Updated meshoptimizer.

3rdparty/meshoptimizer/src/clusterizer.cpp

@@ -13,6 +13,12 @@
 namespace meshopt
 {
 
+// This must be <= 255 since index 0xff is used internally to indice a vertex that doesn't belong to a meshlet
+const size_t kMeshletMaxVertices = 255;
+
+// A reasonable limit is around 2*max_vertices or less
+const size_t kMeshletMaxTriangles = 512;
+
 struct TriangleAdjacency2
 {
 	unsigned int* counts;
@@ -215,7 +221,16 @@ static float computeTriangleCones(Cone* triangles, const unsigned int* indices,
 	return mesh_area;
 }
 
-static bool appendMeshlet(meshopt_Meshlet& meshlet, unsigned int a, unsigned int b, unsigned int c, unsigned char* used, meshopt_Meshlet* destination, size_t offset, size_t max_vertices, size_t max_triangles)
+static void finishMeshlet(meshopt_Meshlet& meshlet, unsigned char* meshlet_triangles)
+{
+	size_t offset = meshlet.triangle_offset + meshlet.triangle_count * 3;
+
+	// fill 4b padding with 0
+	while (offset & 3)
+		meshlet_triangles[offset++] = 0;
+}
+
+static bool appendMeshlet(meshopt_Meshlet& meshlet, unsigned int a, unsigned int b, unsigned int c, unsigned char* used, meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, size_t meshlet_offset, size_t max_vertices, size_t max_triangles)
 {
 	unsigned char& av = used[a];
 	unsigned char& bv = used[b];
@@ -227,37 +242,42 @@ static bool appendMeshlet(meshopt_Meshlet& meshlet, unsigned int a, unsigned int
 
 	if (meshlet.vertex_count + used_extra > max_vertices || meshlet.triangle_count >= max_triangles)
 	{
-		destination[offset] = meshlet;
+		meshlets[meshlet_offset] = meshlet;
 
 		for (size_t j = 0; j < meshlet.vertex_count; ++j)
-			used[meshlet.vertices[j]] = 0xff;
+			used[meshlet_vertices[meshlet.vertex_offset + j]] = 0xff;
+
+		finishMeshlet(meshlet, meshlet_triangles);
 
-		memset(&meshlet, 0, sizeof(meshlet));
+		meshlet.vertex_offset += meshlet.vertex_count;
+		meshlet.triangle_offset += (meshlet.triangle_count * 3 + 3) & ~3; // 4b padding
+		meshlet.vertex_count = 0;
+		meshlet.triangle_count = 0;
 
 		result = true;
 	}
 
 	if (av == 0xff)
 	{
-		av = meshlet.vertex_count;
-		meshlet.vertices[meshlet.vertex_count++] = a;
+		av = (unsigned char)meshlet.vertex_count;
+		meshlet_vertices[meshlet.vertex_offset + meshlet.vertex_count++] = a;
 	}
 
 	if (bv == 0xff)
 	{
-		bv = meshlet.vertex_count;
-		meshlet.vertices[meshlet.vertex_count++] = b;
+		bv = (unsigned char)meshlet.vertex_count;
+		meshlet_vertices[meshlet.vertex_offset + meshlet.vertex_count++] = b;
 	}
 
 	if (cv == 0xff)
 	{
-		cv = meshlet.vertex_count;
-		meshlet.vertices[meshlet.vertex_count++] = c;
+		cv = (unsigned char)meshlet.vertex_count;
+		meshlet_vertices[meshlet.vertex_offset + meshlet.vertex_count++] = c;
 	}
 
-	meshlet.indices[meshlet.triangle_count][0] = av;
-	meshlet.indices[meshlet.triangle_count][1] = bv;
-	meshlet.indices[meshlet.triangle_count][2] = cv;
+	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 0] = av;
+	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 1] = bv;
+	meshlet_triangles[meshlet.triangle_offset + meshlet.triangle_count * 3 + 2] = cv;
 	meshlet.triangle_count++;
 
 	return result;
@@ -348,7 +368,8 @@ static size_t kdtreeBuild(size_t offset, KDNode* nodes, size_t node_count, const
 	}
 
 	// split axis is one where the variance is largest
-	unsigned int axis = vars[0] >= vars[1] && vars[0] >= vars[2] ? 0 : vars[1] >= vars[2] ? 1 : 2;
+	unsigned int axis = vars[0] >= vars[1] && vars[0] >= vars[2] ? 0 : vars[1] >= vars[2] ? 1
+	                                                                                      : 2;
 
 	float split = mean[axis];
 	size_t middle = kdtreePartition(indices, count, points, stride, axis, split);
@@ -419,9 +440,15 @@ static void kdtreeNearest(KDNode* nodes, unsigned int root, const float* points,
 
 size_t meshopt_buildMeshletsBound(size_t index_count, size_t max_vertices, size_t max_triangles)
 {
+	using namespace meshopt;
+
 	assert(index_count % 3 == 0);
-	assert(max_vertices >= 3);
-	assert(max_triangles >= 1);
+	assert(max_vertices >= 3 && max_vertices <= kMeshletMaxVertices);
+	assert(max_triangles >= 1 && max_triangles <= kMeshletMaxTriangles);
+	assert(max_triangles % 4 == 0); // ensures the caller will compute output space properly as index data is 4b aligned
+
+	(void)kMeshletMaxVertices;
+	(void)kMeshletMaxTriangles;
 
 	// meshlet construction is limited by max vertices and max triangles per meshlet
 	// the worst case is that the input is an unindexed stream since this equally stresses both limits
@@ -433,24 +460,19 @@ size_t meshopt_buildMeshletsBound(size_t index_count, size_t max_vertices, size_
 	return meshlet_limit_vertices > meshlet_limit_triangles ? meshlet_limit_vertices : meshlet_limit_triangles;
 }
 
-size_t meshopt_buildMeshlets(meshopt_Meshlet* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight)
+size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight)
 {
 	using namespace meshopt;
 
 	assert(index_count % 3 == 0);
-	assert(max_vertices >= 3);
-	assert(max_triangles >= 1);
 	assert(vertex_positions_stride > 0 && vertex_positions_stride <= 256);
 	assert(vertex_positions_stride % sizeof(float) == 0);
 
-	meshopt_Allocator allocator;
-
-	meshopt_Meshlet meshlet;
-	memset(&meshlet, 0, sizeof(meshlet));
+	assert(max_vertices >= 3 && max_vertices <= kMeshletMaxVertices);
+	assert(max_triangles >= 1 && max_triangles <= kMeshletMaxTriangles);
+	assert(max_triangles % 4 == 0); // ensures the caller will compute output space properly as index data is 4b aligned
 
-	assert(max_vertices <= sizeof(meshlet.vertices) / sizeof(meshlet.vertices[0]));
-	assert(max_vertices <= 255);
-	assert(max_triangles <= sizeof(meshlet.indices) / 3);
+	meshopt_Allocator allocator;
 
 	TriangleAdjacency2 adjacency = {};
 	buildTriangleAdjacency(adjacency, indices, index_count, vertex_count, allocator);
@@ -483,7 +505,8 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* destination, const unsigned int* i
 	unsigned char* used = allocator.allocate<unsigned char>(vertex_count);
 	memset(used, -1, vertex_count);
 
-	size_t offset = 0;
+	meshopt_Meshlet meshlet = {};
+	size_t meshlet_offset = 0;
 
 	Cone meshlet_cone_acc = {};
 
@@ -497,7 +520,7 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* destination, const unsigned int* i
 
 		for (size_t i = 0; i < meshlet.vertex_count; ++i)
 		{
-			unsigned int index = meshlet.vertices[i];
+			unsigned int index = meshlet_vertices[meshlet.vertex_offset + i];
 
 			unsigned int* neighbours = &adjacency.data[0] + adjacency.offsets[index];
 			size_t neighbours_size = adjacency.counts[index];
@@ -566,9 +589,9 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* destination, const unsigned int* i
 		assert(a < vertex_count && b < vertex_count && c < vertex_count);
 
 		// add meshlet to the output; when the current meshlet is full we reset the accumulated bounds
-		if (appendMeshlet(meshlet, a, b, c, used, destination, offset, max_vertices, max_triangles))
+		if (appendMeshlet(meshlet, a, b, c, used, meshlets, meshlet_vertices, meshlet_triangles, meshlet_offset, max_vertices, max_triangles))
 		{
-			offset++;
+			meshlet_offset++;
 			memset(&meshlet_cone_acc, 0, sizeof(meshlet_cone_acc));
 		}
 
@@ -610,35 +633,34 @@ size_t meshopt_buildMeshlets(meshopt_Meshlet* destination, const unsigned int* i
 	}
 
 	if (meshlet.triangle_count)
-		destination[offset++] = meshlet;
+	{
+		finishMeshlet(meshlet, meshlet_triangles);
 
-	assert(offset <= meshopt_buildMeshletsBound(index_count, max_vertices, max_triangles));
+		meshlets[meshlet_offset++] = meshlet;
+	}
 
-	return offset;
+	assert(meshlet_offset <= meshopt_buildMeshletsBound(index_count, max_vertices, max_triangles));
+	return meshlet_offset;
 }
 
-size_t meshopt_buildMeshletsScan(meshopt_Meshlet* destination, const unsigned int* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles)
+size_t meshopt_buildMeshletsScan(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles)
 {
 	using namespace meshopt;
 
 	assert(index_count % 3 == 0);
-	assert(max_vertices >= 3);
-	assert(max_triangles >= 1);
-
-	meshopt_Allocator allocator;
 
-	meshopt_Meshlet meshlet;
-	memset(&meshlet, 0, sizeof(meshlet));
+	assert(max_vertices >= 3 && max_vertices <= kMeshletMaxVertices);
+	assert(max_triangles >= 1 && max_triangles <= kMeshletMaxTriangles);
+	assert(max_triangles % 4 == 0); // ensures the caller will compute output space properly as index data is 4b aligned
 
-	assert(max_vertices <= sizeof(meshlet.vertices) / sizeof(meshlet.vertices[0]));
-	assert(max_vertices <= 255);
-	assert(max_triangles <= sizeof(meshlet.indices) / 3);
+	meshopt_Allocator allocator;
 
 	// index of the vertex in the meshlet, 0xff if the vertex isn't used
 	unsigned char* used = allocator.allocate<unsigned char>(vertex_count);
 	memset(used, -1, vertex_count);
 
-	size_t offset = 0;
+	meshopt_Meshlet meshlet = {};
+	size_t meshlet_offset = 0;
 
 	for (size_t i = 0; i < index_count; i += 3)
 	{
@@ -646,15 +668,18 @@ size_t meshopt_buildMeshletsScan(meshopt_Meshlet* destination, const unsigned in
 		assert(a < vertex_count && b < vertex_count && c < vertex_count);
 
 		// appends triangle to the meshlet and writes previous meshlet to the output if full
-		offset += appendMeshlet(meshlet, a, b, c, used, destination, offset, max_vertices, max_triangles);
+		meshlet_offset += appendMeshlet(meshlet, a, b, c, used, meshlets, meshlet_vertices, meshlet_triangles, meshlet_offset, max_vertices, max_triangles);
 	}
 
 	if (meshlet.triangle_count)
-		destination[offset++] = meshlet;
+	{
+		finishMeshlet(meshlet, meshlet_triangles);
 
-	assert(offset <= meshopt_buildMeshletsBound(index_count, max_vertices, max_triangles));
+		meshlets[meshlet_offset++] = meshlet;
+	}
 
-	return offset;
+	assert(meshlet_offset <= meshopt_buildMeshletsBound(index_count, max_vertices, max_triangles));
+	return meshlet_offset;
 }
 
 meshopt_Bounds meshopt_computeClusterBounds(const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
@@ -662,18 +687,17 @@ meshopt_Bounds meshopt_computeClusterBounds(const unsigned int* indices, size_t
 	using namespace meshopt;
 
 	assert(index_count % 3 == 0);
+	assert(index_count / 3 <= kMeshletMaxTriangles);
 	assert(vertex_positions_stride > 0 && vertex_positions_stride <= 256);
 	assert(vertex_positions_stride % sizeof(float) == 0);
 
-	assert(index_count / 3 <= 256);
-
 	(void)vertex_count;
 
 	size_t vertex_stride_float = vertex_positions_stride / sizeof(float);
 
 	// compute triangle normals and gather triangle corners
-	float normals[256][3];
-	float corners[256][3][3];
+	float normals[kMeshletMaxTriangles][3];
+	float corners[kMeshletMaxTriangles][3][3];
 	size_t triangles = 0;
 
 	for (size_t i = 0; i < index_count; i += 3)
@@ -811,25 +835,23 @@ meshopt_Bounds meshopt_computeClusterBounds(const unsigned int* indices, size_t
 	return bounds;
 }
 
-meshopt_Bounds meshopt_computeMeshletBounds(const meshopt_Meshlet* meshlet, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
+meshopt_Bounds meshopt_computeMeshletBounds(const unsigned int* meshlet_vertices, const unsigned char* meshlet_triangles, size_t triangle_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
 {
+	using namespace meshopt;
+
+	assert(triangle_count <= kMeshletMaxTriangles);
 	assert(vertex_positions_stride > 0 && vertex_positions_stride <= 256);
 	assert(vertex_positions_stride % sizeof(float) == 0);
 
-	unsigned int indices[sizeof(meshlet->indices) / sizeof(meshlet->indices[0][0])];
+	unsigned int indices[kMeshletMaxTriangles * 3];
 
-	for (size_t i = 0; i < meshlet->triangle_count; ++i)
+	for (size_t i = 0; i < triangle_count * 3; ++i)
 	{
-		unsigned int a = meshlet->vertices[meshlet->indices[i][0]];
-		unsigned int b = meshlet->vertices[meshlet->indices[i][1]];
-		unsigned int c = meshlet->vertices[meshlet->indices[i][2]];
-
-		assert(a < vertex_count && b < vertex_count && c < vertex_count);
+		unsigned int index = meshlet_vertices[meshlet_triangles[i]];
+		assert(index < vertex_count);
 
-		indices[i * 3 + 0] = a;
-		indices[i * 3 + 1] = b;
-		indices[i * 3 + 2] = c;
+		indices[i] = index;
 	}
 
-	return meshopt_computeClusterBounds(indices, meshlet->triangle_count * 3, vertex_positions, vertex_count, vertex_positions_stride);
+	return meshopt_computeClusterBounds(indices, triangle_count * 3, vertex_positions, vertex_count, vertex_positions_stride);
 }

3rdparty/meshoptimizer/src/indexgenerator.cpp

@@ -4,6 +4,8 @@
 #include <assert.h>
 #include <string.h>
 
+// This work is based on:
+// John McDonald, Mark Kilgard. Crack-Free Point-Normal Triangles using Adjacent Edge Normals. 2010
 namespace meshopt
 {
 
@@ -83,10 +85,49 @@ struct VertexStreamHasher
 	}
 };
 
+struct EdgeHasher
+{
+	const unsigned int* remap;
+
+	size_t hash(unsigned long long edge) const
+	{
+		unsigned int e0 = unsigned(edge >> 32);
+		unsigned int e1 = unsigned(edge);
+
+		unsigned int h1 = remap[e0];
+		unsigned int h2 = remap[e1];
+
+		const unsigned int m = 0x5bd1e995;
+
+		// MurmurHash64B finalizer
+		h1 ^= h2 >> 18;
+		h1 *= m;
+		h2 ^= h1 >> 22;
+		h2 *= m;
+		h1 ^= h2 >> 17;
+		h1 *= m;
+		h2 ^= h1 >> 19;
+		h2 *= m;
+
+		return h2;
+	}
+
+	bool equal(unsigned long long lhs, unsigned long long rhs) const
+	{
+		unsigned int l0 = unsigned(lhs >> 32);
+		unsigned int l1 = unsigned(lhs);
+
+		unsigned int r0 = unsigned(rhs >> 32);
+		unsigned int r1 = unsigned(rhs);
+
+		return remap[l0] == remap[r0] && remap[l1] == remap[r1];
+	}
+};
+
 static size_t hashBuckets(size_t count)
 {
 	size_t buckets = 1;
-	while (buckets < count)
+	while (buckets < count + count / 4)
 		buckets *= 2;
 
 	return buckets;
@@ -119,6 +160,26 @@ static T* hashLookup(T* table, size_t buckets, const Hash& hash, const T& key, c
 	return 0;
 }
 
+static void buildPositionRemap(unsigned int* remap, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, meshopt_Allocator& allocator)
+{
+	VertexHasher vertex_hasher = {reinterpret_cast<const unsigned char*>(vertex_positions), 3 * sizeof(float), vertex_positions_stride};
+
+	size_t vertex_table_size = hashBuckets(vertex_count);
+	unsigned int* vertex_table = allocator.allocate<unsigned int>(vertex_table_size);
+	memset(vertex_table, -1, vertex_table_size * sizeof(unsigned int));
+
+	for (size_t i = 0; i < vertex_count; ++i)
+	{
+		unsigned int index = unsigned(i);
+		unsigned int* entry = hashLookup(vertex_table, vertex_table_size, vertex_hasher, index, ~0u);
+
+		if (*entry == ~0u)
+			*entry = index;
+
+		remap[index] = *entry;
+	}
+}
+
 } // namespace meshopt
 
 size_t meshopt_generateVertexRemap(unsigned int* destination, const unsigned int* indices, size_t index_count, const void* vertices, size_t vertex_count, size_t vertex_size)
@@ -345,3 +406,146 @@ void meshopt_generateShadowIndexBufferMulti(unsigned int* destination, const uns
 		destination[i] = remap[index];
 	}
 }
+
+void meshopt_generateAdjacencyIndexBuffer(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
+{
+	using namespace meshopt;
+
+	assert(index_count % 3 == 0);
+	assert(vertex_positions_stride > 0 && vertex_positions_stride <= 256);
+	assert(vertex_positions_stride % sizeof(float) == 0);
+
+	meshopt_Allocator allocator;
+
+	static const int next[4] = {1, 2, 0, 1};
+
+	// build position remap: for each vertex, which other (canonical) vertex does it map to?
+	unsigned int* remap = allocator.allocate<unsigned int>(vertex_count);
+	buildPositionRemap(remap, vertex_positions, vertex_count, vertex_positions_stride, allocator);
+
+	// build edge set; this stores all triangle edges but we can look these up by any other wedge
+	EdgeHasher edge_hasher = {remap};
+
+	size_t edge_table_size = hashBuckets(index_count);
+	unsigned long long* edge_table = allocator.allocate<unsigned long long>(edge_table_size);
+	unsigned int* edge_vertex_table = allocator.allocate<unsigned int>(edge_table_size);
+
+	memset(edge_table, -1, edge_table_size * sizeof(unsigned long long));
+	memset(edge_vertex_table, -1, edge_table_size * sizeof(unsigned int));
+
+	for (size_t i = 0; i < index_count; i += 3)
+	{
+		for (int e = 0; e < 3; ++e)
+		{
+			unsigned int i0 = indices[i + e];
+			unsigned int i1 = indices[i + next[e]];
+			unsigned int i2 = indices[i + next[e + 1]];
+			assert(i0 < vertex_count && i1 < vertex_count && i2 < vertex_count);
+
+			unsigned long long edge = ((unsigned long long)i0 << 32) | i1;
+			unsigned long long* entry = hashLookup(edge_table, edge_table_size, edge_hasher, edge, ~0ull);
+
+			if (*entry == ~0ull)
+			{
+				*entry = edge;
+
+				// store vertex opposite to the edge
+				edge_vertex_table[entry - edge_table] = i2;
+			}
+		}
+	}
+
+	// build resulting index buffer: 6 indices for each input triangle
+	for (size_t i = 0; i < index_count; i += 3)
+	{
+		unsigned int patch[6];
+
+		for (int e = 0; e < 3; ++e)
+		{
+			unsigned int i0 = indices[i + e];
+			unsigned int i1 = indices[i + next[e]];
+			assert(i0 < vertex_count && i1 < vertex_count);
+
+			// note: this refers to the opposite edge!
+			unsigned long long edge = ((unsigned long long)i1 << 32) | i0;
+			unsigned long long* oppe = hashLookup(edge_table, edge_table_size, edge_hasher, edge, ~0ull);
+
+			patch[e * 2 + 0] = i0;
+			patch[e * 2 + 1] = (*oppe == ~0ull) ? i0 : edge_vertex_table[oppe - edge_table];
+		}
+
+		memcpy(destination + i * 2, patch, sizeof(patch));
+	}
+}
+
+void meshopt_generateTessellationIndexBuffer(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
+{
+	using namespace meshopt;
+
+	assert(index_count % 3 == 0);
+	assert(vertex_positions_stride > 0 && vertex_positions_stride <= 256);
+	assert(vertex_positions_stride % sizeof(float) == 0);
+
+	meshopt_Allocator allocator;
+
+	static const int next[3] = {1, 2, 0};
+
+	// build position remap: for each vertex, which other (canonical) vertex does it map to?
+	unsigned int* remap = allocator.allocate<unsigned int>(vertex_count);
+	buildPositionRemap(remap, vertex_positions, vertex_count, vertex_positions_stride, allocator);
+
+	// build edge set; this stores all triangle edges but we can look these up by any other wedge
+	EdgeHasher edge_hasher = {remap};
+
+	size_t edge_table_size = hashBuckets(index_count);
+	unsigned long long* edge_table = allocator.allocate<unsigned long long>(edge_table_size);
+	memset(edge_table, -1, edge_table_size * sizeof(unsigned long long));
+
+	for (size_t i = 0; i < index_count; i += 3)
+	{
+		for (int e = 0; e < 3; ++e)
+		{
+			unsigned int i0 = indices[i + e];
+			unsigned int i1 = indices[i + next[e]];
+			assert(i0 < vertex_count && i1 < vertex_count);
+
+			unsigned long long edge = ((unsigned long long)i0 << 32) | i1;
+			unsigned long long* entry = hashLookup(edge_table, edge_table_size, edge_hasher, edge, ~0ull);
+
+			if (*entry == ~0ull)
+				*entry = edge;
+		}
+	}
+
+	// build resulting index buffer: 12 indices for each input triangle
+	for (size_t i = 0; i < index_count; i += 3)
+	{
+		unsigned int patch[12];
+
+		for (int e = 0; e < 3; ++e)
+		{
+			unsigned int i0 = indices[i + e];
+			unsigned int i1 = indices[i + next[e]];
+			assert(i0 < vertex_count && i1 < vertex_count);
+
+			// note: this refers to the opposite edge!
+			unsigned long long edge = ((unsigned long long)i1 << 32) | i0;
+			unsigned long long oppe = *hashLookup(edge_table, edge_table_size, edge_hasher, edge, ~0ull);
+
+			// use the same edge if opposite edge doesn't exist (border)
+			oppe = (oppe == ~0ull) ? edge : oppe;
+
+			// triangle index (0, 1, 2)
+			patch[e] = i0;
+
+			// opposite edge (3, 4; 5, 6; 7, 8)
+			patch[3 + e * 2 + 0] = unsigned(oppe);
+			patch[3 + e * 2 + 1] = unsigned(oppe >> 32);
+
+			// dominant vertex (9, 10, 11)
+			patch[9 + e] = remap[i0];
+		}
+
+		memcpy(destination + i * 4, patch, sizeof(patch));
+	}
+}

3rdparty/meshoptimizer/src/meshoptimizer.h

@@ -97,6 +97,35 @@ MESHOPTIMIZER_API void meshopt_generateShadowIndexBuffer(unsigned int* destinati
  */
 MESHOPTIMIZER_API void meshopt_generateShadowIndexBufferMulti(unsigned int* destination, const unsigned int* indices, size_t index_count, size_t vertex_count, const struct meshopt_Stream* streams, size_t stream_count);
 
+/**
+ * Generate index buffer that can be used as a geometry shader input with triangle adjacency topology
+ * Each triangle is converted into a 6-vertex patch with the following layout:
+ * - 0, 2, 4: original triangle vertices
+ * - 1, 3, 5: vertices adjacent to edges 02, 24 and 40
+ * The resulting patch can be rendered with geometry shaders using e.g. VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY.
+ * This can be used to implement algorithms like silhouette detection/expansion and other forms of GS-driven rendering.
+ *
+ * destination must contain enough space for the resulting index buffer (index_count*2 elements)
+ * vertex_positions should have float3 position in the first 12 bytes of each vertex - similar to glVertexPointer
+ */
+MESHOPTIMIZER_EXPERIMENTAL void meshopt_generateAdjacencyIndexBuffer(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
+
+/**
+ * Generate index buffer that can be used for PN-AEN tessellation with crack-free displacement
+ * Each triangle is converted into a 12-vertex patch with the following layout:
+ * - 0, 1, 2: original triangle vertices
+ * - 3, 4: opposing edge for edge 0, 1
+ * - 5, 6: opposing edge for edge 1, 2
+ * - 7, 8: opposing edge for edge 2, 0
+ * - 9, 10, 11: dominant vertices for corners 0, 1, 2
+ * The resulting patch can be rendered with hardware tessellation using PN-AEN and displacement mapping.
+ * See "Tessellation on Any Budget" (John McDonald, GDC 2011) for implementation details.
+ *
+ * destination must contain enough space for the resulting index buffer (index_count*4 elements)
+ * vertex_positions should have float3 position in the first 12 bytes of each vertex - similar to glVertexPointer
+ */
+MESHOPTIMIZER_EXPERIMENTAL void meshopt_generateTessellationIndexBuffer(unsigned int* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
+
 /**
  * Vertex transform cache optimizer
  * Reorders indices to reduce the number of GPU vertex shader invocations
@@ -373,10 +402,13 @@ MESHOPTIMIZER_API struct meshopt_VertexFetchStatistics meshopt_analyzeVertexFetc
 
 struct meshopt_Meshlet
 {
-	unsigned int vertices[64];
-	unsigned char indices[126][3];
-	unsigned char triangle_count;
-	unsigned char vertex_count;
+	/* offsets within meshlet_vertices and meshlet_triangles arrays with meshlet data */
+	unsigned int vertex_offset;
+	unsigned int triangle_offset;
+
+	/* number of vertices and triangles used in the meshlet; data is stored in consecutive range defined by offset and count */
+	unsigned int vertex_count;
+	unsigned int triangle_count;
 };
 
 /**
@@ -386,13 +418,15 @@ struct meshopt_Meshlet
  * When using buildMeshlets, vertex positions need to be provided to minimize the size of the resulting clusters.
  * When using buildMeshletsScan, for maximum efficiency the index buffer being converted has to be optimized for vertex cache first.
  *
- * destination must contain enough space for all meshlets, worst case size can be computed with meshopt_buildMeshletsBound
+ * meshlets must contain enough space for all meshlets, worst case size can be computed with meshopt_buildMeshletsBound
+ * meshlet_vertices must contain enough space for all meshlets, worst case size is equal to max_meshlets * max_vertices
+ * meshlet_triangles must contain enough space for all meshlets, worst case size is equal to max_meshlets * max_triangles * 3
  * vertex_positions should have float3 position in the first 12 bytes of each vertex - similar to glVertexPointer
- * max_vertices and max_triangles can't exceed limits statically declared in meshopt_Meshlet (max_vertices <= 64, max_triangles <= 126)
+ * max_vertices and max_triangles must not exceed implementation limits (max_vertices <= 255 - not 256!, max_triangles <= 512)
  * cone_weight should be set to 0 when cone culling is not used, and a value between 0 and 1 otherwise to balance between cluster size and cone culling efficiency
  */
-MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_buildMeshlets(struct meshopt_Meshlet* destination, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight);
-MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_buildMeshletsScan(struct meshopt_Meshlet* destination, const unsigned int* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles);
+MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_buildMeshlets(struct meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight);
+MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_buildMeshletsScan(struct meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const unsigned int* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles);
 MESHOPTIMIZER_EXPERIMENTAL size_t meshopt_buildMeshletsBound(size_t index_count, size_t max_vertices, size_t max_triangles);
 
 struct meshopt_Bounds
@@ -430,10 +464,10 @@ struct meshopt_Bounds
  * to do frustum/occlusion culling, the formula that doesn't use the apex may be preferable.
  *
  * vertex_positions should have float3 position in the first 12 bytes of each vertex - similar to glVertexPointer
- * index_count should be less than or equal to 256*3 (the function assumes clusters of limited size)
+ * index_count/3 should be less than or equal to 512 (the function assumes clusters of limited size)
  */
 MESHOPTIMIZER_EXPERIMENTAL struct meshopt_Bounds meshopt_computeClusterBounds(const unsigned int* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
-MESHOPTIMIZER_EXPERIMENTAL struct meshopt_Bounds meshopt_computeMeshletBounds(const struct meshopt_Meshlet* meshlet, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
+MESHOPTIMIZER_EXPERIMENTAL struct meshopt_Bounds meshopt_computeMeshletBounds(const unsigned int* meshlet_vertices, const unsigned char* meshlet_triangles, size_t triangle_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
 
 /**
  * Experimental: Spatial sorter
@@ -517,6 +551,10 @@ inline void meshopt_generateShadowIndexBuffer(T* destination, const T* indices,
 template <typename T>
 inline void meshopt_generateShadowIndexBufferMulti(T* destination, const T* indices, size_t index_count, size_t vertex_count, const meshopt_Stream* streams, size_t stream_count);
 template <typename T>
+inline void meshopt_generateAdjacencyIndexBuffer(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
+template <typename T>
+inline void meshopt_generateTessellationIndexBuffer(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
+template <typename T>
 inline void meshopt_optimizeVertexCache(T* destination, const T* indices, size_t index_count, size_t vertex_count);
 template <typename T>
 inline void meshopt_optimizeVertexCacheStrip(T* destination, const T* indices, size_t index_count, size_t vertex_count);
@@ -551,9 +589,9 @@ inline meshopt_OverdrawStatistics meshopt_analyzeOverdraw(const T* indices, size
 template <typename T>
 inline meshopt_VertexFetchStatistics meshopt_analyzeVertexFetch(const T* indices, size_t index_count, size_t vertex_count, size_t vertex_size);
 template <typename T>
-inline size_t meshopt_buildMeshlets(meshopt_Meshlet* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight);
+inline size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight);
 template <typename T>
-inline size_t meshopt_buildMeshletsScan(meshopt_Meshlet* destination, const T* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles);
+inline size_t meshopt_buildMeshletsScan(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const T* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles);
 template <typename T>
 inline meshopt_Bounds meshopt_computeClusterBounds(const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride);
 template <typename T>
@@ -767,6 +805,24 @@ inline void meshopt_generateShadowIndexBufferMulti(T* destination, const T* indi
 	meshopt_generateShadowIndexBufferMulti(out.data, in.data, index_count, vertex_count, streams, stream_count);
 }
 
+template <typename T>
+inline void meshopt_generateAdjacencyIndexBuffer(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
+{
+	meshopt_IndexAdapter<T> in(0, indices, index_count);
+	meshopt_IndexAdapter<T> out(destination, 0, index_count * 2);
+
+	meshopt_generateAdjacencyIndexBuffer(out.data, in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride);
+}
+
+template <typename T>
+inline void meshopt_generateTessellationIndexBuffer(T* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride)
+{
+	meshopt_IndexAdapter<T> in(0, indices, index_count);
+	meshopt_IndexAdapter<T> out(destination, 0, index_count * 4);
+
+	meshopt_generateTessellationIndexBuffer(out.data, in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride);
+}
+
 template <typename T>
 inline void meshopt_optimizeVertexCache(T* destination, const T* indices, size_t index_count, size_t vertex_count)
 {
@@ -914,19 +970,19 @@ inline meshopt_VertexFetchStatistics meshopt_analyzeVertexFetch(const T* indices
 }
 
 template <typename T>
-inline size_t meshopt_buildMeshlets(meshopt_Meshlet* destination, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight)
+inline size_t meshopt_buildMeshlets(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const T* indices, size_t index_count, const float* vertex_positions, size_t vertex_count, size_t vertex_positions_stride, size_t max_vertices, size_t max_triangles, float cone_weight)
 {
 	meshopt_IndexAdapter<T> in(0, indices, index_count);
 
-	return meshopt_buildMeshlets(destination, in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride, max_vertices, max_triangles, cone_weight);
+	return meshopt_buildMeshlets(meshlets, meshlet_vertices, meshlet_triangles, in.data, index_count, vertex_positions, vertex_count, vertex_positions_stride, max_vertices, max_triangles, cone_weight);
 }
 
 template <typename T>
-inline size_t meshopt_buildMeshletsScan(meshopt_Meshlet* destination, const T* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles)
+inline size_t meshopt_buildMeshletsScan(meshopt_Meshlet* meshlets, unsigned int* meshlet_vertices, unsigned char* meshlet_triangles, const T* indices, size_t index_count, size_t vertex_count, size_t max_vertices, size_t max_triangles)
 {
 	meshopt_IndexAdapter<T> in(0, indices, index_count);
 
-	return meshopt_buildMeshletsScan(destination, in.data, index_count, vertex_count, max_vertices, max_triangles);
+	return meshopt_buildMeshletsScan(meshlets, meshlet_vertices, meshlet_triangles, in.data, index_count, vertex_count, max_vertices, max_triangles);
 }
 
 template <typename T>

3rdparty/meshoptimizer/src/simplifier.cpp

@@ -131,7 +131,7 @@ struct PositionHasher
 static size_t hashBuckets2(size_t count)
 {
 	size_t buckets = 1;
-	while (buckets < count)
+	while (buckets < count + count / 4)
 		buckets *= 2;
 
 	return buckets;