llvmpipe: reintroduce SET_STATE binner command
[mesa.git] / src / gallium / drivers / llvmpipe / lp_setup_tri.c
index 53ffb95505bdd6ae8e4b822a84222e69f5d5eb51..bc48eb8d1b54785287ed94a4a7f47f3f9448c2ee 100644 (file)
@@ -160,8 +160,9 @@ lp_setup_print_triangle(struct lp_setup_context *setup,
 }
 
 
+#define MAX_PLANES 8
 static unsigned
-lp_rast_tri_tab[9] = {
+lp_rast_tri_tab[MAX_PLANES+1] = {
    0,               /* should be impossible */
    LP_RAST_OP_TRIANGLE_1,
    LP_RAST_OP_TRIANGLE_2,
@@ -199,14 +200,16 @@ lp_setup_whole_tile(struct lp_setup_context *setup,
       }
 
       LP_COUNT(nr_shade_opaque_64);
-      return lp_scene_bin_command( scene, tx, ty,
-                                   LP_RAST_OP_SHADE_TILE_OPAQUE,
-                                   lp_rast_arg_inputs(inputs) );
+      return lp_scene_bin_cmd_with_state( scene, tx, ty,
+                                          setup->fs.stored,
+                                          LP_RAST_OP_SHADE_TILE_OPAQUE,
+                                          lp_rast_arg_inputs(inputs) );
    } else {
       LP_COUNT(nr_shade_64);
-      return lp_scene_bin_command( scene, tx, ty,
-                                   LP_RAST_OP_SHADE_TILE,
-                                   lp_rast_arg_inputs(inputs) );
+      return lp_scene_bin_cmd_with_state( scene, tx, ty,
+                                          setup->fs.stored, 
+                                          LP_RAST_OP_SHADE_TILE,
+                                          lp_rast_arg_inputs(inputs) );
    }
 }
 
@@ -227,7 +230,6 @@ do_triangle_ccw(struct lp_setup_context *setup,
    struct lp_rast_triangle *tri;
    int x[3];
    int y[3];
-   int area;
    struct u_rect bbox;
    unsigned tri_bytes;
    int i;
@@ -311,21 +313,8 @@ do_triangle_ccw(struct lp_setup_context *setup,
    tri->plane[1].dcdx = y[1] - y[2];
    tri->plane[2].dcdx = y[2] - y[0];
 
-   area = (tri->plane[0].dcdy * tri->plane[2].dcdx -
-           tri->plane[2].dcdy * tri->plane[0].dcdx);
-
    LP_COUNT(nr_tris);
 
-   /* Cull non-ccw and zero-sized triangles. 
-    *
-    * XXX: subject to overflow??
-    */
-   if (area <= 0) {
-      lp_scene_putback_data( scene, tri_bytes );
-      LP_COUNT(nr_culled_tris);
-      return TRUE;
-   }
-
    /* Setup parameter interpolants:
     */
    lp_setup_tri_coef( setup, &tri->inputs, v0, v1, v2, frontfacing );
@@ -333,7 +322,6 @@ do_triangle_ccw(struct lp_setup_context *setup,
    tri->inputs.facing = frontfacing ? 1.0F : -1.0F;
    tri->inputs.disable = FALSE;
    tri->inputs.opaque = setup->fs.current.variant->opaque;
-   tri->inputs.state = setup->fs.stored;
 
   
    for (i = 0; i < 3; i++) {
@@ -447,8 +435,10 @@ do_triangle_ccw(struct lp_setup_context *setup,
 static INLINE uint32_t 
 floor_pot(uint32_t n)
 {
-   assert(n);
 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
+   if (n == 0)
+      return 0;
+
    __asm__("bsr %1,%0"
           : "=r" (n)
           : "rm" (n));
@@ -484,56 +474,62 @@ lp_setup_bin_triangle( struct lp_setup_context *setup,
    int sz = floor_pot((bbox->x1 - (bbox->x0 & ~3)) |
                      (bbox->y1 - (bbox->y0 & ~3)));
 
-   if (nr_planes == 3) {
-      if (sz < 4 && dx < 64)
-      {
-        /* Triangle is contained in a single 4x4 stamp:
-         */
-        int mask = (bbox->x0 & 63 & ~3) | ((bbox->y0 & 63 & ~3) << 8);
-
-        return lp_scene_bin_command( scene,
-                                     bbox->x0/64, bbox->y0/64,
-                                     LP_RAST_OP_TRIANGLE_3_4,
-                                     lp_rast_arg_triangle(tri, mask) );
-      }
-
-      if (sz < 16 && dx < 64)
-      {
-        int mask = (bbox->x0 & 63 & ~3) | ((bbox->y0 & 63 & ~3) << 8);
-
-        /* Triangle is contained in a single 16x16 block:
-         */
-        return lp_scene_bin_command( scene,
-                                     bbox->x0/64, bbox->y0/64,
-                                      LP_RAST_OP_TRIANGLE_3_16,
-                                      lp_rast_arg_triangle(tri, mask) );
-      }
-   }
-
-
    /* Determine which tile(s) intersect the triangle's bounding box
     */
    if (dx < TILE_SIZE)
    {
       int ix0 = bbox->x0 / TILE_SIZE;
       int iy0 = bbox->y0 / TILE_SIZE;
+      int px = bbox->x0 & 63 & ~3;
+      int py = bbox->y0 & 63 & ~3;
+      int mask = px | (py << 8);
 
       assert(iy0 == bbox->y1 / TILE_SIZE &&
             ix0 == bbox->x1 / TILE_SIZE);
 
+      if (nr_planes == 3) {
+         if (sz < 4)
+         {
+            /* Triangle is contained in a single 4x4 stamp:
+             */
+            return lp_scene_bin_cmd_with_state( scene, ix0, iy0,
+                                                setup->fs.stored,
+                                                LP_RAST_OP_TRIANGLE_3_4,
+                                                lp_rast_arg_triangle(tri, mask) );
+         }
+
+         if (sz < 16)
+         {
+            /* Triangle is contained in a single 16x16 block:
+             */
+            return lp_scene_bin_cmd_with_state( scene, ix0, iy0,
+                                                setup->fs.stored,
+                                                LP_RAST_OP_TRIANGLE_3_16,
+                                                lp_rast_arg_triangle(tri, mask) );
+         }
+      }
+      else if (nr_planes == 4 && sz < 16) 
+      {
+         return lp_scene_bin_cmd_with_state(scene, ix0, iy0,
+                                            setup->fs.stored,
+                                            LP_RAST_OP_TRIANGLE_4_16,
+                                            lp_rast_arg_triangle(tri, mask) );
+      }
+
+
       /* Triangle is contained in a single tile:
        */
-      return lp_scene_bin_command( scene, ix0, iy0,
-                                   lp_rast_tri_tab[nr_planes], 
-                                   lp_rast_arg_triangle(tri, (1<<nr_planes)-1) );
+      return lp_scene_bin_cmd_with_state( scene, ix0, iy0, setup->fs.stored,
+                                          lp_rast_tri_tab[nr_planes], 
+                                          lp_rast_arg_triangle(tri, (1<<nr_planes)-1) );
    }
    else
    {
-      int c[7];
-      int ei[7];
-      int eo[7];
-      int xstep[7];
-      int ystep[7];
+      int c[MAX_PLANES];
+      int ei[MAX_PLANES];
+      int eo[MAX_PLANES];
+      int xstep[MAX_PLANES];
+      int ystep[MAX_PLANES];
       int x, y;
 
       int ix0 = bbox->x0 / TILE_SIZE;
@@ -562,7 +558,7 @@ lp_setup_bin_triangle( struct lp_setup_context *setup,
       for (y = iy0; y <= iy1; y++)
       {
         boolean in = FALSE;  /* are we inside the triangle? */
-        int cx[7];
+        int cx[MAX_PLANES];
 
          for (i = 0; i < nr_planes; i++)
             cx[i] = c[i];
@@ -591,9 +587,11 @@ lp_setup_bin_triangle( struct lp_setup_context *setup,
                 */
                int count = util_bitcount(partial);
                in = TRUE;
-               if (!lp_scene_bin_command( scene, x, y,
-                                          lp_rast_tri_tab[count], 
-                                          lp_rast_arg_triangle(tri, partial) ))
+               
+               if (!lp_scene_bin_cmd_with_state( scene, x, y,
+                                                 setup->fs.stored,
+                                                 lp_rast_tri_tab[count], 
+                                                 lp_rast_arg_triangle(tri, partial) ))
                   goto fail;
 
                LP_COUNT(nr_partially_covered_64);
@@ -632,40 +630,62 @@ fail:
 
 
 /**
- * Draw triangle if it's CW, cull otherwise.
+ * Try to draw the triangle, restart the scene on failure.
  */
-static void triangle_cw( struct lp_setup_context *setup,
-                        const float (*v0)[4],
-                        const float (*v1)[4],
-                        const float (*v2)[4] )
+static void retry_triangle_ccw( struct lp_setup_context *setup,
+                                const float (*v0)[4],
+                                const float (*v1)[4],
+                                const float (*v2)[4],
+                                boolean front)
 {
-   if (!do_triangle_ccw( setup, v1, v0, v2, !setup->ccw_is_frontface ))
+   if (!do_triangle_ccw( setup, v0, v1, v2, front ))
    {
-      lp_setup_flush_and_restart(setup);
+      if (!lp_setup_flush_and_restart(setup))
+         return;
 
-      if (!do_triangle_ccw( setup, v1, v0, v2, !setup->ccw_is_frontface ))
-         assert(0);
+      if (!do_triangle_ccw( setup, v0, v1, v2, front ))
+         return;
    }
 }
 
+static INLINE float
+calc_area(const float (*v0)[4],
+          const float (*v1)[4],
+          const float (*v2)[4])
+{
+   float dx01 = v0[0][0] - v1[0][0];
+   float dy01 = v0[0][1] - v1[0][1];
+   float dx20 = v2[0][0] - v0[0][0];
+   float dy20 = v2[0][1] - v0[0][1];
+   return dx01 * dy20 - dx20 * dy01;
+}
+
 
 /**
- * Draw triangle if it's CCW, cull otherwise.
+ * Draw triangle if it's CW, cull otherwise.
  */
-static void triangle_ccw( struct lp_setup_context *setup,
+static void triangle_cw( struct lp_setup_context *setup,
                         const float (*v0)[4],
                         const float (*v1)[4],
                         const float (*v2)[4] )
 {
-   if (!do_triangle_ccw( setup, v0, v1, v2, setup->ccw_is_frontface ))
-   {
-      lp_setup_flush_and_restart(setup);
-      if (!do_triangle_ccw( setup, v0, v1, v2, setup->ccw_is_frontface ))
-         assert(0);
-   }
+   float area = calc_area(v0, v1, v2);
+
+   if (area < 0.0f) 
+      retry_triangle_ccw(setup, v0, v2, v1, !setup->ccw_is_frontface);
 }
 
 
+static void triangle_ccw( struct lp_setup_context *setup,
+                          const float (*v0)[4],
+                          const float (*v1)[4],
+                          const float (*v2)[4])
+{
+   float area = calc_area(v0, v1, v2);
+
+   if (area > 0.0f) 
+      retry_triangle_ccw(setup, v0, v1, v2, setup->ccw_is_frontface);
+}
 
 /**
  * Draw triangle whether it's CW or CCW.
@@ -675,18 +695,12 @@ static void triangle_both( struct lp_setup_context *setup,
                           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 */
-   const float det = ex * fy - ey * fx;
-   if (det < 0.0f) 
-      triangle_ccw( setup, v0, v1, v2 );
-   else if (det > 0.0f)
-      triangle_cw( setup, v0, v1, v2 );
+   float area = calc_area(v0, v1, v2);
+
+   if (area > 0.0f) 
+      retry_triangle_ccw( setup, v0, v1, v2, setup->ccw_is_frontface );
+   else if (area < 0.0f)
+      retry_triangle_ccw( setup, v0, v2, v1, !setup->ccw_is_frontface );
 }