/
niceSOFT
/
mpfr
Обзор
Документация
Войти
/
niceSOFT
/
mpfr
Код
Задачи
Вики
Пакеты
0
Релизы
0
Аналитика
Безопасность
bug_compound-root
src/mpfr-impl.h
2 765 строк
124 KB
Vincent Lefevre
[src/mpfr-impl.h] Updated FIXME comment for decimal_point/thousands_sep.
31 янв 2023, 18:47
31 янв 2023, 18:47
5c77e28
Код
Авторство
О чём код?
/* Utilities for MPFR developers, not exported. Copyright 1999-2023 Free Software Foundation, Inc. Contributed by the AriC and Caramba projects, INRIA. This file is part of the GNU MPFR Library. The GNU MPFR Library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. The GNU MPFR Library 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 Lesser General Public License for more details. You should have received a copy of the GNU Lesser General Public License along with the GNU MPFR Library; see the file COPYING.LESSER. If not, see https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. */ #ifndef __MPFR_IMPL_H__ #define __MPFR_IMPL_H__ /* Include config.h before using ANY configure macros if needed. */ #ifdef HAVE_CONFIG_H # include "config.h" #endif /****************************************************** ***************** Standard headers ***************** ******************************************************/ /* Let's include some standard headers unconditionally as they are already needed by several source files or when some options are enabled/disabled, and it is easy to forget them (some configure options may hide the error). Note: If some source file must not have such a header included (which is very unlikely and probably means something broken in this source file), we should do that with some macro (that would also force to disable incompatible features). */ #if defined (__cplusplus) # include <cstdio> # include <cstring> #else # include <stdio.h> # include <string.h> #endif /* Since <stdio.h> (<cstdio> for C++) is unconditionally included... */ #define MPFR_USE_FILE #include <stdlib.h> #include <limits.h> #include <float.h> /* for FLT_RADIX, etc., tested below */ /****************************************************** ***************** Include files ******************** ******************************************************/ /* The macros defined in mpfr-cvers.h do not depend on anything, so that it is better to include this header file early: then it can be used by any other header. */ #include "mpfr-cvers.h" #if defined(_MPFR_EXP_FORMAT) && _MPFR_EXP_FORMAT == 4 /* mpfr_exp_t will be defined as intmax_t */ # undef MPFR_NEED_INTMAX_H # define MPFR_NEED_INTMAX_H 1 #endif #ifdef MPFR_NEED_INTMAX_H # include "mpfr-intmax.h" #endif /* Check if we are inside a build of MPFR or inside the test suite. This is needed in mpfr.h to export or import the functions. It matters only for Windows DLL */ #ifndef __MPFR_TEST_H__ # define __MPFR_WITHIN_MPFR 1 #endif /* For the definition of MPFR_THREAD_ATTR. GCC/ICC detection macros are no longer used, as they sometimes gave incorrect information about the support of thread-local variables. A configure check is now done. */ #if defined(MPFR_WANT_SHARED_CACHE) # define MPFR_NEED_THREAD_LOCK 1 #endif #include "mpfr-thread.h" #ifndef MPFR_USE_MINI_GMP #include "gmp.h" #else #include "mini-gmp.h" #endif /* With the current code, MPFR_LONG_WITHIN_LIMB must be defined if an unsigned long fits in a limb. Since one cannot rely on the configure tests entirely (in particular when GMP is involved) and some platforms may not use configure, handle this definition here. A limb (mp_limb_t) is normally defined as an unsigned long, but this may not be the case with mini-gmp (and we can't rely on __GMP_SHORT_LIMB for this reason). So, concerning mp_limb_t, we can only test GMP_NUMB_BITS. Chosen heuristic: define MPFR_LONG_WITHIN_LIMB only when * mp_limb_t and unsigned long have both 32 bits exactly, or * mp_limb_t has at least 64 bits. Since we require that mp_limb_t have a size that is a power of 2, we can currently be wrong only if mini-gmp is used and unsigned long has more than 64 bits, which is unlikely to occur. */ #if GMP_NUMB_BITS >= 64 || (GMP_NUMB_BITS == 32 && ULONG_MAX == 0xffffffff) # undef MPFR_LONG_WITHIN_LIMB # define MPFR_LONG_WITHIN_LIMB 1 #endif #ifdef MPFR_HAVE_GMP_IMPL /* Build with gmp internals */ # ifdef MPFR_USE_MINI_GMP # error "MPFR_HAVE_GMP_IMPL and MPFR_USE_MINI_GMP must not be both defined" # endif # include "gmp-impl.h" # ifdef MPFR_NEED_LONGLONG_H # include "longlong.h" # endif # include "mpfr.h" # include "mpfr-gmp.h" #else /* Build without gmp internals */ /* if using mini-gmp, include missing definitions in mini-gmp */ # ifdef MPFR_USE_MINI_GMP # include "mpfr-mini-gmp.h" # endif # include "mpfr.h" # include "mpfr-gmp.h" # ifdef MPFR_LONG_WITHIN_LIMB /* longlong.h is not valid otherwise */ # ifdef MPFR_NEED_LONGLONG_H # define LONGLONG_STANDALONE # include "mpfr-longlong.h" # endif # endif #endif #undef MPFR_NEED_LONGLONG_H /****************************************************** ************* Attribute definitions **************** ******************************************************/ #if defined(MPFR_HAVE_NORETURN) /* _Noreturn is specified by ISO C11 (Section 6.7.4); in GCC, it is supported as of version 4.7. */ # define MPFR_NORETURN _Noreturn #elif !defined(noreturn) /* A noreturn macro could be defined if <stdnoreturn.h> has been included, in which case it would make sense to #define MPFR_NORETURN noreturn. But this is unlikely, as MPFR_HAVE_NORETURN should have been defined in such a case. So, in doubt, let us avoid any code that would use a noreturn macro, since it could be invalid. */ # if __MPFR_GNUC(3,0) || __MPFR_ICC(8,1,0) # define MPFR_NORETURN __attribute__ ((noreturn)) # elif defined(_MSC_VER) && defined(_WIN32) && (_MSC_VER >= 1200) # define MPFR_NORETURN __declspec (noreturn) # endif #endif #ifndef MPFR_NORETURN # define MPFR_NORETURN #endif #if __MPFR_GNUC(3,0) || __MPFR_ICC(8,1,0) # define MPFR_CONST_FUNCTION_ATTR __attribute__ ((const)) #else # define MPFR_CONST_FUNCTION_ATTR #endif #if __MPFR_GNUC(3,0) || __MPFR_ICC(8,1,0) # define MPFR_PURE_FUNCTION_ATTR __attribute__ ((pure)) #else # define MPFR_PURE_FUNCTION_ATTR #endif /* The hot attribute on a function is used to inform the compiler that the function is a hot spot of the compiled program. */ #if __MPFR_GNUC(4,3) # define MPFR_HOT_FUNCTION_ATTR __attribute__ ((hot)) #else # define MPFR_HOT_FUNCTION_ATTR #endif /* The cold attribute on functions is used to inform the compiler that the function is unlikely to be executed. */ #if __MPFR_GNUC(4,3) # define MPFR_COLD_FUNCTION_ATTR __attribute__ ((cold)) #else # define MPFR_COLD_FUNCTION_ATTR #endif /* Add MPFR_MAYBE_UNUSED after a variable declaration to avoid compiler warnings if it is not used. TODO: To be replaced by the future maybe_unused attribute (C2x) once supported. */ #if __MPFR_GNUC(3,4) #define MPFR_MAYBE_UNUSED __attribute__ ((unused)) #else #define MPFR_MAYBE_UNUSED #endif /* This MPFR_FALLTHROUGH macro allows one to make fallthrough in switch case explicit. Use this macro at the end of a switch case if it falls through, in order to avoid a -Wimplicit-fallthrough warning. */ #if __MPFR_GNUC(7,0) #define MPFR_FALLTHROUGH __attribute__ ((fallthrough)) #else #define MPFR_FALLTHROUGH #endif /****************************************************** *** Global internal variables and related macros *** ******************************************************/ #if defined (__cplusplus) extern "C" { #endif #if defined(MPFR_WANT_SHARED_CACHE) # define MPFR_CACHE_ATTR #else # define MPFR_CACHE_ATTR MPFR_THREAD_ATTR #endif /* Note: The following structure and types depend on the MPFR build options (including compiler options), due to the various locking methods affecting MPFR_DEFERRED_INIT_SLAVE_DECL and MPFR_LOCK_DECL. But since this is only internal, that's OK. */ struct __gmpfr_cache_s { mpfr_t x; int inexact; int (*func)(mpfr_ptr, mpfr_rnd_t); MPFR_DEFERRED_INIT_SLAVE_DECL() MPFR_LOCK_DECL(lock) }; typedef struct __gmpfr_cache_s mpfr_cache_t[1]; typedef struct __gmpfr_cache_s *mpfr_cache_ptr; #if __GMP_LIBGMP_DLL # define MPFR_WIN_THREAD_SAFE_DLL 1 #endif #if defined(__MPFR_WITHIN_MPFR) || !defined(MPFR_WIN_THREAD_SAFE_DLL) extern MPFR_THREAD_ATTR mpfr_flags_t __gmpfr_flags; extern MPFR_THREAD_ATTR mpfr_exp_t __gmpfr_emin; extern MPFR_THREAD_ATTR mpfr_exp_t __gmpfr_emax; extern MPFR_THREAD_ATTR mpfr_prec_t __gmpfr_default_fp_bit_precision; extern MPFR_THREAD_ATTR mpfr_rnd_t __gmpfr_default_rounding_mode; extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_cache_const_euler; extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_cache_const_catalan; # ifndef MPFR_USE_LOGGING extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_cache_const_pi; extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_cache_const_log2; # else /* Two constants are used by the logging functions (via mpfr_fprintf, then mpfr_log, for the base conversion): pi and log(2). Since the mpfr_cache function isn't re-entrant when working on the same cache, we need to define two caches for each constant. */ extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_normal_pi; extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_normal_log2; extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_logging_pi; extern MPFR_CACHE_ATTR mpfr_cache_t __gmpfr_logging_log2; extern MPFR_CACHE_ATTR mpfr_cache_ptr __gmpfr_cache_const_pi; extern MPFR_CACHE_ATTR mpfr_cache_ptr __gmpfr_cache_const_log2; # endif #endif #ifdef MPFR_WIN_THREAD_SAFE_DLL # define MPFR_MAKE_VARFCT(T,N) T * N ## _f (void) { return &N; } __MPFR_DECLSPEC mpfr_flags_t * __gmpfr_flags_f (void); __MPFR_DECLSPEC mpfr_exp_t * __gmpfr_emin_f (void); __MPFR_DECLSPEC mpfr_exp_t * __gmpfr_emax_f (void); __MPFR_DECLSPEC mpfr_prec_t * __gmpfr_default_fp_bit_precision_f (void); __MPFR_DECLSPEC mpfr_rnd_t * __gmpfr_default_rounding_mode_f (void); __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_cache_const_euler_f (void); __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_cache_const_catalan_f (void); # ifndef MPFR_USE_LOGGING __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_cache_const_pi_f (void); __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_cache_const_log2_f (void); # else __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_normal_pi_f (void); __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_normal_log2_f (void); __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_logging_pi_f (void); __MPFR_DECLSPEC mpfr_cache_t * __gmpfr_logging_log2_f (void); __MPFR_DECLSPEC mpfr_cache_ptr * __gmpfr_cache_const_pi_f (void); __MPFR_DECLSPEC mpfr_cache_ptr * __gmpfr_cache_const_log2_f (void); # endif # ifndef __MPFR_WITHIN_MPFR # define __gmpfr_flags (*__gmpfr_flags_f()) # define __gmpfr_emin (*__gmpfr_emin_f()) # define __gmpfr_emax (*__gmpfr_emax_f()) # define __gmpfr_default_fp_bit_precision (*__gmpfr_default_fp_bit_precision_f()) # define __gmpfr_default_rounding_mode (*__gmpfr_default_rounding_mode_f()) # define __gmpfr_cache_const_euler (*__gmpfr_cache_const_euler_f()) # define __gmpfr_cache_const_catalan (*__gmpfr_cache_const_catalan_f()) # ifndef MPFR_USE_LOGGING # define __gmpfr_cache_const_pi (*__gmpfr_cache_const_pi_f()) # define __gmpfr_cache_const_log2 (*__gmpfr_cache_const_log2_f()) # else # define __gmpfr_normal_pi (*__gmpfr_normal_pi_f()) # define __gmpfr_logging_pi (*__gmpfr_logging_pi_f()) # define __gmpfr_logging_log2 (*__gmpfr_logging_log2_f()) # define __gmpfr_cache_const_pi (*__gmpfr_cache_const_pi_f()) # define __gmpfr_cache_const_log2 (*__gmpfr_cache_const_log2_f()) # endif # endif #else # define MPFR_MAKE_VARFCT(T,N) #endif # define MPFR_THREAD_VAR(T,N,V) \ MPFR_THREAD_ATTR T N = (V); \ MPFR_MAKE_VARFCT (T,N) #define BASE_MAX 62 __MPFR_DECLSPEC extern const __mpfr_struct __gmpfr_l2b[BASE_MAX-1][2]; /* Note: do not use the following values when they can be outside the current exponent range, e.g. when the exponent range has not been extended yet; under such a condition, they can be used only in mpfr_cmpabs. */ __MPFR_DECLSPEC extern const mpfr_t __gmpfr_one; __MPFR_DECLSPEC extern const mpfr_t __gmpfr_two; __MPFR_DECLSPEC extern const mpfr_t __gmpfr_four; __MPFR_DECLSPEC extern const mpfr_t __gmpfr_mone; __MPFR_DECLSPEC extern const mpfr_t __gmpfr_const_log2_RNDD; __MPFR_DECLSPEC extern const mpfr_t __gmpfr_const_log2_RNDU; #if defined (__cplusplus) } #endif /* Replace some common functions for direct access to the global vars. The casts prevent these macros from being used as a lvalue (and this method makes sure that the expressions have the correct type). */ #define mpfr_get_emin() ((mpfr_exp_t) __gmpfr_emin) #define mpfr_get_emax() ((mpfr_exp_t) __gmpfr_emax) #define mpfr_get_default_rounding_mode() \ ((mpfr_rnd_t) __gmpfr_default_rounding_mode) #define mpfr_get_default_prec() \ ((mpfr_prec_t) __gmpfr_default_fp_bit_precision) /* Flags related macros. */ /* Note: Function-like macros that modify __gmpfr_flags are not defined because of the risk to break the sequence point rules if two such macros are used in the same expression (without a sequence point between). For instance, mpfr_sgn currently uses mpfr_set_erangeflag, which mustn't be implemented as a macro for this reason. */ #define mpfr_flags_test(mask) \ (__gmpfr_flags & (mpfr_flags_t) (mask)) #if MPFR_FLAGS_ALL <= INT_MAX #define mpfr_underflow_p() \ ((int) (__gmpfr_flags & MPFR_FLAGS_UNDERFLOW)) #define mpfr_overflow_p() \ ((int) (__gmpfr_flags & MPFR_FLAGS_OVERFLOW)) #define mpfr_nanflag_p() \ ((int) (__gmpfr_flags & MPFR_FLAGS_NAN)) #define mpfr_inexflag_p() \ ((int) (__gmpfr_flags & MPFR_FLAGS_INEXACT)) #define mpfr_erangeflag_p() \ ((int) (__gmpfr_flags & MPFR_FLAGS_ERANGE)) #define mpfr_divby0_p() \ ((int) (__gmpfr_flags & MPFR_FLAGS_DIVBY0)) #endif /* Use a do-while statement for the following macros in order to prevent one from using them in an expression, as the sequence point rules could be broken if __gmpfr_flags is assigned twice in the same expression (via macro expansions). For instance, the mpfr_sgn macro currently uses mpfr_set_erangeflag, which mustn't be implemented as a function-like macro for this reason. It is not clear whether an expression with sequence points, like (void) (0, __gmpfr_flags = 0), would avoid UB. */ #define MPFR_CLEAR_FLAGS() \ do __gmpfr_flags = 0; while (0) #define MPFR_CLEAR_UNDERFLOW() \ do __gmpfr_flags &= MPFR_FLAGS_ALL ^ MPFR_FLAGS_UNDERFLOW; while (0) #define MPFR_CLEAR_OVERFLOW() \ do __gmpfr_flags &= MPFR_FLAGS_ALL ^ MPFR_FLAGS_OVERFLOW; while (0) #define MPFR_CLEAR_DIVBY0() \ do __gmpfr_flags &= MPFR_FLAGS_ALL ^ MPFR_FLAGS_DIVBY0; while (0) #define MPFR_CLEAR_NANFLAG() \ do __gmpfr_flags &= MPFR_FLAGS_ALL ^ MPFR_FLAGS_NAN; while (0) #define MPFR_CLEAR_INEXFLAG() \ do __gmpfr_flags &= MPFR_FLAGS_ALL ^ MPFR_FLAGS_INEXACT; while (0) #define MPFR_CLEAR_ERANGEFLAG() \ do __gmpfr_flags &= MPFR_FLAGS_ALL ^ MPFR_FLAGS_ERANGE; while (0) #define MPFR_SET_UNDERFLOW() \ do __gmpfr_flags |= MPFR_FLAGS_UNDERFLOW; while (0) #define MPFR_SET_OVERFLOW() \ do __gmpfr_flags |= MPFR_FLAGS_OVERFLOW; while (0) #define MPFR_SET_DIVBY0() \ do __gmpfr_flags |= MPFR_FLAGS_DIVBY0; while (0) #define MPFR_SET_NANFLAG() \ do __gmpfr_flags |= MPFR_FLAGS_NAN; while (0) #define MPFR_SET_INEXFLAG() \ do __gmpfr_flags |= MPFR_FLAGS_INEXACT; while (0) #define MPFR_SET_ERANGEFLAG() \ do __gmpfr_flags |= MPFR_FLAGS_ERANGE; while (0) /* Testing an exception flag correctly is tricky. There are mainly two pitfalls: First, one needs to remember to clear the corresponding flag, in case it was set before the function call or during some intermediate computations (in practice, one can clear all the flags). Secondly, one needs to test the flag early enough, i.e. before it can be modified by another function. Moreover, it is quite difficult (if not impossible) to reliably check problems with "make check". To avoid these pitfalls, it is recommended to use the following macros. Other use of the exception-flag predicate functions/macros will be detected by mpfrlint. Note: _op can be either a statement or an expression. MPFR_BLOCK_EXCEP should be used only inside a block; it is useful to detect some exception in order to exit the block as soon as possible. */ #define MPFR_BLOCK_DECL(_flags) mpfr_flags_t _flags /* The (void) (_flags) makes sure that _flags is read at least once (it makes sense to use MPFR_BLOCK while _flags will never be read in the source, so that we wish to avoid the corresponding warning). */ #define MPFR_BLOCK(_flags,_op) \ do \ { \ MPFR_CLEAR_FLAGS (); \ _op; \ (_flags) = __gmpfr_flags; \ (void) (_flags); \ } \ while (0) #define MPFR_BLOCK_TEST(_flags,_f) MPFR_UNLIKELY ((_flags) & (_f)) #define MPFR_BLOCK_EXCEP (__gmpfr_flags & (MPFR_FLAGS_UNDERFLOW | \ MPFR_FLAGS_OVERFLOW | \ MPFR_FLAGS_DIVBY0 | \ MPFR_FLAGS_NAN)) /* Let's use a MPFR_ prefix, because e.g. OVERFLOW is defined by glibc's math.h, though this is not a reserved identifier! */ #define MPFR_UNDERFLOW(_flags) MPFR_BLOCK_TEST (_flags, MPFR_FLAGS_UNDERFLOW) #define MPFR_OVERFLOW(_flags) MPFR_BLOCK_TEST (_flags, MPFR_FLAGS_OVERFLOW) #define MPFR_NANFLAG(_flags) MPFR_BLOCK_TEST (_flags, MPFR_FLAGS_NAN) #define MPFR_INEXFLAG(_flags) MPFR_BLOCK_TEST (_flags, MPFR_FLAGS_INEXACT) #define MPFR_ERANGEFLAG(_flags) MPFR_BLOCK_TEST (_flags, MPFR_FLAGS_ERANGE) #define MPFR_DIVBY0(_flags) MPFR_BLOCK_TEST (_flags, MPFR_FLAGS_DIVBY0) /****************************************************** ******************* Assertions ********************* ******************************************************/ /* MPFR_WANT_ASSERT can take 4 values (the default value is 0): -1 (or below): Do not check any assertion. Discouraged, in particular for a shared library (for time-critical applications, LTO with a static library should also be used anyway). 0: Check normal assertions. 1: Check debugging assertions too. 2 (or above): Additional checks that may take time. For instance, some functions may be tested by using two different implementations and comparing the results. */ /* Note: do not use GMP macros ASSERT_ALWAYS and ASSERT as they are not expressions, and as a consequence, they cannot be used in a for(), with a comma operator and so on. */ /* MPFR_ASSERTN(expr): assertions that should normally be checked, otherwise give a hint to the compiler. MPFR_ASSERTD(expr): assertions that should be checked when testing, otherwise give a hint to the compiler. MPFR_DBGRES(assignment): to be used when the result is tested only in an MPFR_ASSERTD expression (in order to avoid a warning, e.g. with GCC's -Wunused-but-set-variable, in non-debug mode). Note: WG14/N2270 proposed a maybe_unused attribute, which could be useful to avoid MPFR_DBGRES. See: https://www.open-std.org/jtc1/sc22/wg14/www/docs/n2270.pdf Note: Evaluating expr might yield side effects, but such side effects must not change the results (except by yielding an assertion failure). */ #ifndef MPFR_WANT_ASSERT # define MPFR_WANT_ASSERT 0 #endif #if MPFR_WANT_ASSERT < 0 # undef MPFR_EXP_CHECK # define MPFR_ASSERTN(expr) MPFR_ASSUME (expr) #else # define MPFR_ASSERTN(expr) \ ((void) ((MPFR_LIKELY(expr)) || (ASSERT_FAIL(expr),MPFR_ASSUME(expr),0))) /* Some explanations: mpfr_assert_fail is not marked as "no return", so that ((void) ((MPFR_LIKELY(expr)) || (ASSERT_FAIL(expr),0))) cannot be a way to tell the compiler that after this code, expr is necessarily true. The MPFR_ASSUME(expr) is a way to tell the compiler that if expr is false, then ASSERT_FAIL(expr) does not return (otherwise they would be a contradiction / UB when MPFR_ASSUME(expr) is reached). Such information is useful to avoid warnings like those from -Wmaybe-uninitialized, e.g. in tests/turandom.c r11663 on t[0] from "mpfr_equal_p (y, t[0])". TODO: Remove the MPFR_ASSUME(expr) once mpfr_assert_fail is marked as "no return". */ #endif #if MPFR_WANT_ASSERT > 0 # define MPFR_EXP_CHECK 1 # define MPFR_ASSERTD(expr) MPFR_ASSERTN (expr) # define MPFR_DBGRES(A) (A) #else # define MPFR_ASSERTD(expr) MPFR_ASSUME (expr) # define MPFR_DBGRES(A) ((void) (A)) #endif /* MPFR_ASSUME is like assert(), but it is a hint to a compiler about a statement of fact in a function call free expression, which allows the compiler to generate better machine code. __builtin_unreachable has been introduced in GCC 4.5 but it works fine since 4.8 only (before it may generate unoptimized code if there are more than one decision). Note: The goal of MPFR_ASSUME() is to allow the compiler to optimize even more. Thus we need to make sure that its use in MPFR will never yield code generation. Since MPFR_ASSUME() may be used by MPFR_ASSERTN() and MPFR_ASSERTD(), whose expression might have side effects, we need to make sure that the expression x is not evaluated in such a case. This is done with __builtin_constant_p (!!(x) || !(x)), whose value is 0 if x has side effects, and normally 1 if the compiler knows that x has no side effects (since here, it can deduce that !!(x) || !(x) is equivalent to the constant 1). In the former case, MPFR_ASSUME(x) will give (void) 0, and in the latter case, it will give: (x) ? (void) 0 : __builtin_unreachable() In the development code, it is better to use MPFR_ASSERTD than MPFR_ASSUME, since it'll check if the condition is true in debug build. */ #if defined(MPFR_HAVE_BUILTIN_UNREACHABLE) && __MPFR_GNUC(4, 8) # define MPFR_ASSUME(x) \ (! __builtin_constant_p (!!(x) || !(x)) || (x) ? \ (void) 0 : __builtin_unreachable()) #elif defined(_MSC_VER) # define MPFR_ASSUME(x) __assume(x) #else # define MPFR_ASSUME(x) ((void) 0) #endif #include "mpfr-sassert.h" /* Code to deal with impossible, for functions returning an int. The "return 0;" avoids an error with current GCC versions and "-Werror=return-type". WARNING: It doesn't use do { } while (0) for Insure++ */ #if defined(HAVE_BUILTIN_UNREACHABLE) # define MPFR_RET_NEVER_GO_HERE() do { __builtin_unreachable(); } while (0) #else # define MPFR_RET_NEVER_GO_HERE() do { MPFR_ASSERTN(0); return 0; } while (0) #endif /****************************************************** ******************* Warnings *********************** ******************************************************/ /* MPFR_WARNING is no longer useful, but let's keep the macro in case it needs to be used again in the future. */ #ifdef MPFR_USE_WARNINGS # define MPFR_WARNING(W) \ do \ { \ char *q = getenv ("MPFR_QUIET"); \ if (q == NULL || *q == 0) \ fprintf (stderr, "MPFR: %s\n", W); \ } \ while (0) #else # define MPFR_WARNING(W) ((void) 0) #endif /****************************************************** ***************** double macros ******************** ******************************************************/ /* Precision used for lower precision computations */ #define MPFR_SMALL_PRECISION 32 /* Definition of constants */ #define LOG2 0.69314718055994528622 /* log(2) rounded to zero on 53 bits */ /* MPFR_DOUBLE_SPEC = 1 if the C type 'double' corresponds to IEEE-754 double precision, 0 if it doesn't, and undefined if one doesn't know. On all the tested machines, MPFR_DOUBLE_SPEC = 1. To have this macro defined here, #include <float.h> is needed. If need be, other values could be defined for other specs (once they are known). */ #if !defined(MPFR_DOUBLE_SPEC) && defined(FLT_RADIX) && \ defined(DBL_MANT_DIG) && defined(DBL_MIN_EXP) && defined(DBL_MAX_EXP) # if FLT_RADIX == 2 && DBL_MANT_DIG == 53 && \ DBL_MIN_EXP == -1021 && DBL_MAX_EXP == 1024 # define MPFR_DOUBLE_SPEC 1 # else # define MPFR_DOUBLE_SPEC 0 # endif #endif /* With -DMPFR_DISABLE_IEEE_FLOATS, exercise non IEEE floats */ #ifdef MPFR_DISABLE_IEEE_FLOATS # ifdef _MPFR_IEEE_FLOATS # undef _MPFR_IEEE_FLOATS # endif # define _MPFR_IEEE_FLOATS 0 # undef HAVE_LDOUBLE_IS_DOUBLE # undef HAVE_LDOUBLE_IEEE_EXT_LITTLE # undef HAVE_LDOUBLE_IEEE_EXT_BIG # undef HAVE_LDOUBLE_IEEE_QUAD_BIG # undef HAVE_LDOUBLE_IEEE_QUAD_LITTLE #endif #ifndef IEEE_DBL_MANT_DIG #define IEEE_DBL_MANT_DIG 53 #endif #define MPFR_LIMBS_PER_DOUBLE ((IEEE_DBL_MANT_DIG-1)/GMP_NUMB_BITS+1) #ifndef IEEE_FLT_MANT_DIG #define IEEE_FLT_MANT_DIG 24 #endif #define MPFR_LIMBS_PER_FLT ((IEEE_FLT_MANT_DIG-1)/GMP_NUMB_BITS+1) /* Visual C++ doesn't support +1.0/0.0, -1.0/0.0 and 0.0/0.0 at compile time. Clang with -fsanitize=undefined is a bit similar due to a bug: https://llvm.org/bugs/show_bug.cgi?id=17381 (fixed on 2015-12-03) but even without its sanitizer, it may be better to use the double_zero version until IEEE 754 division by zero is properly supported: https://llvm.org/bugs/show_bug.cgi?id=17005 Note: DBL_NAN is 0/0, thus its value is a quiet NaN (qNAN). */ #if (defined(_MSC_VER) && defined(_WIN32) && (_MSC_VER >= 1200)) || \ defined(__clang__) static double double_zero = 0.0; # define DBL_NAN (double_zero/double_zero) # define DBL_POS_INF ((double) 1.0/double_zero) # define DBL_NEG_INF ((double)-1.0/double_zero) # define DBL_NEG_ZERO (-double_zero) #else # define DBL_POS_INF ((double) 1.0/0.0) # define DBL_NEG_INF ((double)-1.0/0.0) # define DBL_NAN ((double) 0.0/0.0) # define DBL_NEG_ZERO (-0.0) #endif /* Note: In the past, there was specific code for _MPFR_IEEE_FLOATS, which was based on NaN and Inf memory representations. This code was breaking the aliasing rules (see ISO C99, 6.5#6 and 6.5#7 on the effective type) and for this reason it did not behave correctly with GCC 4.5.0 20091119. The code needed a memory transfer and was probably not better than the macros below with a good compiler (a fix based on the NaN / Inf memory representation would be even worse due to C limitations), and this code could be selected only when MPFR was built with --with-gmp-build, thus introducing a difference (bad for maintaining/testing MPFR); therefore it has been removed. The old code required that the argument x be an lvalue of type double. We still require that, in case one would need to change the macros below, e.g. for some broken compiler. But the LVALUE(x) condition could be removed if really necessary. */ /* Below, the &(x) == &(x) || &(x) != &(x) allows to make sure that x is a lvalue without (probably) any warning from the compiler. The &(x) != &(x) is needed to avoid a failure under Mac OS X 10.4.11 (with Xcode 2.4.1, i.e. the latest one). */ #define LVALUE(x) (&(x) == &(x) || &(x) != &(x)) #define DOUBLE_ISINF(x) (LVALUE(x) && ((x) > DBL_MAX || (x) < -DBL_MAX)) /* The DOUBLE_ISNAN(x) macro must be valid with any real floating type, thus constants must be of integer type (e.g. 0). */ #if defined(MPFR_NANISNAN) || (__MPFR_GNUC(1,0) && !defined(__STRICT_ANSI__)) /* Avoid MIPSpro / IRIX64 / GCC (incorrect) optimizations. The + must not be replaced by a ||. With gcc -ffast-math, NaN is regarded as a positive number or something like that; the second test catches this case. [2016-03-01] Various tests now fail with gcc -ffast-math or just -ffinite-math-only; such options are not supported, but this makes difficult to test MPFR assuming x == x optimization to 1. Anyway support of functions/tests of using native FP and special values for non-IEEE-754 environment will always be on a case-by-case basis. [2018-06-02] Let's use this macro instead of the usual (x) != (x) test with all GCC versions except in ISO C mode[*], as due to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=323 there is no guarantee that (x) != (x) will be true only for NaN. Testing __STRICT_ANSI__ is suggested in: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85995 but this is not safe if the user adds a -f option affecting conformance, in which case this would be a user error (however, note that the configure test associated with MPFR_NANISNAN will catch some issues). */ # define DOUBLE_ISNAN(x) \ (LVALUE(x) && !((((x) >= 0) + ((x) <= 0)) && -(x)*(x) <= 0)) #else # define DOUBLE_ISNAN(x) (LVALUE(x) && (x) != (x)) #endif /****************************************************** ********** long double macros and typedef ********** ******************************************************/ /* We try to get the exact value of the precision of long double (provided by the implementation) in order to provide correct rounding in this case (not guaranteed if the C implementation does not have an adequate long double arithmetic). Note that it may be lower than the precision of some numbers that can be represented in a long double; e.g. on FreeBSD/x86, it is 53 because the processor is configured to round in double precision (even when using the long double type -- this is a limitation of the x87 arithmetic), and on Mac OS X, it is 106 because the implementation is a double-double arithmetic. Otherwise (e.g. in base 10), we get an upper bound of the precision, and correct rounding isn't currently provided. */ /* Definitions are enabled only if <float.h> is included. */ #if defined (FLT_RADIX) #if defined(LDBL_MANT_DIG) && FLT_RADIX == 2 # define MPFR_LDBL_MANT_DIG LDBL_MANT_DIG #else # define MPFR_LDBL_MANT_DIG \ (sizeof(long double)*GMP_NUMB_BITS/sizeof(mp_limb_t)) #endif /* LONGDOUBLE_NAN_ACTION executes the code "action" if x is a NaN. */ /* On hppa2.0n-hp-hpux10 with the unbundled HP cc, the test x!=x on a NaN has been seen false, meaning NaNs are not detected. This seemed to happen only after other comparisons, not sure what's really going on. In any case we can pick apart the bytes to identify a NaN. */ #ifdef HAVE_LDOUBLE_IEEE_QUAD_BIG # define LONGDOUBLE_NAN_ACTION(x, action) \ do { \ union { \ long double ld; \ struct { \ unsigned int sign : 1; \ unsigned int exp : 15; \ unsigned int man3 : 16; \ unsigned int man2 : 32; \ unsigned int man1 : 32; \ unsigned int man0 : 32; \ } s; \ } u; \ u.ld = (x); \ if (u.s.exp == 0x7FFFL \ && (u.s.man0 | u.s.man1 | u.s.man2 | u.s.man3) != 0) \ { action; } \ } while (0) #endif #ifdef HAVE_LDOUBLE_IEEE_QUAD_LITTLE # define LONGDOUBLE_NAN_ACTION(x, action) \ do { \ union { \ long double ld; \ struct { \ unsigned int man0 : 32; \ unsigned int man1 : 32; \ unsigned int man2 : 32; \ unsigned int man3 : 16; \ unsigned int exp : 15; \ unsigned int sign : 1; \ } s; \ } u; \ u.ld = (x); \ if (u.s.exp == 0x7FFFL \ && (u.s.man0 | u.s.man1 | u.s.man2 | u.s.man3) != 0) \ { action; } \ } while (0) #endif /* Under IEEE rules, NaN is not equal to anything, including itself. "volatile" here stops "cc" on mips64-sgi-irix6.5 from optimizing away x!=x. */ #ifndef LONGDOUBLE_NAN_ACTION # define LONGDOUBLE_NAN_ACTION(x, action) \ do { \ volatile long double __x = LONGDOUBLE_VOLATILE (x); \ if ((x) != __x) \ { action; } \ } while (0) /* Some compilers do not have a proper "volatile" and #define volatile to empty (to avoid a build failure with programs using "volatile"), i.e. "volatile" is just ignored and will not prevent optimizations that could potentially break the IEEE rules. In this case, call an external function, hoping that the compiler will not optimize. */ # ifdef volatile __MPFR_DECLSPEC long double __gmpfr_longdouble_volatile (long double) MPFR_CONST_FUNCTION_ATTR; # define LONGDOUBLE_VOLATILE(x) (__gmpfr_longdouble_volatile (x)) # define WANT_GMPFR_LONGDOUBLE_VOLATILE 1 # else # define LONGDOUBLE_VOLATILE(x) (x) # endif #endif /* Some special case for IEEE_EXT Little Endian */ #if HAVE_LDOUBLE_IEEE_EXT_LITTLE typedef union { long double ld; struct { unsigned int manl : 32; unsigned int manh : 32; unsigned int expl : 8 ; unsigned int exph : 7; unsigned int sign : 1; } s; } mpfr_long_double_t; #endif /* HAVE_LDOUBLE_IEEE_EXT_LITTLE */ #endif /* long double macros and typedef */ /****************************************************** ***************** _Float128 support **************** ******************************************************/ /* This is standardized by IEEE 754-2008. */ #define IEEE_FLOAT128_MANT_DIG 113 /****************************************************** ****************** Decimal support ***************** ******************************************************/ #ifdef MPFR_WANT_DECIMAL_FLOATS #if defined(__GNUC__) && \ __FLT64_DECIMAL_DIG__ == 17 && \ __FLT128_DECIMAL_DIG__ == 36 /* GCC has standard _Decimal64 and _Decimal128 support. We may be able to detect the encoding here at compile time. Note: GCC may define __FLT64_DECIMAL_DIG__ and __FLT128_DECIMAL_DIG__ even when it does not support _Decimal64 and _Decimal128, e.g. on aarch64 and sparc64. But since MPFR_WANT_DECIMAL_FLOATS has been defined, we already know that these types should be supported. Determining which encoding is used via macros is not documented, and there is the risk of being wrong. Currently __DECIMAL_BID_FORMAT__ is defined on x86, where the BID encoding is used. But on PowerPC, where the DPD encoding is used, nothing indicating the encoding is defined. A possible reason may be that the decimal support is provided by the hardware (in this case), so that GCC does not need to care about the encoding. Thus the absence of a __DECIMAL_BID_FORMAT__ macro would not necessarily imply DPD, as similarly in the future, GCC could support an ISA-level implementation using the BID encoding. */ #ifdef __DECIMAL_BID_FORMAT__ #if defined(DECIMAL_DPD_FORMAT) # error "Decimal encoding mismatch (DPD assumed, BID detected)" #elif !defined(DECIMAL_GENERIC_CODE) # define DECIMAL_BID_FORMAT 1 #endif #endif /* __DECIMAL_BID_FORMAT__ */ #endif /* __GNUC__ */ #if defined(DECIMAL_GENERIC_CODE) # if defined(DECIMAL_BID_FORMAT) # error "DECIMAL_BID_FORMAT and DECIMAL_GENERIC_CODE both defined" # endif # if defined(DECIMAL_DPD_FORMAT) # error "DECIMAL_DPD_FORMAT and DECIMAL_GENERIC_CODE both defined" # endif #elif defined(DECIMAL_BID_FORMAT) || defined(DECIMAL_DPD_FORMAT) # if defined(DECIMAL_BID_FORMAT) && defined(DECIMAL_DPD_FORMAT) # error "DECIMAL_BID_FORMAT and DECIMAL_DPD_FORMAT both defined" # endif #else # define DECIMAL_GENERIC_CODE 1 #endif /* TODO: The following is ugly and possibly wrong on some platforms. Do something like union ieee_decimal128. */ union ieee_double_decimal64 { double d; _Decimal64 d64; }; /* FIXME: There's no reason to make the _Decimal128 code depend on whether _MPFR_IEEE_FLOATS is defined or not, as _MPFR_IEEE_FLOATS is about binary IEEE-754 floating point only. */ #if _MPFR_IEEE_FLOATS /* TODO: It would be better to define a different structure for DPD, where the t* bit-fields correspond to the declets. And to avoid confusion and detect coding errors, these bit-fields should have different names for BID and DPD. */ union ieee_decimal128 { struct { /* Assume little-endian double implies little-endian for bit-field allocation (C99 says: "The order of allocation of bit-fields within a unit (high-order to low-order or low-order to high-order) is implementation-defined.") */ #if defined(HAVE_DECIMAL128_IEEE_LITTLE_ENDIAN) #define HAVE_DECIMAL128_IEEE 1 unsigned int t3:32; unsigned int t2:32; unsigned int t1:32; unsigned int t0:14; unsigned int comb:17; unsigned int sig:1; #elif defined(HAVE_DECIMAL128_IEEE_BIG_ENDIAN) #define HAVE_DECIMAL128_IEEE 1 unsigned int sig:1; unsigned int comb:17; unsigned int t0:14; unsigned int t1:32; unsigned int t2:32; unsigned int t3:32; #else /* unknown bit-field ordering */ /* This will not be used as HAVE_DECIMAL128_IEEE is not defined. */ unsigned int dummy; #endif } s; _Decimal128 d128; }; #endif /* _MPFR_IEEE_FLOATS */ #endif /* MPFR_WANT_DECIMAL_FLOATS */ /****************************************************** **************** mpfr_t properties ***************** ******************************************************/ #define MPFR_PREC_COND(p) ((p) >= MPFR_PREC_MIN && (p) <= MPFR_PREC_MAX) #define MPFR_PREC_IN_RANGE(p) (MPFR_ASSERTD (MPFR_PREC_COND(p)), (p)) /* In the following macro, p is usually a mpfr_prec_t, but this macro works with other integer types (without integer overflow). Checking that p >= 1 in debug mode is useful here because this macro can be used on a computed precision (in particular, this formula does not work for a degenerate case p = 0, and could give different results on different platforms). But let us not use an assertion checking in the MPFR_LAST_LIMB() and MPFR_LIMB_SIZE() macros below to avoid too much expansion for assertions (in practice, this should be a problem just when testing MPFR with the --enable-assert configure option and the -ansi -pedantic-errors gcc compiler flags). */ #define MPFR_PREC2LIMBS(p) \ (MPFR_ASSERTD ((p) >= 1), ((p) - 1) / GMP_NUMB_BITS + 1) #define MPFR_PREC(x) ((x)->_mpfr_prec) #define MPFR_EXP(x) ((x)->_mpfr_exp) #define MPFR_MANT(x) ((x)->_mpfr_d) #define MPFR_GET_PREC(x) MPFR_PREC_IN_RANGE (MPFR_PREC (x)) #define MPFR_LAST_LIMB(x) ((MPFR_GET_PREC (x) - 1) / GMP_NUMB_BITS) #define MPFR_LIMB_SIZE(x) (MPFR_LAST_LIMB (x) + 1) /****************************************************** *************** Exponent properties **************** ******************************************************/ /* Limits of the mpfr_exp_t type (NOT those of valid exponent values). These macros can be used in preprocessor directives. */ #if _MPFR_EXP_FORMAT == 1 # define MPFR_EXP_MAX (SHRT_MAX) # define MPFR_EXP_MIN (SHRT_MIN) #elif _MPFR_EXP_FORMAT == 2 # define MPFR_EXP_MAX (INT_MAX) # define MPFR_EXP_MIN (INT_MIN) #elif _MPFR_EXP_FORMAT == 3 # define MPFR_EXP_MAX (LONG_MAX) # define MPFR_EXP_MIN (LONG_MIN) #elif _MPFR_EXP_FORMAT == 4 /* Note: MPFR_EXP_MAX and MPFR_EXP_MIN must not be used in #if directives if _MPFR_EXP_FORMAT == 4 and MPFR_HAVE_INTMAX_MAX is not defined. */ # define MPFR_EXP_MAX (MPFR_INTMAX_MAX) # define MPFR_EXP_MIN (MPFR_INTMAX_MIN) #else # error "Invalid MPFR Exp format" #endif /* Before doing a cast to mpfr_uexp_t, make sure that the value is non-negative. */ #define MPFR_UEXP(X) (MPFR_ASSERTD ((X) >= 0), (mpfr_uexp_t) (X)) /* Define mpfr_eexp_t, mpfr_ueexp_t and MPFR_EXP_FSPEC. Warning! MPFR_EXP_FSPEC is the length modifier associated with these types mpfr_eexp_t / mpfr_ueexp_t, not with mpfr_exp_t. (Should we change that, or is this safer to detect bugs, e.g. in the context of an expression with computations with long?) */ #if _MPFR_EXP_FORMAT <= 3 typedef long mpfr_eexp_t; typedef unsigned long mpfr_ueexp_t; # define mpfr_get_exp_t(x,r) mpfr_get_si((x),(r)) # define mpfr_set_exp_t(x,e,r) mpfr_set_si((x),(e),(r)) # define MPFR_EXP_FSPEC "l" #else typedef intmax_t mpfr_eexp_t; typedef uintmax_t mpfr_ueexp_t; # define mpfr_get_exp_t(x,r) mpfr_get_sj((x),(r)) # define mpfr_set_exp_t(x,e,r) mpfr_set_sj((x),(e),(r)) # define MPFR_EXP_FSPEC "j" #endif /* Size of mpfr_exp_t in limbs */ #define MPFR_EXP_LIMB_SIZE \ ((sizeof (mpfr_exp_t) - 1) / MPFR_BYTES_PER_MP_LIMB + 1) /* Invalid exponent value (to track bugs...) */ #define MPFR_EXP_INVALID \ ((mpfr_exp_t) 1 << (GMP_NUMB_BITS*sizeof(mpfr_exp_t)/sizeof(mp_limb_t)-2)) /* Definition of the exponent limits for MPFR numbers. * These limits are chosen so that if e is such an exponent, then 2e-1 and * 2e+1 are representable. This is useful for intermediate computations, * in particular the multiplication. */ #undef MPFR_EMIN_MIN #undef MPFR_EMIN_MAX #undef MPFR_EMAX_MIN #undef MPFR_EMAX_MAX #define MPFR_EMIN_MIN (1-MPFR_EXP_INVALID) #define MPFR_EMIN_MAX (MPFR_EXP_INVALID-1) #define MPFR_EMAX_MIN (1-MPFR_EXP_INVALID) #define MPFR_EMAX_MAX (MPFR_EXP_INVALID-1) /* Use MPFR_GET_EXP and MPFR_SET_EXP instead of MPFR_EXP directly, unless when the exponent may be out-of-range, for instance when setting the exponent before calling mpfr_check_range. MPFR_EXP_CHECK is defined when MPFR_WANT_ASSERT is defined, but if you don't use MPFR_WANT_ASSERT (for speed reasons), you can still define MPFR_EXP_CHECK by setting -DMPFR_EXP_CHECK in $CFLAGS. Note about MPFR_EXP_IN_RANGE and MPFR_SET_EXP: The exp expression is required to have a signed type. To avoid spurious failures, we could cast (exp) to mpfr_exp_t, but this wouldn't allow us to detect some bugs that can occur on particular platforms. Anyway, an unsigned type for exp is suspicious and should be regarded as a bug. */ #define MPFR_EXP_IN_RANGE(e) \ (MPFR_ASSERTD (IS_SIGNED (e)), (e) >= __gmpfr_emin && (e) <= __gmpfr_emax) #ifdef MPFR_EXP_CHECK # define MPFR_GET_EXP(x) (mpfr_get_exp) (x) # define MPFR_SET_EXP(x,e) (MPFR_ASSERTN (MPFR_EXP_IN_RANGE (e)), \ (void) (MPFR_EXP (x) = (e))) # define MPFR_SET_INVALID_EXP(x) ((void) (MPFR_EXP (x) = MPFR_EXP_INVALID)) #else # define MPFR_GET_EXP(x) MPFR_EXP (x) # define MPFR_SET_EXP(x,e) ((void) (MPFR_EXP (x) = (e))) # define MPFR_SET_INVALID_EXP(x) ((void) 0) #endif /* Compare the exponents of two numbers, which can be either MPFR numbers or UBF numbers. */ #define MPFR_UBF_EXP_LESS_P(x,y) \ (MPFR_UNLIKELY (MPFR_IS_UBF (x) || MPFR_IS_UBF (y)) ? \ mpfr_ubf_exp_less_p (x, y) : MPFR_GET_EXP (x) < MPFR_GET_EXP (y)) /****************************************************** ********* Singular values (NAN, INF, ZERO) ********* ******************************************************/ /* Enum special value of exponent. */ # define MPFR_EXP_ZERO (MPFR_EXP_MIN+1) # define MPFR_EXP_NAN (MPFR_EXP_MIN+2) # define MPFR_EXP_INF (MPFR_EXP_MIN+3) # define MPFR_EXP_UBF (MPFR_EXP_MIN+4) #define MPFR_IS_NAN(x) (MPFR_EXP(x) == MPFR_EXP_NAN) #define MPFR_SET_NAN(x) (MPFR_EXP(x) = MPFR_EXP_NAN) #define MPFR_IS_INF(x) (MPFR_EXP(x) == MPFR_EXP_INF) #define MPFR_SET_INF(x) (MPFR_EXP(x) = MPFR_EXP_INF) #define MPFR_IS_ZERO(x) (MPFR_EXP(x) == MPFR_EXP_ZERO) #define MPFR_SET_ZERO(x) (MPFR_EXP(x) = MPFR_EXP_ZERO) #define MPFR_NOTZERO(x) (MPFR_EXP(x) != MPFR_EXP_ZERO) #define MPFR_IS_UBF(x) (MPFR_EXP(x) == MPFR_EXP_UBF) #define MPFR_SET_UBF(x) (MPFR_EXP(x) = MPFR_EXP_UBF) #define MPFR_IS_NORMALIZED(x) \ (MPFR_LIMB_MSB (MPFR_MANT(x)[MPFR_LAST_LIMB(x)]) != 0) #define MPFR_IS_FP(x) (!MPFR_IS_NAN(x) && !MPFR_IS_INF(x)) /* Note: contrary to the MPFR_IS_PURE_*(x) macros, the MPFR_IS_SINGULAR*(x) macros may be used even when x is being constructed, i.e. its exponent field is already set (possibly out-of-range), but its significand field may still contain arbitrary data. Thus MPFR_IS_PURE_FP(x) is not always equivalent to !MPFR_IS_SINGULAR(x); see the code below. */ #define MPFR_IS_SINGULAR(x) (MPFR_EXP(x) <= MPFR_EXP_INF) #define MPFR_IS_SINGULAR_OR_UBF(x) (MPFR_EXP(x) <= MPFR_EXP_UBF) /* The following two macros return true iff the value is a regular number, i.e. it is not a singular number. In debug mode, the format is also checked: valid exponent, but potentially out of range; normalized value. In contexts where UBF's are not accepted or not possible, MPFR_IS_PURE_FP is preferable. If UBF's are accepted, MPFR_IS_PURE_UBF must be used. */ #define MPFR_IS_PURE_FP(x) \ (!MPFR_IS_SINGULAR(x) && \ (MPFR_ASSERTD (MPFR_EXP (x) >= MPFR_EMIN_MIN && \ MPFR_EXP (x) <= MPFR_EMAX_MAX && \ MPFR_IS_NORMALIZED (x)), 1)) #define MPFR_IS_PURE_UBF(x) \ (!MPFR_IS_SINGULAR(x) && \ (MPFR_ASSERTD ((MPFR_IS_UBF (x) || \ (MPFR_EXP (x) >= MPFR_EMIN_MIN && \ MPFR_EXP (x) <= MPFR_EMAX_MAX)) && \ MPFR_IS_NORMALIZED (x)), 1)) /* Ditto for 2 numbers. */ #define MPFR_ARE_SINGULAR(x,y) \ (MPFR_UNLIKELY(MPFR_IS_SINGULAR(x)) || MPFR_UNLIKELY(MPFR_IS_SINGULAR(y))) #define MPFR_ARE_SINGULAR_OR_UBF(x,y) \ (MPFR_UNLIKELY(MPFR_IS_SINGULAR_OR_UBF(x)) || \ MPFR_UNLIKELY(MPFR_IS_SINGULAR_OR_UBF(y))) /****************************************************** ******************** Sign macros ******************* ******************************************************/ /* These are sign macros for MPFR numbers only. */ #define MPFR_SIGN_POS (1) #define MPFR_SIGN_NEG (-1) #define MPFR_IS_STRICTPOS(x) (MPFR_NOTZERO (x) && MPFR_IS_POS (x)) #define MPFR_IS_STRICTNEG(x) (MPFR_NOTZERO (x) && MPFR_IS_NEG (x)) #define MPFR_IS_NEG(x) (MPFR_SIGN(x) < 0) #define MPFR_IS_POS(x) (MPFR_SIGN(x) > 0) #define MPFR_SET_POS(x) (MPFR_SIGN(x) = MPFR_SIGN_POS) #define MPFR_SET_NEG(x) (MPFR_SIGN(x) = MPFR_SIGN_NEG) #define MPFR_CHANGE_SIGN(x) (MPFR_SIGN(x) = -MPFR_SIGN(x)) #define MPFR_SET_SAME_SIGN(x, y) (MPFR_SIGN(x) = MPFR_SIGN(y)) #define MPFR_SET_OPPOSITE_SIGN(x, y) (MPFR_SIGN(x) = -MPFR_SIGN(y)) #define MPFR_ASSERT_SIGN(s) \ (MPFR_ASSERTD((s) == MPFR_SIGN_POS || (s) == MPFR_SIGN_NEG)) #define MPFR_SET_SIGN(x, s) \ (MPFR_ASSERT_SIGN(s), MPFR_SIGN(x) = s) #define MPFR_IS_POS_SIGN(s1) ((s1) > 0) #define MPFR_IS_NEG_SIGN(s1) ((s1) < 0) #define MPFR_MULT_SIGN(s1, s2) ((s1) * (s2)) /* Transform a sign to 1 or -1 */ #define MPFR_FROM_SIGN_TO_INT(s) (s) #define MPFR_INT_SIGN(x) MPFR_FROM_SIGN_TO_INT(MPFR_SIGN(x)) /****************************************************** *************** Ternary value macros *************** ******************************************************/ /* Special inexact value */ #define MPFR_EVEN_INEX 2 /* Note: the addition/subtraction of 2 comparisons below instead of the use of the ?: operator allows GCC and Clang to generate better code in general; this is the case at least with GCC on x86 (32 & 64 bits), PowerPC and Aarch64 (64-bit ARM), and with Clang on x86_64. VSIGN code based on mini-gmp's GMP_CMP macro; adapted for INEXPOS. */ /* Macros for functions returning two inexact values in an 'int' (exact = 0, positive = 1, negative = 2) */ #define INEXPOS(y) (((y) != 0) + ((y) < 0)) #define INEX(y,z) (INEXPOS(y) | (INEXPOS(z) << 2)) /* When returning the ternary inexact value, ALWAYS use one of the following two macros, unless the flag comes from another function returning the ternary inexact value */ #define MPFR_RET(I) return \ (I) != 0 ? ((__gmpfr_flags |= MPFR_FLAGS_INEXACT), (I)) : 0 #define MPFR_RET_NAN return (__gmpfr_flags |= MPFR_FLAGS_NAN), 0 /* Sign of a native value. */ #define VSIGN(I) (((I) > 0) - ((I) < 0)) #define SAME_SIGN(I1,I2) (VSIGN (I1) == VSIGN (I2)) /****************************************************** *************** Rounding mode macros *************** ******************************************************/ /* MPFR_RND_MAX gives the number of supported rounding modes by all functions. */ #define MPFR_RND_MAX ((mpfr_rnd_t)((MPFR_RNDF)+1)) /* We want to test this : * (rnd == MPFR_RNDU && test) || (rnd == RNDD && !test) * i.e. it transforms RNDU or RNDD to away or zero according to the sign. * The argument test must be 0 or 1. */ #define MPFR_IS_RNDUTEST_OR_RNDDNOTTEST(rnd, test) \ (MPFR_ASSERTD ((test) == 0 || (test) == 1), \ ((rnd) + (test)) == MPFR_RNDD) /* We want to test if rnd rounds toward zero or away from zero. 'neg' is 1 if negative, and 0 if positive. */ #define MPFR_IS_LIKE_RNDZ(rnd, neg) \ ((rnd) == MPFR_RNDZ || MPFR_IS_RNDUTEST_OR_RNDDNOTTEST (rnd, neg)) #define MPFR_IS_LIKE_RNDA(rnd, neg) \ ((rnd) == MPFR_RNDA || MPFR_IS_RNDUTEST_OR_RNDDNOTTEST (rnd, (neg) == 0)) #define MPFR_IS_LIKE_RNDU(rnd, sign) \ (((rnd) == MPFR_RNDU) || \ ((rnd) == MPFR_RNDZ && MPFR_IS_NEG_SIGN (sign)) || \ ((rnd) == MPFR_RNDA && MPFR_IS_POS_SIGN (sign))) #define MPFR_IS_LIKE_RNDD(rnd, sign) \ (((rnd) == MPFR_RNDD) || \ ((rnd) == MPFR_RNDZ && MPFR_IS_POS_SIGN (sign)) || \ ((rnd) == MPFR_RNDA && MPFR_IS_NEG_SIGN (sign))) /* Invert RNDU and RNDD; the other rounding modes are unchanged. */ #define MPFR_INVERT_RND(rnd) ((rnd) == MPFR_RNDU ? MPFR_RNDD : \ (rnd) == MPFR_RNDD ? MPFR_RNDU : (rnd)) /* Transform RNDU and RNDD to RNDZ according to test */ #define MPFR_UPDATE_RND_MODE(rnd, test) \ do { \ if (MPFR_UNLIKELY(MPFR_IS_RNDUTEST_OR_RNDDNOTTEST(rnd, test))) \ rnd = MPFR_RNDZ; \ } while (0) /* Transform RNDU and RNDD to RNDZ or RNDA according to sign; leave the other modes unchanged. Usage: MPFR_UPDATE2_RND_MODE (rnd_mode, MPFR_SIGN (x)) */ #define MPFR_UPDATE2_RND_MODE(rnd, sign) \ do { \ if (rnd == MPFR_RNDU) \ rnd = MPFR_IS_POS_SIGN (sign) ? MPFR_RNDA : MPFR_RNDZ; \ else if (rnd == MPFR_RNDD) \ rnd = MPFR_IS_NEG_SIGN (sign) ? MPFR_RNDA : MPFR_RNDZ; \ } while (0) /****************************************************** ****************** Limb macros ********************* ******************************************************/ /* MPFR_LIMB: Cast to mp_limb_t, assuming that x is based on mp_limb_t variables (needed when mp_limb_t is defined as an integer type shorter than int, due to the integer promotion rules, which is possible only if MPFR_LONG_WITHIN_LIMB is not defined). Warning! This will work only when the computed value x is congruent to the expected value modulo MPFR_LIMB_MAX + 1. Be aware that this macro may not solve all the problems related to the integer promotion rules, because it won't have an influence on the evaluation of x itself. Hence the need for... MPFR_LIMB_LSHIFT: Left shift by making sure that the shifted argument is unsigned (use unsigned long due to the MPFR_LONG_WITHIN_LIMB test). For instance, due to the integer promotion rules, if mp_limb_t is defined as a 16-bit unsigned short and an int has 32 bits, then a mp_limb_t will be converted to an int, which is signed. */ #ifdef MPFR_LONG_WITHIN_LIMB #define MPFR_LIMB(x) (x) #define MPFR_LIMB_LSHIFT(x,c) ((x) << (c)) #else #define MPFR_LIMB(x) ((mp_limb_t) (x)) #define MPFR_LIMB_LSHIFT(x,c) MPFR_LIMB((unsigned long) (x) << (c)) #endif /* Definition of simple mp_limb_t constants */ #define MPFR_LIMB_ZERO ((mp_limb_t) 0) #define MPFR_LIMB_ONE ((mp_limb_t) 1) #define MPFR_LIMB_HIGHBIT MPFR_LIMB_LSHIFT (MPFR_LIMB_ONE, GMP_NUMB_BITS - 1) #define MPFR_LIMB_MAX ((mp_limb_t) -1) /* Mask to get the Most Significant Bit of a limb */ #define MPFR_LIMB_MSB(l) ((mp_limb_t) ((l) & MPFR_LIMB_HIGHBIT)) /* Mask for the low 's' bits of a limb */ #define MPFR_LIMB_MASK(s) \ (MPFR_ASSERTD (s >= 0 && s <= GMP_NUMB_BITS), \ s == GMP_NUMB_BITS ? MPFR_LIMB_MAX : \ (mp_limb_t) (MPFR_LIMB_LSHIFT (MPFR_LIMB_ONE, s) - MPFR_LIMB_ONE)) /****************************************************** ********************** Memory ********************** ******************************************************/ #define MPFR_BYTES_PER_MP_LIMB (GMP_NUMB_BITS/CHAR_BIT) /* Heap memory handling -------------------- Memory allocated for a significand (mantissa) has the following format: * A mp_size_t in a mpfr_size_limb_t union (see below). * An array of mp_limb_t (not all of them are necessarily used, as the precision can change without a reallocation). The goal of the mpfr_size_limb_t union is to make sure that size and alignment requirements are satisfied if mp_size_t and mp_limb_t have different sizes and/or alignment requirements. And the casts to void * prevents the compiler from emitting a warning (or error), such as: cast increases required alignment of target type with the -Wcast-align GCC option. Correct alignment is checked by MPFR_SET_MANT_PTR (when setting MPFR_MANT(x), the MPFR code should use this macro or guarantee a correct alignment at this time). Moreover, pointer conversions are not fully specified by the C standard, and the use of a union (and the double casts below) might help even if mp_size_t and mp_limb_t have the same size and the same alignment requirements. Still, there is currently no guarantee that this code is portable. Note that union members are not used at all. */ typedef union { mp_size_t s; mp_limb_t l; } mpfr_size_limb_t; #define MPFR_GET_ALLOC_SIZE(x) \ (((mp_size_t *) (void *) MPFR_MANT(x))[-1] + 0) #define MPFR_SET_ALLOC_SIZE(x, n) \ (((mp_size_t *) (void *) MPFR_MANT(x))[-1] = (n)) #define MPFR_MALLOC_SIZE(s) \ (sizeof(mpfr_size_limb_t) + MPFR_BYTES_PER_MP_LIMB * (size_t) (s)) #define MPFR_SET_MANT_PTR(x,p) \ (MPFR_MANT(x) = (mp_limb_t *) ((mpfr_size_limb_t *) (p) + 1)) #define MPFR_GET_REAL_PTR(x) \ ((void *) ((mpfr_size_limb_t *) (void *) MPFR_MANT(x) - 1)) /* Temporary memory handling */ #ifndef TMP_SALLOC /* GMP 4.1.x or below or internals */ #define MPFR_TMP_DECL TMP_DECL #define MPFR_TMP_MARK TMP_MARK #define MPFR_TMP_ALLOC TMP_ALLOC #define MPFR_TMP_FREE TMP_FREE #else #define MPFR_TMP_DECL(x) TMP_DECL #define MPFR_TMP_MARK(x) TMP_MARK #define MPFR_TMP_ALLOC(s) TMP_ALLOC(s) #define MPFR_TMP_FREE(x) TMP_FREE #endif #define MPFR_TMP_LIMBS_ALLOC(N) \ ((mp_limb_t *) MPFR_TMP_ALLOC ((size_t) (N) * MPFR_BYTES_PER_MP_LIMB)) /* The temporary var doesn't have any size field, but it doesn't matter * since only functions dealing with the Heap care about it */ #define MPFR_TMP_INIT1(xp, x, p) \ ( MPFR_PREC(x) = (p), \ MPFR_MANT(x) = (xp), \ MPFR_SET_POS(x), \ MPFR_SET_INVALID_EXP(x)) #define MPFR_TMP_INIT(xp, x, p, s) \ (xp = MPFR_TMP_LIMBS_ALLOC(s), \ MPFR_TMP_INIT1(xp, x, p)) /* Set y to s*significand(x)*2^e, for example MPFR_ALIAS(y,x,1,MPFR_EXP(x)) sets y to |x|, and MPFR_ALIAS(y,x,MPFR_SIGN(x),0) sets y to x*2^f such that 1/2 <= |y| < 1. Does not check y is in the valid exponent range. WARNING! x and y share the same mantissa. So, some operations are not valid if x has been provided via an argument, e.g., trying to modify the mantissa of y, even temporarily, or calling mpfr_clear on y. */ #define MPFR_ALIAS(y,x,s,e) \ (MPFR_PREC(y) = MPFR_PREC(x), \ MPFR_SIGN(y) = (s), \ MPFR_EXP(y) = (e), \ MPFR_MANT(y) = MPFR_MANT(x)) #define MPFR_TMP_INIT_ABS(y,x) \ MPFR_ALIAS (y, x, MPFR_SIGN_POS, MPFR_EXP (x)) #define MPFR_TMP_INIT_NEG(y,x) \ MPFR_ALIAS (y, x, - MPFR_SIGN (x), MPFR_EXP (x)) /****************************************************** ******************* Cache macros ******************* ******************************************************/ /* Cache struct */ #define mpfr_const_pi(_d,_r) mpfr_cache(_d, __gmpfr_cache_const_pi,_r) #define mpfr_const_log2(_d,_r) mpfr_cache(_d, __gmpfr_cache_const_log2, _r) #define mpfr_const_euler(_d,_r) mpfr_cache(_d, __gmpfr_cache_const_euler, _r) #define mpfr_const_catalan(_d,_r) mpfr_cache(_d,__gmpfr_cache_const_catalan,_r) /* Declare a global cache for a MPFR constant. If the shared cache is enabled, and if the constructor/destructor attributes are available, we need to initialize the shared lock of the cache with a constructor. It is the goal of the macro MPFR_DEFERRED_INIT_MASTER_DECL. FIXME: When MPFR is built with shared cache, the field "lock" is not explicitly initialized, which can yield a warning, e.g. with GCC's -Wmissing-field-initializers (and an error with -Werror). Since one does not know what is behind the associated typedef name, one cannot provide an explicit initialization for such a type. Two possible solutions: 1. Encapsulate the type in a structure or a union and use the universal zero initializer: { 0 } But: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80454 2. Use designated initializers when supported. But this needs a configure test. Using a diagnostic pragma to ignore the warning in this particular case is not really possible, because the warning occurs when the macro is expanded and one cannot put a pragma in the contents of a #define. */ #define MPFR_DECL_INIT_CACHE(_cache,_func) \ MPFR_DEFERRED_INIT_MASTER_DECL(_func, \ MPFR_LOCK_INIT( (_cache)->lock), \ MPFR_LOCK_CLEAR((_cache)->lock)) \ MPFR_CACHE_ATTR mpfr_cache_t _cache = {{ \ {{ 0, MPFR_SIGN_POS, 0, (mp_limb_t *) 0 }}, 0, _func \ MPFR_DEFERRED_INIT_SLAVE_VALUE(_func) \ }}; \ MPFR_MAKE_VARFCT (mpfr_cache_t,_cache) /****************************************************** *************** Threshold parameters *************** ******************************************************/ #include "mparam.h" /****************************************************** ****************** Useful macros ******************* ******************************************************/ /* The MAX, MIN and ABS macros may already be defined if gmp-impl.h has been included. They have the same semantics as in gmp-impl.h, but the expressions may be slightly different. So, it's better to undefine them first, as required by the ISO C standard. */ #undef MAX #undef MIN #undef ABS #define MAX(a, b) (((a) > (b)) ? (a) : (b)) #define MIN(a, b) (((a) < (b)) ? (a) : (b)) #define ABS(x) (((x)>0) ? (x) : -(x)) /* These macros help the compiler to determine if a test is likely or unlikely. The !! is necessary in case x is larger than a long. */ #if defined MPFR_DEBUG_PREDICTION && __MPFR_GNUC(3,0) /* Code to debug branch prediction, based on Ulrich Drepper's paper * "What Every Programmer Should Know About Memory": * https://people.freebsd.org/~lstewart/articles/cpumemory.pdf */ asm (".section predict_data, \"aw\"; .previous\n" ".section predict_line, \"a\"; .previous\n" ".section predict_file, \"a\"; .previous"); # if defined __x86_64__ # define MPFR_DEBUGPRED__(e,E) \ ({ long _e = !!(e); \ asm volatile (".pushsection predict_data\n" \ "..predictcnt%=: .quad 0; .quad 0\n" \ ".section predict_line; .quad %c1\n" \ ".section predict_file; .quad %c2; .popsection\n" \ "addq $1,..predictcnt%=(,%0,8)" \ : : "r" (_e == E), "i" (__LINE__), "i" (__FILE__)); \ __builtin_expect (_e, E); \ }) # elif defined __i386__ # define MPFR_DEBUGPRED__(e,E) \ ({ long _e = !!(e); \ asm volatile (".pushsection predict_data\n" \ "..predictcnt%=: .long 0; .long 0\n" \ ".section predict_line; .long %c1\n" \ ".section predict_file; .long %c2; .popsection\n" \ "incl ..predictcnt%=(,%0,4)" \ : : "r" (_e == E), "i" (__LINE__), "i" (__FILE__)); \ __builtin_expect (_e, E); \ }) # else # error "MPFR_DEBUGPRED__ definition missing" # endif # define MPFR_LIKELY(x) MPFR_DEBUGPRED__ ((x), 1) # define MPFR_UNLIKELY(x) MPFR_DEBUGPRED__ ((x), 0) #elif __MPFR_GNUC(3,0) || __MPFR_ICC(8,1,0) # define MPFR_LIKELY(x) (__builtin_expect(!!(x), 1)) # define MPFR_UNLIKELY(x) (__builtin_expect(!!(x), 0)) #else # define MPFR_LIKELY(x) (x) # define MPFR_UNLIKELY(x) (x) #endif /* Declare that some variable is initialized before being used (without a dummy initialization) in order to avoid some compiler warnings. Use the VAR = VAR trick (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36296#c3) only with gcc as this is undefined behavior, and we don't know what other compilers do (they may also be smarter). This self-initialization trick could be disabled with future gcc versions. It can explicitly be disabled by defining NO_INITIALIZED_TRICK (useful for testing). */ #if defined(__GNUC__) && !defined(NO_INITIALIZED_TRICK) # define INITIALIZED(VAR) VAR = VAR #else # define INITIALIZED(VAR) VAR #endif /* Ceil log 2: If GCC, uses a GCC extension, otherwise calls a function */ /* Warning: * Needs to define MPFR_NEED_LONGLONG. * Computes ceil(log2(x)) only for x integer (unsigned long) * Undefined if x is 0 */ #if __MPFR_GNUC(2,95) || __MPFR_ICC(8,1,0) /* Note: This macro MPFR_INT_CEIL_LOG2 shouldn't be used in an MPFR_ASSERT* macro, either directly or indirectly via other macros, otherwise it can yield an error due to a too large stringized expression in ASSERT_FAIL. A static inline function could be a better solution than this macro. */ /* FIXME: The current code assumes that x fits in an unsigned long (used by __gmpfr_int_ceil_log2) while MPFR_INT_CEIL_LOG2 is used on values that might be larger than ULONG_MAX on some platforms and/or with some build options; a loop could be used if x > ULONG_MAX. If the type of x is <= unsigned long, then no additional code will be generated thanks to obvious compiler optimization. */ #ifdef MPFR_LONG_WITHIN_LIMB # define MPFR_INT_CEIL_LOG2(x) \ (MPFR_UNLIKELY ((x) == 1) ? 0 : \ __extension__ ({ int _b; mp_limb_t _limb; \ MPFR_ASSERTN ((x) > 1); \ _limb = (x) - 1; \ MPFR_ASSERTN (_limb == (x) - 1); \ count_leading_zeros (_b, _limb); \ _b = GMP_NUMB_BITS - _b; \ MPFR_ASSERTD (_b >= 0); \ _b; })) #else # define MPFR_INT_CEIL_LOG2(x) \ (MPFR_UNLIKELY ((x) == 1) ? 0 : \ __extension__ ({ int _c = 0; unsigned long _x = (x) - 1; \ MPFR_ASSERTN ((x) > 1); \ while (_x != 0) \ { \ _x = _x >> 1; \ _c ++; \ }; \ MPFR_ASSERTD (_c >= 0); \ _c; })) #endif /* MPFR_LONG_WITHIN_LIMB */ #else # define MPFR_INT_CEIL_LOG2(x) \ (MPFR_ASSERTN (x <= ULONG_MAX), __gmpfr_int_ceil_log2(x)) #endif /* __MPFR_GNUC(2,95) || __MPFR_ICC(8,1,0) */ /* Add two integers with overflow handling */ /* Example: MPFR_SADD_OVERFLOW (c, a, b, long, unsigned long, * LONG_MIN, LONG_MAX, * goto overflow, goto underflow); */ #define MPFR_UADD_OVERFLOW(c,a,b,ACTION_IF_OVERFLOW) \ do { \ (c) = (a) + (b); \ if ((c) < (a)) ACTION_IF_OVERFLOW; \ } while (0) #define MPFR_SADD_OVERFLOW(c,a,b,STYPE,UTYPE,MIN,MAX,ACTION_IF_POS_OVERFLOW,ACTION_IF_NEG_OVERFLOW) \ do { \ if ((a) >= 0 && (b) >= 0) { \ UTYPE uc,ua,ub; \ ua = (UTYPE) (a); ub = (UTYPE) (b); \ MPFR_UADD_OVERFLOW (uc, ua, ub, ACTION_IF_POS_OVERFLOW); \ if (uc > (UTYPE)(MAX)) ACTION_IF_POS_OVERFLOW; \ else (c) = (STYPE) uc; \ } else if ((a) < 0 && (b) < 0) { \ UTYPE uc,ua,ub; \ ua = -(UTYPE) (a); ub = -(UTYPE) (b); \ MPFR_UADD_OVERFLOW (uc, ua, ub, ACTION_IF_NEG_OVERFLOW); \ if (uc >= -(UTYPE)(MIN) || uc > (UTYPE)(MAX)) { \ if (uc == -(UTYPE)(MIN)) (c) = (MIN); \ else ACTION_IF_NEG_OVERFLOW; } \ else (c) = -(STYPE) uc; \ } else (c) = (a) + (b); \ } while (0) /* Set a number to 1 (Fast) - It doesn't check if 1 is in the exponent range */ #define MPFR_SET_ONE(x) \ do { \ mp_size_t _size = MPFR_LAST_LIMB(x); \ MPFR_SET_POS(x); \ MPFR_EXP(x) = 1; \ MPN_ZERO ( MPFR_MANT(x), _size); \ MPFR_MANT(x)[_size] = MPFR_LIMB_HIGHBIT; \ } while (0) /* Compute s = (-a) % GMP_NUMB_BITS as unsigned */ #define MPFR_UNSIGNED_MINUS_MODULO(s, a) \ do \ { \ if (IS_POW2 (GMP_NUMB_BITS)) \ (s) = (- (unsigned int) (a)) % GMP_NUMB_BITS; \ else \ { \ (s) = (a) % GMP_NUMB_BITS; \ if ((s) != 0) \ (s) = GMP_NUMB_BITS - (s); \ } \ MPFR_ASSERTD ((s) >= 0 && (s) < GMP_NUMB_BITS); \ } \ while (0) /* Test if X (positive) is a power of 2 */ #define IS_POW2(X) (((X) & ((X) - 1)) == 0) #define NOT_POW2(X) (((X) & ((X) - 1)) != 0) /* Safe absolute value and difference (to avoid possible integer overflow) */ /* type is the target (unsigned) type */ #define SAFE_ABS(type,x) ((x) >= 0 ? (type)(x) : -(type)(x)) #define SAFE_DIFF(type,x,y) (MPFR_ASSERTD((x) >= (y)), (type)(x) - (type)(y)) #define ULONG2LONG(U) ((U) > LONG_MAX ? -1 - (long) ~(U) : (long) (U)) /* Check whether an integer type (after integer promotion) is signed. This can be determined at compilation time, but unfortunately, when used in practice, this is not a constant expression (because the argument X is not a constant expression, even though the result does not depend on its value), so that this cannot be used for a static assertion. */ #define IS_SIGNED(X) ((X) * 0 - 1 < 0) #define mpfr_get_d1(x) mpfr_get_d(x,__gmpfr_default_rounding_mode) /* Store in r the size in bits of the mpz_t z */ #define MPFR_MPZ_SIZEINBASE2(r, z) \ do { \ int _cnt; \ mp_size_t _size; \ MPFR_ASSERTD (mpz_sgn (z) != 0); \ _size = ABSIZ(z); \ MPFR_ASSERTD (_size >= 1); \ count_leading_zeros (_cnt, PTR(z)[_size-1]); \ (r) = (mp_bitcnt_t) _size * GMP_NUMB_BITS - _cnt; \ } while (0) /* MPFR_LCONV_DPTS can also be forced to 0 or 1 by the user. */ #ifndef MPFR_LCONV_DPTS # if defined(HAVE_LOCALE_H) && \ defined(HAVE_STRUCT_LCONV_DECIMAL_POINT) && \ defined(HAVE_STRUCT_LCONV_THOUSANDS_SEP) # define MPFR_LCONV_DPTS 1 # else # define MPFR_LCONV_DPTS 0 # endif #endif /* FIXME: Add support for multibyte decimal_point and thousands_sep since this can be found in practice: https://reviews.llvm.org/D27167 says: "I found this problem on FreeBSD 11, where thousands_sep in fr_FR.UTF-8 is a no-break space (U+00A0)." Under Linux, this is U+202F NARROW NO-BREAK SPACE (e2 80 af). And in the ps_AF locale, decimal_point = U+066B ARABIC DECIMAL SEPARATOR (d9 ab) thousands_sep = U+066C ARABIC THOUSANDS SEPARATOR (d9 ac) In the mean time, in case of non-single-byte character, revert to the default value. */ #if MPFR_LCONV_DPTS #include <locale.h> /* Warning! In case of signed char, the value of MPFR_DECIMAL_POINT may be negative (the ISO C99 does not seem to forbid negative values). */ #define MPFR_DECIMAL_POINT \ (localeconv()->decimal_point[1] != '\0' ? \ (char) '.' : localeconv()->decimal_point[0]) #define MPFR_THOUSANDS_SEPARATOR \ (localeconv()->thousands_sep[0] == '\0' || \ localeconv()->thousands_sep[1] != '\0' ? \ (char) '\0' : localeconv()->thousands_sep[0]) #else #define MPFR_DECIMAL_POINT ((char) '.') #define MPFR_THOUSANDS_SEPARATOR ((char) '\0') #endif /* Size of an array, as a constant expression. */ #define numberof_const(x) (sizeof (x) / sizeof ((x)[0])) /* Size of an array, safe version but not a constant expression: Since an array can silently be converted to a pointer, we check that this macro is applied on an array, not a pointer. Also make sure that the type is signed ("long" is sufficient in practice since the sizes come from the MPFR source), so that the value can be used in arbitrary expressions without the risk of silently switching to unsigned arithmetic. */ /* TODO: Make numberof() a constant expression and always use it in the MPFR code instead of numberof_const(). See the tricks at https://gcc.gnu.org/pipermail/gcc/2020-September/233763.html "[PATCH v2] <sys/param.h>: Add nitems() and snitems() macros" by Alejandro Colomar but this needs to be fully tested on various platforms and with various compilers and compilation options. Moreover, change "long" to "ptrdiff_t", as used at the above URL? */ #undef numberof #if 0 /* The following should work with GCC as documented in its manual, but fails: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=38377#c10 Thus disabled for now. */ # define numberof(x) \ ( __extension__ ({ \ int is_array = (void *) &(x) == (void *) &(x)[0]; \ MPFR_STAT_STATIC_ASSERT (__builtin_constant_p (is_array) ? \ is_array : 1); \ MPFR_ASSERTN (is_array); \ (long) numberof_const (x); \ }) ) #else # define numberof(x) \ (MPFR_ASSERTN ((void *) &(x) == (void *) &(x)[0]), \ (long) numberof_const (x)) #endif /* Addition with carry (detected by GCC and other good compilers). */ #define ADD_LIMB(u,v,c) ((u) += (v), (c) = (u) < (v)) /* umul_hi(h, x, y) puts in h the high part of x*y */ /* MPFR_NEED_LONGLONG_H needs to be defined to use it. */ #define umul_hi(h, x, y) \ do { \ mp_limb_t _l; \ umul_ppmm (h, _l, x, y); \ (void) _l; /* unused variable */ \ } while (0) /****************************************************** ************ Save exponent/flags macros ************ ******************************************************/ /* See README.dev for details on how to use the macros. They are used to set the exponent range to the maximum temporarily */ typedef struct { mpfr_flags_t saved_flags; mpfr_exp_t saved_emin; mpfr_exp_t saved_emax; } mpfr_save_expo_t; #define MPFR_SAVE_EXPO_DECL(x) mpfr_save_expo_t x #define MPFR_SAVE_EXPO_MARK(x) \ ((x).saved_flags = __gmpfr_flags, \ (x).saved_emin = __gmpfr_emin, \ (x).saved_emax = __gmpfr_emax, \ __gmpfr_emin = MPFR_EMIN_MIN, \ __gmpfr_emax = MPFR_EMAX_MAX) #define MPFR_SAVE_EXPO_FREE(x) \ (__gmpfr_flags = (x).saved_flags, \ __gmpfr_emin = (x).saved_emin, \ __gmpfr_emax = (x).saved_emax) #define MPFR_SAVE_EXPO_UPDATE_FLAGS(x, flags) \ (x).saved_flags |= (flags) /* Speed up final checking */ #define mpfr_check_range(x,t,r) \ (MPFR_LIKELY (MPFR_EXP_IN_RANGE (MPFR_EXP (x))) \ ? ((t) ? (__gmpfr_flags |= MPFR_FLAGS_INEXACT, (t)) : 0) \ : mpfr_check_range(x,t,r)) /****************************************************** ***************** Inline rounding ****************** ******************************************************/ /* * Note: due to the labels, one cannot use a macro MPFR_RNDRAW* more than * once in a function (otherwise these labels would not be unique). */ /* * Round mantissa (srcp, sprec) to mpfr_t dest using rounding mode rnd * assuming dest's sign is sign. * In rounding to nearest mode, execute MIDDLE_HANDLER when the value * is the middle of two consecutive numbers in dest precision. * Execute OVERFLOW_HANDLER in case of overflow when rounding. * * Note: the exponent field of dest is not used, possibly except by the * handlers. It is the caller (via the handlers) who entirely decides * how to handle it. */ #define MPFR_RNDRAW_GEN(inexact, dest, srcp, sprec, rnd, sign, \ MIDDLE_HANDLER, OVERFLOW_HANDLER) \ do { \ mp_size_t _dests, _srcs; \ mp_limb_t *_destp; \ mpfr_prec_t _destprec, _srcprec; \ \ /* Check Trivial Case when Dest Mantissa has more bits than source */ \ _srcprec = (sprec); \ _destprec = MPFR_PREC (dest); \ MPFR_ASSERTD (_srcprec >= MPFR_PREC_MIN); \ MPFR_ASSERTD (_destprec >= MPFR_PREC_MIN); \ _destp = MPFR_MANT (dest); \ if (MPFR_UNLIKELY (_destprec >= _srcprec)) \ { \ _srcs = MPFR_PREC2LIMBS (_srcprec); \ _dests = MPFR_PREC2LIMBS (_destprec) - _srcs; \ MPN_COPY (_destp + _dests, srcp, _srcs); \ MPN_ZERO (_destp, _dests); \ inexact = 0; \ } \ else \ { \ /* Non trivial case: rounding needed */ \ mpfr_prec_t _sh; \ mp_limb_t *_sp; \ mp_limb_t _rb, _sb, _ulp; \ \ /* Compute Position and shift */ \ _srcs = MPFR_PREC2LIMBS (_srcprec); \ _dests = MPFR_PREC2LIMBS (_destprec); \ MPFR_UNSIGNED_MINUS_MODULO (_sh, _destprec); \ _sp = (srcp) + _srcs - _dests; \ \ /* General case when prec % GMP_NUMB_BITS != 0 */ \ if (MPFR_LIKELY (_sh != 0)) \ { \ mp_limb_t _mask; \ /* Compute Rounding Bit and Sticky Bit */ \ /* Note: in directed rounding modes, if the rounding bit */ \ /* is 1, the behavior does not depend on the sticky bit; */ \ /* thus we will not try to compute it in this case (this */ \ /* can be much faster and avoids to read uninitialized */ \ /* data in the current mpfr_mul implementation). We just */ \ /* make sure that _sb is initialized. */ \ _mask = MPFR_LIMB_ONE << (_sh - 1); \ _rb = _sp[0] & _mask; \ _sb = _sp[0] & (_mask - 1); \ if ((rnd) == MPFR_RNDN || _rb == 0) \ { \ mp_limb_t *_tmp; \ mp_size_t _n; \ for (_tmp = _sp, _n = _srcs - _dests ; \ _n != 0 && _sb == 0 ; _n--) \ _sb = *--_tmp; \ } \ _ulp = 2 * _mask; \ } \ else /* _sh == 0 */ \ { \ MPFR_ASSERTD (_dests < _srcs); \ /* Compute Rounding Bit and Sticky Bit - see note above */ \ _rb = _sp[-1] & MPFR_LIMB_HIGHBIT; \ _sb = _sp[-1] & (MPFR_LIMB_HIGHBIT-1); \ if ((rnd) == MPFR_RNDN || _rb == 0) \ { \ mp_limb_t *_tmp; \ mp_size_t _n; \ for (_tmp = _sp - 1, _n = _srcs - _dests - 1 ; \ _n != 0 && _sb == 0 ; _n--) \ _sb = *--_tmp; \ } \ _ulp = MPFR_LIMB_ONE; \ } \ /* Rounding */ \ if (rnd == MPFR_RNDF) \ { \ inexact = 0; \ goto trunc_doit; \ } \ else if (rnd == MPFR_RNDN) \ { \ if (_rb == 0) \ { \ trunc: \ inexact = MPFR_LIKELY ((_sb | _rb) != 0) ? -sign : 0; \ trunc_doit: \ MPN_COPY (_destp, _sp, _dests); \ _destp[0] &= ~(_ulp - 1); \ } \ else if (MPFR_UNLIKELY (_sb == 0)) \ { /* Middle of two consecutive representable numbers */ \ MIDDLE_HANDLER; \ } \ else \ { \ if (0) \ goto addoneulp_doit; /* dummy code to avoid warning */ \ addoneulp: \ inexact = sign; \ addoneulp_doit: \ if (MPFR_UNLIKELY (mpn_add_1 (_destp, _sp, _dests, _ulp))) \ { \ _destp[_dests - 1] = MPFR_LIMB_HIGHBIT; \ OVERFLOW_HANDLER; \ } \ _destp[0] &= ~(_ulp - 1); \ } \ } \ else \ { /* Directed rounding mode */ \ if (MPFR_IS_LIKE_RNDZ (rnd, MPFR_IS_NEG_SIGN (sign))) \ goto trunc; \ else if (MPFR_UNLIKELY ((_sb | _rb) == 0)) \ { \ inexact = 0; \ goto trunc_doit; \ } \ else \ goto addoneulp; \ } \ } \ } while (0) /* * Round mantissa (srcp, sprec) to mpfr_t dest using rounding mode rnd * assuming dest's sign is sign. * Execute OVERFLOW_HANDLER in case of overflow when rounding. */ #define MPFR_RNDRAW(inexact, dest, srcp, sprec, rnd, sign, OVERFLOW_HANDLER) \ MPFR_RNDRAW_GEN (inexact, dest, srcp, sprec, rnd, sign, \ if ((_sp[0] & _ulp) == 0) \ { \ inexact = -sign; \ goto trunc_doit; \ } \ else \ goto addoneulp; \ , OVERFLOW_HANDLER) /* * Round mantissa (srcp, sprec) to mpfr_t dest using rounding mode rnd * assuming dest's sign is sign. * Execute OVERFLOW_HANDLER in case of overflow when rounding. * Set inexact to +/- MPFR_EVEN_INEX in case of even rounding. */ #define MPFR_RNDRAW_EVEN(inexact, dest, srcp, sprec, rnd, sign, \ OVERFLOW_HANDLER) \ MPFR_RNDRAW_GEN (inexact, dest, srcp, sprec, rnd, sign, \ if ((_sp[0] & _ulp) == 0) \ { \ inexact = -MPFR_EVEN_INEX * sign; \ goto trunc_doit; \ } \ else \ { \ inexact = MPFR_EVEN_INEX * sign; \ goto addoneulp_doit; \ } \ , OVERFLOW_HANDLER) /* Return TRUE if b is non singular and we can round it to precision 'prec' and determine the ternary value, with rounding mode 'rnd', and with error at most 'error' */ #define MPFR_CAN_ROUND(b,err,prec,rnd) \ (!MPFR_IS_SINGULAR (b) && mpfr_round_p (MPFR_MANT (b), MPFR_LIMB_SIZE (b), \ (err), (prec) + ((rnd)==MPFR_RNDN))) /* Copy the sign and the significand, and handle the exponent in exp. */ #define MPFR_SETRAW(inexact,dest,src,exp,rnd) \ if (dest != src) \ { \ MPFR_SET_SIGN (dest, MPFR_SIGN (src)); \ if (MPFR_PREC (dest) == MPFR_PREC (src)) \ { \ MPN_COPY (MPFR_MANT (dest), MPFR_MANT (src), \ MPFR_LIMB_SIZE (src)); \ inexact = 0; \ } \ else \ { \ MPFR_RNDRAW (inexact, dest, MPFR_MANT (src), MPFR_PREC (src), \ rnd, MPFR_SIGN (src), exp++); \ } \ } \ else \ inexact = 0; /* TODO: fix this description (see round_near_x.c). */ /* Assuming that the function has a Taylor expansion which looks like: y=o(f(x)) = o(v + g(x)) with |g(x)| <= 2^(EXP(v)-err) we can quickly set y to v if x is small (ie err > prec(y)+1) in most cases. It assumes that f(x) is not representable exactly as a FP number. v must not be a singular value (NAN, INF or ZERO); usual values are v=1 or v=x. y is the destination (a mpfr_t), v the value to set (a mpfr_t), err1+err2 with 0 <= err2 <= 3 the error term (mpfr_exp_t's), dir (an int) is the direction of the committed error (if dir = 0, it rounds toward 0, if dir=1, it rounds away from 0), rnd the rounding mode. It returns from the function a ternary value in case of success. If you want to free something, you must fill the "extra" field in consequences, otherwise put nothing in it. The test is less restrictive than necessary, but the function will finish the check itself. Note: err1 + err2 is allowed to overflow as mpfr_exp_t, but it must give its real value as mpfr_uexp_t. */ #define MPFR_FAST_COMPUTE_IF_SMALL_INPUT(y,v,err1,err2,dir,rnd,extra) \ do { \ mpfr_ptr _y = (y); \ mpfr_exp_t _err1 = (err1); \ mpfr_exp_t _err2 = (err2); \ if (_err1 > 0) \ { \ mpfr_uexp_t _err = (mpfr_uexp_t) _err1 + _err2; \ if (MPFR_UNLIKELY (_err > MPFR_PREC (_y) + 1)) \ { \ int _inexact = mpfr_round_near_x (_y,(v),_err,(dir),(rnd)); \ if (_inexact != 0) \ { \ extra; \ return _inexact; \ } \ } \ } \ } while (0) /* Variant, to be called somewhere after MPFR_SAVE_EXPO_MARK. This variant is needed when there are some computations before or when some non-zero real constant is used, such as __gmpfr_one for mpfr_cos. */ #define MPFR_SMALL_INPUT_AFTER_SAVE_EXPO(y,v,err1,err2,dir,rnd,expo,extra) \ do { \ mpfr_ptr _y = (y); \ mpfr_exp_t _err1 = (err1); \ mpfr_exp_t _err2 = (err2); \ if (_err1 > 0) \ { \ mpfr_uexp_t _err = (mpfr_uexp_t) _err1 + _err2; \ if (MPFR_UNLIKELY (_err > MPFR_PREC (_y) + 1)) \ { \ int _inexact; \ MPFR_CLEAR_FLAGS (); \ _inexact = mpfr_round_near_x (_y,(v),_err,(dir),(rnd)); \ if (_inexact != 0) \ { \ extra; \ MPFR_SAVE_EXPO_UPDATE_FLAGS (expo, __gmpfr_flags); \ MPFR_SAVE_EXPO_FREE (expo); \ return mpfr_check_range (_y, _inexact, (rnd)); \ } \ } \ } \ } while (0) /****************************************************** ***************** Ziv loop macros ****************** ******************************************************/ /* To safely increase some precision, detecting integer overflows. This macro is particularly useful when determining the initial working precision before Ziv's loop. P is a precision, X is an arbitrary non-negative integer. Note: On 2012-02-23, the MPFR_PREC_MAX value has been decreased by 256 from the maximum value representable in the mpfr_prec_t type, in order to avoid some integer overflows when this macro is not used (if the result is larger than MPFR_PREC_MAX, this should be detected with a later assertion, e.g. in mpfr_init2). But this change is mainly for existing code that has not been updated yet. So, it is advised to always use MPFR_ADD_PREC or MPFR_INC_PREC if the result can be larger than MPFR_PREC_MAX. */ #define MPFR_ADD_PREC(P,X) \ (MPFR_ASSERTN ((X) <= MPFR_PREC_MAX - (P)), (P) + (X)) #define MPFR_INC_PREC(P,X) \ (MPFR_ASSERTN ((X) <= MPFR_PREC_MAX - (P)), (P) += (X)) #ifndef MPFR_USE_LOGGING #define MPFR_ZIV_DECL(_x) mpfr_prec_t _x #define MPFR_ZIV_INIT(_x, _p) (_x) = GMP_NUMB_BITS #define MPFR_ZIV_NEXT(_x, _p) (MPFR_INC_PREC (_p, _x), (_x) = (_p)/2) #define MPFR_ZIV_FREE(x) #else /* The following test on glibc is there mainly for Darwin (Mac OS X), to obtain a better error message. The real test should have been a test concerning nested functions in gcc, which are disabled by default on Darwin; but it is not possible to do that without a configure test. */ # if defined (__cplusplus) || !(__MPFR_GNUC(3,0) && __MPFR_GLIBC(2,0)) # error "Logging not supported (needs gcc >= 3.0 and GNU C Library >= 2.0)." # endif /* Use LOGGING */ /* Note: the mpfr_log_level >= 0 below avoids to take into account Ziv loops used by the MPFR functions called by the mpfr_fprintf in LOG_PRINT. */ #define MPFR_ZIV_DECL(_x) \ mpfr_prec_t _x; \ int _x ## _cpt = 1; \ static unsigned long _x ## _loop = 0, _x ## _bad = 0; \ static const char *_x ## _fname = __func__; \ auto void __attribute__ ((destructor)) x ## _f (void); \ void __attribute__ ((destructor)) x ## _f (void) { \ if (_x ## _loop != 0 && (MPFR_LOG_STAT_F & mpfr_log_type)) { \ fprintf (mpfr_log_file, \ "%s: Ziv failed %2.2f%% (%lu bad cases / %lu calls)\n", \ _x ## _fname, (double) 100.0 * _x ## _bad / _x ## _loop, \ _x ## _bad, _x ## _loop ); \ if (mpfr_log_flush) \ fflush (mpfr_log_file); \ } \ } #define MPFR_ZIV_INIT(_x, _p) \ do \ { \ (_x) = GMP_NUMB_BITS; \ if (mpfr_log_level >= 0) \ _x ## _loop ++; \ LOG_PRINT (MPFR_LOG_BADCASE_F, "%s:ZIV 1st prec=%Pd\n", \ __func__, (mpfr_prec_t) (_p)); \ } \ while (0) #define MPFR_ZIV_NEXT(_x, _p) \ do \ { \ MPFR_INC_PREC (_p, _x); \ (_x) = (_p) / 2; \ if (mpfr_log_level >= 0) \ _x ## _bad += (_x ## _cpt == 1); \ _x ## _cpt ++; \ LOG_PRINT (MPFR_LOG_BADCASE_F, "%s:ZIV new prec=%Pd\n", \ __func__, (mpfr_prec_t) (_p)); \ } \ while (0) #define MPFR_ZIV_FREE(_x) \ do \ if (_x ## _cpt > 1) \ LOG_PRINT (MPFR_LOG_BADCASE_F, "%s:ZIV %d loops\n", \ __func__, _x ## _cpt); \ while (0) #endif /****************************************************** ****************** Logging macros ****************** ******************************************************/ /* The different kind of LOG */ #define MPFR_LOG_INPUT_F 1 #define MPFR_LOG_OUTPUT_F 2 #define MPFR_LOG_INTERNAL_F 4 #define MPFR_LOG_TIME_F 8 #define MPFR_LOG_BADCASE_F 16 #define MPFR_LOG_MSG_F 32 #define MPFR_LOG_STAT_F 64 #ifdef MPFR_USE_LOGGING /* Check if we can support this feature */ # ifdef MPFR_USE_THREAD_SAFE # error "Enable either `Logging' or `thread-safe', not both" # endif # if !__MPFR_GNUC(3,0) # error "Logging not supported (GCC >= 3.0)" # endif #if defined (__cplusplus) extern "C" { #endif __MPFR_DECLSPEC extern FILE *mpfr_log_file; __MPFR_DECLSPEC extern int mpfr_log_flush; __MPFR_DECLSPEC extern int mpfr_log_type; __MPFR_DECLSPEC extern int mpfr_log_level; __MPFR_DECLSPEC extern int mpfr_log_current; __MPFR_DECLSPEC extern mpfr_prec_t mpfr_log_prec; #if defined (__cplusplus) } #endif /* LOG_PRINT calls mpfr_fprintf on mpfr_log_file with logging disabled (recursive logging is not wanted and freezes MPFR). */ #define LOG_PRINT(type, format, ...) \ do \ if ((mpfr_log_type & (type)) && mpfr_log_current <= mpfr_log_level) \ { \ int old_level = mpfr_log_level; \ mpfr_log_level = -1; /* disable logging in mpfr_fprintf */ \ __gmpfr_cache_const_pi = __gmpfr_logging_pi; \ __gmpfr_cache_const_log2 = __gmpfr_logging_log2; \ mpfr_fprintf (mpfr_log_file, format, __VA_ARGS__); \ if (mpfr_log_flush) \ fflush (mpfr_log_file); \ mpfr_log_level = old_level; \ __gmpfr_cache_const_pi = __gmpfr_normal_pi; \ __gmpfr_cache_const_log2 = __gmpfr_normal_log2; \ } \ while (0) #define MPFR_LOG_VAR(x) \ LOG_PRINT (MPFR_LOG_INTERNAL_F, "%s.%d:%s[%#Pu]=%.*Rg\n", __func__, \ (int) __LINE__, #x, mpfr_get_prec (x), mpfr_log_prec, x) #define MPFR_LOG_MSG2(format, ...) \ LOG_PRINT (MPFR_LOG_MSG_F, "%s.%d: "format, __func__, (int) __LINE__, \ __VA_ARGS__) #define MPFR_LOG_MSG(x) MPFR_LOG_MSG2 x #define MPFR_LOG_BEGIN2(format, ...) \ mpfr_log_current ++; \ LOG_PRINT (MPFR_LOG_INPUT_F, "%s:IN flags=%x "format"\n", __func__, \ (unsigned int) __gmpfr_flags, __VA_ARGS__); \ if ((MPFR_LOG_TIME_F & mpfr_log_type) && \ (mpfr_log_current <= mpfr_log_level)) \ __gmpfr_log_time = mpfr_get_cputime (); #define MPFR_LOG_BEGIN(x) \ int __gmpfr_log_time = 0; \ MPFR_LOG_BEGIN2 x #define MPFR_LOG_END2(format, ...) \ LOG_PRINT (MPFR_LOG_TIME_F, "%s:TIM %dms\n", __mpfr_log_fname, \ mpfr_get_cputime () - __gmpfr_log_time); \ LOG_PRINT (MPFR_LOG_OUTPUT_F, "%s:OUT flags=%x "format"\n", \ __mpfr_log_fname, (unsigned int) __gmpfr_flags, \ __VA_ARGS__); \ mpfr_log_current --; #define MPFR_LOG_END(x) \ static const char *__mpfr_log_fname = __func__; \ MPFR_LOG_END2 x #define MPFR_LOG_FUNC(begin,end) \ static const char *__mpfr_log_fname = __func__; \ auto void __mpfr_log_cleanup (int *time); \ void __mpfr_log_cleanup (int *time) { \ int __gmpfr_log_time = *time; \ MPFR_LOG_END2 end; } \ int __gmpfr_log_time __attribute__ ((cleanup (__mpfr_log_cleanup))); \ __gmpfr_log_time = 0; \ MPFR_LOG_BEGIN2 begin #else /* MPFR_USE_LOGGING */ /* Define void macro for logging */ #define MPFR_LOG_VAR(x) #define MPFR_LOG_BEGIN(x) #define MPFR_LOG_END(x) #define MPFR_LOG_MSG(x) #define MPFR_LOG_FUNC(x,y) #endif /* MPFR_USE_LOGGING */ /************************************************************** ************ Group Initialize Functions Macros ************* **************************************************************/ #ifndef MPFR_GROUP_STATIC_SIZE # define MPFR_GROUP_STATIC_SIZE 16 #endif struct mpfr_group_t { size_t alloc; mp_limb_t *mant; #if MPFR_GROUP_STATIC_SIZE != 0 mp_limb_t tab[MPFR_GROUP_STATIC_SIZE]; #else /* In order to detect memory leaks when testing, MPFR_GROUP_STATIC_SIZE can be set to 0, in which case tab will not be used. ISO C does not support zero-length arrays[*], thus let's use a flexible array member (which will be equivalent here). Note: this is new in C99, but this is just used for testing. [*] Zero-length arrays are a GNU extension: https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html and as such an extension is forbidden in ISO C, it triggers an error with -Werror=pedantic. */ mp_limb_t tab[]; #endif }; #define MPFR_GROUP_DECL(g) struct mpfr_group_t g #define MPFR_GROUP_CLEAR(g) do { \ MPFR_LOG_MSG (("GROUP_CLEAR: ptr = 0x%lX, size = %lu\n", \ (unsigned long) (g).mant, \ (unsigned long) (g).alloc)); \ if ((g).alloc != 0) { \ MPFR_ASSERTD ((g).mant != (g).tab); \ mpfr_free_func ((g).mant, (g).alloc); \ }} while (0) #define MPFR_GROUP_INIT_TEMPLATE(g, prec, num, handler) do { \ mpfr_prec_t _prec = (prec); \ mp_size_t _size; \ MPFR_ASSERTD (_prec >= MPFR_PREC_MIN); \ if (MPFR_UNLIKELY (_prec > MPFR_PREC_MAX)) \ mpfr_abort_prec_max (); \ _size = MPFR_PREC2LIMBS (_prec); \ if (_size * (num) > MPFR_GROUP_STATIC_SIZE) \ { \ (g).alloc = (num) * _size * sizeof (mp_limb_t); \ (g).mant = (mp_limb_t *) mpfr_allocate_func ((g).alloc); \ } \ else \ { \ (g).alloc = 0; \ (g).mant = (g).tab; \ } \ MPFR_LOG_MSG (("GROUP_INIT: ptr = 0x%lX, size = %lu\n", \ (unsigned long) (g).mant, (unsigned long) (g).alloc)); \ handler; \ } while (0) #define MPFR_GROUP_TINIT(g, n, x) \ MPFR_TMP_INIT1 ((g).mant + _size * (n), x, _prec) #define MPFR_GROUP_INIT_1(g, prec, x) \ MPFR_GROUP_INIT_TEMPLATE(g, prec, 1, MPFR_GROUP_TINIT(g, 0, x)) #define MPFR_GROUP_INIT_2(g, prec, x, y) \ MPFR_GROUP_INIT_TEMPLATE(g, prec, 2, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y)) #define MPFR_GROUP_INIT_3(g, prec, x, y, z) \ MPFR_GROUP_INIT_TEMPLATE(g, prec, 3, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z)) #define MPFR_GROUP_INIT_4(g, prec, x, y, z, t) \ MPFR_GROUP_INIT_TEMPLATE(g, prec, 4, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z);MPFR_GROUP_TINIT(g, 3, t)) #define MPFR_GROUP_INIT_5(g, prec, x, y, z, t, a) \ MPFR_GROUP_INIT_TEMPLATE(g, prec, 5, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z);MPFR_GROUP_TINIT(g, 3, t); \ MPFR_GROUP_TINIT(g, 4, a)) #define MPFR_GROUP_INIT_6(g, prec, x, y, z, t, a, b) \ MPFR_GROUP_INIT_TEMPLATE(g, prec, 6, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z);MPFR_GROUP_TINIT(g, 3, t); \ MPFR_GROUP_TINIT(g, 4, a);MPFR_GROUP_TINIT(g, 5, b)) #define MPFR_GROUP_REPREC_TEMPLATE(g, prec, num, handler) do { \ mpfr_prec_t _prec = (prec); \ size_t _oalloc = (g).alloc; \ mp_size_t _size; \ MPFR_LOG_MSG (("GROUP_REPREC: oldptr = 0x%lX, oldsize = %lu\n", \ (unsigned long) (g).mant, (unsigned long) _oalloc)); \ MPFR_ASSERTD (_prec >= MPFR_PREC_MIN); \ if (MPFR_UNLIKELY (_prec > MPFR_PREC_MAX)) \ mpfr_abort_prec_max (); \ _size = MPFR_PREC2LIMBS (_prec); \ (g).alloc = (num) * _size * sizeof (mp_limb_t); \ if (_oalloc == 0) \ (g).mant = (mp_limb_t *) mpfr_allocate_func ((g).alloc); \ else \ (g).mant = (mp_limb_t *) \ mpfr_reallocate_func ((g).mant, _oalloc, (g).alloc); \ MPFR_LOG_MSG (("GROUP_REPREC: newptr = 0x%lX, newsize = %lu\n", \ (unsigned long) (g).mant, (unsigned long) (g).alloc)); \ handler; \ } while (0) #define MPFR_GROUP_REPREC_1(g, prec, x) \ MPFR_GROUP_REPREC_TEMPLATE(g, prec, 1, MPFR_GROUP_TINIT(g, 0, x)) #define MPFR_GROUP_REPREC_2(g, prec, x, y) \ MPFR_GROUP_REPREC_TEMPLATE(g, prec, 2, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y)) #define MPFR_GROUP_REPREC_3(g, prec, x, y, z) \ MPFR_GROUP_REPREC_TEMPLATE(g, prec, 3, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z)) #define MPFR_GROUP_REPREC_4(g, prec, x, y, z, t) \ MPFR_GROUP_REPREC_TEMPLATE(g, prec, 4, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z);MPFR_GROUP_TINIT(g, 3, t)) #define MPFR_GROUP_REPREC_5(g, prec, x, y, z, t, a) \ MPFR_GROUP_REPREC_TEMPLATE(g, prec, 5, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z);MPFR_GROUP_TINIT(g, 3, t); \ MPFR_GROUP_TINIT(g, 4, a)) #define MPFR_GROUP_REPREC_6(g, prec, x, y, z, t, a, b) \ MPFR_GROUP_REPREC_TEMPLATE(g, prec, 6, \ MPFR_GROUP_TINIT(g, 0, x);MPFR_GROUP_TINIT(g, 1, y); \ MPFR_GROUP_TINIT(g, 2, z);MPFR_GROUP_TINIT(g, 3, t); \ MPFR_GROUP_TINIT(g, 4, a);MPFR_GROUP_TINIT(g, 5, b)) /****************************************************** *************** Internal functions ***************** ******************************************************/ #if defined (__cplusplus) extern "C" { #endif MPFR_COLD_FUNCTION_ATTR __MPFR_DECLSPEC int mpfr_underflow (mpfr_ptr, mpfr_rnd_t, int); MPFR_COLD_FUNCTION_ATTR __MPFR_DECLSPEC int mpfr_overflow (mpfr_ptr, mpfr_rnd_t, int); __MPFR_DECLSPEC int mpfr_add1 (mpfr_ptr, mpfr_srcptr, mpfr_srcptr, mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_sub1 (mpfr_ptr, mpfr_srcptr, mpfr_srcptr, mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_add1sp (mpfr_ptr, mpfr_srcptr, mpfr_srcptr, mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_sub1sp (mpfr_ptr, mpfr_srcptr, mpfr_srcptr, mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_can_round_raw (const mp_limb_t *, mp_size_t, int, mpfr_exp_t, mpfr_rnd_t, mpfr_rnd_t, mpfr_prec_t); __MPFR_DECLSPEC int mpfr_set_1_2 (mpfr_ptr, mpfr_srcptr, mpfr_rnd_t, int); __MPFR_DECLSPEC int mpfr_cmp2 (mpfr_srcptr, mpfr_srcptr, mpfr_prec_t *); __MPFR_DECLSPEC long __gmpfr_ceil_log2 (double); __MPFR_DECLSPEC long __gmpfr_floor_log2 (double); __MPFR_DECLSPEC double __gmpfr_ceil_exp2 (double); __MPFR_DECLSPEC unsigned long __gmpfr_isqrt (unsigned long); __MPFR_DECLSPEC unsigned long __gmpfr_cuberoot (unsigned long); __MPFR_DECLSPEC int __gmpfr_int_ceil_log2 (unsigned long); __MPFR_DECLSPEC mpfr_exp_t mpfr_ceil_mul (mpfr_exp_t, int, int); __MPFR_DECLSPEC int mpfr_exp_2 (mpfr_ptr, mpfr_srcptr, mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_exp_3 (mpfr_ptr, mpfr_srcptr, mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_powerof2_raw (mpfr_srcptr); __MPFR_DECLSPEC int mpfr_powerof2_raw2 (const mp_limb_t *, mp_size_t); __MPFR_DECLSPEC int mpfr_pow_general (mpfr_ptr, mpfr_srcptr, mpfr_srcptr, mpfr_rnd_t, int, mpfr_save_expo_t *); __MPFR_DECLSPEC void mpfr_setmax (mpfr_ptr, mpfr_exp_t); __MPFR_DECLSPEC void mpfr_setmin (mpfr_ptr, mpfr_exp_t); __MPFR_DECLSPEC int mpfr_mpn_exp (mp_limb_t *, mpfr_exp_t *, int, mpfr_exp_t, size_t); #ifdef _MPFR_H_HAVE_FILE __MPFR_DECLSPEC void mpfr_fdump (FILE *, mpfr_srcptr); #endif __MPFR_DECLSPEC void mpfr_print_mant_binary (const char*, const mp_limb_t*, mpfr_prec_t); __MPFR_DECLSPEC void mpfr_set_str_binary (mpfr_ptr, const char*); __MPFR_DECLSPEC int mpfr_round_raw (mp_limb_t *, const mp_limb_t *, mpfr_prec_t, int, mpfr_prec_t, mpfr_rnd_t, int *); __MPFR_DECLSPEC int mpfr_round_raw_2 (const mp_limb_t *, mpfr_prec_t, int, mpfr_prec_t, mpfr_rnd_t); /* No longer defined (see round_prec.c). Uncomment if it needs to be defined again. __MPFR_DECLSPEC int mpfr_round_raw_3 (const mp_limb_t *, mpfr_prec_t, int, mpfr_prec_t, mpfr_rnd_t, int *); */ __MPFR_DECLSPEC int mpfr_round_raw_4 (mp_limb_t *, const mp_limb_t *, mpfr_prec_t, int, mpfr_prec_t, mpfr_rnd_t); #define mpfr_round_raw2(xp, xn, neg, r, prec) \ mpfr_round_raw_2((xp),(xn)*GMP_NUMB_BITS,(neg),(prec),(r)) __MPFR_DECLSPEC int mpfr_check (mpfr_srcptr); __MPFR_DECLSPEC int mpfr_get_cputime (void); __MPFR_DECLSPEC void mpfr_nexttozero (mpfr_ptr); __MPFR_DECLSPEC void mpfr_nexttoinf (mpfr_ptr); __MPFR_DECLSPEC int mpfr_const_pi_internal (mpfr_ptr,mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_const_log2_internal (mpfr_ptr,mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_const_euler_internal (mpfr_ptr, mpfr_rnd_t); __MPFR_DECLSPEC int mpfr_const_catalan_internal (mpfr_ptr, mpfr_rnd_t); #if 0 __MPFR_DECLSPEC void mpfr_init_cache (mpfr_cache_t, int(*)(mpfr_ptr,mpfr_rnd_t)); #endif __MPFR_DECLSPEC void mpfr_clear_cache (mpfr_cache_t); __MPFR_DECLSPEC int mpfr_cache (mpfr_ptr, mpfr_cache_t, mpfr_rnd_t); __MPFR_DECLSPEC void mpfr_mulhigh_n (mpfr_limb_ptr, mpfr_limb_srcptr, mpfr_limb_srcptr, mp_size_t); __MPFR_DECLSPEC void mpfr_mullow_n (mpfr_limb_ptr, mpfr_limb_srcptr, mpfr_limb_srcptr, mp_size_t); __MPFR_DECLSPEC void mpfr_sqrhigh_n (mpfr_limb_ptr, mpfr_limb_srcptr, mp_size_t); __MPFR_DECLSPEC mp_limb_t mpfr_divhigh_n (mpfr_limb_ptr, mpfr_limb_ptr, mpfr_limb_ptr, mp_size_t); __MPFR_DECLSPEC int mpfr_round_p (mp_limb_t *, mp_size_t, mpfr_exp_t, mpfr_prec_t); __MPFR_DECLSPEC int mpfr_round_near_x (mpfr_ptr, mpfr_srcptr, mpfr_uexp_t, int, mpfr_rnd_t); __MPFR_DECLSPEC MPFR_COLD_FUNCTION_ATTR MPFR_NORETURN void mpfr_abort_prec_max (void); __MPFR_DECLSPEC void mpfr_rand_raw (mpfr_limb_ptr, gmp_randstate_t, mpfr_prec_t); __MPFR_DECLSPEC mpz_srcptr mpfr_bernoulli_cache (unsigned long); __MPFR_DECLSPEC void mpfr_bernoulli_freecache (void); __MPFR_DECLSPEC int mpfr_sincos_fast (mpfr_ptr, mpfr_ptr, mpfr_srcptr, mpfr_rnd_t); __MPFR_DECLSPEC double mpfr_scale2 (double, int); __MPFR_DECLSPEC void mpfr_div_ui2 (mpfr_ptr, mpfr_srcptr, unsigned long, unsigned long, mpfr_rnd_t); __MPFR_DECLSPEC void mpfr_gamma_one_and_two_third (mpfr_ptr, mpfr_ptr, mpfr_prec_t); __MPFR_DECLSPEC void mpfr_mpz_init (mpz_ptr); __MPFR_DECLSPEC void mpfr_mpz_init2 (mpz_ptr, mp_bitcnt_t); __MPFR_DECLSPEC void mpfr_mpz_clear (mpz_ptr); __MPFR_DECLSPEC int mpfr_odd_p (mpfr_srcptr); __MPFR_DECLSPEC int mpfr_nbits_ulong (unsigned long); #ifdef _MPFR_H_HAVE_INTMAX_T __MPFR_DECLSPEC int mpfr_nbits_uj (uintmax_t); #endif #ifdef _MPFR_H_HAVE_VA_LIST /* Declared only if <stdarg.h> has been included. */ __MPFR_DECLSPEC int mpfr_vasnprintf_aux (char**, char*, size_t, const char*, va_list); #endif #if MPFR_WANT_ASSERT >= 2 __MPFR_DECLSPEC void flags_fout (FILE *, mpfr_flags_t); #endif #if defined (__cplusplus) } #endif /***************************************************** *************** Internal mpz_t pool *************** *****************************************************/ /* don't use the mpz_t pool with mini-gmp */ #ifdef MPFR_USE_MINI_GMP # define MPFR_POOL_NENTRIES 0 #endif #ifndef MPFR_POOL_NENTRIES # define MPFR_POOL_NENTRIES 32 /* default number of entries of the pool */ #endif #if MPFR_POOL_NENTRIES && !defined(MPFR_POOL_DONT_REDEFINE) # undef mpz_init # undef mpz_init2 # undef mpz_clear # undef mpz_init_set_ui # undef mpz_init_set # define mpz_init mpfr_mpz_init # define mpz_init2 mpfr_mpz_init2 # define mpz_clear mpfr_mpz_clear # define mpz_init_set_ui(a,b) do { mpz_init (a); mpz_set_ui (a, b); } while (0) # define mpz_init_set(a,b) do { mpz_init (a); mpz_set (a, b); } while (0) #endif /****************************************************** ******** Compute LOG2(LOG2(MPFR_PREC_MAX)) ********* ******************************************************/ #if _MPFR_PREC_FORMAT == 1 # define MPFR_PREC_MAX_TEMP USHRT_MAX #elif _MPFR_PREC_FORMAT == 2 # define MPFR_PREC_MAX_TEMP UINT_MAX #elif _MPFR_PREC_FORMAT == 3 # define MPFR_PREC_MAX_TEMP ULONG_MAX #else # error "Invalid MPFR Prec format" #endif /* Note: In the constants below, it is sufficient to use the suffix U. * A large enough unsigned type will be chosen automatically, but the * exact type doesn't matter here. */ #if MPFR_PREC_MAX_TEMP == 255U # define MPFR_PREC_BITS 8 # define MPFR_LOG2_PREC_BITS 3 #elif MPFR_PREC_MAX_TEMP == 65535U # define MPFR_PREC_BITS 16 # define MPFR_LOG2_PREC_BITS 4 #elif MPFR_PREC_MAX_TEMP == 4294967295U # define MPFR_PREC_BITS 32 # define MPFR_LOG2_PREC_BITS 5 #elif MPFR_PREC_MAX_TEMP == 18446744073709551615U # define MPFR_PREC_BITS 64 # define MPFR_LOG2_PREC_BITS 6 #else # error "Unsupported MPFR_PREC_MAX_TEMP value" #endif /****************************************************** ************* Value coverage checking ************** ******************************************************/ #ifdef MPFR_COV_CHECK /* Variable names should start with the __gmpfr_cov_ prefix. */ #define MPFR_COV_SET(X) (__gmpfr_cov_ ## X = 1) #if defined (__cplusplus) extern "C" { #endif __MPFR_DECLSPEC extern int __gmpfr_cov_div_ui_sb[10][2]; __MPFR_DECLSPEC extern int __gmpfr_cov_sum_tmd[MPFR_RND_MAX][2][2][3][2][2]; #if defined (__cplusplus) } #endif #else /* MPFR_COV_CHECK */ #define MPFR_COV_SET(X) ((void) 0) #endif /* MPFR_COV_CHECK */ /****************************************************** ***************** Unbounded Floats ***************** ******************************************************/ #if defined (__cplusplus) extern "C" { #endif /* An UBF is like a MPFR number, but with an additional mpz_t member, which is assumed to be present (with a value in it) when the usual exponent field has the value MPFR_EXP_UBF. The goal of this compatible representation is to easily be able to support UBF in "normal" code using the public API. This is some form of "subtyping". Unfortunately this breaks aliasing rules, and C does not provide any way to avoid that (except with additional syntax ugliness and API breakage, though there is a workaround -- see the end of this comment): https://news.ycombinator.com/item?id=11753236 The alignment requirement for __mpfr_ubf_struct (UBF) needs to be at least as strong as the one for __mpfr_struct (MPFR number); this is not required by the C standard, but this should always be the case in practice, since __mpfr_ubf_struct starts with the same members as those of __mpfr_struct. If ever this would not be the case with some particular C implementation, an _Alignas alignment attribute (C11) could be added for UBF. When an input of a public function is an UBF, the semantic remains internal to MPFR and can change in the future. UBF arguments need to be explicitly converted to mpfr_ptr (or mpfr_srcptr); be careful with variadic functions, as the compiler will not be able to check in general. See fmma.c as an example of usage. In general, the type used for values that may be UBF must be either mpfr_ubf_t or mpfr_ubf_ptr. The type mpfr_ptr or mpfr_srcptr may be used for UBF only in the case where the pointer has been converted from mpfr_ubf_ptr, in order to ensure valid alignment. For instance, in mpfr_fmma_aux, one uses mpfr_ubf_t to generate the exact products as UBF; then the corresponding pointers are converted to mpfr_srcptr for mpfr_add (even though they point to UBF). Functions that can accept either MPFR arguments (mpfr_ptr type) or UBF arguments (mpfr_ubf_ptr type) must use a pointer type that can always be converted from both, typically mpfr_ptr or mpfr_srcptr. For instance, that's why mpfr_ubf_exp_less_p uses mpfr_srcptr. Note: "void *" could also be used, but mpfr_ptr is more meaningful and practical. Note that functions used for logging need to support UBF (currently done by printing that a number is an UBF, as it may be difficult to do more without significant changes). -------- A workaround to avoid breaking aliasing rules should be to use mpfr_ptr to access the usual mpfr_t members and mpfr_ubf_ptr to access the additional member _mpfr_zexp. And never use __mpfr_ubf_struct as a declared type; otherwise this would force __mpfr_ubf_struct to be the effective type of the whole object. Thus instead of typedef __mpfr_ubf_struct mpfr_ubf_t[1]; one could use the following definition as a trick to allocate an UBF as an automatic variable with the required alignment but without forcing the effective type to __mpfr_ubf_struct. typedef union { __mpfr_ubf_struct u[1]; __mpfr_struct m[1]; } mpfr_ubf_t; Then adapt the related code to select to right member, depending on the context. Unfortunately, this triggers -Wstrict-aliasing false positives with GCC in the MPFR_UBF_CLEAR_EXP expansion: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94337 (see changeset r13820 in the ubf2 branch). So, for the time being, as long as the code does not break, do not change anything. Note: The condition "use mpfr_ptr to access the usual mpfr_t members and mpfr_ubf_ptr to access the additional member _mpfr_zexp" may be ignored if the union type is visible within the function (see ISO C99 6.5.2.3#5 and 6.5.2.3#8 for the example, this implementation being very similar to the valid fragment of this example), which must be the case as the union is declared globally. However, this seems to be buggy in GCC: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=14319 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65892 Alternatively, GCC's may_alias attribute could conditionally be used on the __mpfr_ubf_struct and __mpfr_struct types (though it would be much stronger than needed since only these two types may alias each other). */ typedef struct { mpfr_prec_t _mpfr_prec; mpfr_sign_t _mpfr_sign; mpfr_exp_t _mpfr_exp; mp_limb_t *_mpfr_d; mpz_t _mpfr_zexp; } __mpfr_ubf_struct; typedef __mpfr_ubf_struct mpfr_ubf_t[1]; typedef __mpfr_ubf_struct *mpfr_ubf_ptr; __MPFR_DECLSPEC void mpfr_ubf_mul_exact (mpfr_ubf_ptr, mpfr_srcptr, mpfr_srcptr); __MPFR_DECLSPEC int mpfr_ubf_exp_less_p (mpfr_srcptr, mpfr_srcptr); __MPFR_DECLSPEC mpfr_exp_t mpfr_ubf_zexp2exp (mpz_ptr); __MPFR_DECLSPEC mpfr_exp_t mpfr_ubf_diff_exp (mpfr_srcptr, mpfr_srcptr); #if defined (__cplusplus) } #endif /* Get the _mpfr_zexp field (pointer to a mpz_t) of an UBF object. For practical reasons, the type of the argument x can be either mpfr_ubf_ptr or mpfr_ptr, since the latter is used in functions that accept both MPFR numbers and UBF's; this is checked by the code "(x)->_mpfr_exp" (the "sizeof" prevents an access, which could be invalid when MPFR_ZEXP(x) is used for an assignment, and also avoids breaking the aliasing rules if they are dealt with in the future). This macro can be used when building an UBF. So we do not check that the _mpfr_exp field has the value MPFR_EXP_UBF. */ #define MPFR_ZEXP(x) \ ((void) sizeof ((x)->_mpfr_exp), \ ((mpfr_ubf_ptr) (x))->_mpfr_zexp) /* If x is an UBF, clear its mpz_t exponent. */ #define MPFR_UBF_CLEAR_EXP(x) \ ((void) (MPFR_IS_UBF (x) && (mpz_clear (MPFR_ZEXP (x)), 0))) /* Like MPFR_GET_EXP, but accepts UBF (with exponent saturated to the interval [MPFR_EXP_MIN,MPFR_EXP_MAX]). */ #define MPFR_UBF_GET_EXP(x) \ (MPFR_IS_UBF (x) ? mpfr_ubf_zexp2exp (MPFR_ZEXP (x)) : \ MPFR_GET_EXP ((mpfr_ptr) (x))) #endif /* __MPFR_IMPL_H__ */