Re: [boost] [review] The review of Boost.DoubleEnded starts today: September 21 - September 30

Subject: Re: [boost] [review] The review of Boost.DoubleEnded starts today: September 21 - September 30
From: Thorsten Ottosen (tottosen_at_[hidden])
Date: 2017-10-02 17:30:17

Hi,

Some more points to consider. I tried comparing 
push_back/unsafe_push_back a little. I don't see any performance speedup 
on my system if the loop body is just the push_back. If I make the loop 
a little more complicated than just a push_back, I get some differences:

32 bit:

boost::vector::push_back vs devector::unsafe_push_back

10E3    29.0508 24.8578
10E4    26.9987 22.7853
10E5    26.9468 22.3659
10E6    26.9766 22.3562
10E7    27.1064 22.5215

64 bit:

10E3    28.6106 24.2881
10E4    27.9322 23.6613
10E5    28.0182 23.3318
10E6    28.0493 23.3803
10E7    28.1458 23.5394

32 bit:

boost::deque::push_back vs batch_deque::unsafe_push_back

10E3    34.803  45.6774
10E4    36.0686 46.3888
10E5    36.7773 48.6228
10E6    40.7754 49.6144
10E7    43.1134 49.1949

64 bit:

10E3    33.5997 30.0433
10E4    35.5992 32.8985
10E5    34.4685 48.9641
10E6    40.7974 48.6893
10E7    43.2836 48.9001

I'm not 100% sure why batch_deque is so much slower, but looking a 
little at the assembler, it seems that going from an iterator to a 
pointer may be expensive because the batch_deque::iterator uses a 
segment pointer and an index. (normal push_back in batch_deque therefore 
also appears somewhat slower than in boost::deque).

kind regards

Thorsten