This only fixes the 'symptoms', root of the issue here is that we get
a weird covariance matrix - Blender's own invert function errors on it,
this script's matrix_invert func does give some result, but in any case
final result is not what one would expect...

Don't print tile size on terminal, was only for debug

Also reduce default margin slightly, especially for spacing between vertically-aligned nodes.

Still unsure we are doing the 'right' thing here, but at least let's be consistent
between our importer and exporter!

This sounds suspicious to me, but it’s FBX, so could be the solution, we'll see...

Now, we consider default BU as 1 meter (100 FBX units), instead of 1 FBX unit (1cm)...

Yes... 10h of work, for a oneliner fixing a simple stupid dummy missing connection
between the empty used as armature 'object', and its (void, useless) 'NodeAttribute' data...

Why such basic harmless error breaks something (aparently) completly unrelated
- and why is it breaking anything, btw - is to be added on the Olympus Mons of FBXSDK madness.
Not to mention the usual total lack of parsing warnings/errors from that 'thing'.

Trying to fix the ugly scaling issue with FBX and animated armatures. No luck so far,
all those changes make our generated file closer to those generated by 'official' FBX apps,
but does not seem to fix or solve anything...

Also, fixed wrong '2.75' blender version statement - no need to update that unless mandatory,
this is minimal working version, not 'current' version...

And please, always remove menus *before* unregistering, otherwise UI code may try
to access removed classes (menus, shortcuts, operators, etc.).

Updated Authors, added built in xyz math object presets

Limited, because STL only stores face normals, so we can only fake this by setting
all clnors of a same face to that face normal... Guess use case are rather limited,
but does not hurt to have it either.

Tile size would be recalculated constantly (every scene update) when the calculated size was less than 8 pixels (blender's own hard limit)
diff --git a/render_auto_tile_size.py b/render_auto_tile_size.py
index a773281..00f848e 100644
--- a/render_auto_tile_size.py
+++ b/render_auto_tile_size.py
@@ -20,7 +20,7 @@ bl_info = {
     "name": "Auto Tile Size",
     "description": "Estimate and set the tile size that will render the fastest",
     "author": "Greg Zaal",
-    "version": (3, 0),
+    "version": (3, 1),
     "blender": (2, 74, 0),
     "location": "Render Settings > Performance",
     "warning": "",
@@ -300,6 +300,12 @@ def do_set_tile_size(context):
     else:
         settings.threads_error = False

+    # Make sure tile sizes are within the internal limit
+    tile_x = max(8, tile_x)
+    tile_y = max(8, tile_y)
+    tile_x = min(65536, tile_x)
+    tile_y = min(65536, tile_y)
+
     render.tile_x = tile_x
     render.tile_y = tile_y

 - Let the user choose a specific number for the target tile size (not only limited to powers of 2).
 - If there are more render threads available than tiles, reduce the tile size until all threads will be utilized - this means you can do tiny border-renders and still utilize 100% of your CPU.
 - Number of tiles is now displayed in the UI.
 - Minor tweaks to tooltips and function names.

NOTE: The math of calculating the maximum tile size possible while still using all the threads is a bit wonky - it's functional, but not nice. Occasionally when there are many CPU threads (e.g. 16), only 15 of them will be used. Math is not my forte, so some help in this area would be much appreciated.

diff --git a/render_auto_tile_size.py b/render_auto_tile_size.py
index 8ece451..a773281 100644
--- a/render_auto_tile_size.py
+++ b/render_auto_tile_size.py
@@ -20,8 +20,8 @@ bl_info = {
     "name": "Auto Tile Size",
     "description": "Estimate and set the tile size that will render the fastest",
     "author": "Greg Zaal",
-    "version": (2, 7),
-    "blender": (2, 72, 0),
+    "version": (3, 0),
+    "blender": (2, 74, 0),
     "location": "Render Settings > Performance",
     "warning": "",
     "wiki_url": "http://wiki.blender.org/index.php?title=Extensions:2.6/Py/Scripts/Render/Auto_Tile_Size",
@@ -31,7 +31,7 @@ bl_info = {

 import bpy
 from bpy.app.handlers import persistent
-from math import ceil, floor
+from math import ceil, floor, sqrt

 SUPPORTED_RENDER_ENGINES = {'CYCLES', 'BLENDER_RENDER'}
@@ -55,16 +55,14 @@ class AutoTileSizeSettings(bpy.types.PropertyGroup):
         name="Target GPU Tile Size",
         items=TILE_SIZES,
         default='256',
-        description="Square dimentions of tiles",
+        description="Square dimentions of tiles for GPU rendering",
         update=_update_tile_size)
-
     cpu_choice = bpy.props.EnumProperty(
         name="Target CPU Tile Size",
         items=TILE_SIZES,
         default='32',
-        description="Square dimentions of tiles",
+        description="Square dimentions of tiles for CPU rendering",
         update=_update_tile_size)
-
     bi_choice = bpy.props.EnumProperty(
         name="Target CPU Tile Size",
         items=TILE_SIZES,
@@ -72,6 +70,37 @@ class AutoTileSizeSettings(bpy.types.PropertyGroup):
         description="Square dimentions of tiles",
         update=_update_tile_size)

+    gpu_custom = bpy.props.IntProperty(
+        name="Target Size",
+        default=256,
+        min=8,  # same as blender's own limits
+        max=65536,
+        description="Custom target tile size for GPU rendering",
+        update=_update_tile_size)
+    cpu_custom = bpy.props.IntProperty(
+        name="Target Size",
+        default=32,
+        min=8,  # same as blender's own limits
+        max=65536,
+        description="Custom target tile size for CPU rendering",
+        update=_update_tile_size)
+    bi_custom = bpy.props.IntProperty(
+        name="Target Size",
+        default=64,
+        min=8,  # same as blender's own limits
+        max=65536,
+        description="Custom target tile size",
+        update=_update_tile_size)
+
+    target_type = bpy.props.EnumProperty(
+        name="Target tile size",
+        items=(
+            ('po2', "Po2", "A choice between powers of 2 (16, 32, 64...)"),
+            ('custom', "Custom", "Choose any number as the tile size target")),
+        default='po2',
+        description="Method of choosing the target tile size",
+        update=_update_tile_size)
+
     use_optimal = bpy.props.BoolProperty(
         name="Optimal Tiles",
         default=True,
@@ -89,9 +118,16 @@ class AutoTileSizeSettings(bpy.types.PropertyGroup):
         default=False,
         description="Show extra options for more control over the calculated tile size")

+    thread_error_correct = bpy.props.BoolProperty(
+        name="Fix",
+        default=True,
+        description="Reduce the tile size so that all your available threads are used.",
+        update=_update_tile_size)
+
     # Internally used props (not for GUI)
     first_run = bpy.props.BoolProperty(default=True, options={'HIDDEN'})
     threads_error = bpy.props.BoolProperty(options={'HIDDEN'})
+    num_tiles = bpy.props.IntVectorProperty(default=(0, 0), size=2, options={'HIDDEN'})
     prev_choice = bpy.props.StringProperty(default='', options={'HIDDEN'})
     prev_engine = bpy.props.StringProperty(default='', options={'HIDDEN'})
     prev_device = bpy.props.StringProperty(default='', options={'HIDDEN'})
@@ -109,16 +145,17 @@ def ats_poll(context):
     return True

-def ats_get_engine_is_gpu(engine, device, userpref):
+def engine_is_gpu(engine, device, userpref):
     return engine == 'CYCLES' and device == 'GPU' and userpref.system.compute_device_type != 'NONE'

-def ats_get_tilesize_prop(engine, device, userpref):
-    if ats_get_engine_is_gpu(engine, device, userpref):
-        return "gpu_choice"
+def get_tilesize_prop(engine, device, userpref):
+    target_type = "_choice" if bpy.context.scene.ats_settings.target_type == 'po2' else "_custom"
+    if engine_is_gpu(engine, device, userpref):
+        return ("gpu" + target_type)
     elif engine == 'CYCLES':
-        return "cpu_choice"
-    return "bi_choice"
+        return ("cpu" + target_type)
+    return ("bi" + target_type)

 @persistent
@@ -137,9 +174,9 @@ def on_scene_update(scene):
     # scene.cycles might not always exist (Cycles is an addon)...
     device = scene.cycles.device if engine == 'CYCLES' else settings.prev_device
     border = render.use_border
-    threads = render.threads
+    threads = get_threads(context, device)

-    choice = getattr(settings, ats_get_tilesize_prop(engine, device, userpref))
+    choice = getattr(settings, get_tilesize_prop(engine, device, userpref))

     res = get_actual_res(render)
     actual_ts = (render.tile_x, render.tile_y)
@@ -150,7 +187,7 @@ def on_scene_update(scene):
                  device != settings.prev_device or
                  border != settings.prev_border or
                  threads != settings.prev_threads or
-                 choice != settings.prev_choice or
+                 str(choice) != settings.prev_choice or
                  res != settings.prev_res[:] or
                  border_res != settings.prev_border_res[:] or
                  actual_ts != settings.prev_actual_tile_size[:])
@@ -163,6 +200,61 @@ def get_actual_res(render):
     # floor is implicitly done by int conversion...
     return (int(render.resolution_x * rend_percent), int(render.resolution_y * rend_percent))

+def get_threads(context, device):
+    render = context.scene.render
+    engine = render.engine
+    userpref = context.user_preferences
+
+    if engine_is_gpu(engine, device, userpref):
+        gpu_device_str = userpref.system.compute_device
+        if 'MULTI' in gpu_device_str:
+            threads = int(gpu_device_str.split('_')[-1])
+        else:
+            threads = 1
+    else:
+        threads = render.threads
+
+    return threads
+
+def max_tile_size(threads, xres, yres):
+    ''' Give the largest tile size that will still use all threads '''
+
+    render_area = xres * yres
+    tile_area = render_area / threads
+    tile_length = sqrt(tile_area)
+
+    # lists: num x tiles, num y tiles, squareness, total tiles
+    perfect_attempts = []  # attempts with correct number of tiles
+    attempts = []  # all attempts, even if incorrect number of tiles
+
+    axes = [xres, yres]
+    funcs = [floor, ceil]
+
+    for axis in axes:
+        sec_axis = yres if axis == xres else xres
+        for func in funcs:
+            primary = func(axis / tile_length)
+            if primary > 0:
+                secondary = threads / primary
+                ts_p = axis/primary
+                ts_s = sec_axis/secondary
+                squareness = max(ts_p, ts_s) - min(ts_p, ts_s)
+                attempt = [primary if axis == xres else secondary, primary if axis != xres else secondary, squareness, primary * secondary]
+                if attempt not in attempts:
+                    attempts.append(attempt)
+                    if secondary.is_integer():  # will only be an integer if there are the right number of tiles
+                        perfect_attempts.append(attempt)
+
+    if perfect_attempts:  # prefer to use attempt that has exactly the right number of tiles
+        attempts = perfect_attempts
+
+    attempt = sorted(attempts, key=lambda k: k[2])[0]  # pick set with most square tiles
+    numtiles_x = round(attempt[0])
+    numtiles_y = round(attempt[1])
+    tile_x = ceil(xres / numtiles_x)
+    tile_y = ceil(yres / numtiles_y)
+
+    return (tile_x, tile_y)

 def do_set_tile_size(context):
     if not ats_poll(context):
@@ -176,7 +268,6 @@ def do_set_tile_size(context):
     engine = render.engine
     device = scene.cycles.device if engine == 'CYCLES' else settings.prev_device
     border = render.use_border
-    threads = render.threads

     realxres, realyres = xres, yres = res = get_actual_res(scene.render)

@@ -184,11 +275,12 @@ def do_set_tile_size(context):
         xres = round(xres * (render.border_max_x - render.border_min_x))
         yres = round(yres * (render.border_max_y - render.border_min_y))

-    choice = getattr(settings, ats_get_tilesize_prop(engine, device, userpref))
+    choice = getattr(settings, get_tilesize_prop(engine, device, userpref))
     target = int(choice)

     numtiles_x = ceil(xres / target)
     numtiles_y = ceil(yres / target)
+    settings.num_tiles = (numtiles_x, numtiles_y)
     if settings.use_optimal:
         tile_x = ceil(xres / numtiles_x)
         tile_y = ceil(yres / numtiles_y)
@@ -198,20 +290,25 @@ def do_set_tile_size(context):

     print("Tile size: %dx%d (%dx%d tiles)" % (tile_x, tile_y, ceil(xres / tile_x), ceil(yres / tile_y)))

-    render.tile_x = tile_x
-    render.tile_y = tile_y
-
     # Detect if there are fewer tiles than available threads
-    if ((numtiles_x * numtiles_y) < threads) and not ats_get_engine_is_gpu(engine, device, userpref):
+    threads = get_threads(context, device)
+    if ((numtiles_x * numtiles_y) < threads):
         settings.threads_error = True
+        if settings.thread_error_correct:
+            tile_x, tile_y = max_tile_size(threads, xres, yres)
+            settings.num_tiles = (ceil(xres/tile_x), ceil(yres/tile_y))
     else:
         settings.threads_error = False

+    render.tile_x = tile_x
+    render.tile_y = tile_y
+
+
     settings.prev_engine = engine
     settings.prev_device = device
     settings.prev_border = border
     settings.prev_threads = threads
-    settings.prev_choice = choice
+    settings.prev_choice = str(choice)
     settings.prev_res = res
     settings.prev_border_res = (render.border_min_x, render.border_min_y, render.border_max_x, render.border_max_y)
     settings.prev_actual_tile_size = (tile_x, tile_y)
@@ -252,16 +349,21 @@ def ui_layout(engine, layout, context):
     row.prop(settings, "is_enabled", toggle=True)
     row.prop(settings, "use_advanced_ui", toggle=True, text="", icon='PREFERENCES')

-    sub = col.column(align=True)
+    sub = col.column(align=False)
     sub.enabled = settings.is_enabled

     if settings.use_advanced_ui:
-        sub.label("Target tile size:")
+        row = sub.row(align=True)
+        row.label("Target tile size:")
+        row.separator()
+        row.prop(settings, "target_type", expand=True)

         row = sub.row(align=True)
-        row.prop(settings, ats_get_tilesize_prop(engine, device, userpref), expand=True)
+        row.prop(settings, get_tilesize_prop(engine, device, userpref), expand=True)
         sub.prop(settings, "use_optimal", text="Calculate Optimal Size")

+        sub.label("Number of tiles: %s x %s (Total: %s)" % (settings.num_tiles[0], settings.num_tiles[1], settings.num_tiles[0] * settings.num_tiles[1]))
+
     if settings.first_run:
         sub = layout.column(align=True)
         sub.operator("render.autotilesize_set", text="First-render fix", icon='ERROR')
@@ -272,9 +374,11 @@ def ui_layout(engine, layout, context):
     if (render.tile_x / render.tile_y > 2) or (render.tile_x / render.tile_y < 0.5):  # if not very square tile
         sub.label(text="Warning: Tile size is not very square", icon='ERROR')
         sub.label(text="    Try a slightly different resolution")
-        sub.label(text="    or choose \"Exact\" above")
     if settings.threads_error:
-        sub.label(text="Warning: Fewer tiles than render threads", icon='ERROR')
+        row = sub.row(align=True)
+        row.alignment = 'CENTER'
+        row.label(text="Warning: Fewer tiles than threads.", icon='ERROR')
+        row.prop(settings, 'thread_error_correct')

 def menu_func_cycles(self, context):

This is really a different bug that the one that was reopened
but I didn't know that when I reopened it.
It shouldn't happen that vertices get repeated in a face, but
when doing it happens in extreme cases, and for now am fixing by
just removing those duplicates before calling the make-face routine.

Crash caused by some vertices in the input model being doubled.
The code to test for the maximum inset deduped the vertices
but the faces were not remapped. Fixed by having that test code
not dedup the vertices.

Adding an image sequence now works for Shader nodes too (not sure when this broke?)
Also improved/fixed the logic of figuring out what image sequence the user is trying to add - sometimes they select only one frame, sometimes they select all the frames (Blender itself is inconsistent about this behavior)

Tracker: https://developer.blender.org/T44561

New UI, New Addons, Better organization & removal of 'Lesser Object
Types'

Merged & removed "lesser" & 'larger" add mesh scripts.
Menu updates & re-organization & modernize.
New: Add Vert, Round Cube, Menger Cube, Brilliant Diamond, Parent to
Empty
Removed: Add Mesh: Sqorus, Trapezohedron, Wedge, Polysphere.
Merged add_mesh_symmetrical_empty from contrib
Documentation:
http://wiki.blender.org/index.php/Extensions:2.6/Py/Scripts/Add_Mesh/Add_Mesh_Extra_Objects

It would export animations from all compatible actions - even for objects that were
actually not animated at all! This would lead to undesired behavior esp. for
simple objects scenes with only one or two animated.

Now we only export all compatible actions for a given object if it is actually
animated.

Was a TODO actually, now we try to match better illum values with current mesh material.
Also, using mirror color as Ka instead of ambient one, when mirror is enabled in blen mat.

Patch D1210 by @Leon95