mirror of
https://git.code.sf.net/p/quake/quakeforge
synced 2024-11-26 14:20:59 +00:00
d7bab4e222
The current code is pretty broken when it comes to vector types (losing the vector and bogus errors among other issues). The whole thing needs a rework or even just to be tossed in favor of better DAG processing.
1744 lines
43 KiB
C
1744 lines
43 KiB
C
/*
|
|
constfold.c
|
|
|
|
expression constant folding
|
|
|
|
Copyright (C) 2004 Bill Currie <bill@taniwha.org>
|
|
|
|
Author: Bill Currie <bill@taniwha.org>
|
|
Date: 2004/01/22
|
|
|
|
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
|
|
of the License, 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:
|
|
|
|
Free Software Foundation, Inc.
|
|
59 Temple Place - Suite 330
|
|
Boston, MA 02111-1307, USA
|
|
|
|
*/
|
|
#ifdef HAVE_CONFIG_H
|
|
# include "config.h"
|
|
#endif
|
|
|
|
#ifdef HAVE_STRING_H
|
|
# include <string.h>
|
|
#endif
|
|
#ifdef HAVE_STRINGS_H
|
|
# include <strings.h>
|
|
#endif
|
|
#include <stdlib.h>
|
|
|
|
#include <QF/dstring.h>
|
|
#include <QF/mathlib.h>
|
|
|
|
#include "tools/qfcc/include/diagnostic.h"
|
|
#include "tools/qfcc/include/evaluate.h"
|
|
#include "tools/qfcc/include/expr.h"
|
|
#include "tools/qfcc/include/options.h"
|
|
#include "tools/qfcc/include/qfcc.h"
|
|
#include "tools/qfcc/include/strpool.h"
|
|
#include "tools/qfcc/include/type.h"
|
|
#include "tools/qfcc/include/value.h"
|
|
|
|
#include "tools/qfcc/source/qc-parse.h"
|
|
|
|
typedef expr_t *(*operation_t) (int op, expr_t *e, expr_t *e1, expr_t *e2);
|
|
typedef expr_t *(*unaryop_t) (int op, expr_t *e, expr_t *e1);
|
|
|
|
static __attribute__((pure)) int
|
|
valid_op (int op, int *valid_ops)
|
|
{
|
|
while (*valid_ops && op != *valid_ops)
|
|
valid_ops++;
|
|
return *valid_ops == op;
|
|
}
|
|
|
|
static expr_t *
|
|
cmp_result_expr (int result)
|
|
{
|
|
if (is_float (type_default)) {
|
|
return new_float_expr (result);
|
|
} else {
|
|
return new_int_expr(result);
|
|
}
|
|
}
|
|
|
|
static int
|
|
is_addsub (int op)
|
|
{
|
|
return op == '+' || op == '-';
|
|
}
|
|
|
|
static int
|
|
inv_addsub (int op)
|
|
{
|
|
return op == '+' ? '-' : '+';
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_string (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
const char *s1, *s2;
|
|
static dstring_t *temp_str;
|
|
static int valid[] = {'+', LT, GT, LE, GE, EQ, NE, 0};
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid operand for string");
|
|
|
|
if (is_compare (op) || is_logic (op)) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
} else {
|
|
e->e.expr.type = &type_string;
|
|
}
|
|
|
|
if (!is_constant (e1) || !is_constant (e2))
|
|
return e;
|
|
|
|
s1 = expr_string (e1);
|
|
s2 = expr_string (e2);
|
|
if (!s1)
|
|
s1 = "";
|
|
if (!s2)
|
|
s2 = "";
|
|
|
|
switch (op) {
|
|
case '+':
|
|
if (!temp_str)
|
|
temp_str = dstring_newstr ();
|
|
dstring_clearstr (temp_str);
|
|
dstring_appendstr (temp_str, s1);
|
|
dstring_appendstr (temp_str, s2);
|
|
e = new_string_expr (save_string (temp_str->str));
|
|
break;
|
|
case LT:
|
|
e = cmp_result_expr (strcmp (s1, s2) < 0);
|
|
break;
|
|
case GT:
|
|
e = cmp_result_expr (strcmp (s1, s2) > 0);
|
|
break;
|
|
case LE:
|
|
e = cmp_result_expr (strcmp (s1, s2) <= 0);
|
|
break;
|
|
case GE:
|
|
e = cmp_result_expr (strcmp (s1, s2) >= 0);
|
|
break;
|
|
case EQ:
|
|
e = cmp_result_expr (strcmp (s1, s2) == 0);
|
|
break;
|
|
case NE:
|
|
e = cmp_result_expr (strcmp (s1, s2));
|
|
break;
|
|
default:
|
|
internal_error (e1, 0);
|
|
}
|
|
e->file = e1->file;
|
|
e->line = e1->line;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
convert_to_float (expr_t *e)
|
|
{
|
|
if (is_float(get_type (e)))
|
|
return e;
|
|
|
|
expr_t *n = cast_expr (&type_float, e);
|
|
n->file = e->file;
|
|
n->line = e->line;
|
|
return n;
|
|
}
|
|
|
|
static expr_t *
|
|
convert_to_double (expr_t *e)
|
|
{
|
|
if (is_double(get_type (e)))
|
|
return e;
|
|
|
|
expr_t *n = cast_expr (&type_double, e);
|
|
n->file = e->file;
|
|
n->line = e->line;
|
|
return n;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_float (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
float f1, f2;
|
|
expr_t *conv;
|
|
type_t *type = &type_float;
|
|
static int valid[] = {
|
|
'+', '-', '*', '/', '&', '|', '^', '%',
|
|
SHL, SHR, AND, OR, LT, GT, LE, GE, EQ, NE, 0
|
|
};
|
|
|
|
if (!is_scalar (get_type (e1)) || !is_scalar (get_type (e2))) {
|
|
return e;
|
|
}
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid operator for float");
|
|
|
|
if ((conv = convert_to_float (e1)) != e1) {
|
|
e->e.expr.e1 = e1 = conv;
|
|
}
|
|
if ((conv = convert_to_float (e2)) != e2) {
|
|
e->e.expr.e2 = e2 = conv;
|
|
}
|
|
if (is_compare (op) || is_logic (op)) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
type = &type_int;
|
|
else
|
|
type = &type_float;
|
|
}
|
|
e->e.expr.type = type;
|
|
|
|
if (op == '*' && is_constant (e1) && expr_float (e1) == 1)
|
|
return e2;
|
|
if (op == '*' && is_constant (e2) && expr_float (e2) == 1)
|
|
return e1;
|
|
if (op == '*' && is_constant (e1) && expr_float (e1) == 0)
|
|
return e1;
|
|
if (op == '*' && is_constant (e2) && expr_float (e2) == 0)
|
|
return e2;
|
|
if (op == '/' && is_constant (e2) && expr_float (e2) == 1)
|
|
return e1;
|
|
if (op == '/' && is_constant (e2) && expr_float (e2) == 0)
|
|
return error (e, "division by zero");
|
|
if (op == '/' && is_constant (e1) && expr_float (e1) == 0)
|
|
return e1;
|
|
if (op == '+' && is_constant (e1) && expr_float (e1) == 0)
|
|
return e2;
|
|
if (op == '+' && is_constant (e2) && expr_float (e2) == 0)
|
|
return e1;
|
|
if (op == '-' && is_constant (e2) && expr_float (e2) == 0)
|
|
return e1;
|
|
|
|
if (!is_constant (e1) || !is_constant (e2))
|
|
return e;
|
|
|
|
f1 = expr_float (e1);
|
|
f2 = expr_float (e2);
|
|
|
|
switch (op) {
|
|
case '+':
|
|
e = new_float_expr (f1 + f2);
|
|
break;
|
|
case '-':
|
|
e = new_float_expr (f1 - f2);
|
|
break;
|
|
case '*':
|
|
e = new_float_expr (f1 * f2);
|
|
break;
|
|
case '/':
|
|
if (!f2)
|
|
return error (e1, "divide by zero");
|
|
e = new_float_expr (f1 / f2);
|
|
break;
|
|
case '&':
|
|
e = new_float_expr ((int)f1 & (int)f2);
|
|
break;
|
|
case '|':
|
|
e = new_float_expr ((int)f1 | (int)f2);
|
|
break;
|
|
case '^':
|
|
e = new_float_expr ((int)f1 ^ (int)f2);
|
|
break;
|
|
case '%':
|
|
e = new_float_expr ((int)f1 % (int)f2);
|
|
break;
|
|
case SHL:
|
|
e = new_float_expr ((int)f1 << (int)f2);
|
|
break;
|
|
case SHR:
|
|
e = new_float_expr ((int)f1 >> (int)f2);
|
|
break;
|
|
case AND:
|
|
e = cmp_result_expr (f1 && f2);
|
|
break;
|
|
case OR:
|
|
e = cmp_result_expr (f1 || f2);
|
|
break;
|
|
case LT:
|
|
e = cmp_result_expr (f1 < f2);
|
|
break;
|
|
case GT:
|
|
e = cmp_result_expr (f1 > f2);
|
|
break;
|
|
case LE:
|
|
e = cmp_result_expr (f1 <= f2);
|
|
break;
|
|
case GE:
|
|
e = cmp_result_expr (f1 >= f2);
|
|
break;
|
|
case EQ:
|
|
e = cmp_result_expr (f1 == f2);
|
|
break;
|
|
case NE:
|
|
e = cmp_result_expr (f1 != f2);
|
|
break;
|
|
default:
|
|
internal_error (e1, 0);
|
|
}
|
|
e->file = e1->file;
|
|
e->line = e1->line;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_double (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
double d1, d2;
|
|
expr_t *conv;
|
|
type_t *type = &type_double;
|
|
static int valid[] = {
|
|
'+', '-', '*', '/', '%',
|
|
LT, GT, LE, GE, EQ, NE, 0
|
|
};
|
|
|
|
if (!is_scalar (get_type (e1)) || !is_scalar (get_type (e2))) {
|
|
return e;
|
|
}
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid operator for double");
|
|
|
|
if ((conv = convert_to_double (e1)) != e1) {
|
|
e->e.expr.e1 = e1 = conv;
|
|
}
|
|
if ((conv = convert_to_double (e2)) != e2) {
|
|
e->e.expr.e2 = e2 = conv;
|
|
}
|
|
if (is_compare (op) || is_logic (op)) {
|
|
type = &type_int;
|
|
}
|
|
e->e.expr.type = type;
|
|
|
|
if (op == '*' && is_constant (e1) && expr_double (e1) == 1)
|
|
return e2;
|
|
if (op == '*' && is_constant (e2) && expr_double (e2) == 1)
|
|
return e1;
|
|
if (op == '*' && is_constant (e1) && expr_double (e1) == 0)
|
|
return e1;
|
|
if (op == '*' && is_constant (e2) && expr_double (e2) == 0)
|
|
return e2;
|
|
if (op == '/' && is_constant (e2) && expr_double (e2) == 1)
|
|
return e1;
|
|
if (op == '/' && is_constant (e2) && expr_double (e2) == 0)
|
|
return error (e, "division by zero");
|
|
if (op == '/' && is_constant (e1) && expr_double (e1) == 0)
|
|
return e1;
|
|
if (op == '+' && is_constant (e1) && expr_double (e1) == 0)
|
|
return e2;
|
|
if (op == '+' && is_constant (e2) && expr_double (e2) == 0)
|
|
return e1;
|
|
if (op == '-' && is_constant (e2) && expr_double (e2) == 0)
|
|
return e1;
|
|
|
|
if (!is_constant (e1) || !is_constant (e2))
|
|
return e;
|
|
|
|
d1 = expr_double (e1);
|
|
d2 = expr_double (e2);
|
|
|
|
switch (op) {
|
|
case '+':
|
|
e = new_double_expr (d1 + d2);
|
|
break;
|
|
case '-':
|
|
e = new_double_expr (d1 - d2);
|
|
break;
|
|
case '*':
|
|
e = new_double_expr (d1 * d2);
|
|
break;
|
|
case '/':
|
|
if (!d2)
|
|
return error (e1, "divide by zero");
|
|
e = new_double_expr (d1 / d2);
|
|
break;
|
|
case '%':
|
|
e = new_double_expr ((int)d1 % (int)d2);
|
|
break;
|
|
case LT:
|
|
e = cmp_result_expr (d1 < d2);
|
|
break;
|
|
case GT:
|
|
e = cmp_result_expr (d1 > d2);
|
|
break;
|
|
case LE:
|
|
e = cmp_result_expr (d1 <= d2);
|
|
break;
|
|
case GE:
|
|
e = cmp_result_expr (d1 >= d2);
|
|
break;
|
|
case EQ:
|
|
e = cmp_result_expr (d1 == d2);
|
|
break;
|
|
case NE:
|
|
e = cmp_result_expr (d1 != d2);
|
|
break;
|
|
default:
|
|
internal_error (e1, 0);
|
|
}
|
|
e->file = e1->file;
|
|
e->line = e1->line;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_vector (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
const float *v1, *v2;
|
|
vec3_t v, float_vec;
|
|
static int vv_valid[] = {'+', '-', DOT, HADAMARD, EQ, NE, 0};
|
|
static int vs_valid[] = {SCALE, 0};
|
|
static int sv_valid[] = {SCALE, '/', 0};
|
|
expr_t *t;
|
|
|
|
if (!is_vector(get_type (e1))) {
|
|
|
|
if (!valid_op (op, sv_valid))
|
|
return error (e1, "invalid operator for scalar-vector");
|
|
|
|
t = e1;
|
|
e->e.expr.e1 = e1 = e2;
|
|
e2 = t;
|
|
}
|
|
if (!is_vector(get_type (e2))) {
|
|
e->e.expr.e2 = e2 = convert_to_float (e2);
|
|
if (!valid_op (op, vs_valid))
|
|
return error (e1, "invalid operator for vector");
|
|
} else {
|
|
if (!valid_op (op, vv_valid))
|
|
return error (e1, "invalid operator for vector");
|
|
}
|
|
if (is_compare (op) || is_logic (op)) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
} else if (op == DOT && is_vector(get_type (e2))) {
|
|
e->e.expr.type = &type_float;
|
|
} else if (op == '/' && !is_constant (e1)) {
|
|
e2 = fold_constants (binary_expr ('/', new_float_expr (1), e2));
|
|
e = fold_constants (binary_expr ('*', e1, e2));
|
|
} else {
|
|
e->e.expr.type = &type_vector;
|
|
}
|
|
|
|
if (op == SCALE && is_float_val (e2) && expr_float (e2) == 1)
|
|
return e1;
|
|
if (op == SCALE && is_float_val (e2) && expr_float (e2) == 0)
|
|
return new_vector_expr (vec3_origin);
|
|
if (op == '/' && is_float_val (e2) && expr_float (e2) == 1)
|
|
return e1;
|
|
if (op == '/' && is_float_val (e2) && expr_float (e2) == 0)
|
|
return error (e, "division by zero");
|
|
if (op == '+' && is_constant (e1) && VectorIsZero (expr_vector (e1)))
|
|
return e2;
|
|
if (op == '+' && is_constant (e2) && VectorIsZero (expr_vector (e2)))
|
|
return e1;
|
|
if (op == '-' && is_constant (e1) && VectorIsZero (expr_vector (e1))
|
|
&& is_constant (e2)) {
|
|
vec3_t v;
|
|
VectorNegate (expr_vector (e2), v);
|
|
e = new_vector_expr (v);
|
|
e->file = e2->file;
|
|
e->line = e2->line;
|
|
return e;
|
|
}
|
|
if (op == '-' && is_constant (e2) && VectorIsZero (expr_vector (e2)))
|
|
return e1;
|
|
|
|
if (!is_constant (e1) || !is_constant (e2))
|
|
return e;
|
|
|
|
if (is_float_val (e1)) {
|
|
float_vec[0] = expr_float (e1);
|
|
v2 = float_vec;
|
|
v1 = expr_vector (e2);
|
|
} else if (is_float_val (e2)) {
|
|
float_vec[0] = expr_float (e2);
|
|
v2 = float_vec;
|
|
v1 = expr_vector (e1);
|
|
} else {
|
|
v1 = expr_vector (e1);
|
|
v2 = expr_vector (e2);
|
|
}
|
|
|
|
switch (op) {
|
|
case '+':
|
|
VectorAdd (v1, v2, v);
|
|
e = new_vector_expr (v);
|
|
break;
|
|
case '-':
|
|
VectorSubtract (v1, v2, v);
|
|
e = new_vector_expr (v);
|
|
break;
|
|
case '/':
|
|
if (!v2[0])
|
|
return error (e1, "divide by zero");
|
|
VectorScale (v1, 1 / v2[0], v);
|
|
e = new_vector_expr (v);
|
|
break;
|
|
case DOT:
|
|
e = new_float_expr (DotProduct (v1, v2));
|
|
break;
|
|
case SCALE:
|
|
VectorScale (v1, v2[0], v);
|
|
e = new_vector_expr (v);
|
|
break;
|
|
case EQ:
|
|
e = cmp_result_expr (VectorCompare (v1, v2));
|
|
break;
|
|
case NE:
|
|
e = cmp_result_expr (!VectorCompare (v1, v2));
|
|
break;
|
|
default:
|
|
internal_error (e1, 0);
|
|
}
|
|
e->file = e1->file;
|
|
e->line = e1->line;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_entity (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
type_t *type = get_type (e2);
|
|
|
|
if (op == '.' && type->type == ev_field) {
|
|
return e;
|
|
}
|
|
if (op == EQ || op == NE) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
return e;
|
|
}
|
|
if (!is_entity(type))
|
|
return error (e1, "invalid operator for entity");
|
|
e->e.expr.type = &type_entity;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_field (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
return error (e1, "invalid operator for field");
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_func (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
if (op == EQ || op == NE) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
return e;
|
|
}
|
|
return error (e1, "invalid operator for func");
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_pointer (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
type_t *type;
|
|
static int valid[] = {'-', '.', EQ, NE, 0};
|
|
|
|
if (is_integral (type = get_type (e2)) && (op == '-' || op == '+')) {
|
|
// pointer arithmetic
|
|
expr_t *ptoi = new_alias_expr (type, e1);
|
|
e->e.expr.e1 = ptoi;
|
|
e = fold_constants (e);
|
|
return new_alias_expr (get_type (e1), e);
|
|
}
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid operator for pointer: %s",
|
|
get_op_string (op));
|
|
|
|
if (op == '-') {
|
|
type = get_type (e1);
|
|
if (type != get_type (e2))
|
|
return error (e2, "invalid operands to binary -");
|
|
e1 = new_alias_expr (&type_int, e1);
|
|
e2 = new_alias_expr (&type_int, e2);
|
|
e = binary_expr ('-', e1, e2);
|
|
if (type_size (type) != 1)
|
|
e = binary_expr ('/', e, new_int_expr (type_size (type)));
|
|
return e;
|
|
}
|
|
if (op == EQ || op == NE) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
}
|
|
if (op != '.' && extract_type (e1) != extract_type (e2))
|
|
return type_mismatch (e1, e2, op);
|
|
if (op == '.' && is_uint(get_type (e2)))
|
|
e->e.expr.e2 = cast_expr (&type_int, e2);
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_quatvect (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
if (op != '*')
|
|
return error (e1, "invalid operator for quaternion and vector");
|
|
if (is_constant (e1) && QuatIsZero (expr_quaternion (e1)))
|
|
return new_vector_expr (vec3_origin);
|
|
if (is_constant (e2) && VectorIsZero (expr_vector (e2)))
|
|
return new_vector_expr (vec3_origin);
|
|
if (is_constant (e1) && is_constant (e2)) {
|
|
const vec_t *q, *v;
|
|
vec3_t r;
|
|
|
|
q = expr_quaternion (e1);
|
|
v = expr_vector (e2);
|
|
QuatMultVec (q, v, r);
|
|
return new_vector_expr (r);
|
|
}
|
|
e->e.expr.type = &type_vector;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_quaternion (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
const float *q1, *q2;
|
|
quat_t q, float_quat;
|
|
static int valid[] = {'+', '-', '*', EQ, NE, 0};
|
|
expr_t *t;
|
|
|
|
if (!is_quaternion(get_type (e1))) {
|
|
|
|
if (op != '*' && op != '/')
|
|
return error (e1, "invalid operator for quaternion");
|
|
|
|
if (op == '*') {
|
|
t = e1;
|
|
e->e.expr.e1 = e1 = e2;
|
|
e2 = t;
|
|
}
|
|
}
|
|
if (!is_quaternion(get_type (e2))) {
|
|
e->e.expr.e2 = e2 = convert_to_float (e2);
|
|
if (op != '*' && op != '/')
|
|
return error (e1, "invalid operator for quaternion");
|
|
} else {
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid operator for quaternion");
|
|
}
|
|
if (is_compare (op) || is_logic (op)) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
} else if (op == '/' && !is_constant (e1)) {
|
|
e2 = fold_constants (binary_expr ('/', new_float_expr (1), e2));
|
|
e = fold_constants (binary_expr ('*', e1, e2));
|
|
} else {
|
|
e->e.expr.type = &type_quaternion;
|
|
}
|
|
|
|
if (op == '*' && is_float_val (e2) && expr_float (e2) == 1)
|
|
return e1;
|
|
if (op == '*' && is_float_val (e2) && expr_float (e2) == 0)
|
|
return new_quaternion_expr (quat_origin);
|
|
if (op == '/' && is_float_val (e2) && expr_float (e2) == 1)
|
|
return e1;
|
|
if (op == '/' && is_float_val (e2) && expr_float (e2) == 0)
|
|
return error (e, "division by zero");
|
|
if (op == '+' && is_constant (e1) && QuatIsZero (expr_quaternion (e1)))
|
|
return e2;
|
|
if (op == '+' && is_constant (e2) && QuatIsZero (expr_quaternion (e2)))
|
|
return e1;
|
|
if (op == '-' && is_constant (e2) && QuatIsZero (expr_quaternion (e2)))
|
|
return e1;
|
|
|
|
if (!is_constant (e1) || !is_constant (e2))
|
|
return e;
|
|
|
|
if (is_float_val (e1)) {
|
|
QuatSet (0, 0, 0, expr_float (e1), float_quat);
|
|
q2 = float_quat;
|
|
q1 = expr_quaternion (e2);
|
|
} else if (is_float_val (e2)) {
|
|
QuatSet (0, 0, 0, expr_float (e2), float_quat);
|
|
q2 = float_quat;
|
|
q1 = expr_quaternion (e1);
|
|
} else {
|
|
q1 = expr_quaternion (e1);
|
|
q2 = expr_quaternion (e2);
|
|
}
|
|
|
|
switch (op) {
|
|
case '+':
|
|
QuatAdd (q1, q2, q);
|
|
e = new_quaternion_expr (q);
|
|
break;
|
|
case '-':
|
|
QuatSubtract (q1, q2, q);
|
|
e = new_quaternion_expr (q);
|
|
break;
|
|
case '/':
|
|
if (is_float_val (e2)) {
|
|
QuatScale (q1, 1 / expr_float (e2), q);
|
|
} else {
|
|
QuatInverse (q2, q);
|
|
QuatScale (q2, expr_float (e1), q);
|
|
}
|
|
e = new_quaternion_expr (q);
|
|
break;
|
|
case '*':
|
|
if (is_quaternion(get_type (e2))) {
|
|
QuatMult (q1, q2, q);
|
|
} else {
|
|
QuatScale (q1, q2[3], q);
|
|
}
|
|
e = new_quaternion_expr (q);
|
|
break;
|
|
case EQ:
|
|
e = cmp_result_expr (QuatCompare (q1, q2));
|
|
break;
|
|
case NE:
|
|
e = cmp_result_expr (!QuatCompare (q1, q2));
|
|
break;
|
|
default:
|
|
internal_error (e1, 0);
|
|
}
|
|
e->file = e1->file;
|
|
e->line = e1->line;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_int (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
int isval1 = 0, isval2 = 0;
|
|
int val1 = 0, val2 = 0;
|
|
static int valid[] = {
|
|
'+', '-', '*', '/', '&', '|', '^', '%',
|
|
SHL, SHR, AND, OR, LT, GT, LE, GE, EQ, NE, 0
|
|
};
|
|
|
|
if (!is_scalar (get_type (e1)) || !is_scalar (get_type (e2))) {
|
|
return e;
|
|
}
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid operator for int");
|
|
|
|
if (is_short_val (e1)) {
|
|
isval1 = 1;
|
|
val1 = expr_short (e1);
|
|
}
|
|
if (is_int_val (e1)) {
|
|
isval1 = 1;
|
|
val1 = expr_int (e1);
|
|
}
|
|
|
|
if (is_short_val (e2)) {
|
|
isval2 = 1;
|
|
val2 = expr_short (e2);
|
|
}
|
|
if (is_int_val (e2)) {
|
|
isval2 = 1;
|
|
val2 = expr_int (e2);
|
|
}
|
|
|
|
if (is_compare (op) || is_logic (op)) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
} else {
|
|
e->e.expr.type = &type_int;
|
|
}
|
|
|
|
if (op == '*' && isval1 && val1 == 1)
|
|
return e2;
|
|
if (op == '*' && isval2 && val2 == 1)
|
|
return e1;
|
|
if (op == '*' && isval1 && val1 == 0)
|
|
return e1;
|
|
if (op == '*' && isval2 && val2 == 0)
|
|
return e2;
|
|
if (op == '/' && isval2 && val2 == 1)
|
|
return e1;
|
|
if (op == '/' && isval2 && val2 == 0)
|
|
return error (e, "division by zero");
|
|
if (op == '/' && isval1 && val1 == 0)
|
|
return e1;
|
|
if (op == '+' && isval1 && val1 == 0)
|
|
return e2;
|
|
if (op == '+' && isval2 && val2 == 0)
|
|
return e1;
|
|
if (op == '-' && isval2 && val2 == 0)
|
|
return e1;
|
|
|
|
if (!isval1) {
|
|
if (e1->type == ex_expr) {
|
|
// at most one of the two sub-expressions is constant otherwise
|
|
// e1 would be a constant
|
|
if (is_constant (e1->e.expr.e1)) {
|
|
if ((op == '*' && e1->e.expr.op == '*')
|
|
|| (is_addsub (op) && is_addsub (e1->e.expr.op))) {
|
|
expr_t *c = binary_expr (op, e1->e.expr.e1, e2);
|
|
e = binary_expr (e1->e.expr.op, c, e1->e.expr.e2);
|
|
}
|
|
} else if (is_constant (e1->e.expr.e2)) {
|
|
if ((op == '*' && e1->e.expr.op == '*')
|
|
|| (is_addsub (op) && e1->e.expr.op == '+')) {
|
|
expr_t *c = binary_expr (op, e1->e.expr.e2, e2);
|
|
e = binary_expr (e1->e.expr.op, e1->e.expr.e1, c);
|
|
} else if (is_addsub (op) && e1->e.expr.op == '-') {
|
|
// must ivert op
|
|
expr_t *c = binary_expr (inv_addsub (op),
|
|
e1->e.expr.e2, e2);
|
|
e = binary_expr (e1->e.expr.op, e1->e.expr.e1, c);
|
|
}
|
|
}
|
|
}
|
|
return e;
|
|
} else if (!isval2) {
|
|
if (e2->type == ex_expr) {
|
|
// at most one of the two sub-expressions is constant otherwise
|
|
// e2 would be a constant
|
|
if (is_constant (e2->e.expr.e1)) {
|
|
if ((op == '*' && e2->e.expr.op == '*')
|
|
|| (op == '+' && is_addsub (e2->e.expr.op))) {
|
|
expr_t *c = binary_expr (op, e1, e2->e.expr.e1);
|
|
e = binary_expr (e2->e.expr.op, c, e2->e.expr.e2);
|
|
} else if (op == '-' && is_addsub (e2->e.expr.op)) {
|
|
expr_t *c = binary_expr (op, e1, e2->e.expr.e1);
|
|
c = fold_constants (c);
|
|
e = binary_expr (inv_addsub (e2->e.expr.op),
|
|
c, e2->e.expr.e2);
|
|
}
|
|
} else if (is_constant (e2->e.expr.e2)) {
|
|
if ((op == '*' && e2->e.expr.op == '*')
|
|
|| (op == '+' && is_addsub (e2->e.expr.op))) {
|
|
expr_t *c = binary_expr (e2->e.expr.op,
|
|
e1, e2->e.expr.e2);
|
|
e = binary_expr (op, c, e2->e.expr.e1);
|
|
} else if (op == '-' && is_addsub (e2->e.expr.op)) {
|
|
expr_t *c = binary_expr (inv_addsub (e2->e.expr.op),
|
|
e1, e2->e.expr.e2);
|
|
e = binary_expr (op, c, e2->e.expr.e1);
|
|
}
|
|
}
|
|
}
|
|
return e;
|
|
}
|
|
if (!isval1 || !isval2)
|
|
return e;
|
|
|
|
switch (op) {
|
|
case '+':
|
|
e = new_int_expr (val1 + val2);
|
|
break;
|
|
case '-':
|
|
e = new_int_expr (val1 - val2);
|
|
break;
|
|
case '*':
|
|
e = new_int_expr (val1 * val2);
|
|
break;
|
|
case '/':
|
|
if (options.warnings.integer_divide)
|
|
warning (e2, "%d / %d == %d", val1, val2, val1 / val2);
|
|
e = new_int_expr (val1 / val2);
|
|
break;
|
|
case '&':
|
|
e = new_int_expr (val1 & val2);
|
|
break;
|
|
case '|':
|
|
e = new_int_expr (val1 | val2);
|
|
break;
|
|
case '^':
|
|
e = new_int_expr (val1 ^ val2);
|
|
break;
|
|
case '%':
|
|
e = new_int_expr (val1 % val2);
|
|
break;
|
|
case SHL:
|
|
e = new_int_expr (val1 << val2);
|
|
break;
|
|
case SHR:
|
|
e = new_int_expr (val1 >> val2);
|
|
break;
|
|
case AND:
|
|
e = cmp_result_expr (val1 && val2);
|
|
break;
|
|
case OR:
|
|
e = cmp_result_expr (val1 || val2);
|
|
break;
|
|
case LT:
|
|
e = cmp_result_expr (val1 < val2);
|
|
break;
|
|
case GT:
|
|
e = cmp_result_expr (val1 > val2);
|
|
break;
|
|
case LE:
|
|
e = cmp_result_expr (val1 <= val2);
|
|
break;
|
|
case GE:
|
|
e = cmp_result_expr (val1 >= val2);
|
|
break;
|
|
case EQ:
|
|
e = cmp_result_expr (val1 == val2);
|
|
break;
|
|
case NE:
|
|
e = cmp_result_expr (val1 != val2);
|
|
break;
|
|
default:
|
|
internal_error (e1, 0);
|
|
}
|
|
e->file = e1->file;
|
|
e->line = e1->line;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_uint (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_short (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
short i1, i2;
|
|
static int valid[] = {
|
|
'+', '-', '*', '/', '&', '|', '^', '%',
|
|
SHL, SHR, AND, OR, LT, GT, LE, GE, EQ, NE, 0
|
|
};
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid operator for short");
|
|
|
|
if (is_compare (op) || is_logic (op)) {
|
|
if (options.code.progsversion > PROG_ID_VERSION)
|
|
e->e.expr.type = &type_int;
|
|
else
|
|
e->e.expr.type = &type_float;
|
|
} else {
|
|
e->e.expr.type = &type_short;
|
|
}
|
|
|
|
if (!is_constant (e1) || !is_constant (e2))
|
|
return e;
|
|
|
|
i1 = expr_short (e1);
|
|
i2 = expr_short (e2);
|
|
|
|
switch (op) {
|
|
case '+':
|
|
e = new_short_expr (i1 + i2);
|
|
break;
|
|
case '-':
|
|
e = new_short_expr (i1 - i2);
|
|
break;
|
|
case '*':
|
|
e = new_short_expr (i1 * i2);
|
|
break;
|
|
case '/':
|
|
if (options.warnings.integer_divide)
|
|
warning (e2, "%d / %d == %d", i1, i2, i1 / i2);
|
|
e = new_short_expr (i1 / i2);
|
|
break;
|
|
case '&':
|
|
e = new_short_expr (i1 & i2);
|
|
break;
|
|
case '|':
|
|
e = new_short_expr (i1 | i2);
|
|
break;
|
|
case '^':
|
|
e = new_short_expr (i1 ^ i2);
|
|
break;
|
|
case '%':
|
|
e = new_short_expr (i1 % i2);
|
|
break;
|
|
case SHL:
|
|
e = new_short_expr (i1 << i2);
|
|
break;
|
|
case SHR:
|
|
e = new_short_expr (i1 >> i2);
|
|
break;
|
|
case AND:
|
|
e = new_short_expr (i1 && i2);
|
|
break;
|
|
case OR:
|
|
e = new_short_expr (i1 || i2);
|
|
break;
|
|
case LT:
|
|
e = cmp_result_expr (i1 < i2);
|
|
break;
|
|
case GT:
|
|
e = cmp_result_expr (i1 > i2);
|
|
break;
|
|
case LE:
|
|
e = cmp_result_expr (i1 <= i2);
|
|
break;
|
|
case GE:
|
|
e = cmp_result_expr (i1 >= i2);
|
|
break;
|
|
case EQ:
|
|
e = cmp_result_expr (i1 == i2);
|
|
break;
|
|
case NE:
|
|
e = cmp_result_expr (i1 != i2);
|
|
break;
|
|
default:
|
|
internal_error (e1, 0);
|
|
}
|
|
e->file = e1->file;
|
|
e->line = e1->line;
|
|
return e;
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_struct (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
return error (e1, "invalid operator for struct");
|
|
}
|
|
|
|
static expr_t *
|
|
do_op_compound (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
type_t *t1 = get_type (e1);
|
|
type_t *t2 = get_type (e2);
|
|
if (is_struct (t1) && is_struct (t2))
|
|
return do_op_struct (op, e, e1, e2);
|
|
if (is_union (t1) && is_union (t2))
|
|
return do_op_struct (op, e, e1, e2);
|
|
if (is_scalar (t1) && is_scalar (t2)) {
|
|
if (is_enum (t1)) {
|
|
if (t2->type == ev_double)
|
|
return do_op_float (op, e, e1, e2);
|
|
if (t2->type == ev_double)
|
|
return do_op_float (op, e, e1, e2);
|
|
return do_op_int (op, e, e1, e2);
|
|
}
|
|
if (is_enum (t2)) {
|
|
if (t1->type == ev_double)
|
|
return do_op_double (op, e, e1, e2);
|
|
if (t1->type == ev_float)
|
|
return do_op_float (op, e, e1, e2);
|
|
return do_op_int (op, e, e1, e2);
|
|
}
|
|
}
|
|
return error (e1, "invalid operator for compound");
|
|
}
|
|
|
|
static operation_t *do_op[ev_type_count];
|
|
static expr_t *
|
|
do_op_invalid (int op, expr_t *e, expr_t *e1, expr_t *e2)
|
|
{
|
|
type_t *t1 = get_type (e1);
|
|
type_t *t2 = get_type (e2);
|
|
|
|
if (is_scalar (t1) && is_scalar (t2)) {
|
|
// one or both expressions are an enum, and the other is one of
|
|
// int, float or short. Treat the enum as the other type, or as
|
|
// the default type if both are enum.
|
|
etype_t t;
|
|
if (!is_enum (t1))
|
|
t = t1->type;
|
|
else if (!is_enum (t2))
|
|
t = t2->type;
|
|
else
|
|
t = type_default->type;
|
|
return do_op[t][t] (op, e, e1, e2);
|
|
} else {
|
|
dstring_t *enc1 = dstring_newstr ();
|
|
dstring_t *enc2 = dstring_newstr ();
|
|
|
|
print_type_str (enc1, t1);
|
|
print_type_str (enc2, t2);
|
|
|
|
//print_expr (e);
|
|
e1 = error (e1, "invalid operands for binary %s: %s %s",
|
|
get_op_string (op), enc1->str, enc2->str);
|
|
dstring_delete (enc1);
|
|
dstring_delete (enc2);
|
|
return e1;
|
|
}
|
|
}
|
|
|
|
static operation_t op_void[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_invalid,
|
|
[ev_vector] = do_op_invalid,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_invalid,
|
|
[ev_int] = do_op_invalid,
|
|
[ev_uint] = do_op_invalid,
|
|
[ev_short] = do_op_invalid,
|
|
[ev_double] = do_op_invalid,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_string[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_string,
|
|
[ev_float] = do_op_invalid,
|
|
[ev_vector] = do_op_invalid,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_invalid,
|
|
[ev_int] = do_op_invalid,
|
|
[ev_uint] = do_op_invalid,
|
|
[ev_short] = do_op_invalid,
|
|
[ev_double] = do_op_invalid,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_float[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_float,
|
|
[ev_vector] = do_op_vector,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_quaternion,
|
|
[ev_int] = do_op_float,
|
|
[ev_uint] = do_op_float,
|
|
[ev_short] = do_op_float,
|
|
[ev_double] = do_op_double,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_vector[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_vector,
|
|
[ev_vector] = do_op_vector,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_invalid,
|
|
[ev_int] = do_op_vector,
|
|
[ev_uint] = do_op_vector,
|
|
[ev_short] = do_op_vector,
|
|
[ev_double] = do_op_vector,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_entity[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_invalid,
|
|
[ev_vector] = do_op_invalid,
|
|
[ev_entity] = do_op_entity,
|
|
[ev_field] = do_op_entity,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_invalid,
|
|
[ev_int] = do_op_invalid,
|
|
[ev_uint] = do_op_invalid,
|
|
[ev_short] = do_op_invalid,
|
|
[ev_double] = do_op_invalid,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_field[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_invalid,
|
|
[ev_vector] = do_op_invalid,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_field,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_invalid,
|
|
[ev_int] = do_op_invalid,
|
|
[ev_uint] = do_op_invalid,
|
|
[ev_short] = do_op_invalid,
|
|
[ev_double] = do_op_invalid,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_func[ev_type_count] = {
|
|
[ev_void] = do_op_func,
|
|
[ev_string] = do_op_func,
|
|
[ev_float] = do_op_func,
|
|
[ev_vector] = do_op_func,
|
|
[ev_entity] = do_op_func,
|
|
[ev_field] = do_op_func,
|
|
[ev_func] = do_op_func,
|
|
[ev_ptr] = do_op_func,
|
|
[ev_quaternion] = do_op_func,
|
|
[ev_int] = do_op_func,
|
|
[ev_uint] = do_op_func,
|
|
[ev_short] = do_op_func,
|
|
[ev_double] = do_op_func,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_func,
|
|
};
|
|
|
|
static operation_t op_pointer[ev_type_count] = {
|
|
[ev_void] = do_op_pointer,
|
|
[ev_string] = do_op_pointer,
|
|
[ev_float] = do_op_pointer,
|
|
[ev_vector] = do_op_pointer,
|
|
[ev_entity] = do_op_pointer,
|
|
[ev_field] = do_op_pointer,
|
|
[ev_func] = do_op_pointer,
|
|
[ev_ptr] = do_op_pointer,
|
|
[ev_quaternion] = do_op_pointer,
|
|
[ev_int] = do_op_pointer,
|
|
[ev_uint] = do_op_pointer,
|
|
[ev_short] = do_op_pointer,
|
|
[ev_double] = do_op_pointer,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_pointer,
|
|
};
|
|
|
|
static operation_t op_quaternion[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_quaternion,
|
|
[ev_vector] = do_op_quatvect,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_quaternion,
|
|
[ev_int] = do_op_quaternion,
|
|
[ev_uint] = do_op_quaternion,
|
|
[ev_short] = do_op_quaternion,
|
|
[ev_double] = do_op_quaternion,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_int[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_float,
|
|
[ev_vector] = do_op_vector,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_quaternion,
|
|
[ev_int] = do_op_int,
|
|
[ev_uint] = do_op_uint,
|
|
[ev_short] = do_op_int,
|
|
[ev_double] = do_op_double,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_uint[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_float,
|
|
[ev_vector] = do_op_vector,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_quaternion,
|
|
[ev_int] = do_op_uint,
|
|
[ev_uint] = do_op_uint,
|
|
[ev_short] = do_op_uint,
|
|
[ev_double] = do_op_double,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_short[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_float,
|
|
[ev_vector] = do_op_vector,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_quaternion,
|
|
[ev_int] = do_op_int,
|
|
[ev_uint] = do_op_uint,
|
|
[ev_short] = do_op_short,
|
|
[ev_double] = do_op_double,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_double[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_float,
|
|
[ev_vector] = do_op_vector,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_quaternion,
|
|
[ev_int] = do_op_int,
|
|
[ev_uint] = do_op_uint,
|
|
[ev_short] = do_op_short,
|
|
[ev_double] = do_op_double,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = do_op_invalid,
|
|
};
|
|
|
|
static operation_t op_compound[ev_type_count] = {
|
|
[ev_void] = do_op_invalid,
|
|
[ev_string] = do_op_invalid,
|
|
[ev_float] = do_op_compound,
|
|
[ev_vector] = do_op_invalid,
|
|
[ev_entity] = do_op_invalid,
|
|
[ev_field] = do_op_invalid,
|
|
[ev_func] = do_op_invalid,
|
|
[ev_ptr] = do_op_invalid,
|
|
[ev_quaternion] = do_op_invalid,
|
|
[ev_int] = do_op_compound,
|
|
[ev_uint] = do_op_compound,
|
|
[ev_short] = do_op_compound,
|
|
[ev_double] = do_op_compound,
|
|
[ev_long] = do_op_compound,
|
|
[ev_ulong] = do_op_compound,
|
|
[ev_invalid] = do_op_compound,
|
|
};
|
|
|
|
static operation_t *do_op[ev_type_count] = {
|
|
[ev_void] = op_void,
|
|
[ev_string] = op_string,
|
|
[ev_float] = op_float,
|
|
[ev_vector] = op_vector,
|
|
[ev_entity] = op_entity,
|
|
[ev_field] = op_field,
|
|
[ev_func] = op_func,
|
|
[ev_ptr] = op_pointer,
|
|
[ev_quaternion] = op_quaternion,
|
|
[ev_int] = op_int,
|
|
[ev_uint] = op_uint,
|
|
[ev_short] = op_short,
|
|
[ev_double] = op_double,
|
|
[ev_long] = 0,
|
|
[ev_ulong] = 0,
|
|
[ev_invalid] = op_compound,
|
|
};
|
|
|
|
static unaryop_t do_unary_op[ev_type_count];
|
|
static expr_t *
|
|
uop_invalid (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
type_t *t1 = get_type (e1);
|
|
if (is_scalar (t1)) {
|
|
// The expression is an enum. Treat the enum as the default type.
|
|
etype_t t;
|
|
t = type_default->type;
|
|
return do_unary_op[t] (op, e, e1);
|
|
} else {
|
|
dstring_t *enc1 = dstring_newstr ();
|
|
|
|
print_type_str (enc1, t1);
|
|
|
|
//print_expr (e);
|
|
e1 = error (e1, "invalid operand for unary %s: %s",
|
|
get_op_string (op), enc1->str);
|
|
dstring_delete (enc1);
|
|
return e1;
|
|
}
|
|
}
|
|
|
|
static expr_t *
|
|
uop_string (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
// + - ! ~ *
|
|
static int valid[] = { '!', 0 };
|
|
const char *s;
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for string: %s",
|
|
get_op_string (op));
|
|
|
|
if (!is_constant (e1))
|
|
return e;
|
|
|
|
s = expr_string (e1);
|
|
return cmp_result_expr (!s || !s[0]);
|
|
}
|
|
|
|
static expr_t *
|
|
uop_float (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '+', '-', '!', '~', 'C', 0 };
|
|
type_t *type;
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for float: %s",
|
|
get_op_string (op));
|
|
if (op == '+')
|
|
return e1;
|
|
type = get_type (e);
|
|
if (op == 'C' && !is_int(type) && !is_double(type))
|
|
return error (e1, "invalid cast of float");
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '-':
|
|
return new_float_expr (-expr_float (e1));
|
|
case '!':
|
|
print_type (get_type (e));
|
|
return cmp_result_expr (!expr_float (e1));
|
|
case '~':
|
|
return new_float_expr (~(int) expr_float (e1));
|
|
case 'C':
|
|
if (is_int(type)) {
|
|
return new_int_expr (expr_float (e1));
|
|
} else {
|
|
return new_double_expr (expr_float (e1));
|
|
}
|
|
}
|
|
internal_error (e, "float unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_vector (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '+', '-', '!', 0 };
|
|
vec3_t v;
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for vector: %s",
|
|
get_op_string (op));
|
|
if (op == '+')
|
|
return e1;
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '-':
|
|
VectorNegate (expr_vector (e), v);
|
|
return new_vector_expr (v);
|
|
case '!':
|
|
return cmp_result_expr (!VectorIsZero (expr_vector (e1)));
|
|
}
|
|
internal_error (e, "vector unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_entity (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '!', 0 };
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for entity: %s",
|
|
get_op_string (op));
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '!':
|
|
internal_error (e, "!entity");
|
|
}
|
|
internal_error (e, "entity unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_field (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '!', 0 };
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for field: %s",
|
|
get_op_string (op));
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '!':
|
|
internal_error (e, "!field");
|
|
}
|
|
internal_error (e, "field unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_func (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '!', 0 };
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for func: %s",
|
|
get_op_string (op));
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '!':
|
|
internal_error (e, "!func");
|
|
}
|
|
internal_error (e, "func unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_pointer (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '!', '.', 0 };
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for pointer: %s",
|
|
get_op_string (op));
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '!':
|
|
internal_error (e, "!pointer");
|
|
case '.':
|
|
debug (e, ".pointer");
|
|
return e;
|
|
}
|
|
internal_error (e, "pointer unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_quaternion (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '+', '-', '!', '~', 0 };
|
|
quat_t q;
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for quaternion: %s",
|
|
get_op_string (op));
|
|
if (op == '+')
|
|
return e1;
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '-':
|
|
QuatNegate (expr_vector (e), q);
|
|
return new_quaternion_expr (q);
|
|
case '!':
|
|
return cmp_result_expr (!QuatIsZero (expr_quaternion (e1)));
|
|
case '~':
|
|
QuatConj (expr_vector (e), q);
|
|
return new_quaternion_expr (q);
|
|
}
|
|
internal_error (e, "quaternion unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_int (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '+', '-', '!', '~', 'C', 0 };
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for int: %s",
|
|
get_op_string (op));
|
|
if (op == '+')
|
|
return e1;
|
|
if (op == 'C' && !is_float(get_type (e)))
|
|
return error (e1, "invalid cast of int");
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '-':
|
|
return new_int_expr (-expr_int (e1));
|
|
case '!':
|
|
return cmp_result_expr (!expr_int (e1));
|
|
case '~':
|
|
return new_int_expr (~expr_int (e1));
|
|
case 'C':
|
|
return new_float_expr (expr_int (e1));
|
|
}
|
|
internal_error (e, "int unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_uint (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '+', '-', '!', '~', 0 };
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for uint: %s",
|
|
get_op_string (op));
|
|
if (op == '+')
|
|
return e1;
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '-':
|
|
return new_uint_expr (-expr_uint (e1));
|
|
case '!':
|
|
return cmp_result_expr (!expr_uint (e1));
|
|
case '~':
|
|
return new_uint_expr (~expr_uint (e1));
|
|
}
|
|
internal_error (e, "uint unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_short (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '+', '-', '!', '~', 0 };
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for short: %s",
|
|
get_op_string (op));
|
|
if (op == '+')
|
|
return e1;
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '-':
|
|
return new_short_expr (-expr_short (e1));
|
|
case '!':
|
|
return cmp_result_expr (!expr_short (e1));
|
|
case '~':
|
|
return new_short_expr (~expr_short (e1));
|
|
}
|
|
internal_error (e, "short unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_double (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
static int valid[] = { '+', '-', '!', 'C', 0 };
|
|
type_t *type;
|
|
|
|
if (!valid_op (op, valid))
|
|
return error (e1, "invalid unary operator for double: %s",
|
|
get_op_string (op));
|
|
if (op == '+')
|
|
return e1;
|
|
type = get_type (e);
|
|
if (op == 'C' && !is_int(type) && !is_float(type))
|
|
return error (e1, "invalid cast of double");
|
|
if (!is_constant (e1))
|
|
return e;
|
|
switch (op) {
|
|
case '-':
|
|
return new_double_expr (-expr_double (e1));
|
|
case '!':
|
|
print_type (get_type (e));
|
|
return cmp_result_expr (!expr_double (e1));
|
|
case 'C':
|
|
if (is_int(type)) {
|
|
return new_int_expr (expr_double (e1));
|
|
} else {
|
|
return new_float_expr (expr_double (e1));
|
|
}
|
|
}
|
|
internal_error (e, "float unary op blew up");
|
|
}
|
|
|
|
static expr_t *
|
|
uop_compound (int op, expr_t *e, expr_t *e1)
|
|
{
|
|
type_t *t1 = get_type (e1);
|
|
|
|
if (is_scalar (t1)) {
|
|
if (is_enum (t1)) {
|
|
return uop_int (op, e, e1);
|
|
}
|
|
}
|
|
return error (e1, "invalid operand for unary %s", get_op_string (op));
|
|
}
|
|
|
|
static unaryop_t do_unary_op[ev_type_count] = {
|
|
uop_invalid, // ev_void
|
|
uop_string, // ev_string
|
|
uop_float, // ev_float
|
|
uop_vector, // ev_vector
|
|
uop_entity, // ev_entity
|
|
uop_field, // ev_field
|
|
uop_func, // ev_func
|
|
uop_pointer, // ev_ptr
|
|
uop_quaternion, // ev_quaternion
|
|
uop_int, // ev_int
|
|
uop_uint, // ev_uint
|
|
uop_short, // ev_short
|
|
uop_double, // ev_double
|
|
uop_compound, // ev_invalid
|
|
};
|
|
|
|
expr_t *
|
|
fold_constants (expr_t *e)
|
|
{
|
|
int op;
|
|
expr_t *e1, *e2;
|
|
etype_t t1, t2;
|
|
|
|
if (e->type == ex_uexpr) {
|
|
e1 = e->e.expr.e1;
|
|
if (!e1) {
|
|
return e;
|
|
}
|
|
if (is_math (get_type (e1))) {
|
|
return evaluate_constexpr (e);
|
|
}
|
|
op = e->e.expr.op;
|
|
t1 = extract_type (e1);
|
|
if (t1 >= ev_type_count || !do_unary_op[t1]) {
|
|
print_expr (e);
|
|
internal_error (e, "invalid type: %d", t1);
|
|
}
|
|
return do_unary_op[t1] (op, e, e1);
|
|
} else if (e->type == ex_expr) {
|
|
e1 = e->e.expr.e1;
|
|
e2 = e->e.expr.e2;
|
|
if (!is_constant (e1) && !is_constant (e2)) {
|
|
return e;
|
|
}
|
|
|
|
if (is_math_val (e1) && is_math_val (e2)) {
|
|
return evaluate_constexpr (e);
|
|
}
|
|
|
|
op = e->e.expr.op;
|
|
|
|
t1 = extract_type (e1);
|
|
t2 = extract_type (e2);
|
|
|
|
if (t1 >= ev_type_count || t2 >= ev_type_count
|
|
|| !do_op[t1] || !do_op[t1][t2])
|
|
internal_error (e, "invalid type %d %d", t1, t2);
|
|
return do_op[t1][t2] (op, e, e1, e2);
|
|
}
|
|
return e;
|
|
}
|