Move guts of vertex array drawing into the 'draw' module.

12 files changed, 478 insertions, 372 deletions

diff --git a/src/mesa/pipe/draw/draw_arrays.c b/src/mesa/pipe/draw/draw_arrays.c
new file mode 100644
index 0000000000..59098fe242
--- /dev/null
+++ b/src/mesa/pipe/draw/draw_arrays.c
@@ -0,0 +1,386 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Brian Paul
+ *    Keith Whitwell
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_util.h"
+
+#include "pipe/draw/draw_private.h"
+#include "pipe/draw/draw_context.h"
+#include "pipe/draw/draw_prim.h"
+
+#include "pipe/tgsi/core/tgsi_exec.h"
+#include "pipe/tgsi/core/tgsi_build.h"
+#include "pipe/tgsi/core/tgsi_util.h"
+
+
+#if defined __GNUC__
+#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME[SIZE] __attribute__(( aligned( 16 ) ))
+#define ALIGN16_ASSIGN(P) P
+#else
+#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME[SIZE + 1]
+#define ALIGN16_ASSIGN(P) align16(P)
+#endif
+
+
+
+static INLINE unsigned
+compute_clipmask(float cx, float cy, float cz, float cw)
+{
+   unsigned mask;
+#if defined(macintosh) || defined(__powerpc__)
+   /* on powerpc cliptest is 17% faster in this way. */
+   mask =  (((cw <  cx) << CLIP_RIGHT_SHIFT));
+   mask |= (((cw < -cx) << CLIP_LEFT_SHIFT));
+   mask |= (((cw <  cy) << CLIP_TOP_SHIFT));
+   mask |= (((cw < -cy) << CLIP_BOTTOM_SHIFT));
+   mask |= (((cw <  cz) << CLIP_FAR_SHIFT));
+   mask |= (((cw < -cz) << CLIP_NEAR_SHIFT));
+#else /* !defined(macintosh)) */
+   mask = 0x0;
+   if (-cx + cw < 0) mask |= CLIP_RIGHT_BIT;
+   if ( cx + cw < 0) mask |= CLIP_LEFT_BIT;
+   if (-cy + cw < 0) mask |= CLIP_TOP_BIT;
+   if ( cy + cw < 0) mask |= CLIP_BOTTOM_BIT;
+   if (-cz + cw < 0) mask |= CLIP_FAR_BIT;
+   if ( cz + cw < 0) mask |= CLIP_NEAR_BIT;
+#endif /* defined(macintosh) */
+   return mask;
+}
+
+
+/**
+ * Fetch a float[4] vertex attribute from memory, doing format/type
+ * conversion as needed.
+ * XXX this might be a temporary thing.
+ */
+static void
+fetch_attrib4(const void *ptr, unsigned format, float attrib[4])
+{
+   /* defaults */
+   attrib[1] = 0.0;
+   attrib[2] = 0.0;
+   attrib[3] = 1.0;
+   switch (format) {
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      attrib[3] = ((float *) ptr)[3];
+      /* fall-through */
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      attrib[2] = ((float *) ptr)[2];
+      /* fall-through */
+   case PIPE_FORMAT_R32G32_FLOAT:
+      attrib[1] = ((float *) ptr)[1];
+      /* fall-through */
+   case PIPE_FORMAT_R32_FLOAT:
+      attrib[0] = ((float *) ptr)[0];
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+/**
+ * Transform vertices with the current vertex program/shader
+ * Up to four vertices can be shaded at a time.
+ * \param vbuffer  the input vertex data
+ * \param elts  indexes of four input vertices
+ * \param count  number of vertices to shade [1..4]
+ * \param vOut  array of pointers to four output vertices
+ */
+static void
+run_vertex_program(struct draw_context *draw,
+                   unsigned elts[4], unsigned count,
+                   struct vertex_header *vOut[])
+{
+   struct tgsi_exec_machine machine;
+   unsigned int j;
+
+   ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
+   ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
+   const float *scale = draw->viewport.scale;
+   const float *trans = draw->viewport.translate;
+
+   assert(count <= 4);
+
+#ifdef DEBUG
+   memset( &machine, 0, sizeof( machine ) );
+#endif
+
+   /* init machine state */
+   tgsi_exec_machine_init(&machine,
+                          draw->vertex_shader.tokens,
+                          PIPE_MAX_SAMPLERS,
+                          NULL /*samplers*/ );
+
+   /* Consts does not require 16 byte alignment. */
+   machine.Consts = draw->vertex_shader.constants->constant;
+
+   machine.Inputs = ALIGN16_ASSIGN(inputs);
+   machine.Outputs = ALIGN16_ASSIGN(outputs);
+
+
+   if (0)
+   {
+      unsigned attr;
+      for (attr = 0; attr < 16; attr++) {
+         if (draw->vertex_shader.inputs_read & (1 << attr)) {
+            printf("attr %d: buf_off %d  src_off %d  pitch %d\n",
+                   attr, 
+                   draw->vertex_buffer[attr].buffer_offset,
+                   draw->vertex_element[attr].src_offset,
+                   draw->vertex_buffer[attr].pitch);
+         }
+      }
+   }
+
+   /* load machine inputs */
+   for (j = 0; j < count; j++) {
+      unsigned attr;
+      for (attr = 0; attr < 16; attr++) {
+         if (draw->vertex_shader.inputs_read & (1 << attr)) {
+            const void *src
+               = (const void *) ((const ubyte *) draw->mapped_vbuffer[attr]
+                                 + draw->vertex_buffer[attr].buffer_offset
+                                 + draw->vertex_element[attr].src_offset
+                                 + elts[j] * draw->vertex_buffer[attr].pitch);
+            float p[4];
+
+            fetch_attrib4(src, draw->vertex_element[attr].src_format, p);
+
+            machine.Inputs[attr].xyzw[0].f[j] = p[0]; /*X*/
+            machine.Inputs[attr].xyzw[1].f[j] = p[1]; /*Y*/
+            machine.Inputs[attr].xyzw[2].f[j] = p[2]; /*Z*/
+            machine.Inputs[attr].xyzw[3].f[j] = p[3]; /*W*/
+#if 0
+            if (attr == 0) {
+               printf("Input vertex %d: %f %f %f\n",
+                      j, p[0], p[1], p[2]);
+            }
+#endif
+         }
+      }
+   }
+
+#if 0
+   printf("Consts:\n");
+   for (i = 0; i < 4; i++) {
+      printf(" %d: %f %f %f %f\n", i,
+             machine.Consts[i][0],
+             machine.Consts[i][1],
+             machine.Consts[i][2],
+             machine.Consts[i][3]);
+   }
+#endif
+
+   /* run shader */
+   tgsi_exec_machine_run( &machine );
+
+#if 0
+   printf("VS result: %f %f %f %f\n",
+          outputs[0].xyzw[0].f[0],
+          outputs[0].xyzw[1].f[0],
+          outputs[0].xyzw[2].f[0],
+          outputs[0].xyzw[3].f[0]);
+#endif
+
+   /* store machine results */
+   assert(draw->vertex_shader.outputs_written & (1 << VERT_RESULT_HPOS));
+   for (j = 0; j < count; j++) {
+      unsigned attr, slot;
+      float x, y, z, w;
+
+      /* Handle attr[0] (position) specially: */
+      x = vOut[j]->clip[0] = outputs[0].xyzw[0].f[j];
+      y = vOut[j]->clip[1] = outputs[0].xyzw[1].f[j];
+      z = vOut[j]->clip[2] = outputs[0].xyzw[2].f[j];
+      w = vOut[j]->clip[3] = outputs[0].xyzw[3].f[j];
+
+      vOut[j]->clipmask = compute_clipmask(x, y, z, w);
+      vOut[j]->edgeflag = 1;
+
+      /* divide by w */
+      w = 1.0 / w;
+      x *= w;
+      y *= w;
+      z *= w;
+
+      /* Viewport mapping */
+      vOut[j]->data[0][0] = x * scale[0] + trans[0];
+      vOut[j]->data[0][1] = y * scale[1] + trans[1];
+      vOut[j]->data[0][2] = z * scale[2] + trans[2];
+      vOut[j]->data[0][3] = w;
+#if 0
+      printf("wincoord: %f %f %f\n",
+             vOut[j]->data[0][0],
+             vOut[j]->data[0][1],
+             vOut[j]->data[0][2]);
+#endif
+
+      /* remaining attributes: */
+      /* pack into sequential post-transform attrib slots */
+      slot = 1;
+      for (attr = 1; attr < VERT_RESULT_MAX; attr++) {
+         if (draw->vertex_shader.outputs_written & (1 << attr)) {
+            assert(slot < draw->nr_attrs);
+            vOut[j]->data[slot][0] = outputs[attr].xyzw[0].f[j];
+            vOut[j]->data[slot][1] = outputs[attr].xyzw[1].f[j];
+            vOut[j]->data[slot][2] = outputs[attr].xyzw[2].f[j];
+            vOut[j]->data[slot][3] = outputs[attr].xyzw[3].f[j];
+            slot++;
+         }
+      }
+   }
+
+#if 0
+   memcpy(
+      quad->outputs.color,
+      &machine.Outputs[1].xyzw[0].f[0],
+      sizeof( quad->outputs.color ) );
+#endif
+}
+
+
+/**
+ * Called by the draw module when the vertx cache needs to be flushed.
+ * This involves running the vertex shader.
+ */
+static void vs_flush( struct draw_context *draw )
+{
+   unsigned i, j;
+
+   /* run vertex shader on vertex cache entries, four per invokation */
+   for (i = 0; i < draw->vs.queue_nr; i += 4) {
+      struct vertex_header *dests[4];
+      unsigned elts[4];
+      int n;
+
+      for (j = 0; j < 4; j++) {
+         elts[j] = draw->vs.queue[i + j].elt;
+         dests[j] = draw->vs.queue[i + j].dest;
+      }
+
+      n = MIN2(4, draw->vs.queue_nr - i);
+      assert(n > 0);
+      assert(n <= 4);
+
+      run_vertex_program(draw, elts, n, dests);
+   }
+
+   draw->vs.queue_nr = 0;
+}
+
+
+void draw_set_mapped_vertex_buffer(struct draw_context *draw,
+                                   unsigned attr, const void *buffer)
+{
+   draw->mapped_vbuffer[attr] = buffer;
+}
+
+
+/**
+ * Draw vertex arrays
+ * This is the main entrypoint into the drawing module.
+ * \param prim  one of PIPE_PRIM_x
+ * \param start  index of first vertex to draw
+ * \param count  number of vertices to draw
+ */
+void
+draw_arrays(struct draw_context *draw, unsigned prim,
+            unsigned start, unsigned count)
+{
+   /* tell drawing pipeline we're beginning drawing */
+   draw->pipeline.first->begin( draw->pipeline.first );
+
+   draw->vs_flush = vs_flush;
+
+   draw_invalidate_vcache( draw );
+
+   draw_set_prim( draw, prim );
+
+   /* drawing done here: */
+   draw_prim(draw, start, count);
+
+   /* draw any left-over buffered prims */
+   draw_flush(draw);
+
+   /* tell drawing pipeline we're done drawing */
+   draw->pipeline.first->end( draw->pipeline.first );
+}
+
+
+/* XXX move this into draw_context.c? */
+
+#define EMIT_ATTR( VF_ATTR, STYLE, SIZE )			\
+do {								\
+   if (draw->nr_attrs >= 2)					\
+      draw->vf_attr_to_slot[VF_ATTR] = draw->nr_attrs - 2;	\
+   draw->attrs[draw->nr_attrs].attrib = VF_ATTR;		\
+   draw->attrs[draw->nr_attrs].format = STYLE;			\
+   draw->nr_attrs++;						\
+   draw->vertex_size += SIZE;					\
+} while (0)
+
+
+/**
+ * XXX very similar to same func in draw_vb.c (which will go away)
+ */
+void
+draw_set_vertex_attributes( struct draw_context *draw,
+                            const unsigned *slot_to_vf_attr,
+                            unsigned nr_attrs )
+{
+   unsigned i;
+
+   memset(draw->vf_attr_to_slot, 0, sizeof(draw->vf_attr_to_slot));
+   draw->nr_attrs = 0;
+   draw->vertex_size = 0;
+
+   /*
+    * First three attribs are always the same: header, clip pos, winpos
+    */
+   EMIT_ATTR(VF_ATTRIB_VERTEX_HEADER, EMIT_1F, 1);
+   EMIT_ATTR(VF_ATTRIB_CLIP_POS, EMIT_4F, 4);
+
+   assert(slot_to_vf_attr[0] == VF_ATTRIB_POS);
+   EMIT_ATTR(slot_to_vf_attr[0], EMIT_4F_VIEWPORT, 4);
+
+   /*
+    * Remaining attribs (color, texcoords, etc)
+    */
+   for (i = 1; i < nr_attrs; i++) 
+      EMIT_ATTR(slot_to_vf_attr[i], EMIT_4F, 4);
+
+   draw->vertex_size *= 4; /* floats to bytes */
+}
diff --git a/src/mesa/pipe/draw/draw_context.c b/src/mesa/pipe/draw/draw_context.c
index a808fb7777..cc00576c53 100644
--- a/src/mesa/pipe/draw/draw_context.c
+++ b/src/mesa/pipe/draw/draw_context.c
@@ -201,3 +201,31 @@ void draw_set_viewport_state( struct draw_context *draw,
     * Full pipe will have vertex shader, vertex fetch of its own.
     */
 }
+
+
+void
+draw_set_vertex_buffer(struct draw_context *draw,
+                       unsigned attr,
+                       const struct pipe_vertex_buffer *buffer)
+{
+   assert(attr < PIPE_ATTRIB_MAX);
+   draw->vertex_buffer[attr] = *buffer;
+}
+
+
+void
+draw_set_vertex_element(struct draw_context *draw,
+                        unsigned attr,
+                        const struct pipe_vertex_element *element)
+{
+   assert(attr < PIPE_ATTRIB_MAX);
+   draw->vertex_element[attr] = *element;
+}
+
+
+void
+draw_set_vertex_shader(struct draw_context *draw,
+                       const struct pipe_shader_state *shader)
+{
+   draw->vertex_shader = *shader;
+}
diff --git a/src/mesa/pipe/draw/draw_context.h b/src/mesa/pipe/draw/draw_context.h
index 74fdd46262..2fce1322c5 100644
--- a/src/mesa/pipe/draw/draw_context.h
+++ b/src/mesa/pipe/draw/draw_context.h
@@ -93,14 +93,36 @@ void draw_set_vertex_attributes( struct draw_context *draw,
 				 const unsigned *attrs,
 				 unsigned nr_attrs );
 
-/* XXX temporary */
-void draw_set_vertex_attributes2( struct draw_context *draw,
-				 const unsigned *attrs,
-				 unsigned nr_attrs );
+unsigned draw_prim_info( unsigned prim, unsigned *first, unsigned *incr );
+
+unsigned draw_trim( unsigned count, unsigned first, unsigned incr );
+
+void draw_set_mapped_element_buffer( struct draw_context *draw,
+                                     unsigned eltSize, void *elements );
+
+void draw_set_mapped_vertex_buffer(struct draw_context *draw,
+                                   unsigned attr, const void *buffer);
+
+
+void
+draw_set_vertex_buffer(struct draw_context *draw,
+                       unsigned attr,
+                       const struct pipe_vertex_buffer *buffer);
+
+void
+draw_set_vertex_element(struct draw_context *draw,
+                        unsigned attr,
+                        const struct pipe_vertex_element *element);
+
+void
+draw_set_vertex_shader(struct draw_context *draw,
+                       const struct pipe_shader_state *shader);
+
+
+void
+draw_arrays(struct draw_context *draw, unsigned prim,
+            unsigned start, unsigned count);
 
-/* XXX temporary */
-void draw_vb(struct draw_context *draw,
-	     struct vertex_buffer *VB );
 
 void draw_vertices(struct draw_context *draw,
                    unsigned mode,
diff --git a/src/mesa/pipe/draw/draw_prim.c b/src/mesa/pipe/draw/draw_prim.c
index fb8bc0f36d..fbd0672875 100644
--- a/src/mesa/pipe/draw/draw_prim.c
+++ b/src/mesa/pipe/draw/draw_prim.c
@@ -434,8 +434,8 @@ draw_set_prim( struct draw_context *draw, unsigned prim )
  * \param elements  the element buffer ptr
  */
 void
-draw_set_element_buffer( struct draw_context *draw,
-                         unsigned eltSize, void *elements )
+draw_set_mapped_element_buffer( struct draw_context *draw,
+                                unsigned eltSize, void *elements )
 {
    /* choose the get_vertex() function to use */
    switch (eltSize) {
@@ -456,10 +456,9 @@ draw_set_element_buffer( struct draw_context *draw,
    }
    draw->elts = elements;
    draw->eltSize = eltSize;
-
-
 }
 
+
 unsigned
 draw_prim_info(unsigned prim, unsigned *first, unsigned *incr)
 {
diff --git a/src/mesa/pipe/draw/draw_prim.h b/src/mesa/pipe/draw/draw_prim.h
index 3224989bef..4c55b02978 100644
--- a/src/mesa/pipe/draw/draw_prim.h
+++ b/src/mesa/pipe/draw/draw_prim.h
@@ -8,16 +8,9 @@ void draw_invalidate_vcache( struct draw_context *draw );
 
 void draw_set_prim( struct draw_context *draw, unsigned prim );
 
-void draw_set_element_buffer( struct draw_context *draw,
-                              unsigned eltSize, void *elements );
-
 void draw_prim( struct draw_context *draw, unsigned start, unsigned count );
 
 void draw_flush( struct draw_context *draw );
 
-unsigned draw_prim_info( unsigned prim, unsigned *first, unsigned *incr );
-
-unsigned draw_trim( unsigned count, unsigned first, unsigned incr );
-
 
 #endif /* DRAW_PRIM_H */
diff --git a/src/mesa/pipe/draw/draw_private.h b/src/mesa/pipe/draw/draw_private.h
index 798fa5c3a6..5c3efb80e9 100644
--- a/src/mesa/pipe/draw/draw_private.h
+++ b/src/mesa/pipe/draw/draw_private.h
@@ -158,9 +158,11 @@ struct draw_context
    /* pipe state that we need: */
    struct pipe_setup_state setup;
    struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
+   struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
+   struct pipe_shader_state vertex_shader;
 
-   /** need to know the pipe for vertex flushing/transformation: */
-   struct pipe_context *pipe;
+   const void *mapped_vbuffer[PIPE_ATTRIB_MAX];
 
    /* Clip derived state:
     */
diff --git a/src/mesa/pipe/draw/draw_vb.c b/src/mesa/pipe/draw/draw_vb.c
index 18e48c0a68..0eefb0b250 100644
--- a/src/mesa/pipe/draw/draw_vb.c
+++ b/src/mesa/pipe/draw/draw_vb.c
@@ -221,9 +221,9 @@ void draw_vb(struct draw_context *draw,
    vf_emit_vertices( draw->vf, VB->Count, draw->verts );
 
    if (VB->Elts) 
-      draw_set_element_buffer(draw, sizeof(unsigned), VB->Elts);
+      draw_set_mapped_element_buffer(draw, sizeof(unsigned), VB->Elts);
    else
-      draw_set_element_buffer(draw, 0, NULL);
+      draw_set_mapped_element_buffer(draw, 0, NULL);
 
    for (i = 0; i < VB->PrimitiveCount; i++) {
       const GLenum mode = VB->Primitive[i].mode;
@@ -281,7 +281,7 @@ draw_vertices(struct draw_context *draw,
 
 
    /* no element/index buffer */
-   draw_set_element_buffer(draw, 0, NULL);
+   draw_set_mapped_element_buffer(draw, 0, NULL);
 
    /*draw_prim_info(mode, &first, &incr);*/
    draw_allocate_vertices( draw, numVerts );
@@ -325,7 +325,7 @@ draw_vertices(struct draw_context *draw,
    draw->in_vb = 0;
 }
 
-
+#if 000
 
 /**
  * Accumulate another attribute's info.
@@ -383,3 +383,4 @@ void draw_set_vertex_attributes( struct draw_context *draw,
                                                  draw->nr_attrs, 0 );
 #endif
 }
+#endif
diff --git a/src/mesa/pipe/softpipe/sp_draw_arrays.c b/src/mesa/pipe/softpipe/sp_draw_arrays.c
index 6213286f47..2df07eb7d6 100644
--- a/src/mesa/pipe/softpipe/sp_draw_arrays.c
+++ b/src/mesa/pipe/softpipe/sp_draw_arrays.c
@@ -31,284 +31,15 @@
  */
 
 
-/** TEMP */
-#include "main/context.h"
-#include "main/macros.h"
-
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
 #include "pipe/p_winsys.h"
-
+#include "pipe/p_util.h"
 
 #include "sp_context.h"
 #include "sp_state.h"
 
-#include "pipe/draw/draw_private.h"
 #include "pipe/draw/draw_context.h"
-#include "pipe/draw/draw_prim.h"
-
-#include "pipe/tgsi/core/tgsi_exec.h"
-#include "pipe/tgsi/core/tgsi_build.h"
-#include "pipe/tgsi/core/tgsi_util.h"
-
-
-#if defined __GNUC__
-#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME[SIZE] __attribute__(( aligned( 16 ) ))
-#define ALIGN16_ASSIGN(P) P
-#else
-#define ALIGN16_DECL(TYPE, NAME, SIZE)  TYPE NAME[SIZE + 1]
-#define ALIGN16_ASSIGN(P) align16(P)
-#endif
-
-
-
-static INLINE unsigned
-compute_clipmask(float cx, float cy, float cz, float cw)
-{
-   unsigned mask;
-#if defined(macintosh) || defined(__powerpc__)
-   /* on powerpc cliptest is 17% faster in this way. */
-   mask =  (((cw <  cx) << CLIP_RIGHT_SHIFT));
-   mask |= (((cw < -cx) << CLIP_LEFT_SHIFT));
-   mask |= (((cw <  cy) << CLIP_TOP_SHIFT));
-   mask |= (((cw < -cy) << CLIP_BOTTOM_SHIFT));
-   mask |= (((cw <  cz) << CLIP_FAR_SHIFT));
-   mask |= (((cw < -cz) << CLIP_NEAR_SHIFT));
-#else /* !defined(macintosh)) */
-   mask = 0x0;
-   if (-cx + cw < 0) mask |= CLIP_RIGHT_BIT;
-   if ( cx + cw < 0) mask |= CLIP_LEFT_BIT;
-   if (-cy + cw < 0) mask |= CLIP_TOP_BIT;
-   if ( cy + cw < 0) mask |= CLIP_BOTTOM_BIT;
-   if (-cz + cw < 0) mask |= CLIP_FAR_BIT;
-   if ( cz + cw < 0) mask |= CLIP_NEAR_BIT;
-#endif /* defined(macintosh) */
-   return mask;
-}
-
-
-/**
- * Fetch a float[4] vertex attribute from memory, doing format/type
- * conversion as needed.
- * XXX this might be a temporary thing.
- */
-static void
-fetch_attrib4(const void *ptr, unsigned format, float attrib[4])
-{
-   /* defaults */
-   attrib[1] = 0.0;
-   attrib[2] = 0.0;
-   attrib[3] = 1.0;
-   switch (format) {
-   case PIPE_FORMAT_R32G32B32A32_FLOAT:
-      attrib[3] = ((float *) ptr)[3];
-      /* fall-through */
-   case PIPE_FORMAT_R32G32B32_FLOAT:
-      attrib[2] = ((float *) ptr)[2];
-      /* fall-through */
-   case PIPE_FORMAT_R32G32_FLOAT:
-      attrib[1] = ((float *) ptr)[1];
-      /* fall-through */
-   case PIPE_FORMAT_R32_FLOAT:
-      attrib[0] = ((float *) ptr)[0];
-      break;
-   default:
-      assert(0);
-   }
-}
-
-
-/**
- * Transform vertices with the current vertex program/shader
- * Up to four vertices can be shaded at a time.
- * \param vbuffer  the input vertex data
- * \param elts  indexes of four input vertices
- * \param count  number of vertices to shade [1..4]
- * \param vOut  array of pointers to four output vertices
- */
-static void
-run_vertex_program(struct softpipe_context *sp,
-                   unsigned elts[4], unsigned count,
-                   struct vertex_header *vOut[])
-{
-   struct tgsi_exec_machine machine;
-   unsigned int j;
-
-   ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
-   ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
-   const float *scale = sp->viewport.scale;
-   const float *trans = sp->viewport.translate;
-
-   assert(count <= 4);
-
-#ifdef DEBUG
-   memset( &machine, 0, sizeof( machine ) );
-#endif
-
-   /* init machine state */
-   tgsi_exec_machine_init(
-                          &machine,
-                          sp->vs.tokens,
-                          PIPE_MAX_SAMPLERS,
-                          NULL /*samplers*/ );
-
-   /* Consts does not require 16 byte alignment. */
-   machine.Consts = sp->vs.constants->constant;
-
-   machine.Inputs = ALIGN16_ASSIGN(inputs);
-   machine.Outputs = ALIGN16_ASSIGN(outputs);
-
-
-   if (0)
-   {
-      unsigned attr;
-      for (attr = 0; attr < 16; attr++) {
-         if (sp->vs.inputs_read & (1 << attr)) {
-            printf("attr %d: buf_off %d  src_off %d  pitch %d\n",
-                   attr, 
-                   sp->vertex_buffer[attr].buffer_offset,
-                   sp->vertex_element[attr].src_offset,
-                   sp->vertex_buffer[attr].pitch);
-         }
-      }
-   }
-
-   /* load machine inputs */
-   for (j = 0; j < count; j++) {
-      unsigned attr;
-      for (attr = 0; attr < 16; attr++) {
-         if (sp->vs.inputs_read & (1 << attr)) {
-            const void *src
-               = (const void *) ((const ubyte *) sp->mapped_vbuffer[attr]
-                                 + sp->vertex_buffer[attr].buffer_offset
-                                 + sp->vertex_element[attr].src_offset
-                                 + elts[j] * sp->vertex_buffer[attr].pitch);
-            float p[4];
-
-            fetch_attrib4(src, sp->vertex_element[attr].src_format, p);
-
-            machine.Inputs[attr].xyzw[0].f[j] = p[0]; /*X*/
-            machine.Inputs[attr].xyzw[1].f[j] = p[1]; /*Y*/
-            machine.Inputs[attr].xyzw[2].f[j] = p[2]; /*Z*/
-            machine.Inputs[attr].xyzw[3].f[j] = p[3]; /*W*/
-#if 0
-            if (attr == 0) {
-               printf("Input vertex %d: %f %f %f\n",
-                      j, p[0], p[1], p[2]);
-            }
-#endif
-         }
-      }
-   }
-
-#if 0
-   printf("Consts:\n");
-   for (i = 0; i < 4; i++) {
-      printf(" %d: %f %f %f %f\n", i,
-             machine.Consts[i][0],
-             machine.Consts[i][1],
-             machine.Consts[i][2],
-             machine.Consts[i][3]);
-   }
-#endif
-
-   /* run shader */
-   tgsi_exec_machine_run( &machine );
-
-#if 0
-   printf("VS result: %f %f %f %f\n",
-          outputs[0].xyzw[0].f[0],
-          outputs[0].xyzw[1].f[0],
-          outputs[0].xyzw[2].f[0],
-          outputs[0].xyzw[3].f[0]);
-#endif
-
-   /* store machine results */
-   assert(sp->vs.outputs_written & (1 << VERT_RESULT_HPOS));
-   for (j = 0; j < count; j++) {
-      unsigned attr, slot;
-      float x, y, z, w;
-
-      /* Handle attr[0] (position) specially: */
-      x = vOut[j]->clip[0] = outputs[0].xyzw[0].f[j];
-      y = vOut[j]->clip[1] = outputs[0].xyzw[1].f[j];
-      z = vOut[j]->clip[2] = outputs[0].xyzw[2].f[j];
-      w = vOut[j]->clip[3] = outputs[0].xyzw[3].f[j];
-
-      vOut[j]->clipmask = compute_clipmask(x, y, z, w);
-      vOut[j]->edgeflag = 1;
-
-      /* divide by w */
-      w = 1.0 / w;
-      x *= w;
-      y *= w;
-      z *= w;
-
-      /* Viewport mapping */
-      vOut[j]->data[0][0] = x * scale[0] + trans[0];
-      vOut[j]->data[0][1] = y * scale[1] + trans[1];
-      vOut[j]->data[0][2] = z * scale[2] + trans[2];
-      vOut[j]->data[0][3] = w;
-#if 0
-      printf("wincoord: %f %f %f\n",
-             vOut[j]->data[0][0],
-             vOut[j]->data[0][1],
-             vOut[j]->data[0][2]);
-#endif
-
-      /* remaining attributes: */
-      /* pack into sequential post-transform attrib slots */
-      slot = 1;
-      for (attr = 1; attr < VERT_RESULT_MAX; attr++) {
-         if (sp->vs.outputs_written & (1 << attr)) {
-            assert(slot < sp->nr_attrs);
-            vOut[j]->data[slot][0] = outputs[attr].xyzw[0].f[j];
-            vOut[j]->data[slot][1] = outputs[attr].xyzw[1].f[j];
-            vOut[j]->data[slot][2] = outputs[attr].xyzw[2].f[j];
-            vOut[j]->data[slot][3] = outputs[attr].xyzw[3].f[j];
-            slot++;
-         }
-      }
-   }
-
-#if 0
-   memcpy(
-      quad->outputs.color,
-      &machine.Outputs[1].xyzw[0].f[0],
-      sizeof( quad->outputs.color ) );
-#endif
-}
-
-
-/**
- * Called by the draw module when the vertx cache needs to be flushed.
- * This involves running the vertex shader.
- */
-static void vs_flush( struct draw_context *draw )
-{
-   struct softpipe_context *sp = (struct softpipe_context *) draw->pipe;
-   unsigned i, j;
-
-   /* run vertex shader on vertex cache entries, four per invokation */
-   for (i = 0; i < draw->vs.queue_nr; i += 4) {
-      struct vertex_header *dests[4];
-      unsigned elts[4];
-      int n;
-
-      for (j = 0; j < 4; j++) {
-         elts[j] = draw->vs.queue[i + j].elt;
-         dests[j] = draw->vs.queue[i + j].dest;
-      }
-
-      n = MIN2(4, draw->vs.queue_nr - i);
-      assert(n > 0);
-      assert(n <= 4);
-
-      run_vertex_program(sp, elts, n, dests);
-   }
-
-   draw->vs.queue_nr = 0;
-}
 
 
 
@@ -322,6 +53,10 @@ softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode,
 
 
 /**
+ * Draw vertex arrays, with optional indexing.
+ * Basically, map the vertex buffers (and drawing surfaces), then hand off
+ * the drawing to the 'draw' module.
+ *
  * XXX should the element buffer be specified/bound with a separate function?
  */
 void
@@ -330,7 +65,6 @@ softpipe_draw_elements(struct pipe_context *pipe,
                        unsigned indexSize,
                        unsigned mode, unsigned start, unsigned count)
 {
-
    struct softpipe_context *sp = softpipe_context(pipe);
    struct draw_context *draw = sp->draw;
    unsigned length, first, incr, i;
@@ -353,10 +87,11 @@ softpipe_draw_elements(struct pipe_context *pipe,
     */
    for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
       if (sp->vertex_buffer[i].buffer) {
-         sp->mapped_vbuffer[i]
+         void *buf
             = pipe->winsys->buffer_map(pipe->winsys,
                                        sp->vertex_buffer[i].buffer,
                                        PIPE_BUFFER_FLAG_READ);
+         draw_set_mapped_vertex_buffer(draw, i, buf);
       }
    }
    /* Map index buffer, if present */
@@ -364,30 +99,15 @@ softpipe_draw_elements(struct pipe_context *pipe,
       mapped_indexes = pipe->winsys->buffer_map(pipe->winsys,
                                                 indexBuffer,
                                                 PIPE_BUFFER_FLAG_READ);
-      draw_set_element_buffer(draw, indexSize, mapped_indexes);
+      draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
    }
    else {
-      draw_set_element_buffer(draw, 0, NULL);  /* no index/element buffer */
+      draw_set_mapped_element_buffer(draw, 0, NULL);  /* no index/element buffer */
    }
 
-   /* tell drawing pipeline we're beginning drawing */
-   draw->pipeline.first->begin( draw->pipeline.first );
-
-   draw->vs_flush = vs_flush;
-   draw->pipe = pipe;  /* XXX pass pipe to draw_create() */
-
-   draw_invalidate_vcache( draw );
 
-   draw_set_prim( draw, mode );
+   draw_arrays(draw, mode, start, count);
 
-   /* drawing done here: */
-   draw_prim(draw, start, count);
-
-   /* draw any left-over buffered prims */
-   draw_flush(draw);
-
-   /* tell drawing pipeline we're done drawing */
-   draw->pipeline.first->end( draw->pipeline.first );
 
    /*
     * unmap vertex/index buffers
@@ -395,58 +115,13 @@ softpipe_draw_elements(struct pipe_context *pipe,
    for (i = 0; i < PIPE_ATTRIB_MAX; i++) {
       if (sp->vertex_buffer[i].buffer) {
          pipe->winsys->buffer_unmap(pipe->winsys, sp->vertex_buffer[i].buffer);
+         draw_set_mapped_vertex_buffer(draw, i, NULL);
       }
    }
    if (indexBuffer) {
       pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer);
+      draw_set_mapped_element_buffer(draw, 0, NULL);
    }
 
    softpipe_unmap_surfaces(sp);
 }
-
-
-
-#define EMIT_ATTR( VF_ATTR, STYLE, SIZE )			\
-do {								\
-   if (draw->nr_attrs >= 2)					\
-      draw->vf_attr_to_slot[VF_ATTR] = draw->nr_attrs - 2;	\
-   draw->attrs[draw->nr_attrs].attrib = VF_ATTR;		\
-   draw->attrs[draw->nr_attrs].format = STYLE;			\
-   draw->nr_attrs++;						\
-   draw->vertex_size += SIZE;					\
-} while (0)
-
-
-/**
- * XXX very similar to same func in draw_vb.c (which will go away)
- */
-void
-draw_set_vertex_attributes2( struct draw_context *draw,
-                             const unsigned *slot_to_vf_attr,
-                             unsigned nr_attrs )
-{
-   unsigned i;
-
-   memset(draw->vf_attr_to_slot, 0, sizeof(draw->vf_attr_to_slot));
-   draw->nr_attrs = 0;
-   draw->vertex_size = 0;
-
-   /*
-    * First three attribs are always the same: header, clip pos, winpos
-    */
-   EMIT_ATTR(VF_ATTRIB_VERTEX_HEADER, EMIT_1F, 1);
-   EMIT_ATTR(VF_ATTRIB_CLIP_POS, EMIT_4F, 4);
-
-   assert(slot_to_vf_attr[0] == VF_ATTRIB_POS);
-   EMIT_ATTR(slot_to_vf_attr[0], EMIT_4F_VIEWPORT, 4);
-
-   /*
-    * Remaining attribs (color, texcoords, etc)
-    */
-   for (i = 1; i < nr_attrs; i++) 
-      EMIT_ATTR(slot_to_vf_attr[i], EMIT_4F, 4);
-
-   draw->vertex_size *= 4; /* floats to bytes */
-}
-			    
-
diff --git a/src/mesa/pipe/softpipe/sp_state_derived.c b/src/mesa/pipe/softpipe/sp_state_derived.c
index 95b0cfee03..eba789e9c4 100644
--- a/src/mesa/pipe/softpipe/sp_state_derived.c
+++ b/src/mesa/pipe/softpipe/sp_state_derived.c
@@ -210,22 +210,15 @@ static void calculate_vertex_layout( struct softpipe_context *softpipe )
       }
    }
 
-   /* If the attributes have changed, tell the draw module (which in turn
-    * tells the vf module) about the new vertex layout.
+   /* If the attributes have changed, tell the draw module about
+    * the new vertex layout.
     */
    if (attr_mask != softpipe->attr_mask) {
       softpipe->attr_mask = attr_mask;
 
-#define USE_NEW_DRAW 01
-#if USE_NEW_DRAW
-      draw_set_vertex_attributes2( softpipe->draw,
-				  slot_to_vf_attr,
-				  softpipe->nr_attrs );
-#else
       draw_set_vertex_attributes( softpipe->draw,
-				  slot_to_vf_attr,
-				  softpipe->nr_attrs );
-#endif
+				 slot_to_vf_attr,
+				 softpipe->nr_attrs );
    }
 }
 
diff --git a/src/mesa/pipe/softpipe/sp_state_fs.c b/src/mesa/pipe/softpipe/sp_state_fs.c
index 3505c2f1fb..98d596d4c5 100644
--- a/src/mesa/pipe/softpipe/sp_state_fs.c
+++ b/src/mesa/pipe/softpipe/sp_state_fs.c
@@ -49,4 +49,6 @@ void softpipe_set_vs_state( struct pipe_context *pipe,
    memcpy(&softpipe->vs, vs, sizeof(*vs));
 
    softpipe->dirty |= SP_NEW_VS;
+
+   draw_set_vertex_shader(softpipe->draw, vs);
 }
diff --git a/src/mesa/pipe/softpipe/sp_state_vertex.c b/src/mesa/pipe/softpipe/sp_state_vertex.c
index d985b20c36..18852552eb 100644
--- a/src/mesa/pipe/softpipe/sp_state_vertex.c
+++ b/src/mesa/pipe/softpipe/sp_state_vertex.c
@@ -41,6 +41,8 @@ softpipe_set_vertex_element(struct pipe_context *pipe,
    assert(index < PIPE_ATTRIB_MAX);
    softpipe->vertex_element[index] = *attrib; /* struct copy */
    softpipe->dirty |= SP_NEW_VERTEX;
+
+   draw_set_vertex_element(softpipe->draw, index, attrib);
 }
 
 
@@ -53,4 +55,6 @@ softpipe_set_vertex_buffer(struct pipe_context *pipe,
    assert(index < PIPE_ATTRIB_MAX);
    softpipe->vertex_buffer[index] = *buffer; /* struct copy */
    softpipe->dirty |= SP_NEW_VERTEX;
+
+   draw_set_vertex_buffer(softpipe->draw, index, buffer);
 }
diff --git a/src/mesa/sources b/src/mesa/sources
index d97a34f534..0731e64d8a 100644
--- a/src/mesa/sources
+++ b/src/mesa/sources
@@ -156,6 +156,7 @@ VF_SOURCES = \
 
 
 DRAW_SOURCES = \
+	pipe/draw/draw_arrays.c \
 	pipe/draw/draw_clip.c \
 	pipe/draw/draw_context.c\
 	pipe/draw/draw_cull.c \