X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fllvmpipe%2Flp_rast_tri_tmp.h;h=52f6e9996839f2e46a0f7732b828daf0fbae5c19;hb=2f2d1b3d9b090aeba316d6c425c23e92340b5502;hp=905f3df21353f744ff3ebad3e8e13b51729940aa;hpb=ee0d1c29eeddfa364a18783507acd4d031029ba2;p=mesa.git

diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h b/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h
index 905f3df2135..52f6e999683 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h
@@ -32,7 +32,7 @@
 
 
 /**
- * Prototype for a 7 plane rasterizer function.  Will codegenerate
+ * Prototype for a 8 plane rasterizer function.  Will codegenerate
  * several of these.
  *
  * XXX: Varients for more/fewer planes.
@@ -44,15 +44,15 @@ TAG(do_block_4)(struct lp_rasterizer_task *task,
                 const struct lp_rast_triangle *tri,
                 const struct lp_rast_plane *plane,
                 int x, int y,
-                const int *c)
+                const int64_t *c)
 {
    unsigned mask = 0xffff;
    int j;
 
    for (j = 0; j < NR_PLANES; j++) {
-      mask &= ~build_mask(c[j] - 1, 
-			  -plane[j].dcdx,
-			  plane[j].dcdy);
+      mask &= ~BUILD_MASK_LINEAR(c[j] - 1, 
+				 -plane[j].dcdx,
+				 plane[j].dcdy);
    }
 
    /* Now pass to the shader:
@@ -70,7 +70,7 @@ TAG(do_block_16)(struct lp_rasterizer_task *task,
                  const struct lp_rast_triangle *tri,
                  const struct lp_rast_plane *plane,
                  int x, int y,
-                 const int *c)
+                 const int64_t *c)
 {
    unsigned outmask, inmask, partmask, partial_mask;
    unsigned j;
@@ -79,13 +79,17 @@ TAG(do_block_16)(struct lp_rasterizer_task *task,
    partmask = 0;                /* outside one or more trivial accept planes */
 
    for (j = 0; j < NR_PLANES; j++) {
-      const int dcdx = -plane[j].dcdx * 4;
-      const int dcdy = plane[j].dcdy * 4;
-      const int cox = c[j] + plane[j].eo * 4;
-      const int cio = c[j] + plane[j].ei * 4 - 1;
-
-      outmask |= build_mask(cox, dcdx, dcdy);
-      partmask |= build_mask(cio, dcdx, dcdy);
+      const int64_t dcdx = -IMUL64(plane[j].dcdx, 4);
+      const int64_t dcdy = IMUL64(plane[j].dcdy, 4);
+      const int64_t cox = IMUL64(plane[j].eo, 4);
+      const int64_t ei = plane[j].dcdy - plane[j].dcdx - plane[j].eo;
+      const int64_t cio = IMUL64(ei, 4) - 1;
+
+      BUILD_MASKS(c[j] + cox,
+		  cio - cox,
+		  dcdx, dcdy, 
+		  &outmask,   /* sign bits from c[i][0..15] + cox */
+		  &partmask); /* sign bits from c[i][0..15] + cio */
    }
 
    if (outmask == 0xffff)
@@ -102,19 +106,27 @@ TAG(do_block_16)(struct lp_rasterizer_task *task,
 
    assert((partial_mask & inmask) == 0);
 
+   LP_COUNT_ADD(nr_empty_4, util_bitcount(0xffff & ~(partial_mask | inmask)));
+
    /* Iterate over partials:
     */
    while (partial_mask) {
       int i = ffs(partial_mask) - 1;
-      int px = x + pos_table4[i][0];
-      int py = y + pos_table4[i][1];
-      int cx[NR_PLANES];
-
-      for (j = 0; j < NR_PLANES; j++)
-         cx[j] = c[j] + plane[j].step[i] * 4;
+      int ix = (i & 3) * 4;
+      int iy = (i >> 2) * 4;
+      int px = x + ix;
+      int py = y + iy; 
+      int64_t cx[NR_PLANES];
 
       partial_mask &= ~(1 << i);
 
+      LP_COUNT(nr_partially_covered_4);
+
+      for (j = 0; j < NR_PLANES; j++)
+         cx[j] = (c[j] 
+                  - IMUL64(plane[j].dcdx, ix)
+                  + IMUL64(plane[j].dcdy, iy));
+
       TAG(do_block_4)(task, tri, plane, px, py, cx);
    }
 
@@ -122,11 +134,14 @@ TAG(do_block_16)(struct lp_rasterizer_task *task,
     */
    while (inmask) {
       int i = ffs(inmask) - 1;
-      int px = x + pos_table4[i][0];
-      int py = y + pos_table4[i][1];
+      int ix = (i & 3) * 4;
+      int iy = (i >> 2) * 4;
+      int px = x + ix;
+      int py = y + iy; 
 
       inmask &= ~(1 << i);
 
+      LP_COUNT(nr_fully_covered_4);
       block_full_4(task, tri, px, py);
    }
 }
@@ -142,35 +157,42 @@ TAG(lp_rast_triangle)(struct lp_rasterizer_task *task,
 {
    const struct lp_rast_triangle *tri = arg.triangle.tri;
    unsigned plane_mask = arg.triangle.plane_mask;
+   const struct lp_rast_plane *tri_plane = GET_PLANES(tri);
    const int x = task->x, y = task->y;
    struct lp_rast_plane plane[NR_PLANES];
-   int c[NR_PLANES];
+   int64_t c[NR_PLANES];
    unsigned outmask, inmask, partmask, partial_mask;
-   unsigned j, nr_planes = 0;
+   unsigned j = 0;
 
-   while (plane_mask) {
-      int i = ffs(plane_mask) - 1;
-      plane[nr_planes] = tri->plane[i];
-      plane_mask &= ~(1 << i);
-      nr_planes++;
-   };
+   if (tri->inputs.disable) {
+      /* This triangle was partially binned and has been disabled */
+      return;
+   }
 
-   assert(nr_planes == NR_PLANES);
    outmask = 0;                 /* outside one or more trivial reject planes */
    partmask = 0;                /* outside one or more trivial accept planes */
 
-   for (j = 0; j < NR_PLANES; j++) {
-      c[j] = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
-   }
-
-   for (j = 0; j < NR_PLANES; j++) {
-      const int dcdx = -plane[j].dcdx * 16;
-      const int dcdy = plane[j].dcdy * 16;
-      const int cox = c[j] + plane[j].eo * 16;
-      const int cio = c[j] + plane[j].ei * 16 - 1;
-
-      outmask |= build_mask(cox, dcdx, dcdy);
-      partmask |= build_mask(cio, dcdx, dcdy);
+   while (plane_mask) {
+      int i = ffs(plane_mask) - 1;
+      plane[j] = tri_plane[i];
+      plane_mask &= ~(1 << i);
+      c[j] = plane[j].c + IMUL64(plane[j].dcdy, y) - IMUL64(plane[j].dcdx, x);
+
+      {
+         const int64_t dcdx = -IMUL64(plane[j].dcdx, 16);
+         const int64_t dcdy = IMUL64(plane[j].dcdy, 16);
+         const int64_t cox = IMUL64(plane[j].eo, 16);
+         const int64_t ei = plane[j].dcdy - plane[j].dcdx - plane[j].eo;
+         const int64_t cio = IMUL64(ei, 16) - 1;
+
+         BUILD_MASKS(c[j] + cox,
+                     cio - cox,
+                     dcdx, dcdy,
+                     &outmask,   /* sign bits from c[i][0..15] + cox */
+                     &partmask); /* sign bits from c[i][0..15] + cio */
+      }
+
+      j++;
    }
 
    if (outmask == 0xffff)
@@ -187,16 +209,22 @@ TAG(lp_rast_triangle)(struct lp_rasterizer_task *task,
 
    assert((partial_mask & inmask) == 0);
 
+   LP_COUNT_ADD(nr_empty_16, util_bitcount(0xffff & ~(partial_mask | inmask)));
+
    /* Iterate over partials:
     */
    while (partial_mask) {
       int i = ffs(partial_mask) - 1;
-      int px = x + pos_table16[i][0];
-      int py = y + pos_table16[i][1];
-      int cx[NR_PLANES];
+      int ix = (i & 3) * 16;
+      int iy = (i >> 2) * 16;
+      int px = x + ix;
+      int py = y + iy;
+      int64_t cx[NR_PLANES];
 
       for (j = 0; j < NR_PLANES; j++)
-         cx[j] = c[j] + plane[j].step[i] * 16;
+         cx[j] = (c[j]
+                  - IMUL64(plane[j].dcdx, ix)
+                  + IMUL64(plane[j].dcdy, iy));
 
       partial_mask &= ~(1 << i);
 
@@ -208,8 +236,10 @@ TAG(lp_rast_triangle)(struct lp_rasterizer_task *task,
     */
    while (inmask) {
       int i = ffs(inmask) - 1;
-      int px = x + pos_table16[i][0];
-      int py = y + pos_table16[i][1];
+      int ix = (i & 3) * 16;
+      int iy = (i >> 2) * 16;
+      int px = x + ix;
+      int py = y + iy;
 
       inmask &= ~(1 << i);
 
@@ -218,6 +248,133 @@ TAG(lp_rast_triangle)(struct lp_rasterizer_task *task,
    }
 }
 
+#if defined(PIPE_ARCH_SSE) && defined(TRI_16)
+/* XXX: special case this when intersection is not required.
+ *      - tile completely within bbox,
+ *      - bbox completely within tile.
+ */
+void
+TRI_16(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 = GET_PLANES(tri);
+   unsigned mask = arg.triangle.plane_mask;
+   unsigned outmask, partial_mask;
+   unsigned j;
+   __m128i cstep4[NR_PLANES][4];
+
+   int x = (mask & 0xff);
+   int y = (mask >> 8);
+
+   outmask = 0;                 /* outside one or more trivial reject planes */
+   
+   x += task->x;
+   y += task->y;
+
+   for (j = 0; j < NR_PLANES; j++) {
+      const int dcdx = -plane[j].dcdx * 4;
+      const int dcdy = plane[j].dcdy * 4;
+      __m128i xdcdy = _mm_set1_epi32(dcdy);
+
+      cstep4[j][0] = _mm_setr_epi32(0, dcdx, dcdx*2, dcdx*3);
+      cstep4[j][1] = _mm_add_epi32(cstep4[j][0], xdcdy);
+      cstep4[j][2] = _mm_add_epi32(cstep4[j][1], xdcdy);
+      cstep4[j][3] = _mm_add_epi32(cstep4[j][2], xdcdy);
+
+      {
+	 const int c = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
+	 const int cox = plane[j].eo * 4;
+
+	 outmask |= sign_bits4(cstep4[j], c + cox);
+      }
+   }
+
+   if (outmask == 0xffff)
+      return;
+
+
+   /* Mask of sub-blocks which are inside all trivial reject planes,
+    * but outside at least one trivial accept plane:
+    */
+   partial_mask = 0xffff & ~outmask;
+
+   /* Iterate over partials:
+    */
+   while (partial_mask) {
+      int i = ffs(partial_mask) - 1;
+      int ix = (i & 3) * 4;
+      int iy = (i >> 2) * 4;
+      int px = x + ix;
+      int py = y + iy; 
+      unsigned mask = 0xffff;
+
+      partial_mask &= ~(1 << i);
+
+      for (j = 0; j < NR_PLANES; j++) {
+         const int cx = (plane[j].c - 1
+			 - plane[j].dcdx * px
+			 + plane[j].dcdy * py) * 4;
+
+	 mask &= ~sign_bits4(cstep4[j], cx);
+      }
+
+      if (mask)
+	 lp_rast_shade_quads_mask(task, &tri->inputs, px, py, mask);
+   }
+}
+#endif
+
+#if defined(PIPE_ARCH_SSE) && defined(TRI_4)
+void
+TRI_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 = GET_PLANES(tri);
+   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 < NR_PLANES; 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
+
+
+
 #undef TAG
+#undef TRI_4
+#undef TRI_16
 #undef NR_PLANES