On Wed, Jul 2, 2014 at 12:22 PM, Eliot Miranda <eliot.miranda@gmail.com> wrote:
On Wed, Jul 2, 2014 at 11:43 AM, kilon alios <kilon.alios@gmail.com> wrote:
0.00000000549181 seconds per message is just insanely low number . This is not even in the realm of nanoseconds. Is this number real ? Because I am very skeptical that one can get this kind of performance from a dynamic language even with a JIT VM.
Here's a measurement on my 2.2GHz Mac Book Pro Core i7:
| r | {[r := 34 benchFib] timeToRun. r} #(140 18454929)
benchFib is nfib that adds one for each call, so the answer is the number of calls necessary to compute the answer.
benchFib ^self < 2 ifTrue: [1] ifFalse: [(self-1) benchFib + (self-2) benchFib + 1]
So that's 18 million calls in 140 milliseconds, or about 132 calls per microsecond. That's indeed in the realm of nanoseconds, 13.2 nsecs per call.
Ugh, let me try again... that's 184549290 / 0.14 = 1.3182092142857141e9 calls per second, which is 0.76 nsecs per call.
On Wed, Jul 2, 2014 at 9:08 PM, Camille Teruel <camille.teruel@gmail.com> wrote:
A spur image :) awesome !! Thanks Esteban & Guille! I'll start adapting the new class builder to Spur next wednesday when I come back from holidays.
On 2 juil. 2014, at 16:13, Esteban Lorenzano <estebanlm@gmail.com> wrote:
Hi,
Iâve been working on prepare Pharo to run with the new Spur VM⦠and finally this week Guille and I sit together and make a huge advance :) here a screenshot of pharo 4 running with spur: <Screen Shot 2014-07-02 at 15.58.17.png> for more information, same image before migration, with a regular cogvm gives this numbers: '1297023432 bytecodes/sec; 161029354 sends/secâ, so that means that the tiny benchmarks run at 166% the speed of the old vm⦠Of course this is just one benchmark, but Iâm very impressed :)
Now, it is still not usable but Iâm confident I can put a jenkins job to work very soon :)
Esteban
ps: before you start asking: Spur WILL NOT be available for Pharo3, it will be part of Pharo4.
-- best, Eliot
-- best, Eliot