integer patch: performance
[
Date Prev][
Date Next][
Thread Prev][
Thread Next]
[
Date Index]
[
Thread Index]
- Subject: integer patch: performance
- From: Asko Kauppi <askok@...>
- Date: 2006年1月21日 03:50:34 +0200
I've now finished (software never is..) the Integer patch for Lua 5.1
(rc).
Performance figures (running test/factorial.lua 1000 times in a loop):
The values are 'user' column of 'time' output.
** iMac PowerPC G4 700MHz/512 MB, OS X 10.4.4 **
Sat Jan 21 00:51:29 EET 2006 / Asko Kauppi
float: 1.07 1.10 1.10 1.09 (1.0900 = 100%) 143 792 bytes
float+int: 1.11 1.08 1.08 1.08 (1.0875 = 99.8%) 147 920 bytes (+4128)
double: 1.14 1.12 1.11 1.11 (1.1200 = 102.8%) 143 780 bytes (-12)
double+int: 1.14 1.11 1.11 1.11 (1.1175 = 102.5%) 147 908 bytes (+4116)
** NSLU2 ARM 266MHz/32 MB, Linux unslung 2.4.22 **
Sat Jan 21 02:34:22 EET 2006 / Asko Kauppi
float: 5.17 5.10 5.13 5.07 (5.1175 = 100%) 126 472 bytes
float+int: 3.34 3.39 3.40 3.38 (3.3775 = 66.0%) 129 892 bytes (+3420)
double: 5.78 5.82 5.82 5.86 (5.8200 = 113.7%) 127 224 bytes (+752)
double+int: 3.52 3.56 3.53 3.54 (3.5375 = 69.1%) 130 588 bytes (+4116)
The good news is:
- it remains completely transparent to both Lua C API and scripts
- it does not degrade FPU performance (at least on PowerPC, x86
not tested)
- it gives >30% speed boost on non-FPU platform (ARM)
(or, comparing double vs. float+int, the boost is 42%)
- it has moderate size effect, but not major (4k)
Now, I'd like to gain more testing results, meaning more hardware
where the
thing is run. Anything with basic *nix utilities shall be easy, maybe
needing
some changes in a Makefile and that's it.
Also, real world testing would prove to be more valuable than some
particular
test app (since they're always biased, and simple).
-asko