draw: more flatshade_first changes

- Reduce the number of changes to the normal vertex ordering
- Assume that the hardware knows how to do this in the standard case.
- Add support to the passthrough vcache path.

5 files changed, 144 insertions, 97 deletions

diff --git a/src/gallium/auxiliary/draw/draw_prim.c b/src/gallium/auxiliary/draw/draw_prim.c
index 9779aa8440a..4452376a70e 100644
--- a/src/gallium/auxiliary/draw/draw_prim.c
+++ b/src/gallium/auxiliary/draw/draw_prim.c
@@ -343,21 +343,11 @@ draw_prim( struct draw_context *draw,
       break;
 
    case PIPE_PRIM_LINES:
-      if (flatfirst) {
-         for (i = 0; i+1 < count; i += 2) {
-            do_line( draw, 
-               TRUE,
-               start + i + 1,
-               start + i + 0);
-         }
-      }
-      else {
-         for (i = 0; i+1 < count; i += 2) {
-            do_line( draw, 
-               TRUE,
-               start + i + 0,
-               start + i + 1);
-         }
+      for (i = 0; i+1 < count; i += 2) {
+         do_line( draw, 
+                  TRUE,
+                  start + i + 0,
+                  start + i + 1);
       }
       break;
 
@@ -378,63 +368,31 @@ draw_prim( struct draw_context *draw,
       break;
 
    case PIPE_PRIM_LINE_STRIP:
-      if (flatfirst) {
-         for (i = 1; i < count; i++) {
-            do_line( draw,
-               i == 1,
-               start + i,
-               start + i - 1 );
-         }
-      }
-      else {
-         for (i = 1; i < count; i++) {
-            do_line( draw,
-               i == 1,
-               start + i - 1,
-               start + i );
-         }
+      for (i = 1; i < count; i++) {
+         do_line( draw,
+                  i == 1,
+                  start + i - 1,
+                  start + i );
       }
       break;
 
    case PIPE_PRIM_TRIANGLES:
-      if (flatfirst) {
-         if (unfilled) {
-            for (i = 0; i+2 < count; i += 3) {
-               do_ef_triangle( draw,
-                  1, 
-                  ~0,
-                  start + i + 1,
-                  start + i + 2,
-                  start + i + 0 );
-            }
-         } 
-         else {
-            for (i = 0; i+2 < count; i += 3) {
-               do_triangle( draw,
-                  start + i + 1,
-                  start + i + 2,
-                  start + i + 0 );
-            }
+      if (unfilled) {
+         for (i = 0; i+2 < count; i += 3) {
+            do_ef_triangle( draw,
+                            1, 
+                            ~0,
+                            start + i + 0,
+                            start + i + 1,
+                            start + i + 2 );
          }
-      }
+      } 
       else {
-         if (unfilled) {
-            for (i = 0; i+2 < count; i += 3) {
-               do_ef_triangle( draw,
-                  1, 
-                  ~0,
-                  start + i + 0,
-                  start + i + 1,
-                  start + i + 2 );
-            }
-         } 
-         else {
-            for (i = 0; i+2 < count; i += 3) {
-               do_triangle( draw,
-                  start + i + 0,
-                  start + i + 1,
-                  start + i + 2 );
-            }
+         for (i = 0; i+2 < count; i += 3) {
+            do_triangle( draw,
+                         start + i + 0,
+                         start + i + 1,
+                         start + i + 2 );
          }
       }
       break;
@@ -444,15 +402,15 @@ draw_prim( struct draw_context *draw,
          for (i = 0; i+2 < count; i++) {
             if (i & 1) {
                do_triangle( draw,
+                  start + i + 0,
                   start + i + 2,
-                  start + i + 1,
-                  start + i + 0 );
+                  start + i + 1 );
             }
             else {
                do_triangle( draw,
+                  start + i + 0,
                   start + i + 1,
-                  start + i + 2,
-                  start + i + 0 );
+                  start + i + 2 );
             }
          }
       }
@@ -479,9 +437,9 @@ draw_prim( struct draw_context *draw,
          if (flatfirst) {
             for (i = 0; i+2 < count; i++) {
                do_triangle( draw,
+                  start + i + 1,
                   start + i + 2,
-                  start + 0,
-                  start + i + 1 );
+                  start + 0 );
             }
          }
          else {
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index df306a5fa4f..2ea96c686da 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -191,7 +191,7 @@ boolean draw_pt_init( struct draw_context *draw )
    if (!draw->pt.middle.fetch_emit)
       return FALSE;
 
-   draw->pt.front.vcache = draw_pt_vcache();
+   draw->pt.front.vcache = draw_pt_vcache( draw );
    if (!draw->pt.front.vcache)
       return FALSE;
 
diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h
index 439fa4c8813..f8786160799 100644
--- a/src/gallium/auxiliary/draw/draw_pt.h
+++ b/src/gallium/auxiliary/draw/draw_pt.h
@@ -110,7 +110,7 @@ const void *draw_pt_elt_ptr( struct draw_context *draw,
 
 /* Implementations:
  */
-struct draw_pt_front_end *draw_pt_vcache( void );
+struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw );
 struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw );
 
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c
index da9a3a52aed..16ffedf580c 100644
--- a/src/gallium/auxiliary/draw/draw_pt_vcache.c
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c
@@ -44,6 +44,7 @@
 
 struct vcache_frontend {
    struct draw_pt_front_end base;
+   struct draw_context *draw;
 
    unsigned in[CACHE_MAX];
    ushort out[CACHE_MAX];
@@ -157,14 +158,6 @@ static void vcache_quad( struct vcache_frontend *vcache,
 }
 
 
-static void vcache_prepare( struct draw_pt_front_end *frontend,
-                            struct draw_pt_middle_end *middle )
-{
-   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
-   vcache->middle = middle;
-   middle->prepare( middle );
-}
-
 static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = {
    PIPE_PRIM_POINTS,
    PIPE_PRIM_LINES,
@@ -179,11 +172,11 @@ static unsigned reduced_prim[PIPE_PRIM_POLYGON + 1] = {
 };
 
 
-static void vcache_run( struct draw_pt_front_end *frontend, 
-                         unsigned prim,
-                         pt_elt_func get_elt,
-                         const void *elts,
-                         unsigned count )
+static void vcache_run_pv2( struct draw_pt_front_end *frontend, 
+                            unsigned prim,
+                            pt_elt_func get_elt,
+                            const void *elts,
+                            unsigned count )
 {
    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
    unsigned i;
@@ -309,6 +302,109 @@ static void vcache_run( struct draw_pt_front_end *frontend,
    vcache_flush( vcache );
 }
 
+
+static void vcache_run_pv0( struct draw_pt_front_end *frontend, 
+                            unsigned prim,
+                            pt_elt_func get_elt,
+                            const void *elts,
+                            unsigned count )
+{
+   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
+   unsigned i;
+   
+   /* These are for validation only:
+    */
+   vcache->elt_func = get_elt;
+   vcache->elt_ptr = elts;
+   vcache->output_prim = reduced_prim[prim];
+
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < count; i ++) {
+         vcache_point( vcache,
+                       get_elt(elts, i) );
+      }
+      break;
+
+   case PIPE_PRIM_LINES:
+      for (i = 0; i+1 < count; i += 2) {
+         vcache_line( vcache, 
+                      TRUE,
+                      get_elt(elts, i + 0),
+                      get_elt(elts, i + 1));
+      }
+      break;
+
+   case PIPE_PRIM_LINE_STRIP:
+      for (i = 1; i < count; i++) {
+         vcache_line( vcache,
+                      i == 1,
+                      get_elt(elts, i - 1),
+                      get_elt(elts, i) );
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLES:
+      for (i = 0; i+2 < count; i += 3) {
+         vcache_triangle( vcache,
+                          get_elt(elts, i + 0),
+                          get_elt(elts, i + 1),
+                          get_elt(elts, i + 2) );
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      for (i = 0; i+2 < count; i++) {
+         if (i & 1) {
+            vcache_triangle( vcache,
+                             get_elt(elts, i + 0),
+                             get_elt(elts, i + 2),
+                             get_elt(elts, i + 1) );
+         }
+         else {
+            vcache_triangle( vcache,
+                             get_elt(elts, i + 0),
+                             get_elt(elts, i + 1),
+                             get_elt(elts, i + 2) );
+         }
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_FAN:
+      for (i = 0; i+2 < count; i++) {
+         vcache_triangle( vcache,
+                          get_elt(elts, i + 1),
+                          get_elt(elts, i + 2),
+                          get_elt(elts, 0) );
+      }
+      break;
+
+
+   default:
+      assert(0);
+      break;
+   }
+   
+   vcache_flush( vcache );
+}
+
+static void vcache_prepare( struct draw_pt_front_end *frontend,
+                            struct draw_pt_middle_end *middle )
+{
+   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
+
+   if (vcache->draw->rasterizer->flatshade_first)
+      vcache->base.run = vcache_run_pv0;
+   else
+      vcache->base.run = vcache_run_pv2;
+   
+   vcache->middle = middle;
+   middle->prepare( middle );
+}
+
+
+
+
 static void vcache_finish( struct draw_pt_front_end *frontend )
 {
    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
@@ -322,14 +418,15 @@ static void vcache_destroy( struct draw_pt_front_end *frontend )
 }
 
 
-struct draw_pt_front_end *draw_pt_vcache( void )
+struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw )
 {
    struct vcache_frontend *vcache = CALLOC_STRUCT( vcache_frontend );
  
    vcache->base.prepare = vcache_prepare;
-   vcache->base.run     = vcache_run;
+   vcache->base.run     = NULL;
    vcache->base.finish  = vcache_finish;
    vcache->base.destroy = vcache_destroy;
+   vcache->draw = draw;
    
    memset(vcache->in, ~0, sizeof(vcache->in));
   
diff --git a/src/gallium/auxiliary/draw/draw_validate.c b/src/gallium/auxiliary/draw/draw_validate.c
index ad43f06f73f..e163e078f0f 100644
--- a/src/gallium/auxiliary/draw/draw_validate.c
+++ b/src/gallium/auxiliary/draw/draw_validate.c
@@ -76,10 +76,6 @@ draw_need_pipeline(const struct draw_context *draw,
       /* AA lines */
       if (draw->rasterizer->line_smooth && draw->pipeline.aaline)
          return TRUE;
-
-      /* first-vertex driven flatshading */
-      if (draw->rasterizer->flatshade && draw->rasterizer->flatshade_first)
-         return TRUE;
    }
 
    if (points(prim))
@@ -116,10 +112,6 @@ draw_need_pipeline(const struct draw_context *draw,
       /* two-side lighting */
       if (draw->rasterizer->light_twoside)
          return TRUE;
-
-      /* first-vertex driven flatshading */
-      if (draw->rasterizer->flatshade && draw->rasterizer->flatshade_first)
-         return TRUE;
    }
 
    /* polygon cull - this is difficult - hardware can cull just fine