3

Preamble: I want to convert a pointer to an integer type, e.g. to check alignment. uintptr_t seems to be the correct type, but it is guaranteed only in C, not in C++ (or C++11)

For the following code:

#include <stdint.h>
#ifndef I_WONDER_IF_UINTPR_T_IS_DEFINED
  typedef unsigned long uintptr_t;
#endif

template <typename T>
bool isAligned(unsigned char* p) ///Checks alignment with respect to some data type
{
   return !((uintptr_t) p % sizeof(T));
}

template bool isAligned<uint16_t>(unsigned char* p);
template bool isAligned<uint32_t>(unsigned char* p);
template bool isAligned<uint64_t>(unsigned char* p);

2 questions:

  • Is there a magic and guaranteed word that I can use where I put I_WONDER_IF_UINTPR_T_IS_DEFINED?
  • Should I just use unsigned long and forget about it?

Generated assembly (when uintptr_t available): http://goo.gl/4feUNK

Note 1: I am aware than in C++11 I should use alignof in place of sizeof
Note 2: I am aware of this discussion: <cstdint> vs <stdint.h>

Community
  • 1
  • 1
Antonio
  • 19,451
  • 13
  • 99
  • 197

3 Answers3

1

In general, portable build systems on *nix tend to use the "autotools" approach which is to test-compile a trivial file containing the type in question, and if it compiles, you know you have that type. If you want something simpler, you could either (a) assume uintptr_t is available even in C++ (probably reasonable), or (b) use unsigned long long and then:

static_assert(sizeof(unsigned long long) >= sizeof(void*), "oops");
John Zwinck
  • 239,568
  • 38
  • 324
  • 436
1

If you really want to work with an implementation not providing uintptr_t when including <stdint.h>, consider using uintmax_t instead, and a static_assert for suitability (you already established you might be on a goofy setup, so it might be too small):

#include <stdint.h>
static_assert(sizeof(uintmax_t) >= sizeof(void*), "need a bigger unsigned type.");

// Add code using `uintmax_t` to round-trip data-pointers here

There are two dis-advantages though:

  1. uintmax_t might not be uintptr_t, important for overload-resolution and linking.
  2. uintmax_t might be bigger than neccessary.
Deduplicator
  • 44,692
  • 7
  • 66
  • 118
  • Regarding 2nd disadvantage, I am pretty sure the compiler will take care of that and remove any efficiency. – Antonio Oct 02 '14 at 12:03
  • @Antonio: Depends on what all you do with it. If it can, I sure hope so. In your specific use-case, both dis-advantages should not count at all, the first due to no use, the second due to permitted optimization. – Deduplicator Oct 02 '14 at 12:09
  • Another drawback is that with GCC, one can get a spurious warning if `uintmax_t` is used instead of `uintptr_t` ([discussion](https://gcc.gnu.org/ml/gcc-help/2018-08/msg00046.html)). – vinc17 Sep 03 '18 at 14:41
1

I know this is very old, but, how about using UINTPTR_MAX for your I_WONDER_IF_UINTPR_T_IS_DEFINED ?

R S
  • 162
  • 3
  • 11