diff --git a/examples/obj_sticher/obj_sticher.cc b/examples/obj_sticher/obj_sticher.cc
index f59fee4..47c06eb 100644
--- a/examples/obj_sticher/obj_sticher.cc
+++ b/examples/obj_sticher/obj_sticher.cc
@@ -1,6 +1,8 @@
 //
-// Stiches multiple .obj files into one .obj. 
+// Stiches multiple .obj files into one .obj.
 //
+
+#define TINYOBJLOADER_IMPLEMENTATION
 #include "../../tiny_obj_loader.h"
 #include "obj_writer.h"
 
@@ -11,27 +13,59 @@
 
 typedef std::vector<tinyobj::shape_t> Shape;
 typedef std::vector<tinyobj::material_t> Material;
+typedef tinyobj::attrib_t Attribute;
 
 void
 StichObjs(
+  tinyobj::attrib_t& out_attribute,
   std::vector<tinyobj::shape_t>& out_shape,
   std::vector<tinyobj::material_t>& out_material,
+  const std::vector<Attribute>& attributes,
   const std::vector<Shape>& shapes,
   const std::vector<Material>& materials)
 {
-  int numShapes = 0;
-  for (size_t i = 0; i < shapes.size(); i++) {
-    numShapes += (int)shapes[i].size();
+  // The amount of attributes, shape-vectors and material-vecotrs should be the same.
+  if(attributes.size() != shapes.size() && attributes.size() != materials.size()){
+    std::cerr << "Size of attributes, shapes and Materials don't fit!" << attributes.size() << " " << shapes.size() <<" " << materials.size() << std::endl;;
+    exit(1);
+  }
+  int num_shapes = 0;
+  // 4 values (vertices, normals, texcoords, colors)
+  std::vector<int> num_attributes(4, 0);
+  int num_materials = 0;
+  for(int i = 0; i < shapes.size(); i++){
+    num_shapes += shapes[i].size();
+  }
+  for(int i = 0; i < attributes.size(); i++){
+    num_attributes[0] += attributes[i].vertices.size();
+    num_attributes[1] += attributes[i].normals.size();
+    num_attributes[2] += attributes[i].texcoords.size();
+    num_attributes[3] += attributes[i].colors.size();
+  }
+  for(int i = 0; i < materials.size(); i++){
+    num_materials += materials[i].size();
   }
 
-  printf("Total # of shapes = %d\n", numShapes);
-  int materialIdOffset = 0;
+  // More performant, than push_back
+  out_attribute.vertices.resize(num_attributes[0]);
+  out_attribute.normals.resize(num_attributes[1]);
+  out_attribute.texcoords.resize(num_attributes[2]);
+  out_attribute.colors.resize(num_attributes[3]);
+  out_shape.resize(num_shapes);
+  out_material.resize(num_materials);
 
-  size_t face_offset = 0;
+  int material_id_offset = 0;
+  int shape_id_offset = 0;
+  int vertex_idx_offset = 0;
+  int normal_idx_offset = 0;
+  int texcoord_idx_offset = 0;
+  int color_idx_offset = 0;
+
+  // shapes.size() = attributes.size() = materials.size()
   for (size_t i = 0; i < shapes.size(); i++) {
 
+    // Copy shapes
     for (size_t k = 0; k < shapes[i].size(); k++) {
-
       std::string new_name = shapes[i][k].name;
       // Add suffix
       char buf[1024];
@@ -39,36 +73,51 @@ StichObjs(
       new_name += std::string(buf);
 
       printf("shape[%ld][%ld].name = %s\n", i, k, shapes[i][k].name.c_str());
-      assert((shapes[i][k].mesh.indices.size() % 3) == 0);
-      assert((shapes[i][k].mesh.positions.size() % 3) == 0);
 
       tinyobj::shape_t new_shape = shapes[i][k];
-      // Add offset.
-      for (size_t f = 0; f < new_shape.mesh.material_ids.size(); f++) {
-        new_shape.mesh.material_ids[f] += materialIdOffset;
+      // Add material offset.
+      for(size_t f = 0; f < new_shape.mesh.material_ids.size(); f++) {
+        new_shape.mesh.material_ids[f] += material_id_offset;
+      }
+      // Add indices offset.
+      for(size_t f = 0; f < new_shape.mesh.indices.size(); f++){
+        tinyobj::index_t& ref = new_shape.mesh.indices[f];
+        if(ref.vertex_index > -1){
+          ref.vertex_index += vertex_idx_offset;
+        }
+        if(ref.normal_index > -1){
+          ref.normal_index += normal_idx_offset;
+        }
+        if(ref.texcoord_index > -1){
+          ref.texcoord_index += texcoord_idx_offset;
+        }
       }
 
       new_shape.name = new_name;
       printf("shape[%ld][%ld].new_name = %s\n", i, k, new_shape.name.c_str());
 
-      out_shape.push_back(new_shape);
+      out_shape[shape_id_offset++] = new_shape;
     }
 
-    materialIdOffset += materials[i].size();
-  }
-
-  for (size_t i = 0; i < materials.size(); i++) {
+    // Copy materials
     for (size_t k = 0; k < materials[i].size(); k++) {
-      out_material.push_back(materials[i][k]);
+      out_material[material_id_offset++] = materials[i][k];
     }
-  }
 
+    // Copy attributes (3 floats per vertex, 3 floats per normal, 2 floats per texture-coordinate, 3 floats per color)
+    // You could also include a check here, if the sizes are dividable by 3 (resp. 2), but it's safe to simply assume, they do.
+    std::copy(attributes[i].vertices.begin(), attributes[i].vertices.end(), out_attribute.vertices.begin() + vertex_idx_offset * 3);
+    vertex_idx_offset += attributes[i].vertices.size() / 3;
+    std::copy(attributes[i].normals.begin(), attributes[i].normals.end(), out_attribute.normals.begin() + normal_idx_offset * 3);
+    normal_idx_offset += attributes[i].normals.size() / 3;
+    std::copy(attributes[i].texcoords.begin(), attributes[i].texcoords.end(), out_attribute.texcoords.begin() + texcoord_idx_offset * 2);
+    texcoord_idx_offset += attributes[i].texcoords.size() / 2;
+    std::copy(attributes[i].colors.begin(), attributes[i].colors.end(), out_attribute.colors.begin() + color_idx_offset);
+    color_idx_offset += attributes[i].colors.size();
+  }
 }
 
-int
-main(
-  int argc,
-  char **argv)
+int main(int argc, char **argv)
 {
   if (argc < 3) {
     printf("Usage: obj_sticher input0.obj input1.obj ... output.obj\n");
@@ -78,16 +127,15 @@ main(
   int num_objfiles = argc - 2;
   std::string out_filename = std::string(argv[argc-1]); // last element
 
-  std::vector<Shape> shapes;
-  std::vector<Material> materials;
-  shapes.resize(num_objfiles);
-  materials.resize(num_objfiles);
+  std::vector<Attribute> attributes(num_objfiles);
+  std::vector<Shape> shapes(num_objfiles);
+  std::vector<Material> materials(num_objfiles);
 
   for (int i = 0; i < num_objfiles; i++) {
     std::cout << "Loading " << argv[i+1] << " ... " << std::flush;
     
     std::string err;
-    bool ret = tinyobj::LoadObj(shapes[i], materials[i], err, argv[i+1]);
+    bool ret = tinyobj::LoadObj(&attributes[i], &shapes[i], &materials[i], &err, argv[i+1]);
     if (!err.empty()) {
       std::cerr << err << std::endl;
     }
@@ -98,12 +146,13 @@ main(
     std::cout << "DONE." << std::endl;
   }
 
-  std::vector<tinyobj::shape_t> out_shape;
-  std::vector<tinyobj::material_t> out_material;
-  StichObjs(out_shape, out_material, shapes, materials);
+  Attribute out_attribute;
+  Shape out_shape;
+  Material out_material;
+  StichObjs(out_attribute, out_shape, out_material, attributes, shapes, materials);
 
   bool coordTransform = true;
-  bool ret = WriteObj(out_filename, out_shape, out_material, coordTransform);
+  bool ret = WriteObj(out_filename, out_attribute, out_shape, out_material, coordTransform);
   assert(ret);
 
   return 0;
diff --git a/examples/obj_sticher/obj_writer.cc b/examples/obj_sticher/obj_writer.cc
index 694ce45..9ea8d7c 100644
--- a/examples/obj_sticher/obj_writer.cc
+++ b/examples/obj_sticher/obj_writer.cc
@@ -30,6 +30,8 @@ bool WriteMat(const std::string& filename, const std::vector<tinyobj::material_t
     fprintf(fp, "Ke %f %f %f\n", mat.emission[0], mat.emission[1], mat.emission[2]);
     fprintf(fp, "Ns %f\n", mat.shininess);
     fprintf(fp, "Ni %f\n", mat.ior);
+    fprintf(fp, "illum %d\n", mat.illum);
+    fprintf(fp, "\n");
     // @todo { texture }
   }
   
@@ -38,7 +40,7 @@ bool WriteMat(const std::string& filename, const std::vector<tinyobj::material_t
   return true;
 }
 
-bool WriteObj(const std::string& filename, const std::vector<tinyobj::shape_t>& shapes, const std::vector<tinyobj::material_t>& materials, bool coordTransform) {
+bool WriteObj(const std::string& filename, const tinyobj::attrib_t& attributes, const std::vector<tinyobj::shape_t>& shapes, const std::vector<tinyobj::material_t>& materials, bool coordTransform) {
   FILE* fp = fopen(filename.c_str(), "w");
   if (!fp) {
     fprintf(stderr, "Failed to open file [ %s ] for write.\n", filename.c_str());
@@ -48,17 +50,53 @@ bool WriteObj(const std::string& filename, const std::vector<tinyobj::shape_t>&
   std::string basename = GetFileBasename(filename);
   std::string material_filename = basename + ".mtl";
 
-  int v_offset = 0;
-  int vn_offset = 0;
-  int vt_offset = 0;
   int prev_material_id = -1;
 
-  fprintf(fp, "mtllib %s\n", material_filename.c_str());
+  fprintf(fp, "mtllib %s\n\n", material_filename.c_str());
+
+  // facevarying vtx
+  for (size_t k = 0; k < attributes.vertices.size(); k+=3) {
+    if (coordTransform) {
+      fprintf(fp, "v %f %f %f\n",
+        attributes.vertices[k + 0],
+        attributes.vertices[k + 2],
+        -attributes.vertices[k + 1]);
+    } else {
+      fprintf(fp, "v %f %f %f\n",
+        attributes.vertices[k + 0],
+        attributes.vertices[k + 1],
+        attributes.vertices[k + 2]);
+    }
+  }
+
+  fprintf(fp, "\n");
+
+  // facevarying normal
+  for (size_t k = 0; k < attributes.normals.size(); k += 3) {
+    if (coordTransform) {
+      fprintf(fp, "vn %f %f %f\n",
+        attributes.normals[k + 0],
+        attributes.normals[k + 2],
+        -attributes.normals[k + 1]);
+    } else {
+      fprintf(fp, "vn %f %f %f\n",
+        attributes.normals[k + 0],
+        attributes.normals[k + 1],
+        attributes.normals[k + 2]);
+    }
+  }
+
+  fprintf(fp, "\n");
+
+  // facevarying texcoord
+  for (size_t k = 0; k < attributes.texcoords.size(); k += 2) {
+    fprintf(fp, "vt %f %f\n",
+      attributes.texcoords[k + 0],
+      attributes.texcoords[k + 1]);
+  }
 
   for (size_t i = 0; i < shapes.size(); i++) {
-
-    bool has_vn = false;
-    bool has_vt = false;
+    fprintf(fp, "\n");
 
     if (shapes[i].name.empty()) {
       fprintf(fp, "g Unknown\n");
@@ -66,101 +104,53 @@ bool WriteObj(const std::string& filename, const std::vector<tinyobj::shape_t>&
       fprintf(fp, "g %s\n", shapes[i].name.c_str());
     }
 
-    //if (!shapes[i].material.name.empty()) {
-    //  fprintf(fp, "usemtl %s\n", shapes[i].material.name.c_str());
-    //}
-
-    // facevarying vtx
-    for (size_t k = 0; k < shapes[i].mesh.indices.size() / 3; k++) {
-      for (int j = 0; j < 3; j++) {
-        int idx = shapes[i].mesh.indices[3*k+j];
-        if (coordTransform) {
-          fprintf(fp, "v %f %f %f\n",
-            shapes[i].mesh.positions[3*idx+0],
-            shapes[i].mesh.positions[3*idx+2],
-            -shapes[i].mesh.positions[3*idx+1]);
-        } else {
-          fprintf(fp, "v %f %f %f\n",
-            shapes[i].mesh.positions[3*idx+0],
-            shapes[i].mesh.positions[3*idx+1],
-            shapes[i].mesh.positions[3*idx+2]);
-        }
-      }
+    bool has_vn = false;
+    bool has_vt = false;
+    // Assumes normals and textures are set shape-wise.
+    if(shapes[i].mesh.indices.size() > 0){
+      has_vn = shapes[i].mesh.indices[0].normal_index != -1;
+      has_vt = shapes[i].mesh.indices[0].texcoord_index != -1;
     }
 
-    // facevarying normal
-    if (shapes[i].mesh.normals.size() > 0) {
-      for (size_t k = 0; k < shapes[i].mesh.indices.size() / 3; k++) {
-        for (int j = 0; j < 3; j++) {
-          int idx = shapes[i].mesh.indices[3*k+j];
-          if (coordTransform) {
-            fprintf(fp, "vn %f %f %f\n",
-              shapes[i].mesh.normals[3*idx+0],
-              shapes[i].mesh.normals[3*idx+2],
-              -shapes[i].mesh.normals[3*idx+1]);
-          } else {
-            fprintf(fp, "vn %f %f %f\n",
-              shapes[i].mesh.normals[3*idx+0],
-              shapes[i].mesh.normals[3*idx+1],
-              shapes[i].mesh.normals[3*idx+2]);
-          }
-        }
-      }
-    }
-    if (shapes[i].mesh.normals.size() > 0) has_vn = true;
-
-    // facevarying texcoord
-    if (shapes[i].mesh.texcoords.size() > 0) {
-      for (size_t k = 0; k < shapes[i].mesh.indices.size() / 3; k++) {
-        for (int j = 0; j < 3; j++) {
-          int idx = shapes[i].mesh.indices[3*k+j];
-          fprintf(fp, "vt %f %f\n",
-            shapes[i].mesh.texcoords[2*idx+0],
-            shapes[i].mesh.texcoords[2*idx+1]);
-        }
-      }
-    }
-    if (shapes[i].mesh.texcoords.size() > 0) has_vt = true;
-
     // face
-    for (size_t k = 0; k < shapes[i].mesh.indices.size() / 3; k++) {
-  
-      // Face index is 1-base.
-      //int v0 = shapes[i].mesh.indices[3*k+0] + 1 + v_offset;
-      //int v1 = shapes[i].mesh.indices[3*k+1] + 1 + v_offset;
-      //int v2 = shapes[i].mesh.indices[3*k+2] + 1 + v_offset;
-      int v0 = (3*k + 0) + 1 + v_offset;
-      int v1 = (3*k + 1) + 1 + v_offset;
-      int v2 = (3*k + 2) + 1 + v_offset;
-
-      int vt0 = (3*k + 0) + 1 + vt_offset;
-      int vt1 = (3*k + 1) + 1 + vt_offset;
-      int vt2 = (3*k + 2) + 1 + vt_offset;
-
-      int material_id = shapes[i].mesh.material_ids[k];
+    int face_index = 0;
+    for (size_t k = 0; k < shapes[i].mesh.indices.size(); k += shapes[i].mesh.num_face_vertices[face_index++]) {
+      // Check Materials
+      int material_id = shapes[i].mesh.material_ids[face_index];
       if (material_id != prev_material_id) {
         std::string material_name = materials[material_id].name;
         fprintf(fp, "usemtl %s\n", material_name.c_str());
         prev_material_id = material_id;
       }
 
-      if (has_vn && has_vt) {
-        fprintf(fp, "f %d/%d/%d %d/%d/%d %d/%d/%d\n",
-          v0, vt0, v0, v1, vt1, v1, v2, vt2, v2);
-      } else if (has_vn && !has_vt) {
-        fprintf(fp, "f %d//%d %d//%d %d//%d\n", v0, v0, v1, v1, v2, v2);
-      } else if (!has_vn && has_vt) {
-        fprintf(fp, "f %d/%d %d/%d %d/%d\n", v0, v0, v1, v1, v2, v2);
-      } else {
-        fprintf(fp, "f %d %d %d\n", v0, v1, v2);
+      unsigned char v_per_f = shapes[i].mesh.num_face_vertices[face_index];
+      // Imperformant, but if you want to have variable vertices per face, you need some kind of a dynamic loop.
+      fprintf(fp, "f");
+      for(int l = 0; l < v_per_f; l++){
+        const tinyobj::index_t& ref = shapes[i].mesh.indices[k + l];
+        if(has_vn && has_vt){
+          // v0/t0/vn0
+          fprintf(fp, " %d/%d/%d", ref.vertex_index + 1, ref.texcoord_index + 1, ref.normal_index + 1);
+          continue;
+        }
+        if(has_vn && !has_vt){
+          // v0//vn0
+          fprintf(fp, " %d//%d", ref.vertex_index + 1, ref.normal_index + 1);
+          continue;
+        }
+        if(!has_vn && has_vt){
+          // v0/vt0
+          fprintf(fp, " %d/%d", ref.vertex_index + 1, ref.texcoord_index + 1);
+          continue;
+        }
+        if(!has_vn && !has_vt){
+          // v0 v1 v2
+          fprintf(fp, " %d", ref.vertex_index + 1);
+          continue;
+        }
       }
-      
+      fprintf(fp, "\n");
     }
-
-    v_offset  += shapes[i].mesh.indices.size();
-    //vn_offset += shapes[i].mesh.normals.size() / 3;
-    vt_offset += shapes[i].mesh.texcoords.size() / 2;
-
   }
 
   fclose(fp);
diff --git a/examples/obj_sticher/obj_writer.h b/examples/obj_sticher/obj_writer.h
index bb367b6..262c681 100644
--- a/examples/obj_sticher/obj_writer.h
+++ b/examples/obj_sticher/obj_writer.h
@@ -3,7 +3,6 @@
 
 #include "../../tiny_obj_loader.h"
 
-extern bool WriteObj(const std::string& filename, const std::vector<tinyobj::shape_t>& shapes, const std::vector<tinyobj::material_t>& materials, bool coordTransform = false);
-
+extern bool WriteObj(const std::string& filename, const tinyobj::attrib_t& attributes, const std::vector<tinyobj::shape_t>& shapes, const std::vector<tinyobj::material_t>& materials, bool coordTransform = false);
 
 #endif // __OBJ_WRITER_H__
diff --git a/experimental/README.md b/experimental/README.md
index 99378ce..a0603af 100644
--- a/experimental/README.md
+++ b/experimental/README.md
@@ -6,11 +6,21 @@
 
 * C++-11 compiler
 
+## How to build
+
+```
+$ premak5 gmake
+```
+
 ## Compile options
 
 * zstd compressed .obj support. `--with-zstd` premake option.
 * gzip compressed .obj support. `--with-zlib` premake option.
 
+## Notes on AMD GPU + Linux
+
+You may need to link with libdrm(`-ldrm`).
+
 ## Licenses
 
 * lfpAlloc : MIT license.
diff --git a/experimental/tinyobj_loader_opt.h b/experimental/tinyobj_loader_opt.h
index 154611f..e6c0b20 100644
--- a/experimental/tinyobj_loader_opt.h
+++ b/experimental/tinyobj_loader_opt.h
@@ -1306,15 +1306,17 @@ bool parseObj(attrib_t *attrib, std::vector<shape_t> *shapes,
           end_idx = len - 1;
         }
 
+        // true if the line currently read must be added to the current line info
+        bool new_line_found = (t == 0) || is_line_ending(buf, start_idx - 1, end_idx);
+
         size_t prev_pos = start_idx;
         for (size_t i = start_idx; i < end_idx; i++) {
           if (is_line_ending(buf, i, end_idx)) {
-            if ((t > 0) && (prev_pos == start_idx) &&
-                (!is_line_ending(buf, start_idx - 1, end_idx))) {
+            if (!new_line_found) {
               // first linebreak found in (chunk > 0), and a line before this
               // linebreak belongs to previous chunk, so skip it.
               prev_pos = i + 1;
-              continue;
+              new_line_found = true;
             } else {
               LineInfo info;
               info.pos = prev_pos;
@@ -1329,11 +1331,10 @@ bool parseObj(attrib_t *attrib, std::vector<shape_t> *shapes,
           }
         }
 
-        // Find extra line which spand across chunk boundary.
-        if ((t < num_threads) && (buf[end_idx - 1] != '\n')) {
-          auto extra_span_idx = (std::min)(end_idx - 1 + chunk_size, len);
-          for (size_t i = end_idx; i < extra_span_idx; i++) {
-            if (is_line_ending(buf, i, extra_span_idx)) {
+        // If at least one line started in this chunk, find where it ends in the rest of the buffer
+        if (new_line_found && (t < num_threads) && (buf[end_idx - 1] != '\n')) {
+          for (size_t i = end_idx; i < len; i++) {
+            if (is_line_ending(buf, i, len)) {
               LineInfo info;
               info.pos = prev_pos;
               info.len = i - prev_pos;
@@ -1493,7 +1494,7 @@ bool parseObj(attrib_t *attrib, std::vector<shape_t> *shapes,
     attrib->texcoords.resize(num_vt * 2);
     attrib->indices.resize(num_f);
     attrib->face_num_verts.resize(num_indices);
-    attrib->material_ids.resize(num_indices);
+    attrib->material_ids.resize(num_indices, -1);
 
     size_t v_offsets[kMaxThreads];
     size_t n_offsets[kMaxThreads];
@@ -1524,22 +1525,43 @@ bool parseObj(attrib_t *attrib, std::vector<shape_t> *shapes,
         size_t t_count = t_offsets[t];
         size_t f_count = f_offsets[t];
         size_t face_count = face_offsets[t];
-		int material_id = -1;  // -1 = default unknown material.
 
         for (size_t i = 0; i < commands[t].size(); i++) {
           if (commands[t][i].type == COMMAND_EMPTY) {
             continue;
           } else if (commands[t][i].type == COMMAND_USEMTL) {
             if (commands[t][i].material_name &&
-                commands[t][i].material_name_len > 0) {
-              std::string material_name(commands[t][i].material_name,
-                                        commands[t][i].material_name_len);
-
-              if (material_map.find(material_name) != material_map.end()) {
-                material_id = material_map[material_name];
-              } else {
-                // Assign invalid material ID
-                material_id = -1;
+                commands[t][i].material_name_len > 0 &&
+                // check if there are still faces after this command
+                face_count < num_indices) {
+              // Find next face
+              bool found = false;
+              size_t i_start = i + 1, t_next, i_next;
+              for (t_next = t; t_next < num_threads; t_next++) {
+                for (i_next = i_start; i_next < commands[t_next].size(); i_next++) {
+                  if (commands[t_next][i_next].type == COMMAND_F) {
+                    found = true;
+                    break;
+                  }
+                }
+                if (found)
+                  break;
+                i_start = 0;
+              }
+              // Assign material to this face
+              if (found) {
+                std::string material_name(commands[t][i].material_name,
+                                          commands[t][i].material_name_len);
+                for (size_t k = 0; k < commands[t_next][i_next].f_num_verts.size(); k++) {
+                  if (material_map.find(material_name) != material_map.end()) {
+                      attrib->material_ids[face_count + k] = material_map[material_name];
+                  } else {
+                    // Assign invalid material ID
+                    // Set a different value than the default, to
+                    // prevent following faces from being assigned a valid material
+                    attrib->material_ids[face_count + k] = -2;
+                  }
+                }
               }
             }
           } else if (commands[t][i].type == COMMAND_V) {
@@ -1566,7 +1588,6 @@ bool parseObj(attrib_t *attrib, std::vector<shape_t> *shapes,
                   index_t(vertex_index, texcoord_index, normal_index);
             }
             for (size_t k = 0; k < commands[t][i].f_num_verts.size(); k++) {
-              attrib->material_ids[face_count + k] = material_id;
               attrib->face_num_verts[face_count + k] =
                   commands[t][i].f_num_verts[k];
             }
@@ -1581,19 +1602,13 @@ bool parseObj(attrib_t *attrib, std::vector<shape_t> *shapes,
     for (size_t t = 0; t < workers->size(); t++) {
       workers[t].join();
     }
-	if(material_map.size()>1&& num_threads>1) {
-			for (size_t t = 0; t < num_threads; t++) {
-				size_t face_count = face_offsets[t];
-				if (-1 == attrib->material_ids[face_count]) {
-					int prev_material_id = attrib->material_ids[face_count - 1];
-					size_t max_face_offset = (t == num_threads - 1) ? attrib->material_ids.size() : face_offsets[t + 1];
-					for (int i = face_count; i<max_face_offset; ++i) {
-						if (attrib->material_ids[i] != -1) break;
-						attrib->material_ids[i] = prev_material_id;
-					}
-				}
-			}
-	}
+    
+    // To each face with uninitialized material id,
+    // assign the material id of the last face preceding it that has one
+    for (size_t face_count = 1; face_count < num_indices; ++face_count)
+      if (attrib->material_ids[face_count] == -1)
+        attrib->material_ids[face_count] = attrib->material_ids[face_count - 1];
+    
     auto t_end = std::chrono::high_resolution_clock::now();
     ms_merge = t_end - t_start;
   }
diff --git a/models/invalid-face-definition.mtl b/models/invalid-face-definition.mtl
new file mode 100644
index 0000000..d3a1c7a
--- /dev/null
+++ b/models/invalid-face-definition.mtl
@@ -0,0 +1,24 @@
+newmtl white
+Ka 0 0 0
+Kd 1 1 1
+Ks 0 0 0
+
+newmtl red
+Ka 0 0 0
+Kd 1 0 0
+Ks 0 0 0
+
+newmtl green
+Ka 0 0 0
+Kd 0 1 0
+Ks 0 0 0
+
+newmtl blue
+Ka 0 0 0
+Kd 0 0 1
+Ks 0 0 0
+
+newmtl light
+Ka 20 20 20
+Kd 1 1 1
+Ks 0 0 0
diff --git a/models/invalid-face-definition.obj b/models/invalid-face-definition.obj
new file mode 100644
index 0000000..73ca678
--- /dev/null
+++ b/models/invalid-face-definition.obj
@@ -0,0 +1,18 @@
+mtllib invalid-face-definition.mtl
+
+v 0.000000 2.000000 2.000000
+v 0.000000 0.000000 2.000000
+v 2.000000 0.000000 2.000000
+v 2.000000 2.000000 2.000000
+v 0.000000 2.000000 0.000000
+v 0.000000 0.000000 0.000000
+v 2.000000 0.000000 0.000000
+v 2.000000 2.000000 0.000000
+# 8 vertices
+
+g front cube
+usemtl white
+f 1
+g back cube
+# expects white material
+f 8 7
diff --git a/tests/tester.cc b/tests/tester.cc
index fd5a90f..1d0e34e 100644
--- a/tests/tester.cc
+++ b/tests/tester.cc
@@ -751,6 +751,23 @@ TEST_CASE("smoothing-group", "[Issue162]") {
 
 }
 
+TEST_CASE("invalid-face-definition", "[face]") {
+  tinyobj::attrib_t attrib;
+  std::vector<tinyobj::shape_t> shapes;
+  std::vector<tinyobj::material_t> materials;
+
+  std::string err;
+  bool ret = tinyobj::LoadObj(&attrib, &shapes, &materials, &err, "../models/invalid-face-definition.obj", gMtlBasePath);
+
+  if (!err.empty()) {
+    std::cerr << "[face] " << err << std::endl;
+  }
+
+  REQUIRE(true == ret);
+  REQUIRE(1 == shapes.size());
+  REQUIRE(0 == shapes[0].mesh.indices.size());
+}
+
 // Fuzzer test.
 // Just check if it does not crash.
 
diff --git a/tiny_obj_loader.h b/tiny_obj_loader.h
index eae25b7..fbb31b1 100644
--- a/tiny_obj_loader.h
+++ b/tiny_obj_loader.h
@@ -58,6 +58,9 @@ namespace tinyobj {
 #if __has_warning("-Wzero-as-null-pointer-constant")
 #pragma clang diagnostic ignored "-Wzero-as-null-pointer-constant"
 #endif
+
+#pragma clang diagnostic ignored "-Wpadded"
+
 #endif
 
 // https://en.wikipedia.org/wiki/Wavefront_.obj_file says ...
@@ -370,6 +373,7 @@ void LoadMtl(std::map<std::string, int> *material_map,
 #include <cstdlib>
 #include <cstring>
 #include <utility>
+#include <limits>
 
 #include <fstream>
 #include <sstream>
@@ -867,22 +871,22 @@ static bool ParseTextureNameAndOption(std::string *texname,
   } else {
     texopt->imfchan = 'm';
   }
-  texopt->bump_multiplier = 1.0f;
+  texopt->bump_multiplier = static_cast<real_t>(1.0);
   texopt->clamp = false;
   texopt->blendu = true;
   texopt->blendv = true;
-  texopt->sharpness = 1.0f;
-  texopt->brightness = 0.0f;
-  texopt->contrast = 1.0f;
-  texopt->origin_offset[0] = 0.0f;
-  texopt->origin_offset[1] = 0.0f;
-  texopt->origin_offset[2] = 0.0f;
-  texopt->scale[0] = 1.0f;
-  texopt->scale[1] = 1.0f;
-  texopt->scale[2] = 1.0f;
-  texopt->turbulence[0] = 0.0f;
-  texopt->turbulence[1] = 0.0f;
-  texopt->turbulence[2] = 0.0f;
+  texopt->sharpness = static_cast<real_t>(1.0);
+  texopt->brightness = static_cast<real_t>(0.0);
+  texopt->contrast = static_cast<real_t>(1.0);
+  texopt->origin_offset[0] = static_cast<real_t>(0.0);
+  texopt->origin_offset[1] = static_cast<real_t>(0.0);
+  texopt->origin_offset[2] = static_cast<real_t>(0.0);
+  texopt->scale[0] = static_cast<real_t>(1.0);
+  texopt->scale[1] = static_cast<real_t>(1.0);
+  texopt->scale[2] = static_cast<real_t>(1.0);
+  texopt->turbulence[0] = static_cast<real_t>(0.0);
+  texopt->turbulence[1] = static_cast<real_t>(0.0);
+  texopt->turbulence[2] = static_cast<real_t>(0.0);
   texopt->type = TEXTURE_TYPE_NONE;
 
   const char *token = linebuf;  // Assume line ends with NULL
@@ -968,24 +972,24 @@ static void InitMaterial(material_t *material) {
   material->reflection_texname = "";
   material->alpha_texname = "";
   for (int i = 0; i < 3; i++) {
-    material->ambient[i] = 0.f;
-    material->diffuse[i] = 0.f;
-    material->specular[i] = 0.f;
-    material->transmittance[i] = 0.f;
-    material->emission[i] = 0.f;
+    material->ambient[i] = static_cast<real_t>(0.0);
+    material->diffuse[i] = static_cast<real_t>(0.0);
+    material->specular[i] = static_cast<real_t>(0.0);
+    material->transmittance[i] = static_cast<real_t>(0.0);
+    material->emission[i] = static_cast<real_t>(0.0);
   }
   material->illum = 0;
-  material->dissolve = 1.f;
-  material->shininess = 1.f;
-  material->ior = 1.f;
+  material->dissolve = static_cast<real_t>(1.0);
+  material->shininess = static_cast<real_t>(1.0);
+  material->ior = static_cast<real_t>(1.0);
 
-  material->roughness = 0.f;
-  material->metallic = 0.f;
-  material->sheen = 0.f;
-  material->clearcoat_thickness = 0.f;
-  material->clearcoat_roughness = 0.f;
-  material->anisotropy_rotation = 0.f;
-  material->anisotropy = 0.f;
+  material->roughness = static_cast<real_t>(0.0);
+  material->metallic = static_cast<real_t>(0.0);
+  material->sheen = static_cast<real_t>(0.0);
+  material->clearcoat_thickness = static_cast<real_t>(0.0);
+  material->clearcoat_roughness = static_cast<real_t>(0.0);
+  material->anisotropy_rotation = static_cast<real_t>(0.0);
+  material->anisotropy = static_cast<real_t>(0.0);
   material->roughness_texname = "";
   material->metallic_texname = "";
   material->sheen_texname = "";
@@ -996,8 +1000,9 @@ static void InitMaterial(material_t *material) {
 }
 
 // code from https://wrf.ecse.rpi.edu//Research/Short_Notes/pnpoly.html
-static int pnpoly(int nvert, float *vertx, float *verty, float testx,
-                  float testy) {
+template<typename T>
+static int pnpoly(int nvert, T *vertx, T *verty, T testx,
+                  T testy) {
   int i, j, c = 0;
   for (i = 0, j = nvert - 1; i < nvert; j = i++) {
     if (((verty[i] > testy) != (verty[j] > testy)) &&
@@ -1027,7 +1032,7 @@ static bool exportFaceGroupToShape(shape_t *shape,
     size_t npolys = face.vertex_indices.size();
 
     if (npolys < 3) {
-      // ??? Invalid face definition.
+      // Face must have 3+ vertices.
       continue;
     }
 
@@ -1068,10 +1073,10 @@ static bool exportFaceGroupToShape(shape_t *shape,
         real_t e1x = v2x - v1x;
         real_t e1y = v2y - v1y;
         real_t e1z = v2z - v1z;
-        float cx = std::fabs(e0y * e1z - e0z * e1y);
-        float cy = std::fabs(e0z * e1x - e0x * e1z);
-        float cz = std::fabs(e0x * e1y - e0y * e1x);
-        const float epsilon = 0.0001f;
+        real_t cx = std::fabs(e0y * e1z - e0z * e1y);
+        real_t cy = std::fabs(e0z * e1x - e0x * e1z);
+        real_t cz = std::fabs(e0x * e1y - e0y * e1x);
+        const real_t epsilon = std::numeric_limits<real_t>::epsilon();
         if (cx > epsilon || cy > epsilon || cz > epsilon) {
           // found a corner
           if (cx > cy && cx > cz) {
@@ -1100,7 +1105,7 @@ static bool exportFaceGroupToShape(shape_t *shape,
         real_t v0y = v[vi0 * 3 + axes[1]];
         real_t v1x = v[vi1 * 3 + axes[0]];
         real_t v1y = v[vi1 * 3 + axes[1]];
-        area += (v0x * v1y - v0y * v1x) * 0.5f;
+        area += (v0x * v1y - v0y * v1x) * static_cast<real_t>(0.5);
       }
 
       int maxRounds =
@@ -1136,7 +1141,7 @@ static bool exportFaceGroupToShape(shape_t *shape,
         real_t e1y = vy[2] - vy[1];
         real_t cross = e0x * e1y - e0y * e1x;
         // if an internal angle
-        if (cross * area < 0.0f) {
+        if (cross * area < static_cast<real_t>(0.0)) {
           guess_vert += 1;
           continue;
         }
@@ -1437,7 +1442,7 @@ void LoadMtl(std::map<std::string, int> *material_map,
         // We invert value of Tr(assume Tr is in range [0, 1])
         // NOTE: Interpretation of Tr is application(exporter) dependent. For
         // some application(e.g. 3ds max obj exporter), Tr = d(Issue 43)
-        material.dissolve = 1.0f - parseReal(&token);
+        material.dissolve = static_cast<real_t>(1.0) - parseReal(&token);
       }
       has_tr = true;
       continue;