Comment #14 on issue 3903 by [email protected]: Math.log10 is unexpectedly
slow
https://code.google.com/p/v8/issues/detail?id=3903
Updated again with version of rtoy's fdlibm log port that uses both cached
buffers and inlined reinterpret-cast functions. Together, these improved
performance by ~45x, though the native C Math.log remains 1.4x faster.
Take-aways:
1. For the Taylor/LUT approximation, manual inlining producing a 1.2x
improvement for a reduction of 4 function calls per element.
2. For the fdlibm log port to JS, manual inlining oddly produced a 4.5x
improvement for a reduction of only 3 or 4 function calls per element.
Results: (Chrome 40 / OSX)
------------------------------
n: 262144 | passes: 32
------------------------------
1. Test log*recip: 966 ms.
2. Test Math.log10: 1131 ms.
3. Test log10fc_Inline: 1059 ms.
4. Test log10fc: 1227 ms.
5. Test log10rtoyJS(c): 6103 ms.
6. Test log10rtoyJS(c)(n): 1366 ms.
Attachments:
log_perf_sample.html 28.7 KB
--
You received this message because this project is configured to send all
issue notifications to this address.
You may adjust your notification preferences at:
https://code.google.com/hosting/settings
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
