/**
* Greater-than comparison between two floats. NaNs are not recognized.
*
* @a First quadruple-precision operand.
* @b Second quadruple-precision operand.
* @return 1 if a is greater than b, 0 otherwise.
*/
int is_float128_gt(float128 a, float128 b)
{
uint64_t tmp_hi;
uint64_t tmp_lo;
or128(a.bin.hi, a.bin.lo,
b.bin.hi, b.bin.lo, &tmp_hi, &tmp_lo);
and128(tmp_hi, tmp_lo,
0x7FFFFFFFFFFFFFFFll, 0xFFFFFFFFFFFFFFFFll, &tmp_hi, &tmp_lo);
if (eq128(tmp_hi, tmp_lo, 0x0ll, 0x0ll)) {
/* zeroes are equal with any sign */
return 0;
}
if ((a.parts.sign) && (b.parts.sign)) {
/* if both are negative, greater is that with smaller binary value */
return lt128(a.bin.hi, a.bin.lo, b.bin.hi, b.bin.lo);
}
/*
* lets negate signs - now will be positive numbers always
* bigger than negative (first bit will be set for unsigned
* integer comparison)
*/
a.parts.sign = !a.parts.sign;
b.parts.sign = !b.parts.sign;
return lt128(b.bin.hi, b.bin.lo, a.bin.hi, a.bin.lo);
}
static int reduce_trig_arg(int expDiff, int &zSign, Bit64u &aSig0, Bit64u &aSig1)
{
Bit64u term0, term1, q = 0;
if (expDiff < 0) {
shift128Right(aSig0, 0, 1, &aSig0, &aSig1);
expDiff = 0;
}
if (expDiff > 0) {
q = argument_reduction_kernel(aSig0, expDiff, &aSig0, &aSig1);
}
else {
if (FLOAT_PI_HI <= aSig0) {
aSig0 -= FLOAT_PI_HI;
q = 1;
}
}
shift128Right(FLOAT_PI_HI, FLOAT_PI_LO, 1, &term0, &term1);
if (! lt128(aSig0, aSig1, term0, term1))
{
int lt = lt128(term0, term1, aSig0, aSig1);
int eq = eq128(aSig0, aSig1, term0, term1);
if ((eq && (q & 1)) || lt) {
zSign = !zSign;
++q;
}
if (lt) sub128(FLOAT_PI_HI, FLOAT_PI_LO, aSig0, aSig1, &aSig0, &aSig1);
}
return (int)(q & 3);
}
/**
* Determines whether the given float represents signalling NaN.
*
* @param ld Quadruple-precision float.
* @return 1 if float is signalling NaN, 0 otherwise.
*/
int is_float128_signan(float128 ld)
{
/* SigNaN : exp = 0x7fff and fraction = 0xxxxx..x (binary),
* where at least one x is nonzero */
return ((ld.parts.exp == 0x7FFF) &&
(ld.parts.frac_hi || ld.parts.frac_lo) &&
lt128(ld.parts.frac_hi, ld.parts.frac_lo, 0x800000000000ll, 0x0ll));
}
float128 float32_to_float128(float32 a)
{
float128 result;
uint64_t frac_hi, frac_lo;
uint64_t tmp_hi, tmp_lo;
result.parts.sign = a.parts.sign;
result.parts.frac_hi = 0;
result.parts.frac_lo = a.parts.fraction;
lshift128(result.parts.frac_hi, result.parts.frac_lo,
(FLOAT128_FRACTION_SIZE - FLOAT32_FRACTION_SIZE),
&frac_hi, &frac_lo);
result.parts.frac_hi = frac_hi;
result.parts.frac_lo = frac_lo;
if ((is_float32_infinity(a)) || (is_float32_nan(a))) {
result.parts.exp = FLOAT128_MAX_EXPONENT;
// TODO; check if its correct for SigNaNs
return result;
}
result.parts.exp = a.parts.exp + ((int) FLOAT128_BIAS - FLOAT32_BIAS);
if (a.parts.exp == 0) {
/* normalize denormalized numbers */
if (eq128(result.parts.frac_hi,
result.parts.frac_lo, 0x0ll, 0x0ll)) { /* fix zero */
result.parts.exp = 0;
return result;
}
frac_hi = result.parts.frac_hi;
frac_lo = result.parts.frac_lo;
and128(frac_hi, frac_lo,
FLOAT128_HIDDEN_BIT_MASK_HI, FLOAT128_HIDDEN_BIT_MASK_LO,
&tmp_hi, &tmp_lo);
while (!lt128(0x0ll, 0x0ll, tmp_hi, tmp_lo)) {
lshift128(frac_hi, frac_lo, 1, &frac_hi, &frac_lo);
--result.parts.exp;
}
++result.parts.exp;
result.parts.frac_hi = frac_hi;
result.parts.frac_lo = frac_lo;
}
return result;
}
static floatx80 do_fprem(floatx80 a, floatx80 b, Bit64u &q, int rounding_mode, float_status_t &status)
{
Bit32s aExp, bExp, zExp, expDiff;
Bit64u aSig0, aSig1, bSig;
int aSign;
q = 0;
// handle unsupported extended double-precision floating encodings
if (floatx80_is_unsupported(a) || floatx80_is_unsupported(b))
{
float_raise(status, float_flag_invalid);
return floatx80_default_nan;
}
aSig0 = extractFloatx80Frac(a);
aExp = extractFloatx80Exp(a);
aSign = extractFloatx80Sign(a);
bSig = extractFloatx80Frac(b);
bExp = extractFloatx80Exp(b);
if (aExp == 0x7FFF) {
if ((Bit64u) (aSig0<<1)
|| ((bExp == 0x7FFF) && (Bit64u) (bSig<<1)))
{
return propagateFloatx80NaN(a, b, status);
}
goto invalid;
}
if (bExp == 0x7FFF) {
if ((Bit64u) (bSig<<1)) return propagateFloatx80NaN(a, b, status);
return a;
}
if (bExp == 0) {
if (bSig == 0) {
invalid:
float_raise(status, float_flag_invalid);
return floatx80_default_nan;
}
float_raise(status, float_flag_denormal);
normalizeFloatx80Subnormal(bSig, &bExp, &bSig);
}
if (aExp == 0) {
if ((Bit64u) (aSig0<<1) == 0) return a;
float_raise(status, float_flag_denormal);
normalizeFloatx80Subnormal(aSig0, &aExp, &aSig0);
}
expDiff = aExp - bExp;
aSig1 = 0;
if (expDiff >= 64) {
int n = (expDiff & 0x1f) | 0x20;
remainder_kernel(aSig0, bSig, n, &aSig0, &aSig1);
zExp = aExp - n;
q = (Bit64u) -1;
}
else {
zExp = bExp;
if (expDiff < 0) {
if (expDiff < -1)
return (a.fraction & BX_CONST64(0x8000000000000000)) ?
packFloatx80(aSign, aExp, aSig0) : a;
shift128Right(aSig0, 0, 1, &aSig0, &aSig1);
expDiff = 0;
}
if (expDiff > 0) {
q = remainder_kernel(aSig0, bSig, expDiff, &aSig0, &aSig1);
}
else {
if (bSig <= aSig0) {
aSig0 -= bSig;
q = 1;
}
}
if (rounding_mode == float_round_nearest_even)
{
Bit64u term0, term1;
shift128Right(bSig, 0, 1, &term0, &term1);
if (! lt128(aSig0, aSig1, term0, term1))
{
int lt = lt128(term0, term1, aSig0, aSig1);
int eq = eq128(aSig0, aSig1, term0, term1);
if ((eq && (q & 1)) || lt) {
aSign = !aSign;
++q;
}
if (lt) sub128(bSig, 0, aSig0, aSig1, &aSig0, &aSig1);
}
}
}
return normalizeRoundAndPackFloatx80(80, aSign, zExp, aSig0, aSig1, status);
}