OWB JS test

could you try this url pls?

http://webkit.org/perf/sunspider-0.9/sunspider.html

these are my results (FIREFOX ON WIN):

============================================
RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total: 28859.8ms +/- 3.4%
--------------------------------------------

3d: 4459.0ms +/- 4.2%
cube: 1418.4ms +/- 3.7%
morph: 2250.0ms +/- 5.3%
raytrace: 790.6ms +/- 19.2%

access: 2889.8ms +/- 8.9%
binary-trees: 306.0ms +/- 3.3%
fannkuch: 590.6ms +/- 6.3%
nbody: 1584.0ms +/- 18.0%
nsieve: 409.2ms +/- 42.0%

bitops: 5355.8ms +/- 3.3%
3bit-bits-in-byte: 409.2ms +/- 2.2%
bits-in-byte: 368.8ms +/- 4.7%
bitwise-and: 4005.8ms +/- 2.4%
nsieve-bits: 572.0ms +/- 29.8%

controlflow: 231.0ms +/- 3.6%
recursive: 231.0ms +/- 3.6%

crypto: 1246.6ms +/- 12.4%
aes: 378.0ms +/- 2.2%
md5: 384.2ms +/- 2.7%
sha1: 484.4ms +/- 31.1%

date: 6390.6ms +/- 1.1%
format-tofte: 2584.4ms +/- 1.7%
format-xparb: 3806.2ms +/- 0.7%

math: 2649.8ms +/- 4.9%
cordic: 1121.8ms +/- 8.3%
partial-sums: 922.0ms +/- 1.4%
spectral-norm: 606.0ms +/- 24.0%

regexp: 1028.2ms +/- 5.6%
dna: 1028.2ms +/- 5.6%

string: 4609.0ms +/- 4.7%
base64: 1309.4ms +/- 2.4%
fasta: 765.8ms +/- 22.8%
tagcloud: 749.8ms +/- 4.5%
unpack-code: 1287.2ms +/- 1.7%
validate-input: 496.8ms +/- 5.1%

@afxgroup

OS X and FF2


============================================
RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total: 14205.4ms +/- 0.9%
--------------------------------------------

3d: 2124.4ms +/- 1.0%
cube: 557.2ms +/- 3.4%
morph: 1252.2ms +/- 0.7%
raytrace: 315.0ms +/- 1.5%

access: 1200.4ms +/- 1.3%
binary-trees: 138.0ms +/- 0.6%
fannkuch: 323.4ms +/- 5.9%
nbody: 516.2ms +/- 2.0%
nsieve: 222.8ms +/- 0.7%

bitops: 4081.0ms +/- 1.2%
3bit-bits-in-byte: 260.8ms +/- 1.0%
bits-in-byte: 252.0ms +/- 0.5%
bitwise-and: 3308.8ms +/- 1.1%
nsieve-bits: 259.4ms +/- 5.8%

controlflow: 104.4ms +/- 1.4%
recursive: 104.4ms +/- 1.4%

crypto: 608.6ms +/- 3.9%
aes: 229.4ms +/- 1.9%
md5: 192.4ms +/- 8.2%
sha1: 186.8ms +/- 6.2%

date: 1787.0ms +/- 0.7%
format-tofte: 660.2ms +/- 1.6%
format-xparb: 1126.8ms +/- 0.9%

math: 1189.6ms +/- 6.8%
cordic: 548.8ms +/- 11.8%
partial-sums: 404.8ms +/- 2.3%
spectral-norm: 236.0ms +/- 7.3%

regexp: 545.8ms +/- 0.2%
dna: 545.8ms +/- 0.2%

string: 2564.2ms +/- 2.8%
base64: 589.0ms +/- 1.2%
fasta: 475.2ms +/- 4.1%
tagcloud: 446.2ms +/- 3.0%
unpack-code: 770.0ms +/- 2.5%
validate-input: 283.8ms +/- 9.2%

@afxgroup

OWB 1.15 on ?A1:

@afxgroup

OWB 1.15 A1XEG4

Attach file:

---

  (64.82 KB)


  (142.87 KB)

well, on my A1 global time is 59000 millisecs.. so our Amiga and OWB even if it is a micro or an a1. would be great to see how linux and owb is more fast than Amiga

@AfxGroup

Here are the results with an AmigaOne XE G3/800:

And the results on my AmigaOne XE-G2 933 MHz...

http://murakami.free.fr/Annonce/OWB.jpg

@afxgroup

Quote:Since my linux OWB is a debug build it's not comparable, but here are the linux OWB results with my 750FX CPU anyway, and for comparison (to the AmigaOS4 OWB on a 750FX CPU, not to the linux debug version) the results from Opera 9.50 on my A1.

@thread

PPCMotorola MPC 7447/7457 @ 1266MHz Apollo/AltiVec(tm)
512MB RAM

@Raziel

thats it how to o/clock my A1 to 1.5 ghz so i can get better results than Raziel

@joerg

so is not an owb prblem if you get those results with owb on that cpu.. :-/
and so? what could be?
Another thing. Joking with mlayer in the past, i've seen also that changing gcc options will gain a lot of speed. I don't know if this could be the same with owb but we could give a try to see if the speed increase.

for example. the mplayer CFLAGS are:

-O4 -falign-loops=16 -fsingle-precision-constant -pthread -pipe -fexpensive-optimizations -fschedule-insns2 -fno-strict-aliasing -ffast-math -funroll-all-loops -funswitch-loops -finline-functions -pipe -fomit-frame-pointer

even the singe -funroll-all-loops could give a speed boost even if increase the exe size.
maybe with mplayer is more visible since is a progarm that needs to be optimised in every single part.. but, why don't try?

@poweramiga

Quote:

/me looks up Helge's phone number

@afxgroup

http://freenet-homepage.de/amigo1/owb-js-test.jpg

@afxgroup

Quote:What problem? Others posted the results of this benchmark with the AmigaOS4 OWB on a 750 CPU already, it's about half of the speed of Opera.

As I wrote my linux OWB results are not comparable at all since it's a debug build (compiled with -O0, debugging macros and code enabled).


Quote:Did you roll a dice to get these options?
-O4 doesn't even exist, and AFAIK never did (there only were some proposals for an -O4). GCC supports -O0, -O1, -O2, -Os and -O3. IMHO GCC should abort with an error message for unsupported options, but unfortunately it doesn't but silently treats -Ox with x beeing a number > 3 as -O3 instead. It's probably for compatibility to other compilers which do support -O4, -O5, etc.
-fexpensive-optimizations and -fschedule-insns2 are enabled at -O2 already, -funswitch-loops and -finline-functions with -O3, i.e. they are redundant in your CFLAGS, and -fno-strict-aliasing disables the strict aliasing optimisation enabled by -O2, -Os and -O3.


Quote:You can try if you want, but I very much doubt you'll be able to get anything significantly faster than my -Os -DNDEBUG builds, for such a large and complex program "small" code size is more important than anything else to keep as much of the code in the caches as possible, not only on the classic Amigas with no L2 cache at all, the RAM on the A1 is very slow as well.

I don't have the time to wait about an hour for each possible option combination, especially not for a linux release build just to run this benchmark on linux as well since it's obvious that the results will be identical to the ones of the AmigaOS4 version, there is no rendering or anything else platform specific involved. I only need the linux OWB for testing if something only fails in the AmigaOS4 version, or if it's a bug in all versions of OWB, which was the case for all tests I did.