[MSVC++] [C4146]: unary minus operator applied to unsigned type, result still unsigned

[Matthias Moulin]

When using template methods (e.g., blaze::submatrix) using size_t(-SIMDSIZE) (with SIMDSIZE an unsigned integral type), the Microsoft Visual C++ compiler (/std::c++latest /permissive-) reports a C4146, which is a level W2 warning, and fails compilation.

unsigned int i(-1);
unsigned int j(-2);
unsigned int k(-(1u)); // C4146 
unsigned int l(-(2u)); // C4146

A simple workaround consist of changing size_t(-SIMDSIZE) to size_t(0-SIMDSIZE).

[Klaus Iglberger]

Hi Matthias!

Thanks for creating this issue. We are aware of the problem, but consider this an issue of the MSVC compiler. According to the C++ standard (see [basic.fundamental], note 2), “The range of representable values for the unsigned type is 0 to 2N − 1 (inclusive); arithmetic for the unsigned type is performed modulo 2N . [Note: Unsigned arithmetic does not overflow. Overflow for signed arithmetic yields undefined behavior (7.1). —end note]“. This means that the expression size_t(-SIMDSIZE) is well defined according to the C++ standard. Also, the description and example in C4146 is clearly about a different problem that involved signed integral literals and thus does not fit the expression used in Blaze.

Still, we’ll try to circumvent the problem by coming up with a working expression for MSVC. Until we come up with a solution, please explicitly disable the warning for Blaze by means of \wd4146. Thanks again,

Best regards,

Klaus!

‌

[Matthias Moulin]

Ok will forward to the MSVC++ compiler team

[Klaus Iglberger]

Hi Matthias!

Could you please share a link to the issue for the MSVC compiler team? Thanks a lot,

Best regards,

Klaus!

[Matthias Moulin]

My apologies was still on my backlog: https://developercommunity.visualstudio.com/content/problem/884520/c4146-unary-minus-operator-applied-to-unsigned-typ.html

[Klaus Iglberger]

Hi Matthias!

Thanks for creating the issue at the Microsoft developer community. I analysed the problem and compared compiler behavior by means of Compiler Explorer:

int main()
{
   size_t i;
   std::cin >> i;  // Runtime value

   constexpr size_t N( 4UL );  // Compile time value

   //const size_t res = i - ( i % N );  // (1)
   const size_t res = i & (-N);   // (2)

   std::cout << res;
}

Both GCC and Clang treat lines (1) and (2) equally and resolve it to a single and statement with -4. MSVC does the same for line (2) (without warning), but compiles line (1) into an and statement with 3, followed by a subtraction. Based on the standard specification, this may be a missed optimization opportunity.

Best regards,

Klaus!

[Matthias Moulin]

I noticed that the /sdl compiler flag is the actual culprit treating that specific warning as an error. No idea what the underlying motivation is to do so.

[Klaus Iglberger]

Commit 3f637a3 resolves all MSVC C4146 warnings. The fix is immediately available via cloning the Blaze repository and will be officially released in Blaze 3.7.