benkyd/tinyobjloader
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use a lookup table to speed up float parsing

Browse Source 
The pow() function is pretty expensive, so creating a small lookup
table for the first few negative powers of ten provides a big speedup.

Parse times for some large .obj files (without asan):

        File A  File B  File C
Before  2500ms   573ms   545ms
After   1239ms   294ms   271ms
This commit is contained in:
Björn Steinbrink
2016-10-24 11:53:00 +02:00
parent ebdbd8a231
commit aa670fe91e
1 changed files with 14 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
tiny_obj_loader.h
View File

@@ -418,8 +418,21 @@ static bool tryParseDouble(const char *s, const char *s_end, double *result) {
read = 1; read = 1;
end_not_reached = (curr != s_end); end_not_reached = (curr != s_end);
while (end_not_reached && IS_DIGIT(*curr)) { while (end_not_reached && IS_DIGIT(*curr)) {
static const double pow_lut[] = {
1.0,
0.1,
0.01,
0.001,
0.0001,
0.00001,
0.000001,
0.0000001,
};
const int lut_entries = sizeof pow_lut / sizeof pow_lut[0];
// NOTE: Don't use powf here, it will absolutely murder precision. // NOTE: Don't use powf here, it will absolutely murder precision.
mantissa += static_cast<int>(*curr - 0x30) * pow(10.0, -read); mantissa += static_cast<int>(*curr - 0x30) *
(read < lut_entries ? pow_lut[read] : pow(10.0, -read));
read++; read++;
curr++; curr++;
end_not_reached = (curr != s_end); end_not_reached = (curr != s_end);