remove usage of vertex_header

13 files changed, 1480 insertions, 1344 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index 0806076956..e4e144ef19 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -212,12 +212,6 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
       case EMIT_1F:
          feme->fetch[i].emit = emit_R32_FLOAT;
          break;
-      case EMIT_HEADER:
-         feme->fetch[i].ptr = (const ubyte *)&zero;
-         feme->fetch[i].pitch = 0;
-         feme->fetch[i].fetch = fetch_R32_FLOAT;
-         feme->fetch[i].emit = emit_R32_FLOAT;
-         break;
       case EMIT_1F_PSIZE:
          feme->fetch[i].ptr = (const ubyte *)&feme->draw->rasterizer->point_size;
          feme->fetch[i].pitch = 0;
diff --git a/src/gallium/auxiliary/draw/draw_vbuf.c b/src/gallium/auxiliary/draw/draw_vbuf.c
index f83b441e93..e3216ff711 100644
--- a/src/gallium/auxiliary/draw/draw_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_vbuf.c
@@ -126,13 +126,6 @@ dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data)
       case EMIT_OMIT:
          debug_printf("EMIT_OMIT:");
          break;
-      case EMIT_ALL:
-         assert(i == 0);
-         assert(j == 0);
-         debug_printf("EMIT_ALL:\t");
-         for(k = 0; k < vinfo->size*4; ++k)
-            debug_printf("%02x ", *data++);
-         break;
       case EMIT_1F:
          debug_printf("EMIT_1F:\t");
          debug_printf("%f ", *(float *)data); data += sizeof(float);
@@ -217,19 +210,6 @@ emit_vertex( struct vbuf_stage *vbuf,
          case EMIT_OMIT:
             /* no-op */
             break;
-         case EMIT_ALL:
-            /* just copy the whole vertex as-is to the vbuf */
-            assert(i == 0);
-            assert(j == 0);
-            memcpy(vbuf->vertex_ptr, vertex, vinfo->size * 4);
-            vbuf->vertex_ptr += vinfo->size;
-            count += vinfo->size;
-            break;
-         case EMIT_HEADER:
-            memcpy(vbuf->vertex_ptr, vertex, sizeof(*vertex));
-            *vbuf->vertex_ptr += sizeof(*vertex) / 4;
-            count += sizeof(*vertex) / 4;
-            break;
          case EMIT_1F:
             *vbuf->vertex_ptr++ = fui(vertex->data[j][0]);
             count++;
diff --git a/src/gallium/auxiliary/draw/draw_vertex.c b/src/gallium/auxiliary/draw/draw_vertex.c
index 970adc95e7..168036eee8 100644
--- a/src/gallium/auxiliary/draw/draw_vertex.c
+++ b/src/gallium/auxiliary/draw/draw_vertex.c
@@ -52,9 +52,6 @@ draw_compute_vertex_size(struct vertex_info *vinfo)
       switch (vinfo->emit[i]) {
       case EMIT_OMIT:
          break;
-      case EMIT_HEADER:
-         vinfo->size += sizeof(struct vertex_header) / 4;
-         break;
       case EMIT_4UB:
          /* fall-through */
       case EMIT_1F_PSIZE:
@@ -71,8 +68,6 @@ draw_compute_vertex_size(struct vertex_info *vinfo)
       case EMIT_4F:
          vinfo->size += 4;
          break;
-      case EMIT_ALL:
-         /* fall-through */
       default:
          assert(0);
       }
diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h
index abd2017ed3..65818463ca 100644
--- a/src/gallium/auxiliary/draw/draw_vertex.h
+++ b/src/gallium/auxiliary/draw/draw_vertex.h
@@ -47,8 +47,6 @@
  */
 enum attrib_emit {
    EMIT_OMIT,      /**< don't emit the attribute */
-   EMIT_ALL,       /**< emit whole post-xform vertex, w/ header */
-   EMIT_HEADER,    /**< emit vertex_header struct (XXX temp?) */
    EMIT_1F,
    EMIT_1F_PSIZE,  /**< insert constant point size */
    EMIT_2F,
diff --git a/src/gallium/auxiliary/draw/draw_vf.c b/src/gallium/auxiliary/draw/draw_vf.c
index 7bb34ace7a..9d0154c50d 100644
--- a/src/gallium/auxiliary/draw/draw_vf.c
+++ b/src/gallium/auxiliary/draw/draw_vf.c
@@ -205,7 +205,7 @@ void draw_vf_set_vertex_info( struct draw_vertex_fetch *vf,
                               const struct vertex_info *vinfo,
                               float point_size )
 {
-   unsigned i, j, k;
+   unsigned i, j;
    struct draw_vf_attr *a = vf->attr;
    struct draw_vf_attr_map attrs[PIPE_MAX_SHADER_INPUTS];
    unsigned count = 0;  /* for debug/sanity */
@@ -217,60 +217,6 @@ void draw_vf_set_vertex_info( struct draw_vertex_fetch *vf,
       case EMIT_OMIT:
          /* no-op */
          break;
-      case EMIT_ALL: {
-         /* just copy the whole vertex as-is to the vbuf */
-	 unsigned s = vinfo->size;
-         assert(i == 0);
-         assert(j == 0);
-         /* copy the vertex header */
-         /* XXX: we actually don't copy the header, just pad it */
-	 attrs[nr_attrs].attrib = 0;
-	 attrs[nr_attrs].format = DRAW_EMIT_PAD;
-	 attrs[nr_attrs].offset = offsetof(struct vertex_header, data);
-	 s -= offsetof(struct vertex_header, data)/4;
-         count += offsetof(struct vertex_header, data)/4;
-	 nr_attrs++;
-	 /* copy the vertex data */
-         for(k = 0; k < (s & ~0x3); k += 4) {
-      	    attrs[nr_attrs].attrib = k/4;
-      	    attrs[nr_attrs].format = DRAW_EMIT_4F;
-      	    attrs[nr_attrs].offset = 0;
-      	    nr_attrs++;
-            count += 4;
-         }
-         /* tail */
-         /* XXX: actually, this shouldn't be needed */
- 	 attrs[nr_attrs].attrib = k/4;
-  	 attrs[nr_attrs].offset = 0;
-         switch(s & 0x3) {
-         case 0:
-            break;
-         case 1:
-      	    attrs[nr_attrs].format = DRAW_EMIT_1F;
-      	    nr_attrs++;
-            count += 1;
-            break;
-         case 2:
-      	    attrs[nr_attrs].format = DRAW_EMIT_2F;
-      	    nr_attrs++;
-            count += 2;
-            break;
-         case 3:
-      	    attrs[nr_attrs].format = DRAW_EMIT_3F;
-      	    nr_attrs++;
-            count += 3;
-            break;
-         }
-         break;
-      }
-      case EMIT_HEADER:
-         /* XXX emit new DRAW_EMIT_HEADER attribute??? */
-	 attrs[nr_attrs].attrib = 0;
-	 attrs[nr_attrs].format = DRAW_EMIT_PAD;
-	 attrs[nr_attrs].offset = offsetof(struct vertex_header, data);
-         count += offsetof(struct vertex_header, data)/4;
-	 nr_attrs++;
-         break;
       case EMIT_1F:
 	 attrs[nr_attrs].attrib = j;
 	 attrs[nr_attrs].format = DRAW_EMIT_1F;
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
index 19dfd8c1a7..120bdfd9dd 100644
--- a/src/gallium/drivers/softpipe/Makefile
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -27,6 +27,7 @@ C_SOURCES = \
 	sp_quad_stencil.c \
 	sp_quad_stipple.c \
 	sp_screen.c \
+        sp_setup.c \
 	sp_state_blend.c \
 	sp_state_clip.c \
 	sp_state_derived.c \
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
index 06931fa8d8..c1f7daa8ab 100644
--- a/src/gallium/drivers/softpipe/SConscript
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -14,6 +14,7 @@ softpipe = env.ConvenienceLibrary(
 		'sp_flush.c',
 		'sp_prim_setup.c',
 		'sp_prim_vbuf.c',
+		'sp_setup.c',
 		'sp_quad_alpha_test.c',
 		'sp_quad_blend.c',
 		'sp_quad.c',
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c
index c7eb12b3bb..6fe463b74c 100644
--- a/src/gallium/drivers/softpipe/sp_prim_setup.c
+++ b/src/gallium/drivers/softpipe/sp_prim_setup.c
@@ -26,7 +26,9 @@
  **************************************************************************/
 
 /**
- * \brief  Primitive rasterization/rendering (points, lines, triangles)
+ * \brief A draw stage that drives our triangle setup routines from
+ * within the draw pipeline.  One of two ways to drive setup, the
+ * other being in sp_prim_vbuf.c.
  *
  * \author  Keith Whitwell <keith@tungstengraphics.com>
  * \author  Brian Paul
@@ -34,29 +36,12 @@
 
 
 #include "sp_context.h"
-#include "sp_headers.h"
-#include "sp_quad.h"
+#include "sp_setup.h"
 #include "sp_state.h"
 #include "sp_prim_setup.h"
 #include "draw/draw_private.h"
 #include "draw/draw_vertex.h"
 #include "pipe/p_util.h"
-#include "pipe/p_shader_tokens.h"
-
-#define DEBUG_VERTS 0
-#define DEBUG_FRAGS 0
-
-/**
- * Triangle edge info
- */
-struct edge {
-   float dx;		/**< X(v1) - X(v0), used only during setup */
-   float dy;		/**< Y(v1) - Y(v0), used only during setup */
-   float dxdy;		/**< dx/dy */
-   float sx, sy;	/**< first sample point coord */
-   int lines;		/**< number of lines on this edge */
-};
-
 
 /**
  * Triangle setup info (derived from draw_stage).
@@ -65,39 +50,7 @@ struct edge {
 struct setup_stage {
    struct draw_stage stage; /**< This must be first (base class) */
 
-   struct softpipe_context *softpipe;
-
-   /* Vertices are just an array of floats making up each attribute in
-    * turn.  Currently fixed at 4 floats, but should change in time.
-    * Codegen will help cope with this.
-    */
-   const struct vertex_header *vmax;
-   const struct vertex_header *vmid;
-   const struct vertex_header *vmin;
-   const struct vertex_header *vprovoke;
-
-   struct edge ebot;
-   struct edge etop;
-   struct edge emaj;
-
-   float oneoverarea;
-
-   struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS];
-   struct tgsi_interp_coef posCoef;  /* For Z, W */
-   struct quad_header quad; 
-
-   struct {
-      int left[2];   /**< [0] = row0, [1] = row1 */
-      int right[2];
-      int y;
-      unsigned y_flags;
-      unsigned mask;     /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */
-   } span;
-
-#if DEBUG_FRAGS
-   uint numFragsEmitted;  /**< per primitive */
-   uint numFragsWritten;  /**< per primitive */
-#endif
+   struct setup_context *setup;
 };
 
 
@@ -111,1112 +64,50 @@ static INLINE struct setup_stage *setup_stage( struct draw_stage *stage )
 }
 
 
-/**
- * Clip setup->quad against the scissor/surface bounds.
- */
-static INLINE void
-quad_clip(struct setup_stage *setup)
-{
-   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
-   const int minx = (int) cliprect->minx;
-   const int maxx = (int) cliprect->maxx;
-   const int miny = (int) cliprect->miny;
-   const int maxy = (int) cliprect->maxy;
-
-   if (setup->quad.x0 >= maxx ||
-       setup->quad.y0 >= maxy ||
-       setup->quad.x0 + 1 < minx ||
-       setup->quad.y0 + 1 < miny) {
-      /* totally clipped */
-      setup->quad.mask = 0x0;
-      return;
-   }
-   if (setup->quad.x0 < minx)
-      setup->quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
-   if (setup->quad.y0 < miny)
-      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
-   if (setup->quad.x0 == maxx - 1)
-      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
-   if (setup->quad.y0 == maxy - 1)
-      setup->quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
-}
-
-
-/**
- * Emit a quad (pass to next stage) with clipping.
- */
-static INLINE void
-clip_emit_quad(struct setup_stage *setup)
-{
-   quad_clip(setup);
-   if (setup->quad.mask) {
-      struct softpipe_context *sp = setup->softpipe;
-      sp->quad.first->run(sp->quad.first, &setup->quad);
-   }
-}
-
-
-/**
- * Emit a quad (pass to next stage).  No clipping is done.
- */
-static INLINE void
-emit_quad( struct setup_stage *setup, int x, int y, unsigned mask )
-{
-   struct softpipe_context *sp = setup->softpipe;
-   setup->quad.x0 = x;
-   setup->quad.y0 = y;
-   setup->quad.mask = mask;
-#if DEBUG_FRAGS
-   if (mask & 1) setup->numFragsEmitted++;
-   if (mask & 2) setup->numFragsEmitted++;
-   if (mask & 4) setup->numFragsEmitted++;
-   if (mask & 8) setup->numFragsEmitted++;
-#endif
-   sp->quad.first->run(sp->quad.first, &setup->quad);
-#if DEBUG_FRAGS
-   mask = setup->quad.mask;
-   if (mask & 1) setup->numFragsWritten++;
-   if (mask & 2) setup->numFragsWritten++;
-   if (mask & 4) setup->numFragsWritten++;
-   if (mask & 8) setup->numFragsWritten++;
-#endif
-}
-
-
-/**
- * Given an X or Y coordinate, return the block/quad coordinate that it
- * belongs to.
- */
-static INLINE int block( int x )
-{
-   return x & ~1;
-}
-
-
-/**
- * Compute mask which indicates which pixels in the 2x2 quad are actually inside
- * the triangle's bounds.
- *
- * this is pretty nasty...  may need to rework flush_spans again to
- * fix it, if possible.
- */
-static unsigned calculate_mask( struct setup_stage *setup, int x )
-{
-   unsigned mask = 0x0;
-
-   if (x >= setup->span.left[0] && x < setup->span.right[0]) 
-      mask |= MASK_TOP_LEFT;
-
-   if (x >= setup->span.left[1] && x < setup->span.right[1]) 
-      mask |= MASK_BOTTOM_LEFT;
-      
-   if (x+1 >= setup->span.left[0] && x+1 < setup->span.right[0]) 
-      mask |= MASK_TOP_RIGHT;
-
-   if (x+1 >= setup->span.left[1] && x+1 < setup->span.right[1]) 
-      mask |= MASK_BOTTOM_RIGHT;
-
-   return mask;
-}
-
-
-/**
- * Render a horizontal span of quads
- */
-static void flush_spans( struct setup_stage *setup )
-{
-   int minleft, maxright;
-   int x;
-
-   switch (setup->span.y_flags) {
-   case 0x3:
-      /* both odd and even lines written (both quad rows) */
-      minleft = MIN2(setup->span.left[0], setup->span.left[1]);
-      maxright = MAX2(setup->span.right[0], setup->span.right[1]);
-      break;
-
-   case 0x1:
-      /* only even line written (quad top row) */
-      minleft = setup->span.left[0];
-      maxright = setup->span.right[0];
-      break;
-
-   case 0x2:
-      /* only odd line written (quad bottom row) */
-      minleft = setup->span.left[1];
-      maxright = setup->span.right[1];
-      break;
-
-   default:
-      return;
-   }
-
-   /* XXX this loop could be moved into the above switch cases and
-    * calculate_mask() could be simplified a bit...
-    */
-   for (x = block(minleft); x <= block(maxright); x += 2) {
-      emit_quad( setup, x, setup->span.y, 
-                 calculate_mask( setup, x ) );
-   }
-
-   setup->span.y = 0;
-   setup->span.y_flags = 0;
-   setup->span.right[0] = 0;
-   setup->span.right[1] = 0;
-}
-
-#if DEBUG_VERTS
-static void print_vertex(const struct setup_stage *setup,
-                         const struct vertex_header *v)
-{
-   int i;
-   debug_printf("Vertex: (%p)\n", v);
-   for (i = 0; i < setup->quad.nr_attrs; i++) {
-      debug_printf("  %d: %f %f %f %f\n",  i, 
-              v->data[i][0], v->data[i][1], v->data[i][2], v->data[i][3]);
-   }
-}
-#endif
-
-static boolean setup_sort_vertices( struct setup_stage *setup,
-				      const struct prim_header *prim )
-{
-   const struct vertex_header *v0 = prim->v[0];
-   const struct vertex_header *v1 = prim->v[1];
-   const struct vertex_header *v2 = prim->v[2];
-
-#if DEBUG_VERTS
-   debug_printf("Triangle:\n");
-   print_vertex(setup, v0);
-   print_vertex(setup, v1);
-   print_vertex(setup, v2);
-#endif
-
-   setup->vprovoke = v2;
-
-   /* determine bottom to top order of vertices */
-   {
-      float y0 = v0->data[0][1];
-      float y1 = v1->data[0][1];
-      float y2 = v2->data[0][1];
-      if (y0 <= y1) {
-	 if (y1 <= y2) {
-	    /* y0<=y1<=y2 */
-	    setup->vmin = v0;   
-	    setup->vmid = v1;   
-	    setup->vmax = v2;
-	 }
-	 else if (y2 <= y0) {
-	    /* y2<=y0<=y1 */
-	    setup->vmin = v2;   
-	    setup->vmid = v0;   
-	    setup->vmax = v1;   
-	 }
-	 else {
-	    /* y0<=y2<=y1 */
-	    setup->vmin = v0;   
-	    setup->vmid = v2;   
-	    setup->vmax = v1;  
-	 }
-      }
-      else {
-	 if (y0 <= y2) {
-	    /* y1<=y0<=y2 */
-	    setup->vmin = v1;   
-	    setup->vmid = v0;   
-	    setup->vmax = v2;  
-	 }
-	 else if (y2 <= y1) {
-	    /* y2<=y1<=y0 */
-	    setup->vmin = v2;   
-	    setup->vmid = v1;   
-	    setup->vmax = v0;  
-	 }
-	 else {
-	    /* y1<=y2<=y0 */
-	    setup->vmin = v1;   
-	    setup->vmid = v2;   
-	    setup->vmax = v0;
-	 }
-      }
-   }
-
-   setup->ebot.dx = setup->vmid->data[0][0] - setup->vmin->data[0][0];
-   setup->ebot.dy = setup->vmid->data[0][1] - setup->vmin->data[0][1];
-   setup->emaj.dx = setup->vmax->data[0][0] - setup->vmin->data[0][0];
-   setup->emaj.dy = setup->vmax->data[0][1] - setup->vmin->data[0][1];
-   setup->etop.dx = setup->vmax->data[0][0] - setup->vmid->data[0][0];
-   setup->etop.dy = setup->vmax->data[0][1] - setup->vmid->data[0][1];
-
-   /*
-    * Compute triangle's area.  Use 1/area to compute partial
-    * derivatives of attributes later.
-    *
-    * The area will be the same as prim->det, but the sign may be
-    * different depending on how the vertices get sorted above.
-    *
-    * To determine whether the primitive is front or back facing we
-    * use the prim->det value because its sign is correct.
-    */
-   {
-      const float area = (setup->emaj.dx * setup->ebot.dy - 
-			    setup->ebot.dx * setup->emaj.dy);
-
-      setup->oneoverarea = 1.0f / area;
-      /*
-      debug_printf("%s one-over-area %f  area %f  det %f\n",
-                   __FUNCTION__, setup->oneoverarea, area, prim->det );
-      */
-   }
-
-   /* We need to know if this is a front or back-facing triangle for:
-    *  - the GLSL gl_FrontFacing fragment attribute (bool)
-    *  - two-sided stencil test
-    */
-   setup->quad.facing = (prim->det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
-
-   return TRUE;
-}
-
-
-/**
- * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
- * The value value comes from vertex->data[slot][i].
- * The result will be put into setup->coef[slot].a0[i].
- * \param slot  which attribute slot 
- * \param i  which component of the slot (0..3)
- */
-static void const_coeff( struct setup_stage *setup,
-                         struct tgsi_interp_coef *coef,
-                         uint vertSlot, uint i)
-{
-   assert(i <= 3);
-
-   coef->dadx[i] = 0;
-   coef->dady[i] = 0;
 
-   /* need provoking vertex info!
-    */
-   coef->a0[i] = setup->vprovoke->data[vertSlot][i];
-}
-
-
-/**
- * Compute a0, dadx and dady for a linearly interpolated coefficient,
- * for a triangle.
- */
-static void tri_linear_coeff( struct setup_stage *setup,
-                              struct tgsi_interp_coef *coef,
-                              uint vertSlot, uint i)
-{
-   float botda = setup->vmid->data[vertSlot][i] - setup->vmin->data[vertSlot][i];
-   float majda = setup->vmax->data[vertSlot][i] - setup->vmin->data[vertSlot][i];
-   float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
-   float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
-   float dadx = a * setup->oneoverarea;
-   float dady = b * setup->oneoverarea;
-
-   assert(i <= 3);
-
-   coef->dadx[i] = dadx;
-   coef->dady[i] = dady;
-
-   /* calculate a0 as the value which would be sampled for the
-    * fragment at (0,0), taking into account that we want to sample at
-    * pixel centers, in other words (0.5, 0.5).
-    *
-    * this is neat but unfortunately not a good way to do things for
-    * triangles with very large values of dadx or dady as it will
-    * result in the subtraction and re-addition from a0 of a very
-    * large number, which means we'll end up loosing a lot of the
-    * fractional bits and precision from a0.  the way to fix this is
-    * to define a0 as the sample at a pixel center somewhere near vmin
-    * instead - i'll switch to this later.
-    */
-   coef->a0[i] = (setup->vmin->data[vertSlot][i] - 
-                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
-                   dady * (setup->vmin->data[0][1] - 0.5f)));
-
-   /*
-   debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
-		slot, "xyzw"[i], 
-		setup->coef[slot].a0[i],
-		setup->coef[slot].dadx[i],
-		setup->coef[slot].dady[i]);
-   */
-}
-
-
-/**
- * Compute a0, dadx and dady for a perspective-corrected interpolant,
- * for a triangle.
- * We basically multiply the vertex value by 1/w before computing
- * the plane coefficients (a0, dadx, dady).
- * Later, when we compute the value at a particular fragment position we'll
- * divide the interpolated value by the interpolated W at that fragment.
- */
-static void tri_persp_coeff( struct setup_stage *setup,
-                             struct tgsi_interp_coef *coef,
-                             uint vertSlot, uint i)
-{
-   /* premultiply by 1/w  (v->data[0][3] is always W):
-    */
-   float mina = setup->vmin->data[vertSlot][i] * setup->vmin->data[0][3];
-   float mida = setup->vmid->data[vertSlot][i] * setup->vmid->data[0][3];
-   float maxa = setup->vmax->data[vertSlot][i] * setup->vmax->data[0][3];
-   float botda = mida - mina;
-   float majda = maxa - mina;
-   float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
-   float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
-   float dadx = a * setup->oneoverarea;
-   float dady = b * setup->oneoverarea;
-      
-   /*
-   debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
-          	setup->vmin->data[vertSlot][i],
-          	setup->vmid->data[vertSlot][i],
-       		setup->vmax->data[vertSlot][i]
-          );
-   */
-   assert(i <= 3);
-
-   coef->dadx[i] = dadx;
-   coef->dady[i] = dady;
-   coef->a0[i] = (mina - 
-                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
-                   dady * (setup->vmin->data[0][1] - 0.5f)));
-}
-
-
-/**
- * Special coefficient setup for gl_FragCoord.
- * X and Y are trivial, though Y has to be inverted for OpenGL.
- * Z and W are copied from posCoef which should have already been computed.
- * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
- */
 static void
-setup_fragcoord_coeff(struct setup_stage *setup, uint slot)
-{
-   /*X*/
-   setup->coef[slot].a0[0] = 0;
-   setup->coef[slot].dadx[0] = 1.0;
-   setup->coef[slot].dady[0] = 0.0;
-   /*Y*/
-   if (setup->softpipe->rasterizer->origin_lower_left) {
-      /* y=0=bottom */
-      const int winHeight = setup->softpipe->framebuffer.height;
-      setup->coef[slot].a0[1] = (float) (winHeight - 1);
-      setup->coef[slot].dady[1] = -1.0;
-   }
-   else {
-      /* y=0=top */
-      setup->coef[slot].a0[1] = 0.0;
-      setup->coef[slot].dady[1] = 1.0;
-   }
-   setup->coef[slot].dadx[1] = 0.0;
-   /*Z*/
-   setup->coef[slot].a0[2] = setup->posCoef.a0[2];
-   setup->coef[slot].dadx[2] = setup->posCoef.dadx[2];
-   setup->coef[slot].dady[2] = setup->posCoef.dady[2];
-   /*W*/
-   setup->coef[slot].a0[3] = setup->posCoef.a0[3];
-   setup->coef[slot].dadx[3] = setup->posCoef.dadx[3];
-   setup->coef[slot].dady[3] = setup->posCoef.dady[3];
-}
-
-
-
-/**
- * Compute the setup->coef[] array dadx, dady, a0 values.
- * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized.
- */
-static void setup_tri_coefficients( struct setup_stage *setup )
-{
-   struct softpipe_context *softpipe = setup->softpipe;
-   const struct sp_fragment_shader *spfs = softpipe->fs;
-   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
-   uint fragSlot;
-
-   /* z and w are done by linear interpolation:
-    */
-   tri_linear_coeff(setup, &setup->posCoef, 0, 2);
-   tri_linear_coeff(setup, &setup->posCoef, 0, 3);
-
-   /* setup interpolation for all the remaining attributes:
-    */
-   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->src_index[fragSlot];
-      uint j;
-
-      switch (vinfo->interp_mode[fragSlot]) {
-      case INTERP_CONSTANT:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_LINEAR:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            tri_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_PERSPECTIVE:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_POS:
-         setup_fragcoord_coeff(setup, fragSlot);
-         break;
-      default:
-         assert(0);
-      }
-
-      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
-         /* FOG.y = front/back facing  XXX fix this */
-         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
-         setup->coef[fragSlot].dadx[1] = 0.0;
-         setup->coef[fragSlot].dady[1] = 0.0;
-      }
-   }
-}
-
-
-
-static void setup_tri_edges( struct setup_stage *setup )
-{
-   float vmin_x = setup->vmin->data[0][0] + 0.5f;
-   float vmid_x = setup->vmid->data[0][0] + 0.5f;
-
-   float vmin_y = setup->vmin->data[0][1] - 0.5f;
-   float vmid_y = setup->vmid->data[0][1] - 0.5f;
-   float vmax_y = setup->vmax->data[0][1] - 0.5f;
-
-   setup->emaj.sy = CEILF(vmin_y);
-   setup->emaj.lines = (int) CEILF(vmax_y - setup->emaj.sy);
-   setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy;
-   setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy;
-
-   setup->etop.sy = CEILF(vmid_y);
-   setup->etop.lines = (int) CEILF(vmax_y - setup->etop.sy);
-   setup->etop.dxdy = setup->etop.dx / setup->etop.dy;
-   setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy;
-
-   setup->ebot.sy = CEILF(vmin_y);
-   setup->ebot.lines = (int) CEILF(vmid_y - setup->ebot.sy);
-   setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy;
-   setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy;
-}
-
-
-/**
- * Render the upper or lower half of a triangle.
- * Scissoring/cliprect is applied here too.
- */
-static void subtriangle( struct setup_stage *setup,
-			 struct edge *eleft,
-			 struct edge *eright,
-			 unsigned lines )
-{
-   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
-   const int minx = (int) cliprect->minx;
-   const int maxx = (int) cliprect->maxx;
-   const int miny = (int) cliprect->miny;
-   const int maxy = (int) cliprect->maxy;
-   int y, start_y, finish_y;
-   int sy = (int)eleft->sy;
-
-   assert((int)eleft->sy == (int) eright->sy);
-
-   /* clip top/bottom */
-   start_y = sy;
-   finish_y = sy + lines;
-
-   if (start_y < miny)
-      start_y = miny;
-
-   if (finish_y > maxy)
-      finish_y = maxy;
-
-   start_y -= sy;
-   finish_y -= sy;
-
-   /*
-   debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);  
-   */
-
-   for (y = start_y; y < finish_y; y++) {
-
-      /* avoid accumulating adds as floats don't have the precision to
-       * accurately iterate large triangle edges that way.  luckily we
-       * can just multiply these days.
-       *
-       * this is all drowned out by the attribute interpolation anyway.
-       */
-      int left = (int)(eleft->sx + y * eleft->dxdy);
-      int right = (int)(eright->sx + y * eright->dxdy);
-
-      /* clip left/right */
-      if (left < minx)
-         left = minx;
-      if (right > maxx)
-         right = maxx;
-
-      if (left < right) {
-         int _y = sy + y;
-         if (block(_y) != setup->span.y) {
-            flush_spans(setup);
-            setup->span.y = block(_y);
-         }
-
-         setup->span.left[_y&1] = left;
-         setup->span.right[_y&1] = right;
-         setup->span.y_flags |= 1<<(_y&1);
-      }
-   }
-
-
-   /* save the values so that emaj can be restarted:
-    */
-   eleft->sx += lines * eleft->dxdy;
-   eright->sx += lines * eright->dxdy;
-   eleft->sy += lines;
-   eright->sy += lines;
-}
-
-
-/**
- * Do setup for triangle rasterization, then render the triangle.
- */
-static void setup_tri( struct draw_stage *stage,
-		       struct prim_header *prim )
+do_tri(struct draw_stage *stage, struct prim_header *prim)
 {
    struct setup_stage *setup = setup_stage( stage );
 
-   /*
-   debug_printf("%s\n", __FUNCTION__ );
-   */
-
-#if DEBUG_FRAGS
-   setup->numFragsEmitted = 0;
-   setup->numFragsWritten = 0;
-#endif
-
-   setup_sort_vertices( setup, prim );
-   setup_tri_coefficients( setup );
-   setup_tri_edges( setup );
-
-   setup->quad.prim = PRIM_TRI;
-
-   setup->span.y = 0;
-   setup->span.y_flags = 0;
-   setup->span.right[0] = 0;
-   setup->span.right[1] = 0;
-   /*   setup->span.z_mode = tri_z_mode( setup->ctx ); */
-
-   /*   init_constant_attribs( setup ); */
-      
-   if (setup->oneoverarea < 0.0) {
-      /* emaj on left:
-       */
-      subtriangle( setup, &setup->emaj, &setup->ebot, setup->ebot.lines );
-      subtriangle( setup, &setup->emaj, &setup->etop, setup->etop.lines );
-   }
-   else {
-      /* emaj on right:
-       */
-      subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines );
-      subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines );
-   }
-
-   flush_spans( setup );
-
-#if DEBUG_FRAGS
-   printf("Tri: %u frags emitted, %u written\n",
-          setup->numFragsEmitted,
-          setup->numFragsWritten);
-#endif
-}
-
-
-
-/**
- * Compute a0, dadx and dady for a linearly interpolated coefficient,
- * for a line.
- */
-static void
-line_linear_coeff(struct setup_stage *setup,
-                  struct tgsi_interp_coef *coef,
-                  uint vertSlot, uint i)
-{
-   const float da = setup->vmax->data[vertSlot][i] - setup->vmin->data[vertSlot][i];
-   const float dadx = da * setup->emaj.dx * setup->oneoverarea;
-   const float dady = da * setup->emaj.dy * setup->oneoverarea;
-   coef->dadx[i] = dadx;
-   coef->dady[i] = dady;
-   coef->a0[i] = (setup->vmin->data[vertSlot][i] - 
-                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
-                   dady * (setup->vmin->data[0][1] - 0.5f)));
-}
-
-
-/**
- * Compute a0, dadx and dady for a perspective-corrected interpolant,
- * for a line.
- */
-static void
-line_persp_coeff(struct setup_stage *setup,
-                  struct tgsi_interp_coef *coef,
-                  uint vertSlot, uint i)
-{
-   /* XXX double-check/verify this arithmetic */
-   const float a0 = setup->vmin->data[vertSlot][i] * setup->vmin->data[0][3];
-   const float a1 = setup->vmax->data[vertSlot][i] * setup->vmax->data[0][3];
-   const float da = a1 - a0;
-   const float dadx = da * setup->emaj.dx * setup->oneoverarea;
-   const float dady = da * setup->emaj.dy * setup->oneoverarea;
-   coef->dadx[i] = dadx;
-   coef->dady[i] = dady;
-   coef->a0[i] = (setup->vmin->data[vertSlot][i] - 
-                  (dadx * (setup->vmin->data[0][0] - 0.5f) + 
-                   dady * (setup->vmin->data[0][1] - 0.5f)));
-}
-
-
-/**
- * Compute the setup->coef[] array dadx, dady, a0 values.
- * Must be called after setup->vmin,vmax are initialized.
- */
-static INLINE void
-setup_line_coefficients(struct setup_stage *setup, struct prim_header *prim)
-{
-   struct softpipe_context *softpipe = setup->softpipe;
-   const struct sp_fragment_shader *spfs = softpipe->fs;
-   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
-   uint fragSlot;
-
-   /* use setup->vmin, vmax to point to vertices */
-   setup->vprovoke = prim->v[1];
-   setup->vmin = prim->v[0];
-   setup->vmax = prim->v[1];
-
-   setup->emaj.dx = setup->vmax->data[0][0] - setup->vmin->data[0][0];
-   setup->emaj.dy = setup->vmax->data[0][1] - setup->vmin->data[0][1];
-   /* NOTE: this is not really 1/area */
-   setup->oneoverarea = 1.0f / (setup->emaj.dx * setup->emaj.dx +
-                                setup->emaj.dy * setup->emaj.dy);
-
-   /* z and w are done by linear interpolation:
-    */
-   line_linear_coeff(setup, &setup->posCoef, 0, 2);
-   line_linear_coeff(setup, &setup->posCoef, 0, 3);
-
-   /* setup interpolation for all the remaining attributes:
-    */
-   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->src_index[fragSlot];
-      uint j;
-
-      switch (vinfo->interp_mode[fragSlot]) {
-      case INTERP_CONSTANT:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_LINEAR:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            line_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_PERSPECTIVE:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_POS:
-         setup_fragcoord_coeff(setup, fragSlot);
-         break;
-      default:
-         assert(0);
-      }
-
-      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
-         /* FOG.y = front/back facing  XXX fix this */
-         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
-         setup->coef[fragSlot].dadx[1] = 0.0;
-         setup->coef[fragSlot].dady[1] = 0.0;
-      }
-   }
+   setup_tri( setup->setup,
+              prim->det,
+              prim->v[0]->data,
+              prim->v[1]->data,
+              prim->v[2]->data );
 }
 
-
-/**
- * Plot a pixel in a line segment.
- */
-static INLINE void
-plot(struct setup_stage *setup, int x, int y)
-{
-   const int iy = y & 1;
-   const int ix = x & 1;
-   const int quadX = x - ix;
-   const int quadY = y - iy;
-   const int mask = (1 << ix) << (2 * iy);
-
-   if (quadX != setup->quad.x0 || 
-       quadY != setup->quad.y0) 
-   {
-      /* flush prev quad, start new quad */
-
-      if (setup->quad.x0 != -1)
-         clip_emit_quad(setup);
-
-      setup->quad.x0 = quadX;
-      setup->quad.y0 = quadY;
-      setup->quad.mask = 0x0;
-   }
-
-   setup->quad.mask |= mask;
-}
-
-
-/**
- * Do setup for line rasterization, then render the line.
- * Single-pixel width, no stipple, etc.  We rely on the 'draw' module
- * to handle stippling and wide lines.
- */
 static void
-setup_line(struct draw_stage *stage, struct prim_header *prim)
+do_line(struct draw_stage *stage, struct prim_header *prim)
 {
-   const struct vertex_header *v0 = prim->v[0];
-   const struct vertex_header *v1 = prim->v[1];
    struct setup_stage *setup = setup_stage( stage );
-   int x0 = (int) v0->data[0][0];
-   int x1 = (int) v1->data[0][0];
-   int y0 = (int) v0->data[0][1];
-   int y1 = (int) v1->data[0][1];
-   int dx = x1 - x0;
-   int dy = y1 - y0;
-   int xstep, ystep;
 
-   if (dx == 0 && dy == 0)
-      return;
-
-   setup_line_coefficients(setup, prim);
-
-   if (dx < 0) {
-      dx = -dx;   /* make positive */
-      xstep = -1;
-   }
-   else {
-      xstep = 1;
-   }
-
-   if (dy < 0) {
-      dy = -dy;   /* make positive */
-      ystep = -1;
-   }
-   else {
-      ystep = 1;
-   }
-
-   assert(dx >= 0);
-   assert(dy >= 0);
-
-   setup->quad.x0 = setup->quad.y0 = -1;
-   setup->quad.mask = 0x0;
-   setup->quad.prim = PRIM_LINE;
-   /* XXX temporary: set coverage to 1.0 so the line appears
-    * if AA mode happens to be enabled.
-    */
-   setup->quad.coverage[0] =
-   setup->quad.coverage[1] =
-   setup->quad.coverage[2] =
-   setup->quad.coverage[3] = 1.0;
-
-   if (dx > dy) {
-      /*** X-major line ***/
-      int i;
-      const int errorInc = dy + dy;
-      int error = errorInc - dx;
-      const int errorDec = error - dx;
-
-      for (i = 0; i < dx; i++) {
-         plot(setup, x0, y0);
-
-         x0 += xstep;
-         if (error < 0) {
-            error += errorInc;
-         }
-         else {
-            error += errorDec;
-            y0 += ystep;
-         }
-      }
-   }
-   else {
-      /*** Y-major line ***/
-      int i;
-      const int errorInc = dx + dx;
-      int error = errorInc - dy;
-      const int errorDec = error - dy;
-
-      for (i = 0; i < dy; i++) {
-         plot(setup, x0, y0);
-
-         y0 += ystep;
-         if (error < 0) {
-            error += errorInc;
-         }
-         else {
-            error += errorDec;
-            x0 += xstep;
-         }
-      }
-   }
-
-   /* draw final quad */
-   if (setup->quad.mask) {
-      clip_emit_quad(setup);
-   }
+   setup_line( setup->setup,
+               prim->v[0]->data,
+               prim->v[1]->data );
 }
 
-
 static void
-point_persp_coeff(struct setup_stage *setup,
-                  const struct vertex_header *vert,
-                  struct tgsi_interp_coef *coef,
-                  uint vertSlot, uint i)
-{
-   assert(i <= 3);
-   coef->dadx[i] = 0.0F;
-   coef->dady[i] = 0.0F;
-   coef->a0[i] = vert->data[vertSlot][i] * vert->data[0][3];
-}
-
-
-/**
- * Do setup for point rasterization, then render the point.
- * Round or square points...
- * XXX could optimize a lot for 1-pixel points.
- */
-static void
-setup_point(struct draw_stage *stage, struct prim_header *prim)
+do_point(struct draw_stage *stage, struct prim_header *prim)
 {
    struct setup_stage *setup = setup_stage( stage );
-   struct softpipe_context *softpipe = setup->softpipe;
-   const struct sp_fragment_shader *spfs = softpipe->fs;
-   const struct vertex_header *v0 = prim->v[0];
-   const int sizeAttr = setup->softpipe->psize_slot;
-   const float size
-      = sizeAttr > 0 ? v0->data[sizeAttr][0]
-      : setup->softpipe->rasterizer->point_size;
-   const float halfSize = 0.5F * size;
-   const boolean round = (boolean) setup->softpipe->rasterizer->point_smooth;
-   const float x = v0->data[0][0];  /* Note: data[0] is always position */
-   const float y = v0->data[0][1];
-   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
-   uint fragSlot;
-
-   /* For points, all interpolants are constant-valued.
-    * However, for point sprites, we'll need to setup texcoords appropriately.
-    * XXX: which coefficients are the texcoords???
-    * We may do point sprites as textured quads...
-    *
-    * KW: We don't know which coefficients are texcoords - ultimately
-    * the choice of what interpolation mode to use for each attribute
-    * should be determined by the fragment program, using
-    * per-attribute declaration statements that include interpolation
-    * mode as a parameter.  So either the fragment program will have
-    * to be adjusted for pointsprite vs normal point behaviour, or
-    * otherwise a special interpolation mode will have to be defined
-    * which matches the required behaviour for point sprites.  But -
-    * the latter is not a feature of normal hardware, and as such
-    * probably should be ruled out on that basis.
-    */
-   setup->vprovoke = prim->v[0];
-
-   /* setup Z, W */
-   const_coeff(setup, &setup->posCoef, 0, 2);
-   const_coeff(setup, &setup->posCoef, 0, 3);
-
-   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->src_index[fragSlot];
-      uint j;
-
-      switch (vinfo->interp_mode[fragSlot]) {
-      case INTERP_CONSTANT:
-         /* fall-through */
-      case INTERP_LINEAR:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_PERSPECTIVE:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            point_persp_coeff(setup, setup->vprovoke,
-                              &setup->coef[fragSlot], vertSlot, j);
-         break;
-      case INTERP_POS:
-         setup_fragcoord_coeff(setup, fragSlot);
-         break;
-      default:
-         assert(0);
-      }
-
-      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
-         /* FOG.y = front/back facing  XXX fix this */
-         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
-         setup->coef[fragSlot].dadx[1] = 0.0;
-         setup->coef[fragSlot].dady[1] = 0.0;
-      }
-   }
-
-   setup->quad.prim = PRIM_POINT;
-
-   if (halfSize <= 0.5 && !round) {
-      /* special case for 1-pixel points */
-      const int ix = ((int) x) & 1;
-      const int iy = ((int) y) & 1;
-      setup->quad.x0 = (int) x - ix;
-      setup->quad.y0 = (int) y - iy;
-      setup->quad.mask = (1 << ix) << (2 * iy);
-      clip_emit_quad(setup);
-   }
-   else {
-      if (round) {
-         /* rounded points */
-         const int ixmin = block((int) (x - halfSize));
-         const int ixmax = block((int) (x + halfSize));
-         const int iymin = block((int) (y - halfSize));
-         const int iymax = block((int) (y + halfSize));
-         const float rmin = halfSize - 0.7071F;  /* 0.7071 = sqrt(2)/2 */
-         const float rmax = halfSize + 0.7071F;
-         const float rmin2 = MAX2(0.0F, rmin * rmin);
-         const float rmax2 = rmax * rmax;
-         const float cscale = 1.0F / (rmax2 - rmin2);
-         int ix, iy;
-
-         for (iy = iymin; iy <= iymax; iy += 2) {
-            for (ix = ixmin; ix <= ixmax; ix += 2) {
-               float dx, dy, dist2, cover;
-
-               setup->quad.mask = 0x0;
-
-               dx = (ix + 0.5f) - x;
-               dy = (iy + 0.5f) - y;
-               dist2 = dx * dx + dy * dy;
-               if (dist2 <= rmax2) {
-                  cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_TOP_LEFT;
-               }
-
-               dx = (ix + 1.5f) - x;
-               dy = (iy + 0.5f) - y;
-               dist2 = dx * dx + dy * dy;
-               if (dist2 <= rmax2) {
-                  cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_TOP_RIGHT;
-               }
 
-               dx = (ix + 0.5f) - x;
-               dy = (iy + 1.5f) - y;
-               dist2 = dx * dx + dy * dy;
-               if (dist2 <= rmax2) {
-                  cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_BOTTOM_LEFT;
-               }
-
-               dx = (ix + 1.5f) - x;
-               dy = (iy + 1.5f) - y;
-               dist2 = dx * dx + dy * dy;
-               if (dist2 <= rmax2) {
-                  cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_BOTTOM_RIGHT;
-               }
-
-               if (setup->quad.mask) {
-                  setup->quad.x0 = ix;
-                  setup->quad.y0 = iy;
-                  clip_emit_quad(setup);
-               }
-            }
-         }
-      }
-      else {
-         /* square points */
-         const int xmin = (int) (x + 0.75 - halfSize);
-         const int ymin = (int) (y + 0.25 - halfSize);
-         const int xmax = xmin + (int) size;
-         const int ymax = ymin + (int) size;
-         /* XXX could apply scissor to xmin,ymin,xmax,ymax now */
-         const int ixmin = block(xmin);
-         const int ixmax = block(xmax - 1);
-         const int iymin = block(ymin);
-         const int iymax = block(ymax - 1);
-         int ix, iy;
-
-         /*
-         debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax);
-         */
-         for (iy = iymin; iy <= iymax; iy += 2) {
-            uint rowMask = 0xf;
-            if (iy < ymin) {
-               /* above the top edge */
-               rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
-            }
-            if (iy + 1 >= ymax) {
-               /* below the bottom edge */
-               rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
-            }
-
-            for (ix = ixmin; ix <= ixmax; ix += 2) {
-               uint mask = rowMask;
-
-               if (ix < xmin) {
-                  /* fragment is past left edge of point, turn off left bits */
-                  mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
-               }
-               if (ix + 1 >= xmax) {
-                  /* past the right edge */
-                  mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
-               }
-                  
-               setup->quad.mask = mask;
-               setup->quad.x0 = ix;
-               setup->quad.y0 = iy;
-               clip_emit_quad(setup);
-            }
-         }
-      }
-   }
+   setup_point( setup->setup,
+                prim->v[0]->data );
 }
 
 
 
+
 static void setup_begin( struct draw_stage *stage )
 {
    struct setup_stage *setup = setup_stage(stage);
-   struct softpipe_context *sp = setup->softpipe;
-   const struct sp_fragment_shader *fs = setup->softpipe->fs;
-   uint i;
-
-   if (sp->dirty) {
-      softpipe_update_derived(sp);
-   }
 
-   /* Mark surfaces as defined now */
-   for (i = 0; i < sp->framebuffer.num_cbufs; i++){
-      if (sp->framebuffer.cbufs[i]) {
-         sp->framebuffer.cbufs[i]->status = PIPE_SURFACE_STATUS_DEFINED;
-      }
-   }
-   if (sp->framebuffer.zsbuf) {
-      sp->framebuffer.zsbuf->status = PIPE_SURFACE_STATUS_DEFINED;
-   }
+   setup_prepare( setup->setup );
 
-   setup->quad.nr_attrs = fs->info.num_inputs;
-
-   sp->quad.first->begin(sp->quad.first);
-
-   stage->point = setup_point;
-   stage->line = setup_line;
-   stage->tri = setup_tri;
+   stage->point = do_point;
+   stage->line = do_line;
+   stage->tri = do_tri;
 }
 
 
@@ -1269,19 +160,29 @@ static void render_destroy( struct draw_stage *stage )
  */
 struct draw_stage *sp_draw_render_stage( struct softpipe_context *softpipe )
 {
-   struct setup_stage *setup = CALLOC_STRUCT(setup_stage);
+   struct setup_stage *sstage = CALLOC_STRUCT(setup_stage);
 
-   setup->softpipe = softpipe;
-   setup->stage.draw = softpipe->draw;
-   setup->stage.point = setup_first_point;
-   setup->stage.line = setup_first_line;
-   setup->stage.tri = setup_first_tri;
-   setup->stage.flush = setup_flush;
-   setup->stage.reset_stipple_counter = reset_stipple_counter;
-   setup->stage.destroy = render_destroy;
+   sstage->setup = setup_create_context(softpipe);
+   sstage->stage.draw = softpipe->draw;
+   sstage->stage.point = setup_first_point;
+   sstage->stage.line = setup_first_line;
+   sstage->stage.tri = setup_first_tri;
+   sstage->stage.flush = setup_flush;
+   sstage->stage.reset_stipple_counter = reset_stipple_counter;
+   sstage->stage.destroy = render_destroy;
 
-   setup->quad.coef = setup->coef;
-   setup->quad.posCoef = &setup->posCoef;
+   return (struct draw_stage *)sstage;
+}
 
-   return &setup->stage;
+struct setup_context *
+sp_draw_setup_context( struct draw_stage *stage )
+{
+   struct setup_stage *ssetup = setup_stage(stage);
+   return ssetup->setup;
+}
+
+void
+sp_draw_flush( struct draw_stage *stage )
+{
+   stage->flush( stage, 0 );
 }
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.h b/src/gallium/drivers/softpipe/sp_prim_setup.h
index f3e8a79dd9..49bdd98ed8 100644
--- a/src/gallium/drivers/softpipe/sp_prim_setup.h
+++ b/src/gallium/drivers/softpipe/sp_prim_setup.h
@@ -69,6 +69,12 @@ typedef void (*vbuf_draw_func)( struct pipe_context *pipe,
 extern struct draw_stage *
 sp_draw_render_stage( struct softpipe_context *softpipe );
 
+extern struct setup_context *
+sp_draw_setup_context( struct draw_stage * );
+
+extern void
+sp_draw_flush( struct draw_stage * );
+
 
 extern struct draw_stage *
 sp_draw_vbuf_stage( struct draw_context *draw_context,
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index d940718ed2..025a0113bf 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -1,8 +1,8 @@
 /**************************************************************************
- * 
+ *
  * 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
@@ -10,11 +10,11 @@
  * 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.
@@ -22,7 +22,7 @@
  * 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.
- * 
+ *
  **************************************************************************/
 
 /**
@@ -30,7 +30,7 @@
  * softpipe rendering pipeline.
  * Probably not desired in general, but useful for testing/debuggin.
  * Enabled/Disabled with SP_VBUF env var.
- * 
+ *
  * Authors
  *  Brian Paul
  */
@@ -39,14 +39,16 @@
 #include "sp_context.h"
 #include "sp_state.h"
 #include "sp_prim_vbuf.h"
+#include "sp_prim_setup.h"
+#include "sp_setup.h"
 #include "draw/draw_context.h"
-#include "draw/draw_private.h"
 #include "draw/draw_vbuf.h"
 
 
 #define SP_MAX_VBUF_INDEXES 1024
 #define SP_MAX_VBUF_SIZE    4096
 
+typedef const float (*cptrf4)[4];
 
 /**
  * Subclass of vbuf_render.
@@ -69,7 +71,6 @@ softpipe_vbuf_render(struct vbuf_render *vbr)
 }
 
 
-
 static const struct vertex_info *
 sp_vbuf_get_vertex_info(struct vbuf_render *vbr)
 {
@@ -91,7 +92,7 @@ sp_vbuf_allocate_vertices(struct vbuf_render *vbr,
 
 
 static void
-sp_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices, 
+sp_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices,
                            unsigned vertex_size, unsigned vertices_used)
 {
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
@@ -114,7 +115,7 @@ sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
    else {
       return FALSE;
    }
-      
+
 }
 
 
@@ -122,22 +123,19 @@ sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
  * Recalculate prim's determinant.
  * XXX is this needed?
  */
-static void
-calc_det(struct prim_header *header)
+static float
+calc_det( const float (*v0)[4],
+          const float (*v1)[4],
+          const float (*v2)[4] )
 {
-   /* Window coords: */
-   const float *v0 = header->v[0]->data[0];
-   const float *v1 = header->v[1]->data[0];
-   const float *v2 = header->v[2]->data[0];
-
    /* edge vectors e = v0 - v2, f = v1 - v2 */
-   const float ex = v0[0] - v2[0];
-   const float ey = v0[1] - v2[1];
-   const float fx = v1[0] - v2[0];
-   const float fy = v1[1] - v2[1];
-   
+   const float ex = v0[0][0] - v2[0][0];
+   const float ey = v0[0][1] - v2[0][1];
+   const float fx = v1[0][0] - v2[0][0];
+   const float fy = v1[0][1] - v2[0][1];
+
    /* det = cross(e,f).z */
-   header->det = ex * fy - ey * fx;
+   return ex * fy - ey * fx;
 }
 
 
@@ -147,48 +145,51 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr_indices)
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
    struct softpipe_context *softpipe = cvbr->softpipe;
    struct draw_stage *setup = softpipe->setup;
-   struct prim_header prim;
    unsigned vertex_size = softpipe->vertex_info_vbuf.size * sizeof(float);
    unsigned i, j;
    void *vertex_buffer = cvbr->vertex_buffer;
-
-   prim.det = 0;
-   prim.reset_line_stipple = 0;
-   prim.edgeflags = 0;
-   prim.pad = 0;
+   cptrf4 v[3];
+   struct setup_context *setup_ctx = sp_draw_setup_context(setup);
 
    switch (cvbr->prim) {
    case PIPE_PRIM_TRIANGLES:
       for (i = 0; i < nr_indices; i += 3) {
-         for (j = 0; j < 3; j++) 
-            prim.v[j] = (struct vertex_header *)((char *)vertex_buffer + 
-                                                 indices[i+j] * vertex_size);
-         
-         calc_det(&prim);
-         setup->tri( setup, &prim );
+         for (j = 0; j < 3; j++)
+            v[j] = (cptrf4)((char *)vertex_buffer +
+                             indices[i+j] * vertex_size);
+
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2]);
       }
       break;
 
    case PIPE_PRIM_LINES:
       for (i = 0; i < nr_indices; i += 2) {
-         for (j = 0; j < 2; j++) 
-            prim.v[j] = (struct vertex_header *)((char *)vertex_buffer + 
-                                                 indices[i+j] * vertex_size);
-         
-         setup->line( setup, &prim );
+         for (j = 0; j < 2; j++)
+            v[j] = (cptrf4)((char *)vertex_buffer +
+                            indices[i+j] * vertex_size);
+
+         setup_line( setup_ctx,
+                     v[0],
+                     v[1] );
       }
       break;
 
    case PIPE_PRIM_POINTS:
       for (i = 0; i < nr_indices; i++) {
-         prim.v[0] = (struct vertex_header *)((char *)vertex_buffer + 
-                                              indices[i] * vertex_size);         
-         setup->point( setup, &prim );
+         v[0] = (cptrf4)((char *)vertex_buffer +
+                         indices[i] * vertex_size);
+
+         setup_point( setup_ctx,
+                      v[0] );
       }
       break;
    }
 
-   setup->flush( setup, 0 );
+   sp_draw_flush( setup );
 }
 
 
@@ -202,111 +203,131 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
    struct softpipe_context *softpipe = cvbr->softpipe;
    struct draw_stage *setup = softpipe->setup;
-   struct prim_header prim;
    const void *vertex_buffer = cvbr->vertex_buffer;
    const unsigned vertex_size = softpipe->vertex_info_vbuf.size * sizeof(float);
    unsigned i;
-
-   prim.det = 0;
-   prim.reset_line_stipple = 0;
-   prim.edgeflags = 0;
-   prim.pad = 0;
+   struct setup_context *setup_ctx = sp_draw_setup_context(setup);
+   cptrf4 v[3];
 
 #define VERTEX(I) \
-   (struct vertex_header *) ((char *) vertex_buffer + (I) * vertex_size)
+   (cptrf4) ((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 );
+         v[0] = VERTEX(i);
+         setup_point( setup_ctx, v[0] );
       }
       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 );
+         v[0] = VERTEX(i);
+         v[1] = VERTEX(i + 1);
+         setup_line( setup_ctx, v[0], v[1] );
       }
       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 );
+         v[0] = VERTEX(i - 1);
+         v[1] = VERTEX(i);
+         setup_line( setup_ctx, v[0], v[1] );
       }
       break;
    case PIPE_PRIM_TRIANGLES:
       assert(nr % 3 == 0);
       for (i = 0; i < nr; i += 3) {
-         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 );
+         v[0] = VERTEX(i + 0);
+         v[1] = VERTEX(i + 1);
+         v[2] = VERTEX(i + 2);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
       }
       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 );
+         v[0] = VERTEX(i - 2);
+         v[1] = VERTEX(i - 1);
+         v[2] = VERTEX(i);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
       }
       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 );
+         v[0] = VERTEX(0);
+         v[1] = VERTEX(i - 1);
+         v[2] = VERTEX(i);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
       }
       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 );
+         v[0] = VERTEX(i + 0);
+         v[1] = VERTEX(i + 1);
+         v[2] = VERTEX(i + 2);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
+
+         v[0] = VERTEX(i + 0);
+         v[1] = VERTEX(i + 2);
+         v[2] = VERTEX(i + 3);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
       }
       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 );
+         v[0] = VERTEX(i - 2);
+         v[1] = VERTEX(i);
+         v[2] = VERTEX(i + 1);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
+
+         v[0] = VERTEX(i - 2);
+         v[1] = VERTEX(i + 1);
+         v[2] = VERTEX(i - 1);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
       }
       break;
    case PIPE_PRIM_POLYGON:
       /* draw as tri fan */
       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 );
+         v[0] = VERTEX(0);
+         v[1] = VERTEX(i - 1);
+         v[2] = VERTEX(i);
+         setup_tri( setup_ctx,
+                    calc_det(v[0], v[1], v[2]),
+                    v[0],
+                    v[1],
+                    v[2] );
       }
       break;
    default:
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
new file mode 100644
index 0000000000..48617a66ed
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -0,0 +1,1249 @@
+/**************************************************************************
+ *
+ * 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.
+ *
+ **************************************************************************/
+
+/**
+ * \brief  Primitive rasterization/rendering (points, lines, triangles)
+ *
+ * \author  Keith Whitwell <keith@tungstengraphics.com>
+ * \author  Brian Paul
+ */
+
+#include "sp_setup.h"
+
+#include "sp_context.h"
+#include "sp_headers.h"
+#include "sp_quad.h"
+#include "sp_state.h"
+#include "sp_prim_setup.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vertex.h"
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+
+#define DEBUG_VERTS 0
+#define DEBUG_FRAGS 0
+
+/**
+ * Triangle edge info
+ */
+struct edge {
+   float dx;		/**< X(v1) - X(v0), used only during setup */
+   float dy;		/**< Y(v1) - Y(v0), used only during setup */
+   float dxdy;		/**< dx/dy */
+   float sx, sy;	/**< first sample point coord */
+   int lines;		/**< number of lines on this edge */
+};
+
+
+/**
+ * Triangle setup info (derived from draw_stage).
+ * Also used for line drawing (taking some liberties).
+ */
+struct setup_context {
+   struct softpipe_context *softpipe;
+
+   /* Vertices are just an array of floats making up each attribute in
+    * turn.  Currently fixed at 4 floats, but should change in time.
+    * Codegen will help cope with this.
+    */
+   const float (*vmax)[4];
+   const float (*vmid)[4];
+   const float (*vmin)[4];
+   const float (*vprovoke)[4];
+
+   struct edge ebot;
+   struct edge etop;
+   struct edge emaj;
+
+   float oneoverarea;
+
+   struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS];
+   struct tgsi_interp_coef posCoef;  /* For Z, W */
+   struct quad_header quad;
+
+   struct {
+      int left[2];   /**< [0] = row0, [1] = row1 */
+      int right[2];
+      int y;
+      unsigned y_flags;
+      unsigned mask;     /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */
+   } span;
+
+#if DEBUG_FRAGS
+   uint numFragsEmitted;  /**< per primitive */
+   uint numFragsWritten;  /**< per primitive */
+#endif
+};
+
+
+
+
+
+/**
+ * Recalculate prim's determinant.
+ * XXX is this needed?
+ */
+static INLINE float
+calc_det( const float (*v0)[4],
+          const float (*v1)[4],
+          const float (*v2)[4] )
+{
+   /* edge vectors e = v0 - v2, f = v1 - v2 */
+   const float ex = v0[0][0] - v2[0][0];
+   const float ey = v0[0][1] - v2[0][1];
+   const float fx = v1[0][0] - v2[0][0];
+   const float fy = v1[0][1] - v2[0][1];
+
+   /* det = cross(e,f).z */
+   return ex * fy - ey * fx;
+}
+
+/**
+ * Clip setup->quad against the scissor/surface bounds.
+ */
+static INLINE void
+quad_clip(struct setup_context *setup)
+{
+   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
+   const int minx = (int) cliprect->minx;
+   const int maxx = (int) cliprect->maxx;
+   const int miny = (int) cliprect->miny;
+   const int maxy = (int) cliprect->maxy;
+
+   if (setup->quad.x0 >= maxx ||
+       setup->quad.y0 >= maxy ||
+       setup->quad.x0 + 1 < minx ||
+       setup->quad.y0 + 1 < miny) {
+      /* totally clipped */
+      setup->quad.mask = 0x0;
+      return;
+   }
+   if (setup->quad.x0 < minx)
+      setup->quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
+   if (setup->quad.y0 < miny)
+      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
+   if (setup->quad.x0 == maxx - 1)
+      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
+   if (setup->quad.y0 == maxy - 1)
+      setup->quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
+}
+
+
+/**
+ * Emit a quad (pass to next stage) with clipping.
+ */
+static INLINE void
+clip_emit_quad(struct setup_context *setup)
+{
+   quad_clip(setup);
+   if (setup->quad.mask) {
+      struct softpipe_context *sp = setup->softpipe;
+      sp->quad.first->run(sp->quad.first, &setup->quad);
+   }
+}
+
+
+/**
+ * Emit a quad (pass to next stage).  No clipping is done.
+ */
+static INLINE void
+emit_quad( struct setup_context *setup, int x, int y, unsigned mask )
+{
+   struct softpipe_context *sp = setup->softpipe;
+   setup->quad.x0 = x;
+   setup->quad.y0 = y;
+   setup->quad.mask = mask;
+#if DEBUG_FRAGS
+   if (mask & 1) setup->numFragsEmitted++;
+   if (mask & 2) setup->numFragsEmitted++;
+   if (mask & 4) setup->numFragsEmitted++;
+   if (mask & 8) setup->numFragsEmitted++;
+#endif
+   sp->quad.first->run(sp->quad.first, &setup->quad);
+#if DEBUG_FRAGS
+   mask = setup->quad.mask;
+   if (mask & 1) setup->numFragsWritten++;
+   if (mask & 2) setup->numFragsWritten++;
+   if (mask & 4) setup->numFragsWritten++;
+   if (mask & 8) setup->numFragsWritten++;
+#endif
+}
+
+
+/**
+ * Given an X or Y coordinate, return the block/quad coordinate that it
+ * belongs to.
+ */
+static INLINE int block( int x )
+{
+   return x & ~1;
+}
+
+
+/**
+ * Compute mask which indicates which pixels in the 2x2 quad are actually inside
+ * the triangle's bounds.
+ *
+ * this is pretty nasty...  may need to rework flush_spans again to
+ * fix it, if possible.
+ */
+static unsigned calculate_mask( struct setup_context *setup, int x )
+{
+   unsigned mask = 0x0;
+
+   if (x >= setup->span.left[0] && x < setup->span.right[0])
+      mask |= MASK_TOP_LEFT;
+
+   if (x >= setup->span.left[1] && x < setup->span.right[1])
+      mask |= MASK_BOTTOM_LEFT;
+
+   if (x+1 >= setup->span.left[0] && x+1 < setup->span.right[0])
+      mask |= MASK_TOP_RIGHT;
+
+   if (x+1 >= setup->span.left[1] && x+1 < setup->span.right[1])
+      mask |= MASK_BOTTOM_RIGHT;
+
+   return mask;
+}
+
+
+/**
+ * Render a horizontal span of quads
+ */
+static void flush_spans( struct setup_context *setup )
+{
+   int minleft, maxright;
+   int x;
+
+   switch (setup->span.y_flags) {
+   case 0x3:
+      /* both odd and even lines written (both quad rows) */
+      minleft = MIN2(setup->span.left[0], setup->span.left[1]);
+      maxright = MAX2(setup->span.right[0], setup->span.right[1]);
+      break;
+
+   case 0x1:
+      /* only even line written (quad top row) */
+      minleft = setup->span.left[0];
+      maxright = setup->span.right[0];
+      break;
+
+   case 0x2:
+      /* only odd line written (quad bottom row) */
+      minleft = setup->span.left[1];
+      maxright = setup->span.right[1];
+      break;
+
+   default:
+      return;
+   }
+
+   /* XXX this loop could be moved into the above switch cases and
+    * calculate_mask() could be simplified a bit...
+    */
+   for (x = block(minleft); x <= block(maxright); x += 2) {
+      emit_quad( setup, x, setup->span.y,
+                 calculate_mask( setup, x ) );
+   }
+
+   setup->span.y = 0;
+   setup->span.y_flags = 0;
+   setup->span.right[0] = 0;
+   setup->span.right[1] = 0;
+}
+
+#if DEBUG_VERTS
+static void print_vertex(const struct setup_context *setup,
+                         const float (*v)[4])
+{
+   int i;
+   debug_printf("Vertex: (%p)\n", v);
+   for (i = 0; i < setup->quad.nr_attrs; i++) {
+      debug_printf("  %d: %f %f %f %f\n",  i,
+              v[i][0], v[i][1], v[i][2], v[i][3]);
+   }
+}
+#endif
+
+static boolean setup_sort_vertices( struct setup_context *setup,
+                                    float det,
+                                    const float (*v0)[4],
+                                    const float (*v1)[4],
+                                    const float (*v2)[4] )
+{
+#if DEBUG_VERTS
+   debug_printf("Triangle:\n");
+   print_vertex(setup, v0);
+   print_vertex(setup, v1);
+   print_vertex(setup, v2);
+#endif
+
+   setup->vprovoke = v2;
+
+   /* determine bottom to top order of vertices */
+   {
+      float y0 = v0[0][1];
+      float y1 = v1[0][1];
+      float y2 = v2[0][1];
+      if (y0 <= y1) {
+	 if (y1 <= y2) {
+	    /* y0<=y1<=y2 */
+	    setup->vmin = v0;
+	    setup->vmid = v1;
+	    setup->vmax = v2;
+	 }
+	 else if (y2 <= y0) {
+	    /* y2<=y0<=y1 */
+	    setup->vmin = v2;
+	    setup->vmid = v0;
+	    setup->vmax = v1;
+	 }
+	 else {
+	    /* y0<=y2<=y1 */
+	    setup->vmin = v0;
+	    setup->vmid = v2;
+	    setup->vmax = v1;
+	 }
+      }
+      else {
+	 if (y0 <= y2) {
+	    /* y1<=y0<=y2 */
+	    setup->vmin = v1;
+	    setup->vmid = v0;
+	    setup->vmax = v2;
+	 }
+	 else if (y2 <= y1) {
+	    /* y2<=y1<=y0 */
+	    setup->vmin = v2;
+	    setup->vmid = v1;
+	    setup->vmax = v0;
+	 }
+	 else {
+	    /* y1<=y2<=y0 */
+	    setup->vmin = v1;
+	    setup->vmid = v2;
+	    setup->vmax = v0;
+	 }
+      }
+   }
+
+   setup->ebot.dx = setup->vmid[0][0] - setup->vmin[0][0];
+   setup->ebot.dy = setup->vmid[0][1] - setup->vmin[0][1];
+   setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0];
+   setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1];
+   setup->etop.dx = setup->vmax[0][0] - setup->vmid[0][0];
+   setup->etop.dy = setup->vmax[0][1] - setup->vmid[0][1];
+
+   /*
+    * Compute triangle's area.  Use 1/area to compute partial
+    * derivatives of attributes later.
+    *
+    * The area will be the same as prim->det, but the sign may be
+    * different depending on how the vertices get sorted above.
+    *
+    * To determine whether the primitive is front or back facing we
+    * use the prim->det value because its sign is correct.
+    */
+   {
+      const float area = (setup->emaj.dx * setup->ebot.dy -
+			    setup->ebot.dx * setup->emaj.dy);
+
+      setup->oneoverarea = 1.0f / area;
+      /*
+      debug_printf("%s one-over-area %f  area %f  det %f\n",
+                   __FUNCTION__, setup->oneoverarea, area, det );
+      */
+   }
+
+   /* We need to know if this is a front or back-facing triangle for:
+    *  - the GLSL gl_FrontFacing fragment attribute (bool)
+    *  - two-sided stencil test
+    */
+   setup->quad.facing = (det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
+
+   return TRUE;
+}
+
+
+/**
+ * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
+ * The value value comes from vertex[slot][i].
+ * The result will be put into setup->coef[slot].a0[i].
+ * \param slot  which attribute slot
+ * \param i  which component of the slot (0..3)
+ */
+static void const_coeff( struct setup_context *setup,
+                         struct tgsi_interp_coef *coef,
+                         uint vertSlot, uint i)
+{
+   assert(i <= 3);
+
+   coef->dadx[i] = 0;
+   coef->dady[i] = 0;
+
+   /* need provoking vertex info!
+    */
+   coef->a0[i] = setup->vprovoke[vertSlot][i];
+}
+
+
+/**
+ * Compute a0, dadx and dady for a linearly interpolated coefficient,
+ * for a triangle.
+ */
+static void tri_linear_coeff( struct setup_context *setup,
+                              struct tgsi_interp_coef *coef,
+                              uint vertSlot, uint i)
+{
+   float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i];
+   float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
+   float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
+   float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
+   float dadx = a * setup->oneoverarea;
+   float dady = b * setup->oneoverarea;
+
+   assert(i <= 3);
+
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+
+   /* calculate a0 as the value which would be sampled for the
+    * fragment at (0,0), taking into account that we want to sample at
+    * pixel centers, in other words (0.5, 0.5).
+    *
+    * this is neat but unfortunately not a good way to do things for
+    * triangles with very large values of dadx or dady as it will
+    * result in the subtraction and re-addition from a0 of a very
+    * large number, which means we'll end up loosing a lot of the
+    * fractional bits and precision from a0.  the way to fix this is
+    * to define a0 as the sample at a pixel center somewhere near vmin
+    * instead - i'll switch to this later.
+    */
+   coef->a0[i] = (setup->vmin[vertSlot][i] -
+                  (dadx * (setup->vmin[0][0] - 0.5f) +
+                   dady * (setup->vmin[0][1] - 0.5f)));
+
+   /*
+   debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
+		slot, "xyzw"[i],
+		setup->coef[slot].a0[i],
+		setup->coef[slot].dadx[i],
+		setup->coef[slot].dady[i]);
+   */
+}
+
+
+/**
+ * Compute a0, dadx and dady for a perspective-corrected interpolant,
+ * for a triangle.
+ * We basically multiply the vertex value by 1/w before computing
+ * the plane coefficients (a0, dadx, dady).
+ * Later, when we compute the value at a particular fragment position we'll
+ * divide the interpolated value by the interpolated W at that fragment.
+ */
+static void tri_persp_coeff( struct setup_context *setup,
+                             struct tgsi_interp_coef *coef,
+                             uint vertSlot, uint i)
+{
+   /* premultiply by 1/w  (v[0][3] is always W):
+    */
+   float mina = setup->vmin[vertSlot][i] * setup->vmin[0][3];
+   float mida = setup->vmid[vertSlot][i] * setup->vmid[0][3];
+   float maxa = setup->vmax[vertSlot][i] * setup->vmax[0][3];
+   float botda = mida - mina;
+   float majda = maxa - mina;
+   float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
+   float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
+   float dadx = a * setup->oneoverarea;
+   float dady = b * setup->oneoverarea;
+
+   /*
+   debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
+          	setup->vmin[vertSlot][i],
+          	setup->vmid[vertSlot][i],
+       		setup->vmax[vertSlot][i]
+          );
+   */
+   assert(i <= 3);
+
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+   coef->a0[i] = (mina -
+                  (dadx * (setup->vmin[0][0] - 0.5f) +
+                   dady * (setup->vmin[0][1] - 0.5f)));
+}
+
+
+/**
+ * Special coefficient setup for gl_FragCoord.
+ * X and Y are trivial, though Y has to be inverted for OpenGL.
+ * Z and W are copied from posCoef which should have already been computed.
+ * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
+ */
+static void
+setup_fragcoord_coeff(struct setup_context *setup, uint slot)
+{
+   /*X*/
+   setup->coef[slot].a0[0] = 0;
+   setup->coef[slot].dadx[0] = 1.0;
+   setup->coef[slot].dady[0] = 0.0;
+   /*Y*/
+   if (setup->softpipe->rasterizer->origin_lower_left) {
+      /* y=0=bottom */
+      const int winHeight = setup->softpipe->framebuffer.height;
+      setup->coef[slot].a0[1] = (float) (winHeight - 1);
+      setup->coef[slot].dady[1] = -1.0;
+   }
+   else {
+      /* y=0=top */
+      setup->coef[slot].a0[1] = 0.0;
+      setup->coef[slot].dady[1] = 1.0;
+   }
+   setup->coef[slot].dadx[1] = 0.0;
+   /*Z*/
+   setup->coef[slot].a0[2] = setup->posCoef.a0[2];
+   setup->coef[slot].dadx[2] = setup->posCoef.dadx[2];
+   setup->coef[slot].dady[2] = setup->posCoef.dady[2];
+   /*W*/
+   setup->coef[slot].a0[3] = setup->posCoef.a0[3];
+   setup->coef[slot].dadx[3] = setup->posCoef.dadx[3];
+   setup->coef[slot].dady[3] = setup->posCoef.dady[3];
+}
+
+
+
+/**
+ * Compute the setup->coef[] array dadx, dady, a0 values.
+ * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized.
+ */
+static void setup_tri_coefficients( struct setup_context *setup )
+{
+   struct softpipe_context *softpipe = setup->softpipe;
+   const struct sp_fragment_shader *spfs = softpipe->fs;
+   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
+   uint fragSlot;
+
+   /* z and w are done by linear interpolation:
+    */
+   tri_linear_coeff(setup, &setup->posCoef, 0, 2);
+   tri_linear_coeff(setup, &setup->posCoef, 0, 3);
+
+   /* setup interpolation for all the remaining attributes:
+    */
+   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
+      const uint vertSlot = vinfo->src_index[fragSlot];
+      uint j;
+
+      switch (vinfo->interp_mode[fragSlot]) {
+      case INTERP_CONSTANT:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_LINEAR:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            tri_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_PERSPECTIVE:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_POS:
+         setup_fragcoord_coeff(setup, fragSlot);
+         break;
+      default:
+         assert(0);
+      }
+
+      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
+         /* FOG.y = front/back facing  XXX fix this */
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].dadx[1] = 0.0;
+         setup->coef[fragSlot].dady[1] = 0.0;
+      }
+   }
+}
+
+
+
+static void setup_tri_edges( struct setup_context *setup )
+{
+   float vmin_x = setup->vmin[0][0] + 0.5f;
+   float vmid_x = setup->vmid[0][0] + 0.5f;
+
+   float vmin_y = setup->vmin[0][1] - 0.5f;
+   float vmid_y = setup->vmid[0][1] - 0.5f;
+   float vmax_y = setup->vmax[0][1] - 0.5f;
+
+   setup->emaj.sy = CEILF(vmin_y);
+   setup->emaj.lines = (int) CEILF(vmax_y - setup->emaj.sy);
+   setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy;
+   setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy;
+
+   setup->etop.sy = CEILF(vmid_y);
+   setup->etop.lines = (int) CEILF(vmax_y - setup->etop.sy);
+   setup->etop.dxdy = setup->etop.dx / setup->etop.dy;
+   setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy;
+
+   setup->ebot.sy = CEILF(vmin_y);
+   setup->ebot.lines = (int) CEILF(vmid_y - setup->ebot.sy);
+   setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy;
+   setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy;
+}
+
+
+/**
+ * Render the upper or lower half of a triangle.
+ * Scissoring/cliprect is applied here too.
+ */
+static void subtriangle( struct setup_context *setup,
+			 struct edge *eleft,
+			 struct edge *eright,
+			 unsigned lines )
+{
+   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
+   const int minx = (int) cliprect->minx;
+   const int maxx = (int) cliprect->maxx;
+   const int miny = (int) cliprect->miny;
+   const int maxy = (int) cliprect->maxy;
+   int y, start_y, finish_y;
+   int sy = (int)eleft->sy;
+
+   assert((int)eleft->sy == (int) eright->sy);
+
+   /* clip top/bottom */
+   start_y = sy;
+   finish_y = sy + lines;
+
+   if (start_y < miny)
+      start_y = miny;
+
+   if (finish_y > maxy)
+      finish_y = maxy;
+
+   start_y -= sy;
+   finish_y -= sy;
+
+   /*
+   debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);
+   */
+
+   for (y = start_y; y < finish_y; y++) {
+
+      /* avoid accumulating adds as floats don't have the precision to
+       * accurately iterate large triangle edges that way.  luckily we
+       * can just multiply these days.
+       *
+       * this is all drowned out by the attribute interpolation anyway.
+       */
+      int left = (int)(eleft->sx + y * eleft->dxdy);
+      int right = (int)(eright->sx + y * eright->dxdy);
+
+      /* clip left/right */
+      if (left < minx)
+         left = minx;
+      if (right > maxx)
+         right = maxx;
+
+      if (left < right) {
+         int _y = sy + y;
+         if (block(_y) != setup->span.y) {
+            flush_spans(setup);
+            setup->span.y = block(_y);
+         }
+
+         setup->span.left[_y&1] = left;
+         setup->span.right[_y&1] = right;
+         setup->span.y_flags |= 1<<(_y&1);
+      }
+   }
+
+
+   /* save the values so that emaj can be restarted:
+    */
+   eleft->sx += lines * eleft->dxdy;
+   eright->sx += lines * eright->dxdy;
+   eleft->sy += lines;
+   eright->sy += lines;
+}
+
+
+/**
+ * Do setup for triangle rasterization, then render the triangle.
+ */
+void setup_tri( struct setup_context *setup,
+                float det,
+                const float (*v0)[4],
+                const float (*v1)[4],
+                const float (*v2)[4] )
+{
+   /*
+   debug_printf("%s\n", __FUNCTION__ );
+   */
+
+#if DEBUG_FRAGS
+   setup->numFragsEmitted = 0;
+   setup->numFragsWritten = 0;
+#endif
+
+   setup_sort_vertices( setup, calc_det(v0, v1, v2),
+                        v0, v1, v2 );
+   setup_tri_coefficients( setup );
+   setup_tri_edges( setup );
+
+   setup->quad.prim = PRIM_TRI;
+
+   setup->span.y = 0;
+   setup->span.y_flags = 0;
+   setup->span.right[0] = 0;
+   setup->span.right[1] = 0;
+   /*   setup->span.z_mode = tri_z_mode( setup->ctx ); */
+
+   /*   init_constant_attribs( setup ); */
+
+   if (setup->oneoverarea < 0.0) {
+      /* emaj on left:
+       */
+      subtriangle( setup, &setup->emaj, &setup->ebot, setup->ebot.lines );
+      subtriangle( setup, &setup->emaj, &setup->etop, setup->etop.lines );
+   }
+   else {
+      /* emaj on right:
+       */
+      subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines );
+      subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines );
+   }
+
+   flush_spans( setup );
+
+#if DEBUG_FRAGS
+   printf("Tri: %u frags emitted, %u written\n",
+          setup->numFragsEmitted,
+          setup->numFragsWritten);
+#endif
+}
+
+
+
+/**
+ * Compute a0, dadx and dady for a linearly interpolated coefficient,
+ * for a line.
+ */
+static void
+line_linear_coeff(struct setup_context *setup,
+                  struct tgsi_interp_coef *coef,
+                  uint vertSlot, uint i)
+{
+   const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
+   const float dadx = da * setup->emaj.dx * setup->oneoverarea;
+   const float dady = da * setup->emaj.dy * setup->oneoverarea;
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+   coef->a0[i] = (setup->vmin[vertSlot][i] -
+                  (dadx * (setup->vmin[0][0] - 0.5f) +
+                   dady * (setup->vmin[0][1] - 0.5f)));
+}
+
+
+/**
+ * Compute a0, dadx and dady for a perspective-corrected interpolant,
+ * for a line.
+ */
+static void
+line_persp_coeff(struct setup_context *setup,
+                  struct tgsi_interp_coef *coef,
+                  uint vertSlot, uint i)
+{
+   /* XXX double-check/verify this arithmetic */
+   const float a0 = setup->vmin[vertSlot][i] * setup->vmin[0][3];
+   const float a1 = setup->vmax[vertSlot][i] * setup->vmax[0][3];
+   const float da = a1 - a0;
+   const float dadx = da * setup->emaj.dx * setup->oneoverarea;
+   const float dady = da * setup->emaj.dy * setup->oneoverarea;
+   coef->dadx[i] = dadx;
+   coef->dady[i] = dady;
+   coef->a0[i] = (setup->vmin[vertSlot][i] -
+                  (dadx * (setup->vmin[0][0] - 0.5f) +
+                   dady * (setup->vmin[0][1] - 0.5f)));
+}
+
+
+/**
+ * Compute the setup->coef[] array dadx, dady, a0 values.
+ * Must be called after setup->vmin,vmax are initialized.
+ */
+static INLINE void
+setup_line_coefficients(struct setup_context *setup,
+                        const float (*v0)[4],
+                        const float (*v1)[4])
+{
+   struct softpipe_context *softpipe = setup->softpipe;
+   const struct sp_fragment_shader *spfs = softpipe->fs;
+   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
+   uint fragSlot;
+
+   /* use setup->vmin, vmax to point to vertices */
+   setup->vprovoke = v1;
+   setup->vmin = v0;
+   setup->vmax = v1;
+
+   setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0];
+   setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1];
+   /* NOTE: this is not really 1/area */
+   setup->oneoverarea = 1.0f / (setup->emaj.dx * setup->emaj.dx +
+                                setup->emaj.dy * setup->emaj.dy);
+
+   /* z and w are done by linear interpolation:
+    */
+   line_linear_coeff(setup, &setup->posCoef, 0, 2);
+   line_linear_coeff(setup, &setup->posCoef, 0, 3);
+
+   /* setup interpolation for all the remaining attributes:
+    */
+   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
+      const uint vertSlot = vinfo->src_index[fragSlot];
+      uint j;
+
+      switch (vinfo->interp_mode[fragSlot]) {
+      case INTERP_CONSTANT:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_LINEAR:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            line_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_PERSPECTIVE:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_POS:
+         setup_fragcoord_coeff(setup, fragSlot);
+         break;
+      default:
+         assert(0);
+      }
+
+      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
+         /* FOG.y = front/back facing  XXX fix this */
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].dadx[1] = 0.0;
+         setup->coef[fragSlot].dady[1] = 0.0;
+      }
+   }
+}
+
+
+/**
+ * Plot a pixel in a line segment.
+ */
+static INLINE void
+plot(struct setup_context *setup, int x, int y)
+{
+   const int iy = y & 1;
+   const int ix = x & 1;
+   const int quadX = x - ix;
+   const int quadY = y - iy;
+   const int mask = (1 << ix) << (2 * iy);
+
+   if (quadX != setup->quad.x0 ||
+       quadY != setup->quad.y0)
+   {
+      /* flush prev quad, start new quad */
+
+      if (setup->quad.x0 != -1)
+         clip_emit_quad(setup);
+
+      setup->quad.x0 = quadX;
+      setup->quad.y0 = quadY;
+      setup->quad.mask = 0x0;
+   }
+
+   setup->quad.mask |= mask;
+}
+
+
+/**
+ * Do setup for line rasterization, then render the line.
+ * Single-pixel width, no stipple, etc.  We rely on the 'draw' module
+ * to handle stippling and wide lines.
+ */
+void
+setup_line(struct setup_context *setup,
+           const float (*v0)[4],
+           const float (*v1)[4])
+{
+   int x0 = (int) v0[0][0];
+   int x1 = (int) v1[0][0];
+   int y0 = (int) v0[0][1];
+   int y1 = (int) v1[0][1];
+   int dx = x1 - x0;
+   int dy = y1 - y0;
+   int xstep, ystep;
+
+   if (dx == 0 && dy == 0)
+      return;
+
+   setup_line_coefficients(setup, v0, v1);
+
+   if (dx < 0) {
+      dx = -dx;   /* make positive */
+      xstep = -1;
+   }
+   else {
+      xstep = 1;
+   }
+
+   if (dy < 0) {
+      dy = -dy;   /* make positive */
+      ystep = -1;
+   }
+   else {
+      ystep = 1;
+   }
+
+   assert(dx >= 0);
+   assert(dy >= 0);
+
+   setup->quad.x0 = setup->quad.y0 = -1;
+   setup->quad.mask = 0x0;
+   setup->quad.prim = PRIM_LINE;
+   /* XXX temporary: set coverage to 1.0 so the line appears
+    * if AA mode happens to be enabled.
+    */
+   setup->quad.coverage[0] =
+   setup->quad.coverage[1] =
+   setup->quad.coverage[2] =
+   setup->quad.coverage[3] = 1.0;
+
+   if (dx > dy) {
+      /*** X-major line ***/
+      int i;
+      const int errorInc = dy + dy;
+      int error = errorInc - dx;
+      const int errorDec = error - dx;
+
+      for (i = 0; i < dx; i++) {
+         plot(setup, x0, y0);
+
+         x0 += xstep;
+         if (error < 0) {
+            error += errorInc;
+         }
+         else {
+            error += errorDec;
+            y0 += ystep;
+         }
+      }
+   }
+   else {
+      /*** Y-major line ***/
+      int i;
+      const int errorInc = dx + dx;
+      int error = errorInc - dy;
+      const int errorDec = error - dy;
+
+      for (i = 0; i < dy; i++) {
+         plot(setup, x0, y0);
+
+         y0 += ystep;
+         if (error < 0) {
+            error += errorInc;
+         }
+         else {
+            error += errorDec;
+            x0 += xstep;
+         }
+      }
+   }
+
+   /* draw final quad */
+   if (setup->quad.mask) {
+      clip_emit_quad(setup);
+   }
+}
+
+
+static void
+point_persp_coeff(struct setup_context *setup,
+                  const float (*vert)[4],
+                  struct tgsi_interp_coef *coef,
+                  uint vertSlot, uint i)
+{
+   assert(i <= 3);
+   coef->dadx[i] = 0.0F;
+   coef->dady[i] = 0.0F;
+   coef->a0[i] = vert[vertSlot][i] * vert[0][3];
+}
+
+
+/**
+ * Do setup for point rasterization, then render the point.
+ * Round or square points...
+ * XXX could optimize a lot for 1-pixel points.
+ */
+void
+setup_point( struct setup_context *setup,
+             const float (*v0)[4] )
+{
+   struct softpipe_context *softpipe = setup->softpipe;
+   const struct sp_fragment_shader *spfs = softpipe->fs;
+   const int sizeAttr = setup->softpipe->psize_slot;
+   const float size
+      = sizeAttr > 0 ? v0[sizeAttr][0]
+      : setup->softpipe->rasterizer->point_size;
+   const float halfSize = 0.5F * size;
+   const boolean round = (boolean) setup->softpipe->rasterizer->point_smooth;
+   const float x = v0[0][0];  /* Note: data[0] is always position */
+   const float y = v0[0][1];
+   const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
+   uint fragSlot;
+
+   /* For points, all interpolants are constant-valued.
+    * However, for point sprites, we'll need to setup texcoords appropriately.
+    * XXX: which coefficients are the texcoords???
+    * We may do point sprites as textured quads...
+    *
+    * KW: We don't know which coefficients are texcoords - ultimately
+    * the choice of what interpolation mode to use for each attribute
+    * should be determined by the fragment program, using
+    * per-attribute declaration statements that include interpolation
+    * mode as a parameter.  So either the fragment program will have
+    * to be adjusted for pointsprite vs normal point behaviour, or
+    * otherwise a special interpolation mode will have to be defined
+    * which matches the required behaviour for point sprites.  But -
+    * the latter is not a feature of normal hardware, and as such
+    * probably should be ruled out on that basis.
+    */
+   setup->vprovoke = v0;
+
+   /* setup Z, W */
+   const_coeff(setup, &setup->posCoef, 0, 2);
+   const_coeff(setup, &setup->posCoef, 0, 3);
+
+   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
+      const uint vertSlot = vinfo->src_index[fragSlot];
+      uint j;
+
+      switch (vinfo->interp_mode[fragSlot]) {
+      case INTERP_CONSTANT:
+         /* fall-through */
+      case INTERP_LINEAR:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_PERSPECTIVE:
+         for (j = 0; j < NUM_CHANNELS; j++)
+            point_persp_coeff(setup, setup->vprovoke,
+                              &setup->coef[fragSlot], vertSlot, j);
+         break;
+      case INTERP_POS:
+         setup_fragcoord_coeff(setup, fragSlot);
+         break;
+      default:
+         assert(0);
+      }
+
+      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
+         /* FOG.y = front/back facing  XXX fix this */
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].dadx[1] = 0.0;
+         setup->coef[fragSlot].dady[1] = 0.0;
+      }
+   }
+
+   setup->quad.prim = PRIM_POINT;
+
+   if (halfSize <= 0.5 && !round) {
+      /* special case for 1-pixel points */
+      const int ix = ((int) x) & 1;
+      const int iy = ((int) y) & 1;
+      setup->quad.x0 = (int) x - ix;
+      setup->quad.y0 = (int) y - iy;
+      setup->quad.mask = (1 << ix) << (2 * iy);
+      clip_emit_quad(setup);
+   }
+   else {
+      if (round) {
+         /* rounded points */
+         const int ixmin = block((int) (x - halfSize));
+         const int ixmax = block((int) (x + halfSize));
+         const int iymin = block((int) (y - halfSize));
+         const int iymax = block((int) (y + halfSize));
+         const float rmin = halfSize - 0.7071F;  /* 0.7071 = sqrt(2)/2 */
+         const float rmax = halfSize + 0.7071F;
+         const float rmin2 = MAX2(0.0F, rmin * rmin);
+         const float rmax2 = rmax * rmax;
+         const float cscale = 1.0F / (rmax2 - rmin2);
+         int ix, iy;
+
+         for (iy = iymin; iy <= iymax; iy += 2) {
+            for (ix = ixmin; ix <= ixmax; ix += 2) {
+               float dx, dy, dist2, cover;
+
+               setup->quad.mask = 0x0;
+
+               dx = (ix + 0.5f) - x;
+               dy = (iy + 0.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_TOP_LEFT;
+               }
+
+               dx = (ix + 1.5f) - x;
+               dy = (iy + 0.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_TOP_RIGHT;
+               }
+
+               dx = (ix + 0.5f) - x;
+               dy = (iy + 1.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_BOTTOM_LEFT;
+               }
+
+               dx = (ix + 1.5f) - x;
+               dy = (iy + 1.5f) - y;
+               dist2 = dx * dx + dy * dy;
+               if (dist2 <= rmax2) {
+                  cover = 1.0F - (dist2 - rmin2) * cscale;
+                  setup->quad.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad.mask |= MASK_BOTTOM_RIGHT;
+               }
+
+               if (setup->quad.mask) {
+                  setup->quad.x0 = ix;
+                  setup->quad.y0 = iy;
+                  clip_emit_quad(setup);
+               }
+            }
+         }
+      }
+      else {
+         /* square points */
+         const int xmin = (int) (x + 0.75 - halfSize);
+         const int ymin = (int) (y + 0.25 - halfSize);
+         const int xmax = xmin + (int) size;
+         const int ymax = ymin + (int) size;
+         /* XXX could apply scissor to xmin,ymin,xmax,ymax now */
+         const int ixmin = block(xmin);
+         const int ixmax = block(xmax - 1);
+         const int iymin = block(ymin);
+         const int iymax = block(ymax - 1);
+         int ix, iy;
+
+         /*
+         debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax);
+         */
+         for (iy = iymin; iy <= iymax; iy += 2) {
+            uint rowMask = 0xf;
+            if (iy < ymin) {
+               /* above the top edge */
+               rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
+            }
+            if (iy + 1 >= ymax) {
+               /* below the bottom edge */
+               rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
+            }
+
+            for (ix = ixmin; ix <= ixmax; ix += 2) {
+               uint mask = rowMask;
+
+               if (ix < xmin) {
+                  /* fragment is past left edge of point, turn off left bits */
+                  mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
+               }
+               if (ix + 1 >= xmax) {
+                  /* past the right edge */
+                  mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
+               }
+
+               setup->quad.mask = mask;
+               setup->quad.x0 = ix;
+               setup->quad.y0 = iy;
+               clip_emit_quad(setup);
+            }
+         }
+      }
+   }
+}
+
+void setup_prepare( struct setup_context *setup )
+{
+   struct softpipe_context *sp = setup->softpipe;
+   unsigned i;
+
+   if (sp->dirty) {
+      softpipe_update_derived(sp);
+   }
+
+   /* Mark surfaces as defined now */
+   for (i = 0; i < sp->framebuffer.num_cbufs; i++){
+      if (sp->framebuffer.cbufs[i]) {
+         sp->framebuffer.cbufs[i]->status = PIPE_SURFACE_STATUS_DEFINED;
+      }
+   }
+   if (sp->framebuffer.zsbuf) {
+      sp->framebuffer.zsbuf->status = PIPE_SURFACE_STATUS_DEFINED;
+   }
+
+   {
+      const struct sp_fragment_shader *fs = setup->softpipe->fs;
+      setup->quad.nr_attrs = fs->info.num_inputs;
+      sp->quad.first->begin(sp->quad.first);
+   }
+}
+
+
+
+void setup_destroy_context( struct setup_context *setup )
+{
+   FREE( setup );
+}
+
+
+/**
+ * Create a new primitive setup/render stage.
+ */
+struct setup_context *setup_create_context( struct softpipe_context *softpipe )
+{
+   struct setup_context *setup = CALLOC_STRUCT(setup_context);
+
+   setup->softpipe = softpipe;
+
+   setup->quad.coef = setup->coef;
+   setup->quad.posCoef = &setup->posCoef;
+
+   return setup;
+}
diff --git a/src/gallium/drivers/softpipe/sp_setup.h b/src/gallium/drivers/softpipe/sp_setup.h
new file mode 100644
index 0000000000..3133fc2a3d
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_setup.h
@@ -0,0 +1,53 @@
+/**************************************************************************
+ *
+ * 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 SP_SETUP_H
+#define SP_SETUP_H
+
+struct setup_context;
+struct softpipe_context;
+
+void setup_tri( struct setup_context *setup,
+                float det,
+                const float (*v0)[4],
+                const float (*v1)[4],
+                const float (*v2)[4] );
+
+void
+setup_line(struct setup_context *setup,
+           const float (*v0)[4],
+           const float (*v1)[4]);
+
+void
+setup_point( struct setup_context *setup,
+             const float (*v0)[4] );
+
+
+struct setup_context *setup_create_context( struct softpipe_context *softpipe );
+void setup_prepare( struct setup_context *setup );
+void setup_destroy_context( struct setup_context *setup );
+
+#endif
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
index 14abb20eeb..f10a1fa471 100644
--- a/src/gallium/drivers/softpipe/sp_state_derived.c
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -71,25 +71,16 @@ softpipe_get_vertex_info(struct softpipe_context *softpipe)
           * simply emit the whole post-xform vertex as-is:
           */
          struct vertex_info *vinfo_vbuf = &softpipe->vertex_info_vbuf;
-#if 0
-         vinfo_vbuf->num_attribs = 0;
-         /* special-case to allow memcpy of whole vertex */
-         draw_emit_vertex_attr(vinfo_vbuf, EMIT_ALL, INTERP_NONE, 0);
-         /* size in dwords or floats */
-         vinfo_vbuf->size = 4 * draw_num_vs_outputs(softpipe->draw)
-                          + sizeof(struct vertex_header) / 4;
-#else
-         /* for pass-through mode, we need a more explicit list of attribs */
          const uint num = draw_num_vs_outputs(softpipe->draw);
          uint i;
 
+         /* No longer any need to try and emit draw vertex_header info.
+          */
          vinfo_vbuf->num_attribs = 0;
-         draw_emit_vertex_attr(vinfo_vbuf, EMIT_HEADER, INTERP_NONE, 0);
          for (i = 0; i < num; i++) {
             draw_emit_vertex_attr(vinfo_vbuf, EMIT_4F, INTERP_PERSPECTIVE, i);
          }
          draw_compute_vertex_size(vinfo_vbuf);
-#endif
       }
 
       /*