Intel® oneAPI Threading Building Blocks
Ask questions and share information about adding parallelism to your applications when using this threading library.
The Intel sign-in experience has changed to support enhanced security controls. If you sign in, click here for more information.
2449 Discussions

Asserting binary representations for atomic<T>


I'm looking to utilize the atomic TBB construct across a library API I am developing. I am aware of the C++ ABI dragons that be there (and can work around them), but I do not know very much about the binary layout of the atomic data structure.

In order to make sure I can use atomic in my interface, I'd like to find out if I can assert a fixed binary representation for atomic where T is a known size (e.g., 16- or 32-bit integral).

Is this achievable? Are there compiler settings I need to be aware of that could throw off this assertion (e.g., alignment and/or padding settings)? I'm using gcc on the Mac and MSVC on Windows. Any help here would be greatly appreciated. Thanks in advance!
0 Kudos
3 Replies
Valued Contributor II

A "fixed binary representation" as in "two-complement, little-endian integer?" I'm not sure I get the gist of your question. If you haven't already, you might take a look at the TBBopen source release of include/tbb/atomic.h. You'll find the base type for atomic is atomic_base:

template struct atomic_base { I my_value; };

There are template specializations for various word sizes and for void * and bool, plus a bunch of hair to play nice with various compilers, so atomic covers a variety of word sizes and binary representations. If you're worried about alignment of 16- or 32-bit values, I wouldn't and the only padding issue I can think of would concern over false sharing of adjacent atomics that occupy the same cache line when pulled in from memory.

Is this an answer to your question?


On all current commercially supported platforms, atomic has an alignment andlayout that exactly matches the layout for T.Our regression test test_atomic.cpp checks this property.

There's no guarantee of this property for future platforms, because we are at the mercy of whatever the C++ compiler does, but I would beamazed if a production C++ compiler had a different layout for an atomic than a T.

Thanks for the answers to my question -- they were both helpful!