float128_t
f128_roundToInt( float128_t a, uint_fast8_t roundingMode, bool exact )
{
union ui128_f128 uA;
uint_fast64_t uiA64, uiA0;
int_fast32_t exp;
struct uint128 uiZ;
uint_fast64_t lastBitMask, roundBitsMask;
bool roundNearEven;
union ui128_f128 uZ;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
uA.f = a;
uiA64 = uA.ui.v64;
uiA0 = uA.ui.v0;
exp = expF128UI64( uiA64 );
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
if ( 0x402F <= exp ) {
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
if ( 0x406F <= exp ) {
if ( (exp == 0x7FFF) && (fracF128UI64( uiA64 ) | uiA0) ) {
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, 0, 0 );
goto uiZ;
}
return a;
}
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
lastBitMask = (uint_fast64_t) 2<<(0x406E - exp);
roundBitsMask = lastBitMask - 1;
uiZ.v64 = uiA64;
uiZ.v0 = uiA0;
roundNearEven = (roundingMode == softfloat_round_near_even);
if ( roundNearEven || (roundingMode == softfloat_round_near_maxMag) ) {
if ( exp == 0x402F ) {
if ( UINT64_C( 0x8000000000000000 ) <= uiZ.v0 ) {
++uiZ.v64;
if (
roundNearEven
&& (uiZ.v0 == UINT64_C( 0x8000000000000000 ))
) {
uiZ.v64 &= ~1;
}
}
} else {
uiZ = softfloat_add128( uiZ.v64, uiZ.v0, 0, lastBitMask>>1 );
if ( roundNearEven && ! (uiZ.v0 & roundBitsMask) ) {
uiZ.v0 &= ~lastBitMask;
}
}
} else if ( roundingMode != softfloat_round_minMag ) {
if (
signF128UI64( uiZ.v64 ) ^ (roundingMode == softfloat_round_max)
) {
uiZ = softfloat_add128( uiZ.v64, uiZ.v0, 0, roundBitsMask );
}
}
uiZ.v0 &= ~roundBitsMask;
} else {
float128_t
softfloat_addMagsF128(
uint_fast64_t uiA64,
uint_fast64_t uiA0,
uint_fast64_t uiB64,
uint_fast64_t uiB0,
bool signZ
)
{
int_fast32_t expA;
struct uint128 sigA;
int_fast32_t expB;
struct uint128 sigB;
int_fast32_t expDiff;
struct uint128 uiZ, sigZ;
int_fast32_t expZ;
uint_fast64_t sigZExtra;
struct uint128_extra sig128Extra;
union ui128_f128 uZ;
expA = expF128UI64( uiA64 );
sigA.v64 = fracF128UI64( uiA64 );
sigA.v0 = uiA0;
expB = expF128UI64( uiB64 );
sigB.v64 = fracF128UI64( uiB64 );
sigB.v0 = uiB0;
expDiff = expA - expB;
if ( ! expDiff ) {
if ( expA == 0x7FFF ) {
if ( sigA.v64 | sigA.v0 | sigB.v64 | sigB.v0 ) goto propagateNaN;
uiZ.v64 = uiA64;
uiZ.v0 = uiA0;
goto uiZ;
}
sigZ = softfloat_add128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 );
if ( ! expA ) {
uiZ.v64 = packToF128UI64( signZ, 0, sigZ.v64 );
uiZ.v0 = sigZ.v0;
goto uiZ;
}
expZ = expA;
sigZ.v64 |= UINT64_C( 0x0002000000000000 );
sigZExtra = 0;
goto shiftRight1;
}
if ( expDiff < 0 ) {
if ( expB == 0x7FFF ) {
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
uiZ.v64 = packToF128UI64( signZ, 0x7FFF, 0 );
uiZ.v0 = 0;
goto uiZ;
}
expZ = expB;
if ( expA ) {
sigA.v64 |= UINT64_C( 0x0001000000000000 );
} else {
++expDiff;
sigZExtra = 0;
if ( ! expDiff ) goto newlyAligned;
}
sig128Extra =
softfloat_shiftRightJam128Extra( sigA.v64, sigA.v0, 0, -expDiff );
sigA = sig128Extra.v;
sigZExtra = sig128Extra.extra;
} else {
if ( expA == 0x7FFF ) {
if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
uiZ.v64 = uiA64;
uiZ.v0 = uiA0;
goto uiZ;
}
expZ = expA;
if ( expB ) {
sigB.v64 |= UINT64_C( 0x0001000000000000 );
} else {
--expDiff;
sigZExtra = 0;
if ( ! expDiff ) goto newlyAligned;
}
sig128Extra =
softfloat_shiftRightJam128Extra( sigB.v64, sigB.v0, 0, expDiff );
sigB = sig128Extra.v;
sigZExtra = sig128Extra.extra;
}
newlyAligned:
sigZ =
softfloat_add128(
sigA.v64 | UINT64_C( 0x0001000000000000 ),
sigA.v0,
sigB.v64,
sigB.v0
);
--expZ;
if ( sigZ.v64 < UINT64_C( 0x0002000000000000 ) ) goto roundAndPack;
++expZ;
shiftRight1:
sig128Extra =
softfloat_shortShiftRightJam128Extra(
sigZ.v64, sigZ.v0, sigZExtra, 1 );
sigZ = sig128Extra.v;
sigZExtra = sig128Extra.extra;
roundAndPack:
return
softfloat_roundPackToF128( signZ, expZ, sigZ.v64, sigZ.v0, sigZExtra );
propagateNaN:
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
uiZ:
uZ.ui = uiZ;
return uZ.f;
}
float128_t f128_mul( float128_t a, float128_t b )
{
union ui128_f128 uA;
uint_fast64_t uiA64, uiA0;
bool signA;
int_fast32_t expA;
struct uint128 sigA;
union ui128_f128 uB;
uint_fast64_t uiB64, uiB0;
bool signB;
int_fast32_t expB;
struct uint128 sigB;
bool signZ;
uint_fast64_t magBits;
struct exp32_sig128 normExpSig;
int_fast32_t expZ;
uint64_t sig256Z[4];
uint_fast64_t sigZExtra;
struct uint128 sigZ;
struct uint128_extra sig128Extra;
struct uint128 uiZ;
union ui128_f128 uZ;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
uA.f = a;
uiA64 = uA.ui.v64;
uiA0 = uA.ui.v0;
signA = signF128UI64( uiA64 );
expA = expF128UI64( uiA64 );
sigA.v64 = fracF128UI64( uiA64 );
sigA.v0 = uiA0;
uB.f = b;
uiB64 = uB.ui.v64;
uiB0 = uB.ui.v0;
signB = signF128UI64( uiB64 );
expB = expF128UI64( uiB64 );
sigB.v64 = fracF128UI64( uiB64 );
sigB.v0 = uiB0;
signZ = signA ^ signB;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
if ( expA == 0x7FFF ) {
if (
(sigA.v64 | sigA.v0) || ((expB == 0x7FFF) && (sigB.v64 | sigB.v0))
) {
goto propagateNaN;
}
magBits = expB | sigB.v64 | sigB.v0;
goto infArg;
}
if ( expB == 0x7FFF ) {
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
magBits = expA | sigA.v64 | sigA.v0;
goto infArg;
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
if ( ! expA ) {
if ( ! (sigA.v64 | sigA.v0) ) goto zero;
normExpSig = softfloat_normSubnormalF128Sig( sigA.v64, sigA.v0 );
expA = normExpSig.exp;
sigA = normExpSig.sig;
}
if ( ! expB ) {
if ( ! (sigB.v64 | sigB.v0) ) goto zero;
normExpSig = softfloat_normSubnormalF128Sig( sigB.v64, sigB.v0 );
expB = normExpSig.exp;
sigB = normExpSig.sig;
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
expZ = expA + expB - 0x4000;
sigA.v64 |= UINT64_C( 0x0001000000000000 );
sigB = softfloat_shortShiftLeft128( sigB.v64, sigB.v0, 16 );
softfloat_mul128To256M( sigA.v64, sigA.v0, sigB.v64, sigB.v0, sig256Z );
sigZExtra = sig256Z[indexWord( 4, 1 )] | (sig256Z[indexWord( 4, 0 )] != 0);
sigZ =
softfloat_add128(
sig256Z[indexWord( 4, 3 )], sig256Z[indexWord( 4, 2 )],
sigA.v64, sigA.v0
);
if ( UINT64_C( 0x0002000000000000 ) <= sigZ.v64 ) {
++expZ;
sig128Extra =
softfloat_shortShiftRightJam128Extra(
sigZ.v64, sigZ.v0, sigZExtra, 1 );
sigZ = sig128Extra.v;
sigZExtra = sig128Extra.extra;
}
return
softfloat_roundPackToF128( signZ, expZ, sigZ.v64, sigZ.v0, sigZExtra );
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
propagateNaN:
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
goto uiZ;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
infArg:
if ( ! magBits ) {
softfloat_raiseFlags( softfloat_flag_invalid );
uiZ.v64 = defaultNaNF128UI64;
uiZ.v0 = defaultNaNF128UI0;
goto uiZ;
}
uiZ.v64 = packToF128UI64( signZ, 0x7FFF, 0 );
goto uiZ0;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
zero:
uiZ.v64 = packToF128UI64( signZ, 0, 0 );
uiZ0:
uiZ.v0 = 0;
uiZ:
uZ.ui = uiZ;
return uZ.f;
}
float128_t
softfloat_subMagsF128(
uint_fast64_t uiA64,
uint_fast64_t uiA0,
uint_fast64_t uiB64,
uint_fast64_t uiB0,
bool signZ
)
{
int_fast32_t expA;
struct uint128 sigA;
int_fast32_t expB;
struct uint128 sigB, sigZ;
int_fast32_t expDiff, expZ;
struct uint128 uiZ;
union ui128_f128 uZ;
expA = expF128UI64( uiA64 );
sigA.v64 = fracF128UI64( uiA64 );
sigA.v0 = uiA0;
expB = expF128UI64( uiB64 );
sigB.v64 = fracF128UI64( uiB64 );
sigB.v0 = uiB0;
sigA = softfloat_shortShiftLeft128( sigA.v64, sigA.v0, 4 );
sigB = softfloat_shortShiftLeft128( sigB.v64, sigB.v0, 4 );
expDiff = expA - expB;
if ( 0 < expDiff ) goto expABigger;
if ( expDiff < 0 ) goto expBBigger;
if ( expA == 0x7FFF ) {
if ( sigA.v64 | sigA.v0 | sigB.v64 | sigB.v0 ) goto propagateNaN;
softfloat_raiseFlags( softfloat_flag_invalid );
uiZ.v64 = defaultNaNF128UI64;
uiZ.v0 = defaultNaNF128UI0;
goto uiZ;
}
expZ = expA;
if ( ! expZ ) expZ = 1;
if ( sigB.v64 < sigA.v64 ) goto aBigger;
if ( sigA.v64 < sigB.v64 ) goto bBigger;
if ( sigB.v0 < sigA.v0 ) goto aBigger;
if ( sigA.v0 < sigB.v0 ) goto bBigger;
uiZ.v64 =
packToF128UI64(
(softfloat_roundingMode == softfloat_round_min), 0, 0 );
uiZ.v0 = 0;
goto uiZ;
expBBigger:
if ( expB == 0x7FFF ) {
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
uiZ.v64 = packToF128UI64( signZ ^ 1, 0x7FFF, 0 );
uiZ.v0 = 0;
goto uiZ;
}
if ( expA ) {
sigA.v64 |= UINT64_C( 0x0010000000000000 );
} else {
++expDiff;
if ( ! expDiff ) goto newlyAlignedBBigger;
}
sigA = softfloat_shiftRightJam128( sigA.v64, sigA.v0, -expDiff );
newlyAlignedBBigger:
expZ = expB;
sigB.v64 |= UINT64_C( 0x0010000000000000 );
bBigger:
signZ = ! signZ;
sigZ = softfloat_sub128( sigB.v64, sigB.v0, sigA.v64, sigA.v0 );
goto normRoundPack;
expABigger:
if ( expA == 0x7FFF ) {
if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
uiZ.v64 = uiA64;
uiZ.v0 = uiA0;
goto uiZ;
}
if ( expB ) {
sigB.v64 |= UINT64_C( 0x0010000000000000 );
} else {
--expDiff;
if ( ! expDiff ) goto newlyAlignedABigger;
}
sigB = softfloat_shiftRightJam128( sigB.v64, sigB.v0, expDiff );
newlyAlignedABigger:
expZ = expA;
sigA.v64 |= UINT64_C( 0x0010000000000000 );
aBigger:
sigZ = softfloat_sub128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 );
normRoundPack:
return softfloat_normRoundPackToF128( signZ, expZ - 5, sigZ.v64, sigZ.v0 );
propagateNaN:
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
uiZ:
uZ.ui = uiZ;
return uZ.f;
}
