C++11 memory model and accessing different members of the same struct in different threads

Question

Assume you've got the following definitions:

struct X
{
  char a, b;
};

X x;

And now assume you have two threads, one of which reads and writes x.a but never accesses x.b while the other one reads and writes x.b but never accesses x.a. Neither thread uses any locks or other synchronization primitives. Is this guaranteed to work in C++11? Or does it count as accessing the same object, and therefore need a lock?

It's fine, they are two distinct locations. I'll leave someone with the energy to prove it to answer. — GManNickG, Aug 18 '13 at 20:15
@Mat But every object/primitive in C/C++ is required to be at least 1 byte. And 1 byte is the "smallest addressable unit". — Mysticial, Aug 18 '13 at 20:18
The answer's in there somewhere I think http://isocpp.org/blog/2013/02/atomic-weapons-the-c-memory-model-and-modern-hardware-herb-sutter — Mat, Aug 18 '13 at 20:22
@Mat I remember hearing somewhere that the standard requires that all field members have different addresses. Although this is 3rd hand info I heard from chat. I'll wait for someone to confirm/deny it. — Mysticial, Aug 18 '13 at 20:23
@Mat: Thanks for that interesting link. I just watched part 1; while the answer to my question wasn't in that, I learned a lot of other stuff from it. — celtschk, Aug 18 '13 at 21:59
Note that although this is safe, it will likely result in [false sharing](http://en.wikipedia.org/wiki/False_sharing) and degrade performance. — Cory Nelson, Aug 19 '13 at 00:46

score 27 · Accepted Answer · answered Aug 18 '13 at 20:31

27

It's safe. Quoting C++11:

[intro.memory]p3:

A memory location is either an object of scalar type or a maximal sequence of adjacent bit-fields all having non-zero width. [ Note: Various features of the language, such as references and virtual functions, might involve additional memory locations that are not accessible to programs but are managed by the implementation. —end note ] Two threads of execution (1.10) can update and access separate memory locations without interfering with each other.

[intro.memory]p5:

[ Example: A structure declared as

struct {
  char a;
  int b:5,
  c:11,
  :0,
  d:8;
  struct {int ee:8;} e;
}

contains four separate memory locations: The field a and bit-fields d and e.ee are each separate memory locations, and can be modified concurrently without interfering with each other. The bit-fields b and c together constitute the fourth memory location. The bit-fields b and c cannot be concurrently modified, but b and a, for example, can be. —end example ]

These together mean that the members a and b of X are separate memory locations, and can thus be accessed concurrently.

answered Aug 18 '13 at 20:31

Angew is no longer proud of SO

167,307
17
350
455

What is that `:0`, does this force a byte boundary? – huysentruitw Aug 18 '13 at 20:36
1

@WouterHuysentruit Pretty much. It's a *zero-length bit field.* Quoting `[class.bit]p2`: "As a special case, an unnamed bit-field with a width of zero specifies alignment of the next bit field at an allocation unit boundary." – Angew is no longer proud of SO Aug 18 '13 at 20:40
[Found it](http://stackoverflow.com/questions/13802728/what-is-zero-width-bit-field/13802955#13802955) – huysentruitw Aug 18 '13 at 20:41
What if it was `unsigned int c:11` instead of `int c:11`? Does the fact that the type of the bitfield changes matter? – user541686 Aug 18 '13 at 21:04
@Mehrdad I don't really know - bit fields are weird beasts. – Angew is no longer proud of SO Aug 18 '13 at 21:15
1

It is safe. But because a and b are in the same cache line. Each read and write when the other thread wrote to it last will incur a performance hit because it needs to flush the cache line. – aseesing Jul 08 '15 at 08:45

C++11 memory model and accessing different members of the same struct in different threads

1 Answers1

Linked