/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-hip/checkouts/docs-5.0.0/include/hip/hip_bfloat16.h Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-hip/checkouts/docs-5.0.0/include/hip/hip_bfloat16.h Source File#

HIP Runtime API Reference: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-hip/checkouts/docs-5.0.0/include/hip/hip_bfloat16.h Source File
Go to the documentation of this file.
  
 #ifndef _HIP_BFLOAT16_H_
 #define _HIP_BFLOAT16_H_
  
 #if __cplusplus < 201103L || !defined(__HIPCC__)
  
 // If this is a C compiler, C++ compiler below C++11, or a host-only compiler, we only
 // include a minimal definition of hip_bfloat16
  
 #include <stdint.h>
 typedef struct
 {
     uint16_t data;
 } hip_bfloat16;
  
 #else // __cplusplus < 201103L || !defined(__HIPCC__)
  
 #include <cmath>
 #include <cstddef>
 #include <cstdint>
 #include <hip/hip_runtime.h>
 #include <ostream>
 #include <type_traits>
  
 struct hip_bfloat16
 {
     uint16_t data;
  
     enum truncate_t
     {
         truncate
     };
  
     __host__ __device__ hip_bfloat16() = default;
  
     // round upper 16 bits of IEEE float to convert to bfloat16
     explicit __host__ __device__ hip_bfloat16(float f)
         : data(float_to_bfloat16(f))
     {
     }
  
     explicit __host__ __device__ hip_bfloat16(float f, truncate_t)
         : data(truncate_float_to_bfloat16(f))
     {
     }
  
     // zero extend lower 16 bits of bfloat16 to convert to IEEE float
     __host__ __device__ operator float() const
     {
         union
         {
             uint32_t int32;
             float    fp32;
         } u = {uint32_t(data) << 16};
         return u.fp32;
     }
  
     static  __host__ __device__ hip_bfloat16 round_to_bfloat16(float f)
     {
         hip_bfloat16 output;
         output.data = float_to_bfloat16(f);
         return output;
     }
  
     static  __host__ __device__ hip_bfloat16 round_to_bfloat16(float f, truncate_t)
     {
         hip_bfloat16 output;
         output.data = truncate_float_to_bfloat16(f);
         return output;
     }
  
 private:
     static __host__ __device__ uint16_t float_to_bfloat16(float f)
     {
         union
         {
             float    fp32;
             uint32_t int32;
         } u = {f};
         if(~u.int32 & 0x7f800000)
         {
             // When the exponent bits are not all 1s, then the value is zero, normal,
             // or subnormal. We round the bfloat16 mantissa up by adding 0x7FFF, plus
             // 1 if the least significant bit of the bfloat16 mantissa is 1 (odd).
             // This causes the bfloat16's mantissa to be incremented by 1 if the 16
             // least significant bits of the float mantissa are greater than 0x8000,
             // or if they are equal to 0x8000 and the least significant bit of the
             // bfloat16 mantissa is 1 (odd). This causes it to be rounded to even when
             // the lower 16 bits are exactly 0x8000. If the bfloat16 mantissa already
             // has the value 0x7f, then incrementing it causes it to become 0x00 and
             // the exponent is incremented by one, which is the next higher FP value
             // to the unrounded bfloat16 value. When the bfloat16 value is subnormal
             // with an exponent of 0x00 and a mantissa of 0x7F, it may be rounded up
             // to a normal value with an exponent of 0x01 and a mantissa of 0x00.
             // When the bfloat16 value has an exponent of 0xFE and a mantissa of 0x7F,
             // incrementing it causes it to become an exponent of 0xFF and a mantissa
             // of 0x00, which is Inf, the next higher value to the unrounded value.
             u.int32 += 0x7fff + ((u.int32 >> 16) & 1); // Round to nearest, round to even
         }
         else if(u.int32 & 0xffff)
         {
             // When all of the exponent bits are 1, the value is Inf or NaN.
             // Inf is indicated by a zero mantissa. NaN is indicated by any nonzero
             // mantissa bit. Quiet NaN is indicated by the most significant mantissa
             // bit being 1. Signaling NaN is indicated by the most significant
             // mantissa bit being 0 but some other bit(s) being 1. If any of the
             // lower 16 bits of the mantissa are 1, we set the least significant bit
             // of the bfloat16 mantissa, in order to preserve signaling NaN in case
             // the bloat16's mantissa bits are all 0.
             u.int32 |= 0x10000; // Preserve signaling NaN
         }
         return uint16_t(u.int32 >> 16);
     }
  
     // Truncate instead of rounding, preserving SNaN
     static __host__ __device__ uint16_t truncate_float_to_bfloat16(float f)
     {
         union
         {
             float    fp32;
             uint32_t int32;
         } u = {f};
         return uint16_t(u.int32 >> 16) | (!(~u.int32 & 0x7f800000) && (u.int32 & 0xffff));
     }
 };
  
 typedef struct
 {
     uint16_t data;
 } hip_bfloat16_public;
  
 static_assert(std::is_standard_layout<hip_bfloat16>{},
               "hip_bfloat16 is not a standard layout type, and thus is "
               "incompatible with C.");
  
 static_assert(std::is_trivial<hip_bfloat16>{},
               "hip_bfloat16 is not a trivial type, and thus is "
               "incompatible with C.");
  
 static_assert(sizeof(hip_bfloat16) == sizeof(hip_bfloat16_public)
                   && offsetof(hip_bfloat16, data) == offsetof(hip_bfloat16_public, data),
               "internal hip_bfloat16 does not match public hip_bfloat16");
  
 inline std::ostream& operator<<(std::ostream& os, const hip_bfloat16& bf16)
 {
     return os << float(bf16);
 }
 inline __host__ __device__ hip_bfloat16 operator+(hip_bfloat16 a)
 {
     return a;
 }
 inline __host__ __device__ hip_bfloat16 operator-(hip_bfloat16 a)
 {
     a.data ^= 0x8000;
     return a;
 }
 inline __host__ __device__ hip_bfloat16 operator+(hip_bfloat16 a, hip_bfloat16 b)
 {
     return hip_bfloat16(float(a) + float(b));
 }
 inline __host__ __device__ hip_bfloat16 operator-(hip_bfloat16 a, hip_bfloat16 b)
 {
     return hip_bfloat16(float(a) - float(b));
 }
 inline __host__ __device__ hip_bfloat16 operator*(hip_bfloat16 a, hip_bfloat16 b)
 {
     return hip_bfloat16(float(a) * float(b));
 }
 inline __host__ __device__ hip_bfloat16 operator/(hip_bfloat16 a, hip_bfloat16 b)
 {
     return hip_bfloat16(float(a) / float(b));
 }
 inline __host__ __device__ bool operator<(hip_bfloat16 a, hip_bfloat16 b)
 {
     return float(a) < float(b);
 }
 inline __host__ __device__ bool operator==(hip_bfloat16 a, hip_bfloat16 b)
 {
     return float(a) == float(b);
 }
 inline __host__ __device__ bool operator>(hip_bfloat16 a, hip_bfloat16 b)
 {
     return b < a;
 }
 inline __host__ __device__ bool operator<=(hip_bfloat16 a, hip_bfloat16 b)
 {
     return !(a > b);
 }
 inline __host__ __device__ bool operator!=(hip_bfloat16 a, hip_bfloat16 b)
 {
     return !(a == b);
 }
 inline __host__ __device__ bool operator>=(hip_bfloat16 a, hip_bfloat16 b)
 {
     return !(a < b);
 }
 inline __host__ __device__ hip_bfloat16& operator+=(hip_bfloat16& a, hip_bfloat16 b)
 {
     return a = a + b;
 }
 inline __host__ __device__ hip_bfloat16& operator-=(hip_bfloat16& a, hip_bfloat16 b)
 {
     return a = a - b;
 }
 inline __host__ __device__ hip_bfloat16& operator*=(hip_bfloat16& a, hip_bfloat16 b)
 {
     return a = a * b;
 }
 inline __host__ __device__ hip_bfloat16& operator/=(hip_bfloat16& a, hip_bfloat16 b)
 {
     return a = a / b;
 }
 inline __host__ __device__ hip_bfloat16& operator++(hip_bfloat16& a)
 {
     return a += hip_bfloat16(1.0f);
 }
 inline __host__ __device__ hip_bfloat16& operator--(hip_bfloat16& a)
 {
     return a -= hip_bfloat16(1.0f);
 }
 inline __host__ __device__ hip_bfloat16 operator++(hip_bfloat16& a, int)
 {
     hip_bfloat16 orig = a;
     ++a;
     return orig;
 }
 inline __host__ __device__ hip_bfloat16 operator--(hip_bfloat16& a, int)
 {
     hip_bfloat16 orig = a;
     --a;
     return orig;
 }
  
 namespace std
 {
     constexpr __host__ __device__ bool isinf(hip_bfloat16 a)
     {
         return !(~a.data & 0x7f80) && !(a.data & 0x7f);
     }
     constexpr __host__ __device__ bool isnan(hip_bfloat16 a)
     {
         return !(~a.data & 0x7f80) && +(a.data & 0x7f);
     }
     constexpr __host__ __device__ bool iszero(hip_bfloat16 a)
     {
         return !(a.data & 0x7fff);
     }
 }
  
 #endif // __cplusplus < 201103L || !defined(__HIPCC__)
  
 #endif // _HIP_BFLOAT16_H_