gambas-source-code/main/gbx/gbx_subr_math.c
François Gallo 146f0c41f7 [INTERPRETER]
* BUG: the exp10() function works on FreeBSD now. 



git-svn-id: svn://localhost/gambas/trunk@4556 867c0c6c-44f3-4631-809d-bfa615b0a4ec
2012-03-16 15:47:19 +00:00

734 lines
15 KiB
C

/***************************************************************************
gbx_subr_math.c
(c) 2000-2012 Benoît Minisini <gambas@users.sourceforge.net>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
MA 02110-1301, USA.
***************************************************************************/
#define __GBX_SUBR_MATH_C
#ifdef OS_SOLARIS
/* Make math.h define M_PI and a few other things */
#define __EXTENSIONS__
/* Get definition for finite() */
#include <ieeefp.h>
#endif
#include "gb_common.h"
#include "gbx_value.h"
#include "gbx_subr.h"
#include "gbx_math.h"
#define ABS(x) ((x) < 0 ? (-x) : (x))
#define SMT_NAME SUBR_quo
#define SMT_TYPE 3
#define SMT_OP /
#include "gbx_subr_math_temp.h"
#define SMT_NAME SUBR_rem
#define SMT_TYPE 3
#define SMT_OP %
#include "gbx_subr_math_temp.h"
void SUBR_pi(ushort code)
{
SUBR_ENTER();
if (NPARAM == 0)
{
SP->type = T_FLOAT;
SP->_float.value = M_PI;
SP++;
}
else
{
VALUE_conv_float(PARAM);
PARAM->_float.value = M_PI * PARAM->_float.value;
}
}
void SUBR_randomize(ushort code)
{
SUBR_ENTER();
if (NPARAM == 0)
randomize(FALSE, 0);
else
randomize(TRUE, (uint)SUBR_get_integer(PARAM));
RETURN->type = T_VOID;
SUBR_LEAVE();
}
void SUBR_rnd(ushort code)
{
double min = 0.0, max = 1.0;
SUBR_ENTER();
if (NPARAM >= 1)
{
VALUE_conv_float(&PARAM[0]);
max = PARAM->_float.value;
}
if (NPARAM == 2)
{
min = max;
VALUE_conv_float(&PARAM[1]);
max = PARAM[1]._float.value;
}
RETURN->type = T_FLOAT;
RETURN->_float.value = (rnd() * (max - min)) + min;
SUBR_LEAVE();
}
void SUBR_round(ushort code)
{
int val = 0;
double power;
SUBR_ENTER();
if (NPARAM == 2)
val = SUBR_get_integer(&PARAM[1]);
power = pow(10, val);
VALUE_conv_float(&PARAM[0]);
RETURN->type = T_FLOAT;
/*RETURN->_float.value = rint(PARAM->_float.value / power) * power;*/
RETURN->_float.value = floor(PARAM->_float.value / power + 0.5) * power;
SUBR_LEAVE();
}
void SUBR_math(ushort code)
{
static void *jump[] = {
NULL, &&__FRAC, &&__LOG, &&__EXP, &&__SQRT, &&__SIN, &&__COS, &&__TAN, &&__ATAN, &&__ASIN, &&__ACOS,
&&__DEG, &&__RAD, &&__LOG10, &&__SINH, &&__COSH, &&__TANH, &&__ASINH, &&__ACOSH, &&__ATANH,
&&__EXP2, &&__EXP10, &&__LOG2, &&__CBRT, &&__EXPM1, &&__LOG1P, &&__FLOOR, &&__CEIL
};
SUBR_ENTER_PARAM(1);
VALUE_conv_float(PARAM);
goto *jump[code & 0x1F];
__FRAC: PARAM->_float.value = frac(PARAM->_float.value); goto __END;
__LOG: PARAM->_float.value = __builtin_log(PARAM->_float.value); goto __END;
__EXP: PARAM->_float.value = __builtin_exp(PARAM->_float.value); goto __END;
__SQRT: PARAM->_float.value = __builtin_sqrt(PARAM->_float.value); goto __END;
__SIN: PARAM->_float.value = __builtin_sin(PARAM->_float.value); goto __END;
__COS: PARAM->_float.value = __builtin_cos(PARAM->_float.value); goto __END;
__TAN: PARAM->_float.value = __builtin_tan(PARAM->_float.value); goto __END;
__ATAN: PARAM->_float.value = __builtin_atan(PARAM->_float.value); goto __END;
__ASIN: PARAM->_float.value = __builtin_asin(PARAM->_float.value); goto __END;
__ACOS: PARAM->_float.value = __builtin_acos(PARAM->_float.value); goto __END;
__DEG: PARAM->_float.value = deg(PARAM->_float.value); goto __END;
__RAD: PARAM->_float.value = rad(PARAM->_float.value); goto __END;
__LOG10: PARAM->_float.value = log10(PARAM->_float.value); goto __END;
__SINH: PARAM->_float.value = __builtin_sinh(PARAM->_float.value); goto __END;
__COSH: PARAM->_float.value = __builtin_cosh(PARAM->_float.value); goto __END;
__TANH: PARAM->_float.value = __builtin_tanh(PARAM->_float.value); goto __END;
__ASINH: PARAM->_float.value = __builtin_asinh(PARAM->_float.value); goto __END;
__ACOSH: PARAM->_float.value = __builtin_acosh(PARAM->_float.value); goto __END;
__ATANH: PARAM->_float.value = __builtin_atanh(PARAM->_float.value); goto __END;
__EXP2: PARAM->_float.value = __builtin_exp2(PARAM->_float.value); goto __END;
#ifdef OS_FREEBSD
__EXP10: PARAM->_float.value = exp10(PARAM->_float.value); goto __END;
#else
__EXP10: PARAM->_float.value = __builtin_exp10(PARAM->_float.value); goto __END;
#endif
__LOG2: PARAM->_float.value = __builtin_log2(PARAM->_float.value); goto __END;
__CBRT: PARAM->_float.value = __builtin_cbrt(PARAM->_float.value); goto __END;
__EXPM1: PARAM->_float.value = __builtin_expm1(PARAM->_float.value); goto __END;
__LOG1P: PARAM->_float.value = __builtin_log1p(PARAM->_float.value); goto __END;
__FLOOR: PARAM->_float.value = __builtin_floor(PARAM->_float.value); goto __END;
__CEIL: PARAM->_float.value = __builtin_ceil(PARAM->_float.value); goto __END;
__END:
if (!isfinite(PARAM->_float.value))
THROW(E_MATH);
}
void SUBR_math2(ushort code)
{
static void *jump[] = { NULL, &&__ATAN2, &&__ANG, &&__HYPOT };
SUBR_ENTER_PARAM(2);
VALUE_conv_float(&PARAM[0]);
VALUE_conv_float(&PARAM[1]);
goto *jump[code & 0x1F];
__ATAN2: PARAM->_float.value = __builtin_atan2(PARAM[0]._float.value, PARAM[1]._float.value); goto __END;
__ANG: PARAM->_float.value = __builtin_atan2(PARAM[1]._float.value, PARAM[0]._float.value); goto __END;
__HYPOT: PARAM->_float.value = sqrt(PARAM[0]._float.value * PARAM[0]._float.value + PARAM[1]._float.value * PARAM[1]._float.value); goto __END;
__END:
if (!finite(PARAM->_float.value))
THROW(E_MATH);
SP--;
}
void SUBR_pow(void)
{
SUBR_ENTER_PARAM(2);
VALUE_conv_float(&PARAM[0]);
if (TYPE_is_integer(PARAM[1].type))
{
static void *jump[] = { &&__M4, &&__M3, &&__M2, &&__M1, &&__P0, &&__END, &&__P2, &&__P3, &&__P4 };
int val = PARAM[1]._integer.value;
if (val >= -4 && val <= 4)
goto *jump[val + 4];
else
goto __FLOAT;
__P0: PARAM->_float.value = 1.0; goto __END;
__P2: PARAM->_float.value *= PARAM->_float.value; goto __END;
__P3: PARAM->_float.value *= PARAM->_float.value * PARAM->_float.value; goto __END;
__P4: PARAM->_float.value *= PARAM->_float.value;PARAM->_float.value *= PARAM->_float.value; goto __END;
__M1: PARAM->_float.value = 1.0 / PARAM->_float.value; goto __END;
__M2: PARAM->_float.value = 1.0 / PARAM->_float.value / PARAM->_float.value; goto __END;
__M3: PARAM->_float.value = 1.0 / PARAM->_float.value / PARAM->_float.value / PARAM->_float.value; goto __END;
__M4: PARAM->_float.value = 1.0 / PARAM->_float.value / PARAM->_float.value / PARAM->_float.value / PARAM->_float.value; goto __END;
}
__FLOAT:
VALUE_conv_float(&PARAM[1]);
PARAM->_float.value = pow(PARAM[0]._float.value, PARAM[1]._float.value);
__END:
if (!finite(PARAM->_float.value))
THROW(E_MATH);
SP--;
}
void SUBR_not(ushort code)
{
static void *jump[17] = {
&&__VARIANT, &&__BOOLEAN, &&__BYTE, &&__SHORT, &&__INTEGER, &&__LONG, &&__SINGLE, &&__FLOAT, &&__DATE,
&&__STRING, &&__STRING, &&__ERROR, &&__ERROR, &&__ERROR, &&__ERROR, &&__NULL,
&&__OBJECT
};
VALUE *P1;
void *jump_end;
TYPE type = code & 0x1F;
bool test;
P1 = SP - 1;
jump_end = &&__END;
goto *jump[type];
__BOOLEAN:
P1->_integer.value = P1->_integer.value ? 0 : (-1);
goto *jump_end;
__BYTE:
P1->_integer.value = (unsigned char)~P1->_integer.value;
goto *jump_end;
__SHORT:
P1->_integer.value = (short)~P1->_integer.value;
goto *jump_end;
__INTEGER:
P1->_integer.value = ~P1->_integer.value;
goto *jump_end;
__LONG:
P1->_long.value = ~P1->_long.value;
goto *jump_end;
__SINGLE:
__FLOAT:
__DATE:
goto __ERROR;
__STRING:
__OBJECT:
__NULL:
test = VALUE_is_null(P1);
RELEASE(P1);
P1->_integer.value = test ? (-1) : 0;
P1->type = T_BOOLEAN;
goto *jump_end;
__VARIANT:
type = P1->type;
if (TYPE_is_variant(type))
{
type = P1->_variant.vtype;
jump_end = &&__VARIANT_END;
VARIANT_undo(P1);
}
else if (TYPE_is_object(type))
*PC |= T_OBJECT;
else if (type)
*PC |= type;
else
goto __ERROR;
if (TYPE_is_object(type))
goto __OBJECT;
else
goto *jump[type];
__ERROR:
THROW(E_TYPE, "Number, String or Object", TYPE_get_name(type));
__VARIANT_END:
VALUE_conv_variant(P1);
__END:
return;
}
void SUBR_and_(ushort code)
{
static void *jump[] = {
&&__VARIANT, &&__BOOLEAN, &&__BYTE, &&__SHORT, &&__INTEGER, &&__LONG, &&__ERROR, &&__ERROR, &&__ERROR
};
TYPE type;
VALUE *P1, *P2;
void *jump_end;
short op;
P1 = SP - 2;
P2 = P1 + 1;
jump_end = &&__END;
type = code & 0x0F;
op = (code >> 8) - (C_AND >> 8);
goto *jump[type];
__BYTE:
P1->type = type;
{
static void *exec[] = { &&__AND_C, && __OR_C, &&__XOR_C };
goto *exec[op];
__AND_C: P1->_integer.value = (unsigned char)(P1->_integer.value & P2->_integer.value); goto *jump_end;
__OR_C: P1->_integer.value = (unsigned char)(P1->_integer.value | P2->_integer.value); goto *jump_end;
__XOR_C: P1->_integer.value = (unsigned char)(P1->_integer.value ^ P2->_integer.value); goto *jump_end;
}
goto *jump_end;
__SHORT:
P1->type = type;
{
static void *exec[] = { &&__AND_H, && __OR_H, &&__XOR_H };
goto *exec[op];
__AND_H: P1->_integer.value = (short)(P1->_integer.value & P2->_integer.value); goto *jump_end;
__OR_H: P1->_integer.value = (short)(P1->_integer.value | P2->_integer.value); goto *jump_end;
__XOR_H: P1->_integer.value = (short)(P1->_integer.value ^ P2->_integer.value); goto *jump_end;
}
goto *jump_end;
__BOOLEAN:
__INTEGER:
P1->type = type;
{
static void *exec[] = { &&__AND_I, && __OR_I, &&__XOR_I };
goto *exec[op];
__AND_I: P1->_integer.value &= P2->_integer.value; goto *jump_end;
__OR_I: P1->_integer.value |= P2->_integer.value; goto *jump_end;
__XOR_I: P1->_integer.value ^= P2->_integer.value; goto *jump_end;
}
goto *jump_end;
__LONG:
VALUE_conv(P1, T_LONG);
VALUE_conv(P2, T_LONG);
P1->type = type;
{
static void *exec[] = { &&__AND_L, && __OR_L, &&__XOR_L };
goto *exec[op];
__AND_L: P1->_long.value &= P2->_long.value; goto *jump_end;
__OR_L: P1->_long.value |= P2->_long.value; goto *jump_end;
__XOR_L: P1->_long.value ^= P2->_long.value; goto *jump_end;
}
__VARIANT:
type = Max(P1->type, P2->type);
if (TYPE_is_number_date(type))
{
*PC |= type;
goto *jump[type];
}
if (TYPE_is_variant(P1->type))
VARIANT_undo(P1);
if (TYPE_is_variant(P2->type))
VARIANT_undo(P2);
if (TYPE_is_string(P1->type))
VALUE_convert_boolean(P1);
if (TYPE_is_string(P2->type))
VALUE_convert_boolean(P2);
if (TYPE_is_null(P1->type) || TYPE_is_null(P2->type))
type = T_NULL;
else
type = Max(P1->type, P2->type);
if (TYPE_is_number_date(type))
{
jump_end = &&__VARIANT_END;
goto *jump[type];
}
__ERROR:
THROW(E_TYPE, "Number", TYPE_get_name(type));
__VARIANT_END:
VALUE_conv_variant(P1);
__END:
SP--;
}
#define MANAGE_VARIANT(_func) \
({ \
type = P1->type; \
\
if (TYPE_is_number_date(type)) \
{ \
*PC |= type; \
goto *jump[type]; \
} \
\
if (TYPE_is_variant(type)) \
{ \
type = P1->_variant.vtype; \
if (TYPE_is_number_date(type)) \
{ \
VARIANT_undo(P1); \
(_func)(code | type); \
VALUE_conv_variant(P1); \
return; \
} \
} \
})
void SUBR_sgn(ushort code)
{
static void *jump[] = {
&&__VARIANT, &&__INTEGER, &&__INTEGER, &&__INTEGER, &&__INTEGER, &&__LONG, &&__SINGLE, &&__FLOAT, &&__ERROR
};
VALUE *P1;
TYPE type;
P1 = SP - 1;
type = code & 0x0F;
goto *jump[type];
__INTEGER: P1->_integer.value = lsgn(P1->_integer.value); goto __END;
__LONG: P1->_integer.value = llsgn(P1->_long.value); goto __END;
__SINGLE: P1->_integer.value = (P1->_single.value > 0) ? 1 : ((P1->_single.value < 0) ? (-1) : 0); goto __END;
__FLOAT: P1->_integer.value = fsgn(P1->_float.value); goto __END;
__VARIANT:
type = P1->type;
if (TYPE_is_number(type))
{
*PC |= type;
goto *jump[type];
}
if (TYPE_is_variant(type))
{
type = P1->_variant.vtype;
if (TYPE_is_number(type))
{
VARIANT_undo(P1);
goto *jump[type];
}
}
goto __ERROR;
__ERROR:
THROW(E_TYPE, "Number", TYPE_get_name(type));
__END:
P1->type = T_INTEGER;
}
void SUBR_neg(ushort code)
{
static void *jump[] = {
&&__VARIANT, &&__BOOLEAN, &&__BYTE, &&__SHORT, &&__INTEGER, &&__LONG, &&__SINGLE, &&__FLOAT, &&__ERROR
};
VALUE *P1;
TYPE type;
P1 = SP - 1;
type = code & 0x0F;
goto *jump[type];
__BOOLEAN:
return;
__BYTE:
P1->_integer.value = (unsigned char)(-P1->_integer.value); return;
__SHORT:
P1->_integer.value = (short)(-P1->_integer.value); return;
__INTEGER:
P1->_integer.value = (-P1->_integer.value); return;
__LONG:
P1->_long.value = (-P1->_long.value); return;
__SINGLE:
P1->_single.value = (-P1->_single.value); return;
__FLOAT:
P1->_float.value = (-P1->_float.value); return;
__VARIANT:
MANAGE_VARIANT(SUBR_neg);
__ERROR:
THROW(E_TYPE, "Number", TYPE_get_name(type));
}
void SUBR_abs(ushort code)
{
static void *jump[] = {
&&__VARIANT, &&__BOOLEAN, &&__BYTE, &&__SHORT, &&__INTEGER, &&__LONG, &&__SINGLE, &&__FLOAT, &&__ERROR
};
VALUE *P1;
TYPE type;
P1 = SP - 1;
type = code & 0x0F;
goto *jump[type];
__BOOLEAN:
return;
__BYTE:
P1->_integer.value = (unsigned char)ABS(-P1->_integer.value); return;
__SHORT:
P1->_integer.value = (short)ABS(P1->_integer.value); return;
__INTEGER:
P1->_integer.value = ABS(P1->_integer.value); return;
__LONG:
P1->_long.value = ABS(P1->_long.value); return;
__SINGLE:
P1->_single.value = fabsf(P1->_single.value); return;
__FLOAT:
P1->_float.value = fabs(P1->_float.value); return;
__VARIANT:
MANAGE_VARIANT(SUBR_abs);
__ERROR:
THROW(E_TYPE, "Number", TYPE_get_name(type));
}
void SUBR_int(ushort code)
{
static void *jump[] = {
&&__VARIANT, &&__BOOLEAN, &&__BYTE, &&__SHORT, &&__INTEGER, &&__LONG, &&__SINGLE, &&__FLOAT, &&__ERROR
};
VALUE *P1;
TYPE type;
P1 = SP - 1;
type = code & 0x0F;
goto *jump[type];
__BOOLEAN:
__BYTE:
__SHORT:
__INTEGER:
__LONG:
return;
__SINGLE:
P1->_single.value = floorf(P1->_single.value); return;
__FLOAT:
P1->_float.value = floor(P1->_float.value); return;
__VARIANT:
MANAGE_VARIANT(SUBR_int);
__ERROR:
THROW(E_TYPE, "Number", TYPE_get_name(type));
}
void SUBR_fix(ushort code)
{
static void *jump[] = {
&&__VARIANT, &&__BOOLEAN, &&__BYTE, &&__SHORT, &&__INTEGER, &&__LONG, &&__SINGLE, &&__FLOAT, &&__ERROR
};
VALUE *P1;
TYPE type;
P1 = SP - 1;
type = code & 0x0F;
goto *jump[type];
__BOOLEAN:
__BYTE:
__SHORT:
__INTEGER:
__LONG:
return;
__SINGLE:
P1->_single.value = fixf(P1->_single.value); return;
__FLOAT:
P1->_float.value = fix(P1->_float.value); return;
__VARIANT:
MANAGE_VARIANT(SUBR_fix);
__ERROR:
THROW(E_TYPE, "Number", TYPE_get_name(type));
}