llvmpipe: introduce tri_3_4 for tiny triangles

6 files changed, 127 insertions, 46 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c
index 36068d74be2..d7e6415e139 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast.c
@@ -579,6 +579,7 @@ static lp_rast_cmd_func dispatch[LP_RAST_OP_MAX] =
    lp_rast_triangle_6,
    lp_rast_triangle_7,
    lp_rast_triangle_8,
+   lp_rast_triangle_3_4,
    lp_rast_triangle_3_16,
    lp_rast_shade_tile,
    lp_rast_shade_tile_opaque,
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h
index 3c8dae6b01f..57676679354 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast.h
@@ -236,13 +236,14 @@ lp_rast_arg_null( void )
 #define LP_RAST_OP_TRIANGLE_6        0x7
 #define LP_RAST_OP_TRIANGLE_7        0x8
 #define LP_RAST_OP_TRIANGLE_8        0x9
-#define LP_RAST_OP_TRIANGLE_3_16     0xa
-#define LP_RAST_OP_SHADE_TILE        0xb
-#define LP_RAST_OP_SHADE_TILE_OPAQUE 0xc
-#define LP_RAST_OP_BEGIN_QUERY       0xd
-#define LP_RAST_OP_END_QUERY         0xe
-
-#define LP_RAST_OP_MAX               0xf
+#define LP_RAST_OP_TRIANGLE_3_4      0xa
+#define LP_RAST_OP_TRIANGLE_3_16     0xb
+#define LP_RAST_OP_SHADE_TILE        0xc
+#define LP_RAST_OP_SHADE_TILE_OPAQUE 0xd
+#define LP_RAST_OP_BEGIN_QUERY       0xe
+#define LP_RAST_OP_END_QUERY         0xf
+
+#define LP_RAST_OP_MAX               0x10
 #define LP_RAST_OP_MASK              0xff
 
 void
@@ -252,4 +253,5 @@ lp_debug_draw_bins_by_cmd_length( struct lp_scene *scene );
 void
 lp_debug_draw_bins_by_coverage( struct lp_scene *scene );
 
+
 #endif
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_debug.c b/src/gallium/drivers/llvmpipe/lp_rast_debug.c
index f2ef21f2a98..9fc78645a3a 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_debug.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast_debug.c
@@ -40,6 +40,7 @@ static const char *cmd_names[LP_RAST_OP_MAX] =
    "triangle_6",
    "triangle_7",
    "triangle_8",
+   "triangle_3_4",
    "triangle_3_16",
    "shade_tile",
    "shade_tile_opaque",
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
index 3bcdfd6074b..7370119e966 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
@@ -288,9 +288,11 @@ void lp_rast_triangle_7( struct lp_rasterizer_task *,
 void lp_rast_triangle_8( struct lp_rasterizer_task *, 
                          const union lp_rast_cmd_arg );
 
+void lp_rast_triangle_3_4(struct lp_rasterizer_task *,
+			  const union lp_rast_cmd_arg );
+
 void lp_rast_triangle_3_16( struct lp_rasterizer_task *, 
                             const union lp_rast_cmd_arg );
-
 void
 lp_debug_bin( const struct cmd_bin *bin );
 
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri.c b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
index ab2d7663184..ee5cd71b74a 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
@@ -112,6 +112,23 @@ build_masks(int c,
    *partmask |= build_mask_linear(c + cdiff, dcdx, dcdy);
 }
 
+void
+lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
+                      const union lp_rast_cmd_arg arg)
+{
+   union lp_rast_cmd_arg arg2;
+   arg2.tri = arg.tri;
+   arg2.mask = (1<<3)-1;
+   lp_rast_triangle_3(task, arg2);
+}
+
+void
+lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
+                      const union lp_rast_cmd_arg arg)
+{
+   lp_rast_triangle_3_16(task, arg);
+}
+
 #else
 #include <emmintrin.h>
 #include "util/u_sse.h"
@@ -189,44 +206,6 @@ build_mask_linear(int c, int dcdx, int dcdy)
    return _mm_movemask_epi8(result);
 }
 
-
-#endif
-
-
-
-
-#define TAG(x) x##_1
-#define NR_PLANES 1
-#include "lp_rast_tri_tmp.h"
-
-#define TAG(x) x##_2
-#define NR_PLANES 2
-#include "lp_rast_tri_tmp.h"
-
-#define TAG(x) x##_3
-#define NR_PLANES 3
-#include "lp_rast_tri_tmp.h"
-
-#define TAG(x) x##_4
-#define NR_PLANES 4
-#include "lp_rast_tri_tmp.h"
-
-#define TAG(x) x##_5
-#define NR_PLANES 5
-#include "lp_rast_tri_tmp.h"
-
-#define TAG(x) x##_6
-#define NR_PLANES 6
-#include "lp_rast_tri_tmp.h"
-
-#define TAG(x) x##_7
-#define NR_PLANES 7
-#include "lp_rast_tri_tmp.h"
-
-#define TAG(x) x##_8
-#define NR_PLANES 8
-#include "lp_rast_tri_tmp.h"
-
 static INLINE unsigned
 sign_bits4(const __m128i *cstep, int cdiff)
 {
@@ -342,3 +321,87 @@ lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
       block_full_4(task, tri, px, py);
    }
 }
+
+
+void
+lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
+		     const union lp_rast_cmd_arg arg)
+{
+   const struct lp_rast_triangle *tri = arg.triangle.tri;
+   const struct lp_rast_plane *plane = tri->plane;
+   unsigned mask = arg.triangle.plane_mask;
+   const int x = task->x + (mask & 0xff);
+   const int y = task->y + (mask >> 8);
+   unsigned j;
+
+   /* Iterate over partials:
+    */
+   {
+      unsigned mask = 0xffff;
+
+      for (j = 0; j < 3; j++) {
+	 const int cx = (plane[j].c 
+			 - plane[j].dcdx * x
+			 + plane[j].dcdy * y);
+
+	 const int dcdx = -plane[j].dcdx;
+	 const int dcdy = plane[j].dcdy;
+	 __m128i xdcdy = _mm_set1_epi32(dcdy);
+
+	 __m128i cstep0 = _mm_setr_epi32(cx, cx + dcdx, cx + dcdx*2, cx + dcdx*3);
+	 __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
+	 __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
+	 __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
+
+	 __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
+	 __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
+	 __m128i result = _mm_packs_epi16(cstep01, cstep23);
+
+	 /* Extract the sign bits
+	  */
+	 mask &= ~_mm_movemask_epi8(result);
+      }
+
+      if (mask)
+	 lp_rast_shade_quads_mask(task, &tri->inputs, x, y, mask);
+   }
+}
+
+
+#endif
+
+
+
+
+#define TAG(x) x##_1
+#define NR_PLANES 1
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_2
+#define NR_PLANES 2
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_3
+#define NR_PLANES 3
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_4
+#define NR_PLANES 4
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_5
+#define NR_PLANES 5
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_6
+#define NR_PLANES 6
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_7
+#define NR_PLANES 7
+#include "lp_rast_tri_tmp.h"
+
+#define TAG(x) x##_8
+#define NR_PLANES 8
+#include "lp_rast_tri_tmp.h"
+
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_tri.c b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
index d4ef8f4c9cd..2cd10ed4201 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
@@ -487,6 +487,18 @@ lp_setup_bin_triangle( struct lp_setup_context *setup,
 		     (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);