Merge remote branch 'upstream/gallium-0.1' into nouveau-gallium-0.1

55 files changed, 4319 insertions, 592 deletions

diff --git a/progs/tests/Makefile b/progs/tests/Makefile
index 7053ebc86aa..ea34a708553 100644
--- a/progs/tests/Makefile
+++ b/progs/tests/Makefile
@@ -51,6 +51,7 @@ SOURCES = \
 	manytex.c \
 	minmag.c \
 	mipmap_limits.c \
+	mipmap_view.c \
 	multipal.c \
 	no_s3tc.c \
 	packedpixels.c \
@@ -158,6 +159,14 @@ invert: invert.o readtex.o
 invert.o: invert.c readtex.h
 	$(CC) -c $(INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
 
+mipmap_view: mipmap_view.o readtex.o
+	$(CC) $(CFLAGS) mipmap_view.o readtex.o $(LIBS) -o $@
+
+mipmap_view.o: mipmap_view.c readtex.h
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
+
+
+
 readtex.o: readtex.c
 	$(CC) -c $(INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
 
diff --git a/progs/tests/mipmap_view.c b/progs/tests/mipmap_view.c
new file mode 100644
index 00000000000..d821f432f0b
--- /dev/null
+++ b/progs/tests/mipmap_view.c
@@ -0,0 +1,241 @@
+/*
+ * Test mipmap generation and lod bias.
+ *
+ * Brian Paul
+ * 17 March 2008
+ */
+
+
+#include <assert.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <math.h>
+#include <GL/glut.h>
+#include <GL/glext.h>
+
+#include "readtex.h"
+
+#define TEXTURE_FILE "../images/arch.rgb"
+
+static int TexWidth = 256, TexHeight = 256;
+static int WinWidth = 1044, WinHeight = 900;
+static GLfloat Bias = 0.0;
+static GLboolean ScaleQuads = GL_FALSE;
+
+
+static void
+PrintString(const char *s)
+{
+   while (*s) {
+      glutBitmapCharacter(GLUT_BITMAP_8_BY_13, (int) *s);
+      s++;
+   }
+}
+
+
+static void
+Display(void)
+{
+   int x, y, bias;
+   char str[100];
+   int texWidth = TexWidth, texHeight = TexHeight;
+
+   glClear(GL_COLOR_BUFFER_BIT);
+
+   glMatrixMode(GL_PROJECTION);
+   glLoadIdentity();
+   glOrtho(0, WinWidth, 0, WinHeight, -1, 1);
+   glMatrixMode(GL_MODELVIEW);
+   glLoadIdentity();
+   
+   glColor3f(1,1,1);
+
+   y = WinHeight - 300;
+   x = 4;
+
+   for (bias = -1; bias < 11; bias++) {
+
+      glRasterPos2f(x, y + TexHeight + 5);
+      sprintf(str, "Texture LOD Bias = %d", bias);
+      PrintString(str);
+
+      glPushMatrix();
+      glTranslatef(x, y, 0);
+
+      glEnable(GL_TEXTURE_2D);
+
+      if (ScaleQuads) {
+         if (bias > 0) {
+            texWidth = TexWidth >> bias;
+            texHeight = TexHeight >> bias;
+            if (texWidth < 1)
+               texWidth = 1;
+            if (texHeight < 1)
+               texHeight = 1;
+         }
+         glTexEnvf(GL_TEXTURE_FILTER_CONTROL_EXT, GL_TEXTURE_LOD_BIAS_EXT, 0.0);
+      }
+      else {
+         glTexEnvf(GL_TEXTURE_FILTER_CONTROL_EXT, GL_TEXTURE_LOD_BIAS_EXT, bias);
+      }
+
+      glBegin(GL_POLYGON);
+      glTexCoord2f(0, 0);  glVertex2f(0, 0);
+      glTexCoord2f(1, 0);  glVertex2f(texWidth, 0);
+      glTexCoord2f(1, 1);  glVertex2f(texWidth, texHeight);
+      glTexCoord2f(0, 1);  glVertex2f(0, texHeight);
+      glEnd();
+
+      glPopMatrix();
+
+      glDisable(GL_TEXTURE_2D);
+
+      x += TexWidth + 4;
+      if (x >= WinWidth) {
+         x = 4;
+         y -= 300;
+      }
+   }
+
+   glutSwapBuffers();
+}
+
+
+static void
+Reshape(int width, int height)
+{
+   WinWidth = width;
+   WinHeight = height;
+   glViewport(0, 0, width, height);
+}
+
+
+static void
+Key(unsigned char key, int x, int y)
+{
+   (void) x;
+   (void) y;
+   switch (key) {
+      case 'b':
+         Bias -= 10;
+         break;
+      case 'B':
+         Bias += 10;
+         break;
+      case '0':
+      case '1':
+      case '2':
+      case '3':
+      case '4':
+      case '5':
+      case '6':
+      case '7':
+      case '8':
+      case '9':
+         Bias = 100.0 * (key - '0');
+         break;
+      case 's':
+         ScaleQuads = !ScaleQuads;
+         break;
+      case 27:
+         exit(0);
+         break;
+   }
+   glutPostRedisplay();
+}
+
+
+static void
+Init(void)
+{
+   GLfloat maxBias;
+
+   if (!glutExtensionSupported("GL_EXT_texture_lod_bias")) {
+      printf("Sorry, GL_EXT_texture_lod_bias not supported by this renderer.\n");
+      exit(1);
+   }
+
+   if (!glutExtensionSupported("GL_SGIS_generate_mipmap")) {
+      printf("Sorry, GL_SGIS_generate_mipmap not supported by this renderer.\n");
+      exit(1);
+   }
+
+   glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+
+   if (1) {
+      /* test auto mipmap generation */
+      GLint width, height, i;
+      GLenum format;
+      GLubyte *image = LoadRGBImage(TEXTURE_FILE, &width, &height, &format);
+      if (!image) {
+         printf("Error: could not load texture image %s\n", TEXTURE_FILE);
+         exit(1);
+      }
+      /* resize to TexWidth x TexHeight */
+      if (width != TexWidth || height != TexHeight) {
+         GLubyte *newImage = malloc(TexWidth * TexHeight * 4);
+         gluScaleImage(format, width, height, GL_UNSIGNED_BYTE, image,
+                       TexWidth, TexHeight, GL_UNSIGNED_BYTE, newImage);
+         free(image);
+         image = newImage;
+      }
+      printf("Using GL_SGIS_generate_mipmap\n");
+      glTexParameteri(GL_TEXTURE_2D, GL_GENERATE_MIPMAP_SGIS, GL_TRUE);
+      glTexImage2D(GL_TEXTURE_2D, 0, format, TexWidth, TexHeight, 0,
+                   format, GL_UNSIGNED_BYTE, image);
+      free(image);
+
+      /* make sure mipmap was really generated correctly */
+      width = TexWidth;
+      height = TexHeight;
+      for (i = 0; i < 9; i++) {
+         GLint w, h;
+         glGetTexLevelParameteriv(GL_TEXTURE_2D, i, GL_TEXTURE_WIDTH, &w);
+         glGetTexLevelParameteriv(GL_TEXTURE_2D, i, GL_TEXTURE_HEIGHT, &h);
+         printf("Level %d size: %d x %d\n", i, w, h);
+         assert(w == width);
+         assert(h == height);
+         width /= 2;
+         height /= 2;
+      }
+   }
+   else {
+      if (LoadRGBMipmaps(TEXTURE_FILE, GL_RGB)) {
+         printf("Using gluBuildMipmaps()\n");
+      }
+      else {
+         printf("Error: could not load texture image %s\n", TEXTURE_FILE);
+         exit(1);
+      }
+   }
+
+
+   /* mipmapping required for this extension */
+   glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
+   glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
+   glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);
+
+   glGetFloatv(GL_MAX_TEXTURE_LOD_BIAS_EXT, &maxBias);
+
+   printf("GL_RENDERER: %s\n", (char*) glGetString(GL_RENDERER));
+   printf("LOD bias range: [%g, %g]\n", -maxBias, maxBias);
+
+   printf("Press 's' to toggle quad scaling\n");
+}
+
+
+int
+main(int argc, char *argv[])
+{
+   glutInit(&argc, argv);
+   glutInitWindowPosition(0, 0);
+   glutInitWindowSize(WinWidth, WinHeight);
+   glutInitDisplayMode(GLUT_RGB | GLUT_DOUBLE);
+   glutCreateWindow(argv[0]);
+   glutReshapeFunc(Reshape);
+   glutKeyboardFunc(Key);
+   glutDisplayFunc(Display);
+   Init();
+   glutMainLoop();
+   return 0;
+}
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index fd86bfaca9c..294ac82281e 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -149,6 +149,12 @@ void cso_set_blend(struct cso_context *ctx,
    }
 }
 
+void cso_unset_blend(struct cso_context *ctx)
+{
+   ctx->blend = NULL;
+}
+
+
 void cso_single_sampler(struct cso_context *ctx,
                         unsigned idx,
                         const struct pipe_sampler_state *templ)
@@ -220,6 +226,15 @@ void cso_set_samplers( struct cso_context *ctx,
    cso_single_sampler_done( ctx );
 }
 
+void cso_unset_samplers( struct cso_context *ctx )
+{
+   uint i;
+   for (i = 0; i < ctx->nr_samplers; i++)
+      ctx->samplers[i] = NULL;
+}
+
+
+
 void cso_set_depth_stencil_alpha(struct cso_context *ctx,
                                  const struct pipe_depth_stencil_alpha_state *templ)
 {
@@ -252,6 +267,11 @@ void cso_set_depth_stencil_alpha(struct cso_context *ctx,
    }
 }
 
+void cso_unset_depth_stencil_alpha(struct cso_context *ctx)
+{
+   ctx->depth_stencil = NULL;
+}
+
 
 
 void cso_set_rasterizer(struct cso_context *ctx,
@@ -285,7 +305,10 @@ void cso_set_rasterizer(struct cso_context *ctx,
    }
 }
 
-
+void cso_unset_rasterizer(struct cso_context *ctx)
+{
+   ctx->rasterizer = NULL;
+}
 
 
 
@@ -320,6 +343,12 @@ void cso_set_fragment_shader(struct cso_context *ctx,
    }
 }
 
+void cso_unset_fragment_shader(struct cso_context *ctx)
+{
+   ctx->fragment_shader = NULL;
+}
+
+
 void cso_set_vertex_shader(struct cso_context *ctx,
                            const struct pipe_shader_state *templ)
 {
@@ -350,3 +379,8 @@ void cso_set_vertex_shader(struct cso_context *ctx,
       ctx->pipe->bind_vs_state(ctx->pipe, handle);
    }
 }
+
+void cso_unset_vertex_shader(struct cso_context *ctx)
+{
+   ctx->vertex_shader = NULL;
+}
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
index 1f2a6308043..6aa619abf56 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.h
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -44,16 +44,25 @@ struct cso_context *cso_create_context( struct pipe_context *pipe );
 void cso_set_blend( struct cso_context *cso,
                     const struct pipe_blend_state *blend );
 
+void cso_unset_blend(struct cso_context *cso);
+
 void cso_set_depth_stencil_alpha( struct cso_context *cso,
                                   const struct pipe_depth_stencil_alpha_state *dsa );
 
+void cso_unset_depth_stencil_alpha( struct cso_context *cso );
+
 void cso_set_rasterizer( struct cso_context *cso,
                          const struct pipe_rasterizer_state *rasterizer );
 
+void cso_unset_rasterizer( struct cso_context *cso );
+
 void cso_set_samplers( struct cso_context *cso,
                        unsigned count,
                        const struct pipe_sampler_state **states );
 
+void cso_unset_samplers( struct cso_context *cso );
+
+
 /* Alternate interface to support state trackers that like to modify
  * samplers one at a time:
  */
@@ -72,9 +81,13 @@ void cso_single_sampler_done( struct cso_context *cso );
 void cso_set_fragment_shader( struct cso_context *cso,
                               const struct pipe_shader_state *shader );
 
+void cso_unset_fragment_shader( struct cso_context *cso );
+
 void cso_set_vertex_shader( struct cso_context *cso,
                             const struct pipe_shader_state *shader );
 
+void cso_unset_vertex_shader( struct cso_context *cso );
+
 void cso_destroy_context( struct cso_context *cso );
 
 
diff --git a/src/gallium/auxiliary/draw/draw_passthrough.c b/src/gallium/auxiliary/draw/draw_passthrough.c
index d16f056191c..fdec6a591b8 100644
--- a/src/gallium/auxiliary/draw/draw_passthrough.c
+++ b/src/gallium/auxiliary/draw/draw_passthrough.c
@@ -85,14 +85,52 @@ fetch_store_general( struct draw_context *draw,
    const unsigned *pitch   = draw->vertex_fetch.pitch;
    const ubyte **src       = draw->vertex_fetch.src_ptr;
 
-   for (i = start; i < count; i++) {
+   for (i = start; i < start + count; i++) {
       for (j = 0; j < nr_attrs; j++) {
+         /* vinfo->src_index is the output of the vertex shader
+          * matching this hw-vertex component. 
+          *
+          * In passthrough, we require a 1:1 mapping between vertex
+          * shader outputs and inputs, which in turn correspond to
+          * vertex elements in the state.  So, this is the vertex
+          * element we're interested in...
+          */
          const uint jj = vinfo->src_index[j];
          const enum pipe_format srcFormat  = draw->vertex_element[jj].src_format;
          const ubyte *from = src[jj] + i * pitch[jj];
          float attrib[4];
 
+         /* Except...  When we're not.  Two cases EMIT_HEADER &
+          * EMIT_1F_PSIZE don't consume an input.  Should have some
+          * method for indicating this, or change the logic here
+          * somewhat so it doesn't matter.
+          *
+          * Just hack this up now, do something better about it later.
+          */
+         if (vinfo->emit[j] == EMIT_HEADER) {
+            memset(out, 0, sizeof(struct vertex_header));
+            out += sizeof(struct vertex_header) / 4;
+            continue;
+         }
+         else if (vinfo->emit[j] == EMIT_1F_PSIZE) {
+            out[0] = 1.0;       /* xxx */
+            out += 1;
+            continue;
+         }
+         
+
+         /* The normal fetch/emit code:
+          */
          switch (srcFormat) {
+         case PIPE_FORMAT_B8G8R8A8_UNORM:
+            {
+               ubyte *ub = (ubyte *) from;
+               attrib[0] = UBYTE_TO_FLOAT(ub[0]);
+               attrib[1] = UBYTE_TO_FLOAT(ub[1]);
+               attrib[2] = UBYTE_TO_FLOAT(ub[2]);
+               attrib[3] = UBYTE_TO_FLOAT(ub[3]);
+            }
+            break;
          case PIPE_FORMAT_R32G32B32A32_FLOAT:
             {
                float *f = (float *) from;
@@ -130,14 +168,21 @@ fetch_store_general( struct draw_context *draw,
             }
             break;
          default:
-            abort();
+            assert(0);
          }
 
-         /* XXX this will probably only work for softpipe */
+         debug_printf("attrib %d: %f %f %f %f\n", j, 
+                      attrib[0], attrib[1], attrib[2], attrib[3]);
+
          switch (vinfo->emit[j]) {
-         case EMIT_HEADER:
-            memset(out, 0, sizeof(struct vertex_header));
-            out += sizeof(struct vertex_header) / 4;
+         case EMIT_1F:
+            out[0] = attrib[0];
+            out += 1;
+            break;
+         case EMIT_2F:
+            out[0] = attrib[0];
+            out[1] = attrib[1];
+            out += 2;
             break;
          case EMIT_4F:
             out[0] = attrib[0];
@@ -147,64 +192,15 @@ fetch_store_general( struct draw_context *draw,
             out += 4;
             break;
          default:
-            abort();
+            assert(0);
          }
-
       }
+      debug_printf("\n");
    }
 }
 
 
 
-/* Example of a fetch/emit passthrough shader which could be
- * generated when bypass_clipping is enabled on a passthrough vertex
- * shader.
- */
-static void fetch_xyz_rgb_st( struct draw_context *draw,
-			      float *out,
-			      unsigned start,
-			      unsigned count )
-{
-   const unsigned *pitch   = draw->vertex_fetch.pitch;
-   const ubyte **src       = draw->vertex_fetch.src_ptr;
-   unsigned i;
-
-   const ubyte *xyzw = src[0] + start * pitch[0];
-   const ubyte *rgba = src[1] + start * pitch[1];
-   const ubyte *st = src[2] + start * pitch[2];
-
-   /* loop over vertex attributes (vertex shader inputs)
-    */
-   for (i = 0; i < count; i++) {
-      {
-	 const float *in = (const float *)xyzw; xyzw += pitch[0];
-         /* decode input, encode output.  Assume both are float[4] */
-	 out[0] = in[0];
-	 out[1] = in[1];
-	 out[2] = in[2];
-	 out[3] = in[3];
-      }
-
-      {
-	 const float *in = (const float *)rgba; rgba += pitch[1];
-         /* decode input, encode output.  Assume both are float[4] */
-	 out[4] = in[0];
-	 out[5] = in[1];
-	 out[6] = in[2];
- 	 out[7] = in[3];
-      }
-
-      {
-	 const float *in = (const float *)st; st += pitch[2];
-         /* decode input, encode output.  Assume both are float[2] */
-	 out[8] = in[0];
-	 out[9] = in[1];
-      }
-
-      out += 10;
-   }
-}
-
 static boolean update_shader( struct draw_context *draw )
 {
    const struct vertex_info *vinfo = draw->render->get_vertex_info(draw->render);
@@ -229,70 +225,166 @@ static boolean update_shader( struct draw_context *draw )
 
    draw->pt.hw_vertex_size = vinfo->size * 4;
 
-   /* Just trying to figure out how this would work:
-    */
-   if (draw->rasterizer->bypass_vs ||
-       (nr_attrs == 3 && 0 /* some other tests */))
-   {
-#if 0
-      draw->vertex_fetch.pt_fetch = fetch_xyz_rgb_st;
-#else
-      draw->vertex_fetch.pt_fetch = fetch_store_general;
-#endif
-      /*assert(vinfo->size == 10);*/
+   draw->vertex_fetch.pt_fetch = fetch_store_general;
+   return TRUE;
+}
+
+
+
+
+static boolean split_prim_inplace(unsigned prim, unsigned *first, unsigned *incr)
+{
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      *first = 1;
+      *incr = 1;
+      return TRUE;
+   case PIPE_PRIM_LINES:
+      *first = 2;
+      *incr = 2;
       return TRUE;
+   case PIPE_PRIM_LINE_STRIP:
+      *first = 2;
+      *incr = 1;
+      return TRUE;
+   case PIPE_PRIM_TRIANGLES:
+      *first = 3;
+      *incr = 3;
+      return TRUE;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      *first = 3;
+      *incr = 1;
+      return TRUE;
+   case PIPE_PRIM_QUADS:
+      *first = 4;
+      *incr = 4;
+      return TRUE;
+   case PIPE_PRIM_QUAD_STRIP:
+      *first = 4;
+      *incr = 2;
+      return TRUE;
+   default:
+      *first = 0;
+      *incr = 1;		/* set to one so that count % incr works */
+      return FALSE;
    }
-   
-   return FALSE;
 }
 
 
 
 static boolean set_prim( struct draw_context *draw,
-		      unsigned prim )
+                         unsigned prim,
+                         unsigned count )
 {
    assert(!draw->user.elts);   
 
-   draw->pt.prim = prim;
-
    switch (prim) { 
    case PIPE_PRIM_LINE_LOOP:
+      if (count > 1024)
+         return FALSE;
+      return draw->render->set_primitive( draw->render, PIPE_PRIM_LINE_STRIP );
+
+   case PIPE_PRIM_TRIANGLE_FAN:
+   case PIPE_PRIM_POLYGON:
+      if (count > 1024)
+         return FALSE;
+      return draw->render->set_primitive( draw->render, prim );
+
    case PIPE_PRIM_QUADS:
    case PIPE_PRIM_QUAD_STRIP:
-      return FALSE;
+      return draw->render->set_primitive( draw->render, PIPE_PRIM_TRIANGLES );
+
    default:
-      draw->render->set_primitive( draw->render, prim );
-      return TRUE;
+      return draw->render->set_primitive( draw->render, prim );
+      break;
    }
+
+   return TRUE;
 }
 
 
-boolean
-draw_passthrough_arrays(struct draw_context *draw, 
-                        unsigned prim,
-                        unsigned start, 
-                        unsigned count)
+
+#define INDEX(i) (start + (i))
+static void pt_draw_arrays( struct draw_context *draw,
+			    unsigned start,
+			    unsigned length )
 {
-   float *hw_verts;
+   ushort *tmp = NULL;
+   unsigned i, j;
 
-   if (draw_need_pipeline(draw))
-      return FALSE;
+   switch (draw->pt.prim) {
+   case PIPE_PRIM_LINE_LOOP:
+      tmp = MALLOC( sizeof(ushort) * (length + 1) );
 
-   if (!set_prim(draw, prim))
-      return FALSE;
+      for (i = 0; i < length; i++)
+	 tmp[i] = INDEX(i);
+      tmp[length] = 0;
 
-   if (!update_shader(draw))
-      return FALSE;
+      draw->render->draw( draw->render,
+			  tmp,
+			  length+1 );
+      break;
 
-   hw_verts = draw->render->allocate_vertices( draw->render,
-                                               draw->pt.hw_vertex_size,
-                                               count );
+
+   case PIPE_PRIM_QUAD_STRIP:
+      tmp = MALLOC( sizeof(ushort) * (length / 2 * 6) );
+
+      for (j = i = 0; i + 3 < length; i += 2, j += 6) {
+	 tmp[j+0] = INDEX(i+0);
+	 tmp[j+1] = INDEX(i+1);
+	 tmp[j+2] = INDEX(i+3);
+
+	 tmp[j+3] = INDEX(i+2);
+	 tmp[j+4] = INDEX(i+0);
+	 tmp[j+5] = INDEX(i+3);
+      }
+
+      if (j)
+	 draw->render->draw( draw->render, tmp, j );
+      break;
+
+   case PIPE_PRIM_QUADS:
+      tmp = MALLOC( sizeof(int) * (length / 4 * 6) );
+
+      for (j = i = 0; i + 3 < length; i += 4, j += 6) {
+	 tmp[j+0] = INDEX(i+0);
+	 tmp[j+1] = INDEX(i+1);
+	 tmp[j+2] = INDEX(i+3);
+
+	 tmp[j+3] = INDEX(i+1);
+	 tmp[j+4] = INDEX(i+2);
+	 tmp[j+5] = INDEX(i+3);
+      }
+
+      if (j)
+	 draw->render->draw( draw->render, tmp, j );
+      break;
+
+   default:
+      draw->render->draw_arrays( draw->render,
+                                 start,
+                                 length );
+      break;
+   }
+
+   if (tmp)
+      FREE(tmp);
+}
+
+
+
+static boolean do_draw( struct draw_context *draw, 
+                        unsigned start, unsigned count )
+{
+   float *hw_verts = 
+      draw->render->allocate_vertices( draw->render,
+                                       (ushort)draw->pt.hw_vertex_size,
+                                       (ushort)count );
 
    if (!hw_verts)
       return FALSE;
 					
-   /* Single routine to fetch vertices, run shader and emit HW verts.
-    * Clipping and viewport transformation are done on hardware.
+   /* Single routine to fetch vertices and emit HW verts.
     */
    draw->vertex_fetch.pt_fetch( draw, 
 				hw_verts,
@@ -301,9 +393,9 @@ draw_passthrough_arrays(struct draw_context *draw,
    /* Draw arrays path to avoid re-emitting index list again and
     * again.
     */
-   draw->render->draw_arrays( draw->render,
-                              start,
-                              count );
+   pt_draw_arrays( draw,
+                   0,
+                   count );
    
 
    draw->render->release_vertices( draw->render, 
@@ -314,3 +406,68 @@ draw_passthrough_arrays(struct draw_context *draw,
    return TRUE;
 }
 
+
+boolean
+draw_passthrough_arrays(struct draw_context *draw, 
+                        unsigned prim,
+                        unsigned start, 
+                        unsigned count)
+{
+   unsigned i = 0;
+   unsigned first, incr;
+   
+   //debug_printf("%s prim %d start %d count %d\n", __FUNCTION__, prim, start, count);
+   
+   split_prim_inplace(prim, &first, &incr);
+
+   count -= (count - first) % incr; 
+
+   debug_printf("%s %d %d %d\n", __FUNCTION__, prim, start, count);
+
+   if (draw_need_pipeline(draw))
+      return FALSE;
+
+   debug_printf("%s AAA\n", __FUNCTION__);
+
+   if (!set_prim(draw, prim, count))
+      return FALSE;
+
+   /* XXX: need a single value that reflects the most recent call to
+    * driver->set_primitive:
+    */
+   draw->pt.prim = prim;
+
+   debug_printf("%s BBB\n", __FUNCTION__);
+
+   if (!update_shader(draw))
+      return FALSE;
+
+   debug_printf("%s CCC\n", __FUNCTION__);
+
+   /* Chop this up into bite-sized pieces that a driver should be able
+    * to devour -- problem is we don't have a quick way to query the
+    * driver on the maximum size for this chunk in the current state.
+    */
+   while (i + first <= count) {
+      int nr = MIN2( count - i, 1024 );
+
+      /* snap to prim boundary 
+       */
+      nr -= (nr - first) % incr; 
+
+      if (!do_draw( draw, start + i, nr )) {
+         assert(0);
+         return FALSE;
+      }
+
+      /* increment allowing for repeated vertices
+       */
+      i += nr - (first - incr);
+   }
+
+
+   debug_printf("%s DONE\n", __FUNCTION__);
+   return TRUE;
+}
+
+
diff --git a/src/gallium/auxiliary/draw/draw_prim.c b/src/gallium/auxiliary/draw/draw_prim.c
index 888fa536ea3..cb0277fb6c9 100644
--- a/src/gallium/auxiliary/draw/draw_prim.c
+++ b/src/gallium/auxiliary/draw/draw_prim.c
@@ -169,11 +169,12 @@ void draw_do_flush( struct draw_context *draw, unsigned flags )
       flushing = TRUE;
 
    if (flags >= DRAW_FLUSH_SHADER_QUEUE) {
-      if (draw->vs.queue_nr)
+      if (draw->vs.queue_nr) {
          if (draw->rasterizer->bypass_vs)
             fetch_and_store(draw);
          else
             (*draw->shader_queue_flush)(draw);
+      }
 
       if (flags >= DRAW_FLUSH_PRIM_QUEUE) {
 	 if (draw->pq.queue_nr)
diff --git a/src/gallium/auxiliary/draw/draw_vbuf.h b/src/gallium/auxiliary/draw/draw_vbuf.h
index 5e7de905c16..e90f37872a1 100644
--- a/src/gallium/auxiliary/draw/draw_vbuf.h
+++ b/src/gallium/auxiliary/draw/draw_vbuf.h
@@ -74,9 +74,11 @@ struct vbuf_render {
 
    /**
     * Notify the renderer of the current primitive when it changes.
-    * Prim is restricted to TRIANGLES, LINES and POINTS.
+    * Must succeed for TRIANGLES, LINES and POINTS.  Other prims at
+    * the discretion of the driver, for the benefit of the passthrough
+    * path.
     */
-   void (*set_primitive)( struct vbuf_render *, unsigned prim );
+   boolean (*set_primitive)( struct vbuf_render *, unsigned prim );
 
    /**
     * DrawElements, note indices are ushort:
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
index a996218ce72..24be65bff9d 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
@@ -316,7 +316,9 @@ void spe_init_func(struct spe_function *p, unsigned code_size)
 
 void spe_release_func(struct spe_function *p)
 {
-    align_free(p->store);
+    if (p->store != NULL) {
+        align_free(p->store);
+    }
     p->store = NULL;
     p->csr = NULL;
 }
@@ -326,8 +328,8 @@ int spe_allocate_available_register(struct spe_function *p)
 {
    unsigned i;
    for (i = 0; i < 128; i++) {
-      const uint64_t mask = (1ULL << (i % 128));
-      const unsigned idx = i / 128;
+      const uint64_t mask = (1ULL << (i % 64));
+      const unsigned idx = i / 64;
 
       if ((p->regs[idx] & mask) != 0) {
          p->regs[idx] &= ~mask;
@@ -341,8 +343,8 @@ int spe_allocate_available_register(struct spe_function *p)
 
 int spe_allocate_register(struct spe_function *p, int reg)
 {
-   const unsigned idx = reg / 128;
-   const unsigned bit = reg % 128;
+   const unsigned idx = reg / 64;
+   const unsigned bit = reg % 64;
 
    assert((p->regs[idx] & (1ULL << bit)) != 0);
 
@@ -353,8 +355,8 @@ int spe_allocate_register(struct spe_function *p, int reg)
 
 void spe_release_register(struct spe_function *p, int reg)
 {
-   const unsigned idx = reg / 128;
-   const unsigned bit = reg % 128;
+   const unsigned idx = reg / 64;
+   const unsigned bit = reg % 64;
 
    assert((p->regs[idx] & (1ULL << bit)) == 0);
 
diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile
index 2016c6fb1f7..9b6c2708b62 100644
--- a/src/gallium/auxiliary/util/Makefile
+++ b/src/gallium/auxiliary/util/Makefile
@@ -7,9 +7,13 @@ C_SOURCES = \
 	p_debug.c \
 	p_tile.c \
 	p_util.c \
+	u_blit.c \
+	u_draw_quad.c \
+	u_gen_mipmap.c \
 	u_handle_table.c \
 	u_hash_table.c \
 	u_mm.c \
+	u_simple_shaders.c \
 	u_snprintf.c
 
 include ../../Makefile.template
diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript
index 154a3eca8cc..b44f2d5e398 100644
--- a/src/gallium/auxiliary/util/SConscript
+++ b/src/gallium/auxiliary/util/SConscript
@@ -6,9 +6,13 @@ util = env.ConvenienceLibrary(
 		'p_debug.c',
 		'p_tile.c',
 		'p_util.c',
+		'u_blit.c',
+		'u_draw_quad.c',
+		'u_gen_mipmap.c',
 		'u_handle_table.c',
 		'u_hash_table.c',
 		'u_mm.c',
+		'u_simple_shaders.c',
 		'u_snprintf.c',
 	])
 
diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c
new file mode 100644
index 00000000000..4b4ab8185f2
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_blit.c
@@ -0,0 +1,276 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Copy/blit pixel rect between surfaces
+ *  
+ * @author Brian Paul
+ */
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "util/u_draw_quad.h"
+#include "util/u_blit.h"
+#include "util/u_simple_shaders.h"
+
+
+struct blit_state
+{
+   struct pipe_context *pipe;
+
+   void *blend;
+   void *depthstencil;
+   void *rasterizer;
+   void *samplers[2];  /* one for linear, one for nearest sampling */
+
+   /*struct pipe_viewport_state viewport;*/
+   struct pipe_sampler_state *vs;
+   struct pipe_sampler_state *fs;
+};
+
+
+/**
+ * Create state object for blit.
+ * Intended to be created once and re-used for many blit() calls.
+ */
+struct blit_state *
+util_create_blit(struct pipe_context *pipe)
+{
+   struct pipe_blend_state blend;
+   struct pipe_depth_stencil_alpha_state depthstencil;
+   struct pipe_rasterizer_state rasterizer;
+   struct blit_state *ctx;
+   struct pipe_sampler_state sampler;
+
+   ctx = CALLOC_STRUCT(blit_state);
+   if (!ctx)
+      return NULL;
+
+   ctx->pipe = pipe;
+
+   /* we don't use blending, but need to set valid values */
+   memset(&blend, 0, sizeof(blend));
+   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.colormask = PIPE_MASK_RGBA;
+   ctx->blend = pipe->create_blend_state(pipe, &blend);
+
+   /* depth/stencil/alpha */
+   memset(&depthstencil, 0, sizeof(depthstencil));
+   ctx->depthstencil = pipe->create_depth_stencil_alpha_state(pipe, &depthstencil);
+
+   /* rasterizer */
+   memset(&rasterizer, 0, sizeof(rasterizer));
+   rasterizer.front_winding = PIPE_WINDING_CW;
+   rasterizer.cull_mode = PIPE_WINDING_NONE;
+   rasterizer.bypass_clipping = 1;  /* bypasses viewport too */
+   /*rasterizer.bypass_vs = 1;*/
+   ctx->rasterizer = pipe->create_rasterizer_state(pipe, &rasterizer);
+
+   /* samplers */
+   memset(&sampler, 0, sizeof(sampler));
+   sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+   sampler.min_img_filter = PIPE_TEX_MIPFILTER_NEAREST;
+   sampler.mag_img_filter = PIPE_TEX_MIPFILTER_NEAREST;
+   sampler.normalized_coords = 1;
+   ctx->samplers[0] = pipe->create_sampler_state(pipe, &sampler);
+
+   sampler.min_img_filter = PIPE_TEX_MIPFILTER_LINEAR;
+   sampler.mag_img_filter = PIPE_TEX_MIPFILTER_LINEAR;
+   ctx->samplers[1] = pipe->create_sampler_state(pipe, &sampler);
+
+
+#if 0
+   /* viewport */
+   ctx->viewport.scale[0] = 1.0;
+   ctx->viewport.scale[1] = 1.0;
+   ctx->viewport.scale[2] = 1.0;
+   ctx->viewport.scale[3] = 1.0;
+   ctx->viewport.translate[0] = 0.0;
+   ctx->viewport.translate[1] = 0.0;
+   ctx->viewport.translate[2] = 0.0;
+   ctx->viewport.translate[3] = 0.0;
+#endif
+
+   /* vertex shader */
+   {
+      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
+                                      TGSI_SEMANTIC_GENERIC };
+      const uint semantic_indexes[] = { 0, 0 };
+      ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
+                                                    semantic_indexes);
+   }
+
+   /* fragment shader */
+   ctx->fs = util_make_fragment_tex_shader(pipe);
+
+   return ctx;
+}
+
+
+/**
+ * Destroy a blit context
+ */
+void
+util_destroy_blit(struct blit_state *ctx)
+{
+   struct pipe_context *pipe = ctx->pipe;
+
+   pipe->delete_blend_state(pipe, ctx->blend);
+   pipe->delete_depth_stencil_alpha_state(pipe, ctx->depthstencil);
+   pipe->delete_rasterizer_state(pipe, ctx->rasterizer);
+   pipe->delete_sampler_state(pipe, ctx->samplers[0]);
+   pipe->delete_sampler_state(pipe, ctx->samplers[1]);
+
+   pipe->delete_vs_state(pipe, ctx->vs);
+   pipe->delete_fs_state(pipe, ctx->fs);
+
+   FREE(ctx);
+}
+
+
+/**
+ * Copy pixel block from src surface to dst surface.
+ * Overlapping regions are acceptable.
+ * XXX need some control over blitting Z and/or stencil.
+ */
+void
+util_blit_pixels(struct blit_state *ctx,
+                 struct pipe_surface *src,
+                 int srcX0, int srcY0,
+                 int srcX1, int srcY1,
+                 struct pipe_surface *dst,
+                 int dstX0, int dstY0,
+                 int dstX1, int dstY1,
+                 float z, uint filter)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_texture texTemp, *tex;
+   struct pipe_surface *texSurf;
+   struct pipe_framebuffer_state fb;
+   const int srcW = abs(srcX1 - srcX0);
+   const int srcH = abs(srcY1 - srcY0);
+   const int srcLeft = MIN2(srcX0, srcX1);
+   const int srcTop = MIN2(srcY0, srcY1);
+
+   assert(filter == PIPE_TEX_MIPFILTER_NEAREST ||
+          filter == PIPE_TEX_MIPFILTER_LINEAR);
+
+   if (srcLeft != srcX0) {
+      /* left-right flip */
+      int tmp = dstX0;
+      dstX0 = dstX1;
+      dstX1 = tmp;
+   }
+
+   if (srcTop != srcY0) {
+      /* up-down flip */
+      int tmp = dstY0;
+      dstY0 = dstY1;
+      dstY1 = tmp;
+   }
+
+   /*
+    * XXX for now we're always creating a temporary texture.
+    * Strictly speaking that's not always needed.
+    */
+
+   /* create temp texture */
+   memset(&texTemp, 0, sizeof(texTemp));
+   texTemp.target = PIPE_TEXTURE_2D;
+   texTemp.format = src->format;
+   texTemp.last_level = 0;
+   texTemp.width[0] = srcW;
+   texTemp.height[0] = srcH;
+   texTemp.depth[0] = 1;
+   texTemp.compressed = 0;
+   texTemp.cpp = pf_get_bits(src->format) / 8;
+
+   tex = screen->texture_create(screen, &texTemp);
+   if (!tex)
+      return;
+
+   texSurf = screen->get_tex_surface(screen, tex, 0, 0, 0);
+
+   /* load temp texture */
+   pipe->surface_copy(pipe, FALSE,
+                      texSurf, 0, 0,   /* dest */
+                      src, srcLeft, srcTop, /* src */
+                      srcW, srcH);     /* size */
+
+   /* drawing dest */
+   memset(&fb, 0, sizeof(fb));
+   fb.num_cbufs = 1;
+   fb.cbufs[0] = dst;
+   pipe->set_framebuffer_state(pipe, &fb);
+
+   /* sampler */
+   if (filter == PIPE_TEX_MIPFILTER_NEAREST)
+      pipe->bind_sampler_states(pipe, 1, &ctx->samplers[0]);
+   else
+      pipe->bind_sampler_states(pipe, 1, &ctx->samplers[1]);
+
+   /* texture */
+   pipe->set_sampler_textures(pipe, 1, &tex);
+
+   /* shaders */
+   pipe->bind_fs_state(pipe, ctx->fs);
+   pipe->bind_vs_state(pipe, ctx->vs);
+
+   /* misc state */
+   pipe->bind_blend_state(pipe, ctx->blend);
+   pipe->bind_depth_stencil_alpha_state(pipe, ctx->depthstencil);
+   pipe->bind_rasterizer_state(pipe, ctx->rasterizer);
+
+   /* draw quad */
+   util_draw_texquad(pipe, dstX0, dstY0, dstX1, dstY1, z);
+
+   /* unbind */
+   pipe->set_sampler_textures(pipe, 0, NULL);
+   pipe->bind_sampler_states(pipe, 0, NULL);
+
+   /* free stuff */
+   pipe_surface_reference(&texSurf, NULL);
+   screen->texture_release(screen, &tex);
+
+   /* Note: caller must restore pipe/gallium state at this time */
+}
+
diff --git a/src/gallium/auxiliary/util/u_blit.h b/src/gallium/auxiliary/util/u_blit.h
new file mode 100644
index 00000000000..a349be99ad8
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_blit.h
@@ -0,0 +1,60 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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 U_BLIT_H
+#define U_BLIT_H
+
+
+struct pipe_context;
+struct pipe_surface;
+
+
+struct blit_state;
+
+
+extern struct blit_state *
+util_create_blit(struct pipe_context *pipe);
+
+
+extern void
+util_destroy_blit(struct blit_state *ctx);
+
+
+
+extern void
+util_blit_pixels(struct blit_state *ctx,
+                 struct pipe_surface *src,
+                 int srcX0, int srcY0,
+                 int srcX1, int srcY1,
+                 struct pipe_surface *dst,
+                 int dstX0, int dstY0,
+                 int dstX1, int dstY1,
+                 float z, uint filter);
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_draw_quad.c b/src/gallium/auxiliary/util/u_draw_quad.c
new file mode 100644
index 00000000000..79a69de633d
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_draw_quad.c
@@ -0,0 +1,112 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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 "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_winsys.h"
+#include "util/u_draw_quad.h"
+
+
+/**
+ * Draw screen-aligned textured quad.
+ */
+void 
+util_draw_texquad(struct pipe_context *pipe,
+                  float x0, float y0, float x1, float y1, float z)
+{
+   struct pipe_buffer *vbuf;
+   struct pipe_vertex_buffer vbuffer;
+   struct pipe_vertex_element velement;
+   uint numAttribs = 2, vertexBytes, i, j;
+   float *v;
+
+   vertexBytes = 4 * (4 * numAttribs * sizeof(float));
+
+   /* XXX create one-time */
+   vbuf = pipe->winsys->buffer_create(pipe->winsys, 32,
+                                      PIPE_BUFFER_USAGE_VERTEX, vertexBytes);
+   assert(vbuf);
+
+   v = (float *) pipe->winsys->buffer_map(pipe->winsys, vbuf,
+                                          PIPE_BUFFER_USAGE_CPU_WRITE);
+
+   /*
+    * Load vertex buffer
+    */
+   for (i = j = 0; i < 4; i++) {
+      v[j + 2] = z;   /* z */
+      v[j + 3] = 1.0; /* w */
+      v[j + 6] = 0.0; /* r */
+      v[j + 7] = 1.0; /* q */
+      j += 8;
+   }
+
+   v[0] = x0;
+   v[1] = y0;
+   v[4] = 0.0; /*s*/
+   v[5] = 0.0; /*t*/
+
+   v[8] = x1;
+   v[9] = y0;
+   v[12] = 1.0;
+   v[13] = 0.0;
+
+   v[16] = x1;
+   v[17] = y1;
+   v[20] = 1.0;
+   v[21] = 1.0;
+
+   v[24] = x0;
+   v[25] = y1;
+   v[28] = 0.0;
+   v[29] = 1.0;
+
+   pipe->winsys->buffer_unmap(pipe->winsys, vbuf);
+
+   /* tell pipe about the vertex buffer */
+   vbuffer.buffer = vbuf;
+   vbuffer.pitch = numAttribs * 4 * sizeof(float);  /* vertex size */
+   vbuffer.buffer_offset = 0;
+   pipe->set_vertex_buffer(pipe, 0, &vbuffer);
+
+   /* tell pipe about the vertex attributes */
+   for (i = 0; i < numAttribs; i++) {
+      velement.src_offset = i * 4 * sizeof(float);
+      velement.vertex_buffer_index = 0;
+      velement.src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+      velement.nr_components = 4;
+      pipe->set_vertex_element(pipe, i, &velement);
+   }
+
+   /* draw */
+   pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLE_FAN, 0, 4);
+
+   /* XXX: do one-time */
+   pipe_buffer_reference(pipe->winsys, &vbuf, NULL);
+}
diff --git a/src/gallium/auxiliary/util/u_draw_quad.h b/src/gallium/auxiliary/util/u_draw_quad.h
new file mode 100644
index 00000000000..a97f55d2efd
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_draw_quad.h
@@ -0,0 +1,37 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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 U_DRAWQUAD_H
+#define U_DRAWQUAD_H
+
+
+extern void 
+util_draw_texquad(struct pipe_context *pipe,
+                  float x0, float y0, float x1, float y1, float z);
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c
new file mode 100644
index 00000000000..e18f8ab72ac
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -0,0 +1,877 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Mipmap generation utility
+ *  
+ * @author Brian Paul
+ */
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "util/u_draw_quad.h"
+#include "util/u_gen_mipmap.h"
+#include "util/u_simple_shaders.h"
+
+#include "tgsi/util/tgsi_build.h"
+#include "tgsi/util/tgsi_dump.h"
+#include "tgsi/util/tgsi_parse.h"
+
+
+struct gen_mipmap_state
+{
+   struct pipe_context *pipe;
+
+   void *blend;
+   void *depthstencil;
+   void *rasterizer;
+   /*struct pipe_viewport_state viewport;*/
+   struct pipe_sampler_state *vs;
+   struct pipe_sampler_state *fs;
+};
+
+
+
+enum dtype
+{
+   UBYTE,
+   UBYTE_3_3_2,
+   USHORT,
+   USHORT_4_4_4_4,
+   USHORT_5_6_5,
+   USHORT_1_5_5_5_REV,
+   UINT,
+   FLOAT,
+   HALF_FLOAT
+};
+
+
+typedef ushort half_float;
+
+
+#if 0
+extern half_float
+float_to_half(float f);
+
+extern float
+half_to_float(half_float h);
+#endif
+
+
+/**
+ * Average together two rows of a source image to produce a single new
+ * row in the dest image.  It's legal for the two source rows to point
+ * to the same data.  The source width must be equal to either the
+ * dest width or two times the dest width.
+ * \param datatype  GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
+ * \param comps  number of components per pixel (1..4)
+ */
+static void
+do_row(enum dtype datatype, uint comps, int srcWidth,
+       const void *srcRowA, const void *srcRowB,
+       int dstWidth, void *dstRow)
+{
+   const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
+   const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
+
+   assert(comps >= 1);
+   assert(comps <= 4);
+
+   /* This assertion is no longer valid with non-power-of-2 textures
+   assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
+   */
+
+   if (datatype == UBYTE && comps == 4) {
+      uint i, j, k;
+      const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA;
+      const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB;
+      ubyte(*dst)[4] = (ubyte(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
+      }
+   }
+   else if (datatype == UBYTE && comps == 3) {
+      uint i, j, k;
+      const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA;
+      const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB;
+      ubyte(*dst)[3] = (ubyte(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+      }
+   }
+   else if (datatype == UBYTE && comps == 2) {
+      uint i, j, k;
+      const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA;
+      const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB;
+      ubyte(*dst)[2] = (ubyte(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
+      }
+   }
+   else if (datatype == UBYTE && comps == 1) {
+      uint i, j, k;
+      const ubyte *rowA = (const ubyte *) srcRowA;
+      const ubyte *rowB = (const ubyte *) srcRowB;
+      ubyte *dst = (ubyte *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
+      }
+   }
+
+   else if (datatype == USHORT && comps == 4) {
+      uint i, j, k;
+      const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA;
+      const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB;
+      ushort(*dst)[4] = (ushort(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
+      }
+   }
+   else if (datatype == USHORT && comps == 3) {
+      uint i, j, k;
+      const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA;
+      const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB;
+      ushort(*dst)[3] = (ushort(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+      }
+   }
+   else if (datatype == USHORT && comps == 2) {
+      uint i, j, k;
+      const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA;
+      const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB;
+      ushort(*dst)[2] = (ushort(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+      }
+   }
+   else if (datatype == USHORT && comps == 1) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
+      }
+   }
+
+   else if (datatype == FLOAT && comps == 4) {
+      uint i, j, k;
+      const float(*rowA)[4] = (const float(*)[4]) srcRowA;
+      const float(*rowB)[4] = (const float(*)[4]) srcRowB;
+      float(*dst)[4] = (float(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                      rowB[j][0] + rowB[k][0]) * 0.25F;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                      rowB[j][1] + rowB[k][1]) * 0.25F;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] +
+                      rowB[j][2] + rowB[k][2]) * 0.25F;
+         dst[i][3] = (rowA[j][3] + rowA[k][3] +
+                      rowB[j][3] + rowB[k][3]) * 0.25F;
+      }
+   }
+   else if (datatype == FLOAT && comps == 3) {
+      uint i, j, k;
+      const float(*rowA)[3] = (const float(*)[3]) srcRowA;
+      const float(*rowB)[3] = (const float(*)[3]) srcRowB;
+      float(*dst)[3] = (float(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                      rowB[j][0] + rowB[k][0]) * 0.25F;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                      rowB[j][1] + rowB[k][1]) * 0.25F;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] +
+                      rowB[j][2] + rowB[k][2]) * 0.25F;
+      }
+   }
+   else if (datatype == FLOAT && comps == 2) {
+      uint i, j, k;
+      const float(*rowA)[2] = (const float(*)[2]) srcRowA;
+      const float(*rowB)[2] = (const float(*)[2]) srcRowB;
+      float(*dst)[2] = (float(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                      rowB[j][0] + rowB[k][0]) * 0.25F;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                      rowB[j][1] + rowB[k][1]) * 0.25F;
+      }
+   }
+   else if (datatype == FLOAT && comps == 1) {
+      uint i, j, k;
+      const float *rowA = (const float *) srcRowA;
+      const float *rowB = (const float *) srcRowB;
+      float *dst = (float *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
+      }
+   }
+
+#if 0
+   else if (datatype == HALF_FLOAT && comps == 4) {
+      uint i, j, k, comp;
+      const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA;
+      const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB;
+      half_float(*dst)[4] = (half_float(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         for (comp = 0; comp < 4; comp++) {
+            float aj, ak, bj, bk;
+            aj = half_to_float(rowA[j][comp]);
+            ak = half_to_float(rowA[k][comp]);
+            bj = half_to_float(rowB[j][comp]);
+            bk = half_to_float(rowB[k][comp]);
+            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+         }
+      }
+   }
+   else if (datatype == HALF_FLOAT && comps == 3) {
+      uint i, j, k, comp;
+      const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA;
+      const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB;
+      half_float(*dst)[3] = (half_float(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         for (comp = 0; comp < 3; comp++) {
+            float aj, ak, bj, bk;
+            aj = half_to_float(rowA[j][comp]);
+            ak = half_to_float(rowA[k][comp]);
+            bj = half_to_float(rowB[j][comp]);
+            bk = half_to_float(rowB[k][comp]);
+            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+         }
+      }
+   }
+   else if (datatype == HALF_FLOAT && comps == 2) {
+      uint i, j, k, comp;
+      const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA;
+      const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB;
+      half_float(*dst)[2] = (half_float(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         for (comp = 0; comp < 2; comp++) {
+            float aj, ak, bj, bk;
+            aj = half_to_float(rowA[j][comp]);
+            ak = half_to_float(rowA[k][comp]);
+            bj = half_to_float(rowB[j][comp]);
+            bk = half_to_float(rowB[k][comp]);
+            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+         }
+      }
+   }
+   else if (datatype == HALF_FLOAT && comps == 1) {
+      uint i, j, k;
+      const half_float *rowA = (const half_float *) srcRowA;
+      const half_float *rowB = (const half_float *) srcRowB;
+      half_float *dst = (half_float *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         float aj, ak, bj, bk;
+         aj = half_to_float(rowA[j]);
+         ak = half_to_float(rowA[k]);
+         bj = half_to_float(rowB[j]);
+         bk = half_to_float(rowB[k]);
+         dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F);
+      }
+   }
+#endif
+
+   else if (datatype == UINT && comps == 1) {
+      uint i, j, k;
+      const uint *rowA = (const uint *) srcRowA;
+      const uint *rowB = (const uint *) srcRowB;
+      uint *dst = (uint *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
+      }
+   }
+
+   else if (datatype == USHORT_5_6_5 && comps == 3) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x1f;
+         const int rowAr1 = rowA[k] & 0x1f;
+         const int rowBr0 = rowB[j] & 0x1f;
+         const int rowBr1 = rowB[k] & 0x1f;
+         const int rowAg0 = (rowA[j] >> 5) & 0x3f;
+         const int rowAg1 = (rowA[k] >> 5) & 0x3f;
+         const int rowBg0 = (rowB[j] >> 5) & 0x3f;
+         const int rowBg1 = (rowB[k] >> 5) & 0x3f;
+         const int rowAb0 = (rowA[j] >> 11) & 0x1f;
+         const int rowAb1 = (rowA[k] >> 11) & 0x1f;
+         const int rowBb0 = (rowB[j] >> 11) & 0x1f;
+         const int rowBb1 = (rowB[k] >> 11) & 0x1f;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         dst[i] = (blue << 11) | (green << 5) | red;
+      }
+   }
+   else if (datatype == USHORT_4_4_4_4 && comps == 4) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0xf;
+         const int rowAr1 = rowA[k] & 0xf;
+         const int rowBr0 = rowB[j] & 0xf;
+         const int rowBr1 = rowB[k] & 0xf;
+         const int rowAg0 = (rowA[j] >> 4) & 0xf;
+         const int rowAg1 = (rowA[k] >> 4) & 0xf;
+         const int rowBg0 = (rowB[j] >> 4) & 0xf;
+         const int rowBg1 = (rowB[k] >> 4) & 0xf;
+         const int rowAb0 = (rowA[j] >> 8) & 0xf;
+         const int rowAb1 = (rowA[k] >> 8) & 0xf;
+         const int rowBb0 = (rowB[j] >> 8) & 0xf;
+         const int rowBb1 = (rowB[k] >> 8) & 0xf;
+         const int rowAa0 = (rowA[j] >> 12) & 0xf;
+         const int rowAa1 = (rowA[k] >> 12) & 0xf;
+         const int rowBa0 = (rowB[j] >> 12) & 0xf;
+         const int rowBa1 = (rowB[k] >> 12) & 0xf;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
+         dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
+      }
+   }
+   else if (datatype == USHORT_1_5_5_5_REV && comps == 4) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x1f;
+         const int rowAr1 = rowA[k] & 0x1f;
+         const int rowBr0 = rowB[j] & 0x1f;
+         const int rowBr1 = rowB[k] & 0xf;
+         const int rowAg0 = (rowA[j] >> 5) & 0x1f;
+         const int rowAg1 = (rowA[k] >> 5) & 0x1f;
+         const int rowBg0 = (rowB[j] >> 5) & 0x1f;
+         const int rowBg1 = (rowB[k] >> 5) & 0x1f;
+         const int rowAb0 = (rowA[j] >> 10) & 0x1f;
+         const int rowAb1 = (rowA[k] >> 10) & 0x1f;
+         const int rowBb0 = (rowB[j] >> 10) & 0x1f;
+         const int rowBb1 = (rowB[k] >> 10) & 0x1f;
+         const int rowAa0 = (rowA[j] >> 15) & 0x1;
+         const int rowAa1 = (rowA[k] >> 15) & 0x1;
+         const int rowBa0 = (rowB[j] >> 15) & 0x1;
+         const int rowBa1 = (rowB[k] >> 15) & 0x1;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
+         dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
+      }
+   }
+   else if (datatype == UBYTE_3_3_2 && comps == 3) {
+      uint i, j, k;
+      const ubyte *rowA = (const ubyte *) srcRowA;
+      const ubyte *rowB = (const ubyte *) srcRowB;
+      ubyte *dst = (ubyte *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x3;
+         const int rowAr1 = rowA[k] & 0x3;
+         const int rowBr0 = rowB[j] & 0x3;
+         const int rowBr1 = rowB[k] & 0x3;
+         const int rowAg0 = (rowA[j] >> 2) & 0x7;
+         const int rowAg1 = (rowA[k] >> 2) & 0x7;
+         const int rowBg0 = (rowB[j] >> 2) & 0x7;
+         const int rowBg1 = (rowB[k] >> 2) & 0x7;
+         const int rowAb0 = (rowA[j] >> 5) & 0x7;
+         const int rowAb1 = (rowA[k] >> 5) & 0x7;
+         const int rowBb0 = (rowB[j] >> 5) & 0x7;
+         const int rowBb1 = (rowB[k] >> 5) & 0x7;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         dst[i] = (blue << 5) | (green << 2) | red;
+      }
+   }
+   else {
+      debug_printf("bad format in do_row()");
+   }
+}
+
+
+static void
+format_to_type_comps(enum pipe_format pformat,
+                     enum dtype *datatype, uint *comps)
+{
+   switch (pformat) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      *datatype = UBYTE;
+      *comps = 4;
+      return;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      *datatype = USHORT_1_5_5_5_REV;
+      *comps = 4;
+      return;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      *datatype = USHORT_4_4_4_4;
+      *comps = 4;
+      return;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      *datatype = USHORT_5_6_5;
+      *comps = 3;
+      return;
+   case PIPE_FORMAT_U_L8:
+   case PIPE_FORMAT_U_A8:
+   case PIPE_FORMAT_U_I8:
+      *datatype = UBYTE;
+      *comps = 1;
+      return;
+   case PIPE_FORMAT_U_A8_L8:
+      *datatype = UBYTE;
+      *comps = 2;
+      return;
+   default:
+      assert(0);
+   }
+}
+
+
+static void
+reduce_1d(enum pipe_format pformat,
+          int srcWidth, const ubyte *srcPtr,
+          int dstWidth, ubyte *dstPtr)
+{
+   enum dtype datatype;
+   uint comps;
+
+   format_to_type_comps(pformat, &datatype, &comps);
+
+   /* we just duplicate the input row, kind of hack, saves code */
+   do_row(datatype, comps,
+          srcWidth, srcPtr, srcPtr,
+          dstWidth, dstPtr);
+}
+
+
+/**
+ * Strides are in bytes.  If zero, it'll be computed as width * bpp.
+ */
+static void
+reduce_2d(enum pipe_format pformat,
+          int srcWidth, int srcHeight,
+          int srcRowStride, const ubyte *srcPtr,
+          int dstWidth, int dstHeight,
+          int dstRowStride, ubyte *dstPtr)
+{
+   enum dtype datatype;
+   uint comps;
+   const int bpt = pf_get_size(pformat);
+   const ubyte *srcA, *srcB;
+   ubyte *dst;
+   int row;
+
+   format_to_type_comps(pformat, &datatype, &comps);
+
+   if (!srcRowStride)
+      srcRowStride = bpt * srcWidth;
+
+   if (!dstRowStride)
+      dstRowStride = bpt * dstWidth;
+
+   /* Compute src and dst pointers */
+   srcA = srcPtr;
+   if (srcHeight > 1) 
+      srcB = srcA + srcRowStride;
+   else
+      srcB = srcA;
+   dst = dstPtr;
+
+   for (row = 0; row < dstHeight; row++) {
+      do_row(datatype, comps,
+             srcWidth, srcA, srcB,
+             dstWidth, dst);
+      srcA += 2 * srcRowStride;
+      srcB += 2 * srcRowStride;
+      dst += dstRowStride;
+   }
+}
+
+
+static void
+make_1d_mipmap(struct gen_mipmap_state *ctx,
+               struct pipe_texture *pt,
+               uint face, uint baseLevel, uint lastLevel)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_winsys *winsys = pipe->winsys;
+   const uint zslice = 0;
+   uint dstLevel;
+
+   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
+      const uint srcLevel = dstLevel - 1;
+      struct pipe_surface *srcSurf, *dstSurf;
+      void *srcMap, *dstMap;
+      
+      srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice);
+      dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice);
+
+      srcMap = ((ubyte *) winsys->buffer_map(winsys, srcSurf->buffer,
+                                            PIPE_BUFFER_USAGE_CPU_READ)
+                + srcSurf->offset);
+      dstMap = ((ubyte *) winsys->buffer_map(winsys, dstSurf->buffer,
+                                            PIPE_BUFFER_USAGE_CPU_WRITE)
+                + dstSurf->offset);
+
+      reduce_1d(pt->format,
+                srcSurf->width, srcMap,
+                dstSurf->width, dstMap);
+
+      winsys->buffer_unmap(winsys, srcSurf->buffer);
+      winsys->buffer_unmap(winsys, dstSurf->buffer);
+
+      pipe_surface_reference(&srcSurf, NULL);
+      pipe_surface_reference(&dstSurf, NULL);
+   }
+}
+
+
+static void
+make_2d_mipmap(struct gen_mipmap_state *ctx,
+               struct pipe_texture *pt,
+               uint face, uint baseLevel, uint lastLevel)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_winsys *winsys = pipe->winsys;
+   const uint zslice = 0;
+   uint dstLevel;
+   const int bpt = pf_get_size(pt->format);
+
+   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
+      const uint srcLevel = dstLevel - 1;
+      struct pipe_surface *srcSurf, *dstSurf;
+      ubyte *srcMap, *dstMap;
+      
+      srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice);
+      dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice);
+
+      srcMap = ((ubyte *) winsys->buffer_map(winsys, srcSurf->buffer,
+                                            PIPE_BUFFER_USAGE_CPU_READ)
+                + srcSurf->offset);
+      dstMap = ((ubyte *) winsys->buffer_map(winsys, dstSurf->buffer,
+                                            PIPE_BUFFER_USAGE_CPU_WRITE)
+                + dstSurf->offset);
+
+      reduce_2d(pt->format,
+                srcSurf->width, srcSurf->height,
+                srcSurf->pitch * bpt, srcMap,
+                dstSurf->width, dstSurf->height,
+                dstSurf->pitch * bpt, dstMap);
+
+      winsys->buffer_unmap(winsys, srcSurf->buffer);
+      winsys->buffer_unmap(winsys, dstSurf->buffer);
+
+      pipe_surface_reference(&srcSurf, NULL);
+      pipe_surface_reference(&dstSurf, NULL);
+   }
+}
+
+
+static void
+make_3d_mipmap(struct gen_mipmap_state *ctx,
+               struct pipe_texture *pt,
+               uint face, uint baseLevel, uint lastLevel)
+{
+}
+
+
+static void
+fallback_gen_mipmap(struct gen_mipmap_state *ctx,
+                    struct pipe_texture *pt,
+                    uint face, uint baseLevel, uint lastLevel)
+{
+   switch (pt->target) {
+   case PIPE_TEXTURE_1D:
+      make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      break;
+   case PIPE_TEXTURE_2D:
+   case PIPE_TEXTURE_CUBE:
+      make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      break;
+   case PIPE_TEXTURE_3D:
+      make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+/**
+ * Create a mipmap generation context.
+ * The idea is to create one of these and re-use it each time we need to
+ * generate a mipmap.
+ */
+struct gen_mipmap_state *
+util_create_gen_mipmap(struct pipe_context *pipe)
+{
+   struct pipe_blend_state blend;
+   struct pipe_depth_stencil_alpha_state depthstencil;
+   struct pipe_rasterizer_state rasterizer;
+   struct gen_mipmap_state *ctx;
+
+   ctx = CALLOC_STRUCT(gen_mipmap_state);
+   if (!ctx)
+      return NULL;
+
+   ctx->pipe = pipe;
+
+   /* we don't use blending, but need to set valid values */
+   memset(&blend, 0, sizeof(blend));
+   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.colormask = PIPE_MASK_RGBA;
+   ctx->blend = pipe->create_blend_state(pipe, &blend);
+
+   /* depth/stencil/alpha */
+   memset(&depthstencil, 0, sizeof(depthstencil));
+   ctx->depthstencil = pipe->create_depth_stencil_alpha_state(pipe, &depthstencil);
+
+   /* rasterizer */
+   memset(&rasterizer, 0, sizeof(rasterizer));
+   rasterizer.front_winding = PIPE_WINDING_CW;
+   rasterizer.cull_mode = PIPE_WINDING_NONE;
+   rasterizer.bypass_clipping = 1;  /* bypasses viewport too */
+   //rasterizer.bypass_vs = 1;
+   ctx->rasterizer = pipe->create_rasterizer_state(pipe, &rasterizer);
+
+#if 0
+   /* viewport */
+   ctx->viewport.scale[0] = 1.0;
+   ctx->viewport.scale[1] = 1.0;
+   ctx->viewport.scale[2] = 1.0;
+   ctx->viewport.scale[3] = 1.0;
+   ctx->viewport.translate[0] = 0.0;
+   ctx->viewport.translate[1] = 0.0;
+   ctx->viewport.translate[2] = 0.0;
+   ctx->viewport.translate[3] = 0.0;
+#endif
+
+   /* vertex shader */
+   {
+      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
+                                      TGSI_SEMANTIC_GENERIC };
+      const uint semantic_indexes[] = { 0, 0 };
+      ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
+                                                    semantic_indexes);
+   }
+
+   /* fragment shader */
+   ctx->fs = util_make_fragment_tex_shader(pipe);
+
+   return ctx;
+}
+
+
+/**
+ * Destroy a mipmap generation context
+ */
+void
+util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
+{
+   struct pipe_context *pipe = ctx->pipe;
+
+   pipe->delete_blend_state(pipe, ctx->blend);
+   pipe->delete_depth_stencil_alpha_state(pipe, ctx->depthstencil);
+   pipe->delete_rasterizer_state(pipe, ctx->rasterizer);
+   pipe->delete_vs_state(pipe, ctx->vs);
+   pipe->delete_fs_state(pipe, ctx->fs);
+
+   FREE(ctx);
+}
+
+
+#if 0
+static void
+simple_viewport(struct pipe_context *pipe, uint width, uint height)
+{
+   struct pipe_viewport_state vp;
+
+   vp.scale[0] =  0.5 * width;
+   vp.scale[1] = -0.5 * height;
+   vp.scale[2] = 1.0;
+   vp.scale[3] = 1.0;
+   vp.translate[0] = 0.5 * width;
+   vp.translate[1] = 0.5 * height;
+   vp.translate[2] = 0.0;
+   vp.translate[3] = 0.0;
+
+   pipe->set_viewport_state(pipe, &vp);
+}
+#endif
+
+
+/**
+ * Generate mipmap images.  It's assumed all needed texture memory is
+ * already allocated.
+ *
+ * \param pt  the texture to generate mipmap levels for
+ * \param face  which cube face to generate mipmaps for (0 for non-cube maps)
+ * \param baseLevel  the first mipmap level to use as a src
+ * \param lastLevel  the last mipmap level to generate
+ */
+void
+util_gen_mipmap(struct gen_mipmap_state *ctx,
+                struct pipe_texture *pt,
+                uint face, uint baseLevel, uint lastLevel)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_framebuffer_state fb;
+   struct pipe_sampler_state sampler;
+   void *sampler_cso;
+   uint dstLevel;
+   uint zslice = 0;
+
+   /* check if we can render in the texture's format */
+   if (!screen->is_format_supported(screen, pt->format, PIPE_SURFACE)) {
+      fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      return;
+   }
+
+   /* init framebuffer state */
+   memset(&fb, 0, sizeof(fb));
+   fb.num_cbufs = 1;
+
+   /* sampler state */
+   memset(&sampler, 0, sizeof(sampler));
+   sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
+   sampler.min_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.normalized_coords = 1;
+
+   /* bind our state */
+   pipe->bind_blend_state(pipe, ctx->blend);
+   pipe->bind_depth_stencil_alpha_state(pipe, ctx->depthstencil);
+   pipe->bind_rasterizer_state(pipe, ctx->rasterizer);
+   pipe->bind_vs_state(pipe, ctx->vs);
+   pipe->bind_fs_state(pipe, ctx->fs);
+#if 0
+   pipe->set_viewport_state(pipe, &ctx->viewport);
+#endif
+
+   /*
+    * XXX for small mipmap levels, it may be faster to use the software
+    * fallback path...
+    */
+   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
+      const uint srcLevel = dstLevel - 1;
+
+      /*
+       * Setup framebuffer / dest surface
+       */
+      fb.cbufs[0] = screen->get_tex_surface(screen, pt, face, dstLevel, zslice);
+      pipe->set_framebuffer_state(pipe, &fb);
+
+      /*
+       * Setup sampler state
+       * Note: we should only have to set the min/max LOD clamps to ensure
+       * we grab texels from the right mipmap level.  But some hardware
+       * has trouble with min clamping so we also set the lod_bias to
+       * try to work around that.
+       */
+      sampler.min_lod = sampler.max_lod = (float) srcLevel;
+      sampler.lod_bias = (float) srcLevel;
+      sampler_cso = pipe->create_sampler_state(pipe, &sampler);
+      pipe->bind_sampler_states(pipe, 1, &sampler_cso);
+
+#if 0
+      simple_viewport(pipe, pt->width[dstLevel], pt->height[dstLevel]);
+#endif
+
+      pipe->set_sampler_textures(pipe, 1, &pt);
+
+      /* quad coords in window coords (bypassing clipping, viewport mapping) */
+      util_draw_texquad(pipe,
+                        0.0F, 0.0F, /* x0, y0 */
+                        (float) pt->width[dstLevel], /* x1 */
+                        (float) pt->height[dstLevel], /* y1 */
+                        0.0F);  /* z */
+
+
+      pipe->flush(pipe, PIPE_FLUSH_WAIT);
+
+      /*pipe->texture_update(pipe, pt);  not really needed */
+
+      pipe->delete_sampler_state(pipe, sampler_cso);
+   }
+
+   /* Note: caller must restore pipe/gallium state at this time */
+}
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.h b/src/gallium/auxiliary/util/u_gen_mipmap.h
new file mode 100644
index 00000000000..80496140a24
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.h
@@ -0,0 +1,52 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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 U_GENMIPMAP_H
+#define U_GENMIPMAP_H
+
+#include "pipe/p_state.h"
+
+
+struct gen_mipmap_state;
+
+
+extern struct gen_mipmap_state *
+util_create_gen_mipmap(struct pipe_context *pipe);
+
+
+extern void
+util_destroy_gen_mipmap(struct gen_mipmap_state *ctx);
+
+
+
+extern void
+util_gen_mipmap(struct gen_mipmap_state *ctx,
+                struct pipe_texture *pt,
+                uint face, uint baseLevel, uint lastLevel);
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_handle_table.c b/src/gallium/auxiliary/util/u_handle_table.c
index d25872972aa..0bfb9e1b4ae 100644
--- a/src/gallium/auxiliary/util/u_handle_table.c
+++ b/src/gallium/auxiliary/util/u_handle_table.c
@@ -226,9 +226,13 @@ handle_table_remove(struct handle_table *ht,
 
    index = handle - 1;
    object = ht->objects[index];
-   assert(object);
+   if(!object) {
+      /* XXX: this warning may be noisy for legitimate use -- remove later */
+      debug_warning("removing empty handle");
+      return;
+   }
    
-   if(object && ht->destroy)
+   if(ht->destroy)
       ht->destroy(object);
 
    ht->objects[index] = NULL;
@@ -237,6 +241,28 @@ handle_table_remove(struct handle_table *ht,
 }
 
 
+unsigned
+handle_table_get_next_handle(struct handle_table *ht, 
+                             unsigned handle)
+{
+   unsigned index;
+   
+   for(index = handle; index < ht->size; ++index) {
+      if(!ht->objects[index])
+	 return index + 1;
+   }
+
+   return 0;
+}
+
+
+unsigned
+handle_table_get_first_handle(struct handle_table *ht)
+{
+   return handle_table_get_next_handle(ht, 0);
+}
+
+
 void
 handle_table_destroy(struct handle_table *ht)
 {
diff --git a/src/gallium/auxiliary/util/u_handle_table.h b/src/gallium/auxiliary/util/u_handle_table.h
index a2f1f604ade..d080135c9f4 100644
--- a/src/gallium/auxiliary/util/u_handle_table.h
+++ b/src/gallium/auxiliary/util/u_handle_table.h
@@ -100,6 +100,15 @@ void
 handle_table_destroy(struct handle_table *ht);
 
 
+unsigned
+handle_table_get_first_handle(struct handle_table *ht);
+
+
+unsigned
+handle_table_get_next_handle(struct handle_table *ht,
+                             unsigned handle);
+
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/auxiliary/util/u_hash_table.c b/src/gallium/auxiliary/util/u_hash_table.c
index ac2cb1b540d..f3f16a8d94f 100644
--- a/src/gallium/auxiliary/util/u_hash_table.c
+++ b/src/gallium/auxiliary/util/u_hash_table.c
@@ -187,6 +187,28 @@ hash_table_remove(struct hash_table *ht,
 }
 
 
+enum pipe_error
+hash_table_foreach(struct hash_table *ht,
+                   enum pipe_error (*callback)(void *key, void *value, void *data),
+                   void *data)
+{
+   struct cso_hash_iter iter;
+   struct hash_table_item *item;
+   enum pipe_error result;
+   
+   iter = cso_hash_first_node(ht->cso);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct hash_table_item *)cso_hash_iter_data(iter);
+      result = callback(item->key, item->value, data);
+      if(result != PIPE_OK)
+	 return result;
+      iter = cso_hash_iter_next(iter);
+   }
+
+   return PIPE_OK;
+}
+
+
 void
 hash_table_destroy(struct hash_table *ht)
 {
@@ -196,4 +218,3 @@ hash_table_destroy(struct hash_table *ht)
    
    FREE(ht);
 }
-
diff --git a/src/gallium/auxiliary/util/u_hash_table.h b/src/gallium/auxiliary/util/u_hash_table.h
index d941f2c6b16..1583bd7548c 100644
--- a/src/gallium/auxiliary/util/u_hash_table.h
+++ b/src/gallium/auxiliary/util/u_hash_table.h
@@ -75,6 +75,11 @@ hash_table_remove(struct hash_table *ht,
                   void *key);
 
 
+enum pipe_error
+hash_table_foreach(struct hash_table *ht,
+                   enum pipe_error (*callback)(void *key, void *value, void *data),
+                   void *data);
+
 void
 hash_table_destroy(struct hash_table *ht);
 
diff --git a/src/gallium/auxiliary/util/u_pack_color.h b/src/gallium/auxiliary/util/u_pack_color.h
new file mode 100644
index 00000000000..cd13823985e
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_pack_color.h
@@ -0,0 +1,133 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Functions to produce packed colors/Z from floats.
+ */
+
+
+#ifndef U_PACK_COLOR_H
+#define U_PACK_COLOR_H
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_format.h"
+
+
+/**
+ * Note rgba outside [0,1] will be clamped for int pixel formats.
+ */
+static INLINE void
+util_pack_color(const float rgba[4], enum pipe_format format, void *dest)
+{
+   ubyte r, g, b, a;
+
+   if (pf_size_x(format) <= 8) {
+      /* format uses 8-bit components or less */
+      UNCLAMPED_FLOAT_TO_UBYTE(r, rgba[0]);
+      UNCLAMPED_FLOAT_TO_UBYTE(g, rgba[1]);
+      UNCLAMPED_FLOAT_TO_UBYTE(b, rgba[2]);
+      UNCLAMPED_FLOAT_TO_UBYTE(a, rgba[3]);
+   }
+
+   switch (format) {
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (r << 24) | (g << 16) | (b << 8) | a;
+      }
+      return;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (a << 24) | (r << 16) | (g << 8) | b;
+      }
+      return;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (b << 24) | (g << 16) | (r << 8) | a;
+      }
+      return;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      {
+         ushort *d = (ushort *) dest;
+         *d = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      {
+         float *d = (float *) dest;
+         d[0] = rgba[0];
+         d[1] = rgba[1];
+         d[2] = rgba[2];
+         d[3] = rgba[3];
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      {
+         float *d = (float *) dest;
+         d[0] = rgba[0];
+         d[1] = rgba[1];
+         d[2] = rgba[2];
+      }
+      return;
+   /* XXX lots more cases to add */
+   default:
+      debug_printf("gallium: unhandled format in util_pack_color()");
+   }
+}
+ 
+
+/**
+ * Note: it's assumed that z is in [0,1]
+ */
+static INLINE uint
+util_pack_z(enum pipe_format format, double z)
+{
+   switch (format) {
+   case PIPE_FORMAT_Z16_UNORM:
+      return (uint) (z * 0xffff);
+   case PIPE_FORMAT_Z32_UNORM:
+      /* special-case to avoid overflow */
+      if (z == 1.0)
+         return 0xffffffff;
+      else
+         return (uint) (z * 0xffffffff);
+   case PIPE_FORMAT_S8Z24_UNORM:
+      return (uint) (z * 0xffffff);
+   case PIPE_FORMAT_Z24S8_UNORM:
+      return ((uint) (z * 0xffffff)) << 8;
+   default:
+      debug_printf("gallium: unhandled fomrat in util_pack_z()");
+      return 0;
+   }
+}
+
+
+#endif /* U_PACK_COLOR_H */
diff --git a/src/gallium/auxiliary/util/u_simple_shaders.c b/src/gallium/auxiliary/util/u_simple_shaders.c
new file mode 100644
index 00000000000..88e2ab05bd0
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_shaders.c
@@ -0,0 +1,263 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Simple vertex/fragment shader generators.
+ *  
+ * @author Brian Paul
+ */
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_util.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "util/u_simple_shaders.h"
+
+#include "tgsi/util/tgsi_build.h"
+#include "tgsi/util/tgsi_dump.h"
+#include "tgsi/util/tgsi_parse.h"
+
+
+
+/**
+ * Make simple vertex pass-through shader.
+ */
+void *
+util_make_vertex_passthrough_shader(struct pipe_context *pipe,
+                                    uint num_attribs,
+                                    const uint *semantic_names,
+                                    const uint *semantic_indexes)
+{
+   uint maxTokens = 100;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+   struct tgsi_processor *processor;
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+   const uint procType = TGSI_PROCESSOR_VERTEX;
+   uint ti, i;
+   struct pipe_shader_state shader;
+
+   tokens = (struct tgsi_token *) malloc(maxTokens * sizeof(tokens[0]));
+
+   /* shader header
+    */
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) &tokens[2];
+   *processor = tgsi_build_processor( procType, header );
+
+   ti = 3;
+
+   /* declare inputs */
+   for (i = 0; i < num_attribs; i++) {
+
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_INPUT;
+      /*
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = TGSI_SEMANTIC_POSITION;
+      decl.Semantic.SemanticIndex = 0;
+      */
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = 0;
+      ti += tgsi_build_full_declaration(&decl,
+                                        &tokens[ti],
+                                        header,
+                                        maxTokens - ti);
+   }
+
+   /* declare outputs */
+   for (i = 0; i < num_attribs; i++) {
+
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_OUTPUT;
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = semantic_names[i];
+      decl.Semantic.SemanticIndex = semantic_indexes[i];
+      decl.u.DeclarationRange.First = 
+         decl.u.DeclarationRange.Last = 0;
+      ti += tgsi_build_full_declaration(&decl,
+                                        &tokens[ti],
+                                        header,
+                                        maxTokens - ti);
+
+   }
+
+   /* emit MOV instructions */
+   for (i = 0; i < num_attribs; i++) {
+      /* MOVE out[i], in[i]; */
+      inst = tgsi_default_full_instruction();
+      inst.Instruction.Opcode = TGSI_OPCODE_MOV;
+      inst.Instruction.NumDstRegs = 1;
+      inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      inst.FullDstRegisters[0].DstRegister.Index = i;
+      inst.Instruction.NumSrcRegs = 1;
+      inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      inst.FullSrcRegisters[0].SrcRegister.Index = i;
+      ti += tgsi_build_full_instruction(&inst,
+                                        &tokens[ti],
+                                        header,
+                                        maxTokens - ti );
+   }
+
+   /* END instruction */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_END;
+   inst.Instruction.NumDstRegs = 0;
+   inst.Instruction.NumSrcRegs = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+#if 0 /*debug*/
+   tgsi_dump(tokens, 0);
+#endif
+
+   shader.tokens = tokens;
+   return pipe->create_vs_state(pipe, &shader);
+}
+
+
+
+
+/**
+ * Make simple fragment texture shader:
+ *  TEX OUT[0], IN[0], SAMP[0], 2D;
+ *  END;
+ */
+void *
+util_make_fragment_tex_shader(struct pipe_context *pipe)
+{
+   uint maxTokens = 100;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+   struct tgsi_processor *processor;
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+   const uint procType = TGSI_PROCESSOR_FRAGMENT;
+   uint ti;
+   struct pipe_shader_state shader;
+
+   tokens = (struct tgsi_token *) malloc(maxTokens * sizeof(tokens[0]));
+
+   /* shader header
+    */
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) &tokens[2];
+   *processor = tgsi_build_processor( procType, header );
+
+   ti = 3;
+
+   /* declare TEX[0] input */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_INPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = TGSI_SEMANTIC_GENERIC;
+   decl.Semantic.SemanticIndex = 0;
+   /* XXX this could be linear... */
+   decl.Declaration.Interpolate = 1;
+   decl.Interpolation.Interpolate = TGSI_INTERPOLATE_PERSPECTIVE;
+   decl.u.DeclarationRange.First = 
+   decl.u.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* declare color[0] output */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_OUTPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = TGSI_SEMANTIC_COLOR;
+   decl.Semantic.SemanticIndex = 0;
+   decl.u.DeclarationRange.First = 
+   decl.u.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* declare sampler */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_SAMPLER;
+   decl.u.DeclarationRange.First = 
+   decl.u.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* TEX instruction */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_TEX;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+   inst.FullDstRegisters[0].DstRegister.Index = 0;
+   inst.Instruction.NumSrcRegs = 2;
+   inst.InstructionExtTexture.Texture = TGSI_TEXTURE_2D;
+   inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+   inst.FullSrcRegisters[0].SrcRegister.Index = 0;
+   inst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER;
+   inst.FullSrcRegisters[1].SrcRegister.Index = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+   /* END instruction */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_END;
+   inst.Instruction.NumDstRegs = 0;
+   inst.Instruction.NumSrcRegs = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+#if 0 /*debug*/
+   tgsi_dump(tokens, 0);
+#endif
+
+   shader.tokens = tokens;
+   return pipe->create_fs_state(pipe, &shader);
+}
+
diff --git a/src/gallium/auxiliary/util/u_simple_shaders.h b/src/gallium/auxiliary/util/u_simple_shaders.h
new file mode 100644
index 00000000000..3ef4f288018
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_shaders.h
@@ -0,0 +1,52 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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 U_SIMPLE_SHADERS_H
+#define U_SIMPLE_SHADERS_H
+
+
+#include "pipe/p_compiler.h"
+
+
+struct pipe_context;
+
+
+extern void *
+util_make_vertex_passthrough_shader(struct pipe_context *pipe,
+                                    uint num_attribs,
+                                    const uint *semantic_names,
+                                    const uint *semantic_indexes);
+
+
+extern void *
+util_make_fragment_tex_shader(struct pipe_context *pipe);
+
+
+#endif
+
+
diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index 9a4004535ea..fe93fd8e1a2 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -105,6 +105,16 @@
 
 
 /**
+ */
+struct cell_command_depth_stencil_alpha_test {
+   uint64_t base;               /**< Effective address of code start. */
+   unsigned size;               /**< Size in bytes of test code. */
+   unsigned read_depth;         /**< Flag: should depth be read? */
+   unsigned read_stencil;       /**< Flag: should stencil be read? */
+};
+
+
+/**
  * Tell SPUs about the framebuffer size, location
  */
 struct cell_command_framebuffer
diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile
index d38fa6ce073..0389a9554cf 100644
--- a/src/gallium/drivers/cell/ppu/Makefile
+++ b/src/gallium/drivers/cell/ppu/Makefile
@@ -27,6 +27,7 @@ SOURCES = \
 	cell_flush.c \
 	cell_state_derived.c \
 	cell_state_emit.c \
+	cell_state_per_fragment.c \
 	cell_state_shader.c \
 	cell_pipe_state.c \
 	cell_screen.c \
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index b221424323f..9e79db0acef 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -57,16 +57,37 @@ struct cell_fragment_shader_state
 };
 
 
+struct cell_blend_state {
+   struct pipe_blend_state base;
+
+   /**
+    * Generated code to perform alpha blending
+    */
+   struct spe_function code;
+};
+
+
+struct cell_depth_stencil_alpha_state {
+   struct pipe_depth_stencil_alpha_state   base;
+
+   /**
+    * Generated code to perform alpha, stencil, and depth testing on the SPE
+    */
+   struct spe_function code;
+
+};
+
+
 struct cell_context
 {
    struct pipe_context pipe;
 
    struct cell_winsys *winsys;
 
-   const struct pipe_blend_state *blend;
+   const struct cell_blend_state *blend;
    const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
    uint num_samplers;
-   const struct pipe_depth_stencil_alpha_state   *depth_stencil;
+   const struct cell_depth_stencil_alpha_state   *depth_stencil;
    const struct pipe_rasterizer_state *rasterizer;
    const struct cell_vertex_shader_state *vs;
    const struct cell_fragment_shader_state *fs;
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index 025ed3bbbfe..c880760e4bd 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -36,6 +36,7 @@
 #include "cell_context.h"
 #include "cell_state.h"
 #include "cell_texture.h"
+#include "cell_state_per_fragment.h"
 
 
 
@@ -43,7 +44,12 @@ static void *
 cell_create_blend_state(struct pipe_context *pipe,
                         const struct pipe_blend_state *blend)
 {
-   return mem_dup(blend, sizeof(*blend));
+   struct cell_blend_state *cb = MALLOC(sizeof(struct cell_blend_state));
+
+   (void) memcpy(cb, blend, sizeof(*blend));
+   cb->code.store = NULL;
+
+   return cb;
 }
 
 
@@ -54,7 +60,7 @@ cell_bind_blend_state(struct pipe_context *pipe, void *blend)
 
    draw_flush(cell->draw);
 
-   cell->blend = (const struct pipe_blend_state *)blend;
+   cell->blend = (const struct cell_blend_state *)blend;
 
    cell->dirty |= CELL_NEW_BLEND;
 }
@@ -63,7 +69,10 @@ cell_bind_blend_state(struct pipe_context *pipe, void *blend)
 static void
 cell_delete_blend_state(struct pipe_context *pipe, void *blend)
 {
-   FREE(blend);
+   struct cell_blend_state *cb = (struct cell_blend_state *) blend;
+   
+   spe_release_func(& cb->code);
+   FREE(cb);
 }
 
 
@@ -87,7 +96,13 @@ static void *
 cell_create_depth_stencil_alpha_state(struct pipe_context *pipe,
                  const struct pipe_depth_stencil_alpha_state *depth_stencil)
 {
-   return mem_dup(depth_stencil, sizeof(*depth_stencil));
+   struct cell_depth_stencil_alpha_state *cdsa =
+       MALLOC(sizeof(struct cell_depth_stencil_alpha_state));
+
+   (void) memcpy(cdsa, depth_stencil, sizeof(*depth_stencil));
+   cdsa->code.store = NULL;
+
+   return cdsa;
 }
 
 
@@ -96,12 +111,16 @@ cell_bind_depth_stencil_alpha_state(struct pipe_context *pipe,
                                     void *depth_stencil)
 {
    struct cell_context *cell = cell_context(pipe);
+   struct cell_depth_stencil_alpha_state *cdsa =
+       (struct cell_depth_stencil_alpha_state *) depth_stencil;
 
    draw_flush(cell->draw);
 
-   cell->depth_stencil
-      = (const struct pipe_depth_stencil_alpha_state *) depth_stencil;
+   if ((cdsa != NULL) && (cdsa->code.store == NULL)) {
+      cell_generate_depth_stencil_test(cdsa);
+   }
 
+   cell->depth_stencil = cdsa;
    cell->dirty |= CELL_NEW_DEPTH_STENCIL;
 }
 
@@ -109,7 +128,11 @@ cell_bind_depth_stencil_alpha_state(struct pipe_context *pipe,
 static void
 cell_delete_depth_stencil_alpha_state(struct pipe_context *pipe, void *depth)
 {
-   FREE(depth);
+   struct cell_depth_stencil_alpha_state *cdsa =
+       (struct cell_depth_stencil_alpha_state *) depth;
+
+   spe_release_func(& cdsa->code);
+   FREE(cdsa);
 }
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 670eb26bdd0..4d589bcdbf9 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -71,9 +71,24 @@ cell_emit_state(struct cell_context *cell)
    }
 
    if (cell->dirty & CELL_NEW_DEPTH_STENCIL) {
-      emit_state_cmd(cell, CELL_CMD_STATE_DEPTH_STENCIL,
-                     cell->depth_stencil,
-                     sizeof(struct pipe_depth_stencil_alpha_state));
+      struct cell_command_depth_stencil_alpha_test dsat;
+      
+
+      if (cell->depth_stencil != NULL) {
+	 dsat.base = (intptr_t) cell->depth_stencil->code.store;
+	 dsat.size = (char *) cell->depth_stencil->code.csr
+	     - (char *) cell->depth_stencil->code.store;
+	 dsat.read_depth = TRUE;
+	 dsat.read_stencil = FALSE;
+      } else {
+	 dsat.base = 0;
+	 dsat.size = 0;
+	 dsat.read_depth = FALSE;
+	 dsat.read_stencil = FALSE;
+      }
+
+      emit_state_cmd(cell, CELL_CMD_STATE_DEPTH_STENCIL, &dsat,
+		     sizeof(dsat));
    }
 
    if (cell->dirty & CELL_NEW_SAMPLER) {
diff --git a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
new file mode 100644
index 00000000000..9c479684596
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
@@ -0,0 +1,1075 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 (including the next
+ * paragraph) 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 NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS 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.
+ */
+
+/**
+ * \file
+ * Generate code to perform all per-fragment operations.
+ *
+ * Code generated by these functions perform both alpha, depth, and stencil
+ * testing as well as alpha blending.
+ *
+ * \note
+ * Occlusion query is not supported, but this is the right place to add that
+ * support.
+ *
+ * \author Ian Romanick <[email protected]>
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+
+#include "cell_context.h"
+
+#include "rtasm/rtasm_ppc_spe.h"
+
+
+/**
+ * Generate code to perform alpha testing.
+ *
+ * The code generated by this function uses the register specificed by
+ * \c mask as both an input and an output.
+ *
+ * \param dsa    Current alpha-test state
+ * \param f      Function to which code should be appended
+ * \param mask   Index of register containing active fragment mask
+ * \param alphas Index of register containing per-fragment alpha values
+ *
+ * \note Emits a maximum of 6 instructions.
+ */
+static void
+emit_alpha_test(struct pipe_depth_stencil_alpha_state *dsa,
+                struct spe_function *f, int mask, int alphas)
+{
+   /* If the alpha function is either NEVER or ALWAYS, there is no need to
+    * load the reference value into a register.  ALWAYS is a fairly common
+    * case, and this optimization saves 2 instructions.
+    */
+   if (dsa->alpha.enabled
+       && (dsa->alpha.func != PIPE_FUNC_NEVER)
+       && (dsa->alpha.func != PIPE_FUNC_ALWAYS)) {
+      int ref = spe_allocate_available_register(f);
+      int tmp_a = spe_allocate_available_register(f);
+      int tmp_b = spe_allocate_available_register(f);
+      union {
+	 float f;
+	 unsigned u;
+      } ref_val;
+      boolean complement = FALSE;
+
+      ref_val.f = dsa->alpha.ref;
+
+      spe_il(f, ref, ref_val.u & 0x0000ffff);
+      spe_ilh(f, ref, ref_val.u >> 16);
+
+      switch (dsa->alpha.func) {
+      case PIPE_FUNC_NOTEQUAL:
+	 complement = TRUE;
+	 /* FALLTHROUGH */
+
+      case PIPE_FUNC_EQUAL:
+	 spe_fceq(f, tmp_a, ref, alphas);
+	 break;
+
+      case PIPE_FUNC_LEQUAL:
+	 complement = TRUE;
+	 /* FALLTHROUGH */
+
+      case PIPE_FUNC_GREATER:
+	 spe_fcgt(f, tmp_a, ref, alphas);
+	 break;
+
+      case PIPE_FUNC_LESS:
+	 complement = TRUE;
+	 /* FALLTHROUGH */
+
+      case PIPE_FUNC_GEQUAL:
+	 spe_fcgt(f, tmp_a, ref, alphas);
+	 spe_fceq(f, tmp_b, ref, alphas);
+	 spe_or(f, tmp_a, tmp_b, tmp_a);
+	 break;
+
+      case PIPE_FUNC_ALWAYS:
+      case PIPE_FUNC_NEVER:
+      default:
+	 assert(0);
+	 break;
+      }
+
+      if (complement) {
+	 spe_andc(f, mask, mask, tmp_a);
+      } else {
+	 spe_and(f, mask, mask, tmp_a);
+      }
+
+      spe_release_register(f, ref);
+      spe_release_register(f, tmp_a);
+      spe_release_register(f, tmp_b);
+   } else if (dsa->alpha.enabled && (dsa->alpha.func == PIPE_FUNC_NEVER)) {
+      spe_il(f, mask, 0);
+   }
+}
+
+
+/**
+ * \param dsa        Current depth-test state
+ * \param f          Function to which code should be appended
+ * \param m          Mask of allocated / free SPE registers
+ * \param mask       Index of register to contain depth-pass mask
+ * \param stored     Index of register containing values from depth buffer
+ * \param calculated Index of register containing per-fragment depth values
+ *
+ * \return
+ * If the calculated depth comparison mask is the actual mask, \c FALSE is
+ * returned.  If the calculated depth comparison mask is the compliment of
+ * the actual mask, \c TRUE is returned.
+ *
+ * \note Emits a maximum of 3 instructions.
+ */
+static boolean
+emit_depth_test(struct pipe_depth_stencil_alpha_state *dsa,
+                struct spe_function *f, int mask, int stored, int calculated)
+{
+   unsigned func = (dsa->depth.enabled)
+       ? dsa->depth.func : PIPE_FUNC_ALWAYS;
+   int tmp = spe_allocate_available_register(f);
+   boolean compliment = FALSE;
+
+   switch (func) {
+   case PIPE_FUNC_NEVER:
+      spe_il(f, mask, 0);
+      break;
+
+   case PIPE_FUNC_NOTEQUAL:
+      compliment = TRUE;
+      /* FALLTHROUGH */
+   case PIPE_FUNC_EQUAL:
+      spe_ceq(f, mask, calculated, stored);
+      break;
+
+   case PIPE_FUNC_LEQUAL:
+      compliment = TRUE;
+      /* FALLTHROUGH */
+   case PIPE_FUNC_GREATER:
+      spe_clgt(f, mask, calculated, stored);
+      break;
+
+   case PIPE_FUNC_LESS:
+      compliment = TRUE;
+      /* FALLTHROUGH */
+   case PIPE_FUNC_GEQUAL:
+      spe_clgt(f, mask, calculated, stored);
+      spe_ceq(f, tmp, calculated, stored);
+      spe_or(f, mask, mask, tmp);
+      break;
+
+   case PIPE_FUNC_ALWAYS:
+      spe_il(f, mask, ~0);
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+
+   spe_release_register(f, tmp);
+   return compliment;
+}
+
+
+/**
+ * \note Emits a maximum of 5 instructions.
+ *
+ * \warning
+ * Since \c out and \c in might be the same register, this routine cannot
+ * generate code that uses \c out as a temporary.
+ */
+static void
+emit_stencil_op(struct spe_function *f,
+                int out, int in, int mask, unsigned op, unsigned ref)
+{
+   const int clamp = spe_allocate_available_register(f);
+   const int clamp_mask = spe_allocate_available_register(f);
+   const int result = spe_allocate_available_register(f);
+
+   switch(op) {
+   case PIPE_STENCIL_OP_KEEP:
+      assert(0);
+   case PIPE_STENCIL_OP_ZERO:
+      spe_il(f, result, 0);
+      break;
+   case PIPE_STENCIL_OP_REPLACE:
+      spe_il(f, result, ref);
+      break;
+   case PIPE_STENCIL_OP_INCR:
+      spe_il(f, clamp, 0x0ff);
+      spe_ai(f, result, in, 1);
+      spe_clgti(f, clamp_mask, result, 0x0ff);
+      spe_selb(f, result, result, clamp, clamp_mask);
+      break;
+   case PIPE_STENCIL_OP_DECR:
+      spe_il(f, clamp, 0);
+      spe_ai(f, result, in, -1);
+
+      /* If "(s-1) < 0" in signed arithemtic, then "(s-1) > MAX" in unsigned
+       * arithmetic.
+       */
+      spe_clgti(f, clamp_mask, result, 0x0ff);
+      spe_selb(f, result, result, clamp, clamp_mask);
+      break;
+   case PIPE_STENCIL_OP_INCR_WRAP:
+      spe_ai(f, result, in, 1);
+      break;
+   case PIPE_STENCIL_OP_DECR_WRAP:
+      spe_ai(f, result, in, -1);
+      break;
+   case PIPE_STENCIL_OP_INVERT:
+      spe_nor(f, result, in, in);
+      break;
+   default:
+      assert(0);
+   }
+
+   spe_selb(f, out, in, result, mask);
+
+   spe_release_register(f, result);
+   spe_release_register(f, clamp_mask);
+   spe_release_register(f, clamp);
+}
+
+
+/**
+ * \param dsa        Depth / stencil test state
+ * \param face       0 for front face, 1 for back face
+ * \param f          Function to append instructions to
+ * \param reg_mask   Mask of allocated registers
+ * \param mask       Register containing mask of fragments passing the
+ *                   alpha test
+ * \param depth_mask Register containing mask of fragments passing the
+ *                   depth test
+ * \param depth_compliment  Is \c depth_mask the compliment of the actual mask?
+ * \param stencil    Register containing values from stencil buffer
+ * \param depth_pass Register to store mask of fragments passing stencil test
+ *                   and depth test
+ *
+ * \note
+ * Emits a maximum of 10 + (3 * 5) = 25 instructions.
+ */
+static int
+emit_stencil_test(struct pipe_depth_stencil_alpha_state *dsa,
+                  unsigned face,
+                  struct spe_function *f,
+                  int mask,
+                  int depth_mask,
+                  boolean depth_complement,
+                  int stencil,
+                  int depth_pass)
+{
+   int stencil_fail = spe_allocate_available_register(f);
+   int depth_fail = spe_allocate_available_register(f);
+   int stencil_mask = spe_allocate_available_register(f);
+   int stencil_pass = spe_allocate_available_register(f);
+   int face_stencil = spe_allocate_available_register(f);
+   int stencil_src = stencil;
+   const unsigned ref = (dsa->stencil[face].ref_value
+                         & dsa->stencil[face].value_mask);
+   boolean complement = FALSE;
+   int stored;
+   int tmp = spe_allocate_available_register(f);
+
+
+   if ((dsa->stencil[face].func != PIPE_FUNC_NEVER)
+       && (dsa->stencil[face].func != PIPE_FUNC_ALWAYS)
+       && (dsa->stencil[face].value_mask != 0x0ff)) {
+      stored = spe_allocate_available_register(f);
+      spe_andi(f, stored, stencil, dsa->stencil[face].value_mask);
+   } else {
+      stored = stencil;
+   }
+
+
+   switch (dsa->stencil[face].func) {
+   case PIPE_FUNC_NEVER:
+      spe_il(f, stencil_mask, 0);
+      break;
+
+   case PIPE_FUNC_NOTEQUAL:
+      complement = TRUE;
+      /* FALLTHROUGH */
+   case PIPE_FUNC_EQUAL:
+      spe_ceqi(f, stencil_mask, stored, ref);
+      break;
+
+   case PIPE_FUNC_LEQUAL:
+      complement = TRUE;
+      /* FALLTHROUGH */
+   case PIPE_FUNC_GREATER:
+      spe_clgti(f, stencil_mask, stored, ref);
+      break;
+
+   case PIPE_FUNC_LESS:
+      complement = TRUE;
+      /* FALLTHROUGH */
+   case PIPE_FUNC_GEQUAL:
+      spe_clgti(f, stencil_mask, stored, ref);
+      spe_ceqi(f, tmp, stored, ref);
+      spe_or(f, stencil_mask, stencil_mask, tmp);
+      break;
+
+   case PIPE_FUNC_ALWAYS:
+      /* See comment below. */
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+
+   if (stored != stencil) {
+      spe_release_register(f, stored);
+   }
+   spe_release_register(f, tmp);
+
+
+   /* ALWAYS is a very common stencil-test, so some effort is applied to
+    * optimize that case.  The stencil-pass mask is the same as the input
+    * fragment mask.  This makes the stencil-test (above) a no-op, and the
+    * input fragment mask can be "renamed" the stencil-pass mask.
+    */
+   if (dsa->stencil[face].func == PIPE_FUNC_ALWAYS) {
+      spe_release_register(f, stencil_pass);
+      stencil_pass = mask;
+   } else {
+      if (complement) {
+         spe_andc(f, stencil_pass, mask, stencil_mask);
+      } else {
+         spe_and(f, stencil_pass, mask, stencil_mask);
+      }
+   }
+
+   if (depth_complement) {
+      spe_andc(f, depth_pass, stencil_pass, depth_mask);
+   } else {
+      spe_and(f, depth_pass, stencil_pass, depth_mask);
+   }
+
+
+   /* Conditionally emit code to update the stencil value under various
+    * condititons.  Note that there is no need to generate code under the
+    * following circumstances:
+    *
+    * - Stencil write mask is zero.
+    * - For stencil-fail if the stencil test is ALWAYS
+    * - For depth-fail if the stencil test is NEVER
+    * - For depth-pass if the stencil test is NEVER
+    * - Any of the 3 conditions if the operation is KEEP
+    */
+   if (dsa->stencil[face].write_mask != 0) {
+      if ((dsa->stencil[face].func != PIPE_FUNC_ALWAYS)
+          && (dsa->stencil[face].fail_op != PIPE_STENCIL_OP_KEEP)) {
+         if (complement) {
+            spe_and(f, stencil_fail, mask, stencil_mask);
+         } else {
+            spe_andc(f, stencil_fail, mask, stencil_mask);
+         }
+
+         emit_stencil_op(f, face_stencil, stencil_src, stencil_fail,
+                         dsa->stencil[face].fail_op,
+                         dsa->stencil[face].ref_value);
+
+         stencil_src = face_stencil;
+      }
+
+      if ((dsa->stencil[face].func != PIPE_FUNC_NEVER)
+          && (dsa->stencil[face].zfail_op != PIPE_STENCIL_OP_KEEP)) {
+         if (depth_complement) {
+            spe_and(f, depth_fail, stencil_pass, depth_mask);
+         } else {
+            spe_andc(f, depth_fail, stencil_pass, depth_mask);
+         }
+
+         emit_stencil_op(f, face_stencil, stencil_src, depth_fail,
+                         dsa->stencil[face].zfail_op,
+                         dsa->stencil[face].ref_value);
+         stencil_src = face_stencil;
+      }
+
+      if ((dsa->stencil[face].func != PIPE_FUNC_NEVER)
+          && (dsa->stencil[face].zpass_op != PIPE_STENCIL_OP_KEEP)) {
+         emit_stencil_op(f, face_stencil, stencil_src, depth_pass,
+                         dsa->stencil[face].zpass_op,
+                         dsa->stencil[face].ref_value);
+         stencil_src = face_stencil;
+      }
+   }
+
+   spe_release_register(f, stencil_fail);
+   spe_release_register(f, depth_fail);
+   spe_release_register(f, stencil_mask);
+   if (stencil_pass != mask) {
+      spe_release_register(f, stencil_pass);
+   }
+
+   /* If all of the stencil operations were KEEP or the stencil write mask was
+    * zero, "stencil_src" will still be set to "stencil".  In this case
+    * release the "face_stencil" register.  Otherwise apply the stencil write
+    * mask to select bits from the calculated stencil value and the previous
+    * stencil value.
+    */
+   if (stencil_src == stencil) {
+      spe_release_register(f, face_stencil);
+   } else if (dsa->stencil[face].write_mask != 0x0ff) {
+      int tmp = spe_allocate_available_register(f);
+
+      spe_il(f, tmp, dsa->stencil[face].write_mask);
+      spe_selb(f, stencil_src, stencil, stencil_src, tmp);
+
+      spe_release_register(f, tmp);
+   }
+
+   return stencil_src;
+}
+
+
+void
+cell_generate_depth_stencil_test(struct cell_depth_stencil_alpha_state *cdsa)
+{
+   struct pipe_depth_stencil_alpha_state *const dsa = &cdsa->base;
+   struct spe_function *const f = &cdsa->code;
+
+   /* This code generates a maximum of 6 (alpha test) + 3 (depth test)
+    * + 25 (front stencil) + 25 (back stencil) + 4 = 63 instructions.  Round
+    * up to 64 to make it a happy power-of-two.
+    */
+   spe_init_func(f, 4 * 64);
+
+
+   /* Allocate registers for the function's input parameters.  Cleverly (and
+    * clever code is usually dangerous, but I couldn't resist) the generated
+    * function returns a structure.  Returned structures start with register
+    * 3, and the structure fields are ordered to match up exactly with the
+    * input parameters.
+    */
+   int mask = spe_allocate_register(f, 3);
+   int depth = spe_allocate_register(f, 4);
+   int stencil = spe_allocate_register(f, 5);
+   int zvals = spe_allocate_register(f, 6);
+   int frag_a = spe_allocate_register(f, 7);
+   int facing = spe_allocate_register(f, 8);
+
+   int depth_mask = spe_allocate_available_register(f);
+
+   boolean depth_complement;
+
+
+   emit_alpha_test(dsa, f, mask, frag_a);
+
+   depth_complement = emit_depth_test(dsa, f, depth_mask, depth, zvals);
+
+   if (dsa->stencil[0].enabled) {
+      const int front_depth_pass = spe_allocate_available_register(f);
+      int front_stencil = emit_stencil_test(dsa, 0, f, mask,
+                                            depth_mask, depth_complement,
+                                            stencil, front_depth_pass);
+
+      if (dsa->stencil[1].enabled) {
+         const int back_depth_pass = spe_allocate_available_register(f);
+         int back_stencil = emit_stencil_test(dsa, 1, f, mask,
+                                              depth_mask,  depth_complement,
+                                              stencil, back_depth_pass);
+
+         /* If the front facing stencil value and the back facing stencil
+          * value are stored in the same register, there is no need to select
+          * a value based on the facing.  This can happen if the stencil value
+          * was not modified due to the write masks being zero, the stencil
+          * operations being KEEP, etc.
+          */
+         if (front_stencil != back_stencil) {
+            spe_selb(f, stencil, back_stencil, front_stencil, facing);
+         }
+
+         if (back_stencil != stencil) {
+            spe_release_register(f, back_stencil);
+         }
+
+         if (front_stencil != stencil) {
+            spe_release_register(f, front_stencil);
+         }
+
+         spe_selb(f, mask, back_depth_pass, front_depth_pass, facing);
+
+         spe_release_register(f, back_depth_pass);
+      } else {
+         if (front_stencil != stencil) {
+            spe_or(f, stencil, front_stencil, front_stencil);
+            spe_release_register(f, front_stencil);
+         }
+         spe_or(f, mask, front_depth_pass, front_depth_pass);
+      }
+
+      spe_release_register(f, front_depth_pass);
+   } else if (dsa->depth.enabled) {
+      if (depth_complement) {
+         spe_andc(f, mask, mask, depth_mask);
+      } else {
+         spe_and(f, mask, mask, depth_mask);
+      }
+   }
+
+   if (dsa->depth.writemask) {
+         spe_selb(f, depth, depth, zvals, mask);
+   }
+
+   spe_bi(f, 0, 0, 0);
+
+
+#if 0
+   {
+      const uint32_t *p = f->store;
+      unsigned i;
+
+      printf("# alpha (%sabled)\n",
+             (dsa->alpha.enabled) ? "en" : "dis");
+      printf("#    func: %u\n", dsa->alpha.func);
+      printf("#    ref: %.2f\n", dsa->alpha.ref);
+
+      printf("# depth (%sabled)\n",
+             (dsa->depth.enabled) ? "en" : "dis");
+      printf("#    func: %u\n", dsa->depth.func);
+
+      for (i = 0; i < 2; i++) {
+         printf("# %s stencil (%sabled)\n",
+                (i == 0) ? "front" : "back",
+                (dsa->stencil[i].enabled) ? "en" : "dis");
+
+         printf("#    func: %u\n", dsa->stencil[i].func);
+         printf("#    op (sf, zf, zp): %u %u %u\n",
+                dsa->stencil[i].fail_op,
+                dsa->stencil[i].zfail_op,
+                dsa->stencil[i].zpass_op);
+         printf("#    ref value / value mask / write mask: %02x %02x %02x\n",
+                dsa->stencil[i].ref_value,
+                dsa->stencil[i].value_mask,
+                dsa->stencil[i].write_mask);
+      }
+
+      printf("\t.text\n");
+      for (/* empty */; p < f->csr; p++) {
+         printf("\t.long\t0x%04x\n", *p);
+      }
+      fflush(stdout);
+   }
+#endif
+}
+
+
+/**
+ * \note Emits a maximum of 3 instructions
+ */
+static int
+emit_alpha_factor_calculation(struct spe_function *f,
+                              unsigned factor, float const_alpha,
+                              int src_alpha, int dst_alpha)
+{
+   union {
+      float f;
+      unsigned u;
+   } alpha;
+   int factor_reg;
+   int tmp;
+
+
+   alpha.f = const_alpha;
+
+   switch (factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      factor_reg = -1;
+      break;
+
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      factor_reg = spe_allocate_available_register(f);
+
+      spe_or(f, factor_reg, src_alpha, src_alpha);
+      break;
+
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      factor_reg = dst_alpha;
+      break;
+
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      factor_reg = -1;
+      break;
+
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      const_alpha = 1.0 - const_alpha;
+      /* FALLTHROUGH */
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      factor_reg = spe_allocate_available_register(f);
+
+      spe_il(f, factor_reg, alpha.u & 0x0ffff);
+      spe_ilh(f, factor_reg, alpha.u >> 16);
+      break;
+
+   case PIPE_BLENDFACTOR_ZERO:
+      factor_reg = -1;
+      break;
+
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      tmp = spe_allocate_available_register(f);
+      factor_reg = spe_allocate_available_register(f);
+
+      spe_il(f, tmp, 1);
+      spe_cuflt(f, tmp, tmp, 0);
+      spe_fs(f, factor_reg, tmp, src_alpha);
+
+      spe_release_register(f, tmp);
+      break;
+
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      tmp = spe_allocate_available_register(f);
+      factor_reg = spe_allocate_available_register(f);
+
+      spe_il(f, tmp, 1);
+      spe_cuflt(f, tmp, tmp, 0);
+      spe_fs(f, factor_reg, tmp, dst_alpha);
+
+      spe_release_register(f, tmp);
+      break;
+
+   case PIPE_BLENDFACTOR_SRC1_ALPHA:
+   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+   default:
+      assert(0);
+      factor_reg = -1;
+      break;
+   }
+
+   return factor_reg;
+}
+
+
+/**
+ * \note Emits a maximum of 5 instructions
+ */
+static void
+emit_color_factor_calculation(struct spe_function *f,
+                              unsigned sF, unsigned mask,
+			      const struct pipe_blend_color *blend_color,
+                              const int *src,
+                              const int *dst,
+                              int *factor)
+{
+   union {
+      float f[4];
+      unsigned u[4];
+   } color;
+   int tmp;
+   unsigned i;
+
+
+   color.f[0] = blend_color->color[0];
+   color.f[1] = blend_color->color[1];
+   color.f[2] = blend_color->color[2];
+   color.f[3] = blend_color->color[3];
+
+   factor[0] = -1;
+   factor[1] = -1;
+   factor[2] = -1;
+   factor[3] = -1;
+
+   switch (sF) {
+   case PIPE_BLENDFACTOR_ONE:
+      break;
+
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      for (i = 0; i < 3; ++i) {
+         if ((mask & (1U << i)) != 0) {
+            factor[i] = spe_allocate_available_register(f);
+            spe_or(f, factor[i], src[i], src[i]);
+         }
+      }
+      break;
+
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      factor[0] = spe_allocate_available_register(f);
+      factor[1] = factor[0];
+      factor[2] = factor[0];
+
+      spe_or(f, factor[0], src[3], src[3]);
+      break;
+
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      factor[0] = dst[3];
+      factor[1] = dst[3];
+      factor[2] = dst[3];
+      break;
+
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      factor[0] = dst[0];
+      factor[1] = dst[1];
+      factor[2] = dst[2];
+      break;
+
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      tmp = spe_allocate_available_register(f);
+      factor[0] = spe_allocate_available_register(f);
+      factor[1] = factor[0];
+      factor[2] = factor[0];
+
+      /* Alpha saturate means min(As, 1-Ad).
+       */
+      spe_il(f, tmp, 1);
+      spe_cuflt(f, tmp, tmp, 0);
+      spe_fs(f, tmp, tmp, dst[3]);
+      spe_fcgt(f, factor[0], tmp, src[3]);
+      spe_selb(f, factor[0], src[3], tmp, factor[0]);
+
+      spe_release_register(f, tmp);
+      break;
+
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      color.f[0] = 1.0 - color.f[0];
+      color.f[1] = 1.0 - color.f[1];
+      color.f[2] = 1.0 - color.f[2];
+      /* FALLTHROUGH */
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      for (i = 0; i < 3; i++) {
+	 factor[i] = spe_allocate_available_register(f);
+
+	 spe_il(f, factor[i], color.u[i] & 0x0ffff);
+	 spe_ilh(f, factor[i], color.u[i] >> 16);
+      }
+      break;
+
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+      color.f[3] = 1.0 - color.f[3];
+      /* FALLTHROUGH */
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+      factor[0] = spe_allocate_available_register(f);
+      factor[1] = factor[0];
+      factor[2] = factor[0];
+
+      spe_il(f, factor[0], color.u[3] & 0x0ffff);
+      spe_ilh(f, factor[0], color.u[3] >> 16);
+      break;
+
+   case PIPE_BLENDFACTOR_ZERO:
+      break;
+
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      tmp = spe_allocate_available_register(f);
+
+      spe_il(f, tmp, 1);
+      spe_cuflt(f, tmp, tmp, 0);
+
+      for (i = 0; i < 3; ++i) {
+         if ((mask & (1U << i)) != 0) {
+            factor[i] = spe_allocate_available_register(f);
+            spe_fs(f, factor[i], tmp, src[i]);
+         }
+      }
+
+      spe_release_register(f, tmp);
+      break;
+
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+      tmp = spe_allocate_available_register(f);
+      factor[0] = spe_allocate_available_register(f);
+      factor[1] = factor[0];
+      factor[2] = factor[0];
+
+      spe_il(f, tmp, 1);
+      spe_cuflt(f, tmp, tmp, 0);
+      spe_fs(f, factor[0], tmp, src[3]);
+
+      spe_release_register(f, tmp);
+      break;
+
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+      tmp = spe_allocate_available_register(f);
+      factor[0] = spe_allocate_available_register(f);
+      factor[1] = factor[0];
+      factor[2] = factor[0];
+
+      spe_il(f, tmp, 1);
+      spe_cuflt(f, tmp, tmp, 0);
+      spe_fs(f, factor[0], tmp, dst[3]);
+
+      spe_release_register(f, tmp);
+      break;
+
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      tmp = spe_allocate_available_register(f);
+
+      spe_il(f, tmp, 1);
+      spe_cuflt(f, tmp, tmp, 0);
+
+      for (i = 0; i < 3; ++i) {
+         if ((mask & (1U << i)) != 0) {
+            factor[i] = spe_allocate_available_register(f);
+            spe_fs(f, factor[i], tmp, dst[i]);
+         }
+      }
+
+      spe_release_register(f, tmp);
+      break;
+
+   case PIPE_BLENDFACTOR_SRC1_COLOR:
+   case PIPE_BLENDFACTOR_SRC1_ALPHA:
+   case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
+   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+   default:
+      assert(0);
+   }
+}
+
+
+static void
+emit_blend_calculation(struct spe_function *f,
+                       unsigned func, unsigned sF, unsigned dF,
+                       int src, int src_factor, int dst, int dst_factor)
+{
+   int tmp = spe_allocate_available_register(f);
+
+   switch (func) {
+   case PIPE_BLEND_ADD:
+      if (sF == PIPE_BLENDFACTOR_ONE) {
+         if (dF == PIPE_BLENDFACTOR_ZERO) {
+            /* Do nothing. */
+         } else if (dF == PIPE_BLENDFACTOR_ONE) {
+            spe_fa(f, src, src, dst);
+         }
+      } else if (sF == PIPE_BLENDFACTOR_ZERO) {
+         if (dF == PIPE_BLENDFACTOR_ZERO) {
+            spe_il(f, src, 0);
+         } else if (dF == PIPE_BLENDFACTOR_ONE) {
+            spe_or(f, src, dst, dst);
+         }
+      } else {
+         spe_fm(f, tmp, dst, dst_factor);
+         spe_fma(f, src, src, src_factor, tmp);
+      }
+      break;
+
+   case PIPE_BLEND_SUBTRACT:
+      if (sF == PIPE_BLENDFACTOR_ONE) {
+         if (dF == PIPE_BLENDFACTOR_ZERO) {
+            /* Do nothing. */
+         } else if (dF == PIPE_BLENDFACTOR_ONE) {
+            spe_fs(f, src, src, dst);
+         }
+      } else if (sF == PIPE_BLENDFACTOR_ZERO) {
+         if (dF == PIPE_BLENDFACTOR_ZERO) {
+            spe_il(f, src, 0);
+         } else if (dF == PIPE_BLENDFACTOR_ONE) {
+            spe_il(f, tmp, 0);
+            spe_fs(f, src, tmp, dst);
+         }
+      } else {
+         spe_fm(f, tmp, dst, dst_factor);
+         spe_fms(f, src, src, src_factor, tmp);
+      }
+      break;
+
+   case PIPE_BLEND_REVERSE_SUBTRACT:
+      if (sF == PIPE_BLENDFACTOR_ONE) {
+         if (dF == PIPE_BLENDFACTOR_ZERO) {
+            spe_il(f, tmp, 0);
+            spe_fs(f, src, tmp, src);
+         } else if (dF == PIPE_BLENDFACTOR_ONE) {
+            spe_fs(f, src, dst, src);
+         }
+      } else if (sF == PIPE_BLENDFACTOR_ZERO) {
+         if (dF == PIPE_BLENDFACTOR_ZERO) {
+            spe_il(f, src, 0);
+         } else if (dF == PIPE_BLENDFACTOR_ONE) {
+            spe_or(f, src, dst, dst);
+         }
+      } else {
+         spe_fm(f, tmp, src, src_factor);
+         spe_fms(f, src, src, dst_factor, tmp);
+      }
+      break;
+
+   case PIPE_BLEND_MIN:
+      spe_cgt(f, tmp, src, dst);
+      spe_selb(f, src, dst, src, tmp);
+      break;
+
+   case PIPE_BLEND_MAX:
+      spe_cgt(f, tmp, src, dst);
+      spe_selb(f, src, src, dst, tmp);
+      break;
+
+   default:
+      assert(0);
+   }
+
+   spe_release_register(f, tmp);
+}
+
+
+/**
+ * Generate code to perform alpha blending on the SPE
+ */
+void
+cell_generate_alpha_blend(struct cell_blend_state *cb,
+			  const struct pipe_blend_color *blend_color)
+{
+   struct pipe_blend_state *const b = &cb->base;
+   struct spe_function *const f = &cb->code;
+
+   /* This code generates a maximum of 3 (source alpha factor)
+    * + 3 (destination alpha factor) + (3 * 5) (source color factor)
+    * + (3 * 5) (destination color factor) + (4 * 2) (blend equation)
+    * + 4 (fragment mask) + 1 (return) = 49 instlructions.  Round up to 64 to
+    * make it a happy power-of-two.
+    */
+   spe_init_func(f, 4 * 64);
+
+
+   const int frag[4] = {
+      spe_allocate_register(f, 3),
+      spe_allocate_register(f, 4),
+      spe_allocate_register(f, 5),
+      spe_allocate_register(f, 6),
+   };
+   const int pixel[4] = {
+      spe_allocate_register(f, 7),
+      spe_allocate_register(f, 8),
+      spe_allocate_register(f, 9),
+      spe_allocate_register(f, 10),
+   };
+   const int mask = spe_allocate_register(f, 11);
+   unsigned func[4];
+   unsigned sF[4];
+   unsigned dF[4];
+   unsigned i;
+   int src_factor[4];
+   int dst_factor[4];
+
+
+   /* Does the selected blend mode make use of the source / destination
+    * color (RGB) blend factors?
+    */
+   boolean need_color_factor = b->blend_enable
+       && (b->rgb_func != PIPE_BLEND_MIN)
+       && (b->rgb_func != PIPE_BLEND_MAX);
+
+   /* Does the selected blend mode make use of the source / destination
+    * alpha blend factors?
+    */
+   boolean need_alpha_factor = b->blend_enable
+       && (b->alpha_func != PIPE_BLEND_MIN)
+       && (b->alpha_func != PIPE_BLEND_MAX);
+
+
+   sF[0] = b->rgb_src_factor;
+   sF[1] = sF[0];
+   sF[2] = sF[0];
+   sF[3] = (b->alpha_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE)
+       ? PIPE_BLENDFACTOR_ONE : b->alpha_src_factor;
+
+   dF[0] = b->rgb_dst_factor;
+   dF[1] = dF[0];
+   dF[2] = dF[0];
+   dF[3] = b->rgb_dst_factor;
+
+
+   /* If alpha writing is enabled and the alpha blend mode requires use of
+    * the alpha factor, calculate the alpha factor.
+    */
+   if (((b->colormask & 8) != 0) && need_alpha_factor) {
+      src_factor[3] = emit_alpha_factor_calculation(f, sF[3],
+						    blend_color->color[3],
+                                                    frag[3], pixel[3]);
+
+      /* If the alpha destination blend factor is the same as the alpha source
+       * blend factor, re-use the previously calculated value.
+       */
+      dst_factor[3] = (dF[3] == sF[3])
+          ? src_factor[3]
+          : emit_alpha_factor_calculation(f, dF[3],
+					  blend_color->color[3],
+                                          frag[3], pixel[3]);
+   }
+
+
+   if (sF[0] == sF[3]) {
+      src_factor[0] = src_factor[3];
+      src_factor[1] = src_factor[3];
+      src_factor[2] = src_factor[3];
+   } else if (sF[0] == dF[3]) {
+      src_factor[0] = dst_factor[3];
+      src_factor[1] = dst_factor[3];
+      src_factor[2] = dst_factor[3];
+   } else if (need_color_factor) {
+      emit_color_factor_calculation(f,
+                                    b->rgb_src_factor,
+                                    b->colormask,
+                                    blend_color,
+                                    frag, pixel, src_factor);
+   }
+
+
+   if (dF[0] == sF[3]) {
+      dst_factor[0] = src_factor[3];
+      dst_factor[1] = src_factor[3];
+      dst_factor[2] = src_factor[3];
+   } else if (dF[0] == dF[3]) {
+      dst_factor[0] = dst_factor[3];
+      dst_factor[1] = dst_factor[3];
+      dst_factor[2] = dst_factor[3];
+   } else if (dF[0] == sF[0]) {
+      dst_factor[0] = src_factor[0];
+      dst_factor[1] = src_factor[1];
+      dst_factor[2] = src_factor[2];
+   } else if (need_color_factor) {
+      emit_color_factor_calculation(f,
+                                    b->rgb_dst_factor,
+                                    b->colormask,
+                                    blend_color,
+                                    frag, pixel, dst_factor);
+   }
+
+
+
+   func[0] = b->rgb_func;
+   func[1] = func[0];
+   func[2] = func[0];
+   func[3] = b->alpha_func;
+
+   for (i = 0; i < 4; ++i) {
+      if ((b->colormask & (1U << i)) != 0) {
+         emit_blend_calculation(f,
+                                func[i], sF[i], dF[i],
+                                frag[i], src_factor[i],
+                                pixel[i], dst_factor[i]);
+         spe_selb(f, frag[i], pixel[i], frag[i], mask);
+      } else {
+         spe_or(f, frag[i], pixel[i], pixel[i]);
+      }
+   }
+
+   spe_bi(f, 0, 0, 0);
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.h b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.h
new file mode 100644
index 00000000000..541c3b3be07
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.h
@@ -0,0 +1,35 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 (including the next
+ * paragraph) 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 NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS 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 CELL_STATE_PER_FRAGMENT_H
+#define CELL_STATE_PER_FRAGMENT_H
+
+extern void
+cell_generate_depth_stencil_test(struct cell_depth_stencil_alpha_state *cdsa);
+
+extern void
+cell_generate_alpha_blend(struct cell_blend_state *cb,
+			  const struct pipe_blend_color *blend_color);
+
+#endif /* CELL_STATE_PER_FRAGMENT_H */
diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile
index c071de1900b..115ca8cd901 100644
--- a/src/gallium/drivers/cell/spu/Makefile
+++ b/src/gallium/drivers/cell/spu/Makefile
@@ -19,6 +19,7 @@ SOURCES = \
 	spu_main.c \
 	spu_blend.c \
 	spu_dcache.c \
+	spu_per_fragment_op.c \
 	spu_render.c \
 	spu_texture.c \
 	spu_tile.c \
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 59300028d4d..937962285d0 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -58,6 +58,9 @@ struct spu_vs_context draw;
 static unsigned char attribute_fetch_code_buffer[136 * PIPE_ATTRIB_MAX]
     ALIGN16_ATTRIB;
 
+static unsigned char depth_stencil_code_buffer[4 * 64]
+    ALIGN16_ATTRIB;
+
 /**
  * Tell the PPU that this SPU has finished copying a buffer to
  * local store and that it may be reused by the PPU.
@@ -215,12 +218,19 @@ cmd_state_framebuffer(const struct cell_command_framebuffer *cmd)
    spu.fb.width_tiles = (spu.fb.width + TILE_SIZE - 1) / TILE_SIZE;
    spu.fb.height_tiles = (spu.fb.height + TILE_SIZE - 1) / TILE_SIZE;
 
-   if (spu.fb.depth_format == PIPE_FORMAT_Z32_UNORM)
+   switch (spu.fb.depth_format) {
+   case PIPE_FORMAT_Z32_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
       spu.fb.zsize = 4;
-   else if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM)
+      break;
+   case PIPE_FORMAT_Z16_UNORM:
       spu.fb.zsize = 2;
-   else
+      break;
+   default:
       spu.fb.zsize = 0;
+      break;
+   }
 
    if (spu.fb.color_format == PIPE_FORMAT_A8R8G8B8_UNORM)
       spu.color_shuffle = ((vector unsigned char) {
@@ -248,14 +258,35 @@ cmd_state_blend(const struct pipe_blend_state *state)
 
 
 static void
-cmd_state_depth_stencil(const struct pipe_depth_stencil_alpha_state *state)
+cmd_state_depth_stencil(const struct cell_command_depth_stencil_alpha_test *state)
 {
    if (Debug)
       printf("SPU %u: DEPTH_STENCIL: ztest %d\n",
              spu.init.id,
-             state->depth.enabled);
+             state->read_depth);
+
+   ASSERT_ALIGN16(state->base);
+
+   if (state->size != 0) {
+      mfc_get(depth_stencil_code_buffer,
+	      (unsigned int) state->base,  /* src */
+	      ROUNDUP16(state->size),
+	      TAG_BATCH_BUFFER,
+	      0, /* tid */
+	      0  /* rid */);
+      wait_on_mask(1 << TAG_BATCH_BUFFER);
+   } else {
+      /* If there is no code, emit a return instruction.
+       */
+      depth_stencil_code_buffer[0] = 0x35;
+      depth_stencil_code_buffer[1] = 0x00;
+      depth_stencil_code_buffer[2] = 0x00;
+      depth_stencil_code_buffer[3] = 0x00;
+   }
 
-   memcpy(&spu.depth_stencil, state, sizeof(*state));
+   spu.frag_test = (frag_test_func) depth_stencil_code_buffer;
+   spu.read_depth = state->read_depth;
+   spu.read_stencil = state->read_stencil;
 }
 
 
@@ -415,9 +446,9 @@ cmd_batch(uint opcode)
          pos += (1 + ROUNDUP8(sizeof(struct pipe_blend_state)) / 8);
          break;
       case CELL_CMD_STATE_DEPTH_STENCIL:
-         cmd_state_depth_stencil((struct pipe_depth_stencil_alpha_state *)
+         cmd_state_depth_stencil((struct cell_command_depth_stencil_alpha_test *)
                                  &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct pipe_depth_stencil_alpha_state)) / 8);
+         pos += (1 + ROUNDUP8(sizeof(struct cell_command_depth_stencil_alpha_test)) / 8);
          break;
       case CELL_CMD_STATE_SAMPLER:
          cmd_state_sampler((struct pipe_sampler_state *) &buffer[pos+1]);
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index a13edd17029..444e2186452 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -56,6 +56,17 @@ typedef union {
 #define TILE_STATUS_GETTING 5  /**< mfc_get() called but not yet arrived */
 
 
+struct spu_frag_test_results {
+   qword mask;
+   qword depth;
+   qword stencil;
+};
+
+typedef struct spu_frag_test_results (*frag_test_func)(qword frag_mask,
+    qword pixel_depth, qword pixel_stencil, qword frag_depth,
+    qword frag_alpha, qword facing);
+
+
 struct spu_framebuffer {
    void *color_start;              /**< addr of color surface in main memory */
    void *depth_start;              /**< addr of depth surface in main memory */
@@ -79,8 +90,9 @@ struct spu_global
    struct cell_init_info init;
 
    struct spu_framebuffer fb;
-   struct pipe_blend_state blend_stencil;
-   struct pipe_depth_stencil_alpha_state depth_stencil;
+   boolean read_depth;
+   boolean read_stencil;
+   frag_test_func frag_test;
    struct pipe_blend_state blend;
    struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS];
    struct cell_command_texture texture;
diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
new file mode 100644
index 00000000000..b4cffeeb32a
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
@@ -0,0 +1,211 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 (including the next
+ * paragraph) 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 NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS 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.
+ */
+
+/**
+ * \file spu_per_fragment_op.c
+ * SPU implementation various per-fragment operations.
+ *
+ * \author Ian Romanick <[email protected]>
+ */
+
+#include "pipe/p_format.h"
+#include "spu_main.h"
+#include "spu_per_fragment_op.h"
+
+#define ZERO 0x80
+
+static void
+read_ds_quad(tile_t *buffer, unsigned x, unsigned y,
+             enum pipe_format depth_format, qword *depth,
+             qword *stencil)
+{
+   const int ix = x / 2;
+   const int iy = y / 2;
+
+   switch (depth_format) {
+   case PIPE_FORMAT_Z16_UNORM: {
+      qword *ptr = (qword *) &buffer->us8[iy][ix / 2];
+
+      const qword shuf_vec = (qword) {
+         ZERO, ZERO, 0, 1, ZERO, ZERO, 2, 3,
+         ZERO, ZERO, 4, 5, ZERO, ZERO, 6, 7
+      };
+
+
+      /* At even X values we want the first 4 shorts, and at odd X values we
+       * want the second 4 shorts.
+       */
+      qword bias = (qword) spu_splats((unsigned char) ((ix & 0x01) << 3));
+      qword bias_mask = si_fsmbi(0x3333);
+      qword sv = si_a(shuf_vec, si_and(bias_mask, bias));
+
+      *depth = si_shufb(*ptr, *ptr, sv);
+      *stencil = si_il(0);
+      break;
+   }
+
+
+   case PIPE_FORMAT_Z32_UNORM: {
+      qword *ptr = (qword *) &buffer->ui4[iy][ix];
+
+      *depth = *ptr;
+      *stencil = si_il(0);
+      break;
+   }
+      
+
+   case PIPE_FORMAT_Z24S8_UNORM: {
+      qword *ptr = (qword *) &buffer->ui4[iy][ix];
+      qword mask = si_fsmbi(0xEEEE);
+
+      *depth = si_rotmai(si_and(*ptr, mask), -8);
+      *stencil = si_andc(*ptr, mask);
+      break;
+   }
+
+
+   case PIPE_FORMAT_S8Z24_UNORM: {
+      qword *ptr = (qword *) &buffer->ui4[iy][ix];
+
+      *depth = si_and(*ptr, si_fsmbi(0x7777));
+      *stencil = si_andi(si_roti(*ptr, 8), 0x0ff);
+      break;
+   }
+
+
+   default:
+      assert(0);
+      break;
+   }
+}
+
+
+static void
+write_ds_quad(tile_t *buffer, unsigned x, unsigned y,
+              enum pipe_format depth_format,
+              qword depth, qword stencil)
+{
+   const int ix = x / 2;
+   const int iy = y / 2;
+
+   (void) stencil;
+
+   switch (depth_format) {
+   case PIPE_FORMAT_Z16_UNORM: {
+      qword *ptr = (qword *) &buffer->us8[iy][ix / 2];
+
+      qword sv = ((ix & 0x01) == 0) 
+          ? (qword) { 2, 3, 6, 7, 10, 11, 14, 15,
+                      24, 25, 26, 27, 28, 29, 30, 31 }
+          : (qword) { 16, 17, 18, 19, 20 , 21, 22, 23,
+                      2, 3, 6, 7, 10, 11, 14, 15 };
+      *ptr = si_shufb(depth, *ptr, sv);
+      break;
+   }
+
+
+   case PIPE_FORMAT_Z32_UNORM: {
+      qword *ptr = (qword *) &buffer->ui4[iy][ix];
+      *ptr = depth;
+      break;
+   }
+
+
+   case PIPE_FORMAT_Z24S8_UNORM: {
+      qword *ptr = (qword *) &buffer->ui4[iy][ix];
+      qword mask = si_fsmbi(0xEEEE);
+
+      depth = si_shli(depth, 8);
+      *ptr = si_selb(stencil, depth, mask);
+      break;
+   }
+
+
+   case PIPE_FORMAT_S8Z24_UNORM: {
+      qword *ptr = (qword *) &buffer->ui4[iy][ix];
+      qword mask = si_fsmbi(0x7777);
+
+      stencil = si_shli(stencil, 24);
+      *ptr = si_selb(stencil, depth, mask);
+      break;
+   }
+
+
+   default:
+      assert(0);
+      break;
+   }
+}
+
+
+qword
+spu_do_depth_stencil(int x, int y,
+                     qword frag_mask, qword frag_depth, qword frag_alpha,
+                     qword facing)
+{
+   struct spu_frag_test_results  result;
+   qword pixel_depth;
+   qword pixel_stencil;
+
+   /* All of this preable code (everthing before the call to frag_test) should
+    * be generated on the PPU and upload to the SPU.
+    */
+   if (spu.read_depth || spu.read_stencil) {
+      read_ds_quad(&spu.ztile, x, y, spu.fb.depth_format,
+                   &pixel_depth, &pixel_stencil);
+   }
+   
+   switch (spu.fb.depth_format) {
+   case PIPE_FORMAT_Z16_UNORM:
+      frag_depth = si_fm(frag_depth, (qword)spu_splats((float)(0x0000ffffu)));
+      frag_depth = si_cfltu(frag_depth, 0);
+      break;
+   case PIPE_FORMAT_Z32_UNORM:
+      frag_depth = si_fm(frag_depth, (qword)spu_splats((float)(0xffffffffu)));
+      frag_depth = si_cfltu(frag_depth, 0);
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
+      frag_depth = si_fm(frag_depth, (qword)spu_splats((float)(0x00ffffffu)));
+      frag_depth = si_cfltu(frag_depth, 0);
+      break;
+   default:
+      ASSERT(0);
+      break;
+   }
+
+   result = (*spu.frag_test)(frag_mask, pixel_depth, pixel_stencil,
+                             frag_depth, frag_alpha, facing);
+
+
+   /* This code (everthing after the call to frag_test) should
+    * be generated on the PPU and upload to the SPU.
+    */
+   if (spu.read_depth || spu.read_stencil) {
+      write_ds_quad(&spu.ztile, x, y, spu.fb.depth_format,
+                    result.depth, result.stencil);
+   }
+
+   return result.mask;
+}
diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.h b/src/gallium/drivers/cell/spu/spu_per_fragment_op.h
new file mode 100644
index 00000000000..65712586992
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.h
@@ -0,0 +1,32 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 (including the next
+ * paragraph) 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 NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS 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 SPU_PER_FRAGMENT_OP
+#define SPU_PER_FRAGMENT_OP
+
+extern qword
+spu_do_depth_stencil(int x, int y, qword frag_mask, qword frag_depth,
+		     qword frag_alpha, qword facing);
+
+#endif /* SPU_PER_FRAGMENT_OP */
diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c
index 20e77aa2e63..6df59abd36d 100644
--- a/src/gallium/drivers/cell/spu/spu_render.c
+++ b/src/gallium/drivers/cell/spu/spu_render.c
@@ -98,7 +98,7 @@ my_tile(uint tx, uint ty)
 static INLINE void
 get_cz_tiles(uint tx, uint ty)
 {
-   if (spu.depth_stencil.depth.enabled) {
+   if (spu.read_depth) {
       if (spu.cur_ztile_status != TILE_STATUS_CLEAR) {
          //printf("SPU %u: getting Z tile %u, %u\n", spu.init.id, tx, ty);
          get_tile(tx, ty, &spu.ztile, TAG_READ_TILE_Z, 1);
@@ -153,7 +153,7 @@ static INLINE void
 wait_put_cz_tiles(void)
 {
    wait_on_mask(1 << TAG_WRITE_TILE_COLOR);
-   if (spu.depth_stencil.depth.enabled) {
+   if (spu.read_depth) {
       wait_on_mask(1 << TAG_WRITE_TILE_Z);
    }
 }
diff --git a/src/gallium/drivers/cell/spu/spu_tile.h b/src/gallium/drivers/cell/spu/spu_tile.h
index 3105b848fdc..1b5491112db 100644
--- a/src/gallium/drivers/cell/spu/spu_tile.h
+++ b/src/gallium/drivers/cell/spu/spu_tile.h
@@ -56,13 +56,13 @@ clear_c_tile(tile_t *ctile)
 static INLINE void
 clear_z_tile(tile_t *ztile)
 {
-   if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM) {
+   if (spu.fb.zsize == 2) {
       memset16((ushort*) ztile->us,
                spu.fb.depth_clear_value,
                TILE_SIZE * TILE_SIZE);
    }
    else {
-      ASSERT(spu.fb.depth_format == PIPE_FORMAT_Z32_UNORM);
+      ASSERT(spu.fb.zsize != 0);
       memset32((uint*) ztile->ui,
                spu.fb.depth_clear_value,
                TILE_SIZE * TILE_SIZE);
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index be9624cf7d9..81823f24633 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -38,8 +38,7 @@
 #include "spu_texture.h"
 #include "spu_tile.h"
 #include "spu_tri.h"
-
-#include "spu_ztest.h"
+#include "spu_per_fragment_op.h"
 
 
 /** Masks are uint[4] vectors with each element being 0 or 0xffffffff */
@@ -264,16 +263,12 @@ do_depth_test(int x, int y, mask_t quadmask)
 
    zvals.v = eval_z((float) x, (float) y);
 
-   if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM) {
-      int ix = (x - setup.cliprect_minx) / 4;
-      int iy = (y - setup.cliprect_miny) / 2;
-      mask = spu_z16_test_less(zvals.v, &spu.ztile.us8[iy][ix], x>>1, quadmask);
-   }
-   else {
-      int ix = (x - setup.cliprect_minx) / 2;
-      int iy = (y - setup.cliprect_miny) / 2;
-      mask = spu_z32_test_less(zvals.v, &spu.ztile.ui4[iy][ix], quadmask);
-   }
+   mask = (mask_t) spu_do_depth_stencil(x - setup.cliprect_minx,
+					y - setup.cliprect_miny,
+					(qword) quadmask, 
+					(qword) zvals.v,
+					(qword) spu_splats((unsigned char) 0x0ffu),
+					(qword) spu_splats((unsigned int) 0x01u));
 
    if (spu_extract(spu_orx(mask), 0))
       spu.cur_ztile_status = TILE_STATUS_DIRTY;
@@ -299,7 +294,7 @@ emit_quad( int x, int y, mask_t mask )
    sp->quad.first->run(sp->quad.first, &setup.quad);
 #else
 
-   if (spu.depth_stencil.depth.enabled) {
+   if (spu.read_depth) {
       mask = do_depth_test(x, y, mask);
    }
 
@@ -434,7 +429,7 @@ static void flush_spans( void )
    }
    ASSERT(spu.cur_ctile_status != TILE_STATUS_DEFINED);
 
-   if (spu.depth_stencil.depth.enabled) {
+   if (spu.read_depth) {
       if (spu.cur_ztile_status == TILE_STATUS_GETTING) {
          /* wait for mfc_get() to complete */
          //printf("SPU: %u: waiting for ztile\n", spu.init.id);
diff --git a/src/gallium/drivers/cell/spu/spu_ztest.h b/src/gallium/drivers/cell/spu/spu_ztest.h
deleted file mode 100644
index ce8ad003393..00000000000
--- a/src/gallium/drivers/cell/spu/spu_ztest.h
+++ /dev/null
@@ -1,135 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
- * 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, sub license, 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 (including the
- * next paragraph) 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 NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
- * 
- **************************************************************************/
-
-
-/**
- * Zbuffer/depth test code.
- */
-
-
-#ifndef SPU_ZTEST_H
-#define SPU_ZTEST_H
-
-
-#ifdef __SPU__
-#include <spu_intrinsics.h>
-#endif
-
-
-
-/**
- * Perform Z testing for a 16-bit/value Z buffer.
- *
- * \param zvals  vector of four fragment zvalues as floats
- * \param zbuf   ptr to vector of ushort[8] zbuffer values.  Note that this
- *               contains the Z values for 2 quads, 8 pixels.
- * \param x      x coordinate of quad (only lsbit is significant)
- * \param inMask indicates which fragments in the quad are alive
- * \return new mask indicating which fragments are alive after ztest
- */
-static INLINE vector unsigned int
-spu_z16_test_less(vector float zvals, vector unsigned short *zbuf,
-                  uint x, vector unsigned int inMask)
-{
-#define ZERO 0x80
-   vector unsigned int zvals_ui4, zbuf_ui4, mask;
-
-   /* convert floats to uints in [0, 65535] */
-   zvals_ui4 = spu_convtu(zvals, 32); /* convert to [0, 2^32] */
-   zvals_ui4 = spu_rlmask(zvals_ui4, -16);  /* right shift 16 */
-
-   /* XXX this conditional could be removed with a bit of work */
-   if (x & 1) {
-      /* convert zbuffer values from ushorts to uints */
-      /* gather lower four ushorts */
-      zbuf_ui4 = spu_shuffle((vector unsigned int) *zbuf,
-                             (vector unsigned int) *zbuf,
-                             ((vector unsigned char) {
-                                ZERO, ZERO,  8,  9, ZERO, ZERO, 10, 11,
-                                ZERO, ZERO, 12, 13, ZERO, ZERO, 14, 15}));
-      /* mask = (zbuf_ui4 < zvals_ui4) ? ~0 : 0 */
-      mask = spu_cmpgt(zbuf_ui4, zvals_ui4);
-      /* mask &= inMask */
-      mask = spu_and(mask, inMask);
-      /* zbuf = mask ? zval : zbuf */
-      zbuf_ui4 = spu_sel(zbuf_ui4, zvals_ui4, mask);
-      /* convert zbuffer values from uints back to ushorts, preserve lower 4 */
-      *zbuf = (vector unsigned short)
-         spu_shuffle(zbuf_ui4, (vector unsigned int) *zbuf,
-                     ((vector unsigned char) {
-                        16, 17, 18, 19, 20, 21, 22, 23,
-                        2, 3, 6, 7, 10, 11, 14, 15}));
-   }
-   else {
-      /* convert zbuffer values from ushorts to uints */
-      /* gather upper four ushorts */
-      zbuf_ui4 = spu_shuffle((vector unsigned int) *zbuf,
-                             (vector unsigned int) *zbuf,
-                             ((vector unsigned char) {
-                                ZERO, ZERO, 0, 1, ZERO, ZERO, 2, 3,
-                                ZERO, ZERO, 4, 5, ZERO, ZERO, 6, 7}));
-      /* mask = (zbuf_ui4 < zvals_ui4) ? ~0 : 0 */
-      mask = spu_cmpgt(zbuf_ui4, zvals_ui4);
-      /* mask &= inMask */
-      mask = spu_and(mask, inMask);
-      /* zbuf = mask ? zval : zbuf */
-      zbuf_ui4 = spu_sel(zbuf_ui4, zvals_ui4, mask);
-      /* convert zbuffer values from uints back to ushorts, preserve upper 4 */
-      *zbuf = (vector unsigned short)
-         spu_shuffle(zbuf_ui4, (vector unsigned int) *zbuf,
-                     ((vector unsigned char) {
-                        2, 3, 6, 7, 10, 11, 14, 15,
-                        24, 25, 26, 27, 28, 29, 30, 31}));
-   }
-   return mask;
-#undef ZERO
-}
-
-
-/**
- * As above, but Zbuffer values as 32-bit uints
- */
-static INLINE vector unsigned int
-spu_z32_test_less(vector float zvals, vector unsigned int *zbuf_ptr,
-                  vector unsigned int inMask)
-{
-   vector unsigned int zvals_ui4, mask, zbuf = *zbuf_ptr;
-
-   /* convert floats to uints in [0, 0xffffffff] */
-   zvals_ui4 = spu_convtu(zvals, 32);
-   /* mask = (zbuf < zvals_ui4) ? ~0 : 0 */
-   mask = spu_cmpgt(zbuf, zvals_ui4);
-   /* mask &= inMask */
-   mask = spu_and(mask, inMask);
-   /* zbuf = mask ? zval : zbuf */
-   *zbuf_ptr = spu_sel(zbuf, zvals_ui4, mask);
-
-   return mask;
-}
-
-
-#endif /* SPU_ZTEST_H */
diff --git a/src/gallium/drivers/i915simple/i915_prim_vbuf.c b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
index 9d5f609220a..eb64f51943b 100644
--- a/src/gallium/drivers/i915simple/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
@@ -116,7 +116,7 @@ i915_vbuf_render_allocate_vertices( struct vbuf_render *render,
 }
 
 
-static void 
+static boolean
 i915_vbuf_render_set_primitive( struct vbuf_render *render, 
                                 unsigned prim )
 {
@@ -125,15 +125,17 @@ i915_vbuf_render_set_primitive( struct vbuf_render *render,
    switch(prim) {
    case PIPE_PRIM_POINTS:
       i915_render->hwprim = PRIM3D_POINTLIST;
-      break;
+      return TRUE;
    case PIPE_PRIM_LINES:
       i915_render->hwprim = PRIM3D_LINELIST;
-      break;
+      return TRUE;
    case PIPE_PRIM_TRIANGLES:
       i915_render->hwprim = PRIM3D_TRILIST;
-      break;
+      return TRUE;
    default:
-      assert(0);
+      /* Actually, can handle a lot more just fine...  Fixme.
+       */
+      return FALSE;
    }
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index db0913cb2bc..d940718ed2b 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -101,11 +101,20 @@ sp_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices,
 }
 
 
-static void
+static boolean
 sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
 {
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
-   cvbr->prim = prim;
+   if (prim == PIPE_PRIM_TRIANGLES ||
+       prim == PIPE_PRIM_LINES ||
+       prim == PIPE_PRIM_POINTS) {
+      cvbr->prim = prim;
+      return TRUE;
+   }
+   else {
+      return FALSE;
+   }
+      
 }
 
 
@@ -207,6 +216,27 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
    (struct vertex_header *) ((char *) vertex_buffer + (I) * vertex_size)
 
    switch (cvbr->prim) {
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < nr; i++) {
+         prim.v[0] = VERTEX(i);
+         setup->point( setup, &prim );
+      }
+      break;
+   case PIPE_PRIM_LINES:
+      assert(nr % 2 == 0);
+      for (i = 0; i < nr; i += 2) {
+         prim.v[0] = VERTEX(i);
+         prim.v[1] = VERTEX(i + 1);
+         setup->line( setup, &prim );
+      }
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+      for (i = 1; i < nr; i++) {
+         prim.v[0] = VERTEX(i - 1);
+         prim.v[1] = VERTEX(i);
+         setup->line( setup, &prim );
+      }
+      break;
    case PIPE_PRIM_TRIANGLES:
       assert(nr % 3 == 0);
       for (i = 0; i < nr; i += 3) {
@@ -217,6 +247,58 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
          setup->tri( setup, &prim );
       }
       break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      assert(nr >= 3);
+      for (i = 2; i < nr; i++) {
+         prim.v[0] = VERTEX(i - 2);
+         prim.v[1] = VERTEX(i - 1);
+         prim.v[2] = VERTEX(i);
+         calc_det(&prim);
+         setup->tri( setup, &prim );
+      }
+      break;
+   case PIPE_PRIM_TRIANGLE_FAN:
+      assert(nr >= 3);
+      for (i = 2; i < nr; i++) {
+         prim.v[0] = VERTEX(0);
+         prim.v[1] = VERTEX(i - 1);
+         prim.v[2] = VERTEX(i);
+         calc_det(&prim);
+         setup->tri( setup, &prim );
+      }
+      break;
+   case PIPE_PRIM_QUADS:
+      assert(nr % 4 == 0);
+      for (i = 0; i < nr; i += 4) {
+         prim.v[0] = VERTEX(i + 0);
+         prim.v[1] = VERTEX(i + 1);
+         prim.v[2] = VERTEX(i + 2);
+         calc_det(&prim);
+         setup->tri( setup, &prim );
+
+         prim.v[0] = VERTEX(i + 0);
+         prim.v[1] = VERTEX(i + 2);
+         prim.v[2] = VERTEX(i + 3);
+         calc_det(&prim);
+         setup->tri( setup, &prim );
+      }
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      assert(nr >= 4);
+      for (i = 2; i < nr; i += 2) {
+         prim.v[0] = VERTEX(i - 2);
+         prim.v[1] = VERTEX(i);
+         prim.v[2] = VERTEX(i + 1);
+         calc_det(&prim);
+         setup->tri( setup, &prim );
+
+         prim.v[0] = VERTEX(i - 2);
+         prim.v[1] = VERTEX(i + 1);
+         prim.v[2] = VERTEX(i - 1);
+         calc_det(&prim);
+         setup->tri( setup, &prim );
+      }
+      break;
    case PIPE_PRIM_POLYGON:
       /* draw as tri fan */
       for (i = 2; i < nr; i++) {
diff --git a/src/gallium/include/pipe/p_debug.h b/src/gallium/include/pipe/p_debug.h
index f971ad3adc0..f3dfa062161 100644
--- a/src/gallium/include/pipe/p_debug.h
+++ b/src/gallium/include/pipe/p_debug.h
@@ -158,6 +158,16 @@ void debug_mask_vprintf(uint32_t uuid,
                         const char *format, 
                         va_list ap);
 
+
+#ifdef DEBUG
+#define debug_warning(__msg) \
+   debug_printf("%s:%i:warning: %s\n", __FILE__, __LINE__, (__msg)) 
+#else
+#define debug_warning(__msg) \
+   ((void)0) 
+#endif
+
+
 #ifdef	__cplusplus
 }
 #endif
diff --git a/src/gallium/include/pipe/p_util.h b/src/gallium/include/pipe/p_util.h
index ef36ce75f7d..c2e0f8c6a5d 100644
--- a/src/gallium/include/pipe/p_util.h
+++ b/src/gallium/include/pipe/p_util.h
@@ -88,14 +88,16 @@ FREE( void *ptr )
 static INLINE void *
 REALLOC( void *old_ptr, unsigned old_size, unsigned new_size )
 {
-   void *new_ptr;
-   if( new_size <= old_size ) {
-      return old_ptr;
-   }
-   new_ptr = MALLOC( new_size );
-   if( new_ptr ) {
-      memcpy( new_ptr, old_ptr, old_size );
+   void *new_ptr = NULL;
+
+   if (new_size != 0) {
+      new_ptr = MALLOC( new_size );
+      
+      if( new_ptr && old_ptr ) {
+         memcpy( new_ptr, old_ptr, old_size );
+      }
    }
+
    FREE( old_ptr );
    return new_ptr;
 }
diff --git a/src/mesa/shader/arbprogparse.c b/src/mesa/shader/arbprogparse.c
index e385b9d9973..9e5169eff17 100644
--- a/src/mesa/shader/arbprogparse.c
+++ b/src/mesa/shader/arbprogparse.c
@@ -1576,9 +1576,6 @@ parse_attrib_binding(GLcontext * ctx, const GLubyte ** inst,
    if (err) {
       program_error(ctx, Program->Position, "Bad attribute binding");
    }
-   else {
-      Program->Base.InputsRead |= (1 << *inputReg);
-   }
 
    return err;
 }
@@ -2557,6 +2554,11 @@ parse_src_reg (GLcontext * ctx, const GLubyte ** inst,
          return 1;
    }
 
+   /* Add attributes to InputsRead only if they are used the program.
+    * This avoids the handling of unused ATTRIB declarations in the drivers. */
+   if (*File == PROGRAM_INPUT)
+      Program->Base.InputsRead |= (1 << *Index);
+
    return 0;
 }
 
diff --git a/src/mesa/sources b/src/mesa/sources
index e3d5f228493..287af7121aa 100644
--- a/src/mesa/sources
+++ b/src/mesa/sources
@@ -172,6 +172,7 @@ STATETRACKER_SOURCES = \
 	state_tracker/st_atom_texture.c \
 	state_tracker/st_atom_viewport.c \
 	state_tracker/st_cb_accum.c \
+	state_tracker/st_cb_blit.c \
 	state_tracker/st_cb_bufferobjects.c \
 	state_tracker/st_cb_clear.c \
 	state_tracker/st_cb_flush.c \
diff --git a/src/mesa/state_tracker/st_cb_blit.c b/src/mesa/state_tracker/st_cb_blit.c
new file mode 100644
index 00000000000..dfa79c975c1
--- /dev/null
+++ b/src/mesa/state_tracker/st_cb_blit.c
@@ -0,0 +1,125 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Brian Paul
+  */
+
+#include "main/imports.h"
+#include "main/image.h"
+#include "main/macros.h"
+#include "main/texformat.h"
+#include "shader/program.h"
+#include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
+
+#include "st_context.h"
+#include "st_program.h"
+#include "st_cb_drawpixels.h"
+#include "st_cb_blit.h"
+#include "st_cb_fbo.h"
+
+#include "util/u_blit.h"
+
+#include "cso_cache/cso_context.h"
+
+
+void
+st_init_blit(struct st_context *st)
+{
+   st->blit = util_create_blit(st->pipe);
+}
+
+
+void
+st_destroy_blit(struct st_context *st)
+{
+   util_destroy_blit(st->blit);
+   st->blit = NULL;
+}
+
+
+static void
+st_BlitFramebuffer(GLcontext *ctx,
+                   GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
+                   GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
+                   GLbitfield mask, GLenum filter)
+{
+   struct st_context *st = ctx->st;
+   struct pipe_context *pipe = st->pipe;
+
+   const uint pFilter = ((filter == GL_NEAREST)
+                         ? PIPE_TEX_MIPFILTER_NEAREST
+                         : PIPE_TEX_MIPFILTER_LINEAR);
+
+   if (mask & GL_COLOR_BUFFER_BIT) {
+      struct st_renderbuffer *srcRb = 
+         st_renderbuffer(ctx->ReadBuffer->_ColorReadBuffer);
+      struct st_renderbuffer *dstRb = 
+         st_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[0][0]);
+      struct pipe_surface *srcSurf = srcRb->surface;
+      struct pipe_surface *dstSurf = dstRb->surface;
+
+      srcY0 = srcRb->Base.Height - srcY0;
+      srcY1 = srcRb->Base.Height - srcY1;
+
+      dstY0 = dstRb->Base.Height - dstY0;
+      dstY1 = dstRb->Base.Height - dstY1;
+
+      util_blit_pixels(st->blit,
+                       srcSurf, srcX0, srcY0, srcX1, srcY1,
+                       dstSurf, dstX0, dstY0, dstX1, dstY1,
+                       0.0, pFilter);
+
+   }
+
+#if 0
+   /* XXX is this sufficient? */
+   st_invalidate_state(ctx, _NEW_COLOR | _NEW_TEXTURE);
+#else
+   /* need to "unset" cso state because we went behind the back of the cso
+    * tracker.  Without unset, the _set_ calls would be no-ops.
+    */
+   cso_unset_blend(st->cso_context);
+   cso_unset_depth_stencil_alpha(st->cso_context);
+   cso_unset_rasterizer(st->cso_context);
+   cso_set_blend(st->cso_context, &st->state.blend);
+   cso_set_depth_stencil_alpha(st->cso_context, &st->state.depth_stencil);
+   cso_set_rasterizer(st->cso_context, &st->state.rasterizer);
+   pipe->bind_fs_state(pipe, st->fp->driver_shader);
+   pipe->bind_vs_state(pipe, st->vp->driver_shader);
+#endif
+}
+
+
+
+void
+st_init_blit_functions(struct dd_function_table *functions)
+{
+   functions->BlitFramebuffer = st_BlitFramebuffer;
+}
diff --git a/src/mesa/state_tracker/st_cb_blit.h b/src/mesa/state_tracker/st_cb_blit.h
new file mode 100644
index 00000000000..ed22986b531
--- /dev/null
+++ b/src/mesa/state_tracker/st_cb_blit.h
@@ -0,0 +1,46 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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, sub license, 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 (including the
+ * next paragraph) 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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 ST_CB_BLIT_H
+#define ST_CB_BLIT_H
+
+
+#include "st_context.h"
+
+
+
+extern void
+st_init_blit(struct st_context *st);
+
+extern void
+st_destroy_blit(struct st_context *st);
+
+extern void
+st_init_blit_functions(struct dd_function_table *functions);
+
+
+#endif /* ST_CB_BLIT_H */
diff --git a/src/mesa/state_tracker/st_cb_clear.c b/src/mesa/state_tracker/st_cb_clear.c
index 58650714398..693cddedf75 100644
--- a/src/mesa/state_tracker/st_cb_clear.c
+++ b/src/mesa/state_tracker/st_cb_clear.c
@@ -48,6 +48,7 @@
 #include "pipe/p_state.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_winsys.h"
+#include "util/u_pack_color.h"
 
 #include "cso_cache/cso_context.h"
 
@@ -56,55 +57,6 @@
 #define TEST_DRAW_PASSTHROUGH 0
 
 
-static GLuint
-color_value(enum pipe_format pipeFormat, const GLfloat color[4])
-{
-   GLubyte r, g, b, a;
-
-   UNCLAMPED_FLOAT_TO_UBYTE(r, color[0]);
-   UNCLAMPED_FLOAT_TO_UBYTE(g, color[1]);
-   UNCLAMPED_FLOAT_TO_UBYTE(b, color[2]);
-   UNCLAMPED_FLOAT_TO_UBYTE(a, color[3]);
-
-   switch (pipeFormat) {
-   case PIPE_FORMAT_R8G8B8A8_UNORM:
-      return (r << 24) | (g << 16) | (b << 8) | a;
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
-      return (a << 24) | (r << 16) | (g << 8) | b;
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
-      return (b << 24) | (g << 16) | (r << 8) | a;
-   case PIPE_FORMAT_R5G6B5_UNORM:
-      return ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
-   default:
-      assert(0);
-      return 0;
-   }
-}
- 
-
-static uint
-depth_value(enum pipe_format pipeFormat, GLfloat value)
-{
-   switch (pipeFormat) {
-   case PIPE_FORMAT_Z16_UNORM:
-      return (uint) (value * 0xffff);
-   case PIPE_FORMAT_Z32_UNORM:
-      /* special-case to avoid overflow */
-      if (value == 1.0)
-         return 0xffffffff;
-      else
-         return (uint) (value * 0xffffffff);
-   case PIPE_FORMAT_S8Z24_UNORM:
-      return (uint) (value * 0xffffff);
-   case PIPE_FORMAT_Z24S8_UNORM:
-      return ((uint) (value * 0xffffff)) << 8;
-   default:
-      assert(0);
-      return 0;
-   }
-}
-
-
 static GLboolean
 is_depth_stencil_format(enum pipe_format pipeFormat)
 {
@@ -405,6 +357,8 @@ clear_with_quad(GLcontext *ctx,
    st_invalidate_state(ctx, _NEW_COLOR | _NEW_DEPTH | _NEW_STENCIL);
 #else
    /* Restore pipe state */
+   cso_set_blend(st->cso_context, &st->state.blend);
+   cso_set_depth_stencil_alpha(st->cso_context, &st->state.depth_stencil);
    cso_set_rasterizer(st->cso_context, &st->state.rasterizer);
    pipe->bind_fs_state(pipe, st->fp->driver_shader);
    pipe->bind_vs_state(pipe, st->vp->driver_shader);
@@ -518,7 +472,6 @@ check_clear_stencil_with_quad(GLcontext *ctx, struct gl_renderbuffer *rb)
 
 
 
-
 static void
 clear_color_buffer(GLcontext *ctx, struct gl_renderbuffer *rb)
 {
@@ -527,10 +480,10 @@ clear_color_buffer(GLcontext *ctx, struct gl_renderbuffer *rb)
       clear_with_quad(ctx, GL_TRUE, GL_FALSE, GL_FALSE);
    }
    else {
-      struct st_renderbuffer *strb = st_renderbuffer(rb);
-
       /* clear whole buffer w/out masking */
-      uint clearValue = color_value(strb->surface->format, ctx->Color.ClearColor);
+      struct st_renderbuffer *strb = st_renderbuffer(rb);
+      uint clearValue;
+      util_pack_color(ctx->Color.ClearColor, strb->surface->format, &clearValue);
       ctx->st->pipe->clear(ctx->st->pipe, strb->surface, clearValue);
    }
 }
@@ -547,7 +500,7 @@ clear_depth_buffer(GLcontext *ctx, struct gl_renderbuffer *rb)
       struct st_renderbuffer *strb = st_renderbuffer(rb);
 
       /* simple clear of whole buffer */
-      uint clearValue = depth_value(strb->surface->format, ctx->Depth.Clear);
+      uint clearValue = util_pack_z(strb->surface->format, ctx->Depth.Clear);
       ctx->st->pipe->clear(ctx->st->pipe, strb->surface, clearValue);
    }
 }
@@ -591,7 +544,7 @@ clear_depth_stencil_buffer(GLcontext *ctx, struct gl_renderbuffer *rb)
       struct st_renderbuffer *strb = st_renderbuffer(rb);
 
       /* clear whole buffer w/out masking */
-      GLuint clearValue = depth_value(strb->surface->format, ctx->Depth.Clear);
+      GLuint clearValue = util_pack_z(strb->surface->format, ctx->Depth.Clear);
 
       switch (strb->surface->format) {
       case PIPE_FORMAT_S8Z24_UNORM:
diff --git a/src/mesa/state_tracker/st_context.c b/src/mesa/state_tracker/st_context.c
index 5458ab420e4..e1fc885e0e7 100644
--- a/src/mesa/state_tracker/st_context.c
+++ b/src/mesa/state_tracker/st_context.c
@@ -36,6 +36,7 @@
 #include "st_context.h"
 #include "st_cb_accum.h"
 #include "st_cb_bufferobjects.h"
+#include "st_cb_blit.h"
 #include "st_cb_clear.h"
 #include "st_cb_drawpixels.h"
 #include "st_cb_fbo.h"
@@ -100,6 +101,7 @@ st_create_context_priv( GLcontext *ctx, struct pipe_context *pipe )
    st_init_atoms( st );
    st_init_draw( st );
    st_init_generate_mipmap(st);
+   st_init_blit(st);
 
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
       st->state.sampler_list[i] = &st->state.samplers[i];
@@ -151,6 +153,8 @@ static void st_destroy_context_priv( struct st_context *st )
    draw_destroy(st->draw);
    st_destroy_atoms( st );
    st_destroy_draw( st );
+   st_destroy_generate_mipmap(st);
+   st_destroy_blit(st);
 
    _vbo_DestroyContext(st->ctx);
 
@@ -217,6 +221,7 @@ void st_init_driver_functions(struct dd_function_table *functions)
 
    st_init_accum_functions(functions);
    st_init_bufferobject_functions(functions);
+   st_init_blit_functions(functions);
    st_init_clear_functions(functions);
    st_init_drawpixels_functions(functions);
    st_init_fbo_functions(functions);
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index e81aebba3d2..63150dbeaf6 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -40,6 +40,9 @@ struct draw_context;
 struct draw_stage;
 struct cso_cache;
 struct cso_blend;
+struct gen_mipmap_state;
+struct blit_state;
+
 
 #define ST_NEW_MESA                    0x1 /* Mesa state has changed */
 #define ST_NEW_FRAGMENT_PROGRAM        0x2
@@ -146,18 +149,8 @@ struct st_context
       struct st_fragment_program *combined_prog;
    } bitmap;
 
-   /** For gen/render mipmap feature */
-   struct {
-      struct pipe_blend_state blend;
-      struct pipe_depth_stencil_alpha_state depthstencil;
-      struct pipe_rasterizer_state rasterizer;
-
-      void *blend_cso;
-      void *depthstencil_cso;
-      void *rasterizer_cso;
-      struct st_fragment_program *stfp;
-      struct st_vertex_program *stvp;
-   } gen_mipmap;
+   struct gen_mipmap_state *gen_mipmap;
+   struct blit_state *blit;
 
    struct cso_context *cso_context;
 };
diff --git a/src/mesa/state_tracker/st_extensions.c b/src/mesa/state_tracker/st_extensions.c
index 99d2a5fb9e9..0962b5f74c5 100644
--- a/src/mesa/state_tracker/st_extensions.c
+++ b/src/mesa/state_tracker/st_extensions.c
@@ -143,6 +143,7 @@ void st_init_extensions(struct st_context *st)
    ctx->Extensions.EXT_blend_logic_op = GL_TRUE;
    ctx->Extensions.EXT_blend_minmax = GL_TRUE;
    ctx->Extensions.EXT_blend_subtract = GL_TRUE;
+   ctx->Extensions.EXT_framebuffer_blit = GL_TRUE;
    ctx->Extensions.EXT_framebuffer_object = GL_TRUE;
    ctx->Extensions.EXT_fog_coord = GL_TRUE;
    ctx->Extensions.EXT_multi_draw_arrays = GL_TRUE;
diff --git a/src/mesa/state_tracker/st_gen_mipmap.c b/src/mesa/state_tracker/st_gen_mipmap.c
index 9c4e1032efe..6c3afca1ba1 100644
--- a/src/mesa/state_tracker/st_gen_mipmap.c
+++ b/src/mesa/state_tracker/st_gen_mipmap.c
@@ -37,6 +37,8 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
 #include "pipe/p_winsys.h"
+#include "util/u_gen_mipmap.h"
+
 #include "cso_cache/cso_cache.h"
 #include "cso_cache/cso_context.h"
 
@@ -49,55 +51,6 @@
 #include "st_cb_texture.h"
 
 
-
-static struct st_fragment_program *
-make_tex_fragment_program(GLcontext *ctx)
-{
-   struct st_fragment_program *stfp;
-   struct gl_program *p;
-   GLuint ic = 0;
-
-   p = ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0);
-   if (!p)
-      return NULL;
-
-   p->NumInstructions = 2;
-
-   p->Instructions = _mesa_alloc_instructions(p->NumInstructions);
-   if (!p->Instructions) {
-      ctx->Driver.DeleteProgram(ctx, p);
-      return NULL;
-   }
-   _mesa_init_instructions(p->Instructions, p->NumInstructions);
-
-   /* TEX result.color, fragment.texcoord[0], texture[0], 2D; */
-   p->Instructions[ic].Opcode = OPCODE_TEX;
-   p->Instructions[ic].DstReg.File = PROGRAM_OUTPUT;
-   p->Instructions[ic].DstReg.Index = FRAG_RESULT_COLR;
-   p->Instructions[ic].SrcReg[0].File = PROGRAM_INPUT;
-   p->Instructions[ic].SrcReg[0].Index = FRAG_ATTRIB_TEX0;
-   p->Instructions[ic].TexSrcUnit = 0;
-   p->Instructions[ic].TexSrcTarget = TEXTURE_2D_INDEX;
-   ic++;
-
-   /* END; */
-   p->Instructions[ic++].Opcode = OPCODE_END;
-
-   assert(ic == p->NumInstructions);
-
-   p->InputsRead = FRAG_BIT_TEX0;
-   p->OutputsWritten = (1 << FRAG_RESULT_COLR);
-
-   stfp = (struct st_fragment_program *) p;
-
-   st_translate_fragment_program(ctx->st, stfp, NULL);
-
-   return stfp;
-}
-
-
-
-
 /**
  * one-time init for generate mipmap
  * XXX Note: there may be other times we need no-op/simple state like this.
@@ -106,117 +59,18 @@ make_tex_fragment_program(GLcontext *ctx)
 void
 st_init_generate_mipmap(struct st_context *st)
 {
-   struct pipe_context *pipe = st->pipe;
-   struct pipe_blend_state blend;
-   struct pipe_rasterizer_state rasterizer;
-   struct pipe_depth_stencil_alpha_state depthstencil;
-
-   /* we don't use blending, but need to set valid values */
-   memset(&blend, 0, sizeof(blend));
-   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
-   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
-   blend.colormask = PIPE_MASK_RGBA;
-   st->gen_mipmap.blend = blend;
-   st->gen_mipmap.blend_cso = pipe->create_blend_state(pipe, &blend);
-
-   memset(&depthstencil, 0, sizeof(depthstencil));
-   st->gen_mipmap.depthstencil_cso = pipe->create_depth_stencil_alpha_state(pipe, &depthstencil);
-
-   /* Note: we're assuming zero is valid for all non-specified fields */
-   memset(&rasterizer, 0, sizeof(rasterizer));
-   rasterizer.front_winding = PIPE_WINDING_CW;
-   rasterizer.cull_mode = PIPE_WINDING_NONE;
-   st->gen_mipmap.rasterizer_cso = pipe->create_rasterizer_state(pipe, &rasterizer);
-
-   st->gen_mipmap.stfp = make_tex_fragment_program(st->ctx);
-   st->gen_mipmap.stvp = st_make_passthrough_vertex_shader(st, GL_FALSE);
+   st->gen_mipmap = util_create_gen_mipmap(st->pipe);
 }
 
 
 void
-st_destroy_generate_mipmpap(struct st_context *st)
-{
-   struct pipe_context *pipe = st->pipe;
-
-   pipe->delete_blend_state(pipe, st->gen_mipmap.blend_cso);
-   pipe->delete_depth_stencil_alpha_state(pipe, st->gen_mipmap.depthstencil_cso);
-   pipe->delete_rasterizer_state(pipe, st->gen_mipmap.rasterizer_cso);
-
-   /* XXX free stfp, stvp */
-}
-
-
-static void
-simple_viewport(struct pipe_context *pipe, uint width, uint height)
+st_destroy_generate_mipmap(struct st_context *st)
 {
-   struct pipe_viewport_state vp;
-
-   vp.scale[0] =  0.5 * width;
-   vp.scale[1] = -0.5 * height;
-   vp.scale[2] = 1.0;
-   vp.scale[3] = 1.0;
-   vp.translate[0] = 0.5 * width;
-   vp.translate[1] = 0.5 * height;
-   vp.translate[2] = 0.0;
-   vp.translate[3] = 0.0;
-
-   pipe->set_viewport_state(pipe, &vp);
+   util_destroy_gen_mipmap(st->gen_mipmap);
+   st->gen_mipmap = NULL;
 }
 
 
-
-/*
- * Draw simple [-1,1]x[-1,1] quad
- */
-static void
-draw_quad(GLcontext *ctx)
-{
-   GLfloat verts[4][2][4]; /* four verts, two attribs, XYZW */
-   GLuint i;
-   GLfloat sLeft = 0.0, sRight = 1.0;
-   GLfloat tTop = 1.0, tBot = 0.0;
-   GLfloat x0 = -1.0, x1 = 1.0;
-   GLfloat y0 = -1.0, y1 = 1.0;
-
-   /* upper-left */
-   verts[0][0][0] = x0;    /* attr[0].x */
-   verts[0][0][1] = y0;    /* attr[0].y */
-   verts[0][1][0] = sLeft; /* attr[1].s */
-   verts[0][1][1] = tTop;  /* attr[1].t */
-
-   /* upper-right */
-   verts[1][0][0] = x1;
-   verts[1][0][1] = y0;
-   verts[1][1][0] = sRight;
-   verts[1][1][1] = tTop;
-
-   /* lower-right */
-   verts[2][0][0] = x1;
-   verts[2][0][1] = y1;
-   verts[2][1][0] = sRight;
-   verts[2][1][1] = tBot;
-
-   /* lower-left */
-   verts[3][0][0] = x0;
-   verts[3][0][1] = y1;
-   verts[3][1][0] = sLeft;
-   verts[3][1][1] = tBot;
-
-   /* same for all verts: */
-   for (i = 0; i < 4; i++) {
-      verts[i][0][2] = 0.0; /*Z*/
-      verts[i][0][3] = 1.0; /*W*/
-      verts[i][1][2] = 0.0; /*R*/
-      verts[i][1][3] = 1.0; /*Q*/
-   }
-
-   st_draw_vertices(ctx, PIPE_PRIM_QUADS, 4, (float *) verts, 2, GL_TRUE);
-}
-
-
-
 /**
  * Generate mipmap levels using hardware rendering.
  * \return TRUE if successful, FALSE if not possible
@@ -229,12 +83,7 @@ st_render_mipmap(struct st_context *st,
 {
    struct pipe_context *pipe = st->pipe;
    struct pipe_screen *screen = pipe->screen;
-   struct pipe_framebuffer_state fb;
-   struct pipe_sampler_state sampler;
-   void *sampler_cso;
-   const uint face = _mesa_tex_target_to_face(target), zslice = 0;
-   /*const uint first_level_save = pt->first_level;*/
-   uint dstLevel;
+   const uint face = _mesa_tex_target_to_face(target);
 
    assert(target != GL_TEXTURE_3D); /* not done yet */
 
@@ -243,66 +92,7 @@ st_render_mipmap(struct st_context *st,
       return FALSE;
    }
 
-   /* init framebuffer state */
-   memset(&fb, 0, sizeof(fb));
-   fb.num_cbufs = 1;
-
-   /* sampler state */
-   memset(&sampler, 0, sizeof(sampler));
-   sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-   sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-   sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
-   sampler.min_img_filter = PIPE_TEX_FILTER_LINEAR;
-   sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
-   sampler.normalized_coords = 1;
-
-
-   /* bind state */
-   cso_set_blend(st->cso_context, &st->gen_mipmap.blend);
-   cso_set_depth_stencil_alpha(st->cso_context, &st->gen_mipmap.depthstencil);
-   cso_set_rasterizer(st->cso_context, &st->gen_mipmap.rasterizer);
-
-   /* bind shaders */
-   pipe->bind_fs_state(pipe, st->gen_mipmap.stfp->driver_shader);
-   pipe->bind_vs_state(pipe, st->gen_mipmap.stvp->driver_shader);
-
-   /*
-    * XXX for small mipmap levels, it may be faster to use the software
-    * fallback path...
-    */
-   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
-      const uint srcLevel = dstLevel - 1;
-
-      /*
-       * Setup framebuffer / dest surface
-       */
-      fb.cbufs[0] = screen->get_tex_surface(screen, pt, face, dstLevel, zslice);
-      pipe->set_framebuffer_state(pipe, &fb);
-
-      /*
-       * Setup sampler state
-       */
-      sampler.min_lod = sampler.max_lod = srcLevel;
-      sampler_cso = pipe->create_sampler_state(pipe, &sampler);
-      pipe->bind_sampler_states(pipe, 1, &sampler_cso);
-
-      simple_viewport(pipe, pt->width[dstLevel], pt->height[dstLevel]);
-
-      /*
-       * Setup src texture, override pt->first_level so we sample from
-       * the right mipmap level.
-       */
-      /*pt->first_level = srcLevel;*/
-      pipe->set_sampler_textures(pipe, 1, &pt);
-
-      draw_quad(st->ctx);
-
-      pipe->delete_sampler_state(pipe, sampler_cso);
-   }
-
-   /* restore first_level */
-   /*pt->first_level = first_level_save;*/
+   util_gen_mipmap(st->gen_mipmap, pt, face, baseLevel, lastLevel);
 
    /* restore pipe state */
 #if 0
diff --git a/src/mesa/state_tracker/st_gen_mipmap.h b/src/mesa/state_tracker/st_gen_mipmap.h
index 7668c1e44e3..00fbae93026 100644
--- a/src/mesa/state_tracker/st_gen_mipmap.h
+++ b/src/mesa/state_tracker/st_gen_mipmap.h
@@ -35,7 +35,7 @@ st_init_generate_mipmap(struct st_context *st);
 
 
 extern void
-st_destroy_generate_mipmpap(struct st_context *st);
+st_destroy_generate_mipmap(struct st_context *st);
 
 
 extern void
diff --git a/src/mesa/tnl/t_vertex_sse.c b/src/mesa/tnl/t_vertex_sse.c
index ad4cc62d5f7..a180441a5a6 100644
--- a/src/mesa/tnl/t_vertex_sse.c
+++ b/src/mesa/tnl/t_vertex_sse.c
@@ -348,8 +348,6 @@ static GLboolean build_vertex_emit( struct x86_program *p )
    struct x86_reg vp1 = x86_make_reg(file_XMM, 2);
    GLubyte *fixup, *label;
 
-   x86_init_func(&p->func);
-   
    /* Push a few regs?
     */
    x86_push(&p->func, countEBP);
@@ -641,7 +639,7 @@ void _tnl_generate_sse_emit( GLcontext *ctx )
 
    p.ctx = ctx;
    p.inputs_safe = 0;		/* for now */
-   p.outputs_safe = 1;		/* for now */
+   p.outputs_safe = 0;		/* for now */
    p.have_sse2 = cpu_has_xmm2;
    p.identity = x86_make_reg(file_XMM, 6);
    p.chan0 = x86_make_reg(file_XMM, 7);
diff --git a/src/mesa/vf/vf_sse.c b/src/mesa/vf/vf_sse.c
index 3ce76e2b042..c3a2166578c 100644
--- a/src/mesa/vf/vf_sse.c
+++ b/src/mesa/vf/vf_sse.c
@@ -633,7 +633,7 @@ void vf_generate_sse_emit( struct vertex_fetch *vf )
 
    p.vf = vf;
    p.inputs_safe = 0;		/* for now */
-   p.outputs_safe = 1;		/* for now */
+   p.outputs_safe = 0;		/* for now */
    p.have_sse2 = cpu_has_xmm2;
    p.identity = x86_make_reg(file_XMM, 6);
    p.chan0 = x86_make_reg(file_XMM, 7);