michael@0: /* vim: set ts=8 sw=8 noexpandtab: */
michael@0: //  qcms
michael@0: //  Copyright (C) 2009 Mozilla Corporation
michael@0: //  Copyright (C) 1998-2007 Marti Maria
michael@0: //
michael@0: // Permission is hereby granted, free of charge, to any person obtaining 
michael@0: // a copy of this software and associated documentation files (the "Software"), 
michael@0: // to deal in the Software without restriction, including without limitation 
michael@0: // the rights to use, copy, modify, merge, publish, distribute, sublicense, 
michael@0: // and/or sell copies of the Software, and to permit persons to whom the Software 
michael@0: // is furnished to do so, subject to the following conditions:
michael@0: //
michael@0: // The above copyright notice and this permission notice shall be included in 
michael@0: // all copies or substantial portions of the Software.
michael@0: //
michael@0: // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
michael@0: // EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
michael@0: // THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
michael@0: // NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
michael@0: // LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
michael@0: // OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
michael@0: // WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
michael@0: 
michael@0: #include <stdlib.h>
michael@0: #include <math.h>
michael@0: #include <assert.h>
michael@0: #include <string.h> //memcpy
michael@0: #include "qcmsint.h"
michael@0: #include "transform_util.h"
michael@0: #include "matrix.h"
michael@0: 
michael@0: static struct matrix build_lut_matrix(struct lutType *lut)
michael@0: {
michael@0: 	struct matrix result;
michael@0: 	if (lut) {
michael@0: 		result.m[0][0] = s15Fixed16Number_to_float(lut->e00);
michael@0: 		result.m[0][1] = s15Fixed16Number_to_float(lut->e01);
michael@0: 		result.m[0][2] = s15Fixed16Number_to_float(lut->e02);
michael@0: 		result.m[1][0] = s15Fixed16Number_to_float(lut->e10);
michael@0: 		result.m[1][1] = s15Fixed16Number_to_float(lut->e11);
michael@0: 		result.m[1][2] = s15Fixed16Number_to_float(lut->e12);
michael@0: 		result.m[2][0] = s15Fixed16Number_to_float(lut->e20);
michael@0: 		result.m[2][1] = s15Fixed16Number_to_float(lut->e21);
michael@0: 		result.m[2][2] = s15Fixed16Number_to_float(lut->e22);
michael@0: 		result.invalid = false;
michael@0: 	} else {
michael@0: 		memset(&result, 0, sizeof(struct matrix));
michael@0: 		result.invalid = true;
michael@0: 	}
michael@0: 	return result;
michael@0: }
michael@0: 
michael@0: static struct matrix build_mAB_matrix(struct lutmABType *lut)
michael@0: {
michael@0: 	struct matrix result;
michael@0: 	if (lut) {
michael@0: 		result.m[0][0] = s15Fixed16Number_to_float(lut->e00);
michael@0: 		result.m[0][1] = s15Fixed16Number_to_float(lut->e01);
michael@0: 		result.m[0][2] = s15Fixed16Number_to_float(lut->e02);
michael@0: 		result.m[1][0] = s15Fixed16Number_to_float(lut->e10);
michael@0: 		result.m[1][1] = s15Fixed16Number_to_float(lut->e11);
michael@0: 		result.m[1][2] = s15Fixed16Number_to_float(lut->e12);
michael@0: 		result.m[2][0] = s15Fixed16Number_to_float(lut->e20);
michael@0: 		result.m[2][1] = s15Fixed16Number_to_float(lut->e21);
michael@0: 		result.m[2][2] = s15Fixed16Number_to_float(lut->e22);
michael@0: 		result.invalid = false;
michael@0: 	} else {
michael@0: 		memset(&result, 0, sizeof(struct matrix));
michael@0: 		result.invalid = true;
michael@0: 	}
michael@0: 	return result;
michael@0: }
michael@0: 
michael@0: //Based on lcms cmsLab2XYZ
michael@0: #define f(t) (t <= (24.0f/116.0f)*(24.0f/116.0f)*(24.0f/116.0f)) ? ((841.0/108.0) * t + (16.0/116.0)) : pow(t,1.0/3.0)
michael@0: #define f_1(t) (t <= (24.0f/116.0f)) ? ((108.0/841.0) * (t - (16.0/116.0))) : (t * t * t)
michael@0: static void qcms_transform_module_LAB_to_XYZ(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	// lcms: D50 XYZ values
michael@0: 	float WhitePointX = 0.9642f;
michael@0: 	float WhitePointY = 1.0f;
michael@0: 	float WhitePointZ = 0.8249f;
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float device_L = *src++ * 100.0f;
michael@0: 		float device_a = *src++ * 255.0f - 128.0f;
michael@0: 		float device_b = *src++ * 255.0f - 128.0f;
michael@0: 		float y = (device_L + 16.0f) / 116.0f;
michael@0: 
michael@0: 		float X = f_1((y + 0.002f * device_a)) * WhitePointX;
michael@0: 		float Y = f_1(y) * WhitePointY;
michael@0: 		float Z = f_1((y - 0.005f * device_b)) * WhitePointZ;
michael@0: 		*dest++ = X / (1.0 + 32767.0/32768.0);
michael@0: 		*dest++ = Y / (1.0 + 32767.0/32768.0);
michael@0: 		*dest++ = Z / (1.0 + 32767.0/32768.0);
michael@0: 	}
michael@0: }
michael@0: 
michael@0: //Based on lcms cmsXYZ2Lab
michael@0: static void qcms_transform_module_XYZ_to_LAB(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0:         // lcms: D50 XYZ values
michael@0:         float WhitePointX = 0.9642f;
michael@0:         float WhitePointY = 1.0f;
michael@0:         float WhitePointZ = 0.8249f;
michael@0:         for (i = 0; i < length; i++) {
michael@0:                 float device_x = *src++ * (1.0 + 32767.0/32768.0) / WhitePointX;
michael@0:                 float device_y = *src++ * (1.0 + 32767.0/32768.0) / WhitePointY;
michael@0:                 float device_z = *src++ * (1.0 + 32767.0/32768.0) / WhitePointZ;
michael@0: 
michael@0: 		float fx = f(device_x);
michael@0: 		float fy = f(device_y);
michael@0: 		float fz = f(device_z);
michael@0: 
michael@0:                 float L = 116.0f*fy - 16.0f;
michael@0:                 float a = 500.0f*(fx - fy);
michael@0:                 float b = 200.0f*(fy - fz);
michael@0:                 *dest++ = L / 100.0f;
michael@0:                 *dest++ = (a+128.0f) / 255.0f;
michael@0:                 *dest++ = (b+128.0f) / 255.0f;
michael@0:         }
michael@0: 
michael@0: }
michael@0: 
michael@0: static void qcms_transform_module_clut_only(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	int xy_len = 1;
michael@0: 	int x_len = transform->grid_size;
michael@0: 	int len = x_len * x_len;
michael@0: 	float* r_table = transform->r_clut;
michael@0: 	float* g_table = transform->g_clut;
michael@0: 	float* b_table = transform->b_clut;
michael@0: 
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float linear_r = *src++;
michael@0: 		float linear_g = *src++;
michael@0: 		float linear_b = *src++;
michael@0: 
michael@0: 		int x = floorf(linear_r * (transform->grid_size-1));
michael@0: 		int y = floorf(linear_g * (transform->grid_size-1));
michael@0: 		int z = floorf(linear_b * (transform->grid_size-1));
michael@0: 		int x_n = ceilf(linear_r * (transform->grid_size-1));
michael@0: 		int y_n = ceilf(linear_g * (transform->grid_size-1));
michael@0: 		int z_n = ceilf(linear_b * (transform->grid_size-1));
michael@0: 		float x_d = linear_r * (transform->grid_size-1) - x;
michael@0: 		float y_d = linear_g * (transform->grid_size-1) - y;
michael@0: 		float z_d = linear_b * (transform->grid_size-1) - z;
michael@0: 
michael@0: 		float r_x1 = lerp(CLU(r_table,x,y,z), CLU(r_table,x_n,y,z), x_d);
michael@0: 		float r_x2 = lerp(CLU(r_table,x,y_n,z), CLU(r_table,x_n,y_n,z), x_d);
michael@0: 		float r_y1 = lerp(r_x1, r_x2, y_d);
michael@0: 		float r_x3 = lerp(CLU(r_table,x,y,z_n), CLU(r_table,x_n,y,z_n), x_d);
michael@0: 		float r_x4 = lerp(CLU(r_table,x,y_n,z_n), CLU(r_table,x_n,y_n,z_n), x_d);
michael@0: 		float r_y2 = lerp(r_x3, r_x4, y_d);
michael@0: 		float clut_r = lerp(r_y1, r_y2, z_d);
michael@0: 
michael@0: 		float g_x1 = lerp(CLU(g_table,x,y,z), CLU(g_table,x_n,y,z), x_d);
michael@0: 		float g_x2 = lerp(CLU(g_table,x,y_n,z), CLU(g_table,x_n,y_n,z), x_d);
michael@0: 		float g_y1 = lerp(g_x1, g_x2, y_d);
michael@0: 		float g_x3 = lerp(CLU(g_table,x,y,z_n), CLU(g_table,x_n,y,z_n), x_d);
michael@0: 		float g_x4 = lerp(CLU(g_table,x,y_n,z_n), CLU(g_table,x_n,y_n,z_n), x_d);
michael@0: 		float g_y2 = lerp(g_x3, g_x4, y_d);
michael@0: 		float clut_g = lerp(g_y1, g_y2, z_d);
michael@0: 
michael@0: 		float b_x1 = lerp(CLU(b_table,x,y,z), CLU(b_table,x_n,y,z), x_d);
michael@0: 		float b_x2 = lerp(CLU(b_table,x,y_n,z), CLU(b_table,x_n,y_n,z), x_d);
michael@0: 		float b_y1 = lerp(b_x1, b_x2, y_d);
michael@0: 		float b_x3 = lerp(CLU(b_table,x,y,z_n), CLU(b_table,x_n,y,z_n), x_d);
michael@0: 		float b_x4 = lerp(CLU(b_table,x,y_n,z_n), CLU(b_table,x_n,y_n,z_n), x_d);
michael@0: 		float b_y2 = lerp(b_x3, b_x4, y_d);
michael@0: 		float clut_b = lerp(b_y1, b_y2, z_d);
michael@0: 
michael@0: 		*dest++ = clamp_float(clut_r);
michael@0: 		*dest++ = clamp_float(clut_g);
michael@0: 		*dest++ = clamp_float(clut_b);
michael@0: 	}
michael@0: }
michael@0: 
michael@0: static void qcms_transform_module_clut(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	int xy_len = 1;
michael@0: 	int x_len = transform->grid_size;
michael@0: 	int len = x_len * x_len;
michael@0: 	float* r_table = transform->r_clut;
michael@0: 	float* g_table = transform->g_clut;
michael@0: 	float* b_table = transform->b_clut;
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float device_r = *src++;
michael@0: 		float device_g = *src++;
michael@0: 		float device_b = *src++;
michael@0: 		float linear_r = lut_interp_linear_float(device_r,
michael@0: 				transform->input_clut_table_r, transform->input_clut_table_length);
michael@0: 		float linear_g = lut_interp_linear_float(device_g,
michael@0: 				transform->input_clut_table_g, transform->input_clut_table_length);
michael@0: 		float linear_b = lut_interp_linear_float(device_b,
michael@0: 				transform->input_clut_table_b, transform->input_clut_table_length);
michael@0: 
michael@0: 		int x = floorf(linear_r * (transform->grid_size-1));
michael@0: 		int y = floorf(linear_g * (transform->grid_size-1));
michael@0: 		int z = floorf(linear_b * (transform->grid_size-1));
michael@0: 		int x_n = ceilf(linear_r * (transform->grid_size-1));
michael@0: 		int y_n = ceilf(linear_g * (transform->grid_size-1));
michael@0: 		int z_n = ceilf(linear_b * (transform->grid_size-1));
michael@0: 		float x_d = linear_r * (transform->grid_size-1) - x;
michael@0: 		float y_d = linear_g * (transform->grid_size-1) - y;
michael@0: 		float z_d = linear_b * (transform->grid_size-1) - z;
michael@0: 
michael@0: 		float r_x1 = lerp(CLU(r_table,x,y,z), CLU(r_table,x_n,y,z), x_d);
michael@0: 		float r_x2 = lerp(CLU(r_table,x,y_n,z), CLU(r_table,x_n,y_n,z), x_d);
michael@0: 		float r_y1 = lerp(r_x1, r_x2, y_d);
michael@0: 		float r_x3 = lerp(CLU(r_table,x,y,z_n), CLU(r_table,x_n,y,z_n), x_d);
michael@0: 		float r_x4 = lerp(CLU(r_table,x,y_n,z_n), CLU(r_table,x_n,y_n,z_n), x_d);
michael@0: 		float r_y2 = lerp(r_x3, r_x4, y_d);
michael@0: 		float clut_r = lerp(r_y1, r_y2, z_d);
michael@0: 
michael@0: 		float g_x1 = lerp(CLU(g_table,x,y,z), CLU(g_table,x_n,y,z), x_d);
michael@0: 		float g_x2 = lerp(CLU(g_table,x,y_n,z), CLU(g_table,x_n,y_n,z), x_d);
michael@0: 		float g_y1 = lerp(g_x1, g_x2, y_d);
michael@0: 		float g_x3 = lerp(CLU(g_table,x,y,z_n), CLU(g_table,x_n,y,z_n), x_d);
michael@0: 		float g_x4 = lerp(CLU(g_table,x,y_n,z_n), CLU(g_table,x_n,y_n,z_n), x_d);
michael@0: 		float g_y2 = lerp(g_x3, g_x4, y_d);
michael@0: 		float clut_g = lerp(g_y1, g_y2, z_d);
michael@0: 
michael@0: 		float b_x1 = lerp(CLU(b_table,x,y,z), CLU(b_table,x_n,y,z), x_d);
michael@0: 		float b_x2 = lerp(CLU(b_table,x,y_n,z), CLU(b_table,x_n,y_n,z), x_d);
michael@0: 		float b_y1 = lerp(b_x1, b_x2, y_d);
michael@0: 		float b_x3 = lerp(CLU(b_table,x,y,z_n), CLU(b_table,x_n,y,z_n), x_d);
michael@0: 		float b_x4 = lerp(CLU(b_table,x,y_n,z_n), CLU(b_table,x_n,y_n,z_n), x_d);
michael@0: 		float b_y2 = lerp(b_x3, b_x4, y_d);
michael@0: 		float clut_b = lerp(b_y1, b_y2, z_d);
michael@0: 
michael@0: 		float pcs_r = lut_interp_linear_float(clut_r,
michael@0: 				transform->output_clut_table_r, transform->output_clut_table_length);
michael@0: 		float pcs_g = lut_interp_linear_float(clut_g,
michael@0: 				transform->output_clut_table_g, transform->output_clut_table_length);
michael@0: 		float pcs_b = lut_interp_linear_float(clut_b,
michael@0: 				transform->output_clut_table_b, transform->output_clut_table_length);
michael@0: 
michael@0: 		*dest++ = clamp_float(pcs_r);
michael@0: 		*dest++ = clamp_float(pcs_g);
michael@0: 		*dest++ = clamp_float(pcs_b);
michael@0: 	}
michael@0: }
michael@0: 
michael@0: /* NOT USED
michael@0: static void qcms_transform_module_tetra_clut(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	int xy_len = 1;
michael@0: 	int x_len = transform->grid_size;
michael@0: 	int len = x_len * x_len;
michael@0: 	float* r_table = transform->r_clut;
michael@0: 	float* g_table = transform->g_clut;
michael@0: 	float* b_table = transform->b_clut;
michael@0: 	float c0_r, c1_r, c2_r, c3_r;
michael@0: 	float c0_g, c1_g, c2_g, c3_g;
michael@0: 	float c0_b, c1_b, c2_b, c3_b;
michael@0: 	float clut_r, clut_g, clut_b;
michael@0: 	float pcs_r, pcs_g, pcs_b;
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float device_r = *src++;
michael@0: 		float device_g = *src++;
michael@0: 		float device_b = *src++;
michael@0: 		float linear_r = lut_interp_linear_float(device_r,
michael@0: 				transform->input_clut_table_r, transform->input_clut_table_length);
michael@0: 		float linear_g = lut_interp_linear_float(device_g,
michael@0: 				transform->input_clut_table_g, transform->input_clut_table_length);
michael@0: 		float linear_b = lut_interp_linear_float(device_b,
michael@0: 				transform->input_clut_table_b, transform->input_clut_table_length);
michael@0: 
michael@0: 		int x = floorf(linear_r * (transform->grid_size-1));
michael@0: 		int y = floorf(linear_g * (transform->grid_size-1));
michael@0: 		int z = floorf(linear_b * (transform->grid_size-1));
michael@0: 		int x_n = ceilf(linear_r * (transform->grid_size-1));
michael@0: 		int y_n = ceilf(linear_g * (transform->grid_size-1));
michael@0: 		int z_n = ceilf(linear_b * (transform->grid_size-1));
michael@0: 		float rx = linear_r * (transform->grid_size-1) - x;
michael@0: 		float ry = linear_g * (transform->grid_size-1) - y;
michael@0: 		float rz = linear_b * (transform->grid_size-1) - z;
michael@0: 
michael@0: 		c0_r = CLU(r_table, x, y, z);
michael@0: 		c0_g = CLU(g_table, x, y, z);
michael@0: 		c0_b = CLU(b_table, x, y, z);
michael@0: 		if( rx >= ry ) {
michael@0: 			if (ry >= rz) { //rx >= ry && ry >= rz
michael@0: 				c1_r = CLU(r_table, x_n, y, z) - c0_r;
michael@0: 				c2_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x_n, y, z);
michael@0: 				c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
michael@0: 				c1_g = CLU(g_table, x_n, y, z) - c0_g;
michael@0: 				c2_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x_n, y, z);
michael@0: 				c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
michael@0: 				c1_b = CLU(b_table, x_n, y, z) - c0_b;
michael@0: 				c2_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x_n, y, z);
michael@0: 				c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
michael@0: 			} else {
michael@0: 				if (rx >= rz) { //rx >= rz && rz >= ry
michael@0: 					c1_r = CLU(r_table, x_n, y, z) - c0_r;
michael@0: 					c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
michael@0: 					c3_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x_n, y, z);
michael@0: 					c1_g = CLU(g_table, x_n, y, z) - c0_g;
michael@0: 					c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
michael@0: 					c3_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x_n, y, z);
michael@0: 					c1_b = CLU(b_table, x_n, y, z) - c0_b;
michael@0: 					c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
michael@0: 					c3_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x_n, y, z);
michael@0: 				} else { //rz > rx && rx >= ry
michael@0: 					c1_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x, y, z_n);
michael@0: 					c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
michael@0: 					c3_r = CLU(r_table, x, y, z_n) - c0_r;
michael@0: 					c1_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x, y, z_n);
michael@0: 					c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
michael@0: 					c3_g = CLU(g_table, x, y, z_n) - c0_g;
michael@0: 					c1_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x, y, z_n);
michael@0: 					c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
michael@0: 					c3_b = CLU(b_table, x, y, z_n) - c0_b;
michael@0: 				}
michael@0: 			}
michael@0: 		} else {
michael@0: 			if (rx >= rz) { //ry > rx && rx >= rz
michael@0: 				c1_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x, y_n, z);
michael@0: 				c2_r = CLU(r_table, x_n, y_n, z) - c0_r;
michael@0: 				c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
michael@0: 				c1_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x, y_n, z);
michael@0: 				c2_g = CLU(g_table, x_n, y_n, z) - c0_g;
michael@0: 				c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
michael@0: 				c1_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x, y_n, z);
michael@0: 				c2_b = CLU(b_table, x_n, y_n, z) - c0_b;
michael@0: 				c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
michael@0: 			} else {
michael@0: 				if (ry >= rz) { //ry >= rz && rz > rx 
michael@0: 					c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
michael@0: 					c2_r = CLU(r_table, x, y_n, z) - c0_r;
michael@0: 					c3_r = CLU(r_table, x, y_n, z_n) - CLU(r_table, x, y_n, z);
michael@0: 					c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
michael@0: 					c2_g = CLU(g_table, x, y_n, z) - c0_g;
michael@0: 					c3_g = CLU(g_table, x, y_n, z_n) - CLU(g_table, x, y_n, z);
michael@0: 					c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
michael@0: 					c2_b = CLU(b_table, x, y_n, z) - c0_b;
michael@0: 					c3_b = CLU(b_table, x, y_n, z_n) - CLU(b_table, x, y_n, z);
michael@0: 				} else { //rz > ry && ry > rx
michael@0: 					c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
michael@0: 					c2_r = CLU(r_table, x, y_n, z) - c0_r;
michael@0: 					c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
michael@0: 					c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
michael@0: 					c2_g = CLU(g_table, x, y_n, z) - c0_g;
michael@0: 					c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
michael@0: 					c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
michael@0: 					c2_b = CLU(b_table, x, y_n, z) - c0_b;
michael@0: 					c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
michael@0: 				}
michael@0: 			}
michael@0: 		}
michael@0: 
michael@0: 		clut_r = c0_r + c1_r*rx + c2_r*ry + c3_r*rz;
michael@0: 		clut_g = c0_g + c1_g*rx + c2_g*ry + c3_g*rz;
michael@0: 		clut_b = c0_b + c1_b*rx + c2_b*ry + c3_b*rz;
michael@0: 
michael@0: 		pcs_r = lut_interp_linear_float(clut_r,
michael@0: 				transform->output_clut_table_r, transform->output_clut_table_length);
michael@0: 		pcs_g = lut_interp_linear_float(clut_g,
michael@0: 				transform->output_clut_table_g, transform->output_clut_table_length);
michael@0: 		pcs_b = lut_interp_linear_float(clut_b,
michael@0: 				transform->output_clut_table_b, transform->output_clut_table_length);
michael@0: 		*dest++ = clamp_float(pcs_r);
michael@0: 		*dest++ = clamp_float(pcs_g);
michael@0: 		*dest++ = clamp_float(pcs_b);
michael@0: 	}
michael@0: }
michael@0: */
michael@0: 
michael@0: static void qcms_transform_module_gamma_table(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	float out_r, out_g, out_b;
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float in_r = *src++;
michael@0: 		float in_g = *src++;
michael@0: 		float in_b = *src++;
michael@0: 
michael@0: 		out_r = lut_interp_linear_float(in_r, transform->input_clut_table_r, 256);
michael@0: 		out_g = lut_interp_linear_float(in_g, transform->input_clut_table_g, 256);
michael@0: 		out_b = lut_interp_linear_float(in_b, transform->input_clut_table_b, 256);
michael@0: 
michael@0: 		*dest++ = clamp_float(out_r);
michael@0: 		*dest++ = clamp_float(out_g);
michael@0: 		*dest++ = clamp_float(out_b);
michael@0: 	}
michael@0: }
michael@0: 
michael@0: static void qcms_transform_module_gamma_lut(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	float out_r, out_g, out_b;
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float in_r = *src++;
michael@0: 		float in_g = *src++;
michael@0: 		float in_b = *src++;
michael@0: 
michael@0: 		out_r = lut_interp_linear(in_r,
michael@0: 				transform->output_gamma_lut_r, transform->output_gamma_lut_r_length);
michael@0: 		out_g = lut_interp_linear(in_g,
michael@0: 				transform->output_gamma_lut_g, transform->output_gamma_lut_g_length);
michael@0: 		out_b = lut_interp_linear(in_b,
michael@0: 				transform->output_gamma_lut_b, transform->output_gamma_lut_b_length);
michael@0: 
michael@0: 		*dest++ = clamp_float(out_r);
michael@0: 		*dest++ = clamp_float(out_g);
michael@0: 		*dest++ = clamp_float(out_b);
michael@0: 	}
michael@0: }
michael@0: 
michael@0: static void qcms_transform_module_matrix_translate(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	struct matrix mat;
michael@0: 
michael@0: 	/* store the results in column major mode
michael@0: 	 * this makes doing the multiplication with sse easier */
michael@0: 	mat.m[0][0] = transform->matrix.m[0][0];
michael@0: 	mat.m[1][0] = transform->matrix.m[0][1];
michael@0: 	mat.m[2][0] = transform->matrix.m[0][2];
michael@0: 	mat.m[0][1] = transform->matrix.m[1][0];
michael@0: 	mat.m[1][1] = transform->matrix.m[1][1];
michael@0: 	mat.m[2][1] = transform->matrix.m[1][2];
michael@0: 	mat.m[0][2] = transform->matrix.m[2][0];
michael@0: 	mat.m[1][2] = transform->matrix.m[2][1];
michael@0: 	mat.m[2][2] = transform->matrix.m[2][2];
michael@0: 
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float in_r = *src++;
michael@0: 		float in_g = *src++;
michael@0: 		float in_b = *src++;
michael@0: 
michael@0: 		float out_r = mat.m[0][0]*in_r + mat.m[1][0]*in_g + mat.m[2][0]*in_b + transform->tx;
michael@0: 		float out_g = mat.m[0][1]*in_r + mat.m[1][1]*in_g + mat.m[2][1]*in_b + transform->ty;
michael@0: 		float out_b = mat.m[0][2]*in_r + mat.m[1][2]*in_g + mat.m[2][2]*in_b + transform->tz;
michael@0: 
michael@0: 		*dest++ = clamp_float(out_r);
michael@0: 		*dest++ = clamp_float(out_g);
michael@0: 		*dest++ = clamp_float(out_b);
michael@0: 	}
michael@0: }
michael@0: 
michael@0: static void qcms_transform_module_matrix(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
michael@0: {
michael@0: 	size_t i;
michael@0: 	struct matrix mat;
michael@0: 
michael@0: 	/* store the results in column major mode
michael@0: 	 * this makes doing the multiplication with sse easier */
michael@0: 	mat.m[0][0] = transform->matrix.m[0][0];
michael@0: 	mat.m[1][0] = transform->matrix.m[0][1];
michael@0: 	mat.m[2][0] = transform->matrix.m[0][2];
michael@0: 	mat.m[0][1] = transform->matrix.m[1][0];
michael@0: 	mat.m[1][1] = transform->matrix.m[1][1];
michael@0: 	mat.m[2][1] = transform->matrix.m[1][2];
michael@0: 	mat.m[0][2] = transform->matrix.m[2][0];
michael@0: 	mat.m[1][2] = transform->matrix.m[2][1];
michael@0: 	mat.m[2][2] = transform->matrix.m[2][2];
michael@0: 
michael@0: 	for (i = 0; i < length; i++) {
michael@0: 		float in_r = *src++;
michael@0: 		float in_g = *src++;
michael@0: 		float in_b = *src++;
michael@0: 
michael@0: 		float out_r = mat.m[0][0]*in_r + mat.m[1][0]*in_g + mat.m[2][0]*in_b;
michael@0: 		float out_g = mat.m[0][1]*in_r + mat.m[1][1]*in_g + mat.m[2][1]*in_b;
michael@0: 		float out_b = mat.m[0][2]*in_r + mat.m[1][2]*in_g + mat.m[2][2]*in_b;
michael@0: 
michael@0: 		*dest++ = clamp_float(out_r);
michael@0: 		*dest++ = clamp_float(out_g);
michael@0: 		*dest++ = clamp_float(out_b);
michael@0: 	}
michael@0: }
michael@0: 
michael@0: static struct qcms_modular_transform* qcms_modular_transform_alloc() {
michael@0: 	return calloc(1, sizeof(struct qcms_modular_transform));
michael@0: }
michael@0: 
michael@0: static void qcms_modular_transform_release(struct qcms_modular_transform *transform)
michael@0: {
michael@0: 	struct qcms_modular_transform *next_transform;
michael@0: 	while (transform != NULL) {
michael@0: 		next_transform = transform->next_transform;
michael@0: 		// clut may use a single block of memory.
michael@0: 		// Perhaps we should remove this to simply the code.
michael@0: 		if (transform->input_clut_table_r + transform->input_clut_table_length == transform->input_clut_table_g && transform->input_clut_table_g + transform->input_clut_table_length == transform->input_clut_table_b) {
michael@0: 			if (transform->input_clut_table_r) free(transform->input_clut_table_r);
michael@0: 		} else {
michael@0: 			if (transform->input_clut_table_r) free(transform->input_clut_table_r);
michael@0: 			if (transform->input_clut_table_g) free(transform->input_clut_table_g);
michael@0: 			if (transform->input_clut_table_b) free(transform->input_clut_table_b);
michael@0: 		}
michael@0: 		if (transform->r_clut + 1 == transform->g_clut && transform->g_clut + 1 == transform->b_clut) {
michael@0: 			if (transform->r_clut) free(transform->r_clut);
michael@0: 		} else {
michael@0: 			if (transform->r_clut) free(transform->r_clut);
michael@0: 			if (transform->g_clut) free(transform->g_clut);
michael@0: 			if (transform->b_clut) free(transform->b_clut);
michael@0: 		}
michael@0: 		if (transform->output_clut_table_r + transform->output_clut_table_length == transform->output_clut_table_g && transform->output_clut_table_g+ transform->output_clut_table_length == transform->output_clut_table_b) {
michael@0: 			if (transform->output_clut_table_r) free(transform->output_clut_table_r);
michael@0: 		} else {
michael@0: 			if (transform->output_clut_table_r) free(transform->output_clut_table_r);
michael@0: 			if (transform->output_clut_table_g) free(transform->output_clut_table_g);
michael@0: 			if (transform->output_clut_table_b) free(transform->output_clut_table_b);
michael@0: 		}
michael@0: 		if (transform->output_gamma_lut_r) free(transform->output_gamma_lut_r);
michael@0: 		if (transform->output_gamma_lut_g) free(transform->output_gamma_lut_g);
michael@0: 		if (transform->output_gamma_lut_b) free(transform->output_gamma_lut_b);
michael@0: 		free(transform);
michael@0: 		transform = next_transform;
michael@0: 	}
michael@0: }
michael@0: 
michael@0: /* Set transform to be the next element in the linked list. */
michael@0: static void append_transform(struct qcms_modular_transform *transform, struct qcms_modular_transform ***next_transform)
michael@0: {
michael@0: 	**next_transform = transform;
michael@0: 	while (transform) {
michael@0: 		*next_transform = &(transform->next_transform);
michael@0: 		transform = transform->next_transform;
michael@0: 	}
michael@0: }
michael@0: 
michael@0: /* reverse the transformation list (used by mBA) */
michael@0: static struct qcms_modular_transform* reverse_transform(struct qcms_modular_transform *transform) 
michael@0: {
michael@0: 	struct qcms_modular_transform *prev_transform = NULL;
michael@0: 	while (transform != NULL) {
michael@0: 		struct qcms_modular_transform *next_transform = transform->next_transform;
michael@0: 		transform->next_transform = prev_transform;
michael@0: 		prev_transform = transform;
michael@0: 		transform = next_transform;
michael@0: 	}
michael@0: 	
michael@0: 	return prev_transform;
michael@0: }
michael@0: 
michael@0: #define EMPTY_TRANSFORM_LIST NULL
michael@0: static struct qcms_modular_transform* qcms_modular_transform_create_mAB(struct lutmABType *lut)
michael@0: {
michael@0: 	struct qcms_modular_transform *first_transform = NULL;
michael@0: 	struct qcms_modular_transform **next_transform = &first_transform;
michael@0: 	struct qcms_modular_transform *transform = NULL;
michael@0: 
michael@0: 	if (lut->a_curves[0] != NULL) {
michael@0: 		size_t clut_length;
michael@0: 		float *clut;
michael@0: 
michael@0: 		// If the A curve is present this also implies the 
michael@0: 		// presence of a CLUT.
michael@0: 		if (!lut->clut_table) 
michael@0: 			goto fail;
michael@0: 
michael@0: 		// Prepare A curve.
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform)
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->input_clut_table_r = build_input_gamma_table(lut->a_curves[0]);
michael@0: 		transform->input_clut_table_g = build_input_gamma_table(lut->a_curves[1]);
michael@0: 		transform->input_clut_table_b = build_input_gamma_table(lut->a_curves[2]);
michael@0: 		transform->transform_module_fn = qcms_transform_module_gamma_table;
michael@0: 		if (lut->num_grid_points[0] != lut->num_grid_points[1] ||
michael@0: 			lut->num_grid_points[1] != lut->num_grid_points[2] ) {
michael@0: 			//XXX: We don't currently support clut that are not squared!
michael@0: 			goto fail;
michael@0: 		}
michael@0: 
michael@0: 		// Prepare CLUT
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		clut_length = sizeof(float)*pow(lut->num_grid_points[0], 3)*3;
michael@0: 		clut = malloc(clut_length);
michael@0: 		if (!clut)
michael@0: 			goto fail;
michael@0: 		memcpy(clut, lut->clut_table, clut_length);
michael@0: 		transform->r_clut = clut + 0;
michael@0: 		transform->g_clut = clut + 1;
michael@0: 		transform->b_clut = clut + 2;
michael@0: 		transform->grid_size = lut->num_grid_points[0];
michael@0: 		transform->transform_module_fn = qcms_transform_module_clut_only;
michael@0: 	}
michael@0: 	if (lut->m_curves[0] != NULL) {
michael@0: 		// M curve imples the presence of a Matrix
michael@0: 
michael@0: 		// Prepare M curve
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform)
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->input_clut_table_r = build_input_gamma_table(lut->m_curves[0]);
michael@0: 		transform->input_clut_table_g = build_input_gamma_table(lut->m_curves[1]);
michael@0: 		transform->input_clut_table_b = build_input_gamma_table(lut->m_curves[2]);
michael@0: 		transform->transform_module_fn = qcms_transform_module_gamma_table;
michael@0: 
michael@0: 		// Prepare Matrix
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->matrix = build_mAB_matrix(lut);
michael@0: 		if (transform->matrix.invalid)
michael@0: 			goto fail;
michael@0: 		transform->tx = s15Fixed16Number_to_float(lut->e03);
michael@0: 		transform->ty = s15Fixed16Number_to_float(lut->e13);
michael@0: 		transform->tz = s15Fixed16Number_to_float(lut->e23);
michael@0: 		transform->transform_module_fn = qcms_transform_module_matrix_translate;
michael@0: 	}
michael@0: 	if (lut->b_curves[0] != NULL) {
michael@0: 		// Prepare B curve
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->input_clut_table_r = build_input_gamma_table(lut->b_curves[0]);
michael@0: 		transform->input_clut_table_g = build_input_gamma_table(lut->b_curves[1]);
michael@0: 		transform->input_clut_table_b = build_input_gamma_table(lut->b_curves[2]);
michael@0: 		transform->transform_module_fn = qcms_transform_module_gamma_table;
michael@0: 	} else {
michael@0: 		// B curve is mandatory
michael@0: 		goto fail;
michael@0: 	}
michael@0: 
michael@0: 	if (lut->reversed) {
michael@0: 		// mBA are identical to mAB except that the transformation order
michael@0: 		// is reversed
michael@0: 		first_transform = reverse_transform(first_transform);
michael@0: 	}
michael@0: 
michael@0: 	return first_transform;
michael@0: fail:
michael@0: 	qcms_modular_transform_release(first_transform);
michael@0: 	return NULL;
michael@0: }
michael@0: 
michael@0: static struct qcms_modular_transform* qcms_modular_transform_create_lut(struct lutType *lut)
michael@0: {
michael@0: 	struct qcms_modular_transform *first_transform = NULL;
michael@0: 	struct qcms_modular_transform **next_transform = &first_transform;
michael@0: 	struct qcms_modular_transform *transform = NULL;
michael@0: 
michael@0: 	size_t in_curve_len, clut_length, out_curve_len;
michael@0: 	float *in_curves, *clut, *out_curves;
michael@0: 
michael@0: 	// Prepare Matrix
michael@0: 	transform = qcms_modular_transform_alloc();
michael@0: 	if (!transform) 
michael@0: 		goto fail;
michael@0: 	append_transform(transform, &next_transform);
michael@0: 	transform->matrix = build_lut_matrix(lut);
michael@0: 	if (transform->matrix.invalid)
michael@0: 		goto fail;
michael@0: 	transform->transform_module_fn = qcms_transform_module_matrix;
michael@0: 
michael@0: 	// Prepare input curves
michael@0: 	transform = qcms_modular_transform_alloc();
michael@0: 	if (!transform) 
michael@0: 		goto fail;
michael@0: 	append_transform(transform, &next_transform);
michael@0: 	in_curve_len = sizeof(float)*lut->num_input_table_entries * 3;
michael@0: 	in_curves = malloc(in_curve_len);
michael@0: 	if (!in_curves) 
michael@0: 		goto fail;
michael@0: 	memcpy(in_curves, lut->input_table, in_curve_len);
michael@0: 	transform->input_clut_table_r = in_curves + lut->num_input_table_entries * 0;
michael@0: 	transform->input_clut_table_g = in_curves + lut->num_input_table_entries * 1;
michael@0: 	transform->input_clut_table_b = in_curves + lut->num_input_table_entries * 2;
michael@0: 	transform->input_clut_table_length = lut->num_input_table_entries;
michael@0: 
michael@0: 	// Prepare table
michael@0: 	clut_length = sizeof(float)*pow(lut->num_clut_grid_points, 3)*3;
michael@0: 	clut = malloc(clut_length);
michael@0: 	if (!clut) 
michael@0: 		goto fail;
michael@0: 	memcpy(clut, lut->clut_table, clut_length);
michael@0: 	transform->r_clut = clut + 0;
michael@0: 	transform->g_clut = clut + 1;
michael@0: 	transform->b_clut = clut + 2;
michael@0: 	transform->grid_size = lut->num_clut_grid_points;
michael@0: 
michael@0: 	// Prepare output curves
michael@0: 	out_curve_len = sizeof(float) * lut->num_output_table_entries * 3;
michael@0: 	out_curves = malloc(out_curve_len);
michael@0: 	if (!out_curves) 
michael@0: 		goto fail;
michael@0: 	memcpy(out_curves, lut->output_table, out_curve_len);
michael@0: 	transform->output_clut_table_r = out_curves + lut->num_output_table_entries * 0;
michael@0: 	transform->output_clut_table_g = out_curves + lut->num_output_table_entries * 1;
michael@0: 	transform->output_clut_table_b = out_curves + lut->num_output_table_entries * 2;
michael@0: 	transform->output_clut_table_length = lut->num_output_table_entries;
michael@0: 	transform->transform_module_fn = qcms_transform_module_clut;
michael@0: 
michael@0: 	return first_transform;
michael@0: fail:
michael@0: 	qcms_modular_transform_release(first_transform);
michael@0: 	return NULL;
michael@0: }
michael@0: 
michael@0: struct qcms_modular_transform* qcms_modular_transform_create_input(qcms_profile *in)
michael@0: {
michael@0: 	struct qcms_modular_transform *first_transform = NULL;
michael@0: 	struct qcms_modular_transform **next_transform = &first_transform;
michael@0: 
michael@0: 	if (in->A2B0) {
michael@0: 		struct qcms_modular_transform *lut_transform;
michael@0: 		lut_transform = qcms_modular_transform_create_lut(in->A2B0);
michael@0: 		if (!lut_transform)
michael@0: 			goto fail;
michael@0: 		append_transform(lut_transform, &next_transform);
michael@0: 	} else if (in->mAB && in->mAB->num_in_channels == 3 && in->mAB->num_out_channels == 3) {
michael@0: 		struct qcms_modular_transform *mAB_transform;
michael@0: 		mAB_transform = qcms_modular_transform_create_mAB(in->mAB);
michael@0: 		if (!mAB_transform)
michael@0: 			goto fail;
michael@0: 		append_transform(mAB_transform, &next_transform);
michael@0: 
michael@0: 	} else {
michael@0: 		struct qcms_modular_transform *transform;
michael@0: 
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform)
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->input_clut_table_r = build_input_gamma_table(in->redTRC);
michael@0: 		transform->input_clut_table_g = build_input_gamma_table(in->greenTRC);
michael@0: 		transform->input_clut_table_b = build_input_gamma_table(in->blueTRC);
michael@0: 		transform->transform_module_fn = qcms_transform_module_gamma_table;
michael@0: 		if (!transform->input_clut_table_r || !transform->input_clut_table_g ||
michael@0: 				!transform->input_clut_table_b) {
michael@0: 			goto fail;
michael@0: 		}
michael@0: 
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->matrix.m[0][0] = 1/1.999969482421875f;
michael@0: 		transform->matrix.m[0][1] = 0.f;
michael@0: 		transform->matrix.m[0][2] = 0.f;
michael@0: 		transform->matrix.m[1][0] = 0.f;
michael@0: 		transform->matrix.m[1][1] = 1/1.999969482421875f;
michael@0: 		transform->matrix.m[1][2] = 0.f;
michael@0: 		transform->matrix.m[2][0] = 0.f;
michael@0: 		transform->matrix.m[2][1] = 0.f;
michael@0: 		transform->matrix.m[2][2] = 1/1.999969482421875f;
michael@0: 		transform->matrix.invalid = false;
michael@0: 		transform->transform_module_fn = qcms_transform_module_matrix;
michael@0: 
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->matrix = build_colorant_matrix(in);
michael@0: 		transform->transform_module_fn = qcms_transform_module_matrix;
michael@0: 	}
michael@0: 
michael@0: 	return first_transform;
michael@0: fail:
michael@0: 	qcms_modular_transform_release(first_transform);
michael@0: 	return EMPTY_TRANSFORM_LIST;
michael@0: }
michael@0: static struct qcms_modular_transform* qcms_modular_transform_create_output(qcms_profile *out)
michael@0: {
michael@0: 	struct qcms_modular_transform *first_transform = NULL;
michael@0: 	struct qcms_modular_transform **next_transform = &first_transform;
michael@0: 
michael@0: 	if (out->B2A0) {
michael@0: 		struct qcms_modular_transform *lut_transform;
michael@0: 		lut_transform = qcms_modular_transform_create_lut(out->B2A0);
michael@0: 		if (!lut_transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(lut_transform, &next_transform);
michael@0: 	} else if (out->mBA && out->mBA->num_in_channels == 3 && out->mBA->num_out_channels == 3) {
michael@0: 		struct qcms_modular_transform *lut_transform;
michael@0: 		lut_transform = qcms_modular_transform_create_mAB(out->mBA);
michael@0: 		if (!lut_transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(lut_transform, &next_transform);
michael@0: 	} else if (out->redTRC && out->greenTRC && out->blueTRC) {
michael@0: 		struct qcms_modular_transform *transform;
michael@0: 
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->matrix = matrix_invert(build_colorant_matrix(out));
michael@0: 		transform->transform_module_fn = qcms_transform_module_matrix;
michael@0: 
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		transform->matrix.m[0][0] = 1.999969482421875f;
michael@0: 		transform->matrix.m[0][1] = 0.f;
michael@0: 		transform->matrix.m[0][2] = 0.f;
michael@0: 		transform->matrix.m[1][0] = 0.f;
michael@0: 		transform->matrix.m[1][1] = 1.999969482421875f;
michael@0: 		transform->matrix.m[1][2] = 0.f;
michael@0: 		transform->matrix.m[2][0] = 0.f;
michael@0: 		transform->matrix.m[2][1] = 0.f;
michael@0: 		transform->matrix.m[2][2] = 1.999969482421875f;
michael@0: 		transform->matrix.invalid = false;
michael@0: 		transform->transform_module_fn = qcms_transform_module_matrix;
michael@0: 
michael@0: 		transform = qcms_modular_transform_alloc();
michael@0: 		if (!transform) 
michael@0: 			goto fail;
michael@0: 		append_transform(transform, &next_transform);
michael@0: 		build_output_lut(out->redTRC, &transform->output_gamma_lut_r,
michael@0: 			&transform->output_gamma_lut_r_length);
michael@0: 		build_output_lut(out->greenTRC, &transform->output_gamma_lut_g,
michael@0: 			&transform->output_gamma_lut_g_length);
michael@0: 		build_output_lut(out->blueTRC, &transform->output_gamma_lut_b,
michael@0: 			&transform->output_gamma_lut_b_length);
michael@0: 		transform->transform_module_fn = qcms_transform_module_gamma_lut;
michael@0: 
michael@0: 		if (!transform->output_gamma_lut_r || !transform->output_gamma_lut_g ||
michael@0: 				!transform->output_gamma_lut_b) {
michael@0: 			goto fail;
michael@0: 		}
michael@0: 	} else {
michael@0: 		assert(0 && "Unsupported output profile workflow.");
michael@0: 		return NULL;
michael@0: 	}
michael@0: 
michael@0: 	return first_transform;
michael@0: fail:
michael@0: 	qcms_modular_transform_release(first_transform);
michael@0: 	return EMPTY_TRANSFORM_LIST;
michael@0: }
michael@0: 
michael@0: /* Not Completed
michael@0: // Simplify the transformation chain to an equivalent transformation chain
michael@0: static struct qcms_modular_transform* qcms_modular_transform_reduce(struct qcms_modular_transform *transform)
michael@0: {
michael@0: 	struct qcms_modular_transform *first_transform = NULL;
michael@0: 	struct qcms_modular_transform *curr_trans = transform;
michael@0: 	struct qcms_modular_transform *prev_trans = NULL;
michael@0: 	while (curr_trans) {
michael@0: 		struct qcms_modular_transform *next_trans = curr_trans->next_transform;
michael@0: 		if (curr_trans->transform_module_fn == qcms_transform_module_matrix) {
michael@0: 			if (next_trans && next_trans->transform_module_fn == qcms_transform_module_matrix) {
michael@0: 				curr_trans->matrix = matrix_multiply(curr_trans->matrix, next_trans->matrix);
michael@0: 				goto remove_next;	
michael@0: 			}
michael@0: 		}
michael@0: 		if (curr_trans->transform_module_fn == qcms_transform_module_gamma_table) {
michael@0: 			bool isLinear = true;
michael@0: 			uint16_t i;
michael@0: 			for (i = 0; isLinear && i < 256; i++) {
michael@0: 				isLinear &= (int)(curr_trans->input_clut_table_r[i] * 255) == i;
michael@0: 				isLinear &= (int)(curr_trans->input_clut_table_g[i] * 255) == i;
michael@0: 				isLinear &= (int)(curr_trans->input_clut_table_b[i] * 255) == i;
michael@0: 			}
michael@0: 			goto remove_current;
michael@0: 		}
michael@0: 		
michael@0: next_transform:
michael@0: 		if (!next_trans) break;
michael@0: 		prev_trans = curr_trans;
michael@0: 		curr_trans = next_trans;
michael@0: 		continue;
michael@0: remove_current:
michael@0: 		if (curr_trans == transform) {
michael@0: 			//Update head
michael@0: 			transform = next_trans;
michael@0: 		} else {
michael@0: 			prev_trans->next_transform = next_trans;
michael@0: 		}
michael@0: 		curr_trans->next_transform = NULL;
michael@0: 		qcms_modular_transform_release(curr_trans);
michael@0: 		//return transform;
michael@0: 		return qcms_modular_transform_reduce(transform);
michael@0: remove_next:
michael@0: 		curr_trans->next_transform = next_trans->next_transform;
michael@0: 		next_trans->next_transform = NULL;
michael@0: 		qcms_modular_transform_release(next_trans);
michael@0: 		continue;
michael@0: 	}
michael@0: 	return transform;
michael@0: }
michael@0: */
michael@0: 
michael@0: static struct qcms_modular_transform* qcms_modular_transform_create(qcms_profile *in, qcms_profile *out)
michael@0: {
michael@0: 	struct qcms_modular_transform *first_transform = NULL;
michael@0: 	struct qcms_modular_transform **next_transform = &first_transform;
michael@0: 
michael@0: 	if (in->color_space == RGB_SIGNATURE) {
michael@0: 		struct qcms_modular_transform* rgb_to_pcs;
michael@0: 		rgb_to_pcs = qcms_modular_transform_create_input(in);
michael@0: 		if (!rgb_to_pcs) 
michael@0: 			goto fail;
michael@0: 		append_transform(rgb_to_pcs, &next_transform);
michael@0: 	} else {
michael@0: 		assert(0 && "input color space not supported");
michael@0: 		goto fail;
michael@0: 	}
michael@0: 
michael@0: 	if (in->pcs == LAB_SIGNATURE && out->pcs == XYZ_SIGNATURE) {
michael@0: 		struct qcms_modular_transform* lab_to_pcs;
michael@0: 		lab_to_pcs = qcms_modular_transform_alloc();
michael@0: 		if (!lab_to_pcs) 
michael@0: 			goto fail;
michael@0: 		append_transform(lab_to_pcs, &next_transform);
michael@0: 		lab_to_pcs->transform_module_fn = qcms_transform_module_LAB_to_XYZ;
michael@0: 	}
michael@0: 
michael@0: 	// This does not improve accuracy in practice, something is wrong here.
michael@0: 	//if (in->chromaticAdaption.invalid == false) {
michael@0: 	//	struct qcms_modular_transform* chromaticAdaption;
michael@0: 	//	chromaticAdaption = qcms_modular_transform_alloc();
michael@0: 	//	if (!chromaticAdaption) 
michael@0: 	//		goto fail;
michael@0: 	//	append_transform(chromaticAdaption, &next_transform);
michael@0: 	//	chromaticAdaption->matrix = matrix_invert(in->chromaticAdaption);
michael@0: 	//	chromaticAdaption->transform_module_fn = qcms_transform_module_matrix;
michael@0: 	//}
michael@0: 
michael@0:         if (in->pcs == XYZ_SIGNATURE && out->pcs == LAB_SIGNATURE) {
michael@0: 		struct qcms_modular_transform* pcs_to_lab;
michael@0: 		pcs_to_lab = qcms_modular_transform_alloc();
michael@0: 		if (!pcs_to_lab) 
michael@0: 			goto fail;
michael@0: 		append_transform(pcs_to_lab, &next_transform);
michael@0: 		pcs_to_lab->transform_module_fn = qcms_transform_module_XYZ_to_LAB;
michael@0: 	}
michael@0: 
michael@0: 	if (out->color_space == RGB_SIGNATURE) {
michael@0: 		struct qcms_modular_transform* pcs_to_rgb;
michael@0: 		pcs_to_rgb = qcms_modular_transform_create_output(out);
michael@0: 		if (!pcs_to_rgb) 
michael@0: 			goto fail;
michael@0: 		append_transform(pcs_to_rgb, &next_transform);
michael@0: 	} else {
michael@0: 		assert(0 && "output color space not supported");
michael@0: 		goto fail;
michael@0: 	}
michael@0: 	// Not Completed
michael@0: 	//return qcms_modular_transform_reduce(first_transform);
michael@0: 	return first_transform;
michael@0: fail:
michael@0: 	qcms_modular_transform_release(first_transform);
michael@0: 	return EMPTY_TRANSFORM_LIST;
michael@0: }
michael@0: 
michael@0: static float* qcms_modular_transform_data(struct qcms_modular_transform *transform, float *src, float *dest, size_t len)
michael@0: {
michael@0:         while (transform != NULL) {
michael@0:                 // Keep swaping src/dest when performing a transform to use less memory.
michael@0:                 float *new_src = dest;
michael@0: 		const transform_module_fn_t transform_fn = transform->transform_module_fn;
michael@0: 		if (transform_fn != qcms_transform_module_gamma_table &&
michael@0: 		    transform_fn != qcms_transform_module_gamma_lut &&
michael@0: 		    transform_fn != qcms_transform_module_clut &&
michael@0: 		    transform_fn != qcms_transform_module_clut_only &&
michael@0: 		    transform_fn != qcms_transform_module_matrix &&
michael@0: 		    transform_fn != qcms_transform_module_matrix_translate &&
michael@0: 		    transform_fn != qcms_transform_module_LAB_to_XYZ &&
michael@0: 		    transform_fn != qcms_transform_module_XYZ_to_LAB) {
michael@0: 			assert(0 && "Unsupported transform module");
michael@0: 			return NULL;
michael@0: 		}
michael@0:                 transform->transform_module_fn(transform,src,dest,len);
michael@0:                 dest = src;
michael@0:                 src = new_src;
michael@0:                 transform = transform->next_transform;
michael@0:         }
michael@0:         // The results end up in the src buffer because of the switching
michael@0:         return src;
michael@0: }
michael@0: 
michael@0: float* qcms_chain_transform(qcms_profile *in, qcms_profile *out, float *src, float *dest, size_t lutSize)
michael@0: {
michael@0: 	struct qcms_modular_transform *transform_list = qcms_modular_transform_create(in, out);
michael@0: 	if (transform_list != NULL) {
michael@0: 		float *lut = qcms_modular_transform_data(transform_list, src, dest, lutSize/3);
michael@0: 		qcms_modular_transform_release(transform_list);
michael@0: 		return lut;
michael@0: 	}
michael@0: 	return NULL;
michael@0: }