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

16 files changed, 600 insertions, 116 deletions

diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile
index a0db2e4555e..b28e516396e 100644
--- a/src/gallium/auxiliary/draw/Makefile
+++ b/src/gallium/auxiliary/draw/Makefile
@@ -18,6 +18,7 @@ C_SOURCES = \
 	draw_pt.c \
 	draw_pt_vcache.c \
 	draw_pt_fetch_emit.c \
+	draw_pt_fetch_pipeline.c \
 	draw_pt_elts.c \
 	draw_prim.c \
 	draw_pstipple.c \
diff --git a/src/gallium/auxiliary/draw/SConscript b/src/gallium/auxiliary/draw/SConscript
index 981225a8c2c..9ca41974412 100644
--- a/src/gallium/auxiliary/draw/SConscript
+++ b/src/gallium/auxiliary/draw/SConscript
@@ -17,6 +17,7 @@ draw = env.ConvenienceLibrary(
 		'draw_pt.c',
 		'draw_pt_vcache.c',
 		'draw_pt_fetch_emit.c',
+		'draw_pt_fetch_pipeline.c',
 		'draw_pt_elts.c',
 		'draw_prim.c',
 		'draw_pstipple.c',
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 9a9b25297f1..0c9f9c2e036 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -160,6 +160,11 @@ struct draw_vertex_shader {
 /* Internal function for vertex fetch.
  */
 typedef void (*fetch_func)(const void *ptr, float *attrib);
+
+fetch_func draw_get_fetch_func( enum pipe_format format );
+
+
+
 typedef void (*full_fetch_func)( struct draw_context *draw,
 				 struct tgsi_exec_machine *machine,
 				 const unsigned *elts,
@@ -211,6 +216,7 @@ struct draw_context
 
       struct {
          struct draw_pt_middle_end *fetch_emit;
+         struct draw_pt_middle_end *fetch_pipeline;
          struct draw_pt_middle_end *fetch_shade_emit;
          struct draw_pt_middle_end *fetch_shade_cliptest_pipeline_or_emit;
       } middle;
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 2ea96c686da..f59fb86f781 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -73,7 +73,6 @@ draw_pt_arrays(struct draw_context *draw,
     */
 
 
-#if 0
    if (!cliptest && !pipeline && !shading) {
       /* This is the 'passthrough' path:
        */
@@ -81,6 +80,14 @@ draw_pt_arrays(struct draw_context *draw,
        */
       middle = draw->pt.middle.fetch_emit;
    }
+   else if (!cliptest && !shading) {
+      /* This is the 'passthrough' path targetting the pipeline backend.
+       */
+      /* Fetch user verts, emit pipeline verts, run pipeline:
+       */
+      middle = draw->pt.middle.fetch_pipeline;
+   }
+#if 0
    else if (!cliptest && !pipeline) {
       /* Fetch user verts, run vertex shader, emit hw verts:
        */
@@ -117,10 +124,9 @@ draw_pt_arrays(struct draw_context *draw,
       middle = draw->pt.middle.fetch_shade_cliptest_pipeline;
    }
 #else
-   if (cliptest || pipeline || shading)
+   else {
       return FALSE;
-
-   middle = draw->pt.middle.fetch_emit;
+   }
 #endif
 
 
@@ -171,10 +177,9 @@ draw_pt_arrays(struct draw_context *draw,
     */
    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 
-   frontend->prepare( frontend, middle );
+   frontend->prepare( frontend, prim, middle );
 
    frontend->run( frontend,
-                  prim,
                   draw_pt_elt_func( draw ),
                   draw_pt_elt_ptr( draw, start ),
                   count );
@@ -191,6 +196,10 @@ boolean draw_pt_init( struct draw_context *draw )
    if (!draw->pt.middle.fetch_emit)
       return FALSE;
 
+   draw->pt.middle.fetch_pipeline = draw_pt_fetch_pipeline( draw );
+   if (!draw->pt.middle.fetch_pipeline)
+      return FALSE;
+
    draw->pt.front.vcache = draw_pt_vcache( draw );
    if (!draw->pt.front.vcache)
       return FALSE;
@@ -206,6 +215,11 @@ void draw_pt_destroy( struct draw_context *draw )
       draw->pt.middle.fetch_emit = NULL;
    }
 
+   if (draw->pt.middle.fetch_pipeline) {
+      draw->pt.middle.fetch_pipeline->destroy( draw->pt.middle.fetch_pipeline );
+      draw->pt.middle.fetch_pipeline = NULL;
+   }
+
    if (draw->pt.front.vcache) {
       draw->pt.front.vcache->destroy( draw->pt.front.vcache );
       draw->pt.front.vcache = NULL;
diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h
index f8786160799..800072c5115 100644
--- a/src/gallium/auxiliary/draw/draw_pt.h
+++ b/src/gallium/auxiliary/draw/draw_pt.h
@@ -53,10 +53,10 @@ struct draw_context;
  */
 struct draw_pt_front_end {
    void (*prepare)( struct draw_pt_front_end *,
+                    unsigned prim,
                     struct draw_pt_middle_end * );
 
    void (*run)( struct draw_pt_front_end *,
-                unsigned prim,
                 pt_elt_func elt_func,
                 const void *elt_ptr,
                 unsigned count );
@@ -79,10 +79,10 @@ struct draw_pt_front_end {
  * Currenly only using the passthrough version.
  */
 struct draw_pt_middle_end {
-   void (*prepare)( struct draw_pt_middle_end * );
+   void (*prepare)( struct draw_pt_middle_end *,
+                    unsigned prim );
 
    void (*run)( struct draw_pt_middle_end *,
-                unsigned prim,
                 const unsigned *fetch_elts,
                 unsigned fetch_count,
                 const ushort *draw_elts,
@@ -112,6 +112,7 @@ const void *draw_pt_elt_ptr( struct draw_context *draw,
  */
 struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw );
 struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw );
+struct draw_pt_middle_end *draw_pt_fetch_pipeline( struct draw_context *draw );
 
 
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index 9b098bc173f..39f0b40838f 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -85,45 +85,6 @@ struct fetch_emit_middle_end {
 };
 
 
-static void fetch_B8G8R8A8_UNORM( const void *from,
-                                  float *attrib )
-{
-   ubyte *ub = (ubyte *) from;
-   attrib[2] = UBYTE_TO_FLOAT(ub[0]);
-   attrib[1] = UBYTE_TO_FLOAT(ub[1]);
-   attrib[0] = UBYTE_TO_FLOAT(ub[2]);
-   attrib[3] = UBYTE_TO_FLOAT(ub[3]);
-}
-
-static void fetch_R32G32B32A32_FLOAT( const void *from,
-                                  float *attrib )
-{
-   float *f = (float *) from;
-   attrib[0] = f[0];
-   attrib[1] = f[1];
-   attrib[2] = f[2];
-   attrib[3] = f[3];
-}
-
-static void fetch_R32G32B32_FLOAT( const void *from,
-                                   float *attrib )
-{
-   float *f = (float *) from;
-   attrib[0] = f[0];
-   attrib[1] = f[1];
-   attrib[2] = f[2];
-   attrib[3] = 1.0;
-}
-
-static void fetch_R32G32_FLOAT( const void *from,
-                                float *attrib )
-{
-   float *f = (float *) from;
-   attrib[0] = f[0];
-   attrib[1] = f[1];
-   attrib[2] = 0.0;
-   attrib[3] = 1.0;
-}
 
 static void fetch_R32_FLOAT( const void *from,
                              float *attrib )
@@ -203,16 +164,29 @@ fetch_store_general( struct fetch_emit_middle_end *feme,
 
 
 
-static void fetch_emit_prepare( struct draw_pt_middle_end *middle )
+static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
+                                unsigned prim )
 {
    static const float zero = 0;
    struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
    struct draw_context *draw = feme->draw;
-   const struct vertex_info *vinfo = draw->render->get_vertex_info(draw->render);
-   unsigned nr_attrs = vinfo->num_attribs;
+   const struct vertex_info *vinfo;
    unsigned i;
+   boolean ok;
+
+
+   ok = draw->render->set_primitive( draw->render, 
+                                     prim );
+   if (!ok) {
+      assert(0);
+      return;
+   }
+   
+   /* Must do this after set_primitive() above:
+    */
+   vinfo = draw->render->get_vertex_info(draw->render);
 
-   for (i = 0; i < nr_attrs; i++) {
+   for (i = 0; i < vinfo->num_attribs; i++) {
       unsigned src_element = vinfo->src_index[i];
       unsigned src_buffer = draw->vertex_element[src_element].vertex_buffer_index;
          
@@ -222,28 +196,9 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle )
 
       feme->fetch[i].pitch = draw->vertex_buffer[src_buffer].pitch;
          
-      switch (draw->vertex_element[src_element].src_format) {
-      case PIPE_FORMAT_B8G8R8A8_UNORM:
-         feme->fetch[i].fetch = fetch_B8G8R8A8_UNORM;
-         break;
-      case PIPE_FORMAT_R32G32B32A32_FLOAT:
-         feme->fetch[i].fetch = fetch_R32G32B32A32_FLOAT;
-         break;
-      case PIPE_FORMAT_R32G32B32_FLOAT:
-         feme->fetch[i].fetch = fetch_R32G32B32_FLOAT;
-         break;
-      case PIPE_FORMAT_R32G32_FLOAT:
-         feme->fetch[i].fetch = fetch_R32G32_FLOAT;
-         break;
-      case PIPE_FORMAT_R32_FLOAT:
-         feme->fetch[i].fetch = fetch_R32_FLOAT;
-         break;
-      default:
-         assert(0);
-         feme->fetch[i].fetch = NULL;
-         break;
-      }
-         
+      feme->fetch[i].fetch = draw_get_fetch_func(draw->vertex_element[src_element].src_format);
+
+
       switch (vinfo->emit[i]) {
       case EMIT_4F:
          feme->fetch[i].emit = emit_R32G32B32A32_FLOAT;
@@ -268,6 +223,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle )
          feme->fetch[i].pitch = 0;
          feme->fetch[i].fetch = fetch_R32_FLOAT;
          feme->fetch[i].emit = emit_R32_FLOAT;
+         break;
       default:
          assert(0);
          feme->fetch[i].emit = NULL;
@@ -275,7 +231,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle )
       }
    }
 
-   feme->nr_fetch = nr_attrs;
+   feme->nr_fetch = vinfo->num_attribs;
    feme->hw_vertex_size = vinfo->size * 4;
 }
 
@@ -284,7 +240,6 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle )
 
 
 static void fetch_emit_run( struct draw_pt_middle_end *middle,
-                            unsigned prim,
                             const unsigned *fetch_elts,
                             unsigned fetch_count,
                             const ushort *draw_elts,
@@ -293,16 +248,7 @@ static void fetch_emit_run( struct draw_pt_middle_end *middle,
    struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
    struct draw_context *draw = feme->draw;
    void *hw_verts;
-   boolean ok;
    
-   ok = draw->render->set_primitive( draw->render, 
-                                     prim );
-   if (!ok) {
-      assert(0);
-      return;
-   }
-   
-
    hw_verts = draw->render->allocate_vertices( draw->render,
                                                (ushort)feme->hw_vertex_size,
                                                (ushort)fetch_count );
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c
new file mode 100644
index 00000000000..94e7d01be4b
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c
@@ -0,0 +1,391 @@
+/**************************************************************************
+ * 
+ * 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:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "pipe/p_util.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_pt.h"
+
+/* The simplest 'middle end' in the new vertex code.  
+ * 
+ * The responsibilities of a middle end are to:
+ *  - perform vertex fetch using
+ *       - draw vertex element/buffer state
+ *       - a list of fetch indices we received as an input
+ *  - run the vertex shader
+ *  - cliptest, 
+ *  - clip coord calculation 
+ *  - viewport transformation
+ *  - if necessary, run the primitive pipeline, passing it:
+ *       - a linear array of vertex_header vertices constructed here
+ *       - a set of draw indices we received as an input
+ *  - otherwise, drive the hw backend,
+ *       - allocate space for hardware format vertices
+ *       - translate the vertex-shader output vertices to hw format
+ *       - calling the backend draw functions.
+ *
+ * For convenience, we provide a helper function to drive the hardware
+ * backend given similar inputs to those required to run the pipeline.
+ *
+ * In the case of passthrough mode, many of these actions are disabled
+ * or noops, so we end up doing:
+ *
+ *  - perform vertex fetch
+ *  - drive the hw backend
+ *
+ * IE, basically just vertex fetch to post-vs-format vertices,
+ * followed by a call to the backend helper function.
+ */
+
+
+struct fetch_pipeline_middle_end {
+   struct draw_pt_middle_end base;
+   struct draw_context *draw;
+
+   struct {
+      const ubyte *ptr;
+      unsigned pitch;
+      void (*fetch)( const void *from, float *attrib);
+      void (*emit)( const float *attrib, float **out );
+   } fetch[PIPE_MAX_ATTRIBS];
+   
+   unsigned nr_fetch;
+   unsigned pipeline_vertex_size;
+   unsigned prim;
+};
+
+
+static void fetch_NULL( const void *from,
+                        float *attrib )
+{
+}
+
+
+
+static void emit_R32_FLOAT( const float *attrib,
+                            float **out )
+{
+   (*out)[0] = attrib[0];
+   (*out) += 1;
+}
+
+static void emit_R32G32_FLOAT( const float *attrib,
+                               float **out )
+{
+   (*out)[0] = attrib[0];
+   (*out)[1] = attrib[1];
+   (*out) += 2;
+}
+
+static void emit_R32G32B32_FLOAT( const float *attrib,
+                                  float **out )
+{
+   (*out)[0] = attrib[0];
+   (*out)[1] = attrib[1];
+   (*out)[2] = attrib[2];
+   (*out) += 3;
+}
+
+static void emit_R32G32B32A32_FLOAT( const float *attrib,
+                                     float **out )
+{
+   (*out)[0] = attrib[0];
+   (*out)[1] = attrib[1];
+   (*out)[2] = attrib[2];
+   (*out)[3] = attrib[3];
+   (*out) += 4;
+}
+
+static void emit_header( const float *attrib,
+                         float **out )
+{
+   (*out)[0] = 0;
+   (*out)[1] = 0;
+   (*out)[2] = 0;
+   (*out)[3] = 0;
+   (*out)[3] = 1;
+   (*out) += 5;
+}
+
+/**
+ * General-purpose fetch from user's vertex arrays, emit to driver's
+ * vertex buffer.
+ *
+ * XXX this is totally temporary.
+ */
+static void
+fetch_store_general( struct fetch_pipeline_middle_end *fpme,
+                     void *out_ptr,
+                     const unsigned *fetch_elts,
+                     unsigned count )
+{
+   float *out = (float *)out_ptr;
+   uint i, j;
+
+   for (i = 0; i < count; i++) {
+      unsigned elt = fetch_elts[i];
+      
+      for (j = 0; j < fpme->nr_fetch; j++) {
+         float attrib[4];
+         const ubyte *from = (fpme->fetch[j].ptr +
+                              fpme->fetch[j].pitch * elt);
+         
+         fpme->fetch[j].fetch( from, attrib );
+         fpme->fetch[j].emit( attrib, &out );
+      }
+   }
+}
+
+
+/* We aren't running a vertex shader, but are running the pipeline.
+ * That means the vertices we need to build look like:
+ *
+ * dw0: vertex header (zero?)
+ * dw1: clip coord 0
+ * dw2: clip coord 1
+ * dw3: clip coord 2
+ * dw4: clip coord 4
+ * dw5: screen coord 0
+ * dw6: screen coord 0
+ * dw7: screen coord 0
+ * dw8: screen coord 0
+ * dw9: other attribs...
+ *
+ */
+static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
+                                    unsigned prim )
+{
+   static const float zero = 0;
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+   struct draw_context *draw = fpme->draw;
+   unsigned i, nr = 0;
+
+   fpme->prim = prim;
+
+   /* Emit the vertex header and empty clipspace coord field:
+    */
+   {
+      fpme->fetch[nr].ptr = NULL;
+      fpme->fetch[nr].pitch = 0;
+      fpme->fetch[nr].fetch = fetch_NULL;
+      fpme->fetch[nr].emit = emit_header;
+      nr++;
+   }
+   
+
+   /* Need to look at vertex shader inputs (we know it is a
+    * passthrough shader, so these define the outputs too).  If we
+    * were running a shader, we'd still be looking at the inputs at
+    * this point.
+    */
+   for (i = 0; i < draw->vertex_shader->info.num_inputs; i++) {
+      unsigned buf = draw->vertex_element[i].vertex_buffer_index;
+      enum pipe_format format  = draw->vertex_element[i].src_format;
+
+      fpme->fetch[nr].ptr = ((const ubyte *) draw->user.vbuffer[buf] + 
+                            draw->vertex_buffer[buf].buffer_offset + 
+                            draw->vertex_element[i].src_offset);
+
+      fpme->fetch[nr].pitch = draw->vertex_buffer[buf].pitch;
+      fpme->fetch[nr].fetch = draw_get_fetch_func( format );
+
+      /* Always do this -- somewhat redundant...
+       */
+      fpme->fetch[nr].emit = emit_R32G32B32A32_FLOAT;
+      nr++;
+   }
+
+   fpme->nr_fetch = nr;
+   fpme->pipeline_vertex_size = (5 + (nr-1) * 4) * sizeof(float);
+}
+
+
+
+/**
+ * Add a point to the primitive queue.
+ * \param i0  index into user's vertex arrays
+ */
+static void do_point( struct draw_context *draw,
+		      const char *v0 )
+{
+   struct prim_header prim;
+   
+   prim.reset_line_stipple = 0;
+   prim.edgeflags = 1;
+   prim.pad = 0;
+   prim.v[0] = (struct vertex_header *)v0;
+
+   draw->pipeline.first->point( draw->pipeline.first, &prim );
+}
+
+
+/**
+ * Add a line to the primitive queue.
+ * \param i0  index into user's vertex arrays
+ * \param i1  index into user's vertex arrays
+ */
+static void do_line( struct draw_context *draw,
+		     const char *v0,
+		     const char *v1 )
+{
+   struct prim_header prim;
+   
+   prim.reset_line_stipple = 1; /* fixme */
+   prim.edgeflags = 1;
+   prim.pad = 0;
+   prim.v[0] = (struct vertex_header *)v0;
+   prim.v[1] = (struct vertex_header *)v1;
+
+   draw->pipeline.first->line( draw->pipeline.first, &prim );
+}
+
+/**
+ * Add a triangle to the primitive queue.
+ */
+static void do_triangle( struct draw_context *draw,
+			 char *v0,
+			 char *v1,
+			 char *v2 )
+{
+   struct prim_header prim;
+   
+//   _mesa_printf("tri %d %d %d\n", i0, i1, i2);
+   prim.reset_line_stipple = 1;
+   prim.edgeflags = ~0;
+   prim.pad = 0;
+   prim.v[0] = (struct vertex_header *)v0;
+   prim.v[1] = (struct vertex_header *)v1;
+   prim.v[2] = (struct vertex_header *)v2;
+
+   draw->pipeline.first->tri( draw->pipeline.first, &prim );
+}
+
+
+static void run_pipeline( struct fetch_pipeline_middle_end *fpme,
+                          char *verts,
+                          const ushort *elts,
+                          unsigned count )
+{
+   struct draw_context *draw = fpme->draw;
+   unsigned stride = fpme->pipeline_vertex_size;
+   unsigned i;
+   
+   switch (fpme->prim) {
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < count; i++) 
+         do_point( draw, 
+                   verts + stride * elts[i] );
+      break;
+   case PIPE_PRIM_LINES:
+      for (i = 0; i+1 < count; i += 2) 
+         do_line( draw, 
+                  verts + stride * elts[i+0],
+                  verts + stride * elts[i+1]);
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      for (i = 0; i+2 < count; i += 3)
+         do_triangle( draw, 
+                      verts + stride * elts[i+0],
+                      verts + stride * elts[i+1],
+                      verts + stride * elts[i+2]);
+      break;
+   }
+}
+
+
+
+
+static void fetch_pipeline_run( struct draw_pt_middle_end *middle,
+                            const unsigned *fetch_elts,
+                            unsigned fetch_count,
+                            const ushort *draw_elts,
+                            unsigned draw_count )
+{
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+   struct draw_context *draw = fpme->draw;
+   char *pipeline_verts;
+   
+   pipeline_verts = MALLOC( fpme->pipeline_vertex_size * 
+                            fetch_count );
+   if (!pipeline_verts) {
+      assert(0);
+      return;
+   }
+         
+					
+   /* Single routine to fetch vertices and emit pipeline verts.
+    */
+   fetch_store_general( fpme, 
+                        pipeline_verts,
+                        fetch_elts,
+                        fetch_count );
+
+   
+   run_pipeline( fpme,
+                 pipeline_verts,
+                 draw_elts,
+                 draw_count );
+                 
+
+   /* Done -- that was easy, wasn't it: 
+    */
+   FREE( pipeline_verts );
+}
+
+
+
+static void fetch_pipeline_finish( struct draw_pt_middle_end *middle )
+{
+   /* nothing to do */
+}
+
+static void fetch_pipeline_destroy( struct draw_pt_middle_end *middle )
+{
+   FREE(middle);
+}
+
+
+struct draw_pt_middle_end *draw_pt_fetch_pipeline( struct draw_context *draw )
+{
+   struct fetch_pipeline_middle_end *fetch_pipeline = CALLOC_STRUCT( fetch_pipeline_middle_end );
+ 
+   fetch_pipeline->base.prepare = fetch_pipeline_prepare;
+   fetch_pipeline->base.run     = fetch_pipeline_run;
+   fetch_pipeline->base.finish  = fetch_pipeline_finish;
+   fetch_pipeline->base.destroy = fetch_pipeline_destroy;
+
+   fetch_pipeline->draw = draw;
+     
+   return &fetch_pipeline->base;
+}
+
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c
index 16ffedf580c..5a068761df9 100644
--- a/src/gallium/auxiliary/draw/draw_pt_vcache.c
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c
@@ -59,6 +59,8 @@ struct vcache_frontend {
    const void *elt_ptr;
 
    struct draw_pt_middle_end *middle;
+
+   unsigned input_prim;
    unsigned output_prim;
 };
 
@@ -77,7 +79,6 @@ static void vcache_flush( struct vcache_frontend *vcache )
 
    if (vcache->draw_count) {
       vcache->middle->run( vcache->middle,
-                           vcache->output_prim,
                            vcache->fetch_elts,
                            vcache->fetch_count,
                            vcache->draw_elts,
@@ -173,7 +174,6 @@ static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = {
 
 
 static void vcache_run_pv2( struct draw_pt_front_end *frontend, 
-                            unsigned prim,
                             pt_elt_func get_elt,
                             const void *elts,
                             unsigned count )
@@ -185,9 +185,8 @@ static void vcache_run_pv2( struct draw_pt_front_end *frontend,
     */
    vcache->elt_func = get_elt;
    vcache->elt_ptr = elts;
-   vcache->output_prim = reduced_prim[prim];
 
-   switch (prim) {
+   switch (vcache->input_prim) {
    case PIPE_PRIM_POINTS:
       for (i = 0; i < count; i ++) {
          vcache_point( vcache,
@@ -304,7 +303,6 @@ static void vcache_run_pv2( struct draw_pt_front_end *frontend,
 
 
 static void vcache_run_pv0( struct draw_pt_front_end *frontend, 
-                            unsigned prim,
                             pt_elt_func get_elt,
                             const void *elts,
                             unsigned count )
@@ -316,9 +314,8 @@ static void vcache_run_pv0( struct draw_pt_front_end *frontend,
     */
    vcache->elt_func = get_elt;
    vcache->elt_ptr = elts;
-   vcache->output_prim = reduced_prim[prim];
 
-   switch (prim) {
+   switch (vcache->input_prim) {
    case PIPE_PRIM_POINTS:
       for (i = 0; i < count; i ++) {
          vcache_point( vcache,
@@ -389,6 +386,7 @@ static void vcache_run_pv0( struct draw_pt_front_end *frontend,
 }
 
 static void vcache_prepare( struct draw_pt_front_end *frontend,
+                            unsigned prim,
                             struct draw_pt_middle_end *middle )
 {
    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
@@ -398,8 +396,11 @@ static void vcache_prepare( struct draw_pt_front_end *frontend,
    else
       vcache->base.run = vcache_run_pv2;
    
+   vcache->input_prim = prim;
+   vcache->output_prim = reduced_prim[prim];
+
    vcache->middle = middle;
-   middle->prepare( middle );
+   middle->prepare( middle, vcache->output_prim );
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vertex_fetch.c b/src/gallium/auxiliary/draw/draw_vertex_fetch.c
index 11f99babf65..9041041006e 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_fetch.c
@@ -166,7 +166,7 @@ fetch_B8G8R8A8_UNORM(const void *ptr, float *attrib)
 }
 
 
-static fetch_func get_fetch_func( enum pipe_format format )
+fetch_func draw_get_fetch_func( enum pipe_format format )
 {
 #if 0
    {
@@ -502,7 +502,7 @@ void draw_update_vertex_fetch( struct draw_context *draw )
 						       draw->vertex_element[i].src_offset;
 
       draw->vertex_fetch.pitch[i] = draw->vertex_buffer[buf].pitch;
-      draw->vertex_fetch.fetch[i] = get_fetch_func( format );
+      draw->vertex_fetch.fetch[i] = draw_get_fetch_func( format );
    }
 
    draw->vertex_fetch.nr_attrs = nr_attrs;
diff --git a/src/gallium/auxiliary/draw/draw_wide_point.c b/src/gallium/auxiliary/draw/draw_wide_point.c
index c53f7e6cb37..6fc7c9fcd7a 100644
--- a/src/gallium/auxiliary/draw/draw_wide_point.c
+++ b/src/gallium/auxiliary/draw/draw_wide_point.c
@@ -38,6 +38,11 @@ struct widepoint_stage {
    struct draw_stage stage;
 
    float half_point_size;
+   float point_size_min;
+   float point_size_max;
+
+   float xbias;
+   float ybias;
 
    uint texcoord_slot[PIPE_MAX_SHADER_OUTPUTS];
    uint texcoord_mode[PIPE_MAX_SHADER_OUTPUTS];
@@ -124,20 +129,26 @@ static void widepoint_point( struct draw_stage *stage,
    float *pos2 = v2->data[0];
    float *pos3 = v3->data[0];
 
-   const float xbias = 0.0, ybias = -0.125;
-
    /* point size is either per-vertex or fixed size */
    if (wide->psize_slot >= 0) {
-      half_size = 0.5f * header->v[0]->data[wide->psize_slot][0];
+      half_size = header->v[0]->data[wide->psize_slot][0];
+
+      /* XXX: temporary -- do this in the vertex shader??
+       */
+      half_size = CLAMP(half_size,
+                        wide->point_size_min,
+                        wide->point_size_max);
+      
+      half_size *= 0.5f; 
    }
    else {
       half_size = wide->half_point_size;
    }
 
-   left_adj = -half_size + xbias;
-   right_adj = half_size + xbias;
-   bot_adj = half_size + ybias;
-   top_adj = -half_size + ybias;
+   left_adj = -half_size + wide->xbias;
+   right_adj = half_size + wide->xbias;
+   bot_adj = half_size + wide->ybias;
+   top_adj = -half_size + wide->ybias;
 
    pos0[0] += left_adj;
    pos0[1] += top_adj;
@@ -182,6 +193,14 @@ static void widepoint_first_point( struct draw_stage *stage,
    struct draw_context *draw = stage->draw;
 
    wide->half_point_size = 0.5f * draw->rasterizer->point_size;
+   wide->point_size_min = draw->rasterizer->point_size_min;
+   wide->point_size_max = draw->rasterizer->point_size_max;
+   wide->xbias = 0.0;
+   wide->ybias = 0.0;
+
+   if (draw->rasterizer->gl_rasterization_rules) {
+      wide->ybias = -0.125;
+   }
 
    /* XXX we won't know the real size if it's computed by the vertex shader! */
    if ((draw->rasterizer->point_size > draw->wide_point_threshold) ||
diff --git a/src/gallium/auxiliary/util/p_tile.c b/src/gallium/auxiliary/util/p_tile.c
index 287d7839812..520da5cecd5 100644
--- a/src/gallium/auxiliary/util/p_tile.c
+++ b/src/gallium/auxiliary/util/p_tile.c
@@ -560,6 +560,70 @@ z24s8_get_tile_rgba(unsigned *src,
 }
 
 
+/*** PIPE_FORMAT_YCBCR / PIPE_FORMAT_YCBCR_REV ***/
+
+/**
+ * Convert YCbCr (or YCrCb) to RGBA.
+ */
+static void
+ycbcr_get_tile_rgba(ushort *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride,
+                    boolean rev)
+{
+   const float scale = 1.0f / 255.0f;
+   unsigned i, j;
+
+   /* we're assuming we're being asked for an even number of texels */
+   assert((w & 1) == 0);
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      /* do two texels at a time */
+      for (j = 0; j < w; j += 2, src += 2) {
+         const ushort t0 = src[0];
+         const ushort t1 = src[1];
+         const ubyte y0 = (t0 >> 8) & 0xff;  /* luminance */
+         const ubyte y1 = (t1 >> 8) & 0xff;  /* luminance */
+         ubyte cb, cr;
+         float r, g, b;
+
+         if (rev) {
+            cb = t1 & 0xff;         /* chroma U */
+            cr = t0 & 0xff;         /* chroma V */
+         }
+         else {
+            cb = t0 & 0xff;         /* chroma U */
+            cr = t1 & 0xff;         /* chroma V */
+         }
+
+         /* even pixel: y0,cr,cb */
+         r = 1.164f * (y0-16) + 1.596f * (cr-128);
+         g = 1.164f * (y0-16) - 0.813f * (cr-128) - 0.391f * (cb-128);
+         b = 1.164f * (y0-16) + 2.018f * (cb-128);
+         pRow[0] = r * scale;
+         pRow[1] = g * scale;
+         pRow[2] = b * scale;
+         pRow[3] = 1.0f;
+         pRow += 4;
+
+         /* odd pixel: use y1,cr,cb */
+         r = 1.164f * (y1-16) + 1.596f * (cr-128);
+         g = 1.164f * (y1-16) - 0.813f * (cr-128) - 0.391f * (cb-128);
+         b = 1.164f * (y1-16) + 2.018f * (cb-128);
+         pRow[0] = r * scale;
+         pRow[1] = g * scale;
+         pRow[2] = b * scale;
+         pRow[3] = 1.0f;
+         pRow += 4;
+
+      }
+      p += dst_stride;
+   }
+}
+
+
 void
 pipe_get_tile_rgba(struct pipe_context *pipe,
                    struct pipe_surface *ps,
@@ -622,6 +686,14 @@ pipe_get_tile_rgba(struct pipe_context *pipe,
    case PIPE_FORMAT_Z24S8_UNORM:
       z24s8_get_tile_rgba((unsigned *) packed, w, h, p, dst_stride);
       break;
+   case PIPE_FORMAT_YCBCR:
+      assert((x & 1) == 0);
+      ycbcr_get_tile_rgba((ushort *) packed, w, h, p, dst_stride, FALSE);
+      break;
+   case PIPE_FORMAT_YCBCR_REV:
+      assert((x & 1) == 0);
+      ycbcr_get_tile_rgba((ushort *) packed, w, h, p, dst_stride, TRUE);
+      break;
    default:
       assert(0);
    }
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 1ab1c40379a..e04ffeb9b16 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -343,6 +343,8 @@ cmd_state_texture(const struct cell_command_texture *texture)
    spu.texture[unit].width = width;
    spu.texture[unit].height = height;
 
+   spu.texture[unit].tiles_per_row = width / TILE_SIZE;
+
    spu.texture[unit].tex_size = (vector float) { width, height, 0.0, 0.0};
    spu.texture[unit].tex_size_mask = (vector unsigned int)
          { width - 1, height - 1, 0, 0 };
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index e9e39cbeab3..e962e1426c6 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -105,7 +105,8 @@ struct spu_framebuffer {
 struct spu_texture
 {
    void *start;
-   uint width, height;
+   ushort width, height;
+   ushort tiles_per_row;
    vector float tex_size;
    vector unsigned int tex_size_mask; /**< == int(size - 1) */
    vector unsigned int tex_size_x_mask; /**< == int(size - 1) */
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index ceab2469803..5051774f00c 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -48,25 +48,45 @@ invalidate_tex_cache(void)
 }
 
 
+/**
+ * XXX look into getting texels for all four pixels in a quad at once.
+ */
 static uint
 get_texel(uint unit, vec_uint4 coordinate)
 {
+   /*
+    * XXX we could do the "/ TILE_SIZE" and "% TILE_SIZE" operations as
+    * SIMD since X and Y are already in a SIMD register.
+    */
+   const unsigned texture_ea = (uintptr_t) spu.texture[unit].start;
+   ushort x = spu_extract(coordinate, 0);
+   ushort y = spu_extract(coordinate, 1);
+   unsigned tile_offset = sizeof(tile_t)
+      * ((y / TILE_SIZE * spu.texture[unit].tiles_per_row) + (x / TILE_SIZE));
+   ushort texel_offset = (ushort) 4
+      * (ushort) (((ushort) (y % TILE_SIZE) * (ushort) TILE_SIZE) + (x % TILE_SIZE));
    vec_uint4 tmp;
-   unsigned x = spu_extract(coordinate, 0);
-   unsigned y = spu_extract(coordinate, 1);
-   const unsigned tiles_per_row = spu.texture[unit].width / TILE_SIZE;
-   unsigned tile_offset = sizeof(tile_t) * ((y / TILE_SIZE * tiles_per_row) 
-                                            + (x / TILE_SIZE));
-   unsigned texel_offset = 4 * (((y % TILE_SIZE) * TILE_SIZE)
-                                + (x % TILE_SIZE));
 
    spu_dcache_fetch_unaligned((qword *) & tmp,
-                              spu.texture[unit].start + tile_offset + texel_offset,
+                              texture_ea + tile_offset + texel_offset,
                               4);
    return spu_extract(tmp, 0);
 }
 
 
+/**
+ * Get four texels from locations (x[0], y[0]), (x[1], y[1]) ...
+ *
+ * NOTE: in the typical case of bilinear filtering, the four texels
+ * are in a 2x2 group so we could get by with just two dcache fetches
+ * (two side-by-side texels per fetch).  But when bilinear filtering
+ * wraps around a texture edge, we'll probably need code like we have
+ * now.
+ * FURTHERMORE: since we're rasterizing a quad of 2x2 pixels at a time,
+ * it's quite likely that the four pixels in a quad will need some of the
+ * same texels.  So look into doing texture fetches for four pixels at
+ * a time.
+ */
 static void
 get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 {
@@ -76,7 +96,7 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
    const qword offset_x = si_andi((qword) x, 0x1f);
    const qword offset_y = si_andi((qword) y, 0x1f);
 
-   const qword tiles_per_row = (qword) spu_splats(spu.texture[unit].width / TILE_SIZE);
+   const qword tiles_per_row = (qword) spu_splats(spu.texture[unit].tiles_per_row);
    const qword tile_size = (qword) spu_splats(sizeof(tile_t));
 
    qword tile_offset = si_mpya((qword) tile_y, tiles_per_row, (qword) tile_x);
@@ -97,9 +117,9 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
                               texture_ea + spu_extract(offset, 3), 4);
 }
 
+
 /**
  * Get texture sample at texcoord.
- * XXX this is extremely primitive for now.
  */
 vector float
 sample_texture_nearest(uint unit, vector float texcoord)
@@ -126,17 +146,25 @@ sample_texture_bilinear(uint unit, vector float texcoord)
 
    vec_uint4 texels[4];
    
+   /* setup texcoords for quad:
+    *  +-----+-----+
+    *  |x0,y0|x1,y1|
+    *  +-----+-----+
+    *  |x2,y2|x3,y3|
+    *  +-----+-----+
+    */
    vec_uint4 x = spu_splats(spu_extract(itc, 0));
    vec_uint4 y = spu_splats(spu_extract(itc, 1));
-
    x = spu_add(x, offset_x);
    y = spu_add(y, offset_y);
 
+   /* GL_REPEAT wrap mode: */
    x = spu_and(x, spu.texture[unit].tex_size_x_mask);
    y = spu_and(y, spu.texture[unit].tex_size_y_mask);
 
    get_four_texels(unit, x, y, texels);
 
+   /* integer A8R8G8B8 to float texel conversion */
    vector float texel00 = spu_unpack_A8R8G8B8(spu_extract(texels[0], 0));
    vector float texel01 = spu_unpack_A8R8G8B8(spu_extract(texels[1], 0));
    vector float texel10 = spu_unpack_A8R8G8B8(spu_extract(texels[2], 0));
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index a73028814e3..3593446e1c4 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -115,9 +115,12 @@ struct pipe_rasterizer_state
                             still needed though, to indicate inputs/outputs */
    unsigned origin_lower_left:1;  /**< Is (0,0) the lower-left corner? */
    unsigned flatshade_first:1;   /**< take color attribute from the first vertex of a primitive */
+   unsigned gl_rasterization_rules:1; /**< enable tweaks for GL rasterization?  */
 
    float line_width;
    float point_size;           /**< used when no per-vertex size */
+   float point_size_min;        /* XXX - temporary, will go away */
+   float point_size_max;        /* XXX - temporary, will go away */
    float offset_units;
    float offset_scale;
    ubyte sprite_coord_mode[PIPE_MAX_SHADER_OUTPUTS]; /**< PIPE_SPRITE_COORD_ */
diff --git a/src/gallium/winsys/xlib/SConscript b/src/gallium/winsys/xlib/SConscript
index c38b5be52c8..218b89285f7 100644
--- a/src/gallium/winsys/xlib/SConscript
+++ b/src/gallium/winsys/xlib/SConscript
@@ -22,8 +22,6 @@ sources = [
 	
 drivers = [
 	softpipe,
-	i915simple,
-	i965simple,
 ]
 
 # TODO: write a wrapper function http://www.scons.org/wiki/WrapperFunctions