Major rework of tnl module
New array_cache module
Support 8 texture units in core mesa (now support 8 everywhere)
Rework core mesa statechange operations to avoid flushing on many
noop statechanges.
diff --git a/src/mesa/math/m_eval.c b/src/mesa/math/m_eval.c
new file mode 100644
index 0000000..a4ae039
--- /dev/null
+++ b/src/mesa/math/m_eval.c
@@ -0,0 +1,501 @@
+/* $Id: m_eval.c,v 1.1 2000/12/26 05:09:31 keithw Exp $ */
+
+/*
+ * Mesa 3-D graphics library
+ * Version: 3.5
+ *
+ * Copyright (C) 1999-2000 Brian Paul All Rights Reserved.
+ *
+ * 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
+ * BRIAN PAUL 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.
+ */
+
+
+/*
+ * eval.c was written by
+ * Bernd Barsuhn (bdbarsuh@cip.informatik.uni-erlangen.de) and
+ * Volker Weiss (vrweiss@cip.informatik.uni-erlangen.de).
+ *
+ * My original implementation of evaluators was simplistic and didn't
+ * compute surface normal vectors properly. Bernd and Volker applied
+ * used more sophisticated methods to get better results.
+ *
+ * Thanks guys!
+ */
+
+
+#include "glheader.h"
+#include "config.h"
+#include "m_eval.h"
+
+static GLfloat inv_tab[MAX_EVAL_ORDER];
+
+
+
+/*
+ * Horner scheme for Bezier curves
+ *
+ * Bezier curves can be computed via a Horner scheme.
+ * Horner is numerically less stable than the de Casteljau
+ * algorithm, but it is faster. For curves of degree n
+ * the complexity of Horner is O(n) and de Casteljau is O(n^2).
+ * Since stability is not important for displaying curve
+ * points I decided to use the Horner scheme.
+ *
+ * A cubic Bezier curve with control points b0, b1, b2, b3 can be
+ * written as
+ *
+ * (([3] [3] ) [3] ) [3]
+ * c(t) = (([0]*s*b0 + [1]*t*b1)*s + [2]*t^2*b2)*s + [3]*t^2*b3
+ *
+ * [n]
+ * where s=1-t and the binomial coefficients [i]. These can
+ * be computed iteratively using the identity:
+ *
+ * [n] [n ] [n]
+ * [i] = (n-i+1)/i * [i-1] and [0] = 1
+ */
+
+
+void
+_math_horner_bezier_curve(const GLfloat *cp, GLfloat *out, GLfloat t,
+ GLuint dim, GLuint order)
+{
+ GLfloat s, powert;
+ GLuint i, k, bincoeff;
+
+ if(order >= 2)
+ {
+ bincoeff = order-1;
+ s = 1.0-t;
+
+ for(k=0; k<dim; k++)
+ out[k] = s*cp[k] + bincoeff*t*cp[dim+k];
+
+ for(i=2, cp+=2*dim, powert=t*t; i<order; i++, powert*=t, cp +=dim)
+ {
+ bincoeff *= order-i;
+ bincoeff *= inv_tab[i];
+
+ for(k=0; k<dim; k++)
+ out[k] = s*out[k] + bincoeff*powert*cp[k];
+ }
+ }
+ else /* order=1 -> constant curve */
+ {
+ for(k=0; k<dim; k++)
+ out[k] = cp[k];
+ }
+}
+
+/*
+ * Tensor product Bezier surfaces
+ *
+ * Again the Horner scheme is used to compute a point on a
+ * TP Bezier surface. First a control polygon for a curve
+ * on the surface in one parameter direction is computed,
+ * then the point on the curve for the other parameter
+ * direction is evaluated.
+ *
+ * To store the curve control polygon additional storage
+ * for max(uorder,vorder) points is needed in the
+ * control net cn.
+ */
+
+void
+_math_horner_bezier_surf(GLfloat *cn, GLfloat *out, GLfloat u, GLfloat v,
+ GLuint dim, GLuint uorder, GLuint vorder)
+{
+ GLfloat *cp = cn + uorder*vorder*dim;
+ GLuint i, uinc = vorder*dim;
+
+ if(vorder > uorder)
+ {
+ if(uorder >= 2)
+ {
+ GLfloat s, poweru;
+ GLuint j, k, bincoeff;
+
+ /* Compute the control polygon for the surface-curve in u-direction */
+ for(j=0; j<vorder; j++)
+ {
+ GLfloat *ucp = &cn[j*dim];
+
+ /* Each control point is the point for parameter u on a */
+ /* curve defined by the control polygons in u-direction */
+ bincoeff = uorder-1;
+ s = 1.0-u;
+
+ for(k=0; k<dim; k++)
+ cp[j*dim+k] = s*ucp[k] + bincoeff*u*ucp[uinc+k];
+
+ for(i=2, ucp+=2*uinc, poweru=u*u; i<uorder;
+ i++, poweru*=u, ucp +=uinc)
+ {
+ bincoeff *= uorder-i;
+ bincoeff *= inv_tab[i];
+
+ for(k=0; k<dim; k++)
+ cp[j*dim+k] = s*cp[j*dim+k] + bincoeff*poweru*ucp[k];
+ }
+ }
+
+ /* Evaluate curve point in v */
+ _math_horner_bezier_curve(cp, out, v, dim, vorder);
+ }
+ else /* uorder=1 -> cn defines a curve in v */
+ _math_horner_bezier_curve(cn, out, v, dim, vorder);
+ }
+ else /* vorder <= uorder */
+ {
+ if(vorder > 1)
+ {
+ GLuint i;
+
+ /* Compute the control polygon for the surface-curve in u-direction */
+ for(i=0; i<uorder; i++, cn += uinc)
+ {
+ /* For constant i all cn[i][j] (j=0..vorder) are located */
+ /* on consecutive memory locations, so we can use */
+ /* horner_bezier_curve to compute the control points */
+
+ _math_horner_bezier_curve(cn, &cp[i*dim], v, dim, vorder);
+ }
+
+ /* Evaluate curve point in u */
+ _math_horner_bezier_curve(cp, out, u, dim, uorder);
+ }
+ else /* vorder=1 -> cn defines a curve in u */
+ _math_horner_bezier_curve(cn, out, u, dim, uorder);
+ }
+}
+
+/*
+ * The direct de Casteljau algorithm is used when a point on the
+ * surface and the tangent directions spanning the tangent plane
+ * should be computed (this is needed to compute normals to the
+ * surface). In this case the de Casteljau algorithm approach is
+ * nicer because a point and the partial derivatives can be computed
+ * at the same time. To get the correct tangent length du and dv
+ * must be multiplied with the (u2-u1)/uorder-1 and (v2-v1)/vorder-1.
+ * Since only the directions are needed, this scaling step is omitted.
+ *
+ * De Casteljau needs additional storage for uorder*vorder
+ * values in the control net cn.
+ */
+
+void
+_math_de_casteljau_surf(GLfloat *cn, GLfloat *out, GLfloat *du, GLfloat *dv,
+ GLfloat u, GLfloat v, GLuint dim,
+ GLuint uorder, GLuint vorder)
+{
+ GLfloat *dcn = cn + uorder*vorder*dim;
+ GLfloat us = 1.0-u, vs = 1.0-v;
+ GLuint h, i, j, k;
+ GLuint minorder = uorder < vorder ? uorder : vorder;
+ GLuint uinc = vorder*dim;
+ GLuint dcuinc = vorder;
+
+ /* Each component is evaluated separately to save buffer space */
+ /* This does not drasticaly decrease the performance of the */
+ /* algorithm. If additional storage for (uorder-1)*(vorder-1) */
+ /* points would be available, the components could be accessed */
+ /* in the innermost loop which could lead to less cache misses. */
+
+#define CN(I,J,K) cn[(I)*uinc+(J)*dim+(K)]
+#define DCN(I, J) dcn[(I)*dcuinc+(J)]
+ if(minorder < 3)
+ {
+ if(uorder==vorder)
+ {
+ for(k=0; k<dim; k++)
+ {
+ /* Derivative direction in u */
+ du[k] = vs*(CN(1,0,k) - CN(0,0,k)) +
+ v*(CN(1,1,k) - CN(0,1,k));
+
+ /* Derivative direction in v */
+ dv[k] = us*(CN(0,1,k) - CN(0,0,k)) +
+ u*(CN(1,1,k) - CN(1,0,k));
+
+ /* bilinear de Casteljau step */
+ out[k] = us*(vs*CN(0,0,k) + v*CN(0,1,k)) +
+ u*(vs*CN(1,0,k) + v*CN(1,1,k));
+ }
+ }
+ else if(minorder == uorder)
+ {
+ for(k=0; k<dim; k++)
+ {
+ /* bilinear de Casteljau step */
+ DCN(1,0) = CN(1,0,k) - CN(0,0,k);
+ DCN(0,0) = us*CN(0,0,k) + u*CN(1,0,k);
+
+ for(j=0; j<vorder-1; j++)
+ {
+ /* for the derivative in u */
+ DCN(1,j+1) = CN(1,j+1,k) - CN(0,j+1,k);
+ DCN(1,j) = vs*DCN(1,j) + v*DCN(1,j+1);
+
+ /* for the `point' */
+ DCN(0,j+1) = us*CN(0,j+1,k) + u*CN(1,j+1,k);
+ DCN(0,j) = vs*DCN(0,j) + v*DCN(0,j+1);
+ }
+
+ /* remaining linear de Casteljau steps until the second last step */
+ for(h=minorder; h<vorder-1; h++)
+ for(j=0; j<vorder-h; j++)
+ {
+ /* for the derivative in u */
+ DCN(1,j) = vs*DCN(1,j) + v*DCN(1,j+1);
+
+ /* for the `point' */
+ DCN(0,j) = vs*DCN(0,j) + v*DCN(0,j+1);
+ }
+
+ /* derivative direction in v */
+ dv[k] = DCN(0,1) - DCN(0,0);
+
+ /* derivative direction in u */
+ du[k] = vs*DCN(1,0) + v*DCN(1,1);
+
+ /* last linear de Casteljau step */
+ out[k] = vs*DCN(0,0) + v*DCN(0,1);
+ }
+ }
+ else /* minorder == vorder */
+ {
+ for(k=0; k<dim; k++)
+ {
+ /* bilinear de Casteljau step */
+ DCN(0,1) = CN(0,1,k) - CN(0,0,k);
+ DCN(0,0) = vs*CN(0,0,k) + v*CN(0,1,k);
+ for(i=0; i<uorder-1; i++)
+ {
+ /* for the derivative in v */
+ DCN(i+1,1) = CN(i+1,1,k) - CN(i+1,0,k);
+ DCN(i,1) = us*DCN(i,1) + u*DCN(i+1,1);
+
+ /* for the `point' */
+ DCN(i+1,0) = vs*CN(i+1,0,k) + v*CN(i+1,1,k);
+ DCN(i,0) = us*DCN(i,0) + u*DCN(i+1,0);
+ }
+
+ /* remaining linear de Casteljau steps until the second last step */
+ for(h=minorder; h<uorder-1; h++)
+ for(i=0; i<uorder-h; i++)
+ {
+ /* for the derivative in v */
+ DCN(i,1) = us*DCN(i,1) + u*DCN(i+1,1);
+
+ /* for the `point' */
+ DCN(i,0) = us*DCN(i,0) + u*DCN(i+1,0);
+ }
+
+ /* derivative direction in u */
+ du[k] = DCN(1,0) - DCN(0,0);
+
+ /* derivative direction in v */
+ dv[k] = us*DCN(0,1) + u*DCN(1,1);
+
+ /* last linear de Casteljau step */
+ out[k] = us*DCN(0,0) + u*DCN(1,0);
+ }
+ }
+ }
+ else if(uorder == vorder)
+ {
+ for(k=0; k<dim; k++)
+ {
+ /* first bilinear de Casteljau step */
+ for(i=0; i<uorder-1; i++)
+ {
+ DCN(i,0) = us*CN(i,0,k) + u*CN(i+1,0,k);
+ for(j=0; j<vorder-1; j++)
+ {
+ DCN(i,j+1) = us*CN(i,j+1,k) + u*CN(i+1,j+1,k);
+ DCN(i,j) = vs*DCN(i,j) + v*DCN(i,j+1);
+ }
+ }
+
+ /* remaining bilinear de Casteljau steps until the second last step */
+ for(h=2; h<minorder-1; h++)
+ for(i=0; i<uorder-h; i++)
+ {
+ DCN(i,0) = us*DCN(i,0) + u*DCN(i+1,0);
+ for(j=0; j<vorder-h; j++)
+ {
+ DCN(i,j+1) = us*DCN(i,j+1) + u*DCN(i+1,j+1);
+ DCN(i,j) = vs*DCN(i,j) + v*DCN(i,j+1);
+ }
+ }
+
+ /* derivative direction in u */
+ du[k] = vs*(DCN(1,0) - DCN(0,0)) +
+ v*(DCN(1,1) - DCN(0,1));
+
+ /* derivative direction in v */
+ dv[k] = us*(DCN(0,1) - DCN(0,0)) +
+ u*(DCN(1,1) - DCN(1,0));
+
+ /* last bilinear de Casteljau step */
+ out[k] = us*(vs*DCN(0,0) + v*DCN(0,1)) +
+ u*(vs*DCN(1,0) + v*DCN(1,1));
+ }
+ }
+ else if(minorder == uorder)
+ {
+ for(k=0; k<dim; k++)
+ {
+ /* first bilinear de Casteljau step */
+ for(i=0; i<uorder-1; i++)
+ {
+ DCN(i,0) = us*CN(i,0,k) + u*CN(i+1,0,k);
+ for(j=0; j<vorder-1; j++)
+ {
+ DCN(i,j+1) = us*CN(i,j+1,k) + u*CN(i+1,j+1,k);
+ DCN(i,j) = vs*DCN(i,j) + v*DCN(i,j+1);
+ }
+ }
+
+ /* remaining bilinear de Casteljau steps until the second last step */
+ for(h=2; h<minorder-1; h++)
+ for(i=0; i<uorder-h; i++)
+ {
+ DCN(i,0) = us*DCN(i,0) + u*DCN(i+1,0);
+ for(j=0; j<vorder-h; j++)
+ {
+ DCN(i,j+1) = us*DCN(i,j+1) + u*DCN(i+1,j+1);
+ DCN(i,j) = vs*DCN(i,j) + v*DCN(i,j+1);
+ }
+ }
+
+ /* last bilinear de Casteljau step */
+ DCN(2,0) = DCN(1,0) - DCN(0,0);
+ DCN(0,0) = us*DCN(0,0) + u*DCN(1,0);
+ for(j=0; j<vorder-1; j++)
+ {
+ /* for the derivative in u */
+ DCN(2,j+1) = DCN(1,j+1) - DCN(0,j+1);
+ DCN(2,j) = vs*DCN(2,j) + v*DCN(2,j+1);
+
+ /* for the `point' */
+ DCN(0,j+1) = us*DCN(0,j+1 ) + u*DCN(1,j+1);
+ DCN(0,j) = vs*DCN(0,j) + v*DCN(0,j+1);
+ }
+
+ /* remaining linear de Casteljau steps until the second last step */
+ for(h=minorder; h<vorder-1; h++)
+ for(j=0; j<vorder-h; j++)
+ {
+ /* for the derivative in u */
+ DCN(2,j) = vs*DCN(2,j) + v*DCN(2,j+1);
+
+ /* for the `point' */
+ DCN(0,j) = vs*DCN(0,j) + v*DCN(0,j+1);
+ }
+
+ /* derivative direction in v */
+ dv[k] = DCN(0,1) - DCN(0,0);
+
+ /* derivative direction in u */
+ du[k] = vs*DCN(2,0) + v*DCN(2,1);
+
+ /* last linear de Casteljau step */
+ out[k] = vs*DCN(0,0) + v*DCN(0,1);
+ }
+ }
+ else /* minorder == vorder */
+ {
+ for(k=0; k<dim; k++)
+ {
+ /* first bilinear de Casteljau step */
+ for(i=0; i<uorder-1; i++)
+ {
+ DCN(i,0) = us*CN(i,0,k) + u*CN(i+1,0,k);
+ for(j=0; j<vorder-1; j++)
+ {
+ DCN(i,j+1) = us*CN(i,j+1,k) + u*CN(i+1,j+1,k);
+ DCN(i,j) = vs*DCN(i,j) + v*DCN(i,j+1);
+ }
+ }
+
+ /* remaining bilinear de Casteljau steps until the second last step */
+ for(h=2; h<minorder-1; h++)
+ for(i=0; i<uorder-h; i++)
+ {
+ DCN(i,0) = us*DCN(i,0) + u*DCN(i+1,0);
+ for(j=0; j<vorder-h; j++)
+ {
+ DCN(i,j+1) = us*DCN(i,j+1) + u*DCN(i+1,j+1);
+ DCN(i,j) = vs*DCN(i,j) + v*DCN(i,j+1);
+ }
+ }
+
+ /* last bilinear de Casteljau step */
+ DCN(0,2) = DCN(0,1) - DCN(0,0);
+ DCN(0,0) = vs*DCN(0,0) + v*DCN(0,1);
+ for(i=0; i<uorder-1; i++)
+ {
+ /* for the derivative in v */
+ DCN(i+1,2) = DCN(i+1,1) - DCN(i+1,0);
+ DCN(i,2) = us*DCN(i,2) + u*DCN(i+1,2);
+
+ /* for the `point' */
+ DCN(i+1,0) = vs*DCN(i+1,0) + v*DCN(i+1,1);
+ DCN(i,0) = us*DCN(i,0) + u*DCN(i+1,0);
+ }
+
+ /* remaining linear de Casteljau steps until the second last step */
+ for(h=minorder; h<uorder-1; h++)
+ for(i=0; i<uorder-h; i++)
+ {
+ /* for the derivative in v */
+ DCN(i,2) = us*DCN(i,2) + u*DCN(i+1,2);
+
+ /* for the `point' */
+ DCN(i,0) = us*DCN(i,0) + u*DCN(i+1,0);
+ }
+
+ /* derivative direction in u */
+ du[k] = DCN(1,0) - DCN(0,0);
+
+ /* derivative direction in v */
+ dv[k] = us*DCN(0,2) + u*DCN(1,2);
+
+ /* last linear de Casteljau step */
+ out[k] = us*DCN(0,0) + u*DCN(1,0);
+ }
+ }
+#undef DCN
+#undef CN
+}
+
+
+/*
+ * Do one-time initialization for evaluators.
+ */
+void _math_init_eval( void )
+{
+ GLuint i;
+
+ /* KW: precompute 1/x for useful x.
+ */
+ for (i = 1 ; i < MAX_EVAL_ORDER ; i++)
+ inv_tab[i] = 1.0 / i;
+}
+