35,15 → 35,15 |
float32 subFloat32(float32 a, float32 b) |
{ |
int expdiff; |
__u32 exp1, exp2, mant1, mant2; |
__u32 exp1, exp2, frac1, frac2; |
float32 result; |
|
result.f = 0; |
|
expdiff = a.parts.exp - b.parts.exp; |
if ((expdiff < 0 ) || ((expdiff == 0) && (a.parts.mantisa < b.parts.mantisa))) { |
if ((expdiff < 0 ) || ((expdiff == 0) && (a.parts.fraction < b.parts.fraction))) { |
if (isFloat32NaN(b)) { |
//TODO: fix SigNaN |
/* TODO: fix SigNaN */ |
if (isFloat32SigNaN(b)) { |
}; |
return b; |
56,14 → 56,14 |
|
result.parts.sign = !a.parts.sign; |
|
mant1 = b.parts.mantisa; |
frac1 = b.parts.fraction; |
exp1 = b.parts.exp; |
mant2 = a.parts.mantisa; |
frac2 = a.parts.fraction; |
exp2 = a.parts.exp; |
expdiff *= -1; |
} else { |
if (isFloat32NaN(a)) { |
//TODO: fix SigNaN |
/* TODO: fix SigNaN */ |
if (isFloat32SigNaN(a) || isFloat32SigNaN(b)) { |
}; |
return a; |
72,7 → 72,7 |
if (a.parts.exp == FLOAT32_MAX_EXPONENT) { |
if (b.parts.exp == FLOAT32_MAX_EXPONENT) { |
/* inf - inf => nan */ |
//TODO: fix exception |
/* TODO: fix exception */ |
result.binary = FLOAT32_NAN; |
return result; |
}; |
81,17 → 81,17 |
|
result.parts.sign = a.parts.sign; |
|
mant1 = a.parts.mantisa; |
frac1 = a.parts.fraction; |
exp1 = a.parts.exp; |
mant2 = b.parts.mantisa; |
frac2 = b.parts.fraction; |
exp2 = b.parts.exp; |
}; |
|
if (exp1 == 0) { |
//both are denormalized |
result.parts.mantisa = mant1-mant2; |
if (result.parts.mantisa > mant1) { |
//TODO: underflow exception |
/* both are denormalized */ |
result.parts.fraction = frac1-frac2; |
if (result.parts.fraction > frac1) { |
/* TODO: underflow exception */ |
return result; |
}; |
result.parts.exp = 0; |
99,7 → 99,7 |
}; |
|
/* add hidden bit */ |
mant1 |= FLOAT32_HIDDEN_BIT_MASK; |
frac1 |= FLOAT32_HIDDEN_BIT_MASK; |
|
if (exp2 == 0) { |
/* denormalized */ |
106,36 → 106,36 |
--expdiff; |
} else { |
/* normalized */ |
mant2 |= FLOAT32_HIDDEN_BIT_MASK; |
frac2 |= FLOAT32_HIDDEN_BIT_MASK; |
}; |
|
/* create some space for rounding */ |
mant1 <<= 6; |
mant2 <<= 6; |
frac1 <<= 6; |
frac2 <<= 6; |
|
if (expdiff > FLOAT32_MANTISA_SIZE + 1) { |
if (expdiff > FLOAT32_FRACTION_SIZE + 1) { |
goto done; |
}; |
|
mant1 = mant1 - (mant2 >> expdiff); |
frac1 = frac1 - (frac2 >> expdiff); |
done: |
//TODO: find first nonzero digit and shift result and detect possibly underflow |
while ((exp1 > 0) && (!(mant1 & (FLOAT32_HIDDEN_BIT_MASK << 6 )))) { |
/* TODO: find first nonzero digit and shift result and detect possibly underflow */ |
while ((exp1 > 0) && (!(frac1 & (FLOAT32_HIDDEN_BIT_MASK << 6 )))) { |
--exp1; |
mant1 <<= 1; |
/* TODO: fix underflow - mant1 == 0 does not necessary means underflow... */ |
frac1 <<= 1; |
/* TODO: fix underflow - frac1 == 0 does not necessary means underflow... */ |
}; |
|
/* rounding - if first bit after mantisa is set then round up */ |
mant1 += 0x20; |
/* rounding - if first bit after fraction is set then round up */ |
frac1 += 0x20; |
|
if (mant1 & (FLOAT32_HIDDEN_BIT_MASK << 7)) { |
if (frac1 & (FLOAT32_HIDDEN_BIT_MASK << 7)) { |
++exp1; |
mant1 >>= 1; |
frac1 >>= 1; |
}; |
|
/*Clear hidden bit and shift */ |
result.parts.mantisa = ((mant1 >> 6) & (~FLOAT32_HIDDEN_BIT_MASK)); |
result.parts.fraction = ((frac1 >> 6) & (~FLOAT32_HIDDEN_BIT_MASK)); |
result.parts.exp = exp1; |
|
return result; |
147,15 → 147,15 |
{ |
int expdiff; |
__u32 exp1, exp2; |
__u64 mant1, mant2; |
__u64 frac1, frac2; |
float64 result; |
|
result.d = 0; |
|
expdiff = a.parts.exp - b.parts.exp; |
if ((expdiff < 0 ) || ((expdiff == 0) && (a.parts.mantisa < b.parts.mantisa))) { |
if ((expdiff < 0 ) || ((expdiff == 0) && (a.parts.fraction < b.parts.fraction))) { |
if (isFloat64NaN(b)) { |
//TODO: fix SigNaN |
/* TODO: fix SigNaN */ |
if (isFloat64SigNaN(b)) { |
}; |
return b; |
168,14 → 168,14 |
|
result.parts.sign = !a.parts.sign; |
|
mant1 = b.parts.mantisa; |
frac1 = b.parts.fraction; |
exp1 = b.parts.exp; |
mant2 = a.parts.mantisa; |
frac2 = a.parts.fraction; |
exp2 = a.parts.exp; |
expdiff *= -1; |
} else { |
if (isFloat64NaN(a)) { |
//TODO: fix SigNaN |
/* TODO: fix SigNaN */ |
if (isFloat64SigNaN(a) || isFloat64SigNaN(b)) { |
}; |
return a; |
184,7 → 184,7 |
if (a.parts.exp == FLOAT64_MAX_EXPONENT) { |
if (b.parts.exp == FLOAT64_MAX_EXPONENT) { |
/* inf - inf => nan */ |
//TODO: fix exception |
/* TODO: fix exception */ |
result.binary = FLOAT64_NAN; |
return result; |
}; |
193,17 → 193,17 |
|
result.parts.sign = a.parts.sign; |
|
mant1 = a.parts.mantisa; |
frac1 = a.parts.fraction; |
exp1 = a.parts.exp; |
mant2 = b.parts.mantisa; |
frac2 = b.parts.fraction; |
exp2 = b.parts.exp; |
}; |
|
if (exp1 == 0) { |
//both are denormalized |
result.parts.mantisa = mant1 - mant2; |
if (result.parts.mantisa > mant1) { |
//TODO: underflow exception |
/* both are denormalized */ |
result.parts.fraction = frac1 - frac2; |
if (result.parts.fraction > frac1) { |
/* TODO: underflow exception */ |
return result; |
}; |
result.parts.exp = 0; |
211,7 → 211,7 |
}; |
|
/* add hidden bit */ |
mant1 |= FLOAT64_HIDDEN_BIT_MASK; |
frac1 |= FLOAT64_HIDDEN_BIT_MASK; |
|
if (exp2 == 0) { |
/* denormalized */ |
218,36 → 218,36 |
--expdiff; |
} else { |
/* normalized */ |
mant2 |= FLOAT64_HIDDEN_BIT_MASK; |
frac2 |= FLOAT64_HIDDEN_BIT_MASK; |
}; |
|
/* create some space for rounding */ |
mant1 <<= 6; |
mant2 <<= 6; |
frac1 <<= 6; |
frac2 <<= 6; |
|
if (expdiff > FLOAT64_MANTISA_SIZE + 1) { |
if (expdiff > FLOAT64_FRACTION_SIZE + 1) { |
goto done; |
}; |
|
mant1 = mant1 - (mant2 >> expdiff); |
frac1 = frac1 - (frac2 >> expdiff); |
done: |
//TODO: find first nonzero digit and shift result and detect possibly underflow |
while ((exp1 > 0) && (!(mant1 & (FLOAT64_HIDDEN_BIT_MASK << 6 )))) { |
/* TODO: find first nonzero digit and shift result and detect possibly underflow */ |
while ((exp1 > 0) && (!(frac1 & (FLOAT64_HIDDEN_BIT_MASK << 6 )))) { |
--exp1; |
mant1 <<= 1; |
/* TODO: fix underflow - mant1 == 0 does not necessary means underflow... */ |
frac1 <<= 1; |
/* TODO: fix underflow - frac1 == 0 does not necessary means underflow... */ |
}; |
|
/* rounding - if first bit after mantisa is set then round up */ |
mant1 += 0x20; |
/* rounding - if first bit after fraction is set then round up */ |
frac1 += 0x20; |
|
if (mant1 & (FLOAT64_HIDDEN_BIT_MASK << 7)) { |
if (frac1 & (FLOAT64_HIDDEN_BIT_MASK << 7)) { |
++exp1; |
mant1 >>= 1; |
frac1 >>= 1; |
}; |
|
/*Clear hidden bit and shift */ |
result.parts.mantisa = ((mant1 >> 6) & (~FLOAT64_HIDDEN_BIT_MASK)); |
result.parts.fraction = ((frac1 >> 6) & (~FLOAT64_HIDDEN_BIT_MASK)); |
result.parts.exp = exp1; |
|
return result; |