Merge commit 'origin/gallium-0.1' into gallium-0.2

Conflicts:

	src/mesa/drivers/dri/common/dri_util.c

12 files changed, 1276 insertions, 118 deletions

diff --git a/progs/trivial/Makefile b/progs/trivial/Makefile
index 3f26798ca62..40ee5e81c41 100644
--- a/progs/trivial/Makefile
+++ b/progs/trivial/Makefile
@@ -15,6 +15,7 @@ SOURCES = \
 	clear-fbo.c \
 	clear-scissor.c \
 	clear-undefined.c \
+	clear-repeat.c \
 	clear.c \
 	dlist-dangling.c \
 	dlist-edgeflag-dangling.c \
@@ -92,9 +93,13 @@ SOURCES = \
 	tri-flat-clip.c \
 	tri-flat.c \
 	tri-fog.c \
+	tri-fp.c \
+	tri-fp-const-imm.c \
+	tri-lit.c \
 	tri-mask-tri.c \
 	tri-orig.c \
 	tri-query.c \
+	tri-repeat.c \
 	tri-scissor-tri.c \
 	tri-stencil.c \
 	tri-tex.c \
@@ -103,6 +108,8 @@ SOURCES = \
 	tri-unfilled-edgeflag.c \
 	tri-unfilled-clip.c \
 	tri-unfilled-smooth.c \
+	tri-unfilled-tri.c \
+	tri-unfilled-tri-lit.c \
 	tri-unfilled-userclip-stip.c \
 	tri-unfilled-userclip.c \
 	tri-unfilled.c \
diff --git a/progs/trivial/clear-repeat.c b/progs/trivial/clear-repeat.c
new file mode 100644
index 00000000000..9f9490c6c87
--- /dev/null
+++ b/progs/trivial/clear-repeat.c
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+
+static void Init(void)
+{
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+    glClearColor(0.3, 0.1, 0.3, 0.0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0);
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+   switch (key) {
+   case 27:
+      exit(0);
+   default:
+      glutPostRedisplay();
+      return;
+   }
+}
+
+static void Draw(void)
+{
+   static float f = 0;
+   while (1) {
+      f += .1;
+      glClearColor((sin(f)+1)/2.0,(cos(f)+1)/2.0,0.5,1); 
+      glClear(GL_COLOR_BUFFER_BIT); 
+      glutSwapBuffers();
+   }
+   glutPostRedisplay();
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB | GLUT_ALPHA;
+    type |= GLUT_DOUBLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+    return 0;
+}
diff --git a/progs/trivial/tri-fp-const-imm.c b/progs/trivial/tri-fp-const-imm.c
new file mode 100644
index 00000000000..71113802ed1
--- /dev/null
+++ b/progs/trivial/tri-fp-const-imm.c
@@ -0,0 +1,169 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#define GL_GLEXT_PROTOTYPES
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+
+GLenum doubleBuffer;
+
+static void Init(void)
+{
+   GLint errno;
+   GLuint prognum;
+   static const char *prog1 =
+      "!!ARBfp1.0\n"
+      "TEMP R1;\n"
+      "MOV  R1, state.material.emission;\n"
+      "MUL  R1, R1, {0.9}.x;\n"
+      "MOV  result.color, R1;\n"
+      "END\n";
+
+
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+   /* Setup the fragment program */
+   glGenProgramsARB(1, &prognum);
+   glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, prognum);
+   glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
+                      strlen(prog1), (const GLubyte *)prog1);
+   
+   errno = glGetError();
+   printf("glGetError = 0x%x\n", errno);
+   if (errno != GL_NO_ERROR) {
+      GLint errorpos;
+
+      glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errorpos);
+      printf("errorpos: %d\n", errorpos);
+      printf("glError(GL_PROGRAM_ERROR_STRING_ARB) = %s\n",
+             (char *) glGetString(GL_PROGRAM_ERROR_STRING_ARB));
+   }
+   glEnable(GL_FRAGMENT_PROGRAM_ARB);
+
+    glClearColor(0.0, 0.0, 1.0, 0.0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+/*     glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0); */
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+
+    switch (key) {
+      case 27:
+	exit(1);
+      default:
+	return;
+    }
+
+    glutPostRedisplay();
+}
+
+static void Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glBegin(GL_TRIANGLES);
+   glColor3f(0,0,.7); 
+   glTexCoord3f(0,0,.7); 
+   glVertex3f( 0.9, -0.9, -0.0);
+   glColor3f(.8,0,0); 
+   glTexCoord3f(.8,0,0); 
+   glVertex3f( 0.9,  0.9, -0.0);
+   glColor3f(0,.9,0); 
+   glTexCoord3f(0,.9,0); 
+   glVertex3f(-0.9,  0.0, -0.0);
+   glEnd();
+
+   glFlush();
+
+   if (doubleBuffer) {
+      glutSwapBuffers();
+   }
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    GLint i;
+
+    doubleBuffer = GL_FALSE;
+
+    for (i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "-sb") == 0) {
+	    doubleBuffer = GL_FALSE;
+	} else if (strcmp(argv[i], "-db") == 0) {
+	    doubleBuffer = GL_TRUE;
+	} else {
+	    fprintf(stderr, "%s (Bad option).\n", argv[i]);
+	    return GL_FALSE;
+	}
+    }
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB | GLUT_ALPHA;
+    type |= (doubleBuffer) ? GLUT_DOUBLE : GLUT_SINGLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+	return 0;
+}
diff --git a/progs/trivial/tri-fp.c b/progs/trivial/tri-fp.c
new file mode 100644
index 00000000000..9ff355a4caf
--- /dev/null
+++ b/progs/trivial/tri-fp.c
@@ -0,0 +1,167 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#define GL_GLEXT_PROTOTYPES
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+
+GLenum doubleBuffer;
+
+static void Init(void)
+{
+   GLint errno;
+   GLuint prognum;
+   static const char *prog1 =
+      "!!ARBfp1.0\n"
+      "MOV  result.color, fragment.texcoord[1];\n"
+//      "MOV  result.color, fragment.color;\n"
+      "END\n";
+
+
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+   /* Setup the fragment program */
+   glGenProgramsARB(1, &prognum);
+   glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, prognum);
+   glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
+                      strlen(prog1), (const GLubyte *)prog1);
+   
+   errno = glGetError();
+   printf("glGetError = 0x%x\n", errno);
+   if (errno != GL_NO_ERROR) {
+      GLint errorpos;
+
+      glGetIntegerv(GL_PROGRAM_ERROR_POSITION_ARB, &errorpos);
+      printf("errorpos: %d\n", errorpos);
+      printf("glError(GL_PROGRAM_ERROR_STRING_ARB) = %s\n",
+             (char *) glGetString(GL_PROGRAM_ERROR_STRING_ARB));
+   }
+   glEnable(GL_FRAGMENT_PROGRAM_ARB);
+
+    glClearColor(0.0, 0.0, 1.0, 0.0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+/*     glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0); */
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+
+    switch (key) {
+      case 27:
+	exit(1);
+      default:
+	return;
+    }
+
+    glutPostRedisplay();
+}
+
+static void Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glBegin(GL_TRIANGLES);
+   glColor3f(0,0,.7); 
+   glTexCoord3f(0,0,.7); 
+   glVertex3f( 0.9, -0.9, -0.0);
+   glColor3f(.8,0,0); 
+   glTexCoord3f(.8,0,0); 
+   glVertex3f( 0.9,  0.9, -0.0);
+   glColor3f(0,.9,0); 
+   glTexCoord3f(0,.9,0); 
+   glVertex3f(-0.9,  0.0, -0.0);
+   glEnd();
+
+   glFlush();
+
+   if (doubleBuffer) {
+      glutSwapBuffers();
+   }
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    GLint i;
+
+    doubleBuffer = GL_FALSE;
+
+    for (i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "-sb") == 0) {
+	    doubleBuffer = GL_FALSE;
+	} else if (strcmp(argv[i], "-db") == 0) {
+	    doubleBuffer = GL_TRUE;
+	} else {
+	    fprintf(stderr, "%s (Bad option).\n", argv[i]);
+	    return GL_FALSE;
+	}
+    }
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB | GLUT_ALPHA;
+    type |= (doubleBuffer) ? GLUT_DOUBLE : GLUT_SINGLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+	return 0;
+}
diff --git a/progs/trivial/tri-lit.c b/progs/trivial/tri-lit.c
new file mode 100644
index 00000000000..22f199c9828
--- /dev/null
+++ b/progs/trivial/tri-lit.c
@@ -0,0 +1,139 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+
+GLenum doubleBuffer;
+
+static void Init(void)
+{
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+  glEnable(GL_LIGHTING);
+  glEnable(GL_LIGHT0);
+
+    glClearColor(0.0, 0.0, 1.0, 0.0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+/*     glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0); */
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+
+    switch (key) {
+      case 27:
+	exit(1);
+      default:
+	return;
+    }
+
+    glutPostRedisplay();
+}
+
+static void Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glBegin(GL_TRIANGLES);
+   glNormal3f(0,0,.7); 
+   glVertex3f( 0.9, -0.9, -0.0);
+   glNormal3f(.8,0,0); 
+   glVertex3f( 0.9,  0.9, -0.0);
+   glNormal3f(0,.9,0); 
+   glVertex3f(-0.9,  0.0, -0.0);
+   glEnd();
+
+   glFlush();
+
+   if (doubleBuffer) {
+      glutSwapBuffers();
+   }
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    GLint i;
+
+    doubleBuffer = GL_FALSE;
+
+    for (i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "-sb") == 0) {
+	    doubleBuffer = GL_FALSE;
+	} else if (strcmp(argv[i], "-db") == 0) {
+	    doubleBuffer = GL_TRUE;
+	} else {
+	    fprintf(stderr, "%s (Bad option).\n", argv[i]);
+	    return GL_FALSE;
+	}
+    }
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB | GLUT_ALPHA;
+    type |= (doubleBuffer) ? GLUT_DOUBLE : GLUT_SINGLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+	return 0;
+}
diff --git a/progs/trivial/tri-repeat.c b/progs/trivial/tri-repeat.c
new file mode 100644
index 00000000000..e894269e84b
--- /dev/null
+++ b/progs/trivial/tri-repeat.c
@@ -0,0 +1,118 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+
+static void Init(void)
+{
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+    glClearColor(0.3, 0.1, 0.3, 0.0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0);
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+   switch (key) {
+   case 27:
+      exit(0);
+   default:
+      glutPostRedisplay();
+      return;
+   }
+}
+
+static void Draw(void)
+{
+   static float f = 0;
+   f += .1;
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glBegin(GL_TRIANGLES);
+   glColor3f((sin(f)+1)/2.0,0,0); 
+   glVertex3f(-0.9, -0.9, -30.0);
+   glColor3f(0,(cos(f)+1)/2.0,0); 
+   glVertex3f( 0.9, -0.9, -30.0);
+   glColor3f(0,0,.7); 
+   glVertex3f( 0.0,  0.9, -30.0);
+   glEnd();
+
+   glutSwapBuffers();
+   glutPostRedisplay();
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB | GLUT_ALPHA;
+    type |= GLUT_DOUBLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+    return 0;
+}
diff --git a/progs/trivial/tri-unfilled-tri-lit.c b/progs/trivial/tri-unfilled-tri-lit.c
new file mode 100644
index 00000000000..f9e1d009884
--- /dev/null
+++ b/progs/trivial/tri-unfilled-tri-lit.c
@@ -0,0 +1,178 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+
+GLenum doubleBuffer;
+
+static void Init(void)
+{
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+  glEnable(GL_LIGHTING);
+  glEnable(GL_LIGHT0);
+
+    glClearColor(0.0, 0.0, 1.0, 0.0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+/*     glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0); */
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+
+    switch (key) {
+      case 27:
+	exit(1);
+      default:
+	return;
+    }
+
+    glutPostRedisplay();
+}
+
+static void Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glPolygonMode(GL_FRONT, GL_LINE);
+   glPolygonMode(GL_BACK, GL_LINE);
+
+   glBegin(GL_TRIANGLES);
+   glNormal3f(0,0,.7); 
+   glVertex3f( 0.9, -0.9, -0.0);
+   glNormal3f(.8,0,0); 
+   glVertex3f( 0.9,  0.9, -0.0);
+   glNormal3f(0,.9,0); 
+   glVertex3f(-0.9,  0.0, -0.0);
+   glEnd();
+
+   glPolygonMode(GL_FRONT, GL_FILL);
+   glPolygonMode(GL_BACK, GL_FILL);
+
+   glBegin(GL_TRIANGLES);
+   glNormal3f(0,0,.7); 
+   glVertex3f( 0.8, -0.8, -0.0);
+   glNormal3f(.8,0,0); 
+   glVertex3f( 0.8,  0.8, -0.0);
+   glNormal3f(0,.9,0); 
+   glVertex3f(-0.8,  0.0, -0.0);
+   glEnd();
+
+   glPolygonMode(GL_FRONT, GL_LINE);
+   glPolygonMode(GL_BACK, GL_LINE);
+
+   glBegin(GL_TRIANGLES);
+   glNormal3f(.8,0,0); 
+   glVertex3f( -0.9,  0.9, -0.0);
+   glNormal3f(0,0,.7); 
+   glVertex3f( -0.9, -0.9, -0.0);
+   glNormal3f(0,.9,0); 
+   glVertex3f( 0.9,  0.0, -0.0);
+   glEnd();
+
+   glPolygonMode(GL_FRONT, GL_FILL);
+   glPolygonMode(GL_BACK, GL_FILL);
+
+   glBegin(GL_TRIANGLES);
+   glNormal3f(.8,0,0); 
+   glVertex3f( -0.8,  0.8, -0.0);
+   glNormal3f(0,0,.7); 
+   glVertex3f( -0.8, -0.8, -0.0);
+   glNormal3f(0,.9,0); 
+   glVertex3f( 0.8,  0.0, -0.0);
+   glEnd();
+
+   glFlush();
+
+   if (doubleBuffer) {
+      glutSwapBuffers();
+   }
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    GLint i;
+
+    doubleBuffer = GL_FALSE;
+
+    for (i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "-sb") == 0) {
+	    doubleBuffer = GL_FALSE;
+	} else if (strcmp(argv[i], "-db") == 0) {
+	    doubleBuffer = GL_TRUE;
+	} else {
+	    fprintf(stderr, "%s (Bad option).\n", argv[i]);
+	    return GL_FALSE;
+	}
+    }
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB | GLUT_ALPHA;
+    type |= (doubleBuffer) ? GLUT_DOUBLE : GLUT_SINGLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+	return 0;
+}
diff --git a/progs/trivial/tri-unfilled-tri.c b/progs/trivial/tri-unfilled-tri.c
new file mode 100644
index 00000000000..d9a9faeb9f5
--- /dev/null
+++ b/progs/trivial/tri-unfilled-tri.c
@@ -0,0 +1,175 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+
+GLenum doubleBuffer;
+
+static void Init(void)
+{
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+    glClearColor(0.0, 0.0, 1.0, 0.0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+/*     glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0); */
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+
+    switch (key) {
+      case 27:
+	exit(1);
+      default:
+	return;
+    }
+
+    glutPostRedisplay();
+}
+
+static void Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glPolygonMode(GL_FRONT, GL_LINE);
+   glPolygonMode(GL_BACK, GL_LINE);
+
+   glBegin(GL_TRIANGLES);
+   glColor3f(0,0,.7); 
+   glVertex3f( 0.9, -0.9, -0.0);
+   glColor3f(.8,0,0); 
+   glVertex3f( 0.9,  0.9, -0.0);
+   glColor3f(0,.9,0); 
+   glVertex3f(-0.9,  0.0, -0.0);
+   glEnd();
+
+   glPolygonMode(GL_FRONT, GL_FILL);
+   glPolygonMode(GL_BACK, GL_FILL);
+
+   glBegin(GL_TRIANGLES);
+   glColor3f(0,0,.7); 
+   glVertex3f( 0.8, -0.8, -0.0);
+   glColor3f(.8,0,0); 
+   glVertex3f( 0.8,  0.8, -0.0);
+   glColor3f(0,.9,0); 
+   glVertex3f(-0.8,  0.0, -0.0);
+   glEnd();
+
+   glPolygonMode(GL_FRONT, GL_LINE);
+   glPolygonMode(GL_BACK, GL_LINE);
+
+   glBegin(GL_TRIANGLES);
+   glColor3f(.8,0,0); 
+   glVertex3f( -0.9,  0.9, -0.0);
+   glColor3f(0,0,.7); 
+   glVertex3f( -0.9, -0.9, -0.0);
+   glColor3f(0,.9,0); 
+   glVertex3f( 0.9,  0.0, -0.0);
+   glEnd();
+
+   glPolygonMode(GL_FRONT, GL_FILL);
+   glPolygonMode(GL_BACK, GL_FILL);
+
+   glBegin(GL_TRIANGLES);
+   glColor3f(.8,0,0); 
+   glVertex3f( -0.8,  0.8, -0.0);
+   glColor3f(0,0,.7); 
+   glVertex3f( -0.8, -0.8, -0.0);
+   glColor3f(0,.9,0); 
+   glVertex3f( 0.8,  0.0, -0.0);
+   glEnd();
+
+   glFlush();
+
+   if (doubleBuffer) {
+      glutSwapBuffers();
+   }
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    GLint i;
+
+    doubleBuffer = GL_FALSE;
+
+    for (i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "-sb") == 0) {
+	    doubleBuffer = GL_FALSE;
+	} else if (strcmp(argv[i], "-db") == 0) {
+	    doubleBuffer = GL_TRUE;
+	} else {
+	    fprintf(stderr, "%s (Bad option).\n", argv[i]);
+	    return GL_FALSE;
+	}
+    }
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB | GLUT_ALPHA;
+    type |= (doubleBuffer) ? GLUT_DOUBLE : GLUT_SINGLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+	return 0;
+}
diff --git a/src/gallium/auxiliary/translate/translate.h b/src/gallium/auxiliary/translate/translate.h
index 650cd81fa60..34526eb0617 100644
--- a/src/gallium/auxiliary/translate/translate.h
+++ b/src/gallium/auxiliary/translate/translate.h
@@ -48,8 +48,8 @@ struct translate_element
 {
    enum pipe_format input_format;
    enum pipe_format output_format;
-   unsigned input_buffer;
-   unsigned input_offset;       /* can't really reduce the size of these */
+   unsigned input_buffer:8;
+   unsigned input_offset:24;
    unsigned output_offset;
 };
 
diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c
index 7955186e168..b62db8d8f33 100644
--- a/src/gallium/auxiliary/translate/translate_sse.c
+++ b/src/gallium/auxiliary/translate/translate_sse.c
@@ -29,7 +29,7 @@
 #include "pipe/p_config.h"
 #include "pipe/p_compiler.h"
 #include "util/u_memory.h"
-#include "util/u_simple_list.h"
+#include "util/u_math.h"
 
 #include "translate.h"
 
@@ -56,6 +56,11 @@ typedef void (PIPE_CDECL *run_elts_func)( struct translate *translate,
                                           unsigned count,
                                           void *output_buffer );
 
+struct translate_buffer {
+   const void *base_ptr;
+   unsigned stride;
+   void *ptr;                   /* updated per vertex */
+};
 
 
 struct translate_sse {
@@ -73,14 +78,20 @@ struct translate_sse {
    float float_255[4];
    float inv_255[4];
 
-   struct {
-      char *input_ptr;
-      unsigned input_stride;
-   } attrib[PIPE_MAX_ATTRIBS];
+   struct translate_buffer buffer[PIPE_MAX_ATTRIBS];
+   unsigned nr_buffers;
 
    run_func      gen_run;
    run_elts_func gen_run_elts;
 
+   /* these are actually known values, but putting them in a struct
+    * like this is helpful to keep them in sync across the file.
+    */
+   struct x86_reg tmp_EAX;
+   struct x86_reg idx_EBX;     /* either start+i or &elt[i] */
+   struct x86_reg outbuf_ECX;
+   struct x86_reg machine_EDX;
+   struct x86_reg count_ESI;    /* decrements to zero */
 };
 
 static int get_offset( const void *a, const void *b )
@@ -95,10 +106,6 @@ static struct x86_reg get_identity( struct translate_sse *p )
    struct x86_reg reg = x86_make_reg(file_XMM, 6);
 
    if (!p->loaded_identity) {
-      /* Nasty: 
-       */
-      struct x86_reg translateESI = x86_make_reg(file_REG32, reg_SI);
-
       p->loaded_identity = TRUE;
       p->identity[0] = 0;
       p->identity[1] = 0;
@@ -106,7 +113,7 @@ static struct x86_reg get_identity( struct translate_sse *p )
       p->identity[3] = 1;
 
       sse_movups(p->func, reg, 
-		 x86_make_disp(translateESI, 
+		 x86_make_disp(p->machine_EDX, 
 			       get_offset(p, &p->identity[0])));
    }
 
@@ -115,11 +122,9 @@ static struct x86_reg get_identity( struct translate_sse *p )
 
 static struct x86_reg get_255( struct translate_sse *p )
 {
-   struct x86_reg reg = x86_make_reg(file_XMM, 6);
+   struct x86_reg reg = x86_make_reg(file_XMM, 7);
 
    if (!p->loaded_255) {
-      struct x86_reg translateESI = x86_make_reg(file_REG32, reg_SI);
-
       p->loaded_255 = TRUE;
       p->float_255[0] =
 	 p->float_255[1] =
@@ -127,12 +132,11 @@ static struct x86_reg get_255( struct translate_sse *p )
 	 p->float_255[3] = 255.0f;
 
       sse_movups(p->func, reg, 
-		 x86_make_disp(translateESI, 
+		 x86_make_disp(p->machine_EDX, 
 			       get_offset(p, &p->float_255[0])));
    }
 
    return reg;
-   return x86_make_reg(file_XMM, 7);
 }
 
 static struct x86_reg get_inv_255( struct translate_sse *p )
@@ -140,8 +144,6 @@ static struct x86_reg get_inv_255( struct translate_sse *p )
    struct x86_reg reg = x86_make_reg(file_XMM, 5);
 
    if (!p->loaded_inv_255) {
-      struct x86_reg translateESI = x86_make_reg(file_REG32, reg_SI);
-
       p->loaded_inv_255 = TRUE;
       p->inv_255[0] =
 	 p->inv_255[1] =
@@ -149,7 +151,7 @@ static struct x86_reg get_inv_255( struct translate_sse *p )
 	 p->inv_255[3] = 1.0f / 255.0f;
 
       sse_movups(p->func, reg, 
-		 x86_make_disp(translateESI, 
+		 x86_make_disp(p->machine_EDX, 
 			       get_offset(p, &p->inv_255[0])));
    }
 
@@ -283,28 +285,6 @@ static void emit_store_R8G8B8A8_UNORM( struct translate_sse *p,
 
 
 
-static void get_src_ptr( struct translate_sse *p,
-			 struct x86_reg srcEAX,
-			 struct x86_reg translateREG,
-			 struct x86_reg eltREG,	
-			 unsigned a )
-{
-   struct x86_reg input_ptr = 
-      x86_make_disp(translateREG, 
-		    get_offset(p, &p->attrib[a].input_ptr));
-
-   struct x86_reg input_stride = 
-      x86_make_disp(translateREG, 
-		    get_offset(p, &p->attrib[a].input_stride));
-
-   /* Calculate pointer to current attrib:
-    */
-   x86_mov(p->func, srcEAX, input_stride);
-   x86_imul(p->func, srcEAX, eltREG);	
-   x86_add(p->func, srcEAX, input_ptr);
-}
-
-
 /* Extended swizzles?  Maybe later.
  */  
 static void emit_swizzle( struct translate_sse *p,
@@ -374,12 +354,126 @@ static boolean translate_attr( struct translate_sse *p,
    return TRUE;
 }
 
-/* Build run( struct translate *translate,
+
+static boolean init_inputs( struct translate_sse *p,
+                            boolean linear )
+{
+   unsigned i;
+   if (linear) {
+      for (i = 0; i < p->nr_buffers; i++) {
+         struct x86_reg buf_stride   = x86_make_disp(p->machine_EDX,
+                                                     get_offset(p, &p->buffer[i].stride));
+         struct x86_reg buf_ptr      = x86_make_disp(p->machine_EDX,
+                                                     get_offset(p, &p->buffer[i].ptr));
+         struct x86_reg buf_base_ptr = x86_make_disp(p->machine_EDX,
+                                                     get_offset(p, &p->buffer[i].base_ptr));
+         struct x86_reg elt = p->idx_EBX;
+         struct x86_reg tmp = p->tmp_EAX;
+
+
+         /* Calculate pointer to first attrib:
+          */
+         x86_mov(p->func, tmp, buf_stride);
+         x86_imul(p->func, tmp, elt);
+         x86_add(p->func, tmp, buf_base_ptr);
+
+
+         /* In the linear case, keep the buffer pointer instead of the
+          * index number.
+          */
+         if (p->nr_buffers == 1) 
+            x86_mov( p->func, elt, tmp );
+         else
+            x86_mov( p->func, buf_ptr, tmp );
+      }
+   }
+
+   return TRUE;
+}
+
+
+static struct x86_reg get_buffer_ptr( struct translate_sse *p,
+                                      boolean linear,
+                                      unsigned buf_idx,
+                                      struct x86_reg elt )
+{
+   if (linear && p->nr_buffers == 1) {
+      return p->idx_EBX;
+   }
+   else if (linear) {
+      struct x86_reg ptr = p->tmp_EAX;
+      struct x86_reg buf_ptr = 
+         x86_make_disp(p->machine_EDX, 
+                       get_offset(p, &p->buffer[buf_idx].ptr));
+      
+      x86_mov(p->func, ptr, buf_ptr);
+      return ptr;
+   }
+   else {
+      struct x86_reg ptr = p->tmp_EAX;
+
+      struct x86_reg buf_stride = 
+         x86_make_disp(p->machine_EDX, 
+                       get_offset(p, &p->buffer[buf_idx].stride));
+
+      struct x86_reg buf_base_ptr = 
+         x86_make_disp(p->machine_EDX, 
+                       get_offset(p, &p->buffer[buf_idx].base_ptr));
+
+
+
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(p->func, ptr, buf_stride);
+      x86_imul(p->func, ptr, elt);
+      x86_add(p->func, ptr, buf_base_ptr);
+      return ptr;
+   }
+}
+
+
+
+static boolean incr_inputs( struct translate_sse *p, 
+                            boolean linear )
+{
+   if (linear && p->nr_buffers == 1) {
+      struct x86_reg stride = x86_make_disp(p->machine_EDX,
+                                            get_offset(p, &p->buffer[0].stride));
+
+      x86_add(p->func, p->idx_EBX, stride);
+      sse_prefetchnta(p->func, x86_make_disp(p->idx_EBX, 192));
+   }
+   else if (linear) {
+      unsigned i;
+
+      /* Is this worthwhile??
+       */
+      for (i = 0; i < p->nr_buffers; i++) {
+         struct x86_reg buf_ptr = x86_make_disp(p->machine_EDX,
+                                                get_offset(p, &p->buffer[i].ptr));
+         struct x86_reg buf_stride = x86_make_disp(p->machine_EDX,
+                                                   get_offset(p, &p->buffer[i].stride));
+
+         x86_mov(p->func, p->tmp_EAX, buf_ptr);
+         x86_add(p->func, p->tmp_EAX, buf_stride);
+         if (i == 0) sse_prefetchnta(p->func, x86_make_disp(p->tmp_EAX, 192));
+         x86_mov(p->func, buf_ptr, p->tmp_EAX);
+      }
+   } 
+   else {
+      x86_lea(p->func, p->idx_EBX, x86_make_disp(p->idx_EBX, 4));
+   }
+   
+   return TRUE;
+}
+
+
+/* Build run( struct translate *machine,
  *            unsigned start,
  *            unsigned count,
  *            void *output_buffer )
  * or
- *  run_elts( struct translate *translate,
+ *  run_elts( struct translate *machine,
  *            unsigned *elts,
  *            unsigned count,
  *            void *output_buffer )
@@ -394,14 +488,15 @@ static boolean build_vertex_emit( struct translate_sse *p,
 				  struct x86_function *func,
 				  boolean linear )
 {
-   struct x86_reg vertexECX    = x86_make_reg(file_REG32, reg_AX);
-   struct x86_reg idxEBX       = x86_make_reg(file_REG32, reg_BX);
-   struct x86_reg srcEAX       = x86_make_reg(file_REG32, reg_CX);
-   struct x86_reg countEBP     = x86_make_reg(file_REG32, reg_BP);
-   struct x86_reg translateESI = x86_make_reg(file_REG32, reg_SI);
    int fixup, label;
    unsigned j;
 
+   p->tmp_EAX       = x86_make_reg(file_REG32, reg_AX);
+   p->idx_EBX       = x86_make_reg(file_REG32, reg_BX);
+   p->outbuf_ECX    = x86_make_reg(file_REG32, reg_CX);
+   p->machine_EDX   = x86_make_reg(file_REG32, reg_DX);
+   p->count_ESI     = x86_make_reg(file_REG32, reg_SI);
+
    p->func = func;
    p->loaded_inv_255 = FALSE;
    p->loaded_255 = FALSE;
@@ -411,74 +506,65 @@ static boolean build_vertex_emit( struct translate_sse *p,
 
    /* Push a few regs?
     */
-   x86_push(p->func, countEBP);
-   x86_push(p->func, translateESI);
-   x86_push(p->func, idxEBX);
-
-   /* Get vertex count, compare to zero
-    */
-   x86_xor(p->func, idxEBX, idxEBX);
-   x86_mov(p->func, countEBP, x86_fn_arg(p->func, 3));
-   x86_cmp(p->func, countEBP, idxEBX);
-   fixup = x86_jcc_forward(p->func, cc_E);
-
-   /* If linear, idx is the current element, otherwise it is a pointer
-    * to the current element.
-    */
-   x86_mov(p->func, idxEBX, x86_fn_arg(p->func, 2));
+   x86_push(p->func, p->idx_EBX);
+   x86_push(p->func, p->count_ESI);
 
-   /* Initialize destination register. 
+   /* Load arguments into regs:
     */
-   x86_mov(p->func, vertexECX, x86_fn_arg(p->func, 4));
+   x86_mov(p->func, p->machine_EDX, x86_fn_arg(p->func, 1));
+   x86_mov(p->func, p->idx_EBX, x86_fn_arg(p->func, 2));
+   x86_mov(p->func, p->count_ESI, x86_fn_arg(p->func, 3));
+   x86_mov(p->func, p->outbuf_ECX, x86_fn_arg(p->func, 4));
 
-   /* Move argument 1 (translate_sse pointer) into a reg:
+   /* Get vertex count, compare to zero
     */
-   x86_mov(p->func, translateESI, x86_fn_arg(p->func, 1));
+   x86_xor(p->func, p->tmp_EAX, p->tmp_EAX);
+   x86_cmp(p->func, p->count_ESI, p->tmp_EAX);
+   fixup = x86_jcc_forward(p->func, cc_E);
 
-   
    /* always load, needed or not:
     */
+   init_inputs(p, linear);
 
-   /* Note address for loop jump */
+   /* Note address for loop jump
+    */
    label = x86_get_label(p->func);
-
-
-   for (j = 0; j < p->translate.key.nr_elements; j++) {
-      const struct translate_element *a = &p->translate.key.element[j];
-
-      struct x86_reg destEAX = x86_make_disp(vertexECX, 
-					     a->output_offset);
-
-      /* Figure out source pointer address:
-       */
-      if (linear) {
-	 get_src_ptr(p, srcEAX, translateESI, idxEBX, j);
+   {
+      struct x86_reg elt = linear ? p->idx_EBX : x86_deref(p->idx_EBX);
+      int last_vb = -1;
+      struct x86_reg vb;
+
+      for (j = 0; j < p->translate.key.nr_elements; j++) {
+         const struct translate_element *a = &p->translate.key.element[j];
+
+         /* Figure out source pointer address:
+          */
+         if (a->input_buffer != last_vb) {
+            last_vb = a->input_buffer;
+            vb = get_buffer_ptr(p, linear, a->input_buffer, elt);
+         }
+         
+         if (!translate_attr( p, a, 
+                              x86_make_disp(vb, a->input_offset), 
+                              x86_make_disp(p->outbuf_ECX, a->output_offset)))
+            return FALSE;
       }
-      else {
-	 get_src_ptr(p, srcEAX, translateESI, x86_deref(idxEBX), j);
-      }
-
-      if (!translate_attr( p, a, x86_deref(srcEAX), destEAX ))
-	 return FALSE;
-   }
-
-   /* Next vertex:
-    */
-   x86_lea(p->func, vertexECX, x86_make_disp(vertexECX, p->translate.key.output_stride));
 
-   /* Incr index
-    */ 
-   if (linear) {
-      x86_inc(p->func, idxEBX);
-   } 
-   else {
-      x86_lea(p->func, idxEBX, x86_make_disp(idxEBX, 4));
+      /* Next output vertex:
+       */
+      x86_lea(p->func, 
+              p->outbuf_ECX, 
+              x86_make_disp(p->outbuf_ECX, 
+                            p->translate.key.output_stride));
+
+      /* Incr index
+       */ 
+      incr_inputs( p, linear );
    }
 
    /* decr count, loop if not zero
     */
-   x86_dec(p->func, countEBP);
-   x86_test(p->func, countEBP, countEBP); 
+   x86_dec(p->func, p->count_ESI);
    x86_jcc(p->func, cc_NZ, label);
 
    /* Exit mmx state?
@@ -493,9 +579,8 @@ static boolean build_vertex_emit( struct translate_sse *p,
    /* Pop regs and return
     */
    
-   x86_pop(p->func, idxEBX);
-   x86_pop(p->func, translateESI);
-   x86_pop(p->func, countEBP);
+   x86_pop(p->func, p->count_ESI);
+   x86_pop(p->func, p->idx_EBX);
    x86_ret(p->func);
 
    return TRUE;
@@ -513,15 +598,16 @@ static void translate_sse_set_buffer( struct translate *translate,
 				unsigned stride )
 {
    struct translate_sse *p = (struct translate_sse *)translate;
-   unsigned i;
 
-   for (i = 0; i < p->translate.key.nr_elements; i++) {
-      if (p->translate.key.element[i].input_buffer == buf) {
-	 p->attrib[i].input_ptr = ((char *)ptr +
-				    p->translate.key.element[i].input_offset);
-	 p->attrib[i].input_stride = stride;
-      }
+   if (buf < p->nr_buffers) {
+      p->buffer[buf].base_ptr = (char *)ptr;
+      p->buffer[buf].stride = stride;
    }
+
+   if (0) debug_printf("%s %d/%d: %p %d\n", 
+                       __FUNCTION__, buf, 
+                       p->nr_buffers, 
+                       ptr, stride);
 }
 
 
@@ -565,6 +651,7 @@ static void PIPE_CDECL translate_sse_run( struct translate *translate,
 struct translate *translate_sse2_create( const struct translate_key *key )
 {
    struct translate_sse *p = NULL;
+   unsigned i;
 
    if (!rtasm_cpu_has_sse() || !rtasm_cpu_has_sse2())
       goto fail;
@@ -579,6 +666,11 @@ struct translate *translate_sse2_create( const struct translate_key *key )
    p->translate.run_elts = translate_sse_run_elts;
    p->translate.run = translate_sse_run;
 
+   for (i = 0; i < key->nr_elements; i++) 
+      p->nr_buffers = MAX2( p->nr_buffers, key->element[i].input_buffer + 1 );
+
+   if (0) debug_printf("nr_buffers: %d\n", p->nr_buffers);
+
    if (!build_vertex_emit(p, &p->linear_func, TRUE))
       goto fail;
 
diff --git a/src/gallium/auxiliary/util/u_math.c b/src/gallium/auxiliary/util/u_math.c
index 9c5f616ceb3..2811475fa00 100644
--- a/src/gallium/auxiliary/util/u_math.c
+++ b/src/gallium/auxiliary/util/u_math.c
@@ -30,7 +30,7 @@
 #include "util/u_math.h"
 
 
-/** 2^x, for x in [-1.0, 1.0] */
+/** 2^x, for x in [-1.0, 1.0) */
 float pow2_table[POW2_TABLE_SIZE];
 
 
@@ -43,7 +43,7 @@ init_pow2_table(void)
 }
 
 
-/** log2(x), for x in [1.0, 2.0] */
+/** log2(x), for x in [1.0, 2.0) */
 float log2_table[LOG2_TABLE_SIZE];
 
 
@@ -52,7 +52,7 @@ init_log2_table(void)
 {
    unsigned i;
    for (i = 0; i < LOG2_TABLE_SIZE; i++)
-      log2_table[i] = (float) log2(1.0 + i * (1.0 / LOG2_TABLE_SIZE));
+      log2_table[i] = (float) log2(1.0 + i * (1.0 / LOG2_TABLE_SCALE));
 }
 
 
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index fdaec8df823..653d79ccd5b 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -246,8 +246,9 @@ util_fast_exp(float x)
 }
 
 
-#define LOG2_TABLE_SIZE_LOG2 8
-#define LOG2_TABLE_SIZE (1 << LOG2_TABLE_SIZE_LOG2)
+#define LOG2_TABLE_SIZE_LOG2 16
+#define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)
+#define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)
 extern float log2_table[LOG2_TABLE_SIZE];
 
 
@@ -258,7 +259,8 @@ util_fast_log2(float x)
    float epart, mpart;
    num.f = x;
    epart = (float)(((num.i & 0x7f800000) >> 23) - 127);
-   mpart = log2_table[(num.i & 0x007fffff) >> (23 - LOG2_TABLE_SIZE_LOG2)];
+   /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */
+   mpart = log2_table[((num.i & 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2))) >> (23 - LOG2_TABLE_SIZE_LOG2)];
    return epart + mpart;
 }