Compilation Error in libtensorflow Pointing to float8.h in ml_dtype

[HongMJ1315]

I encountered an error while compiling my program using libtensorflow. The error seems to point to float8.h in ml_dtype. Here is the error information:

In file included from /home/mrjb/tensorflow/tf_env/include/tsl/platform/ml_dtypes.h:19,
                 from /home/mrjb/tensorflow/tf_env/include/tsl/platform/types.h:22,
                 from /home/mrjb/tensorflow/tf_env/include/tensorflow/core/platform/types.h:22,
                 from /home/mrjb/tensorflow/tf_env/include/tensorflow/core/platform/env_time.h:20,
                 from /home/mrjb/tensorflow/tf_env/include/tensorflow/core/platform/env.h:26,
                 from /mnt/c/Users/gonec/Documents/C++/tf/src/main.cc:1:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:871:29: error: ‘get_integer_by_size’ in namespace ‘Eigen::numext’ does not name a template type
  871 |     typename Eigen::numext::get_integer_by_size<kNumBytes>::unsigned_type;
      |                             ^~~~~~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:892:20: error: ‘GetUnsignedInteger’ does not name a type
  892 |   using BitsType = GetUnsignedInteger<sizeof(Float)>;
      |                    ^~~~~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:896:20: error: ‘BitsType’ does not name a type
  896 |   static constexpr BitsType kExponentMask = ((BitsType{1} << kExponentBits) - 1)
      |                    ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:896:59: error: expected unqualified-id before ‘<<’ token
  896 |   static constexpr BitsType kExponentMask = ((BitsType{1} << kExponentBits) - 1)
      |                                                           ^~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:898:20: error: ‘BitsType’ does not name a type
  898 |   static constexpr BitsType kMantissaMask = (BitsType{1} << kMantissaBits) - 1;
      |                    ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:898:58: error: expected unqualified-id before ‘<<’ token
  898 |   static constexpr BitsType kMantissaMask = (BitsType{1} << kMantissaBits) - 1;
      |                                                          ^~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1024:20: error: ‘GetUnsignedInteger’ does not name a type
 1024 |   using WideBits = GetUnsignedInteger<kWideBytes>;
      |                    ^~~~~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In static member function ‘static To ml_dtypes::float8_internal::ConvertImpl<From, To, kSaturate, kTruncate, typename std::enable_if<(! is_same_v<From, To>), void>::type>::run(From)’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1056:9: error: ‘WideBits’ was not declared in this scope; did you mean ‘kWideBits’?
 1056 |         WideBits bits = from_bits;
      |         ^~~~~~~~
      |         kWideBits
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1066:13: error: ‘bits’ was not declared in this scope
 1066 |             bits <<= kExponentOffset;
      |             ^~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1071:11: error: ‘bits’ was not declared in this scope
 1071 |           bits <<= normalization_factor;
      |           ^~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1072:29: error: expected ‘)’ before ‘{’ token
 1072 |           bits &= ~(WideBits{1} << kFromMantissaBits);
      |                             ^
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1072:20: note: to match this ‘(’
 1072 |           bits &= ~(WideBits{1} << kFromMantissaBits);
      |                    ^
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1074:31: error: ‘WideBits’ does not name a type
 1074 |           bits |= static_cast<WideBits>(biased_exponent) << kFromMantissaBits;
      |                               ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1079:11: error: ‘bits’ was not declared in this scope
 1079 |           bits <<= kDigitShift;
      |           ^~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1082:13: error: ‘bits’ was not declared in this scope
 1082 |             bits = RoundBitsToNearestEven(bits, -kDigitShift);
      |             ^~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1084:11: error: ‘bits’ was not declared in this scope
 1084 |           bits >>= -kDigitShift;
      |           ^~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1086:65: error: ‘bits’ was not declared in this scope
 1086 |         To to = Eigen::numext::bit_cast<To>(static_cast<ToBits>(bits));
      |                                                                 ^~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1128:5: error: ‘WideBits’ was not declared in this scope; did you mean ‘kWideBits’?
 1128 |     WideBits rounded_from_bits = from_bits;
      |     ^~~~~~~~
      |     kWideBits
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1131:9: error: ‘rounded_from_bits’ was not declared in this scope
 1131 |         rounded_from_bits = RoundBitsToNearestEven(from_bits, -kDigitShift);
      |         ^~~~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1134:7: error: ‘rounded_from_bits’ was not declared in this scope
 1134 |       rounded_from_bits &= ~((WideBits{1} << (-kDigitShift)) - 1);
      |       ^~~~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1134:39: error: expected ‘)’ before ‘{’ token
 1134 |       rounded_from_bits &= ~((WideBits{1} << (-kDigitShift)) - 1);
      |                                       ^
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1134:30: note: to match this ‘(’
 1134 |       rounded_from_bits &= ~((WideBits{1} << (-kDigitShift)) - 1);
      |                              ^
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1138:5: error: ‘rounded_from_bits’ was not declared in this scope
 1138 |     rounded_from_bits += static_cast<WideBits>(kExponentOffset)
      |     ^~~~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1138:38: error: ‘WideBits’ does not name a type
 1138 |     rounded_from_bits += static_cast<WideBits>(kExponentOffset)
      |                                      ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1144:11: error: ‘WideBits’ does not name a type
 1144 |     const WideBits kToHighestRep =
      |           ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1146:14: error: expected ‘;’ before ‘aligned_highest’
 1146 |     WideBits aligned_highest{kToHighestRep};
      |              ^~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1148:7: error: ‘aligned_highest’ was not declared in this scope
 1148 |       aligned_highest <<= -kDigitShift;
      |       ^~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1163:31: error: ‘aligned_highest’ was not declared in this scope
 1163 |       if (rounded_from_bits > aligned_highest) {
      |                               ^~~~~~~~~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<ml_dtypes::float8_internal::float8_e5m2, ml_dtypes::float8_internal::float8_e5m2fnuz, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1264:65:   required from ‘static Derived ml_dtypes::float8_internal::float8_base<Derived>::ConvertFrom(From) [with bool kSaturate = false; bool kTruncate = false; From = ml_dtypes::float8_internal::float8_e5m2; Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:371:36:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e5m2>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e5m2fnuz>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<ml_dtypes::float8_internal::float8_e4m3b11fnuz, ml_dtypes::float8_internal::float8_e5m2fnuz, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1264:65:   required from ‘static Derived ml_dtypes::float8_internal::float8_base<Derived>::ConvertFrom(From) [with bool kSaturate = false; bool kTruncate = false; From = ml_dtypes::float8_internal::float8_e4m3b11fnuz; Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:373:36:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e4m3b11fnuz>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e5m2fnuz>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<ml_dtypes::float8_internal::float8_e4m3fn, ml_dtypes::float8_internal::float8_e5m2fnuz, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1264:65:   required from ‘static Derived ml_dtypes::float8_internal::float8_base<Derived>::ConvertFrom(From) [with bool kSaturate = false; bool kTruncate = false; From = ml_dtypes::float8_internal::float8_e4m3fn; Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:375:36:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e4m3fn>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e5m2fnuz>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<ml_dtypes::float8_internal::float8_e4m3fnuz, ml_dtypes::float8_internal::float8_e5m2fnuz, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1264:65:   required from ‘static Derived ml_dtypes::float8_internal::float8_base<Derived>::ConvertFrom(From) [with bool kSaturate = false; bool kTruncate = false; From = ml_dtypes::float8_internal::float8_e4m3fnuz; Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:377:36:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e4m3fnuz>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e5m2fnuz>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<ml_dtypes::float8_internal::float8_e4m3b11fnuz, float, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1271:61:   required from ‘static To ml_dtypes::float8_internal::float8_base<Derived>::ConvertTo(Derived) [with To = float; bool kSaturate = false; bool kTruncate = false; Derived = ml_dtypes::float8_internal::float8_e4m3b11fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:87:28:   required from ‘ml_dtypes::float8_internal::float8_base<Derived>::operator float() const [with Derived = ml_dtypes::float8_internal::float8_e4m3b11fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:128:37:   required from ‘Derived ml_dtypes::float8_internal::float8_base<Derived>::operator-(const Derived&) const [with Derived = ml_dtypes::float8_internal::float8_e4m3b11fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:285:27:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e4m3b11fnuz>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<float>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<float, ml_dtypes::float8_internal::float8_e4m3b11fnuz, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1264:65:   required from ‘static Derived ml_dtypes::float8_internal::float8_base<Derived>::ConvertFrom(From) [with bool kSaturate = false; bool kTruncate = false; From = float; Derived = ml_dtypes::float8_internal::float8_e4m3b11fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:70:32:   required from ‘ml_dtypes::float8_internal::float8_base<Derived>::float8_base(T, std::enable_if_t<is_floating_point_v<T>, int>) [with T = float; Derived = ml_dtypes::float8_internal::float8_e4m3b11fnuz; std::enable_if_t<is_floating_point_v<T>, int> = int]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:270:15:   required from ‘Derived ml_dtypes::float8_internal::float8_base<Derived>::operator-(const Derived&) const [with Derived = ml_dtypes::float8_internal::float8_e4m3b11fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:285:27:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<float>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e4m3b11fnuz>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<ml_dtypes::float8_internal::float8_e4m3fnuz, float, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1271:61:   required from ‘static To ml_dtypes::float8_internal::float8_base<Derived>::ConvertTo(Derived) [with To = float; bool kSaturate = false; bool kTruncate = false; Derived = ml_dtypes::float8_internal::float8_e4m3fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:87:28:   required from ‘ml_dtypes::float8_internal::float8_base<Derived>::operator float() const [with Derived = ml_dtypes::float8_internal::float8_e4m3fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:128:37:   required from ‘Derived ml_dtypes::float8_internal::float8_base<Derived>::operator-(const Derived&) const [with Derived = ml_dtypes::float8_internal::float8_e4m3fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:328:27:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e4m3fnuz>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<float>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<float, ml_dtypes::float8_internal::float8_e4m3fnuz, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1264:65:   required from ‘static Derived ml_dtypes::float8_internal::float8_base<Derived>::ConvertFrom(From) [with bool kSaturate = false; bool kTruncate = false; From = float; Derived = ml_dtypes::float8_internal::float8_e4m3fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:70:32:   required from ‘ml_dtypes::float8_internal::float8_base<Derived>::float8_base(T, std::enable_if_t<is_floating_point_v<T>, int>) [with T = float; Derived = ml_dtypes::float8_internal::float8_e4m3fnuz; std::enable_if_t<is_floating_point_v<T>, int> = int]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:313:15:   required from ‘Derived ml_dtypes::float8_internal::float8_base<Derived>::operator-(const Derived&) const [with Derived = ml_dtypes::float8_internal::float8_e4m3fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:328:27:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<float>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e4m3fnuz>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<ml_dtypes::float8_internal::float8_e5m2fnuz, float, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1271:61:   required from ‘static To ml_dtypes::float8_internal::float8_base<Derived>::ConvertTo(Derived) [with To = float; bool kSaturate = false; bool kTruncate = false; Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:87:28:   required from ‘ml_dtypes::float8_internal::float8_base<Derived>::operator float() const [with Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:128:37:   required from ‘Derived ml_dtypes::float8_internal::float8_base<Derived>::operator-(const Derived&) const [with Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:387:27:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e5m2fnuz>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<float>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h: In instantiation of ‘struct ml_dtypes::float8_internal::ConvertImpl<float, ml_dtypes::float8_internal::float8_e5m2fnuz, false, false, void>’:
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1264:65:   required from ‘static Derived ml_dtypes::float8_internal::float8_base<Derived>::ConvertFrom(From) [with bool kSaturate = false; bool kTruncate = false; From = float; Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:70:32:   required from ‘ml_dtypes::float8_internal::float8_base<Derived>::float8_base(T, std::enable_if_t<is_floating_point_v<T>, int>) [with T = float; Derived = ml_dtypes::float8_internal::float8_e5m2fnuz; std::enable_if_t<is_floating_point_v<T>, int> = int]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:367:15:   required from ‘Derived ml_dtypes::float8_internal::float8_base<Derived>::operator-(const Derived&) const [with Derived = ml_dtypes::float8_internal::float8_e5m2fnuz]’
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:387:27:   required from here
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1003:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<float>’
 1003 |   using FromBits = typename FromTraits::BitsType;
      |         ^~~~~~~~
/home/mrjb/tensorflow/tf_env/include/ml_dtypes/include/float8.h:1011:9: error: no type named ‘BitsType’ in ‘struct ml_dtypes::float8_internal::Traits<ml_dtypes::float8_internal::float8_e5m2fnuz>’
 1011 |   using ToBits = typename ToTraits::BitsType;
      |         ^~~~~~

output.txt

Any help would be greatly appreciated. Thank you.

[jakevdp]

Thanks for the question. What version of ml_dtypes are you using, and what compiler are you using?

[HongMJ1315]

Compiler version: g++ (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
I found in tensorflow/third_party/py/ml_dtypes/workspace.bzl that TensorFlow seems to be using this version: 2ca30a2b3c0744625ae3d6988f5596740080bbd0

And for Eigen, I found in tensorflow/third_party/eigen3/workspace.bzl that it seems to be using this version: aa6964bf3a34fd607837dd8123bc42465185c4f8

[jakevdp]

I don't understand where the error would be coming from. In the commit you reference, Traits:BitsType is defined, as you can see here:

ml_dtypes/ml_dtypes/include/float8.h

Lines 908 to 910 in 2ca30a2

	template <typename Float>
	struct TraitsBase {
	using BitsType = GetUnsignedInteger<sizeof(Float)>;

Perhaps the using FromBits = typename FromTraits::BitsType invocation is not supported by g++ 11.4?

[HongMJ1315]

I suspect the error might be originating from the following line:

  871 |     typename Eigen::numext::get_integer_by_size<kNumBytes>::unsigned_type;

I've tried both the Eigen version from tensorflow/third_party/eigen3/workspace.bzl and the latest version of Eigen, but neither could compile. Furthermore, I couldn't find get_integer_by_size in Eigen::numext in either version.

In addition, I noticed that the Eigen library seems to be missing in ml_dtypes/tree/main/third_party. Could the error be due to using different libraries?

[jakevdp]

If eigen is missing from the ml_dtypes package, then perhaps you forgot to run the git submodule commands? There is more info on building from source here: https://github.com/jax-ml/ml_dtypes#installation

[HongMJ1315]

After deleting Eigen and recompiling using the Eigen provided in ml_dtypes/tree/main/third_party, it was successful.
It might indeed be due to different versions of Eigen.
Thank you for your help.