diff options
Diffstat (limited to 'stdlib/integers.c')
| -rw-r--r-- | stdlib/integers.c | 652 |
1 files changed, 0 insertions, 652 deletions
diff --git a/stdlib/integers.c b/stdlib/integers.c deleted file mode 100644 index 4d5d0a80..00000000 --- a/stdlib/integers.c +++ /dev/null @@ -1,652 +0,0 @@ -// Integer type infos and methods -#include <stdio.h> // Must be before gmp.h - -#include <ctype.h> -#include <gc.h> -#include <gmp.h> -#include <stdbool.h> -#include <stdint.h> -#include <stdlib.h> - -#include "arrays.h" -#include "datatypes.h" -#include "integers.h" -#include "optionals.h" -#include "siphash.h" -#include "text.h" -#include "types.h" - -public Text_t Int$value_as_text(Int_t i) { - if (likely(i.small & 1L)) { - return Text$format("%ld", (i.small)>>2L); - } else { - char *str = mpz_get_str(NULL, 10, *i.big); - return Text$from_str(str); - } -} - -public Text_t Int$as_text(const void *i, bool colorize, const TypeInfo_t*) { - if (!i) return Text("Int"); - Text_t text = Int$value_as_text(*(Int_t*)i); - if (colorize) text = Text$concat(Text("\x1b[35m"), text, Text("\x1b[m")); - return text; -} - -static bool Int$is_none(const void *i, const TypeInfo_t*) -{ - return ((Int_t*)i)->small == 0L; -} - -public PUREFUNC int32_t Int$compare_value(const Int_t x, const Int_t y) { - if (likely(x.small & y.small & 1L)) - return (x.small > y.small) - (x.small < y.small); - else if (x.small & 1) - return -mpz_cmp_si(*y.big, x.small); - else if (y.small & 1) - return mpz_cmp_si(*x.big, y.small); - else - return x.big == y.big ? 0 : mpz_cmp(*x.big, *y.big); -} - -public PUREFUNC int32_t Int$compare(const void *x, const void *y, const TypeInfo_t*) { - return Int$compare_value(*(Int_t*)x, *(Int_t*)y); -} - -public PUREFUNC bool Int$equal_value(const Int_t x, const Int_t y) { - if (likely((x.small | y.small) & 1L)) - return x.small == y.small; - else - return x.big == y.big ? 0 : (mpz_cmp(*x.big, *y.big) == 0); -} - -public PUREFUNC bool Int$equal(const void *x, const void *y, const TypeInfo_t*) { - return Int$equal_value(*(Int_t*)x, *(Int_t*)y); -} - -public PUREFUNC uint64_t Int$hash(const void *vx, const TypeInfo_t*) { - Int_t *x = (Int_t*)vx; - if (likely(x->small & 1L)) { - return siphash24((void*)x, sizeof(Int_t)); - } else { - char *str = mpz_get_str(NULL, 16, *x->big); - return siphash24((void*)str, strlen(str)); - } -} - -public Text_t Int$format(Int_t i, Int_t digits_int) { - int64_t digits = Int64$from_int(digits_int, false); - if (likely(i.small & 1L)) { - return Text$format("%0.*ld", digits, (int64_t)((i.small)>>2L)); - } else { - char *str = mpz_get_str(NULL, 10, *i.big); - bool negative = (str[0] == '-'); - int64_t needed_zeroes = digits - (int64_t)strlen(str); - if (needed_zeroes <= 0) - return Text$from_str(str); - - char *zeroes = GC_MALLOC_ATOMIC((size_t)(needed_zeroes)); - memset(zeroes, '0', (size_t)(needed_zeroes)); - if (negative) - return Text$concat(Text("-"), Text$from_str(zeroes), Text$from_str(str + 1)); - else - return Text$concat(Text$from_str(zeroes), Text$from_str(str)); - } -} - -public Text_t Int$hex(Int_t i, Int_t digits_int, bool uppercase, bool prefix) { - if (Int$is_negative(i)) - return Text$concat(Text("-"), Int$hex(Int$negative(i), digits_int, uppercase, prefix)); - - int64_t digits = Int64$from_int(digits_int, false); - if (likely(i.small & 1L)) { - const char *hex_fmt = uppercase ? (prefix ? "0x%0.*lX" : "%0.*lX") : (prefix ? "0x%0.*lx" : "%0.*lx"); - return Text$format(hex_fmt, digits, (i.small)>>2L); - } else { - char *str = mpz_get_str(NULL, 16, *i.big); - if (uppercase) { - for (char *c = str; *c; c++) - *c = (char)toupper(*c); - } - int64_t needed_zeroes = digits - (int64_t)strlen(str); - if (needed_zeroes <= 0) - return prefix ? Text$concat(Text("0x"), Text$from_str(str)) : Text$from_str(str); - - char *zeroes = GC_MALLOC_ATOMIC((size_t)(needed_zeroes)); - memset(zeroes, '0', (size_t)(needed_zeroes)); - if (prefix) - return Text$concat(Text("0x"), Text$from_str(zeroes), Text$from_str(str)); - else - return Text$concat(Text$from_str(zeroes), Text$from_str(str)); - } -} - -public Text_t Int$octal(Int_t i, Int_t digits_int, bool prefix) { - if (Int$is_negative(i)) - return Text$concat(Text("-"), Int$octal(Int$negative(i), digits_int, prefix)); - - int64_t digits = Int64$from_int(digits_int, false); - if (likely(i.small & 1L)) { - const char *octal_fmt = prefix ? "0o%0.*lo" : "%0.*lo"; - return Text$format(octal_fmt, digits, (i.small)>>2L); - } else { - char *str = mpz_get_str(NULL, 8, *i.big); - int64_t needed_zeroes = digits - (int64_t)strlen(str); - if (needed_zeroes <= 0) - return prefix ? Text$concat(Text("0o"), Text$from_str(str)) : Text$from_str(str); - - char *zeroes = GC_MALLOC_ATOMIC((size_t)(needed_zeroes)); - memset(zeroes, '0', (size_t)(needed_zeroes)); - if (prefix) - return Text$concat(Text("0o"), Text$from_str(zeroes), Text$from_str(str)); - else - return Text$concat(Text$from_str(zeroes), Text$from_str(str)); - } -} - -public Int_t Int$slow_plus(Int_t x, Int_t y) { - mpz_t result; - mpz_init_set_int(result, x); - if (y.small & 1L) { - if (y.small < 0L) - mpz_sub_ui(result, result, (uint64_t)(-(y.small >> 2L))); - else - mpz_add_ui(result, result, (uint64_t)(y.small >> 2L)); - } else { - mpz_add(result, result, *y.big); - } - return Int$from_mpz(result); -} - -public Int_t Int$slow_minus(Int_t x, Int_t y) { - mpz_t result; - mpz_init_set_int(result, x); - if (y.small & 1L) { - if (y.small < 0L) - mpz_add_ui(result, result, (uint64_t)(-(y.small >> 2L))); - else - mpz_sub_ui(result, result, (uint64_t)(y.small >> 2L)); - } else { - mpz_sub(result, result, *y.big); - } - return Int$from_mpz(result); -} - -public Int_t Int$slow_times(Int_t x, Int_t y) { - mpz_t result; - mpz_init_set_int(result, x); - if (y.small & 1L) - mpz_mul_si(result, result, y.small >> 2L); - else - mpz_mul(result, result, *y.big); - return Int$from_mpz(result); -} - -public Int_t Int$slow_divided_by(Int_t dividend, Int_t divisor) { - // Euclidean division, see: https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/divmodnote-letter.pdf - mpz_t quotient, remainder; - mpz_init_set_int(quotient, dividend); - mpz_init_set_int(remainder, divisor); - mpz_tdiv_qr(quotient, remainder, quotient, remainder); - if (mpz_sgn(remainder) < 0) { - bool d_positive = likely(divisor.small & 1L) ? divisor.small > 0x1L : mpz_sgn(*divisor.big) > 0; - if (d_positive) - mpz_sub_ui(quotient, quotient, 1); - else - mpz_add_ui(quotient, quotient, 1); - } - return Int$from_mpz(quotient); -} - -public Int_t Int$slow_modulo(Int_t x, Int_t modulus) -{ - mpz_t result; - mpz_init_set_int(result, x); - mpz_t divisor; - mpz_init_set_int(divisor, modulus); - mpz_mod(result, result, divisor); - return Int$from_mpz(result); -} - -public Int_t Int$slow_modulo1(Int_t x, Int_t modulus) -{ - mpz_t result; - mpz_init_set_int(result, x); - mpz_sub_ui(result, result, 1); - mpz_t divisor; - mpz_init_set_int(divisor, modulus); - mpz_mod(result, result, divisor); - mpz_add_ui(result, result, 1); - return Int$from_mpz(result); -} - -public Int_t Int$slow_left_shifted(Int_t x, Int_t y) -{ - mp_bitcnt_t bits = (mp_bitcnt_t)Int64$from_int(y, false); - mpz_t result; - mpz_init_set_int(result, x); - mpz_mul_2exp(result, result, bits); - return Int$from_mpz(result); -} - -public Int_t Int$slow_right_shifted(Int_t x, Int_t y) -{ - mp_bitcnt_t bits = (mp_bitcnt_t)Int64$from_int(y, false); - mpz_t result; - mpz_init_set_int(result, x); - mpz_tdiv_q_2exp(result, result, bits); - return Int$from_mpz(result); -} - -public Int_t Int$slow_bit_and(Int_t x, Int_t y) -{ - mpz_t result; - mpz_init_set_int(result, x); - mpz_t y_mpz; - mpz_init_set_int(y_mpz, y); - mpz_and(result, result, y_mpz); - return Int$from_mpz(result); -} - -public Int_t Int$slow_bit_or(Int_t x, Int_t y) -{ - mpz_t result; - mpz_init_set_int(result, x); - mpz_t y_mpz; - mpz_init_set_int(y_mpz, y); - mpz_ior(result, result, y_mpz); - return Int$from_mpz(result); -} - -public Int_t Int$slow_bit_xor(Int_t x, Int_t y) -{ - mpz_t result; - mpz_init_set_int(result, x); - mpz_t y_mpz; - mpz_init_set_int(y_mpz, y); - mpz_xor(result, result, y_mpz); - return Int$from_mpz(result); -} - -public Int_t Int$slow_negated(Int_t x) -{ - mpz_t result; - mpz_init_set_int(result, x); - mpz_neg(result, result); - mpz_sub_ui(result, result, 1); - return Int$from_mpz(result); -} - -public Int_t Int$slow_negative(Int_t x) -{ - if (likely(x.small & 1L)) - return (Int_t){.small=4L*-((x.small)>>2L) + 1L}; - - mpz_t result; - mpz_init_set_int(result, x); - mpz_neg(result, result); - return Int$from_mpz(result); -} - -public Int_t Int$abs(Int_t x) -{ - if (likely(x.small & 1L)) - return (Int_t){.small=4L*labs((x.small)>>2L) + 1L}; - - mpz_t result; - mpz_init_set_int(result, x); - mpz_abs(result, result); - return Int$from_mpz(result); -} - -public Int_t Int$power(Int_t base, Int_t exponent) -{ - int64_t exp = Int64$from_int(exponent, false); - if (unlikely(exp < 0)) - fail("Cannot take a negative power of an integer!"); - mpz_t result; - mpz_init_set_int(result, base); - mpz_pow_ui(result, result, (uint64_t)exp); - return Int$from_mpz(result); -} - -public Int_t Int$gcd(Int_t x, Int_t y) -{ - if (likely(x.small & y.small & 0x1L)) - return I_small(Int32$gcd(x.small >> 2L, y.small >> 2L)); - - mpz_t result; - mpz_init(result); - if (x.small & 0x1L) - mpz_gcd_ui(result, *y.big, (uint64_t)labs(x.small>>2L)); - else if (y.small & 0x1L) - mpz_gcd_ui(result, *x.big, (uint64_t)labs(y.small>>2L)); - else - mpz_gcd(result, *x.big, *y.big); - return Int$from_mpz(result); -} - -public OptionalInt_t Int$sqrt(Int_t i) -{ - if (Int$compare_value(i, I(0)) < 0) - return NONE_INT; - mpz_t result; - mpz_init_set_int(result, i); - mpz_sqrt(result, result); - return Int$from_mpz(result); -} - -typedef struct { - OptionalInt_t current, last; - Int_t step; -} IntRange_t; - -static OptionalInt_t _next_int(IntRange_t *info) -{ - OptionalInt_t i = info->current; - if (!Int$is_none(&i, &Int$info)) { - Int_t next = Int$plus(i, info->step); - if (!Int$is_none(&info->last, &Int$info) && Int$compare_value(next, info->last) == Int$compare_value(info->step, I(0))) - next = NONE_INT; - info->current = next; - } - return i; -} - -public PUREFUNC Closure_t Int$to(Int_t first, Int_t last, OptionalInt_t step) { - IntRange_t *range = GC_MALLOC(sizeof(IntRange_t)); - range->current = first; - range->last = last; - range->step = Int$is_none(&step, &Int$info) ? - Int$compare_value(last, first) >= 0 ? (Int_t){.small=(1L<<2L)|1L} : (Int_t){.small=(-1L>>2L)|1L} - : step; - return (Closure_t){.fn=_next_int, .userdata=range}; -} - -public PUREFUNC Closure_t Int$onward(Int_t first, Int_t step) { - IntRange_t *range = GC_MALLOC(sizeof(IntRange_t)); - range->current = first; - range->last = NONE_INT; - range->step = step; - return (Closure_t){.fn=_next_int, .userdata=range}; -} - -public Int_t Int$from_str(const char *str) { - mpz_t i; - int result; - if (strncmp(str, "0x", 2) == 0) { - result = mpz_init_set_str(i, str + 2, 16); - } else if (strncmp(str, "0o", 2) == 0) { - result = mpz_init_set_str(i, str + 2, 8); - } else if (strncmp(str, "0b", 2) == 0) { - result = mpz_init_set_str(i, str + 2, 2); - } else { - result = mpz_init_set_str(i, str, 10); - } - if (result != 0) - return NONE_INT; - return Int$from_mpz(i); -} - -public OptionalInt_t Int$parse(Text_t text) { - return Int$from_str(Text$as_c_string(text)); -} - -public bool Int$is_prime(Int_t x, Int_t reps) -{ - mpz_t p; - mpz_init_set_int(p, x); - if (unlikely(Int$compare_value(reps, I(9999)) > 0)) - fail("Number of prime-test repetitions should not be above 9999"); - int reps_int = Int32$from_int(reps, false); - return (mpz_probab_prime_p(p, reps_int) != 0); -} - -public Int_t Int$next_prime(Int_t x) -{ - mpz_t p; - mpz_init_set_int(p, x); - mpz_nextprime(p, p); - return Int$from_mpz(p); -} - -#if __GNU_MP_VERSION >= 6 -#if __GNU_MP_VERSION_MINOR >= 3 -public Int_t Int$prev_prime(Int_t x) -{ - mpz_t p; - mpz_init_set_int(p, x); - if (unlikely(mpz_prevprime(p, p) == 0)) - fail("There is no prime number before %k", (Text_t[1]){Int$as_text(&x, false, &Int$info)}); - return Int$from_mpz(p); -} -#endif -#endif - -public Int_t Int$choose(Int_t n, Int_t k) -{ - if unlikely (Int$compare_value(n, I_small(0)) < 0) - fail("Negative inputs are not supported for choose()"); - - mpz_t ret; - mpz_init(ret); - - int64_t k_i64 = Int64$from_int(k, false); - if unlikely (k_i64 < 0) - fail("Negative inputs are not supported for choose()"); - - if likely (n.small & 1L) { - mpz_bin_uiui(ret, (unsigned long)(n.small >> 2L), (unsigned long)k_i64); - } else { - mpz_t n_mpz; - mpz_init_set_int(n_mpz, n); - mpz_bin_ui(ret, n_mpz, (unsigned long)k_i64); - } - return Int$from_mpz(ret); -} - -public Int_t Int$factorial(Int_t n) -{ - mpz_t ret; - mpz_init(ret); - int64_t n_i64 = Int64$from_int(n, false); - if unlikely (n_i64 < 0) - fail("Factorials are not defined for negative numbers"); - mpz_fac_ui(ret, (unsigned long)n_i64); - return Int$from_mpz(ret); -} - -static void Int$serialize(const void *obj, FILE *out, Table_t *pointers, const TypeInfo_t*) -{ - Int_t i = *(Int_t*)obj; - if (likely(i.small & 1L)) { - fputc(0, out); - int64_t i64 = i.small >> 2L; - Int64$serialize(&i64, out, pointers, &Int64$info); - } else { - fputc(1, out); - mpz_t n; - mpz_init_set_int(n, *(Int_t*)obj); - mpz_out_raw(out, n); - } -} - -static void Int$deserialize(FILE *in, void *obj, Array_t *pointers, const TypeInfo_t*) -{ - if (fgetc(in) == 0) { - int64_t i = 0; - Int64$deserialize(in, &i, pointers, &Int64$info); - *(Int_t*)obj = (Int_t){.small=(i<<2L) | 1L}; - } else { - mpz_t n; - mpz_init(n); - mpz_inp_raw(n, in); - *(Int_t*)obj = Int$from_mpz(n); - } -} - -public const TypeInfo_t Int$info = { - .size=sizeof(Int_t), - .align=__alignof__(Int_t), - .metamethods={ - .compare=Int$compare, - .equal=Int$equal, - .hash=Int$hash, - .as_text=Int$as_text, - .is_none=Int$is_none, - .serialize=Int$serialize, - .deserialize=Int$deserialize, - }, -}; - -public void Int64$serialize(const void *obj, FILE *out, Table_t*, const TypeInfo_t*) -{ - int64_t i = *(int64_t*)obj; - uint64_t z = (uint64_t)((i << 1L) ^ (i >> 63L)); // Zigzag encode - while (z >= 0x80L) { - fputc((uint8_t)(z | 0x80L), out); - z >>= 7L; - } - fputc((uint8_t)z, out); -} - -public void Int64$deserialize(FILE *in, void *outval, Array_t*, const TypeInfo_t*) -{ - uint64_t z = 0; - for(size_t shift = 0; ; shift += 7) { - uint8_t byte = (uint8_t)fgetc(in); - z |= ((uint64_t)(byte & 0x7F)) << shift; - if ((byte & 0x80) == 0) break; - } - *(int64_t*)outval = (int64_t)((z >> 1L) ^ -(z & 1L)); // Zigzag decode -} - -public void Int32$serialize(const void *obj, FILE *out, Table_t*, const TypeInfo_t*) -{ - int32_t i = *(int32_t*)obj; - uint32_t z = (uint32_t)((i << 1) ^ (i >> 31)); // Zigzag encode - while (z >= 0x80) { - fputc((uint8_t)(z | 0x80), out); - z >>= 7; - } - fputc((uint8_t)z, out); -} - -public void Int32$deserialize(FILE *in, void *outval, Array_t*, const TypeInfo_t*) -{ - uint32_t z = 0; - for(size_t shift = 0; ; shift += 7) { - uint8_t byte = (uint8_t)fgetc(in); - z |= ((uint32_t)(byte & 0x7F)) << shift; - if ((byte & 0x80) == 0) break; - } - *(int32_t*)outval = (int32_t)((z >> 1L) ^ -(z & 1L)); // Zigzag decode -} - -// The space savings for smaller ints are not worth having: -#define Int16$serialize NULL -#define Int16$deserialize NULL -#define Int8$serialize NULL -#define Int8$deserialize NULL - -#define DEFINE_INT_TYPE(c_type, KindOfInt, fmt, min_val, max_val, to_attr)\ - public Text_t KindOfInt ## $as_text(const void *i, bool colorize, const TypeInfo_t*) { \ - if (!i) return Text(#KindOfInt); \ - return Text$format(colorize ? "\x1b[35m" fmt "\x1b[m" : fmt, *(c_type*)i); \ - } \ - public PUREFUNC int32_t KindOfInt ## $compare(const void *x, const void *y, const TypeInfo_t*) { \ - return (*(c_type*)x > *(c_type*)y) - (*(c_type*)x < *(c_type*)y); \ - } \ - public PUREFUNC bool KindOfInt ## $equal(const void *x, const void *y, const TypeInfo_t*) { \ - return *(c_type*)x == *(c_type*)y; \ - } \ - public Text_t KindOfInt ## $format(c_type i, Int_t digits_int) { \ - return Text$format("%0*ld", Int32$from_int(digits_int, false), (int64_t)i); \ - } \ - public Text_t KindOfInt ## $hex(c_type i, Int_t digits_int, bool uppercase, bool prefix) { \ - Int_t as_int = Int$from_int64((int64_t)i); \ - return Int$hex(as_int, digits_int, uppercase, prefix); \ - } \ - public Text_t KindOfInt ## $octal(c_type i, Int_t digits_int, bool prefix) { \ - Int_t as_int = Int$from_int64((int64_t)i); \ - return Int$octal(as_int, digits_int, prefix); \ - } \ - public Array_t KindOfInt ## $bits(c_type x) { \ - Array_t bit_array = (Array_t){.data=GC_MALLOC_ATOMIC(sizeof(bool[8*sizeof(c_type)])), .atomic=1, .stride=sizeof(bool), .length=8*sizeof(c_type)}; \ - bool *bits = bit_array.data + sizeof(c_type)*8; \ - for (size_t i = 0; i < 8*sizeof(c_type); i++) { \ - *(bits--) = x & 1; \ - x >>= 1; \ - } \ - return bit_array; \ - } \ - typedef struct { \ - Optional##KindOfInt##_t current, last; \ - KindOfInt##_t step; \ - } KindOfInt##Range_t; \ - static Optional##KindOfInt##_t _next_##KindOfInt(KindOfInt##Range_t *info) \ - { \ - Optional##KindOfInt##_t i = info->current; \ - if (!i.is_none) { \ - KindOfInt##_t next; bool overflow = __builtin_add_overflow(i.i, info->step, &next); \ - if (overflow || (!info->last.is_none && (info->step >= 0 ? next > info->last.i : next < info->last.i))) \ - info->current = (Optional##KindOfInt##_t){.is_none=true}; \ - else \ - info->current = (Optional##KindOfInt##_t){.i=next}; \ - } \ - return i; \ - } \ - public to_attr Closure_t KindOfInt ## $to(c_type first, c_type last, Optional ## KindOfInt ## _t step) { \ - KindOfInt##Range_t *range = GC_MALLOC(sizeof(KindOfInt##Range_t)); \ - range->current = (Optional##KindOfInt##_t){.i=first}; \ - range->last = (Optional##KindOfInt##_t){.i=last}; \ - range->step = step.is_none ? (last >= first ? 1 : -1) : step.i; \ - return (Closure_t){.fn=_next_##KindOfInt, .userdata=range}; \ - } \ - public to_attr Closure_t KindOfInt ## $onward(c_type first, c_type step) { \ - KindOfInt##Range_t *range = GC_MALLOC(sizeof(KindOfInt##Range_t)); \ - range->current = (Optional##KindOfInt##_t){.i=first}; \ - range->last = (Optional##KindOfInt##_t){.is_none=true}; \ - range->step = step; \ - return (Closure_t){.fn=_next_##KindOfInt, .userdata=range}; \ - } \ - public PUREFUNC Optional ## KindOfInt ## _t KindOfInt ## $parse(Text_t text) { \ - OptionalInt_t full_int = Int$parse(text); \ - if (full_int.small == 0L) return (Optional ## KindOfInt ## _t){.is_none=true}; \ - if (Int$compare_value(full_int, I(min_val)) < 0) { \ - return (Optional ## KindOfInt ## _t){.is_none=true}; \ - } \ - if (Int$compare_value(full_int, I(max_val)) > 0) { \ - return (Optional ## KindOfInt ## _t){.is_none=true}; \ - } \ - return (Optional ## KindOfInt ## _t){.i=KindOfInt##$from_int(full_int, true)}; \ - } \ - public CONSTFUNC c_type KindOfInt ## $gcd(c_type x, c_type y) { \ - if (x == 0 || y == 0) return 0; \ - x = KindOfInt##$abs(x); \ - y = KindOfInt##$abs(y); \ - while (x != y) { \ - if (x > y) x -= y; \ - else y -= x; \ - } \ - return x; \ - } \ - public const c_type KindOfInt##$min = min_val; \ - public const c_type KindOfInt##$max = max_val; \ - public const TypeInfo_t KindOfInt##$info = { \ - .size=sizeof(c_type), \ - .align=__alignof__(c_type), \ - .metamethods={ \ - .compare=KindOfInt##$compare, \ - .as_text=KindOfInt##$as_text, \ - .serialize=KindOfInt##$serialize, \ - .deserialize=KindOfInt##$deserialize, \ - }, \ - }; - -DEFINE_INT_TYPE(int64_t, Int64, "%ld", INT64_MIN, INT64_MAX, __attribute__(())) -DEFINE_INT_TYPE(int32_t, Int32, "%d", INT32_MIN, INT32_MAX, CONSTFUNC) -DEFINE_INT_TYPE(int16_t, Int16, "%d", INT16_MIN, INT16_MAX, CONSTFUNC) -DEFINE_INT_TYPE(int8_t, Int8, "%d", INT8_MIN, INT8_MAX, CONSTFUNC) -#undef DEFINE_INT_TYPE - -// vim: ts=4 sw=0 et cino=L2,l1,(0,W4,m1,\:0 |
