void
softfloat_mulAddF128M(
const uint32_t *aWPtr,
const uint32_t *bWPtr,
const uint32_t *cWPtr,
uint32_t *zWPtr,
uint_fast8_t op
)
{
uint32_t uiA96;
int32_t expA;
uint32_t uiB96;
int32_t expB;
uint32_t uiC96;
bool signC;
int32_t expC;
bool signProd, prodIsInfinite;
uint32_t *ptr, uiZ96, sigA[4];
uint_fast8_t shiftCount;
uint32_t sigX[5];
int32_t expProd;
uint32_t sigProd[8], wordSig;
bool doSub;
uint_fast8_t
(*addCarryMRoutinePtr)(
uint_fast8_t,
const uint32_t *,
const uint32_t *,
uint_fast8_t,
uint32_t *
);
int32_t expDiff;
bool signZ;
int32_t expZ;
uint32_t *extSigPtr;
uint_fast8_t carry;
void (*roundPackRoutinePtr)( bool, int32_t, uint32_t *, uint32_t * );
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
uiA96 = aWPtr[indexWordHi( 4 )];
expA = expF128UI96( uiA96 );
uiB96 = bWPtr[indexWordHi( 4 )];
expB = expF128UI96( uiB96 );
uiC96 = cWPtr[indexWordHi( 4 )];
signC = signF128UI96( uiC96 ) ^ (op == softfloat_mulAdd_subC);
expC = expF128UI96( uiC96 );
signProd =
signF128UI96( uiA96 ) ^ signF128UI96( uiB96 )
^ (op == softfloat_mulAdd_subProd);
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
prodIsInfinite = false;
if ( (expA == 0x7FFF) || (expB == 0x7FFF) ) {
if ( softfloat_tryPropagateNaNF128M( aWPtr, bWPtr, zWPtr ) ) {
goto propagateNaN_ZC;
}
ptr = (uint32_t *) aWPtr;
if ( ! (uint32_t) (uiA96<<1) ) goto possibleInvalidProd;
if ( ! (uint32_t) (uiB96<<1) ) {
ptr = (uint32_t *) bWPtr;
possibleInvalidProd:
if (
! (ptr[indexWord( 4, 2 )] | ptr[indexWord( 4, 1 )]
| ptr[indexWord( 4, 0 )])
) {
goto invalid;
}
}
prodIsInfinite = true;
}
if ( expC == 0x7FFF ) {
if (
fracF128UI96( uiC96 )
|| (cWPtr[indexWord( 4, 2 )] | cWPtr[indexWord( 4, 1 )]
| cWPtr[indexWord( 4, 0 )])
) {
zWPtr[indexWordHi( 4 )] = 0;
goto propagateNaN_ZC;
}
if ( prodIsInfinite && (signProd != signC) ) goto invalid;
goto copyC;
}
if ( prodIsInfinite ) {
uiZ96 = packToF128UI96( signProd, 0x7FFF, 0 );
goto uiZ;
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
if ( expA ) {
sigA[indexWordHi( 4 )] = fracF128UI96( uiA96 ) | 0x00010000;
sigA[indexWord( 4, 2 )] = aWPtr[indexWord( 4, 2 )];
sigA[indexWord( 4, 1 )] = aWPtr[indexWord( 4, 1 )];
sigA[indexWord( 4, 0 )] = aWPtr[indexWord( 4, 0 )];
} else {
expA = softfloat_shiftNormSigF128M( aWPtr, 0, sigA );
if ( expA == -128 ) goto zeroProd;
}
if ( expB ) {
sigX[indexWordHi( 4 )] = fracF128UI96( uiB96 ) | 0x00010000;
sigX[indexWord( 4, 2 )] = bWPtr[indexWord( 4, 2 )];
sigX[indexWord( 4, 1 )] = bWPtr[indexWord( 4, 1 )];
sigX[indexWord( 4, 0 )] = bWPtr[indexWord( 4, 0 )];
} else {
expB = softfloat_shiftNormSigF128M( bWPtr, 0, sigX );
if ( expB == -128 ) goto zeroProd;
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
expProd = expA + expB - 0x3FF0;
softfloat_mul128MTo256M( sigA, sigX, sigProd );
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
wordSig = fracF128UI96( uiC96 );
if ( expC ) {
--expC;
wordSig |= 0x00010000;
}
sigX[indexWordHi( 5 )] = wordSig;
sigX[indexWord( 5, 3 )] = cWPtr[indexWord( 4, 2 )];
sigX[indexWord( 5, 2 )] = cWPtr[indexWord( 4, 1 )];
sigX[indexWord( 5, 1 )] = cWPtr[indexWord( 4, 0 )];
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
doSub = (signProd != signC);
addCarryMRoutinePtr =
doSub ? softfloat_addComplCarryM : softfloat_addCarryM;
expDiff = expProd - expC;
if ( expDiff <= 0 ) {
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
signZ = signC;
expZ = expC;
if (
sigProd[indexWord( 8, 2 )]
|| (sigProd[indexWord( 8, 1 )] | sigProd[indexWord( 8, 0 )])
) {
sigProd[indexWord( 8, 3 )] |= 1;
}
extSigPtr = &sigProd[indexMultiwordHi( 8, 5 )];
if ( expDiff ) {
softfloat_shiftRightJam160M( extSigPtr, -expDiff, extSigPtr );
}
carry = 0;
if ( doSub ) {
wordSig = extSigPtr[indexWordLo( 5 )];
extSigPtr[indexWordLo( 5 )] = -wordSig;
carry = ! wordSig;
}
(*addCarryMRoutinePtr)(
4,
&sigX[indexMultiwordHi( 5, 4 )],
extSigPtr + indexMultiwordHi( 5, 4 ),
carry,
extSigPtr + indexMultiwordHi( 5, 4 )
);
wordSig = extSigPtr[indexWordHi( 5 )];
if ( ! expZ ) {
if ( wordSig & 0x80000000 ) {
signZ = ! signZ;
softfloat_negX160M( extSigPtr );
wordSig = extSigPtr[indexWordHi( 5 )];
}
goto checkCancellation;
}
if ( wordSig < 0x00010000 ) {
--expZ;
softfloat_add160M( extSigPtr, extSigPtr, extSigPtr );
goto roundPack;
}
goto extSigReady_noCancellation;
} else {
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
signZ = signProd;
expZ = expProd;
sigX[indexWordLo( 5 )] = 0;
expDiff -= 128;
if ( 0 <= expDiff ) {
/*----------------------------------------------------------------
*----------------------------------------------------------------*/
if ( expDiff ) softfloat_shiftRightJam160M( sigX, expDiff, sigX );
wordSig = sigX[indexWordLo( 5 )];
carry = 0;
if ( doSub ) {
carry = ! wordSig;
wordSig = -wordSig;
}
carry =
(*addCarryMRoutinePtr)(
4,
&sigProd[indexMultiwordLo( 8, 4 )],
&sigX[indexMultiwordHi( 5, 4 )],
carry,
&sigProd[indexMultiwordLo( 8, 4 )]
);
sigProd[indexWord( 8, 2 )] |= wordSig;
ptr = &sigProd[indexWord( 8, 4 )];
} else {
/*----------------------------------------------------------------
*----------------------------------------------------------------*/
shiftCount = expDiff & 31;
if ( shiftCount ) {
softfloat_shortShiftRight160M( sigX, shiftCount, sigX );
}
expDiff >>= 5;
extSigPtr =
&sigProd[indexMultiwordLo( 8, 5 )] - wordIncr
+ expDiff * -wordIncr;
carry =
(*addCarryMRoutinePtr)( 5, extSigPtr, sigX, doSub, extSigPtr );
if ( expDiff == -4 ) {
/*------------------------------------------------------------
*------------------------------------------------------------*/
wordSig = sigProd[indexWordHi( 8 )];
if ( wordSig & 0x80000000 ) {
signZ = ! signZ;
softfloat_negX256M( sigProd );
wordSig = sigProd[indexWordHi( 8 )];
}
/*------------------------------------------------------------
*------------------------------------------------------------*/
if ( wordSig ) goto expProdBigger_noWordShift;
wordSig = sigProd[indexWord( 8, 6 )];
if ( 0x00040000 <= wordSig ) goto expProdBigger_noWordShift;
expZ -= 32;
extSigPtr = &sigProd[indexMultiwordHi( 8, 5 )] - wordIncr;
for (;;) {
if ( wordSig ) break;
wordSig = extSigPtr[indexWord( 5, 3 )];
if ( 0x00040000 <= wordSig ) break;
expZ -= 32;
extSigPtr -= wordIncr;
if ( extSigPtr == &sigProd[indexMultiwordLo( 8, 5 )] ) {
goto checkCancellation;
}
}
/*------------------------------------------------------------
*------------------------------------------------------------*/
ptr = extSigPtr + indexWordLo( 5 );
do {
ptr -= wordIncr;
if ( *ptr ) {
extSigPtr[indexWordLo( 5 )] |= 1;
break;
}
} while ( ptr != &sigProd[indexWordLo( 8 )] );
wordSig = extSigPtr[indexWordHi( 5 )];
goto extSigReady;
}
ptr = extSigPtr + indexWordHi( 5 ) + wordIncr;
}
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
if ( carry != doSub ) {
if ( doSub ) {
do {
wordSig = *ptr;
*ptr = wordSig - 1;
ptr += wordIncr;
} while ( ! wordSig );
} else {
do {
wordSig = *ptr + 1;
*ptr = wordSig;
ptr += wordIncr;
} while ( ! wordSig );
}
}
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
expProdBigger_noWordShift:
if (
sigProd[indexWord( 8, 2 )]
|| (sigProd[indexWord( 8, 1 )] | sigProd[indexWord( 8, 0 )])
) {
sigProd[indexWord( 8, 3 )] |= 1;
}
extSigPtr = &sigProd[indexMultiwordHi( 8, 5 )];
wordSig = extSigPtr[indexWordHi( 5 )];
}
extSigReady:
roundPackRoutinePtr = softfloat_normRoundPackMToF128M;
if ( wordSig < 0x00010000 ) goto doRoundPack;
extSigReady_noCancellation:
if ( 0x00020000 <= wordSig ) {
++expZ;
softfloat_shortShiftRightJam160M( extSigPtr, 1, extSigPtr );
}
roundPack:
roundPackRoutinePtr = softfloat_roundPackMToF128M;
doRoundPack:
(*roundPackRoutinePtr)( signZ, expZ, extSigPtr, zWPtr );
return;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
invalid:
softfloat_invalidF128M( zWPtr );
propagateNaN_ZC:
softfloat_propagateNaNF128M( zWPtr, cWPtr, zWPtr );
return;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
zeroProd:
if (
! (uint32_t) (uiC96<<1) && (signProd != signC)
&& ! cWPtr[indexWord( 4, 2 )]
&& ! (cWPtr[indexWord( 4, 1 )] | cWPtr[indexWord( 4, 0 )])
) {
goto completeCancellation;
}
copyC:
zWPtr[indexWordHi( 4 )] = uiC96;
zWPtr[indexWord( 4, 2 )] = cWPtr[indexWord( 4, 2 )];
zWPtr[indexWord( 4, 1 )] = cWPtr[indexWord( 4, 1 )];
zWPtr[indexWord( 4, 0 )] = cWPtr[indexWord( 4, 0 )];
return;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
checkCancellation:
if (
wordSig
|| (extSigPtr[indexWord( 5, 3 )] | extSigPtr[indexWord( 5, 2 )])
|| (extSigPtr[indexWord( 5, 1 )] | extSigPtr[indexWord( 5, 0 )])
) {
goto extSigReady;
}
completeCancellation:
uiZ96 =
packToF128UI96(
(softfloat_roundingMode == softfloat_round_min), 0, 0 );
uiZ:
zWPtr[indexWordHi( 4 )] = uiZ96;
zWPtr[indexWord( 4, 2 )] = 0;
zWPtr[indexWord( 4, 1 )] = 0;
zWPtr[indexWord( 4, 0 )] = 0;
}
void
f128M_mul( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
{
const uint32_t *aWPtr, *bWPtr;
uint32_t *zWPtr;
uint32_t uiA96;
int32_t expA;
uint32_t uiB96;
int32_t expB;
bool signZ;
const uint32_t *ptr;
uint32_t uiZ96, sigA[4];
uint_fast8_t shiftDist;
uint32_t sigB[4];
int32_t expZ;
uint32_t sigProd[8], *extSigZPtr;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
aWPtr = (const uint32_t *) aPtr;
bWPtr = (const uint32_t *) bPtr;
zWPtr = (uint32_t *) zPtr;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
uiA96 = aWPtr[indexWordHi( 4 )];
expA = expF128UI96( uiA96 );
uiB96 = bWPtr[indexWordHi( 4 )];
expB = expF128UI96( uiB96 );
signZ = signF128UI96( uiA96 ) ^ signF128UI96( uiB96 );
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
if ( (expA == 0x7FFF) || (expB == 0x7FFF) ) {
if ( softfloat_tryPropagateNaNF128M( aWPtr, bWPtr, zWPtr ) ) return;
ptr = aWPtr;
if ( ! expA ) goto possiblyInvalid;
if ( ! expB ) {
ptr = bWPtr;
possiblyInvalid:
if (
! fracF128UI96( ptr[indexWordHi( 4 )] )
&& ! (ptr[indexWord( 4, 2 )] | ptr[indexWord( 4, 1 )]
| ptr[indexWord( 4, 0 )])
) {
softfloat_invalidF128M( zWPtr );
return;
}
}
uiZ96 = packToF128UI96( signZ, 0x7FFF, 0 );
goto uiZ96;
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
if ( expA ) {
sigA[indexWordHi( 4 )] = fracF128UI96( uiA96 ) | 0x00010000;
sigA[indexWord( 4, 2 )] = aWPtr[indexWord( 4, 2 )];
sigA[indexWord( 4, 1 )] = aWPtr[indexWord( 4, 1 )];
sigA[indexWord( 4, 0 )] = aWPtr[indexWord( 4, 0 )];
} else {
expA = softfloat_shiftNormSigF128M( aWPtr, 0, sigA );
if ( expA == -128 ) goto zero;
}
if ( expB ) {
sigB[indexWordHi( 4 )] = fracF128UI96( uiB96 ) | 0x00010000;
sigB[indexWord( 4, 2 )] = bWPtr[indexWord( 4, 2 )];
sigB[indexWord( 4, 1 )] = bWPtr[indexWord( 4, 1 )];
sigB[indexWord( 4, 0 )] = bWPtr[indexWord( 4, 0 )];
} else {
expB = softfloat_shiftNormSigF128M( bWPtr, 0, sigB );
if ( expB == -128 ) goto zero;
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
expZ = expA + expB - 0x4000;
softfloat_mul128MTo256M( sigA, sigB, sigProd );
if (
sigProd[indexWord( 8, 2 )]
|| (sigProd[indexWord( 8, 1 )] | sigProd[indexWord( 8, 0 )])
) {
sigProd[indexWord( 8, 3 )] |= 1;
}
extSigZPtr = &sigProd[indexMultiwordHi( 8, 5 )];
shiftDist = 16;
if ( extSigZPtr[indexWordHi( 5 )] & 2 ) {
++expZ;
shiftDist = 15;
}
softfloat_shortShiftLeft160M( extSigZPtr, shiftDist, extSigZPtr );
softfloat_roundPackMToF128M( signZ, expZ, extSigZPtr, zWPtr );
return;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
zero:
uiZ96 = packToF128UI96( signZ, 0, 0 );
uiZ96:
zWPtr[indexWordHi( 4 )] = uiZ96;
zWPtr[indexWord( 4, 2 )] = 0;
zWPtr[indexWord( 4, 1 )] = 0;
zWPtr[indexWord( 4, 0 )] = 0;
}
void
softfloat_addF128M(
const uint32_t *aWPtr,
const uint32_t *bWPtr,
uint32_t *zWPtr,
bool negateB
)
{
uint32_t uiA96;
int32_t expA;
uint32_t uiB96;
int32_t expB;
uint32_t uiZ96;
bool signZ, signB;
const uint32_t *tempPtr;
uint32_t sig96A, sig96B;
int32_t expDiff;
uint_fast8_t
(*addCarryMRoutinePtr)(
uint_fast8_t,
const uint32_t *,
const uint32_t *,
uint_fast8_t,
uint32_t *
);
uint32_t extSigZ[5], wordSigZ;
uint_fast8_t carry;
void (*roundPackRoutinePtr)( bool, int32_t, uint32_t *, uint32_t * );
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
uiA96 = aWPtr[indexWordHi( 4 )];
expA = expF128UI96( uiA96 );
uiB96 = bWPtr[indexWordHi( 4 )];
expB = expF128UI96( uiB96 );
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
if ( (expA == 0x7FFF) || (expB == 0x7FFF) ) {
if ( softfloat_tryPropagateNaNF128M( aWPtr, bWPtr, zWPtr ) ) return;
uiZ96 = uiA96;
if ( expB == 0x7FFF ) {
uiZ96 = uiB96 ^ packToF128UI96( negateB, 0, 0 );
if ( (expA == 0x7FFF) && (uiZ96 != uiA96) ) {
softfloat_invalidF128M( zWPtr );
return;
}
}
zWPtr[indexWordHi( 4 )] = uiZ96;
zWPtr[indexWord( 4, 2 )] = 0;
zWPtr[indexWord( 4, 1 )] = 0;
zWPtr[indexWord( 4, 0 )] = 0;
return;
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
signZ = signF128UI96( uiA96 );
signB = signF128UI96( uiB96 ) ^ negateB;
negateB = (signZ != signB);
if ( (uint32_t) (uiA96<<1) < (uint32_t) (uiB96<<1) ) {
signZ = signB;
expA = expB;
expB = expF128UI96( uiA96 );
tempPtr = aWPtr;
aWPtr = bWPtr;
bWPtr = tempPtr;
uiA96 = uiB96;
uiB96 = bWPtr[indexWordHi( 4 )];
}
sig96A = fracF128UI96( uiA96 );
sig96B = fracF128UI96( uiB96 );
if ( expA ) {
--expA;
sig96A |= 0x00010000;
if ( expB ) {
--expB;
sig96B |= 0x00010000;
}
}
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
addCarryMRoutinePtr =
negateB ? softfloat_addComplCarryM : softfloat_addCarryM;
expDiff = expA - expB;
if ( expDiff ) {
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
extSigZ[indexWordHi( 5 )] = sig96B;
extSigZ[indexWord( 5, 3 )] = bWPtr[indexWord( 4, 2 )];
extSigZ[indexWord( 5, 2 )] = bWPtr[indexWord( 4, 1 )];
extSigZ[indexWord( 5, 1 )] = bWPtr[indexWord( 4, 0 )];
extSigZ[indexWord( 5, 0 )] = 0;
softfloat_shiftRightJam160M( extSigZ, expDiff, extSigZ );
sig96B = extSigZ[indexWordHi( 5 )];
carry = 0;
if ( negateB ) {
sig96B = ~sig96B;
wordSigZ = extSigZ[indexWordLo( 5 )];
extSigZ[indexWordLo( 5 )] = -wordSigZ;
carry = ! wordSigZ;
}
carry =
(*addCarryMRoutinePtr)(
3,
&aWPtr[indexMultiwordLo( 4, 3 )],
&extSigZ[indexMultiword( 5, 3, 1 )],
carry,
&extSigZ[indexMultiword( 5, 3, 1 )]
);
wordSigZ = sig96A + sig96B + carry;
} else {
/*--------------------------------------------------------------------
*--------------------------------------------------------------------*/
extSigZ[indexWordLo( 5 )] = 0;
carry =
(*addCarryMRoutinePtr)(
3,
&aWPtr[indexMultiwordLo( 4, 3 )],
&bWPtr[indexMultiwordLo( 4, 3 )],
negateB,
&extSigZ[indexMultiword( 5, 3, 1 )]
);
if ( negateB ) {
wordSigZ = sig96A + ~sig96B + carry;
if ( wordSigZ & 0x80000000 ) {
signZ = ! signZ;
carry =
softfloat_addComplCarry96M(
&bWPtr[indexMultiwordLo( 4, 3 )],
&aWPtr[indexMultiwordLo( 4, 3 )],
1,
&extSigZ[indexMultiword( 5, 3, 1 )]
);
wordSigZ = sig96B + ~sig96A + carry;
} else {
if (
! wordSigZ && ! extSigZ[indexWord( 5, 3 )]
&& ! ( extSigZ[indexWord( 5, 2 )]
| extSigZ[indexWord( 5, 1 )]
| extSigZ[indexWord( 5, 0 )]
)
) {
signZ = (softfloat_roundingMode == softfloat_round_min);
zWPtr[indexWordHi( 4 )] = packToF128UI96( signZ, 0, 0 );
zWPtr[indexWord( 4, 2 )] = 0;
zWPtr[indexWord( 4, 1 )] = 0;
zWPtr[indexWord( 4, 0 )] = 0;
return;
}
}
} else {
wordSigZ = sig96A + sig96B + carry;
}
}
extSigZ[indexWordHi( 5 )] = wordSigZ;
/*------------------------------------------------------------------------
*------------------------------------------------------------------------*/
roundPackRoutinePtr = softfloat_normRoundPackMToF128M;
if ( 0x00010000 <= wordSigZ ) {
if ( 0x00020000 <= wordSigZ ) {
++expA;
softfloat_shortShiftRightJam160M( extSigZ, 1, extSigZ );
}
roundPackRoutinePtr = softfloat_roundPackMToF128M;
}
(*roundPackRoutinePtr)( signZ, expA, extSigZ, zWPtr );
}