/*--- poly_l2p1() -----------------------------------------------------------+
| Base 2 logarithm by a polynomial approximation. |
| log2(x+1) |
+---------------------------------------------------------------------------*/
int poly_l2p1(u_char sign0, u_char sign1,
FPU_REG *st0_ptr, FPU_REG *st1_ptr, FPU_REG *dest)
{
u_char tag;
s32 exponent;
Xsig accumulator, yaccum;
if ( exponent16(st0_ptr) < 0 )
{
log2_kernel(st0_ptr, sign0, &accumulator, &exponent);
yaccum.lsw = 0;
XSIG_LL(yaccum) = significand(st1_ptr);
mul_Xsig_Xsig(&accumulator, &yaccum);
exponent += round_Xsig(&accumulator);
exponent += exponent16(st1_ptr) + 1;
if ( exponent < EXP_WAY_UNDER ) exponent = EXP_WAY_UNDER;
significand(dest) = XSIG_LL(accumulator);
setexponent16(dest, exponent);
tag = FPU_round(dest, 1, 0, FULL_PRECISION, sign0 ^ sign1);
FPU_settagi(1, tag);
if ( tag == TAG_Valid )
set_precision_flag_up(); /* 80486 appears to always do this */
}
else
{
/* The magnitude of st0_ptr is far too large. */
if ( sign0 != SIGN_POS )
{
/* Trying to get the log of a negative number. */
#ifdef PECULIAR_486 /* Stupid 80486 doesn't worry about log(negative). */
changesign(st1_ptr);
#else
if ( arith_invalid(1) < 0 )
return 1;
#endif /* PECULIAR_486 */
}
/* 80486 appears to do this */
if ( sign0 == SIGN_NEG )
set_precision_flag_down();
else
set_precision_flag_up();
}
if ( exponent(dest) <= EXP_UNDER )
EXCEPTION(EX_Underflow);
return 0;
}
static void f2xm1(FPU_REG *st0_ptr, u_char tag)
{
FPU_REG a;
clear_C1();
if ( tag == TAG_Valid )
{
/* For an 80486 FPU, the result is undefined if the arg is >= 1.0 */
if ( exponent(st0_ptr) < 0 )
{
denormal_arg:
FPU_to_exp16(st0_ptr, &a);
/* poly_2xm1(x) requires 0 < st(0) < 1. */
poly_2xm1(getsign(st0_ptr), &a, st0_ptr);
}
set_precision_flag_up(); /* 80486 appears to always do this */
return;
}
if ( tag == TAG_Zero )
return;
if ( tag == TAG_Special )
tag = FPU_Special(st0_ptr);
switch ( tag )
{
case TW_Denormal:
if ( denormal_operand() < 0 )
return;
goto denormal_arg;
case TW_Infinity:
if ( signnegative(st0_ptr) )
{
/* -infinity gives -1 (p16-10) */
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
setnegative(st0_ptr);
}
return;
default:
single_arg_error(st0_ptr, tag);
}
}
static void f2xm1(FPU_REG *st0_ptr, u_char tag)
{
FPU_REG a;
clear_C1();
if (tag == TAG_Valid) {
if (exponent(st0_ptr) < 0) {
denormal_arg:
FPU_to_exp16(st0_ptr, &a);
poly_2xm1(getsign(st0_ptr), &a, st0_ptr);
}
set_precision_flag_up();
return;
}
if (tag == TAG_Zero)
return;
if (tag == TAG_Special)
tag = FPU_Special(st0_ptr);
switch (tag) {
case TW_Denormal:
if (denormal_operand() < 0)
return;
goto denormal_arg;
case TW_Infinity:
if (signnegative(st0_ptr)) {
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
setnegative(st0_ptr);
}
return;
default:
single_arg_error(st0_ptr, tag);
}
}
static void fptan(FPU_REG *st0_ptr, u_char st0_tag)
{
FPU_REG *st_new_ptr;
int q;
u_char arg_sign = getsign(st0_ptr);
if (st0_tag == TAG_Empty) {
FPU_stack_underflow();
if (control_word & CW_Invalid) {
st_new_ptr = &st(-1);
push();
FPU_stack_underflow();
}
return;
}
if (STACK_OVERFLOW) {
FPU_stack_overflow();
return;
}
if (st0_tag == TAG_Valid) {
if (exponent(st0_ptr) > -40) {
if ((q = trig_arg(st0_ptr, 0)) == -1) {
return;
}
poly_tan(st0_ptr);
setsign(st0_ptr, (q & 1) ^ (arg_sign != 0));
set_precision_flag_up();
} else {
denormal_arg:
FPU_to_exp16(st0_ptr, st0_ptr);
st0_tag =
FPU_round(st0_ptr, 1, 0, FULL_PRECISION, arg_sign);
FPU_settag0(st0_tag);
}
push();
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
return;
}
if (st0_tag == TAG_Zero) {
push();
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
setcc(0);
return;
}
if (st0_tag == TAG_Special)
st0_tag = FPU_Special(st0_ptr);
if (st0_tag == TW_Denormal) {
if (denormal_operand() < 0)
return;
goto denormal_arg;
}
if (st0_tag == TW_Infinity) {
if (arith_invalid(0) >= 0) {
st_new_ptr = &st(-1);
push();
arith_invalid(0);
}
return;
}
single_arg_2_error(st0_ptr, st0_tag);
}
/*--- poly_atan() -----------------------------------------------------------+
| |
+---------------------------------------------------------------------------*/
void poly_atan(FPU_REG *st0_ptr, u_char st0_tag,
FPU_REG *st1_ptr, u_char st1_tag)
{
u_char transformed, inverted,
sign1, sign2;
int exponent;
long int dummy_exp;
Xsig accumulator, Numer, Denom, accumulatore, argSignif,
argSq, argSqSq;
u_char tag;
sign1 = getsign(st0_ptr);
sign2 = getsign(st1_ptr);
if ( st0_tag == TAG_Valid )
{
exponent = exponent(st0_ptr);
}
else
{
/* This gives non-compatible stack contents... */
FPU_to_exp16(st0_ptr, st0_ptr);
exponent = exponent16(st0_ptr);
}
if ( st1_tag == TAG_Valid )
{
exponent -= exponent(st1_ptr);
}
else
{
/* This gives non-compatible stack contents... */
FPU_to_exp16(st1_ptr, st1_ptr);
exponent -= exponent16(st1_ptr);
}
if ( (exponent < 0) || ((exponent == 0) &&
((st0_ptr->sigh < st1_ptr->sigh) ||
((st0_ptr->sigh == st1_ptr->sigh) &&
(st0_ptr->sigl < st1_ptr->sigl))) ) )
{
inverted = 1;
Numer.lsw = Denom.lsw = 0;
XSIG_LL(Numer) = significand(st0_ptr);
XSIG_LL(Denom) = significand(st1_ptr);
}
else
{
inverted = 0;
exponent = -exponent;
Numer.lsw = Denom.lsw = 0;
XSIG_LL(Numer) = significand(st1_ptr);
XSIG_LL(Denom) = significand(st0_ptr);
}
div_Xsig(&Numer, &Denom, &argSignif);
exponent += norm_Xsig(&argSignif);
if ( (exponent >= -1)
|| ((exponent == -2) && (argSignif.msw > 0xd413ccd0)) )
{
/* The argument is greater than sqrt(2)-1 (=0.414213562...) */
/* Convert the argument by an identity for atan */
transformed = 1;
if ( exponent >= 0 )
{
#ifdef PARANOID
if ( !( (exponent == 0) &&
(argSignif.lsw == 0) && (argSignif.midw == 0) &&
(argSignif.msw == 0x80000000) ) )
{
EXCEPTION(EX_INTERNAL|0x104); /* There must be a logic error */
return;
}
#endif /* PARANOID */
argSignif.msw = 0; /* Make the transformed arg -> 0.0 */
}
else
{
Numer.lsw = Denom.lsw = argSignif.lsw;
XSIG_LL(Numer) = XSIG_LL(Denom) = XSIG_LL(argSignif);
if ( exponent < -1 )
shr_Xsig(&Numer, -1-exponent);
negate_Xsig(&Numer);
shr_Xsig(&Denom, -exponent);
Denom.msw |= 0x80000000;
div_Xsig(&Numer, &Denom, &argSignif);
exponent = -1 + norm_Xsig(&argSignif);
}
}
else
{
transformed = 0;
}
argSq.lsw = argSignif.lsw; argSq.midw = argSignif.midw;
argSq.msw = argSignif.msw;
mul_Xsig_Xsig(&argSq, &argSq);
argSqSq.lsw = argSq.lsw; argSqSq.midw = argSq.midw; argSqSq.msw = argSq.msw;
mul_Xsig_Xsig(&argSqSq, &argSqSq);
accumulatore.lsw = argSq.lsw;
XSIG_LL(accumulatore) = XSIG_LL(argSq);
shr_Xsig(&argSq, 2*(-1-exponent-1));
shr_Xsig(&argSqSq, 4*(-1-exponent-1));
/* Now have argSq etc with binary point at the left
.1xxxxxxxx */
/* Do the basic fixed point polynomial evaluation */
accumulator.msw = accumulator.midw = accumulator.lsw = 0;
polynomial_Xsig(&accumulator, &XSIG_LL(argSqSq),
oddplterms, HIPOWERop-1);
mul64_Xsig(&accumulator, &XSIG_LL(argSq));
negate_Xsig(&accumulator);
polynomial_Xsig(&accumulator, &XSIG_LL(argSqSq), oddnegterms, HIPOWERon-1);
negate_Xsig(&accumulator);
add_two_Xsig(&accumulator, &fixedpterm, &dummy_exp);
mul64_Xsig(&accumulatore, &denomterm);
shr_Xsig(&accumulatore, 1 + 2*(-1-exponent));
accumulatore.msw |= 0x80000000;
div_Xsig(&accumulator, &accumulatore, &accumulator);
mul_Xsig_Xsig(&accumulator, &argSignif);
mul_Xsig_Xsig(&accumulator, &argSq);
shr_Xsig(&accumulator, 3);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &argSignif);
if ( transformed )
{
/* compute pi/4 - accumulator */
shr_Xsig(&accumulator, -1-exponent);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &pi_signif);
exponent = -1;
}
if ( inverted )
{
/* compute pi/2 - accumulator */
shr_Xsig(&accumulator, -exponent);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &pi_signif);
exponent = 0;
}
if ( sign1 )
{
/* compute pi - accumulator */
shr_Xsig(&accumulator, 1 - exponent);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &pi_signif);
exponent = 1;
}
exponent += round_Xsig(&accumulator);
significand(st1_ptr) = XSIG_LL(accumulator);
setexponent16(st1_ptr, exponent);
tag = FPU_round(st1_ptr, 1, 0, FULL_PRECISION, sign2);
FPU_settagi(1, tag);
set_precision_flag_up(); /* We do not really know if up or down,
use this as the default. */
}
static void fptan(FPU_REG *st0_ptr, u_char st0_tag)
{
FPU_REG *st_new_ptr;
int q;
u_char arg_sign = getsign(st0_ptr);
/* Stack underflow has higher priority */
if (st0_tag == TAG_Empty) {
FPU_stack_underflow(); /* Puts a QNaN in st(0) */
if (control_word & CW_Invalid) {
st_new_ptr = &st(-1);
push();
FPU_stack_underflow(); /* Puts a QNaN in the new st(0) */
}
return;
}
if (STACK_OVERFLOW) {
FPU_stack_overflow();
return;
}
if (st0_tag == TAG_Valid) {
if (exponent(st0_ptr) > -40) {
if ((q = trig_arg(st0_ptr, 0)) == -1) {
/* Operand is out of range */
return;
}
poly_tan(st0_ptr);
setsign(st0_ptr, (q & 1) ^ (arg_sign != 0));
set_precision_flag_up(); /* We do not really know if up or down */
} else {
/* For a small arg, the result == the argument */
/* Underflow may happen */
denormal_arg:
FPU_to_exp16(st0_ptr, st0_ptr);
st0_tag =
FPU_round(st0_ptr, 1, 0, FULL_PRECISION, arg_sign);
FPU_settag0(st0_tag);
}
push();
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
return;
}
if (st0_tag == TAG_Zero) {
push();
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
setcc(0);
return;
}
if (st0_tag == TAG_Special)
st0_tag = FPU_Special(st0_ptr);
if (st0_tag == TW_Denormal) {
if (denormal_operand() < 0)
return;
goto denormal_arg;
}
if (st0_tag == TW_Infinity) {
/* The 80486 treats infinity as an invalid operand */
if (arith_invalid(0) >= 0) {
st_new_ptr = &st(-1);
push();
arith_invalid(0);
}
return;
}
single_arg_2_error(st0_ptr, st0_tag);
}
void poly_atan(FPU_REG *st0_ptr, u_char st0_tag,
FPU_REG *st1_ptr, u_char st1_tag)
{
u_char transformed, inverted, sign1, sign2;
int exponent;
long int dummy_exp;
Xsig accumulator, Numer, Denom, accumulatore, argSignif, argSq, argSqSq;
u_char tag;
sign1 = getsign(st0_ptr);
sign2 = getsign(st1_ptr);
if (st0_tag == TAG_Valid) {
exponent = exponent(st0_ptr);
} else {
FPU_to_exp16(st0_ptr, st0_ptr);
exponent = exponent16(st0_ptr);
}
if (st1_tag == TAG_Valid) {
exponent -= exponent(st1_ptr);
} else {
FPU_to_exp16(st1_ptr, st1_ptr);
exponent -= exponent16(st1_ptr);
}
if ((exponent < 0) || ((exponent == 0) &&
((st0_ptr->sigh < st1_ptr->sigh) ||
((st0_ptr->sigh == st1_ptr->sigh) &&
(st0_ptr->sigl < st1_ptr->sigl))))) {
inverted = 1;
Numer.lsw = Denom.lsw = 0;
XSIG_LL(Numer) = significand(st0_ptr);
XSIG_LL(Denom) = significand(st1_ptr);
} else {
inverted = 0;
exponent = -exponent;
Numer.lsw = Denom.lsw = 0;
XSIG_LL(Numer) = significand(st1_ptr);
XSIG_LL(Denom) = significand(st0_ptr);
}
div_Xsig(&Numer, &Denom, &argSignif);
exponent += norm_Xsig(&argSignif);
if ((exponent >= -1)
|| ((exponent == -2) && (argSignif.msw > 0xd413ccd0))) {
transformed = 1;
if (exponent >= 0) {
#ifdef PARANOID
if (!((exponent == 0) &&
(argSignif.lsw == 0) && (argSignif.midw == 0) &&
(argSignif.msw == 0x80000000))) {
EXCEPTION(EX_INTERNAL | 0x104);
return;
}
#endif
argSignif.msw = 0;
} else {
Numer.lsw = Denom.lsw = argSignif.lsw;
XSIG_LL(Numer) = XSIG_LL(Denom) = XSIG_LL(argSignif);
if (exponent < -1)
shr_Xsig(&Numer, -1 - exponent);
negate_Xsig(&Numer);
shr_Xsig(&Denom, -exponent);
Denom.msw |= 0x80000000;
div_Xsig(&Numer, &Denom, &argSignif);
exponent = -1 + norm_Xsig(&argSignif);
}
} else {
transformed = 0;
}
argSq.lsw = argSignif.lsw;
argSq.midw = argSignif.midw;
argSq.msw = argSignif.msw;
mul_Xsig_Xsig(&argSq, &argSq);
argSqSq.lsw = argSq.lsw;
argSqSq.midw = argSq.midw;
argSqSq.msw = argSq.msw;
mul_Xsig_Xsig(&argSqSq, &argSqSq);
accumulatore.lsw = argSq.lsw;
XSIG_LL(accumulatore) = XSIG_LL(argSq);
shr_Xsig(&argSq, 2 * (-1 - exponent - 1));
shr_Xsig(&argSqSq, 4 * (-1 - exponent - 1));
accumulator.msw = accumulator.midw = accumulator.lsw = 0;
polynomial_Xsig(&accumulator, &XSIG_LL(argSqSq),
oddplterms, HIPOWERop - 1);
mul64_Xsig(&accumulator, &XSIG_LL(argSq));
negate_Xsig(&accumulator);
polynomial_Xsig(&accumulator, &XSIG_LL(argSqSq), oddnegterms,
HIPOWERon - 1);
negate_Xsig(&accumulator);
add_two_Xsig(&accumulator, &fixedpterm, &dummy_exp);
mul64_Xsig(&accumulatore, &denomterm);
shr_Xsig(&accumulatore, 1 + 2 * (-1 - exponent));
accumulatore.msw |= 0x80000000;
div_Xsig(&accumulator, &accumulatore, &accumulator);
mul_Xsig_Xsig(&accumulator, &argSignif);
mul_Xsig_Xsig(&accumulator, &argSq);
shr_Xsig(&accumulator, 3);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &argSignif);
if (transformed) {
shr_Xsig(&accumulator, -1 - exponent);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &pi_signif);
exponent = -1;
}
if (inverted) {
shr_Xsig(&accumulator, -exponent);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &pi_signif);
exponent = 0;
}
if (sign1) {
shr_Xsig(&accumulator, 1 - exponent);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &pi_signif);
exponent = 1;
}
exponent += round_Xsig(&accumulator);
significand(st1_ptr) = XSIG_LL(accumulator);
setexponent16(st1_ptr, exponent);
tag = FPU_round(st1_ptr, 1, 0, FULL_PRECISION, sign2);
FPU_settagi(1, tag);
set_precision_flag_up();
}
/*--- poly_l2() -------------------------------------------------------------+
| Base 2 logarithm by a polynomial approximation. |
+---------------------------------------------------------------------------*/
void poly_l2(FPU_REG *st0_ptr, FPU_REG *st1_ptr, u_char st1_sign)
{
long int exponent, expon, expon_expon;
Xsig accumulator, expon_accum, yaccum;
u_char sign, argsign;
FPU_REG x;
int tag;
exponent = exponent16(st0_ptr);
/* From st0_ptr, make a number > sqrt(2)/2 and < sqrt(2) */
if (st0_ptr->sigh > (unsigned)0xb504f334) {
/* Treat as sqrt(2)/2 < st0_ptr < 1 */
significand(&x) = -significand(st0_ptr);
setexponent16(&x, -1);
exponent++;
argsign = SIGN_NEG;
} else {
/* Treat as 1 <= st0_ptr < sqrt(2) */
x.sigh = st0_ptr->sigh - 0x80000000;
x.sigl = st0_ptr->sigl;
setexponent16(&x, 0);
argsign = SIGN_POS;
}
tag = FPU_normalize_nuo(&x);
if (tag == TAG_Zero) {
expon = 0;
accumulator.msw = accumulator.midw = accumulator.lsw = 0;
} else {
log2_kernel(&x, argsign, &accumulator, &expon);
}
if (exponent < 0) {
sign = SIGN_NEG;
exponent = -exponent;
} else
sign = SIGN_POS;
expon_accum.msw = exponent;
expon_accum.midw = expon_accum.lsw = 0;
if (exponent) {
expon_expon = 31 + norm_Xsig(&expon_accum);
shr_Xsig(&accumulator, expon_expon - expon);
if (sign ^ argsign)
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &expon_accum);
} else {
expon_expon = expon;
sign = argsign;
}
yaccum.lsw = 0;
XSIG_LL(yaccum) = significand(st1_ptr);
mul_Xsig_Xsig(&accumulator, &yaccum);
expon_expon += round_Xsig(&accumulator);
if (accumulator.msw == 0) {
FPU_copy_to_reg1(&CONST_Z, TAG_Zero);
return;
}
significand(st1_ptr) = XSIG_LL(accumulator);
setexponent16(st1_ptr, expon_expon + exponent16(st1_ptr) + 1);
tag = FPU_round(st1_ptr, 1, 0, FULL_PRECISION, sign ^ st1_sign);
FPU_settagi(1, tag);
set_precision_flag_up(); /* 80486 appears to always do this */
return;
}
