X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fsoftpipe%2Fsp_setup.c;h=a91e4f588c8e5f83e5c016293f58d22ed1ecedfa;hb=8c9b9aac7d09e65195dca6681d59c10e4ef713d9;hp=de3ae3c36962b6666eb9631cd8536b948ba4b740;hpb=862488075c5537b0613753b0d14c267527fc6199;p=mesa.git

diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index de3ae3c3696..a91e4f588c8 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -18,7 +18,7 @@
  * 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
+ * IN NO EVENT SHALL VMWARE 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.
@@ -28,21 +28,17 @@
 /**
  * \brief  Primitive rasterization/rendering (points, lines, triangles)
  *
- * \author  Keith Whitwell <keith@tungstengraphics.com>
+ * \author  Keith Whitwell <keithw@vmware.com>
  * \author  Brian Paul
  */
 
 #include "sp_context.h"
-#include "sp_prim_setup.h"
 #include "sp_quad.h"
 #include "sp_quad_pipe.h"
 #include "sp_setup.h"
 #include "sp_state.h"
 #include "draw/draw_context.h"
-#include "draw/draw_private.h"
-#include "draw/draw_vertex.h"
 #include "pipe/p_shader_tokens.h"
-#include "pipe/p_thread.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 
@@ -50,6 +46,7 @@
 #define DEBUG_VERTS 0
 #define DEBUG_FRAGS 0
 
+
 /**
  * Triangle edge info
  */
@@ -61,90 +58,17 @@ struct edge {
    int lines;		/**< number of lines on this edge */
 };
 
-#if SP_NUM_QUAD_THREADS > 1
 
-/* Set to 1 if you want other threads to be instantly
- * notified of pending jobs.
+/**
+ * Max number of quads (2x2 pixel blocks) to process per batch.
+ * This can't be arbitrarily increased since we depend on some 32-bit
+ * bitmasks (two bits per quad).
  */
-#define INSTANT_NOTEMPTY_NOTIFY 0
-
-struct thread_info
-{
-   struct setup_context *setup;
-   uint id;
-   pipe_thread handle;
-};
-
-struct quad_job;
+#define MAX_QUADS 16
 
-typedef void (* quad_job_routine)( struct setup_context *setup, uint thread, struct quad_job *job );
-
-struct quad_job
-{
-   struct quad_header_input input;
-   struct quad_header_inout inout;
-   quad_job_routine routine;
-};
-
-#define NUM_QUAD_JOBS 64
-
-struct quad_job_que
-{
-   struct quad_job jobs[NUM_QUAD_JOBS];
-   uint first;
-   uint last;
-   pipe_mutex que_mutex;
-   pipe_condvar que_notfull_condvar;
-   pipe_condvar que_notempty_condvar;
-   uint jobs_added;
-   uint jobs_done;
-   pipe_condvar que_done_condvar;
-};
-
-static void
-add_quad_job( struct quad_job_que *que, struct quad_header *quad, quad_job_routine routine )
-{
-#if INSTANT_NOTEMPTY_NOTIFY
-   boolean empty;
-#endif
-
-   /* Wait for empty slot, see if the que is empty.
-    */
-   pipe_mutex_lock( que->que_mutex );
-   while ((que->last + 1) % NUM_QUAD_JOBS == que->first) {
-#if !INSTANT_NOTEMPTY_NOTIFY
-      pipe_condvar_broadcast( que->que_notempty_condvar );
-#endif
-      pipe_condvar_wait( que->que_notfull_condvar, que->que_mutex );
-   }
-#if INSTANT_NOTEMPTY_NOTIFY
-   empty = que->last == que->first;
-#endif
-   que->jobs_added++;
-   pipe_mutex_unlock( que->que_mutex );
-
-   /* Submit new job.
-    */
-   que->jobs[que->last].input = quad->input;
-   que->jobs[que->last].inout = quad->inout;
-   que->jobs[que->last].routine = routine;
-   que->last = (que->last + 1) % NUM_QUAD_JOBS;
-
-#if INSTANT_NOTEMPTY_NOTIFY
-   /* If the que was empty, notify consumers there's a job to be done.
-    */
-   if (empty) {
-      pipe_mutex_lock( que->que_mutex );
-      pipe_condvar_broadcast( que->que_notempty_condvar );
-      pipe_mutex_unlock( que->que_mutex );
-   }
-#endif
-}
-
-#endif
 
 /**
- * Triangle setup info (derived from draw_stage).
+ * Triangle setup info.
  * Also used for line drawing (taking some liberties).
  */
 struct setup_context {
@@ -164,22 +88,22 @@ struct setup_context {
    struct edge emaj;
 
    float oneoverarea;
+   int facing;
+
+   float pixel_offset;
+   unsigned max_layer;
+
+   struct quad_header quad[MAX_QUADS];
+   struct quad_header *quad_ptrs[MAX_QUADS];
+   unsigned count;
 
    struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS];
    struct tgsi_interp_coef posCoef;  /* For Z, W */
-   struct quad_header quad;
-
-#if SP_NUM_QUAD_THREADS > 1
-   struct quad_job_que que;
-   struct thread_info threads[SP_NUM_QUAD_THREADS];
-#endif
 
    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
@@ -187,104 +111,24 @@ struct setup_context {
    uint numFragsWritten;  /**< per primitive */
 #endif
 
-   unsigned winding;		/* which winding to cull */
+   unsigned cull_face;		/* which faces cull */
+   unsigned nr_vertex_attrs;
 };
 
-#if SP_NUM_QUAD_THREADS > 1
-
-static PIPE_THREAD_ROUTINE( quad_thread, param )
-{
-   struct thread_info *info = (struct thread_info *) param;
-   struct quad_job_que *que = &info->setup->que;
-
-   for (;;) {
-      struct quad_job job;
-      boolean full;
-
-      /* Wait for an available job.
-       */
-      pipe_mutex_lock( que->que_mutex );
-      while (que->last == que->first)
-         pipe_condvar_wait( que->que_notempty_condvar, que->que_mutex );
-
-      /* See if the que is full.
-       */
-      full = (que->last + 1) % NUM_QUAD_JOBS == que->first;
-
-      /* Take a job and remove it from que.
-       */
-      job = que->jobs[que->first];
-      que->first = (que->first + 1) % NUM_QUAD_JOBS;
-
-      /* Notify the producer if the que is not full.
-       */
-      if (full)
-         pipe_condvar_signal( que->que_notfull_condvar );
-      pipe_mutex_unlock( que->que_mutex );
-
-      job.routine( info->setup, info->id, &job );
-
-      /* Notify the producer if that's the last finished job.
-       */
-      pipe_mutex_lock( que->que_mutex );
-      que->jobs_done++;
-      if (que->jobs_added == que->jobs_done)
-         pipe_condvar_signal( que->que_done_condvar );
-      pipe_mutex_unlock( que->que_mutex );
-   }
-
-   return NULL;
-}
-
-#define WAIT_FOR_COMPLETION(setup) \
-   do {\
-      pipe_mutex_lock( setup->que.que_mutex );\
-      if (!INSTANT_NOTEMPTY_NOTIFY)\
-         pipe_condvar_broadcast( setup->que.que_notempty_condvar );\
-      while (setup->que.jobs_added != setup->que.jobs_done)\
-         pipe_condvar_wait( setup->que.que_done_condvar, setup->que.que_mutex );\
-      pipe_mutex_unlock( setup->que.que_mutex );\
-   } while (0)
-
-#else
-
-#define WAIT_FOR_COMPLETION(setup) ((void) 0)
-
-#endif
-
 
 
-/**
- * Do triangle cull test using tri determinant (sign indicates orientation)
- * \return true if triangle is to be culled.
- */
-static INLINE boolean
-cull_tri(const struct setup_context *setup, float det)
-{
-   if (det != 0) {   
-      /* if (det < 0 then Z points toward camera and triangle is 
-       * counter-clockwise winding.
-       */
-      unsigned winding = (det < 0) ? PIPE_WINDING_CCW : PIPE_WINDING_CW;
 
-      if ((winding & setup->winding) == 0)
-	 return FALSE;
-   }
-
-   /* Culled:
-    */
-   return TRUE;
-}
 
 
 
 /**
  * Clip setup->quad against the scissor/surface bounds.
  */
-static INLINE void
-quad_clip( struct setup_context *setup, struct quad_header *quad )
+static inline void
+quad_clip(struct setup_context *setup, struct quad_header *quad)
 {
-   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
+   unsigned viewport_index = quad[0].input.viewport_index;
+   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect[viewport_index];
    const int minx = (int) cliprect->minx;
    const int maxx = (int) cliprect->maxx;
    const int miny = (int) cliprect->miny;
@@ -312,191 +156,119 @@ quad_clip( struct setup_context *setup, struct quad_header *quad )
 /**
  * Emit a quad (pass to next stage) with clipping.
  */
-static INLINE void
-clip_emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread )
+static inline void
+clip_emit_quad(struct setup_context *setup, struct quad_header *quad)
 {
-   quad_clip( setup, quad );
+   quad_clip(setup, quad);
+
    if (quad->inout.mask) {
       struct softpipe_context *sp = setup->softpipe;
 
-      sp->quad[thread].first->run( sp->quad[thread].first, quad );
-   }
-}
-
-#if SP_NUM_QUAD_THREADS > 1
-
-static void
-clip_emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
-{
-   struct quad_header quad;
-
-   quad.input = job->input;
-   quad.inout = job->inout;
-   quad.coef = setup->quad.coef;
-   quad.posCoef = setup->quad.posCoef;
-   quad.nr_attrs = setup->quad.nr_attrs;
-   clip_emit_quad( setup, &quad, thread );
-}
-
-#define CLIP_EMIT_QUAD(setup) add_quad_job( &setup->que, &setup->quad, clip_emit_quad_job )
-
-#else
-
-#define CLIP_EMIT_QUAD(setup) clip_emit_quad( setup, &setup->quad, 0 )
-
-#endif
-
-/**
- * Emit a quad (pass to next stage).  No clipping is done.
- */
-static INLINE void
-emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread )
-{
-   struct softpipe_context *sp = setup->softpipe;
 #if DEBUG_FRAGS
-   uint mask = quad->inout.mask;
+      setup->numFragsEmitted += util_bitcount(quad->inout.mask);
 #endif
 
-#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[thread].first->run( sp->quad[thread].first, quad );
-#if DEBUG_FRAGS
-   mask = quad->inout.mask;
-   if (mask & 1) setup->numFragsWritten++;
-   if (mask & 2) setup->numFragsWritten++;
-   if (mask & 4) setup->numFragsWritten++;
-   if (mask & 8) setup->numFragsWritten++;
-#endif
-}
-
-#if SP_NUM_QUAD_THREADS > 1
-
-static void
-emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
-{
-   struct quad_header quad;
-
-   quad.input = job->input;
-   quad.inout = job->inout;
-   quad.coef = setup->quad.coef;
-   quad.posCoef = setup->quad.posCoef;
-   quad.nr_attrs = setup->quad.nr_attrs;
-   emit_quad( setup, &quad, thread );
+      sp->quad.first->run( sp->quad.first, &quad, 1 );
+   }
 }
 
-#define EMIT_QUAD(setup,x,y,mask) do {\
-      setup->quad.input.x0 = x;\
-      setup->quad.input.y0 = y;\
-      setup->quad.inout.mask = mask;\
-      add_quad_job( &setup->que, &setup->quad, emit_quad_job );\
-   } while (0)
-
-#else
 
-#define EMIT_QUAD(setup,x,y,mask) do {\
-      setup->quad.input.x0 = x;\
-      setup->quad.input.y0 = y;\
-      setup->quad.inout.mask = mask;\
-      emit_quad( setup, &setup->quad, 0 );\
-   } while (0)
-
-#endif
 
 /**
  * Given an X or Y coordinate, return the block/quad coordinate that it
  * belongs to.
  */
-static INLINE int block( int x )
+static inline int
+block(int x)
+{
+   return x & ~(2-1);
+}
+
+
+static inline int
+block_x(int x)
 {
-   return x & ~1;
+   return x & ~(16-1);
 }
 
 
 /**
  * Render a horizontal span of quads
  */
-static void flush_spans( struct setup_context *setup )
+static void
+flush_spans(struct setup_context *setup)
 {
+   const int step = MAX_QUADS;
    const int xleft0 = setup->span.left[0];
    const int xleft1 = setup->span.left[1];
    const int xright0 = setup->span.right[0];
    const int xright1 = setup->span.right[1];
-   int minleft, maxright;
+   struct quad_stage *pipe = setup->softpipe->quad.first;
+
+   const int minleft = block_x(MIN2(xleft0, xleft1));
+   const int maxright = MAX2(xright0, xright1);
    int x;
 
-   switch (setup->span.y_flags) {
-   case 0x3:
-      /* both odd and even lines written (both quad rows) */
-      minleft = block(MIN2(xleft0, xleft1));
-      maxright = block(MAX2(xright0, xright1));
-      for (x = minleft; x <= maxright; x += 2) {
-         /* determine which of the four pixels is inside the span bounds */
-         uint mask = 0x0;
-         if (x >= xleft0 && x < xright0)
-            mask |= MASK_TOP_LEFT;
-         if (x >= xleft1 && x < xright1)
-            mask |= MASK_BOTTOM_LEFT;
-         if (x+1 >= xleft0 && x+1 < xright0)
-            mask |= MASK_TOP_RIGHT;
-         if (x+1 >= xleft1 && x+1 < xright1)
-            mask |= MASK_BOTTOM_RIGHT;
-         if (mask)
-            EMIT_QUAD( setup, x, setup->span.y, mask );
-      }
-      break;
-
-   case 0x1:
-      /* only even line written (quad top row) */
-      minleft = block(xleft0);
-      maxright = block(xright0);
-      for (x = minleft; x <= maxright; x += 2) {
-         uint mask = 0x0;
-         if (x >= xleft0 && x < xright0)
-            mask |= MASK_TOP_LEFT;
-         if (x+1 >= xleft0 && x+1 < xright0)
-            mask |= MASK_TOP_RIGHT;
-         if (mask)
-            EMIT_QUAD( setup, x, setup->span.y, mask );
-      }
-      break;
-
-   case 0x2:
-      /* only odd line written (quad bottom row) */
-      minleft = block(xleft1);
-      maxright = block(xright1);
-      for (x = minleft; x <= maxright; x += 2) {
-         uint mask = 0x0;
-         if (x >= xleft1 && x < xright1)
-            mask |= MASK_BOTTOM_LEFT;
-         if (x+1 >= xleft1 && x+1 < xright1)
-            mask |= MASK_BOTTOM_RIGHT;
-         if (mask)
-            EMIT_QUAD( setup, x, setup->span.y, mask );
-      }
-      break;
+   /* process quads in horizontal chunks of 16 */
+   for (x = minleft; x < maxright; x += step) {
+      unsigned skip_left0 = CLAMP(xleft0 - x, 0, step);
+      unsigned skip_left1 = CLAMP(xleft1 - x, 0, step);
+      unsigned skip_right0 = CLAMP(x + step - xright0, 0, step);
+      unsigned skip_right1 = CLAMP(x + step - xright1, 0, step);
+      unsigned lx = x;
+      unsigned q = 0;
 
-   default:
-      return;
+      unsigned skipmask_left0 = (1U << skip_left0) - 1U;
+      unsigned skipmask_left1 = (1U << skip_left1) - 1U;
+
+      /* These calculations fail when step == 32 and skip_right == 0.
+       */
+      unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0);
+      unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1);
+
+      unsigned mask0 = ~skipmask_left0 & ~skipmask_right0;
+      unsigned mask1 = ~skipmask_left1 & ~skipmask_right1;
+
+      if (mask0 | mask1) {
+         do {
+            unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2);
+            if (quadmask) {
+               setup->quad[q].input.x0 = lx;
+               setup->quad[q].input.y0 = setup->span.y;
+               setup->quad[q].input.facing = setup->facing;
+               setup->quad[q].inout.mask = quadmask;
+               setup->quad_ptrs[q] = &setup->quad[q];
+               q++;
+#if DEBUG_FRAGS
+               setup->numFragsEmitted += util_bitcount(quadmask);
+#endif
+            }
+            mask0 >>= 2;
+            mask1 >>= 2;
+            lx += 2;
+         } while (mask0 | mask1);
+
+         pipe->run( pipe, setup->quad_ptrs, q );
+      }
    }
 
+
    setup->span.y = 0;
-   setup->span.y_flags = 0;
    setup->span.right[0] = 0;
    setup->span.right[1] = 0;
+   setup->span.left[0] = 1000000;     /* greater than right[0] */
+   setup->span.left[1] = 1000000;     /* greater than right[1] */
 }
 
 
 #if DEBUG_VERTS
-static void print_vertex(const struct setup_context *setup,
-                         const float (*v)[4])
+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("   Vertex: (%p)\n", (void *) v);
+   for (i = 0; i < setup->nr_vertex_attrs; i++) {
       debug_printf("     %d: %f %f %f %f\n",  i,
               v[i][0], v[i][1], v[i][2], v[i][3]);
       if (util_is_inf_or_nan(v[i][0])) {
@@ -506,18 +278,23 @@ static void print_vertex(const struct setup_context *setup,
 }
 #endif
 
+
 /**
  * Sort the vertices from top to bottom order, setting up the triangle
  * edge fields (ebot, emaj, etop).
  * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise
  */
-static boolean setup_sort_vertices( struct setup_context *setup,
-                                    float det,
-                                    const float (*v0)[4],
-                                    const float (*v1)[4],
-                                    const float (*v2)[4] )
+static boolean
+setup_sort_vertices(struct setup_context *setup,
+                    float det,
+                    const float (*v0)[4],
+                    const float (*v1)[4],
+                    const float (*v2)[4])
 {
-   setup->vprovoke = v2;
+   if (setup->softpipe->rasterizer->flatshade_first)
+      setup->vprovoke = v0;
+   else
+      setup->vprovoke = v2;
 
    /* determine bottom to top order of vertices */
    {
@@ -600,13 +377,69 @@ static boolean setup_sort_vertices( struct setup_context *setup,
    /* 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
+    * 0 = front-facing, 1 = back-facing
     */
-   setup->quad.input.facing = (det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
+   setup->facing = 
+      ((det < 0.0) ^ 
+       (setup->softpipe->rasterizer->front_ccw));
+
+   {
+      unsigned face = setup->facing == 0 ? PIPE_FACE_FRONT : PIPE_FACE_BACK;
+
+      if (face & setup->cull_face)
+	 return FALSE;
+   }
 
    return TRUE;
 }
 
 
+/* Apply cylindrical wrapping to v0, v1, v2 coordinates, if enabled.
+ * Input coordinates must be in [0, 1] range, otherwise results are undefined.
+ * Some combinations of coordinates produce invalid results,
+ * but this behaviour is acceptable.
+ */
+static void
+tri_apply_cylindrical_wrap(float v0,
+                           float v1,
+                           float v2,
+                           uint cylindrical_wrap,
+                           float output[3])
+{
+   if (cylindrical_wrap) {
+      float delta;
+
+      delta = v1 - v0;
+      if (delta > 0.5f) {
+         v0 += 1.0f;
+      }
+      else if (delta < -0.5f) {
+         v1 += 1.0f;
+      }
+
+      delta = v2 - v1;
+      if (delta > 0.5f) {
+         v1 += 1.0f;
+      }
+      else if (delta < -0.5f) {
+         v2 += 1.0f;
+      }
+
+      delta = v0 - v2;
+      if (delta > 0.5f) {
+         v2 += 1.0f;
+      }
+      else if (delta < -0.5f) {
+         v0 += 1.0f;
+      }
+   }
+
+   output[0] = v0;
+   output[1] = v1;
+   output[2] = v2;
+}
+
+
 /**
  * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
  * The value value comes from vertex[slot][i].
@@ -614,9 +447,10 @@ static boolean setup_sort_vertices( struct setup_context *setup,
  * \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)
+static void
+const_coeff(struct setup_context *setup,
+            struct tgsi_interp_coef *coef,
+            uint vertSlot, uint i)
 {
    assert(i <= 3);
 
@@ -632,13 +466,16 @@ static void const_coeff( struct setup_context *setup,
 /**
  * Compute a0, dadx and dady for a linearly interpolated coefficient,
  * for a triangle.
+ * v[0], v[1] and v[2] are vmin, vmid and vmax, respectively.
  */
-static void tri_linear_coeff( struct setup_context *setup,
-                              struct tgsi_interp_coef *coef,
-                              uint vertSlot, uint i)
+static void
+tri_linear_coeff(struct setup_context *setup,
+                 struct tgsi_interp_coef *coef,
+                 uint i,
+                 const float v[3])
 {
-   float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i];
-   float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
+   float botda = v[1] - v[0];
+   float majda = v[2] - v[0];
    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;
@@ -651,7 +488,7 @@ static void tri_linear_coeff( struct setup_context *setup,
 
    /* 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).
+    * pixel centers, in other words (pixel_offset, pixel_offset).
     *
     * 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
@@ -661,17 +498,9 @@ static void tri_linear_coeff( struct setup_context *setup,
     * 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]);
-   */
+   coef->a0[i] = (v[0] -
+                  (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
+                   dady * (setup->vmin[0][1] - setup->pixel_offset)));
 }
 
 
@@ -682,16 +511,19 @@ static void tri_linear_coeff( struct setup_context *setup,
  * 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.
+ * v[0], v[1] and v[2] are vmin, vmid and vmax, respectively.
  */
-static void tri_persp_coeff( struct setup_context *setup,
-                             struct tgsi_interp_coef *coef,
-                             uint vertSlot, uint i)
+static void
+tri_persp_coeff(struct setup_context *setup,
+                struct tgsi_interp_coef *coef,
+                uint i,
+                const float v[3])
 {
    /* 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 mina = v[0] * setup->vmin[0][3];
+   float mida = v[1] * setup->vmid[0][3];
+   float maxa = v[2] * setup->vmax[0][3];
    float botda = mida - mina;
    float majda = maxa - mina;
    float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
@@ -699,40 +531,41 @@ static void tri_persp_coeff( struct setup_context *setup,
    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)));
+                  (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
+                   dady * (setup->vmin[0][1] - setup->pixel_offset)));
 }
 
 
 /**
  * Special coefficient setup for gl_FragCoord.
- * X and Y are trivial, though Y has to be inverted for OpenGL.
+ * X and Y are trivial, though Y may have 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)
 {
+   const struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info;
+   boolean origin_lower_left =
+         fsInfo->properties[TGSI_PROPERTY_FS_COORD_ORIGIN];
+   boolean pixel_center_integer =
+         fsInfo->properties[TGSI_PROPERTY_FS_COORD_PIXEL_CENTER];
+
    /*X*/
-   setup->coef[slot].a0[0] = 0;
-   setup->coef[slot].dadx[0] = 1.0;
-   setup->coef[slot].dady[0] = 0.0;
+   setup->coef[slot].a0[0] = pixel_center_integer ? 0.0f : 0.5f;
+   setup->coef[slot].dadx[0] = 1.0f;
+   setup->coef[slot].dady[0] = 0.0f;
    /*Y*/
-   setup->coef[slot].a0[1] = 0.0;
-   setup->coef[slot].dadx[1] = 0.0;
-   setup->coef[slot].dady[1] = 1.0;
+   setup->coef[slot].a0[1] =
+		   (origin_lower_left ? setup->softpipe->framebuffer.height-1 : 0)
+		   + (pixel_center_integer ? 0.0f : 0.5f);
+   setup->coef[slot].dadx[1] = 0.0f;
+   setup->coef[slot].dady[1] = origin_lower_left ? -1.0f : 1.0f;
    /*Z*/
    setup->coef[slot].a0[2] = setup->posCoef.a0[2];
    setup->coef[slot].dadx[2] = setup->posCoef.dadx[2];
@@ -749,76 +582,111 @@ setup_fragcoord_coeff(struct setup_context *setup, uint slot)
  * 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 )
+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);
+   const struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info;
+   const struct sp_setup_info *sinfo = &softpipe->setup_info;
    uint fragSlot;
+   float v[3];
+
+   assert(sinfo->valid);
 
    /* z and w are done by linear interpolation:
     */
-   tri_linear_coeff(setup, &setup->posCoef, 0, 2);
-   tri_linear_coeff(setup, &setup->posCoef, 0, 3);
+   v[0] = setup->vmin[0][2];
+   v[1] = setup->vmid[0][2];
+   v[2] = setup->vmax[0][2];
+   tri_linear_coeff(setup, &setup->posCoef, 2, v);
+
+   v[0] = setup->vmin[0][3];
+   v[1] = setup->vmid[0][3];
+   v[2] = setup->vmax[0][3];
+   tri_linear_coeff(setup, &setup->posCoef, 3, v);
 
    /* setup interpolation for all the remaining attributes:
     */
-   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->attrib[fragSlot].src_index;
+   for (fragSlot = 0; fragSlot < fsInfo->num_inputs; fragSlot++) {
+      const uint vertSlot = sinfo->attrib[fragSlot].src_index;
       uint j;
 
-      switch (vinfo->attrib[fragSlot].interp_mode) {
-      case INTERP_CONSTANT:
-         for (j = 0; j < NUM_CHANNELS; j++)
+      switch (sinfo->attrib[fragSlot].interp) {
+      case SP_INTERP_CONSTANT:
+         for (j = 0; j < TGSI_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);
+      case SP_INTERP_LINEAR:
+         for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
+            tri_apply_cylindrical_wrap(setup->vmin[vertSlot][j],
+                                       setup->vmid[vertSlot][j],
+                                       setup->vmax[vertSlot][j],
+                                       fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j),
+                                       v);
+            tri_linear_coeff(setup, &setup->coef[fragSlot], j, v);
+         }
          break;
-      case INTERP_PERSPECTIVE:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+      case SP_INTERP_PERSPECTIVE:
+         for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
+            tri_apply_cylindrical_wrap(setup->vmin[vertSlot][j],
+                                       setup->vmid[vertSlot][j],
+                                       setup->vmax[vertSlot][j],
+                                       fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j),
+                                       v);
+            tri_persp_coeff(setup, &setup->coef[fragSlot], j, v);
+         }
          break;
-      case INTERP_POS:
+      case SP_INTERP_POS:
          setup_fragcoord_coeff(setup, fragSlot);
          break;
       default:
          assert(0);
       }
 
-      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
-         setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing;
+      if (fsInfo->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
+         /* convert 0 to 1.0 and 1 to -1.0 */
+         setup->coef[fragSlot].a0[0] = setup->facing * -2.0f + 1.0f;
          setup->coef[fragSlot].dadx[0] = 0.0;
          setup->coef[fragSlot].dady[0] = 0.0;
       }
+
+      if (0) {
+         for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
+            debug_printf("attr[%d].%c: a0:%f dx:%f dy:%f\n",
+                         fragSlot, "xyzw"[j],
+                         setup->coef[fragSlot].a0[j],
+                         setup->coef[fragSlot].dadx[j],
+                         setup->coef[fragSlot].dady[j]);
+         }
+      }
    }
 }
 
 
-
-static void setup_tri_edges( struct setup_context *setup )
+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_x = setup->vmin[0][0] + setup->pixel_offset;
+   float vmid_x = setup->vmid[0][0] + setup->pixel_offset;
 
-   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;
+   float vmin_y = setup->vmin[0][1] - setup->pixel_offset;
+   float vmid_y = setup->vmid[0][1] - setup->pixel_offset;
+   float vmax_y = setup->vmax[0][1] - setup->pixel_offset;
 
    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.dxdy = setup->emaj.dy ? setup->emaj.dx / setup->emaj.dy : .0f;
    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.dxdy = setup->etop.dy ? setup->etop.dx / setup->etop.dy : .0f;
    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.dxdy = setup->ebot.dy ? setup->ebot.dx / setup->ebot.dy : .0f;
    setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy;
 }
 
@@ -827,12 +695,14 @@ static void setup_tri_edges( struct setup_context *setup )
  * 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 )
+static void
+subtriangle(struct setup_context *setup,
+            struct edge *eleft,
+            struct edge *eright,
+            int lines,
+            unsigned viewport_index)
 {
-   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
+   const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect[viewport_index];
    const int minx = (int) cliprect->minx;
    const int maxx = (int) cliprect->maxx;
    const int miny = (int) cliprect->miny;
@@ -841,14 +711,14 @@ static void subtriangle( struct setup_context *setup,
    int sy = (int)eleft->sy;
 
    assert((int)eleft->sy == (int) eright->sy);
+   assert(lines >= 0);
 
    /* clip top/bottom */
    start_y = sy;
-   finish_y = sy + lines;
-
    if (start_y < miny)
       start_y = miny;
 
+   finish_y = sy + lines;
    if (finish_y > maxy)
       finish_y = maxy;
 
@@ -885,7 +755,6 @@ static void subtriangle( struct setup_context *setup,
 
          setup->span.left[_y&1] = left;
          setup->span.right[_y&1] = right;
-         setup->span.y_flags |= 1<<(_y&1);
       }
    }
 
@@ -905,9 +774,9 @@ static void subtriangle( struct setup_context *setup,
  * calculate it here.
  */
 static float
-calc_det( const float (*v0)[4],
-          const float (*v1)[4],
-          const float (*v2)[4] )
+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];
@@ -923,13 +792,15 @@ calc_det( const float (*v0)[4],
 /**
  * Do setup for triangle rasterization, then render the triangle.
  */
-void setup_tri( struct setup_context *setup,
-                const float (*v0)[4],
-                const float (*v1)[4],
-                const float (*v2)[4] )
+void
+sp_setup_tri(struct setup_context *setup,
+             const float (*v0)[4],
+             const float (*v1)[4],
+             const float (*v2)[4])
 {
    float det;
-
+   uint layer = 0;
+   unsigned viewport_index = 0;
 #if DEBUG_VERTS
    debug_printf("Setup triangle:\n");
    print_vertex(setup, v0);
@@ -937,7 +808,7 @@ void setup_tri( struct setup_context *setup,
    print_vertex(setup, v2);
 #endif
 
-   if (setup->softpipe->no_rast)
+   if (setup->softpipe->no_rast || setup->softpipe->rasterizer->rasterizer_discard)
       return;
    
    det = calc_det(v0, v1, v2);
@@ -950,40 +821,50 @@ void setup_tri( struct setup_context *setup,
    setup->numFragsWritten = 0;
 #endif
 
-   if (cull_tri( setup, det ))
-      return;
-
    if (!setup_sort_vertices( setup, det, v0, v1, v2 ))
       return;
+
    setup_tri_coefficients( setup );
    setup_tri_edges( setup );
 
-   setup->quad.input.prim = QUAD_PRIM_TRI;
+   assert(setup->softpipe->reduced_prim == PIPE_PRIM_TRIANGLES);
 
    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 ); */
+   if (setup->softpipe->layer_slot > 0) {
+      layer = *(unsigned *)setup->vprovoke[setup->softpipe->layer_slot];
+      layer = MIN2(layer, setup->max_layer);
+   }
+   setup->quad[0].input.layer = layer;
+
+   if (setup->softpipe->viewport_index_slot > 0) {
+      unsigned *udata = (unsigned*)v0[setup->softpipe->viewport_index_slot];
+      viewport_index = sp_clamp_viewport_idx(*udata);
+   }
+   setup->quad[0].input.viewport_index = viewport_index;
 
    /*   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 );
+      subtriangle(setup, &setup->emaj, &setup->ebot, setup->ebot.lines, viewport_index);
+      subtriangle(setup, &setup->emaj, &setup->etop, setup->etop.lines, viewport_index);
    }
    else {
       /* emaj on right:
        */
-      subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines );
-      subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines );
+      subtriangle(setup, &setup->ebot, &setup->emaj, setup->ebot.lines, viewport_index);
+      subtriangle(setup, &setup->etop, &setup->emaj, setup->etop.lines, viewport_index);
    }
 
    flush_spans( setup );
 
-   WAIT_FOR_COMPLETION(setup);
+   if (setup->softpipe->active_statistics_queries) {
+      setup->softpipe->pipeline_statistics.c_primitives++;
+   }
 
 #if DEBUG_FRAGS
    printf("Tri: %u frags emitted, %u written\n",
@@ -993,47 +874,75 @@ void setup_tri( struct setup_context *setup,
 }
 
 
+/* Apply cylindrical wrapping to v0, v1 coordinates, if enabled.
+ * Input coordinates must be in [0, 1] range, otherwise results are undefined.
+ */
+static void
+line_apply_cylindrical_wrap(float v0,
+                            float v1,
+                            uint cylindrical_wrap,
+                            float output[2])
+{
+   if (cylindrical_wrap) {
+      float delta;
+
+      delta = v1 - v0;
+      if (delta > 0.5f) {
+         v0 += 1.0f;
+      }
+      else if (delta < -0.5f) {
+         v1 += 1.0f;
+      }
+   }
+
+   output[0] = v0;
+   output[1] = v1;
+}
+
 
 /**
  * Compute a0, dadx and dady for a linearly interpolated coefficient,
  * for a line.
+ * v[0] and v[1] are vmin and vmax, respectively.
  */
 static void
 line_linear_coeff(const struct setup_context *setup,
                   struct tgsi_interp_coef *coef,
-                  uint vertSlot, uint i)
+                  uint i,
+                  const float v[2])
 {
-   const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
+   const float da = v[1] - v[0];
    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)));
+   coef->a0[i] = (v[0] -
+                  (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
+                   dady * (setup->vmin[0][1] - setup->pixel_offset)));
 }
 
 
 /**
  * Compute a0, dadx and dady for a perspective-corrected interpolant,
  * for a line.
+ * v[0] and v[1] are vmin and vmax, respectively.
  */
 static void
 line_persp_coeff(const struct setup_context *setup,
                  struct tgsi_interp_coef *coef,
-                 uint vertSlot, uint i)
+                 uint i,
+                 const float v[2])
 {
-   /* 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 a0 = v[0] * setup->vmin[0][3];
+   const float a1 = v[1] * 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)));
+   coef->a0[i] = (a0 -
+                  (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
+                   dady * (setup->vmin[0][1] - setup->pixel_offset)));
 }
 
 
@@ -1041,16 +950,19 @@ line_persp_coeff(const struct setup_context *setup,
  * Compute the setup->coef[] array dadx, dady, a0 values.
  * Must be called after setup->vmin,vmax are initialized.
  */
-static INLINE boolean
+static boolean
 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);
+   const struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info;
+   const struct sp_setup_info *sinfo = &softpipe->setup_info;
    uint fragSlot;
    float area;
+   float v[2];
+
+   assert(sinfo->valid);
 
    /* use setup->vmin, vmax to point to vertices */
    if (softpipe->rasterizer->flatshade_first)
@@ -1071,37 +983,53 @@ setup_line_coefficients(struct setup_context *setup,
 
    /* z and w are done by linear interpolation:
     */
-   line_linear_coeff(setup, &setup->posCoef, 0, 2);
-   line_linear_coeff(setup, &setup->posCoef, 0, 3);
+   v[0] = setup->vmin[0][2];
+   v[1] = setup->vmax[0][2];
+   line_linear_coeff(setup, &setup->posCoef, 2, v);
+
+   v[0] = setup->vmin[0][3];
+   v[1] = setup->vmax[0][3];
+   line_linear_coeff(setup, &setup->posCoef, 3, v);
 
    /* setup interpolation for all the remaining attributes:
     */
-   for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->attrib[fragSlot].src_index;
+   for (fragSlot = 0; fragSlot < fsInfo->num_inputs; fragSlot++) {
+      const uint vertSlot = sinfo->attrib[fragSlot].src_index;
       uint j;
 
-      switch (vinfo->attrib[fragSlot].interp_mode) {
-      case INTERP_CONSTANT:
-         for (j = 0; j < NUM_CHANNELS; j++)
+      switch (sinfo->attrib[fragSlot].interp) {
+      case SP_INTERP_CONSTANT:
+         for (j = 0; j < TGSI_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);
+      case SP_INTERP_LINEAR:
+         for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
+            line_apply_cylindrical_wrap(setup->vmin[vertSlot][j],
+                                        setup->vmax[vertSlot][j],
+                                        fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j),
+                                        v);
+            line_linear_coeff(setup, &setup->coef[fragSlot], j, v);
+         }
          break;
-      case INTERP_PERSPECTIVE:
-         for (j = 0; j < NUM_CHANNELS; j++)
-            line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
+      case SP_INTERP_PERSPECTIVE:
+         for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
+            line_apply_cylindrical_wrap(setup->vmin[vertSlot][j],
+                                        setup->vmax[vertSlot][j],
+                                        fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j),
+                                        v);
+            line_persp_coeff(setup, &setup->coef[fragSlot], j, v);
+         }
          break;
-      case INTERP_POS:
+      case SP_INTERP_POS:
          setup_fragcoord_coeff(setup, fragSlot);
          break;
       default:
          assert(0);
       }
 
-      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
-         setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing;
+      if (fsInfo->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
+         /* convert 0 to 1.0 and 1 to -1.0 */
+         setup->coef[fragSlot].a0[0] = setup->facing * -2.0f + 1.0f;
          setup->coef[fragSlot].dadx[0] = 0.0;
          setup->coef[fragSlot].dady[0] = 0.0;
       }
@@ -1113,7 +1041,7 @@ setup_line_coefficients(struct setup_context *setup,
 /**
  * Plot a pixel in a line segment.
  */
-static INLINE void
+static inline void
 plot(struct setup_context *setup, int x, int y)
 {
    const int iy = y & 1;
@@ -1122,20 +1050,20 @@ plot(struct setup_context *setup, int x, int y)
    const int quadY = y - iy;
    const int mask = (1 << ix) << (2 * iy);
 
-   if (quadX != setup->quad.input.x0 ||
-       quadY != setup->quad.input.y0)
+   if (quadX != setup->quad[0].input.x0 ||
+       quadY != setup->quad[0].input.y0)
    {
       /* flush prev quad, start new quad */
 
-      if (setup->quad.input.x0 != -1)
-         CLIP_EMIT_QUAD(setup);
+      if (setup->quad[0].input.x0 != -1)
+         clip_emit_quad(setup, &setup->quad[0]);
 
-      setup->quad.input.x0 = quadX;
-      setup->quad.input.y0 = quadY;
-      setup->quad.inout.mask = 0x0;
+      setup->quad[0].input.x0 = quadX;
+      setup->quad[0].input.y0 = quadY;
+      setup->quad[0].inout.mask = 0x0;
    }
 
-   setup->quad.inout.mask |= mask;
+   setup->quad[0].inout.mask |= mask;
 }
 
 
@@ -1145,9 +1073,9 @@ plot(struct setup_context *setup, int x, int y)
  * to handle stippling and wide lines.
  */
 void
-setup_line(struct setup_context *setup,
-           const float (*v0)[4],
-           const float (*v1)[4])
+sp_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];
@@ -1156,6 +1084,8 @@ setup_line(struct setup_context *setup,
    int dx = x1 - x0;
    int dy = y1 - y0;
    int xstep, ystep;
+   uint layer = 0;
+   unsigned viewport_index = 0;
 
 #if DEBUG_VERTS
    debug_printf("Setup line:\n");
@@ -1163,7 +1093,7 @@ setup_line(struct setup_context *setup,
    print_vertex(setup, v1);
 #endif
 
-   if (setup->softpipe->no_rast)
+   if (setup->softpipe->no_rast || setup->softpipe->rasterizer->rasterizer_discard)
       return;
 
    if (dx == 0 && dy == 0)
@@ -1195,17 +1125,29 @@ setup_line(struct setup_context *setup,
 
    assert(dx >= 0);
    assert(dy >= 0);
+   assert(setup->softpipe->reduced_prim == PIPE_PRIM_LINES);
+
+   setup->quad[0].input.x0 = setup->quad[0].input.y0 = -1;
+   setup->quad[0].inout.mask = 0x0;
+   if (setup->softpipe->layer_slot > 0) {
+      layer = *(unsigned *)setup->vprovoke[setup->softpipe->layer_slot];
+      layer = MIN2(layer, setup->max_layer);
+   }
+   setup->quad[0].input.layer = layer;
+
+   if (setup->softpipe->viewport_index_slot > 0) {
+      unsigned *udata = (unsigned*)setup->vprovoke[setup->softpipe->viewport_index_slot];
+      viewport_index = sp_clamp_viewport_idx(*udata);
+   }
+   setup->quad[0].input.viewport_index = viewport_index;
 
-   setup->quad.input.x0 = setup->quad.input.y0 = -1;
-   setup->quad.inout.mask = 0x0;
-   setup->quad.input.prim = QUAD_PRIM_LINE;
    /* XXX temporary: set coverage to 1.0 so the line appears
     * if AA mode happens to be enabled.
     */
-   setup->quad.input.coverage[0] =
-   setup->quad.input.coverage[1] =
-   setup->quad.input.coverage[2] =
-   setup->quad.input.coverage[3] = 1.0;
+   setup->quad[0].input.coverage[0] =
+   setup->quad[0].input.coverage[1] =
+   setup->quad[0].input.coverage[2] =
+   setup->quad[0].input.coverage[3] = 1.0;
 
    if (dx > dy) {
       /*** X-major line ***/
@@ -1249,11 +1191,9 @@ setup_line(struct setup_context *setup,
    }
 
    /* draw final quad */
-   if (setup->quad.inout.mask) {
-      CLIP_EMIT_QUAD(setup);
+   if (setup->quad[0].inout.mask) {
+      clip_emit_quad(setup, &setup->quad[0]);
    }
-
-   WAIT_FOR_COMPLETION(setup);
 }
 
 
@@ -1276,11 +1216,11 @@ point_persp_coeff(const struct setup_context *setup,
  * XXX could optimize a lot for 1-pixel points.
  */
 void
-setup_point( struct setup_context *setup,
-             const float (*v0)[4] )
+sp_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 struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info;
    const int sizeAttr = setup->softpipe->psize_slot;
    const float size
       = sizeAttr > 0 ? v0[sizeAttr][0]
@@ -1289,17 +1229,34 @@ setup_point( struct setup_context *setup,
    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);
+   const struct sp_setup_info *sinfo = &softpipe->setup_info;
    uint fragSlot;
-
+   uint layer = 0;
+   unsigned viewport_index = 0;
 #if DEBUG_VERTS
    debug_printf("Setup point:\n");
    print_vertex(setup, v0);
 #endif
 
-   if (softpipe->no_rast)
+   assert(sinfo->valid);
+
+   if (setup->softpipe->no_rast || setup->softpipe->rasterizer->rasterizer_discard)
       return;
 
+   assert(setup->softpipe->reduced_prim == PIPE_PRIM_POINTS);
+
+   if (setup->softpipe->layer_slot > 0) {
+      layer = *(unsigned *)v0[setup->softpipe->layer_slot];
+      layer = MIN2(layer, setup->max_layer);
+   }
+   setup->quad[0].input.layer = layer;
+
+   if (setup->softpipe->viewport_index_slot > 0) {
+      unsigned *udata = (unsigned*)v0[setup->softpipe->viewport_index_slot];
+      viewport_index = sp_clamp_viewport_idx(*udata);
+   }
+   setup->quad[0].input.viewport_index = viewport_index;
+
    /* For points, all interpolants are constant-valued.
     * However, for point sprites, we'll need to setup texcoords appropriately.
     * XXX: which coefficients are the texcoords???
@@ -1322,46 +1279,46 @@ setup_point( struct setup_context *setup,
    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->attrib[fragSlot].src_index;
+   for (fragSlot = 0; fragSlot < fsInfo->num_inputs; fragSlot++) {
+      const uint vertSlot = sinfo->attrib[fragSlot].src_index;
       uint j;
 
-      switch (vinfo->attrib[fragSlot].interp_mode) {
-      case INTERP_CONSTANT:
+      switch (sinfo->attrib[fragSlot].interp) {
+      case SP_INTERP_CONSTANT:
          /* fall-through */
-      case INTERP_LINEAR:
-         for (j = 0; j < NUM_CHANNELS; j++)
+      case SP_INTERP_LINEAR:
+         for (j = 0; j < TGSI_NUM_CHANNELS; j++)
             const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
          break;
-      case INTERP_PERSPECTIVE:
-         for (j = 0; j < NUM_CHANNELS; j++)
+      case SP_INTERP_PERSPECTIVE:
+         for (j = 0; j < TGSI_NUM_CHANNELS; j++)
             point_persp_coeff(setup, setup->vprovoke,
                               &setup->coef[fragSlot], vertSlot, j);
          break;
-      case INTERP_POS:
+      case SP_INTERP_POS:
          setup_fragcoord_coeff(setup, fragSlot);
          break;
       default:
          assert(0);
       }
 
-      if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
-         setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing;
+      if (fsInfo->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
+         /* convert 0 to 1.0 and 1 to -1.0 */
+         setup->coef[fragSlot].a0[0] = setup->facing * -2.0f + 1.0f;
          setup->coef[fragSlot].dadx[0] = 0.0;
          setup->coef[fragSlot].dady[0] = 0.0;
       }
    }
 
-   setup->quad.input.prim = QUAD_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.input.x0 = (int) x - ix;
-      setup->quad.input.y0 = (int) y - iy;
-      setup->quad.inout.mask = (1 << ix) << (2 * iy);
-      CLIP_EMIT_QUAD(setup);
+      setup->quad[0].input.x0 = (int) x - ix;
+      setup->quad[0].input.y0 = (int) y - iy;
+      setup->quad[0].inout.mask = (1 << ix) << (2 * iy);
+      clip_emit_quad(setup, &setup->quad[0]);
    }
    else {
       if (round) {
@@ -1381,15 +1338,15 @@ setup_point( struct setup_context *setup,
             for (ix = ixmin; ix <= ixmax; ix += 2) {
                float dx, dy, dist2, cover;
 
-               setup->quad.inout.mask = 0x0;
+               setup->quad[0].inout.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.input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_TOP_LEFT;
+                  setup->quad[0].input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_TOP_LEFT;
                }
 
                dx = (ix + 1.5f) - x;
@@ -1397,8 +1354,8 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_TOP_RIGHT;
+                  setup->quad[0].input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_TOP_RIGHT;
                }
 
                dx = (ix + 0.5f) - x;
@@ -1406,8 +1363,8 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_BOTTOM_LEFT;
+                  setup->quad[0].input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_BOTTOM_LEFT;
                }
 
                dx = (ix + 1.5f) - x;
@@ -1415,14 +1372,14 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_BOTTOM_RIGHT;
+                  setup->quad[0].input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_BOTTOM_RIGHT;
                }
 
-               if (setup->quad.inout.mask) {
-                  setup->quad.input.x0 = ix;
-                  setup->quad.input.y0 = iy;
-                  CLIP_EMIT_QUAD(setup);
+               if (setup->quad[0].inout.mask) {
+                  setup->quad[0].input.x0 = ix;
+                  setup->quad[0].input.y0 = iy;
+                  clip_emit_quad(setup, &setup->quad[0]);
                }
             }
          }
@@ -1466,49 +1423,77 @@ setup_point( struct setup_context *setup,
                   mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
                }
 
-               setup->quad.inout.mask = mask;
-               setup->quad.input.x0 = ix;
-               setup->quad.input.y0 = iy;
-               CLIP_EMIT_QUAD(setup);
+               setup->quad[0].inout.mask = mask;
+               setup->quad[0].input.x0 = ix;
+               setup->quad[0].input.y0 = iy;
+               clip_emit_quad(setup, &setup->quad[0]);
             }
          }
       }
    }
-
-   WAIT_FOR_COMPLETION(setup);
 }
 
-void setup_prepare( struct setup_context *setup )
+
+/**
+ * Called by vbuf code just before we start buffering primitives.
+ */
+void
+sp_setup_prepare(struct setup_context *setup)
 {
    struct softpipe_context *sp = setup->softpipe;
-   unsigned i;
-
+   int i;
+   unsigned max_layer = ~0;
    if (sp->dirty) {
-      softpipe_update_derived(sp);
+      softpipe_update_derived(sp, sp->reduced_api_prim);
    }
 
    /* Note: nr_attrs is only used for debugging (vertex printing) */
-   setup->quad.nr_attrs = draw_num_vs_outputs(sp->draw);
+   setup->nr_vertex_attrs = draw_num_shader_outputs(sp->draw);
+
+   /*
+    * Determine how many layers the fb has (used for clamping layer value).
+    * OpenGL (but not d3d10) permits different amount of layers per rt, however
+    * results are undefined if layer exceeds the amount of layers of ANY
+    * attachment hence don't need separate per cbuf and zsbuf max.
+    */
+   for (i = 0; i < setup->softpipe->framebuffer.nr_cbufs; i++) {
+      struct pipe_surface *cbuf = setup->softpipe->framebuffer.cbufs[i];
+      if (cbuf) {
+         max_layer = MIN2(max_layer,
+                          cbuf->u.tex.last_layer - cbuf->u.tex.first_layer);
 
-   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
-      sp->quad[i].first->begin( sp->quad[i].first );
+      }
    }
 
+   /* Prepare pixel offset for rasterisation:
+    *  - pixel center (0.5, 0.5) for GL, or
+    *  - assume (0.0, 0.0) for other APIs.
+    */
+   if (setup->softpipe->rasterizer->half_pixel_center) {
+      setup->pixel_offset = 0.5f;
+   } else {
+      setup->pixel_offset = 0.0f;
+   }
+
+   setup->max_layer = max_layer;
+
+   sp->quad.first->begin( sp->quad.first );
+
    if (sp->reduced_api_prim == PIPE_PRIM_TRIANGLES &&
-       sp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL &&
-       sp->rasterizer->fill_ccw == PIPE_POLYGON_MODE_FILL) {
+       sp->rasterizer->fill_front == PIPE_POLYGON_MODE_FILL &&
+       sp->rasterizer->fill_back == PIPE_POLYGON_MODE_FILL) {
       /* we'll do culling */
-      setup->winding = sp->rasterizer->cull_mode;
+      setup->cull_face = sp->rasterizer->cull_face;
    }
    else {
       /* 'draw' will do culling */
-      setup->winding = PIPE_WINDING_NONE;
+      setup->cull_face = PIPE_FACE_NONE;
    }
 }
 
 
-
-void setup_destroy_context( struct setup_context *setup )
+void
+sp_setup_destroy_context(struct setup_context *setup)
 {
    FREE( setup );
 }
@@ -1517,34 +1502,21 @@ void setup_destroy_context( struct setup_context *setup )
 /**
  * Create a new primitive setup/render stage.
  */
-struct setup_context *setup_create_context( struct softpipe_context *softpipe )
+struct setup_context *
+sp_setup_create_context(struct softpipe_context *softpipe)
 {
    struct setup_context *setup = CALLOC_STRUCT(setup_context);
-#if SP_NUM_QUAD_THREADS > 1
-   uint i;
-#endif
+   unsigned i;
 
    setup->softpipe = softpipe;
 
-   setup->quad.coef = setup->coef;
-   setup->quad.posCoef = &setup->posCoef;
-
-#if SP_NUM_QUAD_THREADS > 1
-   setup->que.first = 0;
-   setup->que.last = 0;
-   pipe_mutex_init( setup->que.que_mutex );
-   pipe_condvar_init( setup->que.que_notfull_condvar );
-   pipe_condvar_init( setup->que.que_notempty_condvar );
-   setup->que.jobs_added = 0;
-   setup->que.jobs_done = 0;
-   pipe_condvar_init( setup->que.que_done_condvar );
-   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
-      setup->threads[i].setup = setup;
-      setup->threads[i].id = i;
-      setup->threads[i].handle = pipe_thread_create( quad_thread, &setup->threads[i] );
+   for (i = 0; i < MAX_QUADS; i++) {
+      setup->quad[i].coef = setup->coef;
+      setup->quad[i].posCoef = &setup->posCoef;
    }
-#endif
+
+   setup->span.left[0] = 1000000;     /* greater than right[0] */
+   setup->span.left[1] = 1000000;     /* greater than right[1] */
 
    return setup;
 }
-