static void process_fastpath_apply_tonecurves(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece,
const void *const ivoid, void *const ovoid,
const dt_iop_roi_t *const roi_in,
const dt_iop_roi_t *const roi_out)
{
const dt_iop_colorout_data_t *const d = (dt_iop_colorout_data_t *)piece->data;
const int ch = piece->colors;
if(!isnan(d->cmatrix[0]))
{
// out is already converted to RGB from Lab.
// do we have any lut to apply, or is this a linear profile?
if((d->lut[0][0] >= 0.0f) && (d->lut[1][0] >= 0.0f) && (d->lut[2][0] >= 0.0f))
{ // apply profile
float *const out = (float *const)ovoid;
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none)
#endif
for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch)
{
for(int c = 0; c < 3; c++)
{
out[k + c] = (out[k + c] < 1.0f) ? lerp_lut(d->lut[c], out[k + c])
: dt_iop_eval_exp(d->unbounded_coeffs[c], out[k + c]);
}
}
}
else if((d->lut[0][0] >= 0.0f) || (d->lut[1][0] >= 0.0f) || (d->lut[2][0] >= 0.0f))
{ // apply profile
float *const out = (float *const)ovoid;
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none)
#endif
for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch)
{
for(int c = 0; c < 3; c++)
{
if(d->lut[c][0] >= 0.0f)
{
out[k + c] = (out[k + c] < 1.0f) ? lerp_lut(d->lut[c], out[k + c])
: dt_iop_eval_exp(d->unbounded_coeffs[c], out[k + c]);
}
}
}
}
}
}
void commit_params (struct dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pixelpipe_t *pipe, dt_dev_pixelpipe_iop_t *piece)
{
dt_iop_colorout_params_t *p = (dt_iop_colorout_params_t *)p1;
dt_iop_colorout_data_t *d = (dt_iop_colorout_data_t *)piece->data;
gchar *overprofile = dt_conf_get_string("plugins/lighttable/export/iccprofile");
const int overintent = dt_conf_get_int("plugins/lighttable/export/iccintent");
const int high_quality_processing = dt_conf_get_bool("plugins/lighttable/export/force_lcms2");
gchar *outprofile=NULL;
int outintent = 0;
/* cleanup profiles */
if (d->output)
dt_colorspaces_cleanup_profile(d->output);
d->output = NULL;
if (d->softproof_enabled)
dt_colorspaces_cleanup_profile(d->softproof);
d->softproof = NULL;
d->softproof_enabled = p->softproof_enabled;
if(self->dev->gui_attached && self->gui_data != NULL)
{
dt_iop_colorout_gui_data_t *g = (dt_iop_colorout_gui_data_t *)self->gui_data;
g->softproof_enabled = p->softproof_enabled;
}
if (d->xform)
{
cmsDeleteTransform(d->xform);
d->xform = 0;
}
d->cmatrix[0] = NAN;
d->lut[0][0] = -1.0f;
d->lut[1][0] = -1.0f;
d->lut[2][0] = -1.0f;
piece->process_cl_ready = 1;
/* if we are exporting then check and set usage of override profile */
if (pipe->type == DT_DEV_PIXELPIPE_EXPORT)
{
if (overprofile && strcmp(overprofile, "image"))
snprintf(p->iccprofile, DT_IOP_COLOR_ICC_LEN, "%s", overprofile);
if (overintent >= 0)
p->intent = overintent;
outprofile = p->iccprofile;
outintent = p->intent;
}
else
{
/* we are not exporting, using display profile as output */
outprofile = p->displayprofile;
outintent = p->displayintent;
}
/*
* Setup transform flags
*/
uint32_t transformFlags = 0;
/* creating output profile */
d->output = _create_profile(outprofile);
/* creating softproof profile if softproof is enabled */
if (d->softproof_enabled && pipe->type == DT_DEV_PIXELPIPE_FULL)
{
d->softproof = _create_profile(p->softproofprofile);
/* TODO: the use of bpc should be userconfigurable either from module or preference pane */
/* softproof flag and black point compensation */
transformFlags |= cmsFLAGS_SOFTPROOFING|cmsFLAGS_NOCACHE|cmsFLAGS_BLACKPOINTCOMPENSATION;
if(d->softproof_enabled == DT_SOFTPROOF_GAMUTCHECK)
transformFlags |= cmsFLAGS_GAMUTCHECK;
}
/* get matrix from profile, if softproofing or high quality exporting always go xform codepath */
if (d->softproof_enabled || (pipe->type == DT_DEV_PIXELPIPE_EXPORT && high_quality_processing) ||
dt_colorspaces_get_matrix_from_output_profile (d->output, d->cmatrix, d->lut[0], d->lut[1], d->lut[2], LUT_SAMPLES))
{
d->cmatrix[0] = NAN;
piece->process_cl_ready = 0;
d->xform = cmsCreateProofingTransform(d->Lab,
TYPE_Lab_FLT,
d->output,
TYPE_RGB_FLT,
d->softproof,
outintent,
INTENT_RELATIVE_COLORIMETRIC,
transformFlags);
}
// user selected a non-supported output profile, check that:
if (!d->xform && isnan(d->cmatrix[0]))
{
dt_control_log(_("unsupported output profile has been replaced by sRGB!"));
if (d->output)
dt_colorspaces_cleanup_profile(d->output);
d->output = dt_colorspaces_create_srgb_profile();
if (d->softproof_enabled || dt_colorspaces_get_matrix_from_output_profile (d->output, d->cmatrix, d->lut[0], d->lut[1], d->lut[2], LUT_SAMPLES))
{
d->cmatrix[0] = NAN;
piece->process_cl_ready = 0;
d->xform = cmsCreateProofingTransform(d->Lab,
TYPE_Lab_FLT,
d->output,
TYPE_RGB_FLT,
d->softproof,
outintent,
INTENT_RELATIVE_COLORIMETRIC,
transformFlags);
}
}
// now try to initialize unbounded mode:
// we do extrapolation for input values above 1.0f.
// unfortunately we can only do this if we got the computation
// in our hands, i.e. for the fast builtin-dt-matrix-profile path.
for(int k=0; k<3; k++)
{
// omit luts marked as linear (negative as marker)
if(d->lut[k][0] >= 0.0f)
{
const float x[4] = {0.7f, 0.8f, 0.9f, 1.0f};
const float y[4] = {lerp_lut(d->lut[k], x[0]),
lerp_lut(d->lut[k], x[1]),
lerp_lut(d->lut[k], x[2]),
lerp_lut(d->lut[k], x[3])
};
dt_iop_estimate_exp(x, y, 4, d->unbounded_coeffs[k]);
}
else d->unbounded_coeffs[k][0] = -1.0f;
}
//fprintf(stderr, " Output profile %s, softproof %s%s%s\n", outprofile, d->softproof_enabled?"enabled ":"disabled",d->softproof_enabled?"using profile ":"",d->softproof_enabled?p->softproofprofile:"");
g_free(overprofile);
}
void
process (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, void *ivoid, void *ovoid, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
const dt_iop_colorout_data_t *const d = (dt_iop_colorout_data_t *)piece->data;
const int ch = piece->colors;
const int gamutcheck = (d->softproof_enabled == DT_SOFTPROOF_GAMUTCHECK);
if(!isnan(d->cmatrix[0]))
{
//fprintf(stderr,"Using cmatrix codepath\n");
// convert to rgb using matrix
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none) shared(roi_in,roi_out, ivoid, ovoid)
#endif
for(int j=0; j<roi_out->height; j++)
{
float *in = (float*)ivoid + ch*roi_in->width *j;
float *out = (float*)ovoid + ch*roi_out->width*j;
const __m128 m0 = _mm_set_ps(0.0f,d->cmatrix[6],d->cmatrix[3],d->cmatrix[0]);
const __m128 m1 = _mm_set_ps(0.0f,d->cmatrix[7],d->cmatrix[4],d->cmatrix[1]);
const __m128 m2 = _mm_set_ps(0.0f,d->cmatrix[8],d->cmatrix[5],d->cmatrix[2]);
for(int i=0; i<roi_out->width; i++, in+=ch, out+=ch )
{
const __m128 xyz = dt_Lab_to_XYZ_SSE(_mm_load_ps(in));
const __m128 t = _mm_add_ps(_mm_mul_ps(m0,_mm_shuffle_ps(xyz,xyz,_MM_SHUFFLE(0,0,0,0))),_mm_add_ps(_mm_mul_ps(m1,_mm_shuffle_ps(xyz,xyz,_MM_SHUFFLE(1,1,1,1))),_mm_mul_ps(m2,_mm_shuffle_ps(xyz,xyz,_MM_SHUFFLE(2,2,2,2)))));
_mm_stream_ps(out,t);
}
}
_mm_sfence();
// apply profile
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none) shared(roi_in,roi_out, ivoid, ovoid)
#endif
for(int j=0; j<roi_out->height; j++)
{
float *in = (float*)ivoid + ch*roi_in->width *j;
float *out = (float*)ovoid + ch*roi_out->width*j;
for(int i=0; i<roi_out->width; i++, in+=ch, out+=ch )
{
for(int i=0; i<3; i++)
if (d->lut[i][0] >= 0.0f)
{
out[i] = (out[i] < 1.0f) ? lerp_lut(d->lut[i], out[i]) : dt_iop_eval_exp(d->unbounded_coeffs[i], out[i]);
}
}
}
}
else
{
float *in = (float*)ivoid;
float *out = (float*)ovoid;
const int rowsize=roi_out->width * 3;
//fprintf(stderr,"Using xform codepath\n");
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none) shared(out, roi_out, in)
#endif
for (int k=0; k<roi_out->height; k++)
{
float Lab[rowsize];
float rgb[rowsize];
const int m=(k*(roi_out->width*ch));
for (int l=0; l<roi_out->width; l++)
{
int li=3*l,ii=ch*l;
Lab[li+0] = in[m+ii+0];
Lab[li+1] = in[m+ii+1];
Lab[li+2] = in[m+ii+2];
}
cmsDoTransform (d->xform, Lab, rgb, roi_out->width);
for (int l=0; l<roi_out->width; l++)
{
int oi=ch*l, ri=3*l;
if(gamutcheck && (rgb[ri+0] < 0.0f || rgb[ri+1] < 0.0f || rgb[ri+2] < 0.0f))
{
out[m+oi+0] = 0.0f;
out[m+oi+1] = 1.0f;
out[m+oi+2] = 1.0f;
}
else
{
out[m+oi+0] = rgb[ri+0];
out[m+oi+1] = rgb[ri+1];
out[m+oi+2] = rgb[ri+2];
}
}
}
}
if(piece->pipe->mask_display)
dt_iop_alpha_copy(ivoid, ovoid, roi_out->width, roi_out->height);
}
void commit_params(struct dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pixelpipe_t *pipe,
dt_dev_pixelpipe_iop_t *piece)
{
dt_iop_colorout_params_t *p = (dt_iop_colorout_params_t *)p1;
dt_iop_colorout_data_t *d = (dt_iop_colorout_data_t *)piece->data;
const dt_colorspaces_color_profile_type_t over_type = dt_conf_get_int("plugins/lighttable/export/icctype");
gchar *over_filename = dt_conf_get_string("plugins/lighttable/export/iccprofile");
const dt_iop_color_intent_t over_intent = dt_conf_get_int("plugins/lighttable/export/iccintent");
const int force_lcms2 = dt_conf_get_bool("plugins/lighttable/export/force_lcms2");
dt_colorspaces_color_profile_type_t out_type = DT_COLORSPACE_SRGB;
gchar *out_filename = NULL;
dt_iop_color_intent_t out_intent = DT_INTENT_PERCEPTUAL;
const cmsHPROFILE Lab = dt_colorspaces_get_profile(DT_COLORSPACE_LAB, "", DT_PROFILE_DIRECTION_ANY)->profile;
cmsHPROFILE output = NULL;
cmsHPROFILE softproof = NULL;
d->mode = pipe->type == DT_DEV_PIXELPIPE_FULL ? darktable.color_profiles->mode : DT_PROFILE_NORMAL;
if(d->xform)
{
cmsDeleteTransform(d->xform);
d->xform = NULL;
}
d->cmatrix[0] = NAN;
d->lut[0][0] = -1.0f;
d->lut[1][0] = -1.0f;
d->lut[2][0] = -1.0f;
piece->process_cl_ready = 1;
/* if we are exporting then check and set usage of override profile */
if(pipe->type == DT_DEV_PIXELPIPE_EXPORT)
{
if(over_type != DT_COLORSPACE_NONE)
{
p->type = over_type;
g_strlcpy(p->filename, over_filename, sizeof(p->filename));
}
if((unsigned int)over_intent < DT_INTENT_LAST) p->intent = over_intent;
out_type = p->type;
out_filename = p->filename;
out_intent = p->intent;
}
else if(pipe->type == DT_DEV_PIXELPIPE_THUMBNAIL)
{
out_type = dt_mipmap_cache_get_colorspace();
out_filename = (out_type == DT_COLORSPACE_DISPLAY ? darktable.color_profiles->display_filename : "");
out_intent = darktable.color_profiles->display_intent;
}
else
{
/* we are not exporting, using display profile as output */
out_type = darktable.color_profiles->display_type;
out_filename = darktable.color_profiles->display_filename;
out_intent = darktable.color_profiles->display_intent;
}
/*
* Setup transform flags
*/
uint32_t transformFlags = 0;
/* creating output profile */
if(out_type == DT_COLORSPACE_DISPLAY) pthread_rwlock_rdlock(&darktable.color_profiles->xprofile_lock);
const dt_colorspaces_color_profile_t *out_profile
= dt_colorspaces_get_profile(out_type, out_filename, DT_PROFILE_DIRECTION_OUT | DT_PROFILE_DIRECTION_DISPLAY);
output = out_profile->profile;
/* creating softproof profile if softproof is enabled */
if(d->mode != DT_PROFILE_NORMAL && pipe->type == DT_DEV_PIXELPIPE_FULL)
{
softproof = dt_colorspaces_get_profile(darktable.color_profiles->softproof_type,
darktable.color_profiles->softproof_filename,
DT_PROFILE_DIRECTION_OUT | DT_PROFILE_DIRECTION_DISPLAY)->profile;
/* TODO: the use of bpc should be userconfigurable either from module or preference pane */
/* softproof flag and black point compensation */
transformFlags |= cmsFLAGS_SOFTPROOFING | cmsFLAGS_NOCACHE | cmsFLAGS_BLACKPOINTCOMPENSATION;
if(d->mode == DT_PROFILE_GAMUTCHECK) transformFlags |= cmsFLAGS_GAMUTCHECK;
}
/*
* NOTE: theoretically, we should be passing
* UsedDirection = LCMS_USED_AS_PROOF into
* dt_colorspaces_get_matrix_from_output_profile() so that
* dt_colorspaces_get_matrix_from_profile() knows it, but since we do not try
* to use our matrix codepath when softproof is enabled, this seemed redundant.
*/
/* get matrix from profile, if softproofing or high quality exporting always go xform codepath */
if(d->mode != DT_PROFILE_NORMAL || force_lcms2
|| dt_colorspaces_get_matrix_from_output_profile(output, d->cmatrix, d->lut[0], d->lut[1], d->lut[2],
LUT_SAMPLES, out_intent))
{
d->cmatrix[0] = NAN;
piece->process_cl_ready = 0;
d->xform = cmsCreateProofingTransform(Lab, TYPE_LabA_FLT, output, TYPE_RGBA_FLT, softproof,
out_intent, INTENT_RELATIVE_COLORIMETRIC, transformFlags);
}
// user selected a non-supported output profile, check that:
if(!d->xform && isnan(d->cmatrix[0]))
{
dt_control_log(_("unsupported output profile has been replaced by sRGB!"));
fprintf(stderr, "unsupported output profile `%s' has been replaced by sRGB!\n", out_profile->name);
output = dt_colorspaces_get_profile(DT_COLORSPACE_SRGB, "", DT_PROFILE_DIRECTION_OUT)->profile;
if(d->mode != DT_PROFILE_NORMAL
|| dt_colorspaces_get_matrix_from_output_profile(output, d->cmatrix, d->lut[0], d->lut[1],
d->lut[2], LUT_SAMPLES, out_intent))
{
d->cmatrix[0] = NAN;
piece->process_cl_ready = 0;
d->xform = cmsCreateProofingTransform(Lab, TYPE_LabA_FLT, output, TYPE_RGBA_FLT, softproof,
out_intent, INTENT_RELATIVE_COLORIMETRIC, transformFlags);
}
}
if(out_type == DT_COLORSPACE_DISPLAY) pthread_rwlock_unlock(&darktable.color_profiles->xprofile_lock);
// now try to initialize unbounded mode:
// we do extrapolation for input values above 1.0f.
// unfortunately we can only do this if we got the computation
// in our hands, i.e. for the fast builtin-dt-matrix-profile path.
for(int k = 0; k < 3; k++)
{
// omit luts marked as linear (negative as marker)
if(d->lut[k][0] >= 0.0f)
{
const float x[4] = { 0.7f, 0.8f, 0.9f, 1.0f };
const float y[4] = { lerp_lut(d->lut[k], x[0]), lerp_lut(d->lut[k], x[1]), lerp_lut(d->lut[k], x[2]),
lerp_lut(d->lut[k], x[3]) };
dt_iop_estimate_exp(x, y, 4, d->unbounded_coeffs[k]);
}
else
d->unbounded_coeffs[k][0] = -1.0f;
}
g_free(over_filename);
}
void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, void *ivoid, void *ovoid,
const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
const dt_iop_colorout_data_t *const d = (dt_iop_colorout_data_t *)piece->data;
const int ch = piece->colors;
const int gamutcheck = (d->mode == DT_PROFILE_GAMUTCHECK);
if(!isnan(d->cmatrix[0]))
{
// fprintf(stderr,"Using cmatrix codepath\n");
// convert to rgb using matrix
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none) shared(roi_in, roi_out, ivoid, ovoid)
#endif
for(int j = 0; j < roi_out->height; j++)
{
float *in = (float *)ivoid + (size_t)ch * roi_in->width * j;
float *out = (float *)ovoid + (size_t)ch * roi_out->width * j;
const __m128 m0 = _mm_set_ps(0.0f, d->cmatrix[6], d->cmatrix[3], d->cmatrix[0]);
const __m128 m1 = _mm_set_ps(0.0f, d->cmatrix[7], d->cmatrix[4], d->cmatrix[1]);
const __m128 m2 = _mm_set_ps(0.0f, d->cmatrix[8], d->cmatrix[5], d->cmatrix[2]);
for(int i = 0; i < roi_out->width; i++, in += ch, out += ch)
{
const __m128 xyz = dt_Lab_to_XYZ_SSE(_mm_load_ps(in));
const __m128 t
= _mm_add_ps(_mm_mul_ps(m0, _mm_shuffle_ps(xyz, xyz, _MM_SHUFFLE(0, 0, 0, 0))),
_mm_add_ps(_mm_mul_ps(m1, _mm_shuffle_ps(xyz, xyz, _MM_SHUFFLE(1, 1, 1, 1))),
_mm_mul_ps(m2, _mm_shuffle_ps(xyz, xyz, _MM_SHUFFLE(2, 2, 2, 2)))));
_mm_stream_ps(out, t);
}
}
_mm_sfence();
// apply profile
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none) shared(roi_in, roi_out, ivoid, ovoid)
#endif
for(int j = 0; j < roi_out->height; j++)
{
float *in = (float *)ivoid + (size_t)ch * roi_in->width * j;
float *out = (float *)ovoid + (size_t)ch * roi_out->width * j;
for(int i = 0; i < roi_out->width; i++, in += ch, out += ch)
{
for(int i = 0; i < 3; i++)
if(d->lut[i][0] >= 0.0f)
{
out[i] = (out[i] < 1.0f) ? lerp_lut(d->lut[i], out[i])
: dt_iop_eval_exp(d->unbounded_coeffs[i], out[i]);
}
}
}
}
else
{
// fprintf(stderr,"Using xform codepath\n");
const __m128 outofgamutpixel = _mm_set_ps(0.0f, 1.0f, 1.0f, 0.0f);
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none) shared(ivoid, ovoid, roi_out)
#endif
for(int k = 0; k < roi_out->height; k++)
{
const float *in = ((float *)ivoid) + (size_t)ch * k * roi_out->width;
float *out = ((float *)ovoid) + (size_t)ch * k * roi_out->width;
if(!gamutcheck)
{
cmsDoTransform(d->xform, in, out, roi_out->width);
}
else
{
void *rgb = dt_alloc_align(16, 4 * sizeof(float) * roi_out->width);
cmsDoTransform(d->xform, in, rgb, roi_out->width);
float *rgbptr = (float *)rgb;
for(int j = 0; j < roi_out->width; j++, rgbptr += 4, out += 4)
{
const __m128 pixel = _mm_load_ps(rgbptr);
__m128 ingamut = _mm_cmplt_ps(pixel, _mm_set_ps(-FLT_MAX, 0.0f, 0.0f, 0.0f));
ingamut = _mm_or_ps(_mm_unpacklo_ps(ingamut, ingamut), _mm_unpackhi_ps(ingamut, ingamut));
ingamut = _mm_or_ps(_mm_unpacklo_ps(ingamut, ingamut), _mm_unpackhi_ps(ingamut, ingamut));
const __m128 result
= _mm_or_ps(_mm_and_ps(ingamut, outofgamutpixel), _mm_andnot_ps(ingamut, pixel));
_mm_stream_ps(out, result);
}
dt_free_align(rgb);
}
}
_mm_sfence();
}
if(piece->pipe->mask_display) dt_iop_alpha_copy(ivoid, ovoid, roi_out->width, roi_out->height);
}
