liz/rawTherapee
liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

iplab2rgb.cc: speedup for copyAndClamp(), #5964

Browse Source
This commit is contained in:
Ingo Weyrich 2020-10-25 13:46:34 +01:00
parent fd9f3246f0
commit b4f68adb64
3 changed files with 97 additions and 66 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
rtengine/color.cc
View File

@ -1015,23 +1015,6 @@ void Color::xyz2r (float x, float y, float z, float &r, const double rgb_xyz[3][
r = ((rgb_xyz[0][0] * x + rgb_xyz[0][1] * y + rgb_xyz[0][2] * z)) ;
}
// same for float
void Color::xyz2rgb (float x, float y, float z, float &r, float &g, float &b, const float rgb_xyz[3][3])
{
r = ((rgb_xyz[0][0] * x + rgb_xyz[0][1] * y + rgb_xyz[0][2] * z)) ;
g = ((rgb_xyz[1][0] * x + rgb_xyz[1][1] * y + rgb_xyz[1][2] * z)) ;
b = ((rgb_xyz[2][0] * x + rgb_xyz[2][1] * y + rgb_xyz[2][2] * z)) ;
}
#ifdef __SSE2__
void Color::xyz2rgb (vfloat x, vfloat y, vfloat z, vfloat &r, vfloat &g, vfloat &b, const vfloat rgb_xyz[3][3])
{
r = ((rgb_xyz[0][0] * x + rgb_xyz[0][1] * y + rgb_xyz[0][2] * z)) ;
g = ((rgb_xyz[1][0] * x + rgb_xyz[1][1] * y + rgb_xyz[1][2] * z)) ;
b = ((rgb_xyz[2][0] * x + rgb_xyz[2][1] * y + rgb_xyz[2][2] * z)) ;
}
#endif // __SSE2__
#ifdef __SSE2__
void Color::trcGammaBW (float &r, float &g, float &b, float gammabwr, float gammabwg, float gammabwb)
{
@ -1646,19 +1629,6 @@ void Color::gammanf2lut (LUTf &gammacurve, float gamma, float divisor, float fac
#endif
}
void Color::Lab2XYZ(float L, float a, float b, float &x, float &y, float &z)
{
float LL = L / 327.68f;
float aa = a / 327.68f;
float bb = b / 327.68f;
float fy = (c1By116 * LL) + c16By116; // (L+16)/116
float fx = (0.002f * aa) + fy;
float fz = fy - (0.005f * bb);
x = 65535.0f * f2xyz(fx) * D50x;
z = 65535.0f * f2xyz(fz) * D50z;
y = (LL > epskap) ? 65535.0f * fy * fy * fy : 65535.0f * LL / kappa;
}
float Color::L2Y(float L)
{
const float LL = L / 327.68f;
@ -1676,27 +1646,6 @@ void Color::L2XYZ(float L, float &x, float &y, float &z) // for black & white
y = (LL > epskap) ? 65535.0f * fy * fy * fy : 65535.0f * LL / kappa;
}
#ifdef __SSE2__
void Color::Lab2XYZ(vfloat L, vfloat a, vfloat b, vfloat &x, vfloat &y, vfloat &z)
{
vfloat c327d68 = F2V(327.68f);
L /= c327d68;
a /= c327d68;
b /= c327d68;
vfloat fy = F2V(c1By116) * L + F2V(c16By116);
vfloat fx = F2V(0.002f) * a + fy;
vfloat fz = fy - (F2V(0.005f) * b);
vfloat c65535 = F2V(65535.f);
x = c65535 * f2xyz(fx) * F2V(D50x);
z = c65535 * f2xyz(fz) * F2V(D50z);
vfloat res1 = fy * fy * fy;
vfloat res2 = L / F2V(kappa);
y = vself(vmaskf_gt(L, F2V(epskap)), res1, res2);
y *= c65535;
}
#endif // __SSE2__
inline float Color::computeXYZ2Lab(float f)
{
if (f < 0.f) {

47
rtengine/color.h
View File

@ -570,9 +570,20 @@ public:
*/
static void xyz2rgb (float x, float y, float z, float &r, float &g, float &b, const double rgb_xyz[3][3]);
static void xyz2r (float x, float y, float z, float &r, const double rgb_xyz[3][3]);
static void xyz2rgb (float x, float y, float z, float &r, float &g, float &b, const float rgb_xyz[3][3]);
static inline void xyz2rgb (float x, float y, float z, float &r, float &g, float &b, const float rgb_xyz[3][3])
{
r = ((rgb_xyz[0][0] * x + rgb_xyz[0][1] * y + rgb_xyz[0][2] * z)) ;
g = ((rgb_xyz[1][0] * x + rgb_xyz[1][1] * y + rgb_xyz[1][2] * z)) ;
b = ((rgb_xyz[2][0] * x + rgb_xyz[2][1] * y + rgb_xyz[2][2] * z)) ;
}
#ifdef __SSE2__
static void xyz2rgb (vfloat x, vfloat y, vfloat z, vfloat &r, vfloat &g, vfloat &b, const vfloat rgb_xyz[3][3]);
static inline void xyz2rgb (vfloat x, vfloat y, vfloat z, vfloat &r, vfloat &g, vfloat &b, const vfloat rgb_xyz[3][3])
{
r = ((rgb_xyz[0][0] * x + rgb_xyz[0][1] * y + rgb_xyz[0][2] * z)) ;
g = ((rgb_xyz[1][0] * x + rgb_xyz[1][1] * y + rgb_xyz[1][2] * z)) ;
b = ((rgb_xyz[2][0] * x + rgb_xyz[2][1] * y + rgb_xyz[2][2] * z)) ;
}
#endif
@ -603,12 +614,40 @@ public:
* @param y Y coordinate [0 ; 65535] ; can be negative! (return value)
* @param z Z coordinate [0 ; 65535] ; can be negative! (return value)
*/
static void Lab2XYZ(float L, float a, float b, float &x, float &y, float &z);
static inline void Lab2XYZ(float L, float a, float b, float &x, float &y, float &z)
{
float LL = L / 327.68f;
float aa = a / 327.68f;
float bb = b / 327.68f;
float fy = (c1By116 * LL) + c16By116; // (L+16)/116
float fx = (0.002f * aa) + fy;
float fz = fy - (0.005f * bb);
x = 65535.f * f2xyz(fx) * D50x;
z = 65535.f * f2xyz(fz) * D50z;
y = (LL > epskapf) ? 65535.f * fy * fy * fy : 65535.f * LL / kappaf;
}
static void L2XYZ(float L, float &x, float &y, float &z);
static float L2Y(float L);
#ifdef __SSE2__
static void Lab2XYZ(vfloat L, vfloat a, vfloat b, vfloat &x, vfloat &y, vfloat &z);
static inline void Lab2XYZ(vfloat L, vfloat a, vfloat b, vfloat &x, vfloat &y, vfloat &z)
{
vfloat c327d68 = F2V(327.68f);
L /= c327d68;
a /= c327d68;
b /= c327d68;
vfloat fy = F2V(c1By116) * L + F2V(c16By116);
vfloat fx = F2V(0.002f) * a + fy;
vfloat fz = fy - (F2V(0.005f) * b);
vfloat c65535 = F2V(65535.f);
x = c65535 * f2xyz(fx) * F2V(D50x);
z = c65535 * f2xyz(fz) * F2V(D50z);
vfloat res1 = fy * fy * fy;
vfloat res2 = L / F2V(kappa);
y = vself(vmaskf_gt(L, F2V(epskap)), res1, res2);
y *= c65535;
}
#endif // __SSE2__
/**

65
rtengine/iplab2rgb.cc
View File

@ -32,8 +32,6 @@
namespace rtengine
{
extern void filmlike_clip(float *r, float *g, float *b);
namespace {
inline void copyAndClampLine(const float *src, unsigned char *dst, const int W)
@ -46,9 +44,26 @@ inline void copyAndClampLine(const float *src, unsigned char *dst, const int W)
inline void copyAndClamp(const LabImage *src, unsigned char *dst, const double rgb_xyz[3][3], bool multiThread)
{
int W = src->W;
int H = src->H;
const int W = src->W;
const int H = src->H;
float rgb_xyzf[3][3];
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
rgb_xyzf[i][j] = rgb_xyz[i][j];
}
}
#ifdef __SSE2__
vfloat rgb_xyzv[3][3];
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
rgb_xyzv[i][j] = F2V(rgb_xyzf[i][j]);
}
}
#endif
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic,16) if (multiThread)
#endif
@ -58,17 +73,47 @@ inline void copyAndClamp(const LabImage *src, unsigned char *dst, const double r
float* rb = src->b[i];
int ix = i * 3 * W;
float R, G, B;
float x_, y_, z_;
for (int j = 0; j < W; ++j) {
#ifdef __SSE2__
float rbuffer[W] ALIGNED16;
float gbuffer[W] ALIGNED16;
float bbuffer[W] ALIGNED16;
int j = 0;
for (; j < W - 3; j += 4) {
vfloat R, G, B;
vfloat x_, y_, z_;
Color::Lab2XYZ(LVFU(rL[j]), LVFU(ra[j]), LVFU(rb[j]), x_, y_, z_ );
Color::xyz2rgb(x_, y_, z_, R, G, B, rgb_xyzv);
STVF(rbuffer[j], Color::gamma2curve[R]);
STVF(gbuffer[j], Color::gamma2curve[G]);
STVF(bbuffer[j], Color::gamma2curve[B]);
}
for (; j < W; ++j) {
float R, G, B;
float x_, y_, z_;
Color::Lab2XYZ(rL[j], ra[j], rb[j], x_, y_, z_ );
Color::xyz2rgb(x_, y_, z_, R, G, B, rgb_xyz);
Color::xyz2rgb(x_, y_, z_, R, G, B, rgb_xyzf);
rbuffer[j] = Color::gamma2curve[R];
gbuffer[j] = Color::gamma2curve[G];
bbuffer[j] = Color::gamma2curve[B];
}
for (int j = 0; j < W; ++j) {
dst[ix++] = uint16ToUint8Rounded(rbuffer[j]);
dst[ix++] = uint16ToUint8Rounded(gbuffer[j]);
dst[ix++] = uint16ToUint8Rounded(bbuffer[j]);
}
#else
for (int j = 0; j < W; ++j) {
float R, G, B;
float x_, y_, z_;
Color::Lab2XYZ(rL[j], ra[j], rb[j], x_, y_, z_ );
Color::xyz2rgb(x_, y_, z_, R, G, B, rgb_xyzf);
dst[ix++] = uint16ToUint8Rounded(Color::gamma2curve[R]);
dst[ix++] = uint16ToUint8Rounded(Color::gamma2curve[G]);
dst[ix++] = uint16ToUint8Rounded(Color::gamma2curve[B]);
}
#endif
}
}
@ -151,8 +196,6 @@ void ImProcFunctions::lab2monitorRgb(LabImage* lab, Image8* image)
// otherwise divide by 327.68, convert to xyz and apply the RGB transform, before converting with gamma2curve
Image8* ImProcFunctions::lab2rgb(LabImage* lab, int cx, int cy, int cw, int ch, const procparams::ColorManagementParams &icm, bool consider_histogram_settings)
{
//gamutmap(lab);
if (cx < 0) {
cx = 0;
}