4 * Declarations for Target Values.
10 Discussion of new interface, proposals by Prof. Waite:
12 > 1. You say that you plan to replace the tv module. That replacement is
13 > absolutely essential for an ANSI C translator: Section 6.1.3.2 of the
14 > standard says that the representation of an integer_constant depends
15 > upon its value as well as any suffixes that are attached to it. The
16 > possible Firm modes for such a constant are i, I, l, and L. The
17 > current tv module provides only one integer conversion routine, and
18 > that requires conversion by the client. Since the type of the value
19 > argument is long, this may preclude the representation of an unsigned
22 > There is a similar problem with floating constants. Floating
23 > constants can be suffixed in C, and the mode depends upon the suffix.
24 > It can indicate that the constant is of type long double, which your
25 > current tv module is incapable of representing.
27 > Your tv module interface accepts two kinds of information: modes and
28 > values. Values obtained from the program text might be uninterpreted
29 > strings, strings interpreted as integers, and strings interpreted as
30 > reals. Values provided by the compiler are usually integers. Modes are
31 > always Firm modes. It seems to me that the tv module should provide
32 > tarval* constructors for three of the four kinds of values. Each of these
33 > constructors should have an ir_mode parameter and one or more parameters
34 > appropriate for the kind of value. As is currently the case, one
35 > constructor should be provided for both compiler-generated integers and
36 > source strings interpreted as integers. (This avoids problems of
37 > different conversion radices -- the client does the conversion.) For
38 > symmetry, the constructor for source strings interpreted as reals should
39 > accept a long double parameter and require the client to do the
49 # include "irnode.h" /* for pnc_number enum */
54 * tv -- TargetValue, short tarval.
55 * Internal representation for machine values.
58 * Christian von Roques
62 * Tarvals represent target machine values. They are typed by modes.
63 * Tarvals only represent values of mode_sort:
70 * In case of references the module accepts an entity to represent the
72 * Furthermore, computations and conversions of these values can
76 * This module is closely related to the irmode module, as the modes
77 * defined there are thoroughly used throughout the whole module.
78 * Also, the comparison functions rely on the definition of comparison
79 * values in the irnode module.
82 * The original tv module originated in the fiasco compiler written ...
83 * This is the new version, described in the tech report 1999-14 by ...
87 * irmode.h for the modes definitions
88 * irnode.h for the pnc_numbers table
90 * tarval_init1 and tarval_init2 for initialization of the
95 #ifndef _TARVAL_TYPEDEF_
96 #define _TARVAL_TYPEDEF_
97 typedef struct tarval tarval;
100 /* ************************ Constructors for tarvals ************************ */
103 * Constructor function for new tarvals.
105 * @param str The string representing the target value
106 * @param len The length of the string
107 * @param mode The mode requested for the result tarval
109 * This function creates a new tarval representing the value represented
110 * by a CString, aka char array. If a tarval representing this value already
111 * exists, this tarval is returned instead of a new one. So tarvals are
112 * directly comparable since their representation is unique.
114 * This function accepts the following strings:
116 * if mode is int_number:
117 * - 0(x|X)[0-9a-fA-F]+ (hexadecimal representation)
118 * - 0[0-7]* (octal representation)
119 * - (+|-)?[1-9][0-9]* (decimal representation)
121 * if mode if float_number:
122 * - (+|-)?(decimal int) (. (decimal int))? ((e|E)(+|-)?(decimal int))?
124 * if mode is boolean: true, True, TRUE ... False... 0, 1,
126 * if mode is reference: hexadecimal of decimal number as int
128 * if mode is character: hex or dec
130 * Leading and/or trailing spaces are ignored
133 * A tarval of proper type representing the requested value is returned.
134 * Tarvals are unique, so for any value/mode pair at most one tarval will
135 * exist, which will be returned upon further requests with an identical
139 * If the string is not representable in the given mode an assertion is
140 * thrown in assert build.
143 * irmode.h for predefined modes
144 * new_tarval_from_long()
145 * new_tarval_from_double()
147 tarval *new_tarval_from_str(const char *str, size_t len, ir_mode *mode);
150 * Constructor function for new tarvals
152 * @param l The long representing the value
153 * @param mode The mode requested for the result tarval
155 * This function creates a new tarval representing the value represented
156 * by a long integer. If a tarval representing this value already exists,
157 * this tarval is returned instead of a new one. So tarvals are directly
158 * comparable since their representation is unique.
161 * A tarval of proper type representing the requested value is returned.
162 * Tarvals are unique, so for any value/mode pair at most one tarval will
163 * exist, which will be returned upon further requests with an identical
167 * If the long is not representable in the given mode an assertion is
168 * thrown in assert build.
171 * irmode.h for predefined modes
172 * new_tarval_from_str()
173 * new_tarval_from_double()
176 tarval *new_tarval_from_long(long l, ir_mode *mode);
179 * This returns a long int with the value represented value, or
180 * gibberish, depending on the size of long int and the size of the
181 * stored value. It works for e.g. 1 as mode_Ls, but might not work for
182 * get_mode_max(mode_Ls).
183 * This will overflow silently, so use only if you know what
184 * you are doing! (better check with tarval_is_long()...)
186 long tarval_to_long(tarval *tv);
189 * This validates if tarval_to_long() will return a satisfying
190 * result. I.e. if tv is an int_number and between min, max
191 * of long int (signed!)
193 int tarval_is_long(tarval *tv);
196 * Constructor function for new tarvals.
198 * @param d The long double representing the value
199 * @param mode The mode requested for the result tarval
201 * This function creates a new tarval representing the value represented
202 * by a long double. If a tarval representing this value already exists,
203 * this tarval is returned instead of a new one. So tarvals are directly
204 * comparable since their representation is unique.
205 * Only modes of sort float_number can be constructed this way.
208 * A tarval of proper type representing the requested value is returned.
209 * Tarvals are unique, so for any value/mode pair at most one tarval will
210 * exist, which will be returned upon further requests with an identical
214 * If the long double is not representable in the given mode an assertion
215 * is thrown. This will happen for any mode not of sort float_number.
218 * irmode.h for predefined values
219 * new_tarval_from_str()
220 * new_tarval_from_long()
222 tarval *new_tarval_from_double(long double d, ir_mode *mode);
225 * This returns a double with the value represented value, or
226 * gibberish, depending on the size of double and the size of the
228 * This will overflow silently, so use only if you know what
229 * you are doing! (better check with tarval_is_long...)
231 long double tarval_to_double(tarval *tv);
234 * This validates if tarval_to_double() will return a satisfying
235 * result. I.e. if tv is an float_number and between min, max
238 int tarval_is_double(tarval *tv);
241 * Construct a tarval that represents the address of the entity.
243 * The address must be constant, the entity must have as owner the global type.
245 tarval *new_tarval_from_entity (entity *ent, ir_mode *mode);
248 * Returns the associated entity of a tarval.
250 entity *tarval_to_entity(tarval *tv);
253 * Returns non-zero if a the given tarval represents an entity.
255 int tarval_is_entity(tarval *tv);
257 /** ********** Access routines for tarval fields ********** **/
265 * ir_mode *get_tarval_mode(tarval *tv)
269 * These are access function for tarval struct members. It is encouraged
270 * to use them instead of direct access to the struct fields.
273 * tv - The tarval to access fields of
276 * get_tv_mode: The mode of the tarval
282 /** Returns the mode of the tarval. */
284 /* Testing properties of the represented values */
286 /** Returns 0 if tv is positive, else > 0.
291 int tarval_is_negative(tarval *a);
293 /** The 'bad' tarval. */
294 extern tarval *tarval_bad;
295 /** Returns the 'bad tarval. */
296 tarval *get_tarval_bad(void);
298 /** The 'undefined' tarval. */
299 extern tarval *tarval_undefined;
300 /** Returns the 'undefined' tarval. */
301 tarval *get_tarval_undefined(void);
303 /** The mode_b tarval 'false'. */
304 extern tarval *tarval_b_false;
305 /** Returns the mode_b tarval 'false'. */
306 tarval *get_tarval_b_false(void);
308 /** The mode_b tarval 'true'. */
309 extern tarval *tarval_b_true;
310 /** Returns the mode_b tarval 'true'. */
311 tarval *get_tarval_b_true(void);
313 /** The 'void' pointer tarval. */
314 extern tarval *tarval_P_void;
315 /** Returns the 'void' pointer tarval. */
316 tarval *get_tarval_P_void(void);
318 /* These functions calculate and return a tarval representing the requested
320 * The functions get_mode_{Max,Min,...} return tarvals retrieved from these
321 * functions, but these are stored on initialization of the irmode module and
322 * therefore the irmode functions should be prefered to the functions below. */
324 /** Returns the maximum value of a given mode. */
325 tarval *get_tarval_max(ir_mode *mode);
327 /** Returns the minimum value of a given mode. */
328 tarval *get_tarval_min(ir_mode *mode);
330 /** Returns the 0 value (additive neutral) of a given mode. */
331 tarval *get_tarval_null(ir_mode *mode);
333 /** Returns the 1 value (multiplicative neutral) of a given mode. */
334 tarval *get_tarval_one(ir_mode *mode);
336 /** Return quite nan for float_number modes. */
337 tarval *get_tarval_nan(ir_mode *mode);
339 /** Return +inf for float_number modes. */
340 tarval *get_tarval_inf(ir_mode *mode);
342 /* ******************** Arithmethic operations on tarvals ******************** */
345 * Compares two tarvals
347 * Compare a with b and return a pnc_number describing the relation
348 * between a and b. This is either Uo, Lt, Eq, Gt, or False if a or b
349 * are symbolic pointers which can not be compared at all.
351 * @param a A tarval to be compared
352 * @param b A tarval to be compared
355 * The pnc_number best describing the relation between a and b is returned.
356 * This means the mode with the least bits set is returned, e.g. if the
357 * tarvals are equal the pnc_number 'Eq' is returned, not 'Ge' which
358 * indicates 'greater or equal'
361 * irnode.h for the definition of pnc_numbers
363 pnc_number tarval_cmp(tarval *a, tarval *b);
366 * Converts a tarval to another mode.
368 * Convert tarval 'src' to mode 'mode', this will suceed if and only if mode
369 * 'mode' is wider than the mode of src, as defined in the firm documentation
370 * and as returned by the function mode_is_smaller defined in irmode.h.
372 * @param src The tarval to convert
373 * @param mode Tho mode to convert to
376 * If a tarval of mode 'mode' with the result of the conversion of the 'src'
377 * tarvals value already exists, it will be returned, else a new tarval is
378 * constructed and returned
381 * Illegal conversations will trigger an assertion
384 * FIRM documentation for conversion rules
385 * mode_is_smaller defined in irmode.h
387 tarval *tarval_convert_to(tarval *src, ir_mode *m);
390 * These function implement basic computations representable as opcodes
396 * a - the tarval to operate on
399 * a - the first operand tarval
400 * b - the second operand tarval
403 * If neccessary a new tarval is constructed for the resulting value,
404 * or the one already carrying the computation result is retrieved and
405 * returned as result.
408 * The order the arguments are given in is important, imagine postfix
410 * Illegal operations will trigger an assertion.
411 * The sort member of the struct mode defines which operations are valid
414 /** Negation of a tarval. */
415 tarval *tarval_neg(tarval *a);
417 /** Addition of two tarvals. */
418 tarval *tarval_add(tarval *a, tarval *b);
420 /** Subtraction from a tarval. */
421 tarval *tarval_sub(tarval *a, tarval *b);
423 /** Multiplication of tarvals. */
424 tarval *tarval_mul(tarval *a, tarval *b);
426 /** 'Exact' division. */
427 tarval *tarval_quo(tarval *a, tarval *b);
429 /** Integer division. */
430 tarval *tarval_div(tarval *a, tarval *b);
432 /** Remainder of integer division. */
433 tarval *tarval_mod(tarval *a, tarval *b);
435 /** Absolute value. */
436 tarval *tarval_abs(tarval *a);
439 tarval *tarval_and(tarval *a, tarval *b);
442 tarval *tarval_or (tarval *a, tarval *b);
444 /** Bitwise exclusive or. */
445 tarval *tarval_eor(tarval *a, tarval *b);
448 tarval *tarval_shl(tarval *a, tarval *b);
450 /** Unsigned (logical) right shift. */
451 tarval *tarval_shr(tarval *a, tarval *b);
453 /** Signed (arithmetic) right shift. */
454 tarval *tarval_shrs(tarval *a, tarval *b);
457 tarval *tarval_rot(tarval *a, tarval *b);
459 /* *********** Output of tarvals *********** */
462 * Returns Bit representation of a tarval value, as string of '0' and '1'
464 * @param tv The tarval
466 * This function returns a printable bit representation of any value
467 * stored as tarval. This representation is a null terminated C string.
470 * As usual in C a pointer to a char is returned. The length of the
471 * returned string if fixed, just read as many chars as the mode defines
475 * The string is allocated using malloc() and is free()ed on the next call
477 * The string consists of the ascii characters '0' and '1' and is
481 * irmode.h for the definition of the ir_mode struct
482 * the size member of aforementioned struct
484 char *tarval_bitpattern(tarval *tv);
487 * Returns bitpattern [from, to[.
489 char *tarval_sub_bitpattern(tarval *tv, int from, int to);
492 * Returns the bitpattern of the bytes_ofs byte.
494 * This function succeeds even if the mode of the tarval uses lesser bits
495 * than requested, in that case the bitpattern is filled with zero bits.
497 * To query a 32bit value the following code can be used:
499 * val0 = tarval_sub_bits(tv, 0);
500 * val1 = tarval_sub_bits(tv, 1);
501 * val2 = tarval_sub_bits(tv, 2);
502 * val3 = tarval_sub_bits(tv, 3);
504 * Because this is the bit representation of the target machine, only the following
505 * operations are legal on the result:
507 * - concatenation (endian dependance MUST be handled by the CALLER)
508 * - bitwise logical operations to select/mask bits
510 * @param tv the tarval
511 * @param byte_ofs the byte offset
514 * The result of this funcion is undefined if the mode is neither integer nor float.
516 unsigned char tarval_sub_bits(tarval *tv, unsigned byte_ofs);
519 * Identifying some tarvals ???
522 * - 0 for additive neutral,
523 * - +1 for multiplicative neutral,
524 * - -1 for bitwise-and neutral
528 * This function is deprecated and its use strongly discouraged.
529 * Implemented for completeness.
531 long tarval_classify(tarval *tv);
534 * Initialization of the tarval module.
536 * Call before init_mode().
538 void init_tarval_1(void);
541 * Initialization of the tarval module.
543 * Call after init_mode().
545 void init_tarval_2(void);
547 typedef int printf_func (void* , const char *, ...) ;
548 int tarval_xprintf(printf_func , void * , tarval *);