pre-increment vs post-increment on std::atomic<int>

Question

The common rule of thumb is to prefer using pre-increment on STL iterators in cases where the value is not immediately evaluated (i.e. you just want to increment the object/iterator). This is because in general the implementation of pre-increment is more efficient than post increment.

But what about std::atomic? If I run static analysis (using PVS studio) I get a warning saying that pre-increment should be more efficient, but when I look at the implementation of pre-increment (on Visual Studio 2015) it looks less efficient than post-increment?

Is there a general rule for using pre-increment over post-increment on STL atomic values or would it be implementation specific?

"it looks less efficient than post-increment?" How does it look less efficient? — eerorika
– eerorika, Commented Nov 29, 2016 at 10:37

Joseph Ireland · Accepted Answer · 2016-11-29 11:01:37Z

I don't have access to windows right now, but in gcc, the difference is between

__atomic_fetch_add(&_M_i, 1, memory_order_seq_cst);

and

__atomic_add_fetch(&_M_i, 1, memory_order_seq_cst);

which are builtins, so you can assume the compiler knows how to optimise it.

if you don't use the result, gcc -O3 yields

 lock add DWORD PTR [rdi], 1

for both.

If you do use the result, then gcc -O3 yields

 mov DWORD PTR [rsp-24], edi mov eax, 1 lock xadd DWORD PTR [rsp-24], eax add eax, 1 ret

for preincrement, and omits the add eax, 1 for post increment, so technically, yes, pre-increment is less efficient by one add, but in reality that is dwarfed by the sequential memory access.

TLDR: don't worry about it

Martin Bonner supports Monica · Accepted Answer · 2016-11-29 10:41:26Z

1

Efficiency (at this level) is always implementation specific.

For example, as a counter to the rule of thumb "prefer pre-increment for STL iterators", it turns out that in practise few compilers will generate different code for pre-increment and post-increment. (The spurious copy of the post-increment just gets optimized away to nothing.)

answered Nov 29, 2016 at 10:41

Martin Bonner supports Monica

29.1k3 gold badges56 silver badges91 bronze badges

2 Comments

eerorika Over a year ago

How does the generated code for post increment differ from the pre increment if the spurious copy is optimized away? EDIT: ah, you meant that it is rare that the generated code is different. I misunderstood.

Martin Bonner supports Monica Over a year ago

@user2079303: It doesn't. That is my point. it++; and ++it; will usually generate identical code. (Of course, if you use the result of the increment then the code will obviously be different.)

Collectives™ on Stack Overflow

pre-increment vs post-increment on std::atomic<int>

2 Answers 2

Comments

2 Comments

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

2 Comments

Related