OpenCl and the tools for it. Reviews and impressions. - page 3

New comment
 
This is the use of RAM to store or exchange data, e.g. between two terminals or even several applications.

It simply uses reserved areas of RAM instead of files. The idea is the same, but it works much faster.

 

Mapping is mapping, it's just that a file object can be created directly in RAM.

I just don't understand why? Alexei clearly wrote that everything is in the init.

 

I repeat, I proposed it not for Alexey's problem, but as a way to speed up calculations in general)

That is, organizing a cache of sorts to use the data in subsequent iterations.

 
OnGoing:
This is the use of RAM to store or exchange data, e.g. between two terminals or even several applications.

It simply uses reserved areas of RAM instead of files. The idea is the same, but it works much faster.


forget about this stripped-down DLL from it.

You have a fully functional version to work with mapping.

 
Mathemat:

And how do you propose to do it - without referring to coding external to MQL4?
I already suggested - through files. But it's long and bothersome.

Alternatively, you can make a special dll.
The set of parameters in memory will be associated with a certain result.
You can even make the mapping trickier, like this:
DateTime - {Calculation parameters} - Result[s].
This already caches calculations on every bar.

Fuck mapping and files. What we need is associative storage.
 
sergeev:

forget about this stripped-down DLL from it.

You have a fully functional version to work with mapping.

Wow, thanks. Looking forward to examples on how to use it)
 
jartmailru: Alternatively, you could make a special dll.

Asked how without a dll.

OK, never mind. It's easier to make heavy C/C++ computations and speed up execution by several times using SSE*, Open MP and IPP. And instead of 10 seconds on MQL4 you'll get 1 second - even on my budget dual-core.

 
Mathemat:

I asked him how he could work without dll.

Okay, never mind. It's easier to implement heavy computations in C/C++ and speed up execution by several times using SSE*, Open MP and IPP. And instead of 10 seconds in MQL4 you'll get 1 second - even on my budget dual core.

MQL5 is 20 times faster than MQL4.

C++ is 6 times faster than MQL5 (when using libraries that automatically parallelize execution).

Total: 20*6=120 times.

If you use GPU calculations, it will be even faster.

TOTAL: 10/120=0.083c.

something like this.

 
joo:

MQL5 is 20 times faster than MQL4.

I would like to believe it. But in practice optimization in a multicore CPU in MT5 is much slower than single-threaded optimization in MT4.

It is quite possible that it only seems as if the code in MT5 is barely wiggling?

I should take some benchmark-algorithm and use it in the start() function of MT4 and OnTick() of MT5 and run it on a single core in Expert Advisors with one dummy input parameter changing from 0 to 9999 without genetics to make sure more exactly. But where to get this single core?

 
Reshetov:

Except that in practice optimization on a multi-core CPU in MT5 is noticeably slower than single-threaded optimization in MT4.

No, this is a tester. Before the Championship we measured the run time - the run time of an economical light Expert Advisor was much less than the time spent for testing.

However this information is not 100%, I may lie, I do not remember.

12345678910...29
New comment