From 51c0bc94efe34deb66b695447c2592870aecb638 Mon Sep 17 00:00:00 2001
From: Thomas Barlow <github@mysterymayhem.co.uk>
Date: Fri, 30 Jun 2023 17:34:48 +0100
Subject: [PATCH] FBX Import: Read known sequential bytes simultaneously

The start of each element and the start of each array property have a
consistent series of values with fixed sizes. Reading these series of
values with a single read call and single unpack call is slightly faster
than reading them individually.

This reduces the time taken to parse fbx files to about 90% to 95% of
the original time. Array decompression makes up about half of the time
taken currently, but this could be multithreaded, in which case, this
patch would reduce the time to about 80% to 90% instead.
---
 io_scene_fbx/parse_fbx.py | 43 +++++++++++++++++++++++----------------
 1 file changed, 26 insertions(+), 17 deletions(-)

diff --git a/io_scene_fbx/parse_fbx.py b/io_scene_fbx/parse_fbx.py
index 154bbf494..19cd1f78c 100644
--- a/io_scene_fbx/parse_fbx.py
+++ b/io_scene_fbx/parse_fbx.py
@@ -20,7 +20,7 @@ from . import data_types
 # that the sub-scope exists (i.e. to distinguish between P: and P : {})
 _BLOCK_SENTINEL_LENGTH = ...
 _BLOCK_SENTINEL_DATA = ...
-read_fbx_elem_uint = ...
+read_fbx_elem_start = ...
 _IS_BIG_ENDIAN = (__import__("sys").byteorder != 'little')
 _HEAD_MAGIC = b'Kaydara FBX Binary\x20\x20\x00\x1a\x00'
 from collections import namedtuple
@@ -32,10 +32,6 @@ def read_uint(read):
     return unpack(b'<I', read(4))[0]
 
 
-def read_uint64(read):
-    return unpack(b'<Q', read(8))[0]
-
-
 def read_ubyte(read):
     return unpack(b'B', read(1))[0]
 
@@ -46,10 +42,24 @@ def read_string_ubyte(read):
     return data
 
 
+def read_array_params(read):
+    return unpack(b'<III', read(12))
+
+
+def read_elem_start32(read):
+    end_offset, prop_count, _prop_length, elem_id_size = unpack(b'<IIIB', read(13))
+    elem_id = read(elem_id_size) if elem_id_size else b""
+    return end_offset, prop_count, elem_id
+
+
+def read_elem_start64(read):
+    end_offset, prop_count, _prop_length, elem_id_size = unpack(b'<QQQB', read(25))
+    elem_id = read(elem_id_size) if elem_id_size else b""
+    return end_offset, prop_count, elem_id
+
+
 def unpack_array(read, array_type, array_stride, array_byteswap):
-    length = read_uint(read)
-    encoding = read_uint(read)
-    comp_len = read_uint(read)
+    length, encoding, comp_len = read_array_params(read)
 
     data = read(comp_len)
 
@@ -89,18 +99,17 @@ read_data_dict = {
 #   * The NULL block marking end of nested stuff switches from 13 bytes long to 25 bytes long.
 #   * The FBX element metadata (end_offset, prop_count and prop_length) switch from uint32 to uint64.
 def init_version(fbx_version):
-    global _BLOCK_SENTINEL_LENGTH, _BLOCK_SENTINEL_DATA, read_fbx_elem_uint
+    global _BLOCK_SENTINEL_LENGTH, _BLOCK_SENTINEL_DATA, read_fbx_elem_start
 
     _BLOCK_SENTINEL_LENGTH = ...
     _BLOCK_SENTINEL_DATA = ...
-    read_fbx_elem_uint = ...
 
     if fbx_version < 7500:
         _BLOCK_SENTINEL_LENGTH = 13
-        read_fbx_elem_uint = read_uint
+        read_fbx_elem_start = read_elem_start32
     else:
         _BLOCK_SENTINEL_LENGTH = 25
-        read_fbx_elem_uint = read_uint64
+        read_fbx_elem_start = read_elem_start64
     _BLOCK_SENTINEL_DATA = (b'\0' * _BLOCK_SENTINEL_LENGTH)
 
 
@@ -108,14 +117,14 @@ def read_elem(read, tell, use_namedtuple):
     # [0] the offset at which this block ends
     # [1] the number of properties in the scope
     # [2] the length of the property list
-    end_offset = read_fbx_elem_uint(read)
+    # [3] elem name length
+    # [4] elem name of the scope/key
+    # read_fbx_elem_start does not return [2] because we don't use it and does not return [3] because it is only used to
+    # get [4].
+    end_offset, prop_count, elem_id = read_fbx_elem_start(read)
     if end_offset == 0:
         return None
 
-    prop_count = read_fbx_elem_uint(read)
-    prop_length = read_fbx_elem_uint(read)
-
-    elem_id = read_string_ubyte(read)        # elem name of the scope/key
     elem_props_type = bytearray(prop_count)  # elem property types
     elem_props_data = [None] * prop_count    # elem properties (if any)
     elem_subtree = []                        # elem children (if any)
-- 
2.30.2