,m)
|
||
|
#define VXM3l(to,v,m) VXM3safe(long,to,v,m)
|
||
|
#define VXM3r(to,v,m) VXM3safe(real,to,v,m)
|
||
|
#define MXV3safe(type,to,m,v) \
|
||
|
do {type _vec_h_temp_[3]; \
|
||
|
MXV3(_vec_h_temp_,m,v); \
|
||
|
SET3(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define MXV3d(to,m,v) MXV3safe(double,to,m,v)
|
||
|
#define MXV3i(to,m,v) MXV3safe(int,to,m,v)
|
||
|
#define MXV3l(to,m,v) MXV3safe(long,to,m,v)
|
||
|
#define MXV3r(to,m,v) MXV3safe(real,to,m,v)
|
||
|
#define VXV3safe(type,to,v1,v2) \
|
||
|
do {type _vec_h_temp_[3]; \
|
||
|
VXV3(_vec_h_temp_,v1,v2); \
|
||
|
SET3(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define VXV3d(to,v1,v2) VXV3safe(double,to,v1,v2)
|
||
|
#define VXV3i(to,v1,v2) VXV3safe(int,to,v1,v2)
|
||
|
#define VXV3l(to,v1,v2) VXV3safe(long,to,v1,v2)
|
||
|
#define VXV3r(to,v1,v2) VXV3safe(real,to,v1,v2)
|
||
|
#define M2XM3safe(type,to3,m2,m3) \
|
||
|
do {type _vec_h_temp_[3][3]; \
|
||
|
M2XM3(_vec_h_temp_,m2,m3); \
|
||
|
SETMAT3(to3, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define M2XM3d(to3,m2,m3) M2XM3safe(double,to3,m2,m3)
|
||
|
#define M2XM3i(to3,m2,m3) M2XM3safe(int,to3,m2,m3)
|
||
|
#define M2XM3l(to3,m2,m3) M2XM3safe(long,to3,m2,m3)
|
||
|
#define M2XM3r(to3,m2,m3) M2XM3safe(real,to3,m2,m3)
|
||
|
#define M3XM2safe(type,to3,m3,m2) \
|
||
|
do {type _vec_h_temp_[3][3]; \
|
||
|
M3XM2(_vec_h_temp_,m3,m2); \
|
||
|
SETMAT3(to3, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define M3XM2d(to3,m3,m2) M3XM2safe(double,to3,m3,m2)
|
||
|
#define M3XM2i(to3,m3,m2) M3XM2safe(int,to3,m3,m2)
|
||
|
#define M3XM2l(to3,m3,m2) M3XM2safe(long,to3,m3,m2)
|
||
|
#define M3XM2r(to3,m3,m2) M3XM2safe(real,to3,m3,m2)
|
||
|
#define V3XM4safe(type,to3,v3,m4) \
|
||
|
do {type _vec_h_temp_[3]; \
|
||
|
V3XM4(_vec_h_temp_,v3,m4); \
|
||
|
SET3(to3, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define V3XM4d(to3,v3,m4) V3XM4safe(double,to3,v3,m4)
|
||
|
#define V3XM4i(to3,v3,m4) V3XM4safe(int,to3,v3,m4)
|
||
|
#define V3XM4l(to3,v3,m4) V3XM4safe(long,to3,v3,m4)
|
||
|
#define V3XM4r(to3,v3,m4) V3XM4safe(real,to3,v3,m4)
|
||
|
#define M4XV3safe(type,to3,m4,v3) \
|
||
|
do {type _vec_h_temp_[3]; \
|
||
|
M4XV3(_vec_h_temp_,m4,v3); \
|
||
|
SET3(to3, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define M4XV3d(to3,m4,v3) M4XV3safe(double,to3,m4,v3)
|
||
|
#define M4XV3i(to3,m4,v3) M4XV3safe(int,to3,m4,v3)
|
||
|
#define M4XV3l(to3,m4,v3) M4XV3safe(long,to3,m4,v3)
|
||
|
#define M4XV3r(to3,m4,v3) M4XV3safe(real,to3,m4,v3)
|
||
|
#define ADJOINT3safe(type,to,m) \
|
||
|
do {type _vec_h_temp_[3][3]; \
|
||
|
ADJOINT3(_vec_h_temp_,m); \
|
||
|
SETMAT3(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define ADJOINT3d(to,m) ADJOINT3safe(double,to,m)
|
||
|
#define ADJOINT3i(to,m) ADJOINT3safe(int,to,m)
|
||
|
#define ADJOINT3l(to,m) ADJOINT3safe(long,to,m)
|
||
|
#define ADJOINT3r(to,m) ADJOINT3safe(real,to,m)
|
||
|
#define INVERTMAT3safe(type,to,from) \
|
||
|
do {type _vec_h_temp_[3][3]; \
|
||
|
ADJOINT3(_vec_h_temp_, from); \
|
||
|
type _vec_h_temp_invdet_ = (type)1/(type)DET3(from); \
|
||
|
SXM3(to, _vec_h_temp_invdet_, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define INVERTMAT3d(to,from) INVERTMAT3safe(double,to,from)
|
||
|
#define INVERTMAT3i(to,from) INVERTMAT3safe(int,to,from)
|
||
|
#define INVERTMAT3l(to,from) INVERTMAT3safe(long,to,from)
|
||
|
#define INVERTMAT3r(to,from) INVERTMAT3safe(real,to,from)
|
||
|
#define TRANSPOSE4safe(type,to,from) \
|
||
|
do {type _vec_h_temp_[4][4]; \
|
||
|
TRANSPOSE4(_vec_h_temp_,from); \
|
||
|
SETMAT4(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define TRANSPOSE4d(to,from) TRANSPOSE4safe(double,to,from)
|
||
|
#define TRANSPOSE4i(to,from) TRANSPOSE4safe(int,to,from)
|
||
|
#define TRANSPOSE4l(to,from) TRANSPOSE4safe(long,to,from)
|
||
|
#define TRANSPOSE4r(to,from) TRANSPOSE4safe(real,to,from)
|
||
|
#define MXM4safe(type,to,m1,m2) \
|
||
|
do {type _vec_h_temp_[4][4]; \
|
||
|
MXM4(_vec_h_temp_,m1,m2); \
|
||
|
SETMAT4(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define MXM4d(to,m1,m2) MXM4safe(double,to,m1,m2)
|
||
|
#define MXM4i(to,m1,m2) MXM4safe(int,to,m1,m2)
|
||
|
#define MXM4l(to,m1,m2) MXM4safe(long,to,m1,m2)
|
||
|
#define MXM4r(to,m1,m2) MXM4safe(real,to,m1,m2)
|
||
|
#define VXM4safe(type,to,v,m) \
|
||
|
do {type _vec_h_temp_[4]; \
|
||
|
VXM4(_vec_h_temp_,v,m); \
|
||
|
SET4(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define VXM4d(to,v,m) VXM4safe(double,to,v,m)
|
||
|
#define VXM4i(to,v,m) VXM4safe(int,to,v,m)
|
||
|
#define VXM4l(to,v,m) VXM4safe(long,to,v,m)
|
||
|
#define VXM4r(to,v,m) VXM4safe(real,to,v,m)
|
||
|
#define MXV4safe(type,to,m,v) \
|
||
|
do {type _vec_h_temp_[4]; \
|
||
|
MXV4(_vec_h_temp_,m,v); \
|
||
|
SET4(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define MXV4d(to,m,v) MXV4safe(double,to,m,v)
|
||
|
#define MXV4i(to,m,v) MXV4safe(int,to,m,v)
|
||
|
#define MXV4l(to,m,v) MXV4safe(long,to,m,v)
|
||
|
#define MXV4r(to,m,v) MXV4safe(real,to,m,v)
|
||
|
#define VXVXV4safe(type,to,v1,v2,v3) \
|
||
|
do {type _vec_h_temp_[4]; \
|
||
|
VXVXV4(_vec_h_temp_,v1,v2,v3); \
|
||
|
SET4(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define VXVXV4d(to,v1,v2,v3) VXVXV4safe(double,to,v1,v2,v3)
|
||
|
#define VXVXV4i(to,v1,v2,v3) VXVXV4safe(int,to,v1,v2,v3)
|
||
|
#define VXVXV4l(to,v1,v2,v3) VXVXV4safe(long,to,v1,v2,v3)
|
||
|
#define VXVXV4r(to,v1,v2,v3) VXVXV4safe(real,to,v1,v2,v3)
|
||
|
#define M3XM4safe(type,to4,m3,m4) \
|
||
|
do {type _vec_h_temp_[4][4]; \
|
||
|
M3XM4(_vec_h_temp_,m3,m4); \
|
||
|
SETMAT4(to4, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define M3XM4d(to4,m3,m4) M3XM4safe(double,to4,m3,m4)
|
||
|
#define M3XM4i(to4,m3,m4) M3XM4safe(int,to4,m3,m4)
|
||
|
#define M3XM4l(to4,m3,m4) M3XM4safe(long,to4,m3,m4)
|
||
|
#define M3XM4r(to4,m3,m4) M3XM4safe(real,to4,m3,m4)
|
||
|
#define M4XM3safe(type,to4,m4,m3) \
|
||
|
do {type _vec_h_temp_[4][4]; \
|
||
|
M4XM3(_vec_h_temp_,m4,m3); \
|
||
|
SETMAT4(to4, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define M4XM3d(to4,m4,m3) M4XM3safe(double,to4,m4,m3)
|
||
|
#define M4XM3i(to4,m4,m3) M4XM3safe(int,to4,m4,m3)
|
||
|
#define M4XM3l(to4,m4,m3) M4XM3safe(long,to4,m4,m3)
|
||
|
#define M4XM3r(to4,m4,m3) M4XM3safe(real,to4,m4,m3)
|
||
|
#define ADJOINT4safe(type,to,m) \
|
||
|
do {type _vec_h_temp_[4][4]; \
|
||
|
ADJOINT4(_vec_h_temp_,m); \
|
||
|
SETMAT4(to, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define ADJOINT4d(to,m) ADJOINT4safe(double,to,m)
|
||
|
#define ADJOINT4i(to,m) ADJOINT4safe(int,to,m)
|
||
|
#define ADJOINT4l(to,m) ADJOINT4safe(long,to,m)
|
||
|
#define ADJOINT4r(to,m) ADJOINT4safe(real,to,m)
|
||
|
#define INVERTMAT4safe(type,to,from) \
|
||
|
do {type _vec_h_temp_[4][4]; \
|
||
|
ADJOINT4(_vec_h_temp_, from); \
|
||
|
type _vec_h_temp_invdet_ = (type)1/(type)DET4(from); \
|
||
|
SXM4(to, _vec_h_temp_invdet_, _vec_h_temp_); \
|
||
|
} while (0)
|
||
|
#define INVERTMAT4d(to,from) INVERTMAT4safe(double,to,from)
|
||
|
#define INVERTMAT4i(to,from) INVERTMAT4safe(int,to,from)
|
||
|
#define INVERTMAT4l(to,from) INVERTMAT4safe(long,to,from)
|
||
|
#define INVERTMAT4r(to,from) INVERTMAT4safe(real,to,from)
|
||
|
#endif /* VEC_H */
|