Compilation of MQL5 programmes with AVX / AVX2 + FMA3 / AVX512 + FMA3 instruction set from build 3902 - page 10

New comment
 

The nastiest thing happened (b4040): backtest result depends on compilation instruction set selection.

The sizes of tst files are 1.5 times different: 13 (AVX) and 17 (X64 Regular) megs. To say that it sucks is nothing to say.

 
fxsaber #:

The nastiest thing happened (b4040): the result of the backtest depends on the choice of compilation instruction set.

The sizes of tst-files differ 1.5 times: 13 (AVX) and 17 (X64 Regular) megs. To say that it sucks is nothing to say.

Perhaps it is because of optimisation. Optimisation may break the program logic. Try without optimisation, then the logic will be in the style of your text.
 
fxsaber #:

The nastiest thing happened (b4040): the result of the backtest depends on the choice of compilation instruction set.

The sizes of tst-files differ 1.5 times: 13 (AVX) and 17 (X64 Regular) megs. To say that it sucks is nothing to say.

Does the test just work correctly?

 
fxsaber #:

The nastiest thing happened (b4040): the result of the backtest depends on the choice of compilation instruction set.

The sizes of tst-files differ 1.5 times: 13 (AVX) and 17 (X64 Regular) megs. To say that it sucks is not to say anything.

If the different test behaviour is different, look for UB. If only the size bothers you, it's normal)
 
Aleksey Vyazmikin #:

Is just the test working properly?

Need a benchmark to answer this question.

 
Vladimir Simakov #:
If different test behaviour, look for your UB. If it's just the size that confuses you, that's fine).

What is it?

 
fxsaber #:

What is it?

Undefined behaviour, it looks like. It's undefined behaviour.
 
traveller00 #:
Undefined behaviour, it seems. It's undefined behaviour.

Then the advice is worthless.

 
Let's get to the bottom of this - it shouldn't happen
 
fxsaber #:

Need a benchmark to answer this question.

I'm not talking about where it is correct, but about the testing mode.

You can try to start with simple Expert Advisors trading at the opening, i.e. exclude the influence of tick history.

On a small number of trades compare the report and visually study the discrepancy.

1...3456789101112
New comment