llvmpipe: drop scissor planes early if the tri is fully inside them

If the tri is fully inside a scissor edge (or rather, we just use the
bounding box of the tri for the comparison), then we can drop these
additional scissor "planes" early. We do not even need to allocate
space for them in the tri.

The math actually appears to be slightly iffy due to bounding boxes
being rounded, but it doesn't matter in the end.

Those scissor rects are costly - the 4 planes from the scissor are
already more expensive to calculate than the 3 planes from the tri itself,
and it also prevents us from using the specialized raster code for small
tris.

This helps openarena performance by about 8% or so. Of course, it helps
there that while openarena often enables scissoring (and even moves the
scissor rect around) I have not seen a single tri actually hit the
scissor rect, ever.

v2: drop individual scissor edges, and do it earlier, not even allocating
space for them.
v3: help the compiler a bit with simpler code, suggested by Brian.

Reviewed-by: Brian Paul <brianp@vmware.com>

2 files changed, 110 insertions, 69 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_setup_line.c b/src/gallium/drivers/llvmpipe/lp_setup_line.c
index f425825fc2..f6e1198d03 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_line.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_line.c
@@ -336,13 +336,6 @@ try_setup_line( struct lp_setup_context *setup,
       layer = MIN2(layer, scene->fb_max_layer);
    }
 
-   if (setup->scissor_test) {
-      nr_planes = 8;
-   }
-   else {
-      nr_planes = 4;
-   }
-
    dx = v1[0][0] - v2[0][0];
    dy = v1[0][1] - v2[0][1];
    area = (dx * dx  + dy * dy);
@@ -591,6 +584,20 @@ try_setup_line( struct lp_setup_context *setup,
    bbox.x0 = MAX2(bbox.x0, 0);
    bbox.y0 = MAX2(bbox.y0, 0);
 
+   nr_planes = 4;
+   /*
+    * Determine how many scissor planes we need, that is drop scissor
+    * edges if the bounding box of the tri is fully inside that edge.
+    */
+   if (setup->scissor_test) {
+      /* why not just use draw_regions */
+      struct u_rect *scissor = &setup->scissors[viewport_index];
+      nr_planes += (bbox.x0 < scissor->x0);
+      nr_planes += (bbox.x1 > scissor->x1);
+      nr_planes += (bbox.y0 < scissor->y0);
+      nr_planes += (bbox.y1 > scissor->y1);
+   }
+
    line = lp_setup_alloc_triangle(scene,
                                   key->num_inputs,
                                   nr_planes,
@@ -708,30 +715,44 @@ try_setup_line( struct lp_setup_context *setup,
     * Note that otherwise, the scissor planes only vary in 'C' value,
     * and even then only on state-changes.  Could alternatively store
     * these planes elsewhere.
+    * (Or only store the c value together with a bit indicating which
+    * scissor edge this is, so rasterization would treat them differently
+    * (easier to evaluate) to ordinary planes.)
     */
-   if (nr_planes == 8) {
-      const struct u_rect *scissor =
-         &setup->scissors[viewport_index];
-
-      plane[4].dcdx = -1 << 8;
-      plane[4].dcdy = 0;
-      plane[4].c = (1-scissor->x0) << 8;
-      plane[4].eo = 1 << 8;
-
-      plane[5].dcdx = 1 << 8;
-      plane[5].dcdy = 0;
-      plane[5].c = (scissor->x1+1) << 8;
-      plane[5].eo = 0;
-
-      plane[6].dcdx = 0;
-      plane[6].dcdy = 1 << 8;
-      plane[6].c = (1-scissor->y0) << 8;
-      plane[6].eo = 1 << 8;
-
-      plane[7].dcdx = 0;
-      plane[7].dcdy = -1 << 8;
-      plane[7].c = (scissor->y1+1) << 8;
-      plane[7].eo = 0;
+   if (nr_planes > 4) {
+      /* why not just use draw_regions */
+      struct u_rect *scissor = &setup->scissors[viewport_index];
+      struct lp_rast_plane *plane_s = &plane[4];
+
+      if (bbox.x0 < scissor->x0) {
+         plane_s->dcdx = -1 << 8;
+         plane_s->dcdy = 0;
+         plane_s->c = (1-scissor->x0) << 8;
+         plane_s->eo = 1 << 8;
+         plane_s++;
+      }
+      if (bbox.x1 > scissor->x1) {
+         plane_s->dcdx = 1 << 8;
+         plane_s->dcdy = 0;
+         plane_s->c = (scissor->x1+1) << 8;
+         plane_s->eo = 0 << 8;
+         plane_s++;
+      }
+      if (bbox.y0 < scissor->y0) {
+         plane_s->dcdx = 0;
+         plane_s->dcdy = 1 << 8;
+         plane_s->c = (1-scissor->y0) << 8;
+         plane_s->eo = 1 << 8;
+         plane_s++;
+      }
+      if (bbox.y1 > scissor->y1) {
+         plane_s->dcdx = 0;
+         plane_s->dcdy = -1 << 8;
+         plane_s->c = (scissor->y1+1) << 8;
+         plane_s->eo = 0;
+         plane_s++;
+      }
+      assert(plane_s == &plane[nr_planes]);
    }
 
    return lp_setup_bin_triangle(setup, line, &bbox, nr_planes, viewport_index);
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_tri.c b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
index 1e3a7501ed..7b0088912b 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
@@ -302,13 +302,6 @@ do_triangle_ccw(struct lp_setup_context *setup,
       layer = MIN2(layer, scene->fb_max_layer);
    }
 
-   if (setup->scissor_test) {
-      nr_planes = 7;
-   }
-   else {
-      nr_planes = 3;
-   }
-
    /* Bounding rectangle (in pixels) */
    {
       /* Yes this is necessary to accurately calculate bounding boxes
@@ -347,6 +340,20 @@ do_triangle_ccw(struct lp_setup_context *setup,
    bbox.x0 = MAX2(bbox.x0, 0);
    bbox.y0 = MAX2(bbox.y0, 0);
 
+   nr_planes = 3;
+   /*
+    * Determine how many scissor planes we need, that is drop scissor
+    * edges if the bounding box of the tri is fully inside that edge.
+    */
+   if (setup->scissor_test) {
+      /* why not just use draw_regions */
+      struct u_rect *scissor = &setup->scissors[viewport_index];
+      nr_planes += (bbox.x0 < scissor->x0);
+      nr_planes += (bbox.x1 > scissor->x1);
+      nr_planes += (bbox.y0 < scissor->y0);
+      nr_planes += (bbox.y1 > scissor->y1);
+   }
+
    tri = lp_setup_alloc_triangle(scene,
                                  key->num_inputs,
                                  nr_planes,
@@ -367,13 +374,11 @@ do_triangle_ccw(struct lp_setup_context *setup,
 
    /* Setup parameter interpolants:
     */
-   setup->setup.variant->jit_function( v0,
-				       v1,
-				       v2,
-				       frontfacing,
-				       GET_A0(&tri->inputs),
-				       GET_DADX(&tri->inputs),
-				       GET_DADY(&tri->inputs) );
+   setup->setup.variant->jit_function(v0, v1, v2,
+                                      frontfacing,
+                                      GET_A0(&tri->inputs),
+                                      GET_DADX(&tri->inputs),
+                                      GET_DADY(&tri->inputs));
 
    tri->inputs.frontfacing = frontfacing;
    tri->inputs.disable = FALSE;
@@ -383,9 +388,9 @@ do_triangle_ccw(struct lp_setup_context *setup,
 
    if (0)
       lp_dump_setup_coef(&setup->setup.variant->key,
-			 (const float (*)[4])GET_A0(&tri->inputs),
-			 (const float (*)[4])GET_DADX(&tri->inputs),
-			 (const float (*)[4])GET_DADY(&tri->inputs));
+                         (const float (*)[4])GET_A0(&tri->inputs),
+                         (const float (*)[4])GET_DADX(&tri->inputs),
+                         (const float (*)[4])GET_DADY(&tri->inputs));
 
    plane = GET_PLANES(tri);
 
@@ -672,29 +677,44 @@ do_triangle_ccw(struct lp_setup_context *setup,
     * Note that otherwise, the scissor planes only vary in 'C' value,
     * and even then only on state-changes.  Could alternatively store
     * these planes elsewhere.
+    * (Or only store the c value together with a bit indicating which
+    * scissor edge this is, so rasterization would treat them differently
+    * (easier to evaluate) to ordinary planes.)
     */
-   if (nr_planes == 7) {
-      const struct u_rect *scissor = &setup->scissors[viewport_index];
-
-      plane[3].dcdx = -1 << 8;
-      plane[3].dcdy = 0;
-      plane[3].c = (1-scissor->x0) << 8;
-      plane[3].eo = 1 << 8;
-
-      plane[4].dcdx = 1 << 8;
-      plane[4].dcdy = 0;
-      plane[4].c = (scissor->x1+1) << 8;
-      plane[4].eo = 0;
-
-      plane[5].dcdx = 0;
-      plane[5].dcdy = 1 << 8;
-      plane[5].c = (1-scissor->y0) << 8;
-      plane[5].eo = 1 << 8;
-
-      plane[6].dcdx = 0;
-      plane[6].dcdy = -1 << 8;
-      plane[6].c = (scissor->y1+1) << 8;
-      plane[6].eo = 0;
+   if (nr_planes > 3) {
+      /* why not just use draw_regions */
+      struct u_rect *scissor = &setup->scissors[viewport_index];
+      struct lp_rast_plane *plane_s = &plane[3];
+
+      if (bbox.x0 < scissor->x0) {
+         plane_s->dcdx = -1 << 8;
+         plane_s->dcdy = 0;
+         plane_s->c = (1-scissor->x0) << 8;
+         plane_s->eo = 1 << 8;
+         plane_s++;
+      }
+      if (bbox.x1 > scissor->x1) {
+         plane_s->dcdx = 1 << 8;
+         plane_s->dcdy = 0;
+         plane_s->c = (scissor->x1+1) << 8;
+         plane_s->eo = 0 << 8;
+         plane_s++;
+      }
+      if (bbox.y0 < scissor->y0) {
+         plane_s->dcdx = 0;
+         plane_s->dcdy = 1 << 8;
+         plane_s->c = (1-scissor->y0) << 8;
+         plane_s->eo = 1 << 8;
+         plane_s++;
+      }
+      if (bbox.y1 > scissor->y1) {
+         plane_s->dcdx = 0;
+         plane_s->dcdy = -1 << 8;
+         plane_s->c = (scissor->y1+1) << 8;
+         plane_s->eo = 0;
+         plane_s++;
+      }
+      assert(plane_s == &plane[nr_planes]);
    }
 
    return lp_setup_bin_triangle(setup, tri, &bbox, nr_planes, viewport_index);