diff options
author | Jörg Frings-Fürst <debian@jff-webhosting.net> | 2018-07-11 22:20:14 +0200 |
---|---|---|
committer | Jörg Frings-Fürst <debian@jff-webhosting.net> | 2018-07-11 22:20:14 +0200 |
commit | 7beb00cd8d28c3d5893ce3db907a828d64afdea9 (patch) | |
tree | 395a3dee2fe197b8284dee02c5f527889df78413 /icc/icc.h | |
parent | e2d30e0583c047a4bedf4c8d0c86320f1b3fd8ed (diff) | |
parent | a0442ed58dee48a521ea053083ea967894507898 (diff) |
Update upstream source from tag 'upstream/2.0.1+repack'
Update to upstream version '2.0.1+repack'
with Debian dir 6edb5dd2df9aca152662fc4a8f72bd6d86f8552e
Diffstat (limited to 'icc/icc.h')
-rwxr-xr-x | icc/icc.h | 62 |
1 files changed, 36 insertions, 26 deletions
@@ -1784,6 +1784,12 @@ extern ICCLIB_API unsigned int icmCSSig2chanNames( icColorSpaceSignature sig, ch /* Copy a 2 vector */ #define icmCpy2(d_ary, s_ary) ((d_ary)[0] = (s_ary)[0], (d_ary)[1] = (s_ary)[1]) +#define icmAdd2(d_ary, s1_ary, s2_ary) ((d_ary)[0] = (s1_ary)[0] + (s2_ary)[0], \ + (d_ary)[1] = (s1_ary)[1] + (s2_ary)[1]) + +#define icmSub2(d_ary, s1_ary, s2_ary) ((d_ary)[0] = (s1_ary)[0] - (s2_ary)[0], \ + (d_ary)[1] = (s1_ary)[1] - (s2_ary)[1]) + /* Copy a 3 vector */ #define icmCpy3(d_ary, s_ary) ((d_ary)[0] = (s_ary)[0], (d_ary)[1] = (s_ary)[1], \ (d_ary)[2] = (s_ary)[2]) @@ -1792,10 +1798,12 @@ extern ICCLIB_API unsigned int icmCSSig2chanNames( icColorSpaceSignature sig, ch #define icmCpy4(d_ary, s_ary) ((d_ary)[0] = (s_ary)[0], (d_ary)[1] = (s_ary)[1], \ (d_ary)[2] = (s_ary)[2], (d_ary)[3] = (s_ary)[3]) +/* - - - - - - - - - - - - - - */ + /* Clamp a 3 vector to be +ve */ void icmClamp3(double out[3], double in[3]); -/* Invert a 3 vector */ +/* Invert (negate) a 3 vector */ void icmInv3(double out[3], double in[3]); /* Add two 3 vectors */ @@ -1832,8 +1840,6 @@ double icmDot3(double in1[3], double in2[3]); /* Compute the cross product of two 3D vectors, out = in1 x in2 */ void icmCross3(double out[3], double in1[3], double in2[3]); -#define icmNorm2(i) sqrt((i)[0] * (i)[0] + (i)[1] * (i)[1]) - /* Compute the norm squared (length squared) of a 3 vector */ double icmNorm3sq(double in[3]); @@ -1869,9 +1875,6 @@ double icmClip3marg(double out[3], double in[3]); /* Normalise a 3 vector to the given length. Return nz if not normalisable */ int icmNormalize3(double out[3], double in[3], double len); -/* Compute the norm (length) of of a vector defined by two points */ -double icmNorm22(double in1[2], double in0[2]); - /* Compute the norm squared (length squared) of a vector defined by two points */ double icmNorm33sq(double in1[3], double in0[3]); @@ -1983,30 +1986,24 @@ double icmClip4marg(double out[4], double in[4]); /* - - - - - - - - - - - - - - - - - - - - - - - */ -/* Multiply 5 vector by 5x5 transform matrix */ -/* Organization is mat[out][in] */ -void icmMulBy5x5(double out[5], double mat[5][5], double in[5]); +#define icmNorm2(i) sqrt((i)[0] * (i)[0] + (i)[1] * (i)[1]) -/* Transpose a 5x5 matrix */ -void icmTranspose5x5(double out[5][5], double in[5][5]); - -/* Clip a vector to the range 0.0 .. 1.0 */ -/* and return any clipping margine */ -double icmClip5marg(double out[5], double in[5]); +#define icmNorm2sq(i) ((i)[0] * (i)[0] + (i)[1] * (i)[1]) +/* Compute the norm (length) of of a vector defined by two points */ +double icmNorm22(double in1[2], double in0[2]); -/* Multiply 6 vector by 6x6 transform matrix */ -/* Organization is mat[out][in] */ -void icmMulBy6x6(double out[6], double mat[6][6], double in[6]); +/* Compute the norm (length) squared of of a vector defined by two points */ +double icmNorm22sq(double in1[2], double in0[2]); -/* Transpose a 6x6 matrix */ -void icmTranspose6x6(double out[6][6], double in[6][6]); +/* Compute the dot product of two 2 vectors */ +double icmDot2(double in1[2], double in2[2]); -/* Clip a vector to the range 0.0 .. 1.0 */ -/* and return any clipping margine */ -double icmClip6marg(double out[6], double in[6]); +#define ICMDOT2(o, i, j) ((o) = (i)[0] * (j)[0] + (i)[1] * (j)[1]) -/* - - - - - - - - - - - - - - - - - - - - - - - */ +/* Compute the dot product of two 2 vectors defined by 4 points */ +/* 1->2 and 3->4 */ +double icmDot22(double in1[2], double in2[2], double in3[2], double in4[2]); /* Given 2 2D points, compute a plane equation. */ /* The normal will be right handed given the order of the points */ @@ -2038,13 +2035,26 @@ int icmLinePointClosest2(double cp[2], double *pa, int icmLineIntersect2(double res[2], double p1[2], double p2[2], double p3[2], double p4[2]); /* Given two finite 2D lines define by 4 points, compute their paramaterized intersection. */ -/* aprm may be NULL */ -/* Return nz if there is no intersection (lines are parallel or do not cross in length) */ +/* aprm may be NULL. Param is prop. from p1 -> p2, p3 -> p4 */ +/* Return 2 if there is no intersection (lines are parallel) */ +/* Return 1 lines do not cross within their length */ int icmParmLineIntersect2(double ares[2], double aprm[2], double p1[2], double p2[2], double p3[2], double p4[2]); +/* Invert a 2x2 transform matrix. Return 1 if error. */ +int icmInverse2x2(double out[2][2], double in[2][2]); + /* Multiply 2 array by 2x2 transform matrix */ void icmMulBy2x2(double out[2], double mat[2][2], double in[2]); +/* Compute a blend between in0 and in1 */ +void icmBlend2(double out[2], double in0[2], double in1[2], double bf); + +/* Scale a 2 vector by the given ratio */ +void icmScale2(double out[2], double in[2], double rat); + +/* Scale a 2 vector by the given ratio and add it */ +void icmScaleAdd2(double out[2], double in2[3], double in1[2], double rat); + /* - - - - - - - - - - - - - - */ /* Simple macro to transfer an array to an XYZ number */ |