MiniFloatConverter::ReduceMantissaToNbitsRounding Class Reference

#include <libminifloat.h>

Public Member Functions
float	operator() (float f) const
	ReduceMantissaToNbitsRounding (int bits)

Private Attributes
const uint32_t	mask
const uint32_t	maxn
const int	shift
const uint32_t	test

Detailed Description

Definition at line 71 of file libminifloat.h.

Constructor & Destructor Documentation

ReduceMantissaToNbitsRounding()

MiniFloatConverter::ReduceMantissaToNbitsRounding::ReduceMantissaToNbitsRounding ( int  bits )

inline

Definition at line 73 of file libminifloat.h.

        : shift(23 - bits), mask((0xFFFFFFFF >> (shift)) << (shift)), test(1 << (shift - 1)), maxn((1 << bits) - 2) {
      assert(bits <= 23);  // "max mantissa size is 23 bits"
    }

References cms::cuda::assert().

Member Function Documentation

operator()()

float MiniFloatConverter::ReduceMantissaToNbitsRounding::operator() ( float  f ) const

inline

Definition at line 77 of file libminifloat.h.

                                    {
      constexpr uint32_t low23 = (0x007FFFFF);  // mask to keep lowest 23 bits = mantissa
      constexpr uint32_t hi9 = (0xFF800000);    // mask to keep highest 9 bits = the rest
      union {
        float flt;
        uint32_t i32;
      } conv;
      conv.flt = f;
      if (conv.i32 & test) {  // need to round
        uint32_t mantissa = (conv.i32 & low23) >> shift;
        if (mantissa < maxn)
          mantissa++;
        conv.i32 = (conv.i32 & hi9) | (mantissa << shift);
      } else {
        conv.i32 &= mask;
      }
      return conv.flt;
    }

References conv, f, mask, maxn, and shift.

Member Data Documentation

mask

const uint32_t MiniFloatConverter::ReduceMantissaToNbitsRounding::mask

private

Definition at line 98 of file libminifloat.h.

Referenced by operator()().

maxn

const uint32_t MiniFloatConverter::ReduceMantissaToNbitsRounding::maxn

private

Definition at line 98 of file libminifloat.h.

Referenced by operator()().

shift

const int MiniFloatConverter::ReduceMantissaToNbitsRounding::shift

private

Definition at line 97 of file libminifloat.h.

Referenced by operator()().

test

const uint32_t MiniFloatConverter::ReduceMantissaToNbitsRounding::test

private

Definition at line 98 of file libminifloat.h.

Referenced by edmIntegrityCheck.IntegrityCheck::report(), and edmIntegrityCheck.IntegrityCheck::structured().