Explicit comparison with infinity in fast floating point mode with intel compiler
Closed this issue · 1 comments
pps83 commented
I use vs 2022 with Intel(R) C++ Compiler Integration 2024. That's the ICX compiler from intel (free extension that can be installed from visual studio).
With intel compiler, if I use boost I end up with lots of warnings that look like this:
3>In file included from D:\work-pps\backtest-engine\src\SplitsInfo.cpp:5:
3>In file included from D:\work-pps\boost_1_86_0\boost\multiprecision\cpp_int.hpp:19:
3>In file included from D:\work-pps\boost_1_86_0\boost\multiprecision\number.hpp:12:
3>In file included from D:\work-pps\boost_1_86_0\boost\multiprecision\detail\generic_interconvert.hpp:12:
3>In file included from D:\work-pps\boost_1_86_0\boost\multiprecision\detail\default_ops.hpp:24:
3>In file included from D:\work-pps\boost_1_86_0\boost\math\special_functions\next.hpp:17:
3>In file included from D:\work-pps\boost_1_86_0\boost\math\special_functions\trunc.hpp:23:
3>In file included from D:\work-pps\boost_1_86_0\boost\math\ccmath\ldexp.hpp:17:
3>In file included from D:\work-pps\boost_1_86_0\boost\math\ccmath\abs.hpp:19:
3>D:\work-pps\boost_1_86_0\boost\math\ccmath\isinf.hpp(29,22): warning : explicit comparison with infinity in fast floating point mode [-Wtautological-constant-compare]
3> 29 | return x == std::numeric_limits<T>::infinity() || -x == std::numeric_limits<T>::infinity();
3> | ~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3>D:\work-pps\boost_1_86_0\boost\math\ccmath\ldexp.hpp(50,38): note: in instantiation of function template specialization 'boost::math::ccmath::isinf<double>' requested here
3> 50 | (boost::math::ccmath::isinf)(arg) ? arg :
3> | ^
3>D:\work-pps\boost_1_86_0\boost\math\ccmath\ldexp.hpp(64,33): note: in instantiation of function template specialization 'boost::math::ccmath::ldexp<double, true>' requested here
3> 64 | return boost::math::ccmath::ldexp(static_cast<double>(arg), exp);
3> | ^
Note, I do not use any special compiler specific options/flags. When I compile with clang-cl I don't get any of these warnings, with icx (which is also clang based) I get tons of warnings as shown above. Temporary, to avoid overwhelming warnings I fixed it this way:
--- D:\work-pps\boost_1_86_0\boost\math\ccmath\isinf - Copy.hpp Fri Oct 25 00:43:06 2024 UTC
+++ D:\work-pps\boost_1_86_0\boost\math\ccmath\isinf.hpp Fri Oct 25 00:44:56 2024 UTC
@@ -22,14 +22,7 @@
{
if constexpr (std::numeric_limits<T>::is_signed)
{
-#ifdef __clang__
-#pragma clang diagnostic push
-#pragma clang diagnostic ignored "-Wtautological-constant-compare"
-#endif
return x == std::numeric_limits<T>::infinity() || -x == std::numeric_limits<T>::infinity();
-#ifdef __clang__
-#pragma clang diagnostic pop
-#endif
}
else
{
@@ -39,7 +32,7 @@
else
{
using boost::math::isinf;
-
+
if constexpr (!std::is_integral_v<T>)
{
return (isinf)(x);