2

I was wondering what does the standard say about how the following structure will be aligned and allocated:

struct A {
   uint8_t b[16];
};

To what extend can I expect that A will be of size 16 bytes, aligned on 16 byte boundary in an array (and size of A[n] equal to 16*n, e.g. no padding) and the address of A will be the first element of b?

And just to narrow the scope a bit, the platforms I am concerned with are x86/x64, ARM M, R, A and A50.

EDIT: A clarification - by "alignment" I mean how will the compiler align the objects in an array on the stack, e.g. will each next element be 16 bytes away, naturally I do not expect malloc to be aligned on anything but the platform default boundary.

EDIT 2: After reading on an answer, I feel inclined to ask, to what extend can I expect platform-specific explicit alignment intrinsics to work?

Improve this question
3
  • Address of struct A is always the address of the its first element. Commented Jan 31, 2014 at 13:58
  • Yes, to everything except the 16 byte alignment, which is not guaranteed. It will just be the minimum alignment for your platform, usually 4 bytes for a 32 bit platform or 8 bytes for a 64 bit platform. Commented Jan 31, 2014 at 14:01
  • Interesting to see what size you get on the platforms of interest with struct xx { uint8_t c; struct A d; uint8_t e; struct A f; uint8_t g[14];}. 48 or something else? Commented Jan 31, 2014 at 14:23

1 Answer 1

Reset to default
1

The size of the struct will be at least 16. It will be 16 in normal C implementations, but I do not see that the C standard requires this, absent extenuating circumstances such as special requests from the user. E.g., if the user uses GCC’s extensions to specify that the struct must have an alignment of 32 bytes, then the struct would have to be padded to 32 bytes.

The struct cannot be expected to have an alignment requirement of 16. None of its members have an alignment requirement greater than 1, and the C standard does not require that a struct have any greater alignment requirement than its members. I do not see that the C standard requires a structure not to have a greater alignment requirement than the minimum needed to satisfy the requirements of its members, so it might be possible that a C implementation could give the struct an alignment of 16. I am unaware of C implementations that do so absent a special request from the user.

Technically, a pointer to the struct cannot be the same as a pointer to the first element of b, because the former is a pointer to a struct and the latter is a pointer to char. However, when the pointer to the struct is converted to a pointer to a uint8_t [16], it will be the same as (will compare equal to) the address of b.

In every C implementation I am familiar with, when either a pointer to the struct or a pointer to the array is further converted to a pointer to uint8_t, it will be the same as (will compare equal to) a pointer to the first element of b. However, I do not see that the C standard guarantees this. Certainly there is no padding at the start of the struct or the array, so we know the “address” of the struct is the same as the “address” of the array and the “address” of the first element, but that is not the same as a guarantee that converting pointers of different types produces the proper value.

Improve this answer
Sign up to request clarification or add additional context in comments.

4 Comments

That is why I did not say pointer but address, e.g. as if cast to void *
@ddriver: I am aware that you used the term “address”, but C only defines the behavior of pointers. Essentially, addresses only exist as values within the C model of pointers. Examining pointers is limited to the operations you can perform on them: Compare them to other pointers, do some arithmetic within bounds, compare them. It may well be that the bytes that represent a pointer to a struct are different from the bytes that represent a pointer to a uint8_t. However, when one is converted to the other, as described in my answer, they will compare equal.
OK, a little "side question", GCC has long had alignment intrinsics, considering that the toolchain for ARM M0 is at version 4.8, can I expect the alignment intrinsics to work on such a highly embedded platform?
@ddriver: ARM’s compiler appears to support the aligned attribute.. I did not check the toolchain version covered by that documentation, but I think it likely you will find it applies in this regard.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.