On Wed, 6 Aug 2025, Tobias Burnus wrote: > This patch updates contrib/download_prerequisites to download > newer versions (the latest) of GMP, MPFR and MPC. > > Main reason is a newer MPFR that provides C23 functions, > additionally, the usual bug fixes or the new "away from zero" > rounding in MPC might turn out to be useful. GMP also claims > better support for some newer archs. > > Current GCC mainline already uses the MPFR 4.2+ functions for > (a)sinpi, (a)cospi, (a)tanpi, and atan2pi, where (e.g.) > 'sinpi(n) == sin(n*pi)': In the middle end as C23 builtin > and additionally in the gfortran FE as those are also F2023 > functions. There are additional C23 functions that will profit > from MPFR support, but still have to be added. See PR 118592 > (esp. comment 2) for the list - and below for the list of > new MPFR functions 4.2+ brings. > > Note: ISL (0.24, latest 0.27) and gettext (0.22, latest 0.26) > have not been updated; that's deferred to a future patch (soon, > if deemed useful, or eventually). See below for some details. > > * * * > > Remarks: > > * download_prerequisites uses > http://gcc.gnu.org/pub/gcc/infrastructure/ > i.e. before the commit, the new files have to be placed > there. > > * download_prerequisites is just a convenience script; > users could freely use a newer or older version of the > libraries (or none for optional ones like ISL) - what > counts is: https://gcc.gnu.org/install/prerequisites.html > > Still, it makes sense that it works for most or all > out of the box, esp. on older, non-standard systems. > > * Newer ISL versions require C++17, such that a build will > fail with older compiler like, e.g., GCC-10.5 fails (albeit > some older ones work with -std=c++1y). - GCC currently > requires an ISO C++14 compiler (>= GCC 5.4). As there are > no ISL changes that seem to be extremely important, an > update does not seem to be sensible. > > * Newer gettext version: Gettext 0.26 (released Jul 20, 2025) > contains a long-overdue update of the gcc-internal-format > and gfc-internal-format checkings, useful to avoid argument > bugs in the .po translation. > Thus, there is some merit in doing an update, but not in > this patch - and some more testing/checking seems to be > sensible before doing it. > > * For GMP, I see that for Fedora, a C23 patch exists for > 'configure', but for SUSE not. It also seems to build fine > with GCC 15 here, i.e. the patch does not seem to be > required and the old, currently downloaded GMP has > the same configure check, unsurprisingly. > There are a few other changes to GMP in the distros, but > none seem to be really affected by 6.2 → 6.3, but I might > have missed something. > MPFR and MPC do not seem to have any changes in distros. > > [As all libraries are out for a while, the hope is that > serious build errors would have lead to a new minor release, > but that's just a hope.] > > * * * > > The GMP, MPFR, MPC version, used for the patch and to be added > to http://gcc.gnu.org/pub/gcc/infrastructure/ are: > > * GMP 6.3.0 (2023-07-30, latest) [currently: 6.2.0] > - https://gmplib.org/ > - For better compatibility, not the .xz file but the .bz2 file > has been used → https://gmplib.org/download/gmp/ > > * MPFR 4.2.2 (2025-03-20, latest) [currently: 4.1.0]: > - https://www.mpfr.org/mpfr-current/ > > * MPC 1.3.1 (December 2022, latest) [currently: 1.2.1]: > - https://www.multiprecision.org/mpc/download.html > - Requires GMP 5.0.0+ and MPFR 4.1.0+ > > * * * > > Testing: > > An in-tree bootstrap GCC mainline with only the three libraries > suggested has been done on: > > * x86-64 GNU Linux with a GCC 15 system compiler > + running the testsuite (which hopefully picked up the just > build libraries and not the system's ones). > > * on cfarm13 5.10.0-32-amd64 with GCC 10.2.1 (Debian 11/bullseye) > + a partial testsuite run > > * on cfarm70 4.9.0-19-amd64 (Debian 9.13 (stretch); gcc 6.3.0) > successfully building the stage1 compiler but stopping half-way > through stage1 library build. (Slow system!) > > > Otherwise, on #gcc IRC on Tue Jul 15 2025: > > - Iain pointed out to the ISL issue and remarked: > "AFAIK the other pre-reqs build fine in-tree (I've been using gmp-6.3, > mpfr 4.2.1, mpc-1.2.1 and gettext-0.22.5) - but sure I tried mpc-1.3.1 > on at least one system." > > * * * > > OK for mainline? (Requires a prior update of 'infrastructure').
-gmp='gmp-6.2.1.tar.bz2' -mpfr='mpfr-4.1.0.tar.bz2' -mpc='mpc-1.2.1.tar.gz' +gmp='gmp-6.3.0.tar.xz' +mpfr='mpfr-4.2.2.tar.bz2' +mpc='mpc-1.3.0.tar.gz' isl='isl-0.24.tar.bz2' gettext='gettext-0.22.tar.gz' diff --git a/contrib/prerequisites.md5 b/contrib/prerequisites.md5 index 716a9ff910c..96b9802d901 100644 --- a/contrib/prerequisites.md5 +++ b/contrib/prerequisites.md5 @@ -1,5 +1,5 @@ -28971fc21cf028042d4897f02fd355ea gmp-6.2.1.tar.bz2 -44b892bc5a45bafb4294d134e13aad1d mpfr-4.1.0.tar.bz2 -9f16c976c25bb0f76b50be749cd7a3a8 mpc-1.2.1.tar.gz +c1cd6ef33085e9cb818b9b08371f9000 gmp-6.3.0.tar.bz2 above you have gmp-6.3.0.tar.xz but the md5sums have gmp-6.3.0.tar.bz2 I have placed the respective .bz2 files in infrastructure. Richard. > * * * > > Tobias > > PS: CHANGES from the respective NEW files. > I additionally note that MPFR's configure now uses '#include "gmp.c"' > instead of '#include <gmp.h>' which together with other configure > changes (see below) might actually improve the build experience. > > > GMP 6.2.0 to 6.3.1 - from the NEWS file: > > BUGS FIXED > * A possible overflow of type int is avoided for mpz_cmp on huge operands. > * A possible error condition when a malformed file is read with > mpz_inp_raw is now correctly handled. > > FEATURES > * New public function mpz_prevprime, companion of the existing > mpz_nextprime. > * New documented pointer types mpz_ptr, mpz_srcptr, and similar for > other GMP types. Refer to the manual for full list and suggested > usage. These types have been present in gmp.h at least since > GMP-4.0, but previously not advertised to users. > * Support for 64-bit Arm under Macos. > * Support for the loongarch64 CPU family. > * Support for building with LTO, link-time optimisations. > > SPEEDUPS > * New special code for base = 2 in mpz_powm reduces the average time > for the functions that test primality. > * Speedup for the function mpz_nextprime on large operands. > * Speedup for multiplications (some sizes only) thanks to new > internal functions to compute small negacyclic products. > * Special assembly code for IBM z13 and later "mainframe" CPUs, resulting in > a huge speedup. > * Improved assembly for several 64-bit x86 CPUs, Risc-V, 64-bit Arm. > > > MPFR 4.1.0 to 4.2.2 - from the NEWS file: > > - In order to resolve a portability issue with the _Float128 fallback to > __float128 for binary128 support (e.g. with Clang and glibc 2.41), the > prototypes of the corresponding conversion functions had to be changed, > with _Float128 replaced by mpfr_float128, where mpfr_float128 is a macro > defined as _Float128 by default. This changes neither the ABI nor the API > (except that the end user of MPFR would need to define mpfr_float128 as > the actual type for the binary128 format if this is not the standard > _Float128 type). > - Other bug fixes (see <https://www.mpfr.org/mpfr-4.2.1/#fixed> and/or the > ChangeLog file). In particular, the formatted output functions behaved > incorrectly with %c on the value 0; such a use is uncommon, but this bug > may have security implications. > - Improved MPFR manual. > - Detect the use of GMP's buggy vsnprintf replacement at configure time. > With it, the tests of "%a" will be disabled to avoid an assertion failure > in the MPFR testsuite. A warning will be displayed in the configure output > in such a case. > - Bug fixes (see <https://www.mpfr.org/mpfr-4.2.0/#fixed> and/or the > ChangeLog file). > - Improved MPFR manual. > - Configure tests: replaced the test of the link with GMP, in order to > avoid the use of a function without a prototype (Autoconf issue), as > this is obsolescent in ISO C. The new test should be more robust. > - The "fondue savoyarde" release. > - Binary compatible with MPFR 4.0.* and 4.1.*, though some minor changes in > the behavior of the formatted output functions may be visible, regarded > as underspecified behavior or bug fixes (see below). > - New functions mpfr_cosu, mpfr_sinu, mpfr_tanu, mpfr_acosu, mpfr_asinu, > mpfr_atanu and mpfr_atan2u. > - New functions mpfr_cospi, mpfr_sinpi, mpfr_tanpi, mpfr_acospi, mpfr_asinpi, > mpfr_atanpi and mpfr_atan2pi. > - New functions mpfr_log2p1, mpfr_log10p1, mpfr_exp2m1, mpfr_exp10m1 and > mpfr_compound_si. > - New functions mpfr_fmod_ui, mpfr_powr, mpfr_pown, mpfr_pow_uj, mpfr_pow_sj > and mpfr_rootn_si (mpfr_pown is actually a macro defined as an alias for > mpfr_pow_sj). > - Bug fixes. > In particular, for the formatted output functions (mpfr_printf, etc.), > the case where the precision consists only of a period has been fixed > to be like ".0" as specified in the ISO C standard, and the manual has > been corrected and clarified. > The macros of the custom interface have also been fixed: they now behave > like functions (except a minor limitation for mpfr_custom_init_set). > > > MPC 1.2.1 to 1.3.1 - from the NEWS file: > > - Bug fix: It is again possible to include mpc.h without including > stdio.h. > - New function: mpc_agm > - New rounding modes "away from zero", indicated by the letter "A" and > corresponding to MPFR_RNDA on the designated real or imaginary part. > - New experimental ball arithmetic. > - New experimental function: mpc_eta_fund > - Bug fixes: > - mpc_asin for asin(z) with small |Re(z)| and tiny |Im(z)| > - mpc_pow_fr: sign of zero part of result when the base has up to sign > the same real and imaginary part, and the exponent is an even > positive integer > - mpc_fma: the returned 'int' value was incorrect in some cases > (indicating > whether the rounded real/imaginary parts were smaller/equal/greater > than the exact values), but the computed complex value was correct. > - Remove the unmaintained Makefile.vc; build files for Visual Studio > can be found at https://github.com/BrianGladman/mpc . > -- Richard Biener <rguent...@suse.de> SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg, Germany; GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)
