Half.h File Reference

#include <c10/macros/Macros.h>
#include <c10/util/C++17.h>
#include <c10/util/TypeSafeSignMath.h>
#include <c10/util/complex.h>
#include <type_traits>
#include <math.h>
#include <stdint.h>
#include <complex>
#include <cstdint>
#include <cstring>
#include <iosfwd>
#include <limits>
#include <sstream>
#include <stdexcept>
#include <string>
#include <utility>
#include <typeinfo>
#include <c10/util/Half-inl.h>

Go to the source code of this file.

Classes
struct	c10::Half
struct	c10::Half::from_bits_t
struct	c10::complex< Half >

Namespaces
namespace	c10
namespace	c10::detail

Macros
#define	C10_DEVICE_HOST_FUNCTION
	Defines the Half type (half-precision floating-point) including conversions to standard C types and basic arithmetic operations. More...

Functions
float	c10::detail::fp32_from_bits (uint32_t w)
uint32_t	c10::detail::fp32_to_bits (float f)
uint32_t	c10::detail::fp16_ieee_to_fp32_bits (uint16_t h)
float	c10::detail::fp16_ieee_to_fp32_value (uint16_t h)
uint16_t	c10::detail::fp16_ieee_from_fp32_value (float f)
template<typename To , typename From >
std::enable_if< std::is_same< From, bool >::value, bool >::type	c10::overflows (From)
template<typename To , typename From >
std::enable_if< std::is_integral< From >::value &&!std::is_same< From, bool >::value, bool >::type	c10::overflows (From f)
template<typename To , typename From >
std::enable_if< std::is_floating_point< From >::value, bool >::type	c10::overflows (From f)
template<typename To , typename From >
std::enable_if< is_complex< From >::value, bool >::type	c10::overflows (From f)
std::ostream &	c10::operator<< (std::ostream &out, const Half &value)

Macro Definition Documentation

C10_DEVICE_HOST_FUNCTION

#define C10_DEVICE_HOST_FUNCTION

Defines the Half type (half-precision floating-point) including conversions to standard C types and basic arithmetic operations.

Note that arithmetic operations are implemented by converting to floating point and performing the operation in float32, instead of using CUDA half intrinsics. Most uses of this type within ATen are memory bound, including the element-wise kernels, and the half intrinsics aren't efficient on all GPUs. If you are writing a compute bound kernel, you can use the CUDA half intrinsics directly on the Half type from device code.