void
fxch_i()
{
/* fxch st(i) */
FPU_REG t;
register FPU_REG *sti_ptr = &st(FPU_rm);
if (FPU_st0_tag == TW_Empty) {
if (sti_ptr->tag == TW_Empty) {
stack_underflow();
stack_underflow_i(FPU_rm);
return;
}
reg_move(sti_ptr, FPU_st0_ptr);
stack_underflow_i(FPU_rm);
return;
}
if (sti_ptr->tag == TW_Empty) {
reg_move(FPU_st0_ptr, sti_ptr);
stack_underflow();
return;
}
reg_move(FPU_st0_ptr, &t);
reg_move(sti_ptr, FPU_st0_ptr);
reg_move(&t, sti_ptr);
}
static int
trig_arg(FPU_REG * X)
{
FPU_REG tmp, quot;
int rv;
long long q;
int old_cw = control_word;
control_word &= ~CW_RC;
control_word |= RC_CHOP;
reg_move(X, ");
reg_div(", &CONST_PI2, ", FULL_PRECISION);
reg_move(", &tmp);
round_to_int(&tmp);
if (tmp.sigh & 0x80000000)
return -1; /* |Arg| is >= 2^63 */
tmp.exp = EXP_BIAS + 63;
q = *(long long *) &(tmp.sigl);
normalize(&tmp);
reg_sub(", &tmp, X, FULL_PRECISION);
rv = q & 7;
control_word = old_cw;
return rv;;
}
static void
fptan(void)
{
FPU_REG *st_new_ptr;
int q;
char arg_sign = FPU_st0_ptr->sign;
if (STACK_OVERFLOW) {
stack_overflow();
return;
}
switch (FPU_st0_tag) {
case TW_Valid:
#ifdef DENORM_OPERAND
if ((FPU_st0_ptr->exp <= EXP_UNDER) && (denormal_operand()))
return;
#endif /* DENORM_OPERAND */
FPU_st0_ptr->sign = SIGN_POS;
if ((q = trig_arg(FPU_st0_ptr)) != -1) {
if (q & 1)
reg_sub(&CONST_1, FPU_st0_ptr, FPU_st0_ptr, FULL_PRECISION);
poly_tan(FPU_st0_ptr, FPU_st0_ptr);
FPU_st0_ptr->sign = (q & 1) ^ arg_sign;
if (FPU_st0_ptr->exp <= EXP_UNDER)
arith_underflow(FPU_st0_ptr);
push();
reg_move(&CONST_1, FPU_st0_ptr);
setcc(0);
} else {
/* Operand is out of range */
setcc(SW_C2);
FPU_st0_ptr->sign = arg_sign; /* restore st(0) */
return;
}
break;
case TW_Infinity:
/* Operand is out of range */
setcc(SW_C2);
FPU_st0_ptr->sign = arg_sign; /* restore st(0) */
return;
case TW_Zero:
push();
reg_move(&CONST_1, FPU_st0_ptr);
setcc(0);
break;
default:
single_arg_error();
break;
}
}
void
fstp_i()
{
/* fstp st(i) */
reg_move(FPU_st0_ptr, &st(FPU_rm));
pop();
}
void
search(short n, boolean is_auto)
{
short s, i, j, row, col, t;
short shown = 0, found = 0;
static boolean reg_search;
for (i = -1; i <= 1; i++) {
for (j = -1; j <= 1; j++) {
row = rogue.row + i;
col = rogue.col + j;
if ((row < MIN_ROW) || (row >= (ROGUE_LINES - 1)) ||
(col < 0) || (col >= ROGUE_COLUMNS)) {
continue;
}
if (dungeon[row][col] & HIDDEN) {
found++;
}
}
}
for (s = 0; s < n; s++) {
for (i = -1; i <= 1; i++) {
for (j = -1; j <= 1; j++) {
row = rogue.row + i;
col = rogue.col + j;
if ((row < MIN_ROW) || (row >= (ROGUE_LINES - 1)) ||
(col < 0) || (col >= ROGUE_COLUMNS)) {
continue;
}
if (dungeon[row][col] & HIDDEN) {
if (rand_percent(17 + (rogue.exp + ring_exp))) {
dungeon[row][col] &= (~HIDDEN);
if ((!blind) && ((row != rogue.row) ||
(col != rogue.col))) {
mvaddch_rogue(row, col,
get_dungeon_char(row, col));
}
shown++;
if (dungeon[row][col] & TRAP) {
t = trap_at(row, col);
message(trap_strings[t * 2], 1);
}
}
}
if (((shown == found) && (found > 0)) || interrupted) {
return;
}
}
}
if ((!is_auto) && (reg_search = !reg_search)) {
(void) reg_move();
}
}
}
static void
fld_const(FPU_REG * c)
{
FPU_REG *st_new_ptr;
if (STACK_OVERFLOW) {
stack_overflow();
return;
}
push();
reg_move(c, FPU_st0_ptr);
status_word &= ~SW_C1;
}
static void
f2xm1(void)
{
switch (FPU_st0_tag) {
case TW_Valid:
{
FPU_REG rv, tmp;
#ifdef DENORM_OPERAND
if ((FPU_st0_ptr->exp <= EXP_UNDER) && (denormal_operand()))
return;
#endif /* DENORM_OPERAND */
if (FPU_st0_ptr->sign == SIGN_POS) {
/* poly_2xm1(x) requires 0 < x < 1. */
if (poly_2xm1(FPU_st0_ptr, &rv))
return; /* error */
reg_mul(&rv, FPU_st0_ptr, FPU_st0_ptr, FULL_PRECISION);
} else {
/* **** Should change poly_2xm1() to at least handle numbers near 0 */
/* poly_2xm1(x) doesn't handle negative
* numbers. */
/* So we compute (poly_2xm1(x+1)-1)/2, for -1
* < x < 0 */
reg_add(FPU_st0_ptr, &CONST_1, &tmp, FULL_PRECISION);
poly_2xm1(&tmp, &rv);
reg_mul(&rv, &tmp, &tmp, FULL_PRECISION);
reg_sub(&tmp, &CONST_1, FPU_st0_ptr, FULL_PRECISION);
FPU_st0_ptr->exp--;
if (FPU_st0_ptr->exp <= EXP_UNDER)
arith_underflow(FPU_st0_ptr);
}
return;
}
case TW_Zero:
return;
case TW_Infinity:
if (FPU_st0_ptr->sign == SIGN_NEG) {
/* -infinity gives -1 (p16-10) */
reg_move(&CONST_1, FPU_st0_ptr);
FPU_st0_ptr->sign = SIGN_NEG;
}
return;
default:
single_arg_error();
}
}
void
vanish(object *obj, short rm, object *pack)
{
if (obj->quantity > 1) {
obj->quantity--;
} else {
if (obj->in_use_flags & BEING_WIELDED) {
unwield(obj);
} else if (obj->in_use_flags & BEING_WORN) {
unwear(obj);
} else if (obj->in_use_flags & ON_EITHER_HAND) {
un_put_on(obj);
}
take_from_pack(obj, pack);
free_object(obj);
}
if (rm) {
reg_move();
}
}
/* Convert a long to register */
static void
convert_l2reg(long *arg, FPU_REG * dest)
{
long num = *arg;
if (num == 0) {
reg_move(&CONST_Z, dest);
return;
}
if (num > 0)
dest->sign = SIGN_POS;
else {
num = -num;
dest->sign = SIGN_NEG;
}
dest->sigh = num;
dest->sigl = 0;
dest->exp = EXP_BIAS + 31;
dest->tag = TW_Valid;
normalize(dest);
}
void
fld_i_()
{
FPU_REG *st_new_ptr;
if (STACK_OVERFLOW) {
stack_overflow();
return;
}
/* fld st(i) */
if (NOT_EMPTY(FPU_rm)) {
reg_move(&st(FPU_rm), st_new_ptr);
push();
} else {
if (control_word & EX_Invalid) {
/* The masked response */
push();
stack_underflow();
} else
EXCEPTION(EX_StackUnder);
}
}
/*--- poly_atan() -----------------------------------------------------------+
| |
+---------------------------------------------------------------------------*/
void
poly_atan(FPU_REG * arg)
{
char recursions = 0;
short exponent;
FPU_REG odd_poly, even_poly, pos_poly, neg_poly;
FPU_REG argSq;
long long arg_signif, argSqSq;
#ifdef PARANOID
if (arg->sign != 0) { /* Can't hack a number < 0.0 */
arith_invalid(arg);
return;
} /* Need a positive number */
#endif /* PARANOID */
exponent = arg->exp - EXP_BIAS;
if (arg->tag == TW_Zero) {
/* Return 0.0 */
reg_move(&CONST_Z, arg);
return;
}
if (exponent >= -2) {
/* argument is in the range [0.25 .. 1.0] */
if (exponent >= 0) {
#ifdef PARANOID
if ((exponent == 0) &&
(arg->sigl == 0) && (arg->sigh == 0x80000000))
#endif /* PARANOID */
{
reg_move(&CONST_PI4, arg);
return;
}
#ifdef PARANOID
EXCEPTION(EX_INTERNAL | 0x104); /* There must be a logic
* error */
#endif /* PARANOID */
}
/* If the argument is greater than sqrt(2)-1 (=0.414213562...) */
/* convert the argument by an identity for atan */
if ((exponent >= -1) || (arg->sigh > 0xd413ccd0)) {
FPU_REG numerator, denom;
recursions++;
arg_signif = *(long long *) &(arg->sigl);
if (exponent < -1) {
if (shrx(&arg_signif, -1 - exponent) >= (unsigned)0x80000000)
arg_signif++; /* round up */
}
*(long long *) &(numerator.sigl) = -arg_signif;
numerator.exp = EXP_BIAS - 1;
normalize(&numerator); /* 1 - arg */
arg_signif = *(long long *) &(arg->sigl);
if (shrx(&arg_signif, -exponent) >= (unsigned)0x80000000)
arg_signif++; /* round up */
*(long long *) &(denom.sigl) = arg_signif;
denom.sigh |= 0x80000000; /* 1 + arg */
arg->exp = numerator.exp;
reg_u_div(&numerator, &denom, arg, FULL_PRECISION);
exponent = arg->exp - EXP_BIAS;
}
}
*(long long *) &arg_signif = *(long long *) &(arg->sigl);
#ifdef PARANOID
/* This must always be true */
if (exponent >= -1) {
EXCEPTION(EX_INTERNAL | 0x120); /* There must be a logic error */
}
#endif /* PARANOID */
/* shift the argument right by the required places */
if (shrx(&arg_signif, -1 - exponent) >= (unsigned)0x80000000)
arg_signif++; /* round up */
/* Now have arg_signif with binary point at the left .1xxxxxxxx */
mul64(&arg_signif, &arg_signif, (long long *) (&argSq.sigl));
mul64((long long *) (&argSq.sigl), (long long *) (&argSq.sigl), &argSqSq);
/* will be a valid positive nr with expon = 0 */
*(short *) &(pos_poly.sign) = 0;
pos_poly.exp = EXP_BIAS;
/* Do the basic fixed point polynomial evaluation */
polynomial(&pos_poly.sigl, (unsigned *) &argSqSq,
(unsigned short (*)[4]) oddplterms, HIPOWERop - 1);
mul64((long long *) (&argSq.sigl), (long long *) (&pos_poly.sigl),
(long long *) (&pos_poly.sigl));
/* will be a valid positive nr with expon = 0 */
*(short *) &(neg_poly.sign) = 0;
neg_poly.exp = EXP_BIAS;
/* Do the basic fixed point polynomial evaluation */
polynomial(&neg_poly.sigl, (unsigned *) &argSqSq,
(unsigned short (*)[4]) oddnegterms, HIPOWERon - 1);
/* Subtract the mantissas */
*((long long *) (&pos_poly.sigl)) -= *((long long *) (&neg_poly.sigl));
reg_move(&pos_poly, &odd_poly);
poly_add_1(&odd_poly);
/* The complete odd polynomial */
reg_u_mul(&odd_poly, arg, &odd_poly, FULL_PRECISION);
/* will be a valid positive nr with expon = 0 */
*(short *) &(even_poly.sign) = 0;
mul64((long long *) (&argSq.sigl),
(long long *) (&denomterm), (long long *) (&even_poly.sigl));
poly_add_1(&even_poly);
reg_div(&odd_poly, &even_poly, arg, FULL_PRECISION);
if (recursions)
reg_sub(&CONST_PI4, arg, arg, FULL_PRECISION);
}
/* Subtract b from a. (a-b) -> dest */
void reg_sub(FPU_REG *a, FPU_REG *b, FPU_REG *dest, int control_w)
{
int diff;
if ( !(a->tag | b->tag) )
{
/* Both registers are valid */
diff = a->exp - b->exp;
if (!diff)
{
diff = a->sigh - b->sigh; /* Works only if ms bits are identical */
if (!diff)
{
diff = a->sigl > b->sigl;
if (!diff)
diff = -(a->sigl < b->sigl);
}
}
switch (a->sign*2 + b->sign)
{
case 0: /* P - P */
case 3: /* N - N */
if (diff > 0)
{
reg_u_sub(a, b, dest, control_w);
dest->sign = a->sign;
}
else if ( diff == 0 )
{
reg_move(&CONST_Z, dest);
/* sign depends upon rounding mode */
dest->sign = ((control_w & CW_RC) != RC_DOWN)
? SIGN_POS : SIGN_NEG;
}
else
{
reg_u_sub(b, a, dest, control_w);
dest->sign = a->sign ^ SIGN_POS^SIGN_NEG;
}
return;
case 1: /* P - N */
reg_u_add(a, b, dest, control_w);
dest->sign = SIGN_POS;
return;
case 2: /* N - P */
reg_u_add(a, b, dest, control_w);
dest->sign = SIGN_NEG;
return;
}
}
else
{
if ( (a->tag == TW_NaN) || (b->tag == TW_NaN) )
{
real_2op_NaN(a, b, dest);
return;
}
else if (b->tag == TW_Zero)
{
if (a->tag == TW_Zero)
{
char same_signs = !(a->sign ^ b->sign);
/* Both are zero, result will be zero. */
reg_move(a, dest); /* Answer for different signs. */
if (same_signs)
{
/* Sign depends upon rounding mode */
dest->sign = ((control_w & CW_RC) != RC_DOWN)
? SIGN_POS : SIGN_NEG;
}
}
else
reg_move(a, dest);
return;
}
else if (a->tag == TW_Zero)
{
reg_move(b, dest);
dest->sign ^= SIGN_POS^SIGN_NEG;
return;
}
else if (a->tag == TW_Infinity)
{
if (b->tag != TW_Infinity)
{
reg_move(a, dest);
return;
}
if (a->sign == b->sign)
{
reg_move(&CONST_QNaN, dest);
return;
}
reg_move(a, dest);
return;
}
else if (b->tag == TW_Infinity)
{
reg_move(b, dest);
dest->sign ^= SIGN_POS^SIGN_NEG;
return;
}
}
#ifdef PARANOID
EXCEPTION(EX_INTERNAL|0x110);
#endif
}
void load_store_instr(char type)
{
FPU_REG *pop_ptr; /* We need a version of FPU_st0_ptr which won't change. */
pop_ptr = NULL; /* Initialized just to stop compiler warnings. */
switch ( type_table[(int) (unsigned) type] )
{
case _NONE_:
break;
case _REG0_:
pop_ptr = &st(0); /* Some of these instructions pop after
storing */
FPU_st0_ptr = pop_ptr; /* Set the global variables. */
FPU_st0_tag = FPU_st0_ptr->tag;
break;
case _PUSH_:
{
pop_ptr = &st(-1);
if ( pop_ptr->tag != TW_Empty )
{ stack_overflow(); return; }
top--;
}
break;
case _null_:
return Un_impl();
#ifdef PARANOID
default:
return EXCEPTION(EX_INTERNAL);
#endif PARANOID
}
switch ( type )
{
case 000: /* fld m32real */
reg_load_single();
reg_move(&FPU_loaded_data, pop_ptr);
break;
case 001: /* fild m32int */
reg_load_int32();
reg_move(&FPU_loaded_data, pop_ptr);
break;
case 002: /* fld m64real */
reg_load_double();
reg_move(&FPU_loaded_data, pop_ptr);
break;
case 003: /* fild m16int */
reg_load_int16();
reg_move(&FPU_loaded_data, pop_ptr);
break;
case 010: /* fst m32real */
reg_store_single();
break;
case 011: /* fist m32int */
reg_store_int32();
break;
case 012: /* fst m64real */
reg_store_double();
break;
case 013: /* fist m16int */
reg_store_int16();
break;
case 014: /* fstp m32real */
if ( reg_store_single() )
pop_0(); /* pop only if the number was actually stored
(see the 80486 manual p16-28) */
break;
case 015: /* fistp m32int */
if ( reg_store_int32() )
pop_0(); /* pop only if the number was actually stored
(see the 80486 manual p16-28) */
break;
case 016: /* fstp m64real */
if ( reg_store_double() )
pop_0(); /* pop only if the number was actually stored
(see the 80486 manual p16-28) */
break;
case 017: /* fistp m16int */
if ( reg_store_int16() )
pop_0(); /* pop only if the number was actually stored
(see the 80486 manual p16-28) */
break;
case 020: /* fldenv m14/28byte */
fldenv();
break;
case 022: /* frstor m94/108byte */
frstor();
break;
case 023: /* fbld m80dec */
reg_load_bcd();
reg_move(&FPU_loaded_data, pop_ptr);
break;
case 024: /* fldcw */
RE_ENTRANT_CHECK_OFF
control_word = get_fs_word((unsigned short *) FPU_data_address);
RE_ENTRANT_CHECK_ON
#ifdef NO_UNDERFLOW_TRAP
if ( !(control_word & EX_Underflow) )
{
control_word |= EX_Underflow;
}
#endif
FPU_data_address = (void *)data_operand_offset; /* We want no net effect */
FPU_entry_eip = ip_offset; /* We want no net effect */
break;
case 025: /* fld m80real */
reg_load_extended();
reg_move(&FPU_loaded_data, pop_ptr);
break;
case 027: /* fild m64int */
reg_load_int64();
reg_move(&FPU_loaded_data, pop_ptr);
break;
case 030: /* fstenv m14/28byte */
fstenv();
FPU_data_address = (void *)data_operand_offset; /* We want no net effect */
FPU_entry_eip = ip_offset; /* We want no net effect */
break;
case 032: /* fsave */
fsave();
FPU_data_address = (void *)data_operand_offset; /* We want no net effect */
FPU_entry_eip = ip_offset; /* We want no net effect */
break;
case 033: /* fbstp m80dec */
if ( reg_store_bcd() )
pop_0(); /* pop only if the number was actually stored
(see the 80486 manual p16-28) */
break;
case 034: /* fstcw m16int */
RE_ENTRANT_CHECK_OFF
verify_area(VERIFY_WRITE,FPU_data_address,2);
put_fs_word(control_word, (short *) FPU_data_address);
RE_ENTRANT_CHECK_ON
FPU_data_address = (void *)data_operand_offset; /* We want no net effect */
FPU_entry_eip = ip_offset; /* We want no net effect */
break;
case 035: /* fstp m80real */
if ( reg_store_extended() )
pop_0(); /* pop only if the number was actually stored
(see the 80486 manual p16-28) */
break;
case 036: /* fstsw m2byte */
status_word &= ~SW_TOP;
status_word |= (top&7) << SW_TOPS;
RE_ENTRANT_CHECK_OFF
verify_area(VERIFY_WRITE,FPU_data_address,2);
put_fs_word(status_word,(short *) FPU_data_address);
RE_ENTRANT_CHECK_ON
FPU_data_address = (void *)data_operand_offset; /* We want no net effect */
FPU_entry_eip = ip_offset; /* We want no net effect */
break;
case 037: /* fistp m64int */
if ( reg_store_int64() )
pop_0(); /* pop only if the number was actually stored
(see the 80486 manual p16-28) */
break;
}
}
void reg_add(FPU_REG *a, FPU_REG *b, FPU_REG *dest, int control_w)
{
int diff;
if ( !(a->tag | b->tag) )
{
/* Both registers are valid */
if (!(a->sign ^ b->sign))
{
/* signs are the same */
reg_u_add(a, b, dest, control_w);
dest->sign = a->sign;
return;
}
/* The signs are different, so do a subtraction */
diff = a->exp - b->exp;
if (!diff)
{
diff = a->sigh - b->sigh; /* Works only if ms bits are identical */
if (!diff)
{
diff = a->sigl > b->sigl;
if (!diff)
diff = -(a->sigl < b->sigl);
}
}
if (diff > 0)
{
reg_u_sub(a, b, dest, control_w);
dest->sign = a->sign;
}
else if ( diff == 0 )
{
reg_move(&CONST_Z, dest);
/* sign depends upon rounding mode */
dest->sign = ((control_w & CW_RC) != RC_DOWN)
? SIGN_POS : SIGN_NEG;
}
else
{
reg_u_sub(b, a, dest, control_w);
dest->sign = b->sign;
}
return;
}
else
{
if ( (a->tag == TW_NaN) || (b->tag == TW_NaN) )
{
real_2op_NaN(a, b, dest);
return;
}
else if (a->tag == TW_Zero)
{
if (b->tag == TW_Zero)
{
char different_signs = a->sign ^ b->sign;
/* Both are zero, result will be zero. */
reg_move(a, dest);
if (different_signs)
{
/* Signs are different. */
/* Sign of answer depends upon rounding mode. */
dest->sign = ((control_w & CW_RC) != RC_DOWN)
? SIGN_POS : SIGN_NEG;
}
}
else
reg_move(b, dest);
return;
}
else if (b->tag == TW_Zero)
{
reg_move(a, dest);
return;
}
else if (a->tag == TW_Infinity)
{
if (b->tag != TW_Infinity)
{
reg_move(a, dest);
return;
}
/* They are both + or - infinity */
if (a->sign == b->sign)
{
reg_move(a, dest);
return;
}
reg_move(&CONST_QNaN, dest); /* inf - inf is undefined. */
return;
}
else if (b->tag == TW_Infinity)
{
reg_move(b, dest);
return;
}
}
#ifdef PARANOID
EXCEPTION(EX_INTERNAL|0x101);
#endif
}
void
fst_i_()
{
/* fst st(i) */
reg_move(FPU_st0_ptr, &st(FPU_rm));
}
/*--- poly_sine() -----------------------------------------------------------+
| |
+---------------------------------------------------------------------------*/
void poly_sine(FPU_REG const *arg, FPU_REG *result)
{
int exponent, echange;
Xsig accumulator, argSqrd, argTo4;
unsigned long fix_up, adj;
unsigned long long fixed_arg;
#ifdef PARANOID
if ( arg->tag == TW_Zero )
{
/* Return 0.0 */
reg_move(&CONST_Z, result);
return;
}
#endif PARANOID
exponent = arg->exp - EXP_BIAS;
accumulator.lsw = accumulator.midw = accumulator.msw = 0;
/* Split into two ranges, for arguments below and above 1.0 */
/* The boundary between upper and lower is approx 0.88309101259 */
if ( (exponent < -1) || ((exponent == -1) && (arg->sigh <= 0xe21240aa)) )
{
/* The argument is <= 0.88309101259 */
argSqrd.msw = arg->sigh; argSqrd.midw = arg->sigl; argSqrd.lsw = 0;
mul64_Xsig(&argSqrd, &significand(arg));
shr_Xsig(&argSqrd, 2*(-1-exponent));
argTo4.msw = argSqrd.msw; argTo4.midw = argSqrd.midw;
argTo4.lsw = argSqrd.lsw;
mul_Xsig_Xsig(&argTo4, &argTo4);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), neg_terms_l,
N_COEFF_N-1);
mul_Xsig_Xsig(&accumulator, &argSqrd);
negate_Xsig(&accumulator);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), pos_terms_l,
N_COEFF_P-1);
shr_Xsig(&accumulator, 2); /* Divide by four */
accumulator.msw |= 0x80000000; /* Add 1.0 */
mul64_Xsig(&accumulator, &significand(arg));
mul64_Xsig(&accumulator, &significand(arg));
mul64_Xsig(&accumulator, &significand(arg));
/* Divide by four, FPU_REG compatible, etc */
exponent = 3*exponent + EXP_BIAS;
/* The minimum exponent difference is 3 */
shr_Xsig(&accumulator, arg->exp - exponent);
negate_Xsig(&accumulator);
XSIG_LL(accumulator) += significand(arg);
echange = round_Xsig(&accumulator);
result->exp = arg->exp + echange;
}
else
{
/* The argument is > 0.88309101259 */
/* We use sin(arg) = cos(pi/2-arg) */
fixed_arg = significand(arg);
if ( exponent == 0 )
{
/* The argument is >= 1.0 */
/* Put the binary point at the left. */
fixed_arg <<= 1;
}
/* pi/2 in hex is: 1.921fb54442d18469 898CC51701B839A2 52049C1 */
fixed_arg = 0x921fb54442d18469LL - fixed_arg;
XSIG_LL(argSqrd) = fixed_arg; argSqrd.lsw = 0;
mul64_Xsig(&argSqrd, &fixed_arg);
XSIG_LL(argTo4) = XSIG_LL(argSqrd); argTo4.lsw = argSqrd.lsw;
mul_Xsig_Xsig(&argTo4, &argTo4);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), neg_terms_h,
N_COEFF_NH-1);
mul_Xsig_Xsig(&accumulator, &argSqrd);
negate_Xsig(&accumulator);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), pos_terms_h,
N_COEFF_PH-1);
negate_Xsig(&accumulator);
mul64_Xsig(&accumulator, &fixed_arg);
mul64_Xsig(&accumulator, &fixed_arg);
shr_Xsig(&accumulator, 3);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &argSqrd);
shr_Xsig(&accumulator, 1);
accumulator.lsw |= 1; /* A zero accumulator here would cause problems */
negate_Xsig(&accumulator);
/* The basic computation is complete. Now fix the answer to
compensate for the error due to the approximation used for
pi/2
*/
/* This has an exponent of -65 */
fix_up = 0x898cc517;
/* The fix-up needs to be improved for larger args */
if ( argSqrd.msw & 0xffc00000 )
{
/* Get about 32 bit precision in these: */
mul_32_32(0x898cc517, argSqrd.msw, &adj);
fix_up -= adj/6;
}
mul_32_32(fix_up, LL_MSW(fixed_arg), &fix_up);
adj = accumulator.lsw; /* temp save */
accumulator.lsw -= fix_up;
if ( accumulator.lsw > adj )
XSIG_LL(accumulator) --;
echange = round_Xsig(&accumulator);
result->exp = EXP_BIAS - 1 + echange;
}
significand(result) = XSIG_LL(accumulator);
result->tag = TW_Valid;
result->sign = arg->sign;
#ifdef PARANOID
if ( (result->exp >= EXP_BIAS)
&& (significand(result) > 0x8000000000000000LL) )
{
EXCEPTION(EX_INTERNAL|0x150);
}
#endif PARANOID
}
/*--- poly_sine() -----------------------------------------------------------+
| |
+---------------------------------------------------------------------------*/
void poly_sine(FPU_REG const *arg, FPU_REG *result)
{
short exponent;
FPU_REG fixed_arg, arg_sqrd, arg_to_4, accum, negaccum;
exponent = arg->exp - EXP_BIAS;
if ( arg->tag == TW_Zero )
{
/* Return 0.0 */
reg_move(&CONST_Z, result);
return;
}
#ifdef PARANOID
if ( arg->sign != 0 ) /* Can't hack a number < 0.0 */
{
EXCEPTION(EX_Invalid);
reg_move(&CONST_QNaN, result);
return;
}
if ( exponent >= 0 ) /* Can't hack a number > 1.0 */
{
if ( (exponent == 0) && (arg->sigl == 0) && (arg->sigh == 0x80000000) )
{
reg_move(&CONST_1, result);
return;
}
EXCEPTION(EX_Invalid);
reg_move(&CONST_QNaN, result);
return;
}
#endif PARANOID
fixed_arg.sigl = arg->sigl;
fixed_arg.sigh = arg->sigh;
if ( exponent < -1 )
{
/* shift the argument right by the required places */
if ( shrx(&(fixed_arg.sigl), -1-exponent) >= 0x80000000U )
significand(&fixed_arg)++; /* round up */
}
mul64(&significand(&fixed_arg), &significand(&fixed_arg),
&significand(&arg_sqrd));
mul64(&significand(&arg_sqrd), &significand(&arg_sqrd),
&significand(&arg_to_4));
/* will be a valid positive nr with expon = 0 */
*(short *)&(accum.sign) = 0;
accum.exp = 0;
/* Do the basic fixed point polynomial evaluation */
polynomial(&(accum.sigl), &(arg_to_4.sigl), lterms, HIPOWER-1);
/* will be a valid positive nr with expon = 0 */
*(short *)&(negaccum.sign) = 0;
negaccum.exp = 0;
/* Do the basic fixed point polynomial evaluation */
polynomial(&(negaccum.sigl), &(arg_to_4.sigl), negterms, HIPOWER-1);
mul64(&significand(&arg_sqrd), &significand(&negaccum),
&significand(&negaccum));
/* Subtract the mantissas */
significand(&accum) -= significand(&negaccum);
/* Convert to 64 bit signed-compatible */
accum.exp = EXP_BIAS - 1 + accum.exp;
reg_move(&accum, result);
normalize(result);
reg_mul(result, arg, result, FULL_PRECISION);
reg_u_add(result, arg, result, FULL_PRECISION);
if ( result->exp >= EXP_BIAS )
{
/* A small overflow may be possible... but an illegal result. */
if ( (result->exp > EXP_BIAS) /* Larger or equal 2.0 */
|| (result->sigl > 1) /* Larger than 1.0+msb */
|| (result->sigh != 0x80000000) /* Much > 1.0 */
)
{
#ifdef DEBUGGING
RE_ENTRANT_CHECK_OFF;
printk("\nEXP=%d, MS=%08x, LS=%08x\n", result->exp,
result->sigh, result->sigl);
RE_ENTRANT_CHECK_ON;
#endif DEBUGGING
EXCEPTION(EX_INTERNAL|0x103);
}
#ifdef DEBUGGING
RE_ENTRANT_CHECK_OFF;
printk("\n***CORRECTING ILLEGAL RESULT*** in poly_sin() computation\n");
printk("EXP=%d, MS=%08x, LS=%08x\n", result->exp,
result->sigh, result->sigl);
RE_ENTRANT_CHECK_ON;
#endif DEBUGGING
result->sigl = 0; /* Truncate the result to 1.00 */
}
}
static void
fxtract(void)
{
FPU_REG *st_new_ptr;
register FPU_REG *st1_ptr = FPU_st0_ptr; /* anticipate */
if (STACK_OVERFLOW) {
stack_overflow();
return;
}
if (!(FPU_st0_tag ^ TW_Valid)) {
long e;
#ifdef DENORM_OPERAND
if ((FPU_st0_ptr->exp <= EXP_UNDER) && (denormal_operand()))
return;
#endif /* DENORM_OPERAND */
push();
reg_move(st1_ptr, FPU_st0_ptr);
FPU_st0_ptr->exp = EXP_BIAS;
e = st1_ptr->exp - EXP_BIAS;
convert_l2reg(&e, st1_ptr);
return;
} else if (FPU_st0_tag == TW_Zero) {
char sign = FPU_st0_ptr->sign;
divide_by_zero(SIGN_NEG, FPU_st0_ptr);
push();
reg_move(&CONST_Z, FPU_st0_ptr);
FPU_st0_ptr->sign = sign;
return;
} else if (FPU_st0_tag == TW_Infinity) {
char sign = FPU_st0_ptr->sign;
FPU_st0_ptr->sign = SIGN_POS;
push();
reg_move(&CONST_INF, FPU_st0_ptr);
FPU_st0_ptr->sign = sign;
return;
} else if (FPU_st0_tag == TW_NaN) {
if (!(FPU_st0_ptr->sigh & 0x40000000)) { /* Signaling ? */
EXCEPTION(EX_Invalid);
/* Convert to a QNaN */
FPU_st0_ptr->sigh |= 0x40000000;
}
push();
reg_move(st1_ptr, FPU_st0_ptr);
return;
} else if (FPU_st0_tag == TW_Empty) {
/* Is this the correct
* behaviour? */
if (control_word & EX_Invalid) {
stack_underflow();
push();
stack_underflow();
} else
EXCEPTION(EX_StackUnder);
}
#ifdef PARANOID
else
EXCEPTION(EX_INTERNAL | 0x119);
#endif /* PARANOID */
}
/*--- poly_tan() ------------------------------------------------------------+
| |
+---------------------------------------------------------------------------*/
void
poly_tan(FPU_REG * arg, FPU_REG * y_reg)
{
char invert = 0;
short exponent;
FPU_REG odd_poly, even_poly, pos_poly, neg_poly;
FPU_REG argSq;
long long arg_signif, argSqSq;
exponent = arg->exp - EXP_BIAS;
if (arg->tag == TW_Zero) {
/* Return 0.0 */
reg_move(&CONST_Z, y_reg);
return;
}
if (exponent >= -1) {
/* argument is in the range [0.5 .. 1.0] */
if (exponent >= 0) {
#ifdef PARANOID
if ((exponent == 0) &&
(arg->sigl == 0) && (arg->sigh == 0x80000000))
#endif /* PARANOID */
{
arith_overflow(y_reg);
return;
}
#ifdef PARANOID
EXCEPTION(EX_INTERNAL | 0x104); /* There must be a logic
* error */
return;
#endif /* PARANOID */
}
/* The argument is in the range [0.5 .. 1.0) */
/* Convert the argument to a number in the range (0.0 .. 0.5] */
*((long long *) (&arg->sigl)) = -*((long long *) (&arg->sigl));
normalize(arg); /* Needed later */
exponent = arg->exp - EXP_BIAS;
invert = 1;
}
#ifdef PARANOID
if (arg->sign != 0) { /* Can't hack a number < 0.0 */
arith_invalid(y_reg);
return;
} /* Need a positive number */
#endif /* PARANOID */
*(long long *) &arg_signif = *(long long *) &(arg->sigl);
if (exponent < -1) {
/* shift the argument right by the required places */
if (shrx(&arg_signif, -1 - exponent) >= (unsigned)0x80000000)
arg_signif++; /* round up */
}
mul64(&arg_signif, &arg_signif, (long long *) (&argSq.sigl));
mul64((long long *) (&argSq.sigl), (long long *) (&argSq.sigl), &argSqSq);
/* will be a valid positive nr with expon = 0 */
*(short *) &(pos_poly.sign) = 0;
pos_poly.exp = EXP_BIAS;
/* Do the basic fixed point polynomial evaluation */
polynomial((u_int *) &pos_poly.sigl, (unsigned *) &argSqSq, oddplterms, HIPOWERop - 1);
/* will be a valid positive nr with expon = 0 */
*(short *) &(neg_poly.sign) = 0;
neg_poly.exp = EXP_BIAS;
/* Do the basic fixed point polynomial evaluation */
polynomial((u_int *) &neg_poly.sigl, (unsigned *) &argSqSq, oddnegterms, HIPOWERon - 1);
mul64((long long *) (&argSq.sigl), (long long *) (&neg_poly.sigl),
(long long *) (&neg_poly.sigl));
/* Subtract the mantissas */
*((long long *) (&pos_poly.sigl)) -= *((long long *) (&neg_poly.sigl));
/* Convert to 64 bit signed-compatible */
pos_poly.exp -= 1;
reg_move(&pos_poly, &odd_poly);
normalize(&odd_poly);
reg_mul(&odd_poly, arg, &odd_poly, FULL_PRECISION);
reg_u_add(&odd_poly, arg, &odd_poly, FULL_PRECISION); /* This is just the odd
* polynomial */
/* will be a valid positive nr with expon = 0 */
*(short *) &(pos_poly.sign) = 0;
pos_poly.exp = EXP_BIAS;
/* Do the basic fixed point polynomial evaluation */
polynomial((u_int *) &pos_poly.sigl, (unsigned *) &argSqSq, evenplterms, HIPOWERep - 1);
mul64((long long *) (&argSq.sigl),
(long long *) (&pos_poly.sigl), (long long *) (&pos_poly.sigl));
/* will be a valid positive nr with expon = 0 */
*(short *) &(neg_poly.sign) = 0;
neg_poly.exp = EXP_BIAS;
/* Do the basic fixed point polynomial evaluation */
polynomial((u_int *) &neg_poly.sigl, (unsigned *) &argSqSq, evennegterms, HIPOWERen - 1);
/* Subtract the mantissas */
*((long long *) (&neg_poly.sigl)) -= *((long long *) (&pos_poly.sigl));
/* and multiply by argSq */
/* Convert argSq to a valid reg number */
*(short *) &(argSq.sign) = 0;
argSq.exp = EXP_BIAS - 1;
normalize(&argSq);
/* Convert to 64 bit signed-compatible */
neg_poly.exp -= 1;
reg_move(&neg_poly, &even_poly);
normalize(&even_poly);
reg_mul(&even_poly, &argSq, &even_poly, FULL_PRECISION);
reg_add(&even_poly, &argSq, &even_poly, FULL_PRECISION);
reg_sub(&CONST_1, &even_poly, &even_poly, FULL_PRECISION); /* This is just the even
* polynomial */
/* Now ready to copy the results */
if (invert) {
reg_div(&even_poly, &odd_poly, y_reg, FULL_PRECISION);
} else {
reg_div(&odd_poly, &even_poly, y_reg, FULL_PRECISION);
}
}
/*--- poly_cos() ------------------------------------------------------------+
| |
+---------------------------------------------------------------------------*/
void poly_cos(FPU_REG const *arg, FPU_REG *result)
{
long int exponent, exp2, echange;
Xsig accumulator, argSqrd, fix_up, argTo4;
unsigned long adj;
unsigned long long fixed_arg;
#ifdef PARANOID
if ( arg->tag == TW_Zero )
{
/* Return 1.0 */
reg_move(&CONST_1, result);
return;
}
if ( (arg->exp > EXP_BIAS)
|| ((arg->exp == EXP_BIAS)
&& (significand(arg) > 0xc90fdaa22168c234LL)) )
{
EXCEPTION(EX_Invalid);
reg_move(&CONST_QNaN, result);
return;
}
#endif PARANOID
exponent = arg->exp - EXP_BIAS;
accumulator.lsw = accumulator.midw = accumulator.msw = 0;
if ( (exponent < -1) || ((exponent == -1) && (arg->sigh <= 0xb00d6f54)) )
{
/* arg is < 0.687705 */
argSqrd.msw = arg->sigh; argSqrd.midw = arg->sigl; argSqrd.lsw = 0;
mul64_Xsig(&argSqrd, &significand(arg));
if ( exponent < -1 )
{
/* shift the argument right by the required places */
shr_Xsig(&argSqrd, 2*(-1-exponent));
}
argTo4.msw = argSqrd.msw; argTo4.midw = argSqrd.midw;
argTo4.lsw = argSqrd.lsw;
mul_Xsig_Xsig(&argTo4, &argTo4);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), neg_terms_h,
N_COEFF_NH-1);
mul_Xsig_Xsig(&accumulator, &argSqrd);
negate_Xsig(&accumulator);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), pos_terms_h,
N_COEFF_PH-1);
negate_Xsig(&accumulator);
mul64_Xsig(&accumulator, &significand(arg));
mul64_Xsig(&accumulator, &significand(arg));
shr_Xsig(&accumulator, -2*(1+exponent));
shr_Xsig(&accumulator, 3);
negate_Xsig(&accumulator);
add_Xsig_Xsig(&accumulator, &argSqrd);
shr_Xsig(&accumulator, 1);
/* It doesn't matter if accumulator is all zero here, the
following code will work ok */
negate_Xsig(&accumulator);
if ( accumulator.lsw & 0x80000000 )
XSIG_LL(accumulator) ++;
if ( accumulator.msw == 0 )
{
/* The result is 1.0 */
reg_move(&CONST_1, result);
}
else
{
significand(result) = XSIG_LL(accumulator);
/* will be a valid positive nr with expon = -1 */
*(short *)&(result->sign) = 0;
result->exp = EXP_BIAS - 1;
}
}
else
{
fixed_arg = significand(arg);
if ( exponent == 0 )
{
/* The argument is >= 1.0 */
/* Put the binary point at the left. */
fixed_arg <<= 1;
}
/* pi/2 in hex is: 1.921fb54442d18469 898CC51701B839A2 52049C1 */
fixed_arg = 0x921fb54442d18469LL - fixed_arg;
exponent = -1;
exp2 = -1;
/* A shift is needed here only for a narrow range of arguments,
i.e. for fixed_arg approx 2^-32, but we pick up more... */
if ( !(LL_MSW(fixed_arg) & 0xffff0000) )
{
fixed_arg <<= 16;
exponent -= 16;
exp2 -= 16;
}
XSIG_LL(argSqrd) = fixed_arg; argSqrd.lsw = 0;
mul64_Xsig(&argSqrd, &fixed_arg);
if ( exponent < -1 )
{
/* shift the argument right by the required places */
shr_Xsig(&argSqrd, 2*(-1-exponent));
}
argTo4.msw = argSqrd.msw; argTo4.midw = argSqrd.midw;
argTo4.lsw = argSqrd.lsw;
mul_Xsig_Xsig(&argTo4, &argTo4);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), neg_terms_l,
N_COEFF_N-1);
mul_Xsig_Xsig(&accumulator, &argSqrd);
negate_Xsig(&accumulator);
polynomial_Xsig(&accumulator, &XSIG_LL(argTo4), pos_terms_l,
N_COEFF_P-1);
shr_Xsig(&accumulator, 2); /* Divide by four */
accumulator.msw |= 0x80000000; /* Add 1.0 */
mul64_Xsig(&accumulator, &fixed_arg);
mul64_Xsig(&accumulator, &fixed_arg);
mul64_Xsig(&accumulator, &fixed_arg);
/* Divide by four, FPU_REG compatible, etc */
exponent = 3*exponent;
/* The minimum exponent difference is 3 */
shr_Xsig(&accumulator, exp2 - exponent);
negate_Xsig(&accumulator);
XSIG_LL(accumulator) += fixed_arg;
/* The basic computation is complete. Now fix the answer to
compensate for the error due to the approximation used for
pi/2
*/
/* This has an exponent of -65 */
XSIG_LL(fix_up) = 0x898cc51701b839a2ll;
fix_up.lsw = 0;
/* The fix-up needs to be improved for larger args */
if ( argSqrd.msw & 0xffc00000 )
{
/* Get about 32 bit precision in these: */
mul_32_32(0x898cc517, argSqrd.msw, &adj);
fix_up.msw -= adj/2;
mul_32_32(0x898cc517, argTo4.msw, &adj);
fix_up.msw += adj/24;
}
exp2 += norm_Xsig(&accumulator);
shr_Xsig(&accumulator, 1); /* Prevent overflow */
exp2++;
shr_Xsig(&fix_up, 65 + exp2);
add_Xsig_Xsig(&accumulator, &fix_up);
echange = round_Xsig(&accumulator);
result->exp = exp2 + EXP_BIAS + echange;
*(short *)&(result->sign) = 0; /* Is a valid positive nr */
significand(result) = XSIG_LL(accumulator);
}
#ifdef PARANOID
if ( (result->exp >= EXP_BIAS)
&& (significand(result) > 0x8000000000000000LL) )
{
EXCEPTION(EX_INTERNAL|0x151);
}
#endif PARANOID
}
