switched off bounds checking in inner loop; added makefile; used curr…

[floswald]

hi @jesusfv

i thought its time for an update on those results. this PR contains the same code but uses the current julia release v0.6. compared to the 1.92 seconds it took on v0.2, now it takes 1.16 seconds on v0.6 on my computer.

cheers
florian

[jesusfv]

Hi @floswald

Thanks for the suggestions. I updated today my own version with some minor changes to make it current with Julia 0.6.2 and use \alpha and \beta unicode characters to make the code easier to read.

My improvements in speed are not so large. I go from 1.91 seconds to 1.60 in my code and, if I add your @inbounds, to 1.49 (over 10 runs).

Any suggestion why your improvements are so large?

[floswald]

hmm, that is interesting (and I have no clue as to why that differs so much). it's interesting that our starting times (1.92) are identical on v0.2. I ran that on an iMac 4Ghz processor, you? i have to say that there was some variation in my timings, but they were always in the range 1.16-1.3. It could be that julia got smarter in emitting code for the specific processor type in the most recent version v0.6, but that's just a guess.

[jesusfv]

Processor Name: Intel Core i7
Processor Speed: 2.8 GHz

The biggest change in relative performance is in R, which now operates a byte code compiler. The basic R code runs now in around 20 seconds, instead of 345.

Matlab has also dropped to around 3.5 seconds, as they have improved their JIT.