diff --git a/chain.c b/chain.c
new file mode 100644
index 0000000..9958ab1
--- /dev/null
+++ b/chain.c
@@ -0,0 +1,988 @@
+/* vim: set ts=8 sw=8 noexpandtab: */
+//  qcms
+//  Copyright (C) 2009 Mozilla Corporation
+//  Copyright (C) 1998-2007 Marti Maria
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+#include <stdlib.h>
+#include <math.h>
+#include <assert.h>
+#include <string.h> //memcpy
+#include "qcmsint.h"
+#include "transform_util.h"
+#include "matrix.h"
+
+static struct matrix build_lut_matrix(struct lutType *lut)
+{
+	struct matrix result;
+	if (lut) {
+		result.m[0][0] = s15Fixed16Number_to_float(lut->e00);
+		result.m[0][1] = s15Fixed16Number_to_float(lut->e01);
+		result.m[0][2] = s15Fixed16Number_to_float(lut->e02);
+		result.m[1][0] = s15Fixed16Number_to_float(lut->e10);
+		result.m[1][1] = s15Fixed16Number_to_float(lut->e11);
+		result.m[1][2] = s15Fixed16Number_to_float(lut->e12);
+		result.m[2][0] = s15Fixed16Number_to_float(lut->e20);
+		result.m[2][1] = s15Fixed16Number_to_float(lut->e21);
+		result.m[2][2] = s15Fixed16Number_to_float(lut->e22);
+		result.invalid = false;
+	} else {
+		memset(&result, 0, sizeof(struct matrix));
+		result.invalid = true;
+	}
+	return result;
+}
+
+static struct matrix build_mAB_matrix(struct lutmABType *lut)
+{
+	struct matrix result;
+	if (lut) {
+		result.m[0][0] = s15Fixed16Number_to_float(lut->e00);
+		result.m[0][1] = s15Fixed16Number_to_float(lut->e01);
+		result.m[0][2] = s15Fixed16Number_to_float(lut->e02);
+		result.m[1][0] = s15Fixed16Number_to_float(lut->e10);
+		result.m[1][1] = s15Fixed16Number_to_float(lut->e11);
+		result.m[1][2] = s15Fixed16Number_to_float(lut->e12);
+		result.m[2][0] = s15Fixed16Number_to_float(lut->e20);
+		result.m[2][1] = s15Fixed16Number_to_float(lut->e21);
+		result.m[2][2] = s15Fixed16Number_to_float(lut->e22);
+		result.invalid = false;
+	} else {
+		memset(&result, 0, sizeof(struct matrix));
+		result.invalid = true;
+	}
+	return result;
+}
+
+//Based on lcms cmsLab2XYZ
+#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)
+#define f_1(t) (t <= (24.0f/116.0f)) ? ((108.0/841.0) * (t - (16.0/116.0))) : (t * t * t)
+static void qcms_transform_module_LAB_to_XYZ(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	// lcms: D50 XYZ values
+	float WhitePointX = 0.9642f;
+	float WhitePointY = 1.0f;
+	float WhitePointZ = 0.8249f;
+	for (i = 0; i < length; i++) {
+		float device_L = *src++ * 100.0f;
+		float device_a = *src++ * 255.0f - 128.0f;
+		float device_b = *src++ * 255.0f - 128.0f;
+		float y = (device_L + 16.0f) / 116.0f;
+
+		float X = f_1((y + 0.002f * device_a)) * WhitePointX;
+		float Y = f_1(y) * WhitePointY;
+		float Z = f_1((y - 0.005f * device_b)) * WhitePointZ;
+		*dest++ = X / (1.0 + 32767.0/32768.0);
+		*dest++ = Y / (1.0 + 32767.0/32768.0);
+		*dest++ = Z / (1.0 + 32767.0/32768.0);
+	}
+}
+
+//Based on lcms cmsXYZ2Lab
+static void qcms_transform_module_XYZ_to_LAB(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+        // lcms: D50 XYZ values
+        float WhitePointX = 0.9642f;
+        float WhitePointY = 1.0f;
+        float WhitePointZ = 0.8249f;
+        for (i = 0; i < length; i++) {
+                float device_x = *src++ * (1.0 + 32767.0/32768.0) / WhitePointX;
+                float device_y = *src++ * (1.0 + 32767.0/32768.0) / WhitePointY;
+                float device_z = *src++ * (1.0 + 32767.0/32768.0) / WhitePointZ;
+
+		float fx = f(device_x);
+		float fy = f(device_y);
+		float fz = f(device_z);
+
+                float L = 116.0f*fy - 16.0f;
+                float a = 500.0f*(fx - fy);
+                float b = 200.0f*(fy - fz);
+                *dest++ = L / 100.0f;
+                *dest++ = (a+128.0f) / 255.0f;
+                *dest++ = (b+128.0f) / 255.0f;
+        }
+
+}
+
+static void qcms_transform_module_clut_only(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	int xy_len = 1;
+	int x_len = transform->grid_size;
+	int len = x_len * x_len;
+	float* r_table = transform->r_clut;
+	float* g_table = transform->g_clut;
+	float* b_table = transform->b_clut;
+
+	for (i = 0; i < length; i++) {
+		float linear_r = *src++;
+		float linear_g = *src++;
+		float linear_b = *src++;
+
+		int x = floor(linear_r * (transform->grid_size-1));
+		int y = floor(linear_g * (transform->grid_size-1));
+		int z = floor(linear_b * (transform->grid_size-1));
+		int x_n = ceil(linear_r * (transform->grid_size-1));
+		int y_n = ceil(linear_g * (transform->grid_size-1));
+		int z_n = ceil(linear_b * (transform->grid_size-1));
+		float x_d = linear_r * (transform->grid_size-1) - x;
+		float y_d = linear_g * (transform->grid_size-1) - y;
+		float z_d = linear_b * (transform->grid_size-1) - z;
+
+		float r_x1 = lerp(CLU(r_table,x,y,z), CLU(r_table,x_n,y,z), x_d);
+		float r_x2 = lerp(CLU(r_table,x,y_n,z), CLU(r_table,x_n,y_n,z), x_d);
+		float r_y1 = lerp(r_x1, r_x2, y_d);
+		float r_x3 = lerp(CLU(r_table,x,y,z_n), CLU(r_table,x_n,y,z_n), x_d);
+		float r_x4 = lerp(CLU(r_table,x,y_n,z_n), CLU(r_table,x_n,y_n,z_n), x_d);
+		float r_y2 = lerp(r_x3, r_x4, y_d);
+		float clut_r = lerp(r_y1, r_y2, z_d);
+
+		float g_x1 = lerp(CLU(g_table,x,y,z), CLU(g_table,x_n,y,z), x_d);
+		float g_x2 = lerp(CLU(g_table,x,y_n,z), CLU(g_table,x_n,y_n,z), x_d);
+		float g_y1 = lerp(g_x1, g_x2, y_d);
+		float g_x3 = lerp(CLU(g_table,x,y,z_n), CLU(g_table,x_n,y,z_n), x_d);
+		float g_x4 = lerp(CLU(g_table,x,y_n,z_n), CLU(g_table,x_n,y_n,z_n), x_d);
+		float g_y2 = lerp(g_x3, g_x4, y_d);
+		float clut_g = lerp(g_y1, g_y2, z_d);
+
+		float b_x1 = lerp(CLU(b_table,x,y,z), CLU(b_table,x_n,y,z), x_d);
+		float b_x2 = lerp(CLU(b_table,x,y_n,z), CLU(b_table,x_n,y_n,z), x_d);
+		float b_y1 = lerp(b_x1, b_x2, y_d);
+		float b_x3 = lerp(CLU(b_table,x,y,z_n), CLU(b_table,x_n,y,z_n), x_d);
+		float b_x4 = lerp(CLU(b_table,x,y_n,z_n), CLU(b_table,x_n,y_n,z_n), x_d);
+		float b_y2 = lerp(b_x3, b_x4, y_d);
+		float clut_b = lerp(b_y1, b_y2, z_d);
+
+		*dest++ = clamp_float(clut_r);
+		*dest++ = clamp_float(clut_g);
+		*dest++ = clamp_float(clut_b);
+	}
+}
+
+static void qcms_transform_module_clut(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	int xy_len = 1;
+	int x_len = transform->grid_size;
+	int len = x_len * x_len;
+	float* r_table = transform->r_clut;
+	float* g_table = transform->g_clut;
+	float* b_table = transform->b_clut;
+	for (i = 0; i < length; i++) {
+		float device_r = *src++;
+		float device_g = *src++;
+		float device_b = *src++;
+		float linear_r = lut_interp_linear_float(device_r,
+				transform->input_clut_table_r, transform->input_clut_table_length);
+		float linear_g = lut_interp_linear_float(device_g,
+				transform->input_clut_table_g, transform->input_clut_table_length);
+		float linear_b = lut_interp_linear_float(device_b,
+				transform->input_clut_table_b, transform->input_clut_table_length);
+
+		int x = floor(linear_r * (transform->grid_size-1));
+		int y = floor(linear_g * (transform->grid_size-1));
+		int z = floor(linear_b * (transform->grid_size-1));
+		int x_n = ceil(linear_r * (transform->grid_size-1));
+		int y_n = ceil(linear_g * (transform->grid_size-1));
+		int z_n = ceil(linear_b * (transform->grid_size-1));
+		float x_d = linear_r * (transform->grid_size-1) - x;
+		float y_d = linear_g * (transform->grid_size-1) - y;
+		float z_d = linear_b * (transform->grid_size-1) - z;
+
+		float r_x1 = lerp(CLU(r_table,x,y,z), CLU(r_table,x_n,y,z), x_d);
+		float r_x2 = lerp(CLU(r_table,x,y_n,z), CLU(r_table,x_n,y_n,z), x_d);
+		float r_y1 = lerp(r_x1, r_x2, y_d);
+		float r_x3 = lerp(CLU(r_table,x,y,z_n), CLU(r_table,x_n,y,z_n), x_d);
+		float r_x4 = lerp(CLU(r_table,x,y_n,z_n), CLU(r_table,x_n,y_n,z_n), x_d);
+		float r_y2 = lerp(r_x3, r_x4, y_d);
+		float clut_r = lerp(r_y1, r_y2, z_d);
+
+		float g_x1 = lerp(CLU(g_table,x,y,z), CLU(g_table,x_n,y,z), x_d);
+		float g_x2 = lerp(CLU(g_table,x,y_n,z), CLU(g_table,x_n,y_n,z), x_d);
+		float g_y1 = lerp(g_x1, g_x2, y_d);
+		float g_x3 = lerp(CLU(g_table,x,y,z_n), CLU(g_table,x_n,y,z_n), x_d);
+		float g_x4 = lerp(CLU(g_table,x,y_n,z_n), CLU(g_table,x_n,y_n,z_n), x_d);
+		float g_y2 = lerp(g_x3, g_x4, y_d);
+		float clut_g = lerp(g_y1, g_y2, z_d);
+
+		float b_x1 = lerp(CLU(b_table,x,y,z), CLU(b_table,x_n,y,z), x_d);
+		float b_x2 = lerp(CLU(b_table,x,y_n,z), CLU(b_table,x_n,y_n,z), x_d);
+		float b_y1 = lerp(b_x1, b_x2, y_d);
+		float b_x3 = lerp(CLU(b_table,x,y,z_n), CLU(b_table,x_n,y,z_n), x_d);
+		float b_x4 = lerp(CLU(b_table,x,y_n,z_n), CLU(b_table,x_n,y_n,z_n), x_d);
+		float b_y2 = lerp(b_x3, b_x4, y_d);
+		float clut_b = lerp(b_y1, b_y2, z_d);
+
+		float pcs_r = lut_interp_linear_float(clut_r,
+				transform->output_clut_table_r, transform->output_clut_table_length);
+		float pcs_g = lut_interp_linear_float(clut_g,
+				transform->output_clut_table_g, transform->output_clut_table_length);
+		float pcs_b = lut_interp_linear_float(clut_b,
+				transform->output_clut_table_b, transform->output_clut_table_length);
+
+		*dest++ = clamp_float(pcs_r);
+		*dest++ = clamp_float(pcs_g);
+		*dest++ = clamp_float(pcs_b);
+	}
+}
+
+/* NOT USED
+static void qcms_transform_module_tetra_clut(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	int xy_len = 1;
+	int x_len = transform->grid_size;
+	int len = x_len * x_len;
+	float* r_table = transform->r_clut;
+	float* g_table = transform->g_clut;
+	float* b_table = transform->b_clut;
+	float c0_r, c1_r, c2_r, c3_r;
+	float c0_g, c1_g, c2_g, c3_g;
+	float c0_b, c1_b, c2_b, c3_b;
+	float clut_r, clut_g, clut_b;
+	float pcs_r, pcs_g, pcs_b;
+	for (i = 0; i < length; i++) {
+		float device_r = *src++;
+		float device_g = *src++;
+		float device_b = *src++;
+		float linear_r = lut_interp_linear_float(device_r,
+				transform->input_clut_table_r, transform->input_clut_table_length);
+		float linear_g = lut_interp_linear_float(device_g,
+				transform->input_clut_table_g, transform->input_clut_table_length);
+		float linear_b = lut_interp_linear_float(device_b,
+				transform->input_clut_table_b, transform->input_clut_table_length);
+
+		int x = floor(linear_r * (transform->grid_size-1));
+		int y = floor(linear_g * (transform->grid_size-1));
+		int z = floor(linear_b * (transform->grid_size-1));
+		int x_n = ceil(linear_r * (transform->grid_size-1));
+		int y_n = ceil(linear_g * (transform->grid_size-1));
+		int z_n = ceil(linear_b * (transform->grid_size-1));
+		float rx = linear_r * (transform->grid_size-1) - x;
+		float ry = linear_g * (transform->grid_size-1) - y;
+		float rz = linear_b * (transform->grid_size-1) - z;
+
+		c0_r = CLU(r_table, x, y, z);
+		c0_g = CLU(g_table, x, y, z);
+		c0_b = CLU(b_table, x, y, z);
+		if( rx >= ry ) {
+			if (ry >= rz) { //rx >= ry && ry >= rz
+				c1_r = CLU(r_table, x_n, y, z) - c0_r;
+				c2_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x_n, y, z);
+				c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
+				c1_g = CLU(g_table, x_n, y, z) - c0_g;
+				c2_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x_n, y, z);
+				c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
+				c1_b = CLU(b_table, x_n, y, z) - c0_b;
+				c2_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x_n, y, z);
+				c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
+			} else {
+				if (rx >= rz) { //rx >= rz && rz >= ry
+					c1_r = CLU(r_table, x_n, y, z) - c0_r;
+					c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
+					c3_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x_n, y, z);
+					c1_g = CLU(g_table, x_n, y, z) - c0_g;
+					c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
+					c3_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x_n, y, z);
+					c1_b = CLU(b_table, x_n, y, z) - c0_b;
+					c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
+					c3_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x_n, y, z);
+				} else { //rz > rx && rx >= ry
+					c1_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x, y, z_n);
+					c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
+					c3_r = CLU(r_table, x, y, z_n) - c0_r;
+					c1_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x, y, z_n);
+					c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
+					c3_g = CLU(g_table, x, y, z_n) - c0_g;
+					c1_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x, y, z_n);
+					c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
+					c3_b = CLU(b_table, x, y, z_n) - c0_b;
+				}
+			}
+		} else {
+			if (rx >= rz) { //ry > rx && rx >= rz
+				c1_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x, y_n, z);
+				c2_r = CLU(r_table, x_n, y_n, z) - c0_r;
+				c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
+				c1_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x, y_n, z);
+				c2_g = CLU(g_table, x_n, y_n, z) - c0_g;
+				c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
+				c1_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x, y_n, z);
+				c2_b = CLU(b_table, x_n, y_n, z) - c0_b;
+				c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
+			} else {
+				if (ry >= rz) { //ry >= rz && rz > rx 
+					c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
+					c2_r = CLU(r_table, x, y_n, z) - c0_r;
+					c3_r = CLU(r_table, x, y_n, z_n) - CLU(r_table, x, y_n, z);
+					c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
+					c2_g = CLU(g_table, x, y_n, z) - c0_g;
+					c3_g = CLU(g_table, x, y_n, z_n) - CLU(g_table, x, y_n, z);
+					c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
+					c2_b = CLU(b_table, x, y_n, z) - c0_b;
+					c3_b = CLU(b_table, x, y_n, z_n) - CLU(b_table, x, y_n, z);
+				} else { //rz > ry && ry > rx
+					c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
+					c2_r = CLU(r_table, x, y_n, z) - c0_r;
+					c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
+					c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
+					c2_g = CLU(g_table, x, y_n, z) - c0_g;
+					c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
+					c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
+					c2_b = CLU(b_table, x, y_n, z) - c0_b;
+					c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
+				}
+			}
+		}
+
+		clut_r = c0_r + c1_r*rx + c2_r*ry + c3_r*rz;
+		clut_g = c0_g + c1_g*rx + c2_g*ry + c3_g*rz;
+		clut_b = c0_b + c1_b*rx + c2_b*ry + c3_b*rz;
+
+		pcs_r = lut_interp_linear_float(clut_r,
+				transform->output_clut_table_r, transform->output_clut_table_length);
+		pcs_g = lut_interp_linear_float(clut_g,
+				transform->output_clut_table_g, transform->output_clut_table_length);
+		pcs_b = lut_interp_linear_float(clut_b,
+				transform->output_clut_table_b, transform->output_clut_table_length);
+		*dest++ = clamp_float(pcs_r);
+		*dest++ = clamp_float(pcs_g);
+		*dest++ = clamp_float(pcs_b);
+	}
+}
+*/
+
+static void qcms_transform_module_gamma_table(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	float out_r, out_g, out_b;
+	for (i = 0; i < length; i++) {
+		float in_r = *src++;
+		float in_g = *src++;
+		float in_b = *src++;
+
+		out_r = lut_interp_linear_float(in_r, transform->input_clut_table_r, 256);
+		out_g = lut_interp_linear_float(in_g, transform->input_clut_table_g, 256);
+		out_b = lut_interp_linear_float(in_b, transform->input_clut_table_b, 256);
+
+		*dest++ = clamp_float(out_r);
+		*dest++ = clamp_float(out_g);
+		*dest++ = clamp_float(out_b);
+	}
+}
+
+static void qcms_transform_module_gamma_lut(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	float out_r, out_g, out_b;
+	for (i = 0; i < length; i++) {
+		float in_r = *src++;
+		float in_g = *src++;
+		float in_b = *src++;
+
+		out_r = lut_interp_linear(in_r,
+				transform->output_gamma_lut_r, transform->output_gamma_lut_r_length);
+		out_g = lut_interp_linear(in_g,
+				transform->output_gamma_lut_g, transform->output_gamma_lut_g_length);
+		out_b = lut_interp_linear(in_b,
+				transform->output_gamma_lut_b, transform->output_gamma_lut_b_length);
+
+		*dest++ = clamp_float(out_r);
+		*dest++ = clamp_float(out_g);
+		*dest++ = clamp_float(out_b);
+	}
+}
+
+static void qcms_transform_module_matrix_translate(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	struct matrix mat;
+
+	/* store the results in column major mode
+	 * this makes doing the multiplication with sse easier */
+	mat.m[0][0] = transform->matrix.m[0][0];
+	mat.m[1][0] = transform->matrix.m[0][1];
+	mat.m[2][0] = transform->matrix.m[0][2];
+	mat.m[0][1] = transform->matrix.m[1][0];
+	mat.m[1][1] = transform->matrix.m[1][1];
+	mat.m[2][1] = transform->matrix.m[1][2];
+	mat.m[0][2] = transform->matrix.m[2][0];
+	mat.m[1][2] = transform->matrix.m[2][1];
+	mat.m[2][2] = transform->matrix.m[2][2];
+
+	for (i = 0; i < length; i++) {
+		float in_r = *src++;
+		float in_g = *src++;
+		float in_b = *src++;
+
+		float out_r = mat.m[0][0]*in_r + mat.m[1][0]*in_g + mat.m[2][0]*in_b + transform->tx;
+		float out_g = mat.m[0][1]*in_r + mat.m[1][1]*in_g + mat.m[2][1]*in_b + transform->ty;
+		float out_b = mat.m[0][2]*in_r + mat.m[1][2]*in_g + mat.m[2][2]*in_b + transform->tz;
+
+		*dest++ = clamp_float(out_r);
+		*dest++ = clamp_float(out_g);
+		*dest++ = clamp_float(out_b);
+	}
+}
+
+static void qcms_transform_module_matrix(struct qcms_modular_transform *transform, float *src, float *dest, size_t length)
+{
+	size_t i;
+	struct matrix mat;
+
+	/* store the results in column major mode
+	 * this makes doing the multiplication with sse easier */
+	mat.m[0][0] = transform->matrix.m[0][0];
+	mat.m[1][0] = transform->matrix.m[0][1];
+	mat.m[2][0] = transform->matrix.m[0][2];
+	mat.m[0][1] = transform->matrix.m[1][0];
+	mat.m[1][1] = transform->matrix.m[1][1];
+	mat.m[2][1] = transform->matrix.m[1][2];
+	mat.m[0][2] = transform->matrix.m[2][0];
+	mat.m[1][2] = transform->matrix.m[2][1];
+	mat.m[2][2] = transform->matrix.m[2][2];
+
+	for (i = 0; i < length; i++) {
+		float in_r = *src++;
+		float in_g = *src++;
+		float in_b = *src++;
+
+		float out_r = mat.m[0][0]*in_r + mat.m[1][0]*in_g + mat.m[2][0]*in_b;
+		float out_g = mat.m[0][1]*in_r + mat.m[1][1]*in_g + mat.m[2][1]*in_b;
+		float out_b = mat.m[0][2]*in_r + mat.m[1][2]*in_g + mat.m[2][2]*in_b;
+
+		*dest++ = clamp_float(out_r);
+		*dest++ = clamp_float(out_g);
+		*dest++ = clamp_float(out_b);
+	}
+}
+
+static struct qcms_modular_transform* qcms_modular_transform_alloc() {
+	return calloc(1, sizeof(struct qcms_modular_transform));
+}
+
+static void qcms_modular_transform_release(struct qcms_modular_transform *transform)
+{
+	struct qcms_modular_transform *next_transform;
+	while (transform != NULL) {
+		next_transform = transform->next_transform;
+		// clut may use a single block of memory.
+		// Perhaps we should remove this to simply the code.
+		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) {
+			if (transform->input_clut_table_r) free(transform->input_clut_table_r);
+		} else {
+			if (transform->input_clut_table_r) free(transform->input_clut_table_r);
+			if (transform->input_clut_table_g) free(transform->input_clut_table_g);
+			if (transform->input_clut_table_b) free(transform->input_clut_table_b);
+		}
+		if (transform->r_clut + 1 == transform->g_clut && transform->g_clut + 1 == transform->b_clut) {
+			if (transform->r_clut) free(transform->r_clut);
+		} else {
+			if (transform->r_clut) free(transform->r_clut);
+			if (transform->g_clut) free(transform->g_clut);
+			if (transform->b_clut) free(transform->b_clut);
+		}
+		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) {
+			if (transform->output_clut_table_r) free(transform->output_clut_table_r);
+		} else {
+			if (transform->output_clut_table_r) free(transform->output_clut_table_r);
+			if (transform->output_clut_table_g) free(transform->output_clut_table_g);
+			if (transform->output_clut_table_b) free(transform->output_clut_table_b);
+		}
+		if (transform->output_gamma_lut_r) free(transform->output_gamma_lut_r);
+		if (transform->output_gamma_lut_g) free(transform->output_gamma_lut_g);
+		if (transform->output_gamma_lut_b) free(transform->output_gamma_lut_b);
+		free(transform);
+		transform = next_transform;
+	}
+}
+
+/* Set transform to be the next element in the linked list. */
+static void append_transform(struct qcms_modular_transform *transform, struct qcms_modular_transform ***next_transform)
+{
+	**next_transform = transform;
+	while (transform) {
+		*next_transform = &(transform->next_transform);
+		transform = transform->next_transform;
+	}
+}
+
+/* reverse the transformation list (used by mBA) */
+static struct qcms_modular_transform* reverse_transform(struct qcms_modular_transform *transform) 
+{
+	struct qcms_modular_transform *prev_transform = NULL;
+	while (transform != NULL) {
+		struct qcms_modular_transform *next_transform = transform->next_transform;
+		transform->next_transform = prev_transform;
+		prev_transform = transform;
+		transform = next_transform;
+	}
+	
+	return prev_transform;
+}
+
+#define EMPTY_TRANSFORM_LIST NULL
+static struct qcms_modular_transform* qcms_modular_transform_create_mAB(struct lutmABType *lut)
+{
+	struct qcms_modular_transform *first_transform = NULL;
+	struct qcms_modular_transform **next_transform = &first_transform;
+	struct qcms_modular_transform *transform = NULL;
+
+	if (lut->a_curves[0] != NULL) {
+		size_t clut_length;
+		float *clut;
+
+		// If the A curve is present this also implies the 
+		// presence of a CLUT.
+		if (!lut->clut_table) 
+			goto fail;
+
+		// Prepare A curve.
+		transform = qcms_modular_transform_alloc();
+		if (!transform)
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->input_clut_table_r = build_input_gamma_table(lut->a_curves[0]);
+		transform->input_clut_table_g = build_input_gamma_table(lut->a_curves[1]);
+		transform->input_clut_table_b = build_input_gamma_table(lut->a_curves[2]);
+		transform->transform_module_fn = qcms_transform_module_gamma_table;
+		if (lut->num_grid_points[0] != lut->num_grid_points[1] ||
+			lut->num_grid_points[1] != lut->num_grid_points[2] ) {
+			//XXX: We don't currently support clut that are not squared!
+			goto fail;
+		}
+
+		// Prepare CLUT
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		clut_length = sizeof(float)*pow(lut->num_grid_points[0], 3)*3;
+		clut = malloc(clut_length);
+		if (!clut)
+			goto fail;
+		memcpy(clut, lut->clut_table, clut_length);
+		transform->r_clut = clut + 0;
+		transform->g_clut = clut + 1;
+		transform->b_clut = clut + 2;
+		transform->grid_size = lut->num_grid_points[0];
+		transform->transform_module_fn = qcms_transform_module_clut_only;
+	}
+	if (lut->m_curves[0] != NULL) {
+		// M curve imples the presence of a Matrix
+
+		// Prepare M curve
+		transform = qcms_modular_transform_alloc();
+		if (!transform)
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->input_clut_table_r = build_input_gamma_table(lut->m_curves[0]);
+		transform->input_clut_table_g = build_input_gamma_table(lut->m_curves[1]);
+		transform->input_clut_table_b = build_input_gamma_table(lut->m_curves[2]);
+		transform->transform_module_fn = qcms_transform_module_gamma_table;
+
+		// Prepare Matrix
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->matrix = build_mAB_matrix(lut);
+		if (transform->matrix.invalid)
+			goto fail;
+		transform->tx = s15Fixed16Number_to_float(lut->e03);
+		transform->ty = s15Fixed16Number_to_float(lut->e13);
+		transform->tz = s15Fixed16Number_to_float(lut->e23);
+		transform->transform_module_fn = qcms_transform_module_matrix_translate;
+	}
+	if (lut->b_curves[0] != NULL) {
+		// Prepare B curve
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->input_clut_table_r = build_input_gamma_table(lut->b_curves[0]);
+		transform->input_clut_table_g = build_input_gamma_table(lut->b_curves[1]);
+		transform->input_clut_table_b = build_input_gamma_table(lut->b_curves[2]);
+		transform->transform_module_fn = qcms_transform_module_gamma_table;
+	} else {
+		// B curve is mandatory
+		goto fail;
+	}
+
+	if (lut->reversed) {
+		// mBA are identical to mAB except that the transformation order
+		// is reversed
+		first_transform = reverse_transform(first_transform);
+	}
+
+	return first_transform;
+fail:
+	qcms_modular_transform_release(first_transform);
+	return NULL;
+}
+
+static struct qcms_modular_transform* qcms_modular_transform_create_lut(struct lutType *lut)
+{
+	struct qcms_modular_transform *first_transform = NULL;
+	struct qcms_modular_transform **next_transform = &first_transform;
+	struct qcms_modular_transform *transform = NULL;
+
+	size_t in_curve_len, clut_length, out_curve_len;
+	float *in_curves, *clut, *out_curves;
+
+	// Prepare Matrix
+	transform = qcms_modular_transform_alloc();
+	if (!transform) 
+		goto fail;
+	append_transform(transform, &next_transform);
+	transform->matrix = build_lut_matrix(lut);
+	if (transform->matrix.invalid)
+		goto fail;
+	transform->transform_module_fn = qcms_transform_module_matrix;
+
+	// Prepare input curves
+	transform = qcms_modular_transform_alloc();
+	if (!transform) 
+		goto fail;
+	append_transform(transform, &next_transform);
+	in_curve_len = sizeof(float)*lut->num_input_table_entries * 3;
+	in_curves = malloc(in_curve_len);
+	if (!in_curves) 
+		goto fail;
+	memcpy(in_curves, lut->input_table, in_curve_len);
+	transform->input_clut_table_r = in_curves + lut->num_input_table_entries * 0;
+	transform->input_clut_table_g = in_curves + lut->num_input_table_entries * 1;
+	transform->input_clut_table_b = in_curves + lut->num_input_table_entries * 2;
+	transform->input_clut_table_length = lut->num_input_table_entries;
+
+	// Prepare table
+	clut_length = sizeof(float)*pow(lut->num_clut_grid_points, 3)*3;
+	clut = malloc(clut_length);
+	if (!clut) 
+		goto fail;
+	memcpy(clut, lut->clut_table, clut_length);
+	transform->r_clut = clut + 0;
+	transform->g_clut = clut + 1;
+	transform->b_clut = clut + 2;
+	transform->grid_size = lut->num_clut_grid_points;
+
+	// Prepare output curves
+	out_curve_len = sizeof(float) * lut->num_output_table_entries * 3;
+	out_curves = malloc(out_curve_len);
+	if (!out_curves) 
+		goto fail;
+	memcpy(out_curves, lut->output_table, out_curve_len);
+	transform->output_clut_table_r = out_curves + lut->num_output_table_entries * 0;
+	transform->output_clut_table_g = out_curves + lut->num_output_table_entries * 1;
+	transform->output_clut_table_b = out_curves + lut->num_output_table_entries * 2;
+	transform->output_clut_table_length = lut->num_output_table_entries;
+	transform->transform_module_fn = qcms_transform_module_clut;
+
+	return first_transform;
+fail:
+	qcms_modular_transform_release(first_transform);
+	return NULL;
+}
+
+struct qcms_modular_transform* qcms_modular_transform_create_input(qcms_profile *in)
+{
+	struct qcms_modular_transform *first_transform = NULL;
+	struct qcms_modular_transform **next_transform = &first_transform;
+
+	if (in->A2B0) {
+		struct qcms_modular_transform *lut_transform;
+		lut_transform = qcms_modular_transform_create_lut(in->A2B0);
+		if (!lut_transform)
+			goto fail;
+		append_transform(lut_transform, &next_transform);
+	} else if (in->mAB && in->mAB->num_in_channels == 3 && in->mAB->num_out_channels == 3) {
+		struct qcms_modular_transform *mAB_transform;
+		mAB_transform = qcms_modular_transform_create_mAB(in->mAB);
+		if (!mAB_transform)
+			goto fail;
+		append_transform(mAB_transform, &next_transform);
+
+	} else {
+		struct qcms_modular_transform *transform;
+
+		transform = qcms_modular_transform_alloc();
+		if (!transform)
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->input_clut_table_r = build_input_gamma_table(in->redTRC);
+		transform->input_clut_table_g = build_input_gamma_table(in->greenTRC);
+		transform->input_clut_table_b = build_input_gamma_table(in->blueTRC);
+		transform->transform_module_fn = qcms_transform_module_gamma_table;
+		if (!transform->input_clut_table_r || !transform->input_clut_table_g ||
+				!transform->input_clut_table_b) {
+			goto fail;
+		}
+
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->matrix.m[0][0] = 1/1.999969482421875f;
+		transform->matrix.m[0][1] = 0.f;
+		transform->matrix.m[0][2] = 0.f;
+		transform->matrix.m[1][0] = 0.f;
+		transform->matrix.m[1][1] = 1/1.999969482421875f;
+		transform->matrix.m[1][2] = 0.f;
+		transform->matrix.m[2][0] = 0.f;
+		transform->matrix.m[2][1] = 0.f;
+		transform->matrix.m[2][2] = 1/1.999969482421875f;
+		transform->matrix.invalid = false;
+		transform->transform_module_fn = qcms_transform_module_matrix;
+
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->matrix = build_colorant_matrix(in);
+		transform->transform_module_fn = qcms_transform_module_matrix;
+	}
+
+	return first_transform;
+fail:
+	qcms_modular_transform_release(first_transform);
+	return EMPTY_TRANSFORM_LIST;
+}
+static struct qcms_modular_transform* qcms_modular_transform_create_output(qcms_profile *out)
+{
+	struct qcms_modular_transform *first_transform = NULL;
+	struct qcms_modular_transform **next_transform = &first_transform;
+
+	if (out->B2A0) {
+		struct qcms_modular_transform *lut_transform;
+		lut_transform = qcms_modular_transform_create_lut(out->B2A0);
+		if (!lut_transform) 
+			goto fail;
+		append_transform(lut_transform, &next_transform);
+	} else if (out->mBA && out->mBA->num_in_channels == 3 && out->mBA->num_out_channels == 3) {
+		struct qcms_modular_transform *lut_transform;
+		lut_transform = qcms_modular_transform_create_mAB(out->mBA);
+		if (!lut_transform) 
+			goto fail;
+		append_transform(lut_transform, &next_transform);
+	} else if (out->redTRC && out->greenTRC && out->blueTRC) {
+		struct qcms_modular_transform *transform;
+
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->matrix = matrix_invert(build_colorant_matrix(out));
+		transform->transform_module_fn = qcms_transform_module_matrix;
+
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		transform->matrix.m[0][0] = 1.999969482421875f;
+		transform->matrix.m[0][1] = 0.f;
+		transform->matrix.m[0][2] = 0.f;
+		transform->matrix.m[1][0] = 0.f;
+		transform->matrix.m[1][1] = 1.999969482421875f;
+		transform->matrix.m[1][2] = 0.f;
+		transform->matrix.m[2][0] = 0.f;
+		transform->matrix.m[2][1] = 0.f;
+		transform->matrix.m[2][2] = 1.999969482421875f;
+		transform->matrix.invalid = false;
+		transform->transform_module_fn = qcms_transform_module_matrix;
+
+		transform = qcms_modular_transform_alloc();
+		if (!transform) 
+			goto fail;
+		append_transform(transform, &next_transform);
+		build_output_lut(out->redTRC, &transform->output_gamma_lut_r,
+			&transform->output_gamma_lut_r_length);
+		build_output_lut(out->greenTRC, &transform->output_gamma_lut_g,
+			&transform->output_gamma_lut_g_length);
+		build_output_lut(out->blueTRC, &transform->output_gamma_lut_b,
+			&transform->output_gamma_lut_b_length);
+		transform->transform_module_fn = qcms_transform_module_gamma_lut;
+
+		if (!transform->output_gamma_lut_r || !transform->output_gamma_lut_g ||
+				!transform->output_gamma_lut_b) {
+			goto fail;
+		}
+	} else {
+		assert(0 && "Unsupported output profile workflow.");
+		return NULL;
+	}
+
+	return first_transform;
+fail:
+	qcms_modular_transform_release(first_transform);
+	return EMPTY_TRANSFORM_LIST;
+}
+
+/* Not Completed
+// Simply the transformation chain an equivilent transformation chain
+static struct qcms_modular_transform* qcms_modular_transform_reduce(struct qcms_modular_transform *transform)
+{
+	struct qcms_modular_transform *first_transform = NULL;
+	struct qcms_modular_transform *curr_trans = transform;
+	struct qcms_modular_transform *prev_trans = NULL;
+	while (curr_trans) {
+		struct qcms_modular_transform *next_trans = curr_trans->next_transform;
+		if (curr_trans->transform_module_fn == qcms_transform_module_matrix) {
+			if (next_trans && next_trans->transform_module_fn == qcms_transform_module_matrix) {
+				curr_trans->matrix = matrix_multiply(curr_trans->matrix, next_trans->matrix);
+				goto remove_next;	
+			}
+		}
+		if (curr_trans->transform_module_fn == qcms_transform_module_gamma_table) {
+			bool isLinear = true;
+			uint16_t i;
+			for (i = 0; isLinear && i < 256; i++) {
+				isLinear &= (int)(curr_trans->input_clut_table_r[i] * 255) == i;
+				isLinear &= (int)(curr_trans->input_clut_table_g[i] * 255) == i;
+				isLinear &= (int)(curr_trans->input_clut_table_b[i] * 255) == i;
+			}
+			goto remove_current;
+		}
+		
+next_transform:
+		if (!next_trans) break;
+		prev_trans = curr_trans;
+		curr_trans = next_trans;
+		continue;
+remove_current:
+		if (curr_trans == transform) {
+			//Update head
+			transform = next_trans;
+		} else {
+			prev_trans->next_transform = next_trans;
+		}
+		curr_trans->next_transform = NULL;
+		qcms_modular_transform_release(curr_trans);
+		//return transform;
+		return qcms_modular_transform_reduce(transform);
+remove_next:
+		curr_trans->next_transform = next_trans->next_transform;
+		next_trans->next_transform = NULL;
+		qcms_modular_transform_release(next_trans);
+		continue;
+	}
+	return transform;
+}
+*/
+
+static struct qcms_modular_transform* qcms_modular_transform_create(qcms_profile *in, qcms_profile *out)
+{
+	struct qcms_modular_transform *first_transform = NULL;
+	struct qcms_modular_transform **next_transform = &first_transform;
+
+	if (in->color_space == RGB_SIGNATURE) {
+		struct qcms_modular_transform* rgb_to_xyz;
+		rgb_to_xyz = qcms_modular_transform_create_input(in);
+		if (!rgb_to_xyz) 
+			goto fail;
+		append_transform(rgb_to_xyz, &next_transform);
+	} else {
+		assert(0 && "input color space not supported");
+		goto fail;
+	}
+
+	if (in->pcs == LAB_SIGNATURE && out->pcs == XYZ_SIGNATURE) {
+		struct qcms_modular_transform* lab_to_xyz;
+		lab_to_xyz = qcms_modular_transform_alloc();
+		if (!lab_to_xyz) 
+			goto fail;
+		append_transform(lab_to_xyz, &next_transform);
+		lab_to_xyz->transform_module_fn = qcms_transform_module_LAB_to_XYZ;
+	}
+
+	//if (in->chromaticAdaption.invalid == false) {
+	//	struct qcms_modular_transform* chromaticAdaption;
+	//	chromaticAdaption = qcms_modular_transform_alloc();
+	//	if (!chromaticAdaption) 
+	//		goto fail;
+	//	append_transform(chromaticAdaption, &next_transform);
+	//	chromaticAdaption->matrix = matrix_invert(in->chromaticAdaption);
+	//	chromaticAdaption->transform_module_fn = qcms_transform_module_matrix;
+	//}
+
+        if (in->pcs == XYZ_SIGNATURE && out->pcs == LAB_SIGNATURE) {
+		struct qcms_modular_transform* xyz_to_lab;
+		xyz_to_lab = qcms_modular_transform_alloc();
+		if (!xyz_to_lab) 
+			goto fail;
+		append_transform(xyz_to_lab, &next_transform);
+		xyz_to_lab->transform_module_fn = qcms_transform_module_XYZ_to_LAB;
+	}
+
+	if (out->color_space == RGB_SIGNATURE) {
+		struct qcms_modular_transform* xyz_to_rgb;
+		xyz_to_rgb = qcms_modular_transform_create_output(out);
+		if (!xyz_to_rgb) 
+			goto fail;
+		append_transform(xyz_to_rgb, &next_transform);
+	} else {
+		assert(0 && "output color space not supported");
+		goto fail;
+	}
+	// Not Completed
+	//return qcms_modular_transform_reduce(first_transform);
+	return first_transform;
+fail:
+	qcms_modular_transform_release(first_transform);
+	return EMPTY_TRANSFORM_LIST;
+}
+
+static float* qcms_modular_transform_data(struct qcms_modular_transform *transform, float *src, float *dest, size_t len)
+{
+        while (transform != NULL) {
+                // Keep swaping src/dest when performing a transform to use less memory.
+                float *new_src = dest;
+		const void *transform_fn = transform->transform_module_fn;
+		if (transform_fn != qcms_transform_module_gamma_table &&
+		    transform_fn != qcms_transform_module_gamma_lut &&
+		    transform_fn != qcms_transform_module_clut &&
+		    transform_fn != qcms_transform_module_clut_only &&
+		    transform_fn != qcms_transform_module_matrix &&
+		    transform_fn != qcms_transform_module_matrix_translate &&
+		    transform_fn != qcms_transform_module_LAB_to_XYZ &&
+		    transform_fn != qcms_transform_module_XYZ_to_LAB) {
+			assert(0 && "Unsupported transform module");
+			return NULL;
+		}
+                transform->transform_module_fn(transform,src,dest,len);
+                dest = src;
+                src = new_src;
+                transform = transform->next_transform;
+        }
+        // The results end up in the src buffer because of the switching
+        return src;
+}
+
+float* qcms_chain_transform(qcms_profile *in, qcms_profile *out, float *src, float *dest, size_t lutSize)
+{
+	struct qcms_modular_transform *transform_list = qcms_modular_transform_create(in, out);
+	if (transform_list != NULL) {
+		float *lut = qcms_modular_transform_data(transform_list, src, dest, lutSize/3);
+		qcms_modular_transform_release(transform_list);
+		return lut;
+	}
+	return NULL;
+}
diff --git a/chain.h b/chain.h
new file mode 100644
index 0000000..bdc6c88
--- /dev/null
+++ b/chain.h
@@ -0,0 +1,30 @@
+/* vim: set ts=8 sw=8 noexpandtab: */
+//  qcms
+//  Copyright (C) 2009 Mozilla Foundation
+//  Copyright (C) 1998-2007 Marti Maria
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+#ifndef _QCMS_CHAIN_H
+#define _QCMS_CHAIN_H
+
+// Generates and returns a 3D LUT with lutSize^3 samples using the provided src/dest.
+float* qcms_chain_transform(qcms_profile *in, qcms_profile *out, float *src, float *dest, size_t lutSize);
+
+#endif
diff --git a/matrix.c b/matrix.c
new file mode 100644
index 0000000..0ce5bd6
--- /dev/null
+++ b/matrix.c
@@ -0,0 +1,136 @@
+/* vim: set ts=8 sw=8 noexpandtab: */
+//  qcms
+//  Copyright (C) 2009 Mozilla Foundation
+//  Copyright (C) 1998-2007 Marti Maria
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+#include <stdlib.h>
+#include "qcmsint.h"
+#include "matrix.h"
+
+struct vector matrix_eval(struct matrix mat, struct vector v)
+{
+	struct vector result;
+	result.v[0] = mat.m[0][0]*v.v[0] + mat.m[0][1]*v.v[1] + mat.m[0][2]*v.v[2];
+	result.v[1] = mat.m[1][0]*v.v[0] + mat.m[1][1]*v.v[1] + mat.m[1][2]*v.v[2];
+	result.v[2] = mat.m[2][0]*v.v[0] + mat.m[2][1]*v.v[1] + mat.m[2][2]*v.v[2];
+	return result;
+}
+
+//XXX: should probably pass by reference and we could
+//probably reuse this computation in matrix_invert
+float matrix_det(struct matrix mat)
+{
+	float det;
+	det = mat.m[0][0]*mat.m[1][1]*mat.m[2][2] +
+		mat.m[0][1]*mat.m[1][2]*mat.m[2][0] +
+		mat.m[0][2]*mat.m[1][0]*mat.m[2][1] -
+		mat.m[0][0]*mat.m[1][2]*mat.m[2][1] -
+		mat.m[0][1]*mat.m[1][0]*mat.m[2][2] -
+		mat.m[0][2]*mat.m[1][1]*mat.m[2][0];
+	return det;
+}
+
+/* from pixman and cairo and Mathematics for Game Programmers */
+/* lcms uses gauss-jordan elimination with partial pivoting which is
+ * less efficient and not as numerically stable. See Mathematics for
+ * Game Programmers. */
+struct matrix matrix_invert(struct matrix mat)
+{
+	struct matrix dest_mat;
+	int i,j;
+	static int a[3] = { 2, 2, 1 };
+	static int b[3] = { 1, 0, 0 };
+
+	/* inv  (A) = 1/det (A) * adj (A) */
+	float det = matrix_det(mat);
+
+	if (det == 0) {
+		dest_mat.invalid = true;
+	} else {
+		dest_mat.invalid = false;
+	}
+
+	det = 1/det;
+
+	for (j = 0; j < 3; j++) {
+		for (i = 0; i < 3; i++) {
+			double p;
+			int ai = a[i];
+			int aj = a[j];
+			int bi = b[i];
+			int bj = b[j];
+
+			p = mat.m[ai][aj] * mat.m[bi][bj] -
+				mat.m[ai][bj] * mat.m[bi][aj];
+			if (((i + j) & 1) != 0)
+				p = -p;
+
+			dest_mat.m[j][i] = det * p;
+		}
+	}
+	return dest_mat;
+}
+
+struct matrix matrix_identity(void)
+{
+	struct matrix i;
+	i.m[0][0] = 1;
+	i.m[0][1] = 0;
+	i.m[0][2] = 0;
+	i.m[1][0] = 0;
+	i.m[1][1] = 1;
+	i.m[1][2] = 0;
+	i.m[2][0] = 0;
+	i.m[2][1] = 0;
+	i.m[2][2] = 1;
+	i.invalid = false;
+	return i;
+}
+
+struct matrix matrix_invalid(void)
+{
+	struct matrix inv = matrix_identity();
+	inv.invalid = true;
+	return inv;
+}
+
+
+/* from pixman */
+/* MAT3per... */
+struct matrix matrix_multiply(struct matrix a, struct matrix b)
+{
+	struct matrix result;
+	int dx, dy;
+	int o;
+	for (dy = 0; dy < 3; dy++) {
+		for (dx = 0; dx < 3; dx++) {
+			double v = 0;
+			for (o = 0; o < 3; o++) {
+				v += a.m[dy][o] * b.m[o][dx];
+			}
+			result.m[dy][dx] = v;
+		}
+	}
+	result.invalid = a.invalid || b.invalid;
+	return result;
+}
+
+
diff --git a/matrix.h b/matrix.h
new file mode 100644
index 0000000..5011988
--- /dev/null
+++ b/matrix.h
@@ -0,0 +1,39 @@
+/* vim: set ts=8 sw=8 noexpandtab: */
+//  qcms
+//  Copyright (C) 2009 Mozilla Foundation
+//  Copyright (C) 1998-2007 Marti Maria
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+#ifndef _QCMS_MATRIX_H
+#define _QCMS_MATRIX_H
+
+struct vector {
+        float v[3];
+};
+
+struct vector matrix_eval(struct matrix mat, struct vector v);
+float matrix_det(struct matrix mat);
+struct matrix matrix_identity(void);
+struct matrix matrix_multiply(struct matrix a, struct matrix b);
+struct matrix matrix_invert(struct matrix mat);
+
+struct matrix matrix_invalid(void);
+
+#endif
diff --git a/transform_util.c b/transform_util.c
new file mode 100644
index 0000000..c29a98a
--- /dev/null
+++ b/transform_util.c
@@ -0,0 +1,534 @@
+#include <math.h>
+#include <assert.h>
+#include <string.h> //memcpy
+#include "qcmsint.h"
+#include "transform_util.h"
+#include "matrix.h"
+
+#define PARAMETRIC_CURVE_TYPE 0x70617261 //'para'
+
+/* value must be a value between 0 and 1 */
+//XXX: is the above a good restriction to have?
+float lut_interp_linear(double value, uint16_t *table, int length)
+{
+	int upper, lower;
+	value = value * (length - 1); // scale to length of the array
+	upper = ceil(value);
+	lower = floor(value);
+	//XXX: can we be more performant here?
+	value = table[upper]*(1. - (upper - value)) + table[lower]*(upper - value);
+	/* scale the value */
+	return value * (1./65535.);
+}
+
+/* same as above but takes and returns a uint16_t value representing a range from 0..1 */
+uint16_t lut_interp_linear16(uint16_t input_value, uint16_t *table, int length)
+{
+	/* Start scaling input_value to the length of the array: 65535*(length-1).
+	 * We'll divide out the 65535 next */
+	uint32_t value = (input_value * (length - 1));
+	uint32_t upper = (value + 65534) / 65535; /* equivalent to ceil(value/65535) */
+	uint32_t lower = value / 65535;           /* equivalent to floor(value/65535) */
+	/* interp is the distance from upper to value scaled to 0..65535 */
+	uint32_t interp = value % 65535;
+
+	value = (table[upper]*(interp) + table[lower]*(65535 - interp))/65535; // 0..65535*65535
+
+	return value;
+}
+
+/* same as above but takes an input_value from 0..PRECACHE_OUTPUT_MAX
+ * and returns a uint8_t value representing a range from 0..1 */
+static
+uint8_t lut_interp_linear_precache_output(uint32_t input_value, uint16_t *table, int length)
+{
+	/* Start scaling input_value to the length of the array: PRECACHE_OUTPUT_MAX*(length-1).
+	 * We'll divide out the PRECACHE_OUTPUT_MAX next */
+	uint32_t value = (input_value * (length - 1));
+
+	/* equivalent to ceil(value/PRECACHE_OUTPUT_MAX) */
+	uint32_t upper = (value + PRECACHE_OUTPUT_MAX-1) / PRECACHE_OUTPUT_MAX;
+	/* equivalent to floor(value/PRECACHE_OUTPUT_MAX) */
+	uint32_t lower = value / PRECACHE_OUTPUT_MAX;
+	/* interp is the distance from upper to value scaled to 0..PRECACHE_OUTPUT_MAX */
+	uint32_t interp = value % PRECACHE_OUTPUT_MAX;
+
+	/* the table values range from 0..65535 */
+	value = (table[upper]*(interp) + table[lower]*(PRECACHE_OUTPUT_MAX - interp)); // 0..(65535*PRECACHE_OUTPUT_MAX)
+
+	/* round and scale */
+	value += (PRECACHE_OUTPUT_MAX*65535/255)/2;
+        value /= (PRECACHE_OUTPUT_MAX*65535/255); // scale to 0..255
+	return value;
+}
+
+/* value must be a value between 0 and 1 */
+//XXX: is the above a good restriction to have?
+float lut_interp_linear_float(float value, float *table, int length)
+{
+        int upper, lower;
+        value = value * (length - 1);
+        upper = ceil(value);
+        lower = floor(value);
+        //XXX: can we be more performant here?
+        value = table[upper]*(1. - (upper - value)) + table[lower]*(upper - value);
+        /* scale the value */
+        return value;
+}
+
+#if 0
+/* if we use a different representation i.e. one that goes from 0 to 0x1000 we can be more efficient
+ * because we can avoid the divisions and use a shifting instead */
+/* same as above but takes and returns a uint16_t value representing a range from 0..1 */
+uint16_t lut_interp_linear16(uint16_t input_value, uint16_t *table, int length)
+{
+	uint32_t value = (input_value * (length - 1));
+	uint32_t upper = (value + 4095) / 4096; /* equivalent to ceil(value/4096) */
+	uint32_t lower = value / 4096;           /* equivalent to floor(value/4096) */
+	uint32_t interp = value % 4096;
+
+	value = (table[upper]*(interp) + table[lower]*(4096 - interp))/4096; // 0..4096*4096
+
+	return value;
+}
+#endif
+
+void compute_curve_gamma_table_type1(float gamma_table[256], double gamma)
+{
+	unsigned int i;
+	for (i = 0; i < 256; i++) {
+		gamma_table[i] = pow(i/255., gamma);
+	}
+}
+
+void compute_curve_gamma_table_type2(float gamma_table[256], uint16_t *table, int length)
+{
+	unsigned int i;
+	for (i = 0; i < 256; i++) {
+		gamma_table[i] = lut_interp_linear(i/255., table, length);
+	}
+}
+
+void compute_curve_gamma_table_type_parametric(float gamma_table[256], float parameter[7], int count)
+{
+        size_t X;
+        float interval;
+        float a, b, c, e, f;
+        float y = parameter[0];
+        if (count == 0) {
+                a = 1;
+                b = 0;
+                c = 0;
+                e = 0;
+                f = 0;
+                interval = -INFINITY;
+        } else if(count == 1) {
+                a = parameter[1];
+                b = parameter[2];
+                c = 0;
+                e = 0;
+                f = 0;
+                interval = -1 * parameter[2] / parameter[1];
+        } else if(count == 2) {
+                a = parameter[1];
+                b = parameter[2];
+                c = 0;
+                e = parameter[3];
+                f = parameter[3];
+                interval = -1 * parameter[2] / parameter[1];
+        } else if(count == 3) {
+                a = parameter[1];
+                b = parameter[2];
+                c = parameter[3];
+                e = -c;
+                f = 0;
+                interval = parameter[4];
+        } else if(count == 4) {
+                a = parameter[1];
+                b = parameter[2];
+                c = parameter[3];
+                e = parameter[5] - c;
+                f = parameter[6];
+                interval = parameter[4];
+        } else {
+                assert(0 && "invalid parametric function type.");
+                a = 1;
+                b = 0;
+                c = 0;
+                e = 0;
+                f = 0;
+                interval = -INFINITY;
+        }       
+        for (X = 0; X < 256; X++) {
+                if (X >= interval) {
+                        // XXX The equations are not exactly as definied in the spec but are
+                        //     algebraic equivilent.
+                        // TODO Should division by 255 be for the whole expression.
+                        gamma_table[X] = pow(a * X / 255. + b, y) + c + e;
+                } else {
+                        gamma_table[X] = c * X / 255. + f;
+                }
+        }
+}
+
+void compute_curve_gamma_table_type0(float gamma_table[256])
+{
+	unsigned int i;
+	for (i = 0; i < 256; i++) {
+		gamma_table[i] = i/255.;
+	}
+}
+
+
+float clamp_float(float a)
+{
+        if (a > 1.)
+                return 1.;
+        else if (a < 0)
+                return 0;
+        else
+                return a;
+}
+
+unsigned char clamp_u8(float v)
+{
+	if (v > 255.)
+		return 255;
+	else if (v < 0)
+		return 0;
+	else
+		return floor(v+.5);
+}
+
+float u8Fixed8Number_to_float(uint16_t x)
+{
+	// 0x0000 = 0.
+	// 0x0100 = 1.
+	// 0xffff = 255  + 255/256
+	return x/256.;
+}
+
+float *build_input_gamma_table(struct curveType *TRC)
+{
+	float *gamma_table;
+
+	if (!TRC) return NULL;
+	gamma_table = malloc(sizeof(float)*256);
+	if (gamma_table) {
+		if (TRC->type == PARAMETRIC_CURVE_TYPE) {
+			compute_curve_gamma_table_type_parametric(gamma_table, TRC->parameter, TRC->count);
+		} else {
+			if (TRC->count == 0) {
+				compute_curve_gamma_table_type0(gamma_table);
+			} else if (TRC->count == 1) {
+				compute_curve_gamma_table_type1(gamma_table, u8Fixed8Number_to_float(TRC->data[0]));
+			} else {
+				compute_curve_gamma_table_type2(gamma_table, TRC->data, TRC->count);
+			}
+		}
+	}
+        return gamma_table;
+}
+
+struct matrix build_colorant_matrix(qcms_profile *p)
+{
+	struct matrix result;
+	result.m[0][0] = s15Fixed16Number_to_float(p->redColorant.X);
+	result.m[0][1] = s15Fixed16Number_to_float(p->greenColorant.X);
+	result.m[0][2] = s15Fixed16Number_to_float(p->blueColorant.X);
+	result.m[1][0] = s15Fixed16Number_to_float(p->redColorant.Y);
+	result.m[1][1] = s15Fixed16Number_to_float(p->greenColorant.Y);
+	result.m[1][2] = s15Fixed16Number_to_float(p->blueColorant.Y);
+	result.m[2][0] = s15Fixed16Number_to_float(p->redColorant.Z);
+	result.m[2][1] = s15Fixed16Number_to_float(p->greenColorant.Z);
+	result.m[2][2] = s15Fixed16Number_to_float(p->blueColorant.Z);
+	result.invalid = false;
+	return result;
+}
+
+/* The following code is copied nearly directly from lcms.
+ * I think it could be much better. For example, Argyll seems to have better code in
+ * icmTable_lookup_bwd and icmTable_setup_bwd. However, for now this is a quick way
+ * to a working solution and allows for easy comparing with lcms. */
+uint16_fract_t lut_inverse_interp16(uint16_t Value, uint16_t LutTable[], int length)
+{
+        int l = 1;
+        int r = 0x10000;
+        int x = 0, res;       // 'int' Give spacing for negative values
+        int NumZeroes, NumPoles;
+        int cell0, cell1;
+        double val2;
+        double y0, y1, x0, x1;
+        double a, b, f;
+
+        // July/27 2001 - Expanded to handle degenerated curves with an arbitrary
+        // number of elements containing 0 at the begining of the table (Zeroes)
+        // and another arbitrary number of poles (FFFFh) at the end.
+        // First the zero and pole extents are computed, then value is compared.
+
+        NumZeroes = 0;
+        while (LutTable[NumZeroes] == 0 && NumZeroes < length-1)
+                        NumZeroes++;
+
+        // There are no zeros at the beginning and we are trying to find a zero, so
+        // return anything. It seems zero would be the less destructive choice
+	/* I'm not sure that this makes sense, but oh well... */
+        if (NumZeroes == 0 && Value == 0)
+            return 0;
+
+        NumPoles = 0;
+        while (LutTable[length-1- NumPoles] == 0xFFFF && NumPoles < length-1)
+                        NumPoles++;
+
+        // Does the curve belong to this case?
+        if (NumZeroes > 1 || NumPoles > 1)
+        {               
+                int a, b;
+
+                // Identify if value fall downto 0 or FFFF zone             
+                if (Value == 0) return 0;
+               // if (Value == 0xFFFF) return 0xFFFF;
+
+                // else restrict to valid zone
+
+                a = ((NumZeroes-1) * 0xFFFF) / (length-1);               
+                b = ((length-1 - NumPoles) * 0xFFFF) / (length-1);
+                                                                
+                l = a - 1;
+                r = b + 1;
+        }
+
+
+        // Seems not a degenerated case... apply binary search
+
+        while (r > l) {
+
+                x = (l + r) / 2;
+
+		res = (int) lut_interp_linear16((uint16_fract_t) (x-1), LutTable, length);
+
+                if (res == Value) {
+
+                    // Found exact match. 
+                    
+                    return (uint16_fract_t) (x - 1);
+                }
+
+                if (res > Value) r = x - 1;
+                else l = x + 1;
+        }
+
+        // Not found, should we interpolate?
+
+                
+        // Get surrounding nodes
+        
+        val2 = (length-1) * ((double) (x - 1) / 65535.0);
+
+        cell0 = (int) floor(val2);
+        cell1 = (int) ceil(val2);
+           
+        if (cell0 == cell1) return (uint16_fract_t) x;
+
+        y0 = LutTable[cell0] ;
+        x0 = (65535.0 * cell0) / (length-1); 
+
+        y1 = LutTable[cell1] ;
+        x1 = (65535.0 * cell1) / (length-1);
+
+        a = (y1 - y0) / (x1 - x0);
+        b = y0 - a * x0;
+
+        if (fabs(a) < 0.01) return (uint16_fract_t) x;
+
+        f = ((Value - b) / a);
+
+        if (f < 0.0) return (uint16_fract_t) 0;
+        if (f >= 65535.0) return (uint16_fract_t) 0xFFFF;
+
+        return (uint16_fract_t) floor(f + 0.5);                        
+
+}
+
+/*
+ The number of entries needed to invert a lookup table should not
+ necessarily be the same as the original number of entries.  This is
+ especially true of lookup tables that have a small number of entries.
+
+ For example:
+ Using a table like:
+    {0, 3104, 14263, 34802, 65535}
+ invert_lut will produce an inverse of:
+    {3, 34459, 47529, 56801, 65535}
+ which has an maximum error of about 9855 (pixel difference of ~38.346)
+
+ For now, we punt the decision of output size to the caller. */
+static uint16_t *invert_lut(uint16_t *table, int length, int out_length)
+{
+        int i;
+        /* for now we invert the lut by creating a lut of size out_length
+         * and attempting to lookup a value for each entry using lut_inverse_interp16 */
+        uint16_t *output = malloc(sizeof(uint16_t)*out_length);
+        if (!output)
+                return NULL;
+
+        for (i = 0; i < out_length; i++) {
+                double x = ((double) i * 65535.) / (double) (out_length - 1);
+                uint16_fract_t input = floor(x + .5);
+                output[i] = lut_inverse_interp16(input, table, length);
+        }
+        return output;
+}
+
+static void compute_precache_pow(uint8_t *output, float gamma)
+{
+        uint32_t v = 0;
+        for (v = 0; v <= 0xffff; v++) {
+                //XXX: don't do integer/float conversion... and round?
+                output[v] = 255. * pow(v/65535., gamma);
+        }
+}
+
+void compute_precache_lut(uint8_t *output, uint16_t *table, int length)
+{
+        uint32_t v = 0;
+        for (v = 0; v <= 0xffff; v++) {
+                //XXX: don't do integer/float conversion... round?
+                output[v] = lut_interp_linear16(v, table, length) >> 8;
+        }
+}
+
+void compute_precache_linear(uint8_t *output)
+{
+        uint32_t v = 0;
+        for (v = 0; v <= 0xffff; v++) {
+                //XXX: round?
+                output[v] = v >> 8;
+        }
+}
+
+qcms_bool compute_precache(struct curveType *trc, uint8_t *output)
+{
+        
+        if (trc->type == PARAMETRIC_CURVE_TYPE) {
+                        float gamma_table[256];
+                        uint16_t gamma_table_uint[256];
+                        uint16_t i;
+                        uint16_t *inverted;
+                        int inverted_size = 256;
+
+                        compute_curve_gamma_table_type_parametric(gamma_table, trc->parameter, trc->count);
+                        for(i = 0; i < 256; i++) {
+                                gamma_table_uint[i] = (uint16_t)(gamma_table[i] * 65535);
+                        }
+
+                        //XXX: the choice of a minimum of 256 here is not backed by any theory, 
+                        //     measurement or data, howeve r it is what lcms uses.
+                        //     the maximum number we would need is 65535 because that's the 
+                        //     accuracy used for computing the pre cache table
+                        if (inverted_size < 256)
+                                inverted_size = 256;
+
+                        inverted = invert_lut(gamma_table_uint, 256, inverted_size);
+                        if (!inverted)
+                                return false;
+                        compute_precache_lut(output, inverted, inverted_size);
+                        free(inverted);
+        } else {
+                if (trc->count == 0) {
+                        compute_precache_linear(output);
+                } else if (trc->count == 1) {
+                        compute_precache_pow(output, 1./u8Fixed8Number_to_float(trc->data[0]));
+                } else {
+                        uint16_t *inverted;
+                        int inverted_size = trc->count;
+                        //XXX: the choice of a minimum of 256 here is not backed by any theory, 
+                        //     measurement or data, howeve r it is what lcms uses.
+                        //     the maximum number we would need is 65535 because that's the 
+                        //     accuracy used for computing the pre cache table
+                        if (inverted_size < 256)
+                                inverted_size = 256;
+
+                        inverted = invert_lut(trc->data, trc->count, inverted_size);
+                        if (!inverted)
+                                return false;
+                        compute_precache_lut(output, inverted, inverted_size);
+                        free(inverted);
+                }
+        }
+        return true;
+}
+
+
+static uint16_t *build_linear_table(int length)
+{
+        int i;
+        uint16_t *output = malloc(sizeof(uint16_t)*length);
+        if (!output)
+                return NULL;
+
+        for (i = 0; i < length; i++) {
+                double x = ((double) i * 65535.) / (double) (length - 1);
+                uint16_fract_t input = floor(x + .5);
+                output[i] = input;
+        }
+        return output;
+}
+
+static uint16_t *build_pow_table(float gamma, int length)
+{
+        int i;
+        uint16_t *output = malloc(sizeof(uint16_t)*length);
+        if (!output)
+                return NULL;
+
+        for (i = 0; i < length; i++) {
+                uint16_fract_t result;
+                double x = ((double) i) / (double) (length - 1);
+                x = pow(x, gamma);                //XXX turn this conversion into a function
+                result = floor(x*65535. + .5);
+                output[i] = result;
+        }
+        return output;
+}
+
+void build_output_lut(struct curveType *trc,
+                uint16_t **output_gamma_lut, size_t *output_gamma_lut_length)
+{
+        if (trc->type == PARAMETRIC_CURVE_TYPE) {
+                float gamma_table[256];
+                uint16_t i;
+                uint16_t *output = malloc(sizeof(uint16_t)*256);
+
+                if (!output) {
+                        *output_gamma_lut = NULL;
+                        return;
+                }
+
+                compute_curve_gamma_table_type_parametric(gamma_table, trc->parameter, trc->count);
+                *output_gamma_lut_length = 256;
+                for(i = 0; i < 256; i++) {
+                        output[i] = (uint16_t)(gamma_table[i] * 65535);
+                }
+                *output_gamma_lut = output;
+        } else {
+                if (trc->count == 0) {
+                        *output_gamma_lut = build_linear_table(4096);
+                        *output_gamma_lut_length = 4096;
+                } else if (trc->count == 1) {
+                        float gamma = 1./u8Fixed8Number_to_float(trc->data[0]);
+                        *output_gamma_lut = build_pow_table(gamma, 4096);
+                        *output_gamma_lut_length = 4096;
+                } else {
+                        //XXX: the choice of a minimum of 256 here is not backed by any theory, 
+                        //     measurement or data, however it is what lcms uses.
+                        *output_gamma_lut_length = trc->count;
+                        if (*output_gamma_lut_length < 256)
+                                *output_gamma_lut_length = 256;
+
+                        *output_gamma_lut = invert_lut(trc->data, trc->count, *output_gamma_lut_length);
+                }
+        }
+
+}
+
diff --git a/transform_util.h b/transform_util.h
new file mode 100644
index 0000000..8f358a8
--- /dev/null
+++ b/transform_util.h
@@ -0,0 +1,59 @@
+/* vim: set ts=8 sw=8 noexpandtab: */
+//  qcms
+//  Copyright (C) 2009 Mozilla Foundation
+//  Copyright (C) 1998-2007 Marti Maria
+//
+// Permission is hereby granted, free of charge, to any person obtaining 
+// a copy of this software and associated documentation files (the "Software"), 
+// to deal in the Software without restriction, including without limitation 
+// the rights to use, copy, modify, merge, publish, distribute, sublicense, 
+// and/or sell copies of the Software, and to permit persons to whom the Software 
+// is furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in 
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO 
+// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE 
+// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 
+// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+#ifndef _QCMS_TRANSFORM_UTIL_H
+#define _QCMS_TRANSFORM_UTIL_H
+
+#include <stdlib.h>
+
+#define CLU(table,x,y,z) table[(x*len + y*x_len + z*xy_len)*3]
+
+//XXX: could use a bettername
+typedef uint16_t uint16_fract_t;
+
+float lut_interp_linear(double value, uint16_t *table, int length);
+float lut_interp_linear_float(float value, float *table, int length);
+uint16_t lut_interp_linear16(uint16_t input_value, uint16_t *table, int length);
+
+
+static inline float lerp(float a, float b, float t)
+{
+        return a*(1.f-t) + b*t;
+}
+
+unsigned char clamp_u8(float v);
+float clamp_float(float a);
+
+float u8Fixed8Number_to_float(uint16_t x);
+
+
+float *build_input_gamma_table(struct curveType *TRC);
+struct matrix build_colorant_matrix(qcms_profile *p);
+void build_output_lut(struct curveType *trc,
+                      uint16_t **output_gamma_lut, size_t *output_gamma_lut_length);
+
+struct matrix matrix_invert(struct matrix mat);
+qcms_bool compute_precache(struct curveType *trc, uint8_t *output);
+
+
+#endif