Why after enabling floating point exceptions I got multiple traps error

Question

Environment Windows 10 with MSVC2015 professional, compile with /EHa

What I'm doing: enabling floating point exception to be able to catch exceptions when some bad things happen, just for debuging

code:

namespace debug_details
{
void (*defaultStructuredExceptionFunc)(unsigned int, PEXCEPTION_POINTERS) = nullptr;

void my_trans_func(unsigned int exceptionCode, PEXCEPTION_POINTERS pExpInfo)
{
    switch (exceptionCode)
    {
    case EXCEPTION_FLT_DENORMAL_OPERAND:
    case EXCEPTION_FLT_DIVIDE_BY_ZERO:
    case EXCEPTION_FLT_INEXACT_RESULT:
    case EXCEPTION_FLT_INVALID_OPERATION:
    case EXCEPTION_FLT_OVERFLOW:
    case EXCEPTION_FLT_STACK_CHECK:
    case EXCEPTION_FLT_UNDERFLOW:
    {
        _clearfp();
        std::stringstream ss;
        ss << "floating-point structured exception: 0x" << std::hex << exceptionCode;
        throw std::runtime_error(ss.str());
    }
    default:
        if (defaultStructuredExceptionFunc != nullptr)
        {
            defaultStructuredExceptionFunc(exceptionCode, pExpInfo);
        }
    };
};

void EnableFloatingPointExceptions()
{
    unsigned int fe_value = ~(/*_EM_INVALID | _EM_DENORMAL |*/ _EM_ZERODIVIDE | _EM_OVERFLOW | _EM_UNDERFLOW /* | _EM_INEXACT*/);
    unsigned int mask = _MCW_EM;
    unsigned int currentWord = 0;
    _clearfp();
    errno_t result = _controlfp_s(&currentWord, fe_value, mask); // https://msdn.microsoft.com/en-us/library/c9676k6h.aspx
    DVASSERT(result == 0);
    debug_details::defaultStructuredExceptionFunc = _set_se_translator(&debug_details::my_trans_func); // https://msdn.microsoft.com/en-us/library/5z4bw5h5.aspx

    float32 div = 0.f;
    float32 f = 15.f / div;
    float32 f2 = 30.f * div;
}

} // end namespace debug_details

I expect EXCEPTION_FLT_DIVIDE_BY_ZERO, but have

Please help to understand what is going on? Thanks in advance!

score 1 · Accepted Answer · answered Nov 30 '15 at 14:48

I found answer on MSDN, problem with x86 platform, on x64 we can get correct exception code from thread context register see ref - // https://social.msdn.microsoft.com/Forums/en-US/48f63378-19be-413f-88a5-0f24aa72d3c8/the-exceptions-statusfloatmultipletraps-and-statusfloatmultiplefaults-is-needed-in-more

score -1 · Answer 2 · answered Nov 27 '15 at 18:17

There are several things to know about FP exceptions:

controlfp's flags mask (i.e. disable) the exceptions you set, not enable them. So to enable only division by 0, you'd need set everything but in EM_ZERODIVIDE in the arguments
exceptions in SSE code are not affected by controlfp, you'd need to use _mm_setctr (so if you have SSE/AVX enabled in the compiler settings, most of your code will likely use SSE, even scalar)
if an exception happened when it was masked, it'll break if you enable it later (unless some code clears the status word)

Why after enabling floating point exceptions I got multiple traps error

2 Answers2

Linked