GPU Buffers:

This commit begins implementation of the idea about hidden face
separation outlined in

http://code.blender.org/2015/06/optimizing-blenders-real-time-mesh-

We split hidden and visible faces to different parts of the triangle
buffer.
Mapped drawing will now skip iterating through hidden polys.
Of course the final target, when all derived mesh types use
VBO sorting, is to skip checking for hide flag per face
completely. All faces will be pre-sorted anyway and we'll
be able to draw them with one draw call.

1 files changed, 58 insertions, 36 deletions

diff --git a/source/blender/blenkernel/intern/cdderivedmesh.c b/source/blender/blenkernel/intern/cdderivedmesh.c
index 060ffff5a12..8f04c191481 100644
--- a/source/blender/blenkernel/intern/cdderivedmesh.c
+++ b/source/blender/blenkernel/intern/cdderivedmesh.c
@@ -719,7 +719,7 @@ static void cdDM_drawMappedFaces(
 			GPUBufferMaterial *bufmat = dm->drawObject->materials + mat_index;
 			DMDrawOption draw_option = DM_DRAW_OPTION_NORMAL;
 			int next_actualFace = bufmat->polys[0];
-			totpoly = bufmat->totpolys;
+			totpoly = useHide ? bufmat->totvisiblepolys : bufmat->totpolys;
 
 			tot_element = 0;
 			start_element = 0;
@@ -1229,32 +1229,70 @@ static void cdDM_drawMappedEdges(DerivedMesh *dm, DMSetDrawOptions setDrawOption
 	glEnd();
 }
 
+typedef struct FaceCount {
+	unsigned int i_visible;
+	unsigned int i_hidden;
+	unsigned int i_tri_visible;
+	unsigned int i_tri_hidden;
+} FaceCount;
+
 static void cdDM_buffer_copy_triangles(
         DerivedMesh *dm, unsigned int *varray,
         const int *mat_orig_to_new)
 {
-	GPUBufferMaterial *gpumat;
-	int i, findex = 0;
+	GPUBufferMaterial *gpumat, *gpumaterials = dm->drawObject->materials;
+	int i, j, start;
 
-	const MPoly *mpoly;
+	const int totmat = dm->drawObject->totmaterial;
+	const MPoly *mpoly = dm->getPolyArray(dm);
 	const MLoopTri *lt = dm->getLoopTriArray(dm);
-	const int tottri = dm->getNumLoopTri(dm);
+	const int totpoly = dm->getNumPolys(dm);
 
-	mpoly = dm->getPolyArray(dm);
+	FaceCount *fc = MEM_mallocN(sizeof(*fc) * totmat, "gpumaterial.facecount");
 
-	for (i = 0; i < tottri; i++, lt++) {
-		int start;
-		gpumat = dm->drawObject->materials + mat_orig_to_new[mpoly[lt->poly].mat_nr];
-		start = gpumat->counter;
+	for (i = 0; i < totmat; i++) {
+		fc[i].i_visible = 0;
+		fc[i].i_tri_visible = 0;
+		fc[i].i_hidden = gpumaterials[i].totpolys - 1;
+		fc[i].i_tri_hidden = gpumaterials[i].totelements - 1;
+	}
 
-		/* v1 v2 v3 */
-		varray[start++] = lt->tri[0];
-		varray[start++] = lt->tri[1];
-		varray[start++] = lt->tri[2];
+	for (i = 0; i < totpoly; i++) {
+		int tottri = ME_POLY_TRI_TOT(&mpoly[i]);
+		int mati = mat_orig_to_new[mpoly[i].mat_nr];
+		gpumat = gpumaterials + mati;
 
-		gpumat->counter += 3;
-		findex += 3;
+		if (mpoly[i].flag & ME_HIDE) {
+			for (j = 0; j < tottri; j++, lt++) {
+				start = gpumat->start + fc[mati].i_tri_hidden;
+				/* v1 v2 v3 */
+				varray[start--] = lt->tri[2];
+				varray[start--] = lt->tri[1];
+				varray[start--] = lt->tri[0];
+				fc[mati].i_tri_hidden -= 3;
+			}
+			gpumat->polys[fc[mati].i_hidden--] = i;
+		}
+		else {
+			for (j = 0; j < tottri; j++, lt++) {
+				start = gpumat->start + fc[mati].i_tri_visible;
+				/* v1 v2 v3 */
+				varray[start++] = lt->tri[0];
+				varray[start++] = lt->tri[1];
+				varray[start++] = lt->tri[2];
+				fc[mati].i_tri_visible += 3;
+			}
+			gpumat->polys[fc[mati].i_visible++] = i;
+		}
+	}
+
+	/* set the visible polygons */
+	for (i = 0; i < totmat; i++) {
+		gpumaterials[i].totvisiblepolys = fc[i].i_visible;
 	}
+
+
+	MEM_freeN(fc);
 }
 
 static void cdDM_buffer_copy_vertex(
@@ -1594,15 +1632,11 @@ static void cdDM_drawobject_add_vert_point(GPUDrawObject *gdo, int vert_index, i
 static void cdDM_drawobject_init_vert_points(
         GPUDrawObject *gdo,
         const MPoly *mpoly, const MLoop *mloop,
-        int tot_poly,
-        int totmat)
+        int tot_poly)
 {
-	GPUBufferMaterial *mat;
-	int i, *mat_orig_to_new;
+	int i;
 	int tot_loops = 0;
 
-	mat_orig_to_new = MEM_callocN(sizeof(*mat_orig_to_new) * totmat,
-	                                             "GPUDrawObject.mat_orig_to_new");
 	/* allocate the array and space for links */
 	gdo->vert_points = MEM_mallocN(sizeof(GPUVertPointLink) * gdo->totvert,
 	                               "GPUDrawObject.vert_points");
@@ -1612,13 +1646,6 @@ static void cdDM_drawobject_init_vert_points(
 	gdo->vert_points_usage = 0;
 #endif
 
-	/* build a map from the original material indices to the new
-	 * GPUBufferMaterial indices */
-	for (i = 0; i < gdo->totmaterial; i++) {
-		mat_orig_to_new[gdo->materials[i].mat_nr] = i;
-		gdo->materials[i].counter = 0;
-	}
-
 	/* -1 indicates the link is not yet used */
 	for (i = 0; i < gdo->totvert; i++) {
 #ifdef USE_GPU_POINT_LINK
@@ -1630,9 +1657,6 @@ static void cdDM_drawobject_init_vert_points(
 	for (i = 0; i < tot_poly; i++) {
 		int j;
 		const MPoly *mp = &mpoly[i];
-		mat = &gdo->materials[mat_orig_to_new[mp->mat_nr]];
-
-		mat->polys[mat->counter++] = i;
 
 		/* assign unique indices to vertices of the mesh */
 		for (j = 0; j < mp->totloop; j++) {
@@ -1648,8 +1672,6 @@ static void cdDM_drawobject_init_vert_points(
 			gdo->tot_loose_point++;
 		}
 	}
-
-	MEM_freeN(mat_orig_to_new);
 }
 
 /* see GPUDrawObject's structure definition for a description of the
@@ -1692,9 +1714,9 @@ static GPUDrawObject *cdDM_GPUobject_new(DerivedMesh *dm)
 	gdo->tot_loop_verts = totloops;
 
 	/* store total number of points used for triangles */
-	gdo->tot_triangle_point = dm->getNumLoopTri(dm) * 3;
+	gdo->tot_triangle_point = poly_to_tri_count(totpolys, totloops) * 3;
 
-	cdDM_drawobject_init_vert_points(gdo, mpoly, mloop, totpolys, totmat);
+	cdDM_drawobject_init_vert_points(gdo, mpoly, mloop, totpolys);
 
 	return gdo;
 }