[Bf-extensions-cvs] [440216cf] blender2.8: OBJ Import: Improve parsing performances (overall 5% speedup).
Bastien Montagne
noreply at git.blender.org
Tue Nov 20 12:35:58 CET 2018
Commit: 440216cf89612793f3afa7d5d2da37457f3c219b
Author: Bastien Montagne
Date: Tue Nov 20 12:18:21 2018 +0100
Branches: blender2.8
https://developer.blender.org/rBA440216cf89612793f3afa7d5d2da37457f3c219b
OBJ Import: Improve parsing performances (overall 5% speedup).
This commits mainly add a specific, optimized handling for vertex data
(coos, nors, UVs) expressed on single lines. Nearly all OBJ files I've
come accross use a single line per vertex data.
This code cand be up to 40% quicker on vertex data parsing, making
typical parsing about 8% quicker, and typical import about 5% quicker.
Note that I tried applying same kind of optimizations on faces parsing,
but we need to perform rather more complex checks there, so could not
get any interesting improvements.
Also, fixing some logical flaw in invalid nor/uvtex indices, which also
gives marginal speed improvements.
===================================================================
M io_scene_obj/__init__.py
M io_scene_obj/import_obj.py
===================================================================
diff --git a/io_scene_obj/__init__.py b/io_scene_obj/__init__.py
index 1f2ff4e9..721802e1 100644
--- a/io_scene_obj/__init__.py
+++ b/io_scene_obj/__init__.py
@@ -21,7 +21,7 @@
bl_info = {
"name": "Wavefront OBJ format",
"author": "Campbell Barton, Bastien Montagne",
- "version": (3, 4, 4),
+ "version": (3, 5, 0),
"blender": (2, 80, 0),
"location": "File > Import-Export",
"description": "Import-Export OBJ, Import OBJ mesh, UV's, materials and textures",
diff --git a/io_scene_obj/import_obj.py b/io_scene_obj/import_obj.py
index d4007c78..01bd0e67 100644
--- a/io_scene_obj/import_obj.py
+++ b/io_scene_obj/import_obj.py
@@ -943,12 +943,16 @@ def load(context,
face_vert_nor_indices = None
face_vert_tex_indices = None
face_vert_nor_valid = face_vert_tex_valid = False
+ verts_loc_len = verts_nor_len = verts_tex_len = 0
face_items_usage = set()
face_invalid_blenpoly = None
prev_vidx = None
face = None
vec = []
+ quick_vert_failures = 0
+ skip_quick_vert = False
+
progress.enter_substeps(3, "Parsing OBJ file...")
with open(filepath, 'rb') as f:
for line in f:
@@ -959,17 +963,39 @@ def load(context,
line_start = line_split[0] # we compare with this a _lot_
- if line_start == b'v' or context_multi_line == b'v':
- context_multi_line = handle_vec(line_start, context_multi_line, line_split, b'v', verts_loc, vec, 3)
-
- elif line_start == b'vn' or context_multi_line == b'vn':
- context_multi_line = handle_vec(line_start, context_multi_line, line_split, b'vn', verts_nor, vec, 3)
+ # Handling vertex data are pretty similar, factorize that.
+ # Also, most OBJ files store all those on a single line, so try fast parsing for that first,
+ # and only fallback to full multi-line parsing when needed, this gives significant speed-up
+ # (~40% on affected code).
+ if line_start == b'v':
+ vdata, vdata_len, do_quick_vert = (verts_loc, 3, not skip_quick_vert)
+ elif line_start == b'vn':
+ vdata, vdata_len, do_quick_vert = (verts_nor, 3, not skip_quick_vert)
+ elif line_start == b'vt':
+ vdata, vdata_len, do_quick_vert = verts_tex, 2, not skip_quick_vert
+ elif context_multi_line == b'v':
+ vdata, vdata_len, do_quick_vert = verts_loc, 3, False
+ elif context_multi_line == b'vn':
+ vdata, vdata_len, do_quick_vert = verts_nor, 3, False
+ elif context_multi_line == b'vt':
+ vdata, vdata_len, do_quick_vert = verts_tex, 2, False
+ else:
+ vdata_len = 0
+
+ if vdata_len:
+ if do_quick_vert:
+ try:
+ vdata.append(tuple(map(float_func, line_split[1:vdata_len + 1])))
+ except:
+ do_quick_vert = False
+ # In case we get too many failures on quick parsing, force fallback to full multi-line one.
+ # Exception handling can become costly...
+ quick_vert_failures += 1
+ if quick_vert_failures > 10000:
+ skip_quick_vert = True
+ if not do_quick_vert:
+ context_multi_line = handle_vec(line_start, context_multi_line, line_split, b'v', vdata, vec, vdata_len)
- elif line_start == b'vt' or context_multi_line == b'vt':
- context_multi_line = handle_vec(line_start, context_multi_line, line_split, b'vt', verts_tex, vec, 2)
-
- # Handle faces lines (as faces) and the second+ lines of fa multiline face here
- # use 'f' not 'f ' because some objs (very rare have 'fo ' for faces)
elif line_start == b'f' or context_multi_line == b'f':
if not context_multi_line:
line_split = line_split[1:]
@@ -979,14 +1005,17 @@ def load(context,
_1, _2, _3, face_invalid_blenpoly) = face
faces.append(face)
face_items_usage.clear()
+ verts_loc_len = len(verts_loc)
+ verts_nor_len = len(verts_nor)
+ verts_tex_len = len(verts_tex)
# Else, use face_vert_loc_indices and face_vert_tex_indices previously defined and used the obj_face
context_multi_line = b'f' if strip_slash(line_split) else b''
for v in line_split:
obj_vert = v.split(b'/')
- idx = int(obj_vert[0]) - 1
- vert_loc_index = (idx + len(verts_loc) + 1) if (idx < 0) else idx
+ idx = int(obj_vert[0]) # Note that we assume here we cannot get OBJ invalid 0 index...
+ vert_loc_index = (idx + verts_loc_len) if (idx < 1) else idx - 1
# Add the vertex to the current group
# *warning*, this wont work for files that have groups defined around verts
if use_groups_as_vgroups and context_vgroup:
@@ -1004,15 +1033,15 @@ def load(context,
# formatting for faces with normals and textures is
# loc_index/tex_index/nor_index
if len(obj_vert) > 1 and obj_vert[1] and obj_vert[1] != b'0':
- idx = int(obj_vert[1]) - 1
- face_vert_tex_indices.append((idx + len(verts_tex) + 1) if (idx < 0) else idx)
+ idx = int(obj_vert[1])
+ face_vert_tex_indices.append((idx + verts_tex_len) if (idx < 1) else idx - 1)
face_vert_tex_valid = True
else:
face_vert_tex_indices.append(0)
if len(obj_vert) > 2 and obj_vert[2] and obj_vert[2] != b'0':
- idx = int(obj_vert[2]) - 1
- face_vert_nor_indices.append((idx + len(verts_nor) + 1) if (idx < 0) else idx)
+ idx = int(obj_vert[2])
+ face_vert_nor_indices.append((idx + verts_nor_len) if (idx < 1) else idx - 1)
face_vert_nor_valid = True
else:
face_vert_nor_indices.append(0)
More information about the Bf-extensions-cvs
mailing list