I don't know what the exact reason is, but if you apply the following patch to Boost.Integer, crc_test and crc_test2 fail:
--- include/boost/integer.hpp
+++ include/boost/integer.hpp
@@ -48,6 +48,13 @@ namespace boost
typedef fast type;
}; // imps may specialize
+ template<>
+ struct int_fast_t<unsigned short>
+ {
+ typedef unsigned int fast;
+ typedef unsigned int type;
+ };
+
namespace detail{
// convert category to type
Best I can tell, this is the correct way to tell boost that it is faster to perform operations on ints than on shorts on a particular platform, even if you only need 16 bit.
Background: I was trying to convert Boost.CRC int a standalone version without any boost dependencies using the std::uint_fast_XX_t / _least_XX_t type alieases instead of the boost::uint_t<XX>::fast/least ones.
At least on msvc x64, std::uint_fast16_t is unsigned int, not unsigned short. After some investigation I could reproduce this even without touching boost crc but only changing boost::uint_t<16> in Boost.Integer as above.
I don't know what the exact reason is, but if you apply the following patch to Boost.Integer, crc_test and crc_test2 fail:
Best I can tell, this is the correct way to tell boost that it is faster to perform operations on ints than on shorts on a particular platform, even if you only need 16 bit.
Background: I was trying to convert Boost.CRC int a standalone version without any boost dependencies using the
std::uint_fast_XX_t / _least_XX_ttype alieases instead of theboost::uint_t<XX>::fast/leastones.At least on msvc x64,
std::uint_fast16_tisunsigned int, notunsigned short. After some investigation I could reproduce this even without touching boost crc but only changingboost::uint_t<16>in Boost.Integer as above.