emerge crossdev amd then crossdev -t nvptx fails leads to GNU nano 8.2 /var/tmp/portage/cross-gcc14+/binutils-2.42-r2/temp/build.log ^[[32;01m*^[[0m Applying ppc64le/2.4.4 patch ... ^[[32;01m*^[[0m Applying clang-runtime/2.4.7 patch ... ^[[32;01m*^[[0m Running elibtoolize in: binutils-2.42/gprofng/libcollector/ ^[[32;01m*^[[0m Applying ppc64le/2.4.4 patch ... ^[[32;01m*^[[0m Applying clang-runtime/2.4.7 patch ... ^[[32;01m*^[[0m Running elibtoolize in: binutils-2.42/ld/ ^[[32;01m*^[[0m Applying ppc64le/2.4.4 patch ... ^[[32;01m*^[[0m Applying clang-runtime/2.4.7 patch ... ^[[32;01m*^[[0m Running elibtoolize in: binutils-2.42/libctf/ ^[[32;01m*^[[0m Applying ppc64le/2.4.4 patch ... ^[[32;01m*^[[0m Running elibtoolize in: binutils-2.42/libiberty/ ^[[32;01m*^[[0m Running elibtoolize in: binutils-2.42/libsframe/ ^[[32;01m*^[[0m Applying ppc64le/2.4.4 patch ... ^[[32;01m*^[[0m Running elibtoolize in: binutils-2.42/opcodes/ ^[[32;01m*^[[0m Applying ppc64le/2.4.4 patch ... ^[[32;01m*^[[0m Running elibtoolize in: binutils-2.42/zlib/ ^[[32;01m*^[[0m Applying ppc64le/2.4.4 patch ... >>> Source prepared. >>> Configuring source in /var/tmp/portage/cross-gcc14+/binutils-2.42-r2/work/binutils-2.42 ... ^[[32m*^[[0m CATEGORY: cross-gcc14+ ^[[32m*^[[0m CBUILD: x86_64-pc-linux-gnu ^[[32m*^[[0m CHOST: x86_64-pc-linux-gnu ^[[32m*^[[0m CTARGET: gcc14+ ^[[32m*^[[0m CFLAGS: -march=native -O3 -pipe -std=gnu17 ^[[32m*^[[0m LDFLAGS: -Wl,-O1 -Wl,--as-needed -Wl,-z,pack-relative-relocs -Wl,--undefined-version * econf: updating binutils-2.42/config.sub with /usr/share/gnuconfig/config.sub * econf: updating binutils-2.42/config.guess with /usr/share/gnuconfig/config.guess /var/tmp/portage/cross-gcc14+/binutils-2.42-r2/work/binutils-2.42/configure --prefix=/usr --build=x86_64-pc-linux-gnu --host=x86_64-pc-linux-gnu --target=gcc14+ --mandir=/usr/share/man --infodir=/usr/share/info --datadir=/usr/share --s> configure: loading site script /usr/share/config.site configure: loading site script /usr/share/config.site.d/80crossdev.conf checking build system type... x86_64-pc-linux-gnu checking host system type... x86_64-pc-linux-gnu checking target system type... Invalid configuration 'gcc14+': machine 'gcc14+-unknown' not recognized configure: error: /bin/sh /var/tmp/portage/cross-gcc14+/binutils-2.42-r2/work/binutils-2.42/config.sub gcc14+ failed !!! Please attach the following file when seeking support: !!! /var/tmp/portage/cross-gcc14+/binutils-2.42-r2/work/build/config.log ^[[31;01m*^[[0m ERROR: cross-gcc14+/binutils-2.42-r2::escpr2 failed (configure phase): ^[[31;01m*^[[0m econf failed ^[[31;01m*^[[0m ^[[31;01m*^[[0m Call stack: ^[[31;01m*^[[0m ebuild.sh, line 136: Called src_configure ^[[31;01m*^[[0m environment, line 1815: Called econf '--enable-plugins' '--without-included-gettext' '--with-system-zlib' '--build=x86_64-pc-linux-gnu' '--with-sysroot=/usr/gcc14+' '--enable-poison-system-directories' '--> ^[[31;01m*^[[0m phase-helpers.sh, line 647: Called __helpers_die 'econf failed' ^[[31;01m*^[[0m isolated-functions.sh, line 112: Called die ^[[31;01m*^[[0m The specific snippet of code: ^[[31;01m*^[[0m die "$@" Reproducible: Always emerge --info Portage 3.0.66.1 (python 3.12.7-final-0, default/linux/amd64/23.0/desktop/plasma, gcc-14, glibc-2.40-r5, 6.11.6-gentoo-x86_64 x86_64) ================================================================= System uname: Linux-6.11.6-gentoo-x86_64-x86_64-AMD_Ryzen_9_3900X_12-Core_Processor-with-glibc2.40 KiB Mem: 32792468 total, 15808008 free KiB Swap: 31249404 total, 29989884 free Timestamp of repository gentoo: Fri, 29 Nov 2024 03:00:00 +0000 Head commit of repository gentoo: abeee5fedd10e3af2bc45ec6f84be9896889671c Timestamp of repository escpr2: Thu, 31 Oct 2024 18:34:26 +0000 Head commit of repository escpr2: f2c923b8c651f1d14744975f879c2c78f6e5f8f9 sh bash 5.2_p37 ld GNU ld (Gentoo 2.43 p3) 2.43.1 app-misc/pax-utils: 1.3.8::gentoo app-shells/bash: 5.2_p37::gentoo dev-build/autoconf: 2.13-r8::gentoo, 2.71-r7::gentoo, 2.72-r1::gentoo dev-build/automake: 1.17-r1::gentoo dev-build/cmake: 3.31.1::gentoo dev-build/libtool: 2.5.4::gentoo dev-build/make: 4.4.1-r100::gentoo dev-build/meson: 1.6.0::gentoo dev-java/java-config: 2.3.4::gentoo dev-lang/perl: 5.40.0::gentoo dev-lang/python: 3.12.7_p1::gentoo, 3.13.0::gentoo dev-lang/rust-bin: 1.81.0-r100::gentoo, 1.82.0-r100::gentoo sys-apps/baselayout: 2.17::gentoo sys-apps/openrc: 0.55.1::gentoo sys-apps/sandbox: 2.40::gentoo sys-devel/binutils: 2.43-r2::gentoo sys-devel/binutils-config: 5.5.2::gentoo sys-devel/clang: 18.1.8-r6::gentoo sys-devel/gcc: 13.3.1_p20241115::gentoo, 14.2.1_p20241116::gentoo sys-devel/gcc-config: 2.11::gentoo sys-devel/lld: 18.1.8::gentoo, 19.1.4::gentoo sys-devel/llvm: 18.1.8-r6::gentoo, 19.1.4::gentoo sys-kernel/linux-headers: 6.11::gentoo (virtual/os-headers) sys-libs/glibc: 2.40-r5::gentoo Repositories: gentoo location: /var/db/repos/gentoo sync-type: rsync sync-uri: rsync://rsync.gentoo.org/gentoo-portage priority: -1000 volatile: False sync-rsync-verify-max-age: 3 sync-rsync-extra-opts: sync-rsync-verify-jobs: 1 sync-rsync-verify-metamanifest: yes escpr2 location: /var/db/repos/escpr2 sync-type: git sync-uri: https://github.com/gentoo-mirror/escpr2.git masters: gentoo volatile: False Binary Repositories: gentoobinhost priority: 9999 sync-uri: https://distfiles.gentoo.org/releases/amd64/binpackages/23.0/x86-64 ACCEPT_KEYWORDS="amd64 ~amd64" ACCEPT_LICENSE="@FREE @BINARY-REDISTRIBUTABLE" CBUILD="x86_64-pc-linux-gnu" CFLAGS="-march=native -O3 -pipe" CHOST="x86_64-pc-linux-gnu" CONFIG_PROTECT="/etc /usr/lib64/libreoffice/program/sofficerc /usr/share/config /usr/share/gnupg/qualified.txt" CONFIG_PROTECT_MASK="/etc/ca-certificates.conf /etc/dconf /etc/env.d /etc/fonts/fonts.conf /etc/gconf /etc/gentoo-release /etc/revdep-rebuild /etc/sandbox.d /etc/texmf/language.dat.d /etc/texmf/language.def.d /etc/texmf/updmap.d /etc/texmf/web2c" CXXFLAGS="-march=native -O3 -pipe" DISTDIR="/var/cache/distfiles" ENV_UNSET="CARGO_HOME DBUS_SESSION_BUS_ADDRESS DISPLAY GDK_PIXBUF_MODULE_FILE GOBIN GOPATH PERL5LIB PERL5OPT PERLPREFIX PERL_CORE PERL_MB_OPT PERL_MM_OPT XAUTHORITY XDG_CACHE_HOME XDG_CONFIG_HOME XDG_DATA_HOME XDG_RUNTIME_DIR XDG_STATE_HOME" FCFLAGS="-march=native -O3 -pipe" FEATURES="assume-digests binpkg-docompress binpkg-dostrip binpkg-logs binpkg-multi-instance binpkg-request-signature buildpkg-live config-protect-if-modified distlocks ebuild-locks fixlafiles getbinpkg ipc-sandbox merge-sync merge-wait multilib-strict network-sandbox news parallel-fetch pid-sandbox pkgdir-index-trusted preserve-libs protect-owned qa-unresolved-soname-deps sandbox sfperms strict unknown-features-warn unmerge-logs unmerge-orphans userfetch userpriv usersandbox usersync xattr" FFLAGS="-march=native -O3 -pipe" GENTOO_MIRRORS="http://distfiles.gentoo.org" LANG="de_DE.utf8" LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,-z,pack-relative-relocs" LEX="flex" PKGDIR="/var/cache/binpkgs" PORTAGE_CONFIGROOT="/" PORTAGE_RSYNC_OPTS="--recursive --links --safe-links --perms --times --omit-dir-times --compress --force --whole-file --delete --stats --human-readable --timeout=180 --exclude=/distfiles --exclude=/local --exclude=/packages --exclude=/.git" PORTAGE_TMPDIR="/var/tmp" SHELL="/bin/bash" USE="X a52 aac acl acpi activities alsa amd64 bluetooth branding bzip2 cairo cdda cdr cet contrib crypt cuda cudnn cups dbus declarative dri dts dvd dvdr dvi elogind encode eps exif fits flac fortran gdbm gdbui gif gnutls go gphoto2 gpm graphite gtk gui iconv icu ipv6 jit jpeg kde kf6compat kwallet lcms libcxx libnotify libtirpc llvm lm-sensors lto mad mng modules-sign mp3 mp4 mpeg multilib ncurses networkmanager nls nvenc objc objc++ offload ogg ompt opencv opengl openmp pam pango pcre pdf pipewire plasma png policykit postscript ppds pulseaudio qml qt5 qt6 raw readline rust screencast sdl seccomp semantic-desktop sound spell ssl startup-notification svg test-rust tiff tpm truetype udev udisks uefi unicode upower usb vorbis vulkan wayland wcs widgets wxwidgets x264 xattr xcb xft xml xv xvid zlib" ABI_X86="64" ADA_TARGET="gcc_12" APACHE2_MODULES="authn_core authz_core socache_shmcb unixd actions alias auth_basic authn_anon authn_dbm authn_file authz_dbm authz_groupfile authz_host authz_owner authz_user autoindex cache cgi cgid dav dav_fs dav_lock deflate dir env expires ext_filter file_cache filter headers include info log_config logio mime mime_magic negotiation rewrite setenvif speling status unique_id userdir usertrack vhost_alias" CALLIGRA_FEATURES="karbon sheets words" COLLECTD_PLUGINS="df interface irq load memory rrdtool swap syslog" CPU_FLAGS_X86="mmx mmxext sse sse2 aes avx avx2 f16c fma3 pclmul popcnt rdrand sha sse3 sse4_1 sse4_2 sse4a ssse3" CURL_QUIC="ngtcp2" CURL_SSL="gnutls" ELIBC="glibc" GPSD_PROTOCOLS="ashtech aivdm earthmate evermore fv18 garmin garmintxt gpsclock greis isync itrax navcom oceanserver oncore rtcm104v2 rtcm104v3 sirf skytraq superstar2 tsip tripmate tnt ublox" GUILE_SINGLE_TARGET="3-0" GUILE_TARGETS="3-0" INPUT_DEVICES="libinput" KERNEL="linux" L10N="de" LCD_DEVICES="bayrad cfontz glk hd44780 lb216 lcdm001 mtxorb text" LUA_SINGLE_TARGET="lua5-1" LUA_TARGETS="lua5-1" OFFICE_IMPLEMENTATION="libreoffice" PHP_TARGETS="php8-2" POSTGRES_TARGETS="postgres16" PYTHON_SINGLE_TARGET="python3_12" PYTHON_TARGETS="python3_12" RUBY_TARGETS="ruby32" VIDEO_CARDS="nvidia" XTABLES_ADDONS="quota2 psd pknock lscan length2 ipv4options ipp2p iface geoip fuzzy condition tarpit sysrq proto logmark ipmark dhcpmac delude chaos account" Unset: ADDR2LINE, AR, ARFLAGS, AS, ASFLAGS, CC, CCLD, CONFIG_SHELL, CPP, CPPFLAGS, CTARGET, CXX, CXXFILT, ELFEDIT, EMERGE_DEFAULT_OPTS, EXTRA_ECONF, F77FLAGS, FC, GCOV, GPROF, INSTALL_MASK, LC_ALL, LD, LFLAGS, LIBTOOL, LINGUAS, MAKE, MAKEFLAGS, MAKEOPTS, NM, OBJCOPY, OBJDUMP, PORTAGE_BINHOST, PORTAGE_BUNZIP2_COMMAND, PORTAGE_COMPRESS, PORTAGE_COMPRESS_FLAGS, PORTAGE_RSYNC_EXTRA_OPTS, PYTHONPATH, RANLIB, READELF, RUSTFLAGS, SIZE, STRINGS, STRIP, YACC, YFLAGS
Created attachment 912586 [details] build.log
That output doesn't match the target. The one you pasted in the other bug was right.
yes you are right... I'm generating a new log by emerging it with the right target again...
Created attachment 912605 [details] info.log
Created attachment 912606 [details] stage1 logfile in tar gz format
Created attachment 912607 [details] config.log in tar gz format
* crossdev version: 20240921 * Host Portage ARCH: amd64 * Host Portage System: x86_64-pc-linux-gnu (i686-pc-linux-gnu x86_64-pc-linux-gnu) * Target Portage ARCH: * * Target System: nvptx * Stage: 3 (C compiler & libc) * USE=multilib: no * Target ABIs: default * binutils: nvptx-tools-[latest] * gcc: gcc-[latest] * headers: linux-headers-[latest] * libc: newlib-[latest] * CROSSDEV_OVERLAY: /var/db/repos/escpr2 * PORT_LOGDIR: /var/log/portage * PORTAGE_CONFIGROOT: / * Portage flags: _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ * leaving metadata/layout.conf alone in /var/db/repos/escpr2 _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ * Log: /var/log/portage/cross-nvptx-nvptx-tools.log * Emerging cross-nvptx-tools ... [ ok ] * Log: /var/log/portage/cross-nvptx-gcc-stage1.log * Emerging cross-gcc-stage1 ... * error: gcc failed :( * * If you file a bug, please attach the following logfiles: * /var/log/portage/cross-nvptx-info.log * /var/log/portage/cross-nvptx-gcc-stage1.log.xz * /var/tmp/portage/cross-nvptx/gcc*/temp/gcc-config.logs.tar.xz
Created attachment 912674 [details] build.log
So, I've found the build log now of this... it appears there were file collisions?.....
the file collisions came from a previous install attempt. after removing them, it still fails, but at the gcc stage.
Created attachment 912686 [details] gcc-stage1- build log
>> Source prepared. >>> Configuring source in /var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/gcc-14-20241116 ... ^[[32m*^[[0m ADAFLAGS="" ^[[32m*^[[0m CFLAGS="-O2 -pipe" ^[[32m*^[[0m CPPFLAGS=" -D_GENTOO_TIME64_FORCE=0" ^[[32m*^[[0m CXXFLAGS="-O2 -pipe" ^[[32m*^[[0m CCASFLAGS="" ^[[32m*^[[0m FFLAGS="-O2 -pipe" ^[[32m*^[[0m FCFLAGS="-O2 -pipe" ^[[32m*^[[0m LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,-z,pack-relative-relocs" ^[[32m*^[[0m PREFIX: /usr ^[[32m*^[[0m BINPATH: /usr/x86_64-pc-linux-gnu/nvptx-tools/gcc-bin/14 ^[[32m*^[[0m LIBPATH: /usr/lib/gcc/nvptx-tools/14 ^[[32m*^[[0m DATAPATH: /usr/share/gcc-data/nvptx-tools/14 ^[[32m*^[[0m STDCXX_INCDIR: /usr/lib/gcc/nvptx-tools/14/include/g++-v14 ^[[32m*^[[0m Languages: c,rust ^[[32m*^[[0m GCC version: 14.2.1 ^[[32m*^[[0m /bin/sh /var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/gcc-14-20241116/configure --host=x86_64-pc-linux-gnu --target=nvptx-tools --build=x86_64-pc-linux-gnu --prefix=/usr --bindir=/usr/x86_64-pc-linux-gnu/nv> configure: loading site script /usr/share/config.site configure: loading site script /usr/share/config.site.d/80crossdev.conf checking build system type... x86_64-pc-linux-gnu checking host system type... x86_64-pc-linux-gnu checking target system type... Invalid configuration 'nvptx-tools': OS 'tools' not recognized configure: error: /bin/sh /var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/gcc-14-20241116/config.sub nvptx-tools failed ^[[31;01m*^[[0m ERROR: cross-nvptx-tools/gcc-14.2.1_p20241116::escpr2 failed (configure phase): ^[[31;01m*^[[0m Failed to run command: /bin/sh /var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/gcc-14-20241116/configure --host=x86_64-pc-linux-gnu --target=nvptx-tools --build=x86_64-pc-linux-gnu --prefix=/usr --bindir> ^[[31;01m*^[[0m ^[[31;01m*^[[0m Call stack: ^[[31;01m*^[[0m ebuild.sh, line 136: Called src_configure ^[[31;01m*^[[0m environment, line 3602: Called toolchain_src_configure ^[[31;01m*^[[0m environment, line 5330: Called edo '/bin/sh' '/var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/gcc-14-20241116/configure' '--host=x86_64-pc-linux-gnu' '--target=nvptx-tools' '--build=x86_64-pc-linux-gnu'> ^[[31;01m*^[[0m environment, line 1609: Called die ^[[31;01m*^[[0m The specific snippet of code: ^[[31;01m*^[[0m "$@" || die -n "Failed to run command: $@" ^[[31;01m*^[[0m ^[[31;01m*^[[0m If you need support, post the output of `emerge --info '=cross-nvptx-tools/gcc-14.2.1_p20241116::escpr2'`, ^[[31;01m*^[[0m the complete build log and the output of `emerge -pqv '=cross-nvptx-tools/gcc-14.2.1_p20241116::escpr2'`. ^[[31;01m*^[[0m ^[[31;01m*^[[0m Please include /var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/gcc-build-logs.tar.xz in your bug report. ^[[31;01m*^[[0m ^[[31;01m*^[[0m The complete build log is located at '/var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/temp/build.log'. ^[[31;01m*^[[0m The ebuild environment file is located at '/var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/temp/environment'. ^[[31;01m*^[[0m Working directory: '/var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/build' ^[[31;01m*^[[0m S: '/var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/work/gcc-14-20241116' >>> Failed to emerge cross-nvptx-tools/gcc-14.2.1_p20241116, Log file: >>> '/var/tmp/portage/cross-nvptx-tools/gcc-14.2.1_p20241116/temp/build.log' * Messages for package cross-nvptx-tools/gcc-14.2.1_p20241116: * ERROR: cross-nvptx-tools/gcc-14.2.1_p20241116::escpr2 failed (configure
hm, if i, after removing the file collisions, emerge the target nvptx instead of nvptx-tools, i get a more meaningful build report. But it also fails...
Created attachment 912689 [details] target nvptx gcc-stage1 build log
checking host system type... nvptx-unknown-none checking for --enable-version-specific-runtime-libs... no checking for a BSD-compatible install... /usr/lib/portage/python3.12/ebuild-helpers/xattr/install -c checking for gawk... gawk checking for nvptx-ar... nvptx-ar checking for nvptx-lipo... nvptx-lipo checking for nvptx-nm... /var/tmp/portage/cross-nvptx/gcc-14.2.1_p20241116/work/build/./gcc/nm checking for nvptx-ranlib... nvptx-ranlib checking for nvptx-strip... nvptx-strip checking whether ln -s works... yes checking for nvptx-gcc... /var/tmp/portage/cross-nvptx/gcc-14.2.1_p20241116/work/build/./gcc/xgcc -B/var/tmp/portage/cross-nvptx/gcc-14.2.1_p20241116/work/build/./gcc/ -B/usr/nvptx/bin/ -B/usr/nvptx/lib/ -isystem /usr/nvptx/include -isystem /usr/nvptx/sys-include checking for suffix of object files... configure: error: in `/var/tmp/portage/cross-nvptx/gcc-14.2.1_p20241116/work/build/nvptx/libgcc': configure: error: cannot compute suffix of object files: cannot compile See `config.log' for more details make[1]: *** [Makefile:12426: configure-target-libgcc] Error 1 make[1]: Leaving directory '/var/tmp/portage/cross-nvptx/gcc-14.2.1_p20241116/work/build' I guess that seems to be the error, finally...
hm, cross-nvptx/newlib also fails, in direct emerge. If I follow this tutorial https://gcc.gnu.org/wiki/Offloading#How_to_build_an_offloading-enabled_GCC i would have to install newlib. nfiguring in nvptx/newlib configure: creating cache ./config.cache checking build system type... x86_64-pc-linux-gnu checking host system type... nvptx-unknown-none checking for a BSD-compatible install... /usr/lib/portage/python3.12/ebuild-helpers/xattr/install -c checking whether build environment is sane... yes checking for nvptx-strip... nvptx-strip checking for a thread-safe mkdir -p... /usr/bin/mkdir -p checking for gawk... gawk checking whether make sets $(MAKE)... yes checking whether make supports nested variables... yes checking for style of include used by make... GNU checking for nvptx-gcc... nvptx-cc -B/var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/build/nvptx/newlib/ -isystem /var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/build/nvptx/newlib/targ-include -isystem /var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/newlib-4.4.0.20231231/newlib/libc/include -B/var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/build/nvptx/libgloss/nvptx -L/var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/build/nvptx/libgloss/libnosys -L/var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/newlib-4.4.0.20231231/libgloss/nvptx checking whether we are cross compiling... yes checking for suffix of object files... configure: error: in `/var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/build/nvptx/newlib': configure: error: cannot compute suffix of object files: cannot compile See `config.log' for more details make[1]: *** [Makefile:8448: configure-target-newlib] Error 1 make[1]: Leaving directory '/var/tmp/portage/cross-nvptx/newlib-4.4.0.20231231/work/build' make[1]: *** Waiting for unfinished jobs.
Looking, but it may take a bit of time.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=7c053ac40241806d3cdb2e367368c8b66cda447e commit 7c053ac40241806d3cdb2e367368c8b66cda447e Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-29 21:59:46 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-29 21:59:46 +0000 sys-devel/nvptx-tools: add 0_pre20240809 Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/nvptx-tools/Manifest | 1 + .../nvptx-tools/nvptx-tools-0_pre20240809.ebuild | 52 ++++++++++++++++++++++ 2 files changed, 53 insertions(+)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=f212c4740e1e16748d965bbd80155fcdfd4786a1 commit f212c4740e1e16748d965bbd80155fcdfd4786a1 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 02:06:58 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 02:08:20 +0000 crossdev: nvptx target fixes * We don't need linux-headers * Set the same GMASK/GUSE as newlib (not sure how useful the multilib bit here is, mind) Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 10 ++++++++++ 1 file changed, 10 insertions(+) https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=ae9adb13dc23de2fb45a858ba7f18998fb238b62 commit ae9adb13dc23de2fb45a858ba7f18998fb238b62 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 02:06:18 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 02:08:20 +0000 wrappers/emerge-wrapper: add nvptx* Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> wrappers/emerge-wrapper | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=2b44881a2d2defadcdcd30260f535b92458da9b5 commit 2b44881a2d2defadcdcd30260f535b92458da9b5 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 01:46:37 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 02:08:19 +0000 crossdev: no SSP for nvptx Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=2cfb62836bcd010f49e186bfa9679ea469e72b32 commit 2cfb62836bcd010f49e186bfa9679ea469e72b32 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 02:05:12 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 02:08:36 +0000 toolchain.eclass: wire up nvptx more We need to handle nvptx* in CTARGET in a few more places so we propagate that we're using newlib (which is obviously right but also means we don't then fail when building libatomic b/c libg.a is missing). Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> eclass/toolchain.eclass | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=02fc58c7c363a44a3d8ba718935e83050fb26294 commit 02fc58c7c363a44a3d8ba718935e83050fb26294 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 02:19:48 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 02:19:48 +0000 sys-devel/crossdev: add 20241130 Michal Rostecki (4): crossdev: Use libunwind and libc++ in LLVM environments ci: Test crossdev in GitHub Actions ci: Run the container explicitly with `docker run` ci: Test more targets, move the testing logic to a script Sam James (4): crossdev: tweak GUSE_DISABLE* crossdev: no SSP for nvptx wrappers/emerge-wrapper: add nvptx* crossdev: nvptx target fixes Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/crossdev/Manifest | 1 + sys-devel/crossdev/crossdev-20241130.ebuild | 40 +++++++++++++++++++++++++++++ 2 files changed, 41 insertions(+)
OK, please try the following: * emerge --sync in an hour or two to let mirrors catch up * do a world upgrade or at least make sure you have the new crossdev version I just pushed * crossdev -C nvptx * crossdev -C nvptx-none * crossdev -C blahblah where blahblah is any other target you tried to make in error (like 'gcc14+') * crossdev -t nvptx-none If it fails, let me know with full commands used and new logs. I was able to get a GCC built at least. Thanks.
(In reply to Sam James from comment #22) > OK, please try the following: > * emerge --sync in an hour or two to let mirrors catch up > * do a world upgrade or at least make sure you have the new crossdev version > I just pushed > * crossdev -C nvptx > * crossdev -C nvptx-none > * crossdev -C blahblah where blahblah is any other target you tried to make > in error (like 'gcc14+') You need to do crossdev -C nvptx-tools too, from looking at your earlier comments.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=c272f335f8b0df15f724e9ee3e2289a9deddbd11 commit c272f335f8b0df15f724e9ee3e2289a9deddbd11 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 02:28:09 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 02:52:12 +0000 crossdev: don't turn off openmp/fortran for nvptx openmp and fortran are rather useful for offloading. Not clear to me whether or not we can turn this on for newlib in general wrt bug #489798 and bug #589672. Bug: https://bugs.gentoo.org/489798 Bug: https://bugs.gentoo.org/589672 Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
I am making some more changes I'm testing now. Please wait.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=6d9336278643e0ad9fc99cfd497d50c3a8f30319 commit 6d9336278643e0ad9fc99cfd497d50c3a8f30319 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 02:59:57 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 05:10:29 +0000 toolchain.eclass: wire up offloading for nvptx 'crossdev -t nvptx-none' creates a standalone cross-compiler which is needed anyway, but by itself, it's less useful than offloading support within host gcc, where GCC can choose to offload into that cross-compiler as appropriate. This could include amdgcn-amdhsa support in the future. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> eclass/toolchain.eclass | 32 ++++++++++++++++++++++++++++++-- 1 file changed, 30 insertions(+), 2 deletions(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=5fcf21142c9bc90ecae60826e0a2e70357769666 commit 5fcf21142c9bc90ecae60826e0a2e70357769666 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 04:40:54 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 04:41:20 +0000 crossdev: build openmp, fortran by default for nvptx Followup to c272f335f8b0df15f724e9ee3e2289a9deddbd11. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=4f1807da7cb79c43875db9aad2e5a6b558b341be commit 4f1807da7cb79c43875db9aad2e5a6b558b341be Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 05:13:05 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 05:13:05 +0000 sys-devel/crossdev: add 20241130.1 Sam James (2): crossdev: don't turn off openmp/fortran for nvptx crossdev: build openmp, fortran by default for nvptx Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/crossdev/Manifest | 1 + sys-devel/crossdev/crossdev-20241130.1.ebuild | 40 +++++++++++++++++++++++++++ 2 files changed, 41 insertions(+)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=2cef8f3d996277a8f44a49b7c220ab8ba2a9e435 commit 2cef8f3d996277a8f44a49b7c220ab8ba2a9e435 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 10:16:55 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 10:18:24 +0000 crossdev: nvptx needs multilib See https://gcc.gnu.org/legacy-ml/gcc-help/2020-01/msg00106.html. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=0e3008ddccb1bccfa7225fb20a9db7694d2752d9 commit 0e3008ddccb1bccfa7225fb20a9db7694d2752d9 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 10:21:26 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 10:37:38 +0000 sys-libs/newlib: pass -Wa,--no-verify for nvptx nvptx-tools performs verification if CUDA is installed, otherwise it skips it. newlib fails to build with that veriifcation, so punt it for now. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-libs/newlib/newlib-4.4.0.20231231.ebuild | 24 ++++++++++++++++++++++++ sys-libs/newlib/newlib-9999.ebuild | 24 ++++++++++++++++++++++++ 2 files changed, 48 insertions(+)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=b2540825f96c83e449d9ffe74894a0cc173ae47e commit b2540825f96c83e449d9ffe74894a0cc173ae47e Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 11:01:26 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 11:03:22 +0000 15.0.0: don't pass -fcf-protection for offload Well, actually, disable it. Bug: https://bugs.gentoo.org/945296 Bug: https://gcc.gnu.org/PR114717 Signed-off-by: Sam James <sam@gentoo.org> 15.0.0/gentoo/26_all_enable-cet.patch | 2 +- 15.0.0/gentoo/README.history | 1 + 2 files changed, 2 insertions(+), 1 deletion(-)
I've spent the day on this and made some pretty good progress but there's more to be done. I'll work on it more tomorrow too.
Will let you know when it is ready to try. Remaining things so far (may add more): * Fix needing to add -fno-stack-protector * Propagate CET patch fix once tested * Figure out gomp_nvptx_main undefined reference issue (ask Alexander if get stuck as he seems to have implemented it)
(In reply to Sam James from comment #33) > Will let you know when it is ready to try. > > Remaining things so far (may add more): > * Fix needing to add -fno-stack-protector > * Propagate CET patch fix once tested > * Figure out gomp_nvptx_main undefined reference issue (ask Alexander if get > stuck as he seems to have implemented it) * Properly test my gcc-config hack and commit it
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=d066fc8f8de408b4e8d9d12a93f19d60ec08e250 commit d066fc8f8de408b4e8d9d12a93f19d60ec08e250 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 11:29:18 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 11:29:58 +0000 15.0.0: pass -fno-stack-protector for offloading --enable-default-ssp doesn't seem to cover this. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> 15.0.0/gentoo/04_all_nossp-on-nostdlib.patch | 2 +- 15.0.0/gentoo/README.history | 1 + 2 files changed, 2 insertions(+), 1 deletion(-)
(In reply to Sam James from comment #34) > (In reply to Sam James from comment #33) > > Will let you know when it is ready to try. > > > > Remaining things so far (may add more): > > * Fix needing to add -fno-stack-protector > > * Propagate CET patch fix once tested > > * Figure out gomp_nvptx_main undefined reference issue (ask Alexander if get > > stuck as he seems to have implemented it) > > * Properly test my gcc-config hack and commit it * Figure out the libgomp.{spec,a} issue (possibly related to the undefined ref) where I need to move it
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=52a39b64a029c40995c6490a307f0c51e9897f2b commit 52a39b64a029c40995c6490a307f0c51e9897f2b Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 12:49:09 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 13:09:02 +0000 toolchain.eclass: move nvptx files to accel/ subdir This is what Fedora does and it seems to have fixed the issue I was having with needing to manually move libgomp.{a,spec} and then after that, getting undefined references to gomp_nvptx_main (which I assume is because the plugin wasn't being loaded). Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> eclass/toolchain.eclass | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)
thanks for the help. I am not too well versed on these kinds of gentoo internals. Also, as I never have used offloading with gcc, i would need some instructions what to do, besides crosstev -t nvptx and then start gcc with -foffload=nvptx-none by the way, here is an interesting instruction how to configure gcc manually for nvidia offloading It has three pages and details all steps: https://www.heise.de/hintergrund/Accelerator-Offloading-mit-GCC-3317330.html I do not know if that would be helpful for devs creating install scripts which do these commands to build the offload compiler and these nvptx files...
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=407e8b720858c57c802614439242364d3cc630a3 commit 407e8b720858c57c802614439242364d3cc630a3 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 14:35:40 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 14:35:40 +0000 toolchain.eclass: move more for nvptx Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> eclass/toolchain.eclass | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
(In reply to Benjamin Schulz from comment #38) > thanks for the help. I am not too well versed on these kinds of gentoo > internals. Also, as I never have used offloading with gcc, i would need some > instructions what to do, besides crosstev -t nvptx and then start gcc with > You're welcome! It's been fun and a learning experience for me too. Still figuring out some bits, but I did just get a "hello world"-style thing working :) > https://www.heise.de/hintergrund/Accelerator-Offloading-mit-GCC-3317330.html > > I do not know if that would be helpful for devs creating install scripts > which do these commands to build the offload compiler and these nvptx > files... Thanks. I think it is useful because there's not that much documentation out there, so even little tidbits can help put pieces together.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=e1214306a0f10fff29dd052617a54489b2e88380 commit e1214306a0f10fff29dd052617a54489b2e88380 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-11-30 14:48:35 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-11-30 14:48:35 +0000 sys-devel/crossdev: add 20241130.3 Sam James (2): crossdev: nvptx needs multilib Makefile: fix tarball generation Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/crossdev/Manifest | 1 + sys-devel/crossdev/crossdev-20241130.3.ebuild | 40 +++++++++++++++++++++++++++ 2 files changed, 41 insertions(+)
Remaining issues: * Fix needing to add -fno-stack-protector (my attempt doesn't *seem* to have worked, need to investigate) * Properly test my gcc-config hack and commit it * Fix compat w/ dash * Propagate the gcc-patches changes to at least 14 once they're all settled And of course, once this is all done, I may play with amd, although I may be a bit fed up by then :p
By the way, Many of these machine learning libraries that are on the market for python are, once you look into the code, quite horrible. For example, this is what tensorflow became after a rename: https://github.com/jax-ml/jax/issues Here is the reason why numba is so fast: https://github.com/numba/numba/issues/9730 So, given that state, one basically has to do everything by oneself. As a result, gpu offloading for gcc is a rather important feature, if one wants to do any computation. Just tell me when I can test anything. At the moment, on my system, it still does not work with the unmasked 9999 crossdev and nvptx packages, it seems.
Created attachment 912904 [details, diff] gcc-config-bug945296-offloading.patch (In reply to Benjamin Schulz from comment #43) > As a result, gpu offloading for gcc is a rather important feature, if one > wants to do any computation. Just tell me when I can test anything. > Yeah, I agree it's important. I'm actually quite surprised you're only the 2nd person to ask about it. The 1st person was about a year ago when I did that very initial wiring of nvptx-none (and hoped they would do some of the further work on it as they were interested in it). You're the 2nd. But I think a lot of people don't think to ask and just try something and then leave it, rather than let us know. Just make sure to give me a tip when you make some money off it ;) > At the moment, on my system, it still does not work with the unmasked 9999 > crossdev and nvptx packages, it seems. You should be able to try it out if you do the following (make sure you follow the precise order): 1) Make sure you have latest crossdev and have run emerge --sync too to pick up eclass and ebuild changes 2) crossdev -C everything nvptx related 3) Patch sys-devel/gcc-config with the attached patch (it's not committed yet for a reason, it's a hack and I need to test it properly) and rebuild it: 4) Rebuild your regular GCC first (sys-devel/gcc:14 or whatever) to pick up the eclass changes to enable offloading detection 5) crossdev -t nvptx-none 6) git clone https://github.com/matthiasdiener/omptest && cd omptest, then edit the Makefile to add -foffload=nvptx-none to CXXFLAGS and FCFLAGS, then `make` For me, step 6) shows my nvidia card being picked up now and it using it for offloading. If any step prior to 6) fails, please stop and tell me. I still have some more patches to tidy up and commit like the gcc-config one, but that offloading example actually works for me now (I got to that point quite late yesterday/earlier).
Also, the GCC version that crossdev builds a toolchain for must match the compiler you'll be using. But by default both of those should be 14.
You might have to also add -fno-stack-protector in *FLAGS in the Makefile too for omptest, but if you get to that point, it's obvious (as it fails with some __stack_chk_fail undefined ref).
I am currently rebuilding my gcc with offload patches from sam. Its taking a bit of time since there is a cron job also working on my system... Once an offload compiler is there, I can test it myself with a short c++ file. For the curious: here are enough examples for offloading. https://enccs.github.io/openmp-gpu/target/ No need for a large omp target testing project project. a few short lines in a cpp file should make clear if it works.
Created attachment 912998 [details] warnings from gcc with gcc-config patch recompiling gcc with emerge gcc worked but created the attached warnings below. I am posting them because some mention a "payload" and I do not know if that has something to do with offloading now trying out crossdev -t nvptx-none
crossdev -t nvptx-none ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ * crossdev version: baba6164d1aa2fd4eee6b52ef10f70b999ee9393 * Host Portage ARCH: amd64 * Host Portage System: x86_64-pc-linux-gnu (i686-pc-linux-gnu x86_64-pc-linux-gnu) * Target Portage ARCH: * * Target System: nvptx-none * Stage: 4 (C/C++ compiler) * USE=multilib: yes * Target ABIs: default * binutils: nvptx-tools-[latest] * gcc: gcc-[latest] * libc: newlib-[latest] * CROSSDEV_OVERLAY: /var/db/repos/escpr2 * PORT_LOGDIR: /var/log/portage * PORTAGE_CONFIGROOT: / * Portage flags: _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - * leaving metadata/layout.conf alone in /var/db/repos/escpr2 _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - !!! WARNING - Cannot auto-configure CHOST nvptx-none; !!! You should edit /usr/nvptx-none/etc/portage/make.conf /usr/nvptx-none/etc/portage/profile/make.defaults /usr/nvptx-none/etc/portage/profile/use.force !!! by hand to complete your configuration. !!! No ARCH is known for this target. !!! No KERNEL is known for this target. * Log: /var/log/portage/cross-nvptx-none-nvptx-tools.log * Emerging cross-nvptx-tools ... [ ok ] * Log: /var/log/portage/cross-nvptx-none-gcc-stage1.log * Emerging cross-gcc-stage1 ... [ ok ] * Log: /var/log/portage/cross-nvptx-none-newlib.log * Emerging cross-newlib ... [ ok ] * Log: /var/log/portage/cross-nvptx-none-gcc-stage2.log * Emerging cross-gcc-stage2 ... * error: gcc failed :( * * If you file a bug, please attach the following logfiles: * /var/log/portage/cross-nvptx-none-info.log * /var/log/portage/cross-nvptx-none-gcc-stage2.log.xz * /var/tmp/portage/cross-nvptx-none/gcc*/temp/gcc-config.logs.tar.xz localhost /home/benni #
Created attachment 912999 [details] logfile in tar gz format
Created attachment 913000 [details] logfile
Created attachment 913001 [details] more logfiles
Did I something wrong? it seems stage2 of the cross build failed. That is at least better than before, where it failed at stage 1....
this here seems to be the problem: none/gcc-14.2.1_p20241116/work/build/nvptx-none/libstdc++-v3/libsupc++/.libs -B/usr/nvptx-none/bin/ -B/usr/nvptx-none/lib/ -isystem /usr/nvptx-none/include -isystem /usr/nvptx-none/sys-include linker (/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build/./gcc/collect-ld) supports shared libraries... no libtool.m4: error: problem compiling CXX test program checking for /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build/./gcc/xg++ -B/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build/./gcc/ -nostdinc++ -funconfigured-libstdc++-v3 -L/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build/nvptx-none/libstdc++-v3/src -L/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build/nvptx-none/libstdc++-v3/src/.libs -L/var/tmp/portage/cross-nvptx-none/gcc-14.2.
and this one configure: error: "libffi has not been ported to nvptx-unknown-none." make[1]: *** [Makefile:15559: configure-target-libffi] Error 1 make[1]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build' make[1]: *** Waiting for unfinished jobs.... make[1]: Entering directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build' Checking multilib configuration for libatomic... mkdir -p -- nvptx-none/libatomic
these here were the useflags for my original re emerged gcc sys-devel/gcc-14.2.1_p20241116:14::gentoo USE="cet (cxx) (default-stack-clash-protection) (default-znow) fortran go graphite jit lto (multilib) nls objc objc++ openmp (pie) rust sanitize ssp zstd -ada (-custom-cflags) -d -debug -doc (-fixed-point) -hardened (-ieee-long-double) (-libssp) -modula2 -objc-gc (-pch) -pgo -systemtap -test (-time64) -valgrind -vanilla -vtv" 0 KiB emerge --info Portage 3.0.66.1 (python 3.12.7-final-0, default/linux/amd64/23.0/desktop/plasma, gcc-14, glibc-2.40-r5, 6.11.6-gentoo-x86_64 x86_64) ================================================================= System uname: Linux-6.11.6-gentoo-x86_64-x86_64-AMD_Ryzen_9_3900X_12-Core_Processor-with-glibc2.40 KiB Mem: 32792460 total, 22229424 free KiB Swap: 31249404 total, 29974780 free Timestamp of repository gentoo: Sun, 01 Dec 2024 17:00:00 +0000 Head commit of repository gentoo: dc16a37cb65e40fc1e900ec2c27b1dfd14ee6e02 Timestamp of repository escpr2: Thu, 31 Oct 2024 18:34:26 +0000 Head commit of repository escpr2: f2c923b8c651f1d14744975f879c2c78f6e5f8f9 sh bash 5.2_p37 ld GNU ld (Gentoo 2.43 p3) 2.43.1 app-misc/pax-utils: 1.3.8::gentoo app-shells/bash: 5.2_p37::gentoo dev-build/autoconf: 2.13-r8::gentoo, 2.71-r7::gentoo, 2.72-r1::gentoo dev-build/automake: 1.17-r1::gentoo dev-build/cmake: 3.31.1::gentoo dev-build/libtool: 2.5.4::gentoo dev-build/make: 4.4.1-r100::gentoo dev-build/meson: 1.6.0::gentoo dev-java/java-config: 2.3.4::gentoo dev-lang/perl: 5.40.0::gentoo dev-lang/python: 3.12.7_p1::gentoo, 3.13.0::gentoo dev-lang/rust-bin: 1.81.0-r100::gentoo, 1.82.0-r100::gentoo, 1.83.0::gentoo sys-apps/baselayout: 2.17::gentoo sys-apps/openrc: 0.55.1::gentoo sys-apps/sandbox: 2.40::gentoo sys-devel/binutils: 2.43-r2::gentoo sys-devel/binutils-config: 5.5.2::gentoo sys-devel/clang: 18.1.8-r6::gentoo, 19.1.4::gentoo sys-devel/gcc: 13.3.1_p20241115::gentoo, 14.2.1_p20241116::gentoo sys-devel/gcc-config: 2.11::gentoo sys-devel/lld: 18.1.8::gentoo, 19.1.4::gentoo sys-devel/llvm: 18.1.8-r6::gentoo, 19.1.4::gentoo sys-kernel/linux-headers: 6.11::gentoo (virtual/os-headers) sys-libs/glibc: 2.40-r5::gentoo Repositories: gentoo location: /var/db/repos/gentoo sync-type: rsync sync-uri: rsync://rsync.gentoo.org/gentoo-portage priority: -1000 volatile: False sync-rsync-extra-opts: sync-rsync-verify-jobs: 1 sync-rsync-verify-max-age: 3 sync-rsync-verify-metamanifest: yes escpr2 location: /var/db/repos/escpr2 sync-type: git sync-uri: https://github.com/gentoo-mirror/escpr2.git masters: gentoo volatile: False Binary Repositories: gentoobinhost priority: 9999 sync-uri: https://distfiles.gentoo.org/releases/amd64/binpackages/23.0/x86-64 ACCEPT_KEYWORDS="amd64 ~amd64" ACCEPT_LICENSE="@FREE @BINARY-REDISTRIBUTABLE NVIDIA-CUDA NVIDIA EPSON-EULA NVIDIA-cuDNN google-chrome all-rights-reserved android" CBUILD="x86_64-pc-linux-gnu" CFLAGS="-march=native -O3 -pipe" CHOST="x86_64-pc-linux-gnu" CONFIG_PROTECT="/etc /usr/lib64/libreoffice/program/sofficerc /usr/share/config /usr/share/gnupg/qualified.txt" CONFIG_PROTECT_MASK="/etc/ca-certificates.conf /etc/dconf /etc/env.d /etc/fonts/fonts.conf /etc/gconf /etc/gentoo-release /etc/revdep-rebuild /etc/sandbox.d /etc/texmf/language.dat.d /etc/texmf/language.def.d /etc/texmf/updmap.d /etc/texmf/web2c" CXXFLAGS="-march=native -O3 -pipe" DISTDIR="/var/cache/distfiles" ENV_UNSET="CARGO_HOME DBUS_SESSION_BUS_ADDRESS DISPLAY GDK_PIXBUF_MODULE_FILE GOBIN GOPATH PERL5LIB PERL5OPT PERLPREFIX PERL_CORE PERL_MB_OPT PERL_MM_OPT XAUTHORITY XDG_CACHE_HOME XDG_CONFIG_HOME XDG_DATA_HOME XDG_RUNTIME_DIR XDG_STATE_HOME" FCFLAGS="-march=native -O3 -pipe" FEATURES="assume-digests binpkg-docompress binpkg-dostrip binpkg-logs binpkg-multi-instance binpkg-request-signature buildpkg-live config-protect-if-modified distlocks ebuild-locks fixlafiles getbinpkg ipc-sandbox merge-sync merge-wait multilib-strict network-sandbox news parallel-fetch pid-sandbox pkgdir-index-trusted preserve-libs protect-owned qa-unresolved-soname-deps sandbox sfperms strict unknown-features-warn unmerge-logs unmerge-orphans userfetch userpriv usersandbox usersync xattr" FFLAGS="-march=native -O3 -pipe" GENTOO_MIRRORS="http://distfiles.gentoo.org" LANG="de_DE.utf8" LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,-z,pack-relative-relocs" LEX="flex" PKGDIR="/var/cache/binpkgs" PORTAGE_CONFIGROOT="/" PORTAGE_RSYNC_OPTS="--recursive --links --safe-links --perms --times --omit-dir-times --compress --force --whole-file --delete --stats --human-readable --timeout=180 --exclude=/distfiles --exclude=/local --exclude=/packages --exclude=/.git" PORTAGE_TMPDIR="/var/tmp" SHELL="/bin/bash" USE="X a52 aac acl acpi activities alsa amd64 bluetooth branding bzip2 cairo cdda cdr cet contrib crypt cuda cudnn cups dbus declarative dri dts dvd dvdr dvi elogind encode eps exif fits flac fortran gdbm gdbui gif gnutls go gphoto2 gpm graphite gtk gui iconv icu ipv6 jit jpeg kde kf6compat kwallet lcms libnotify libtirpc lm-sensors lto mad mng modules-sign mp3 mp4 mpeg multilib ncurses networkmanager nls nvenc objc objc++ offload ogg ompt opencv opengl openmp pam pango pcre pdf pipewire plasma png policykit postscript ppds pulseaudio qml qt5 qt6 raw readline rust screencast sdl seccomp semantic-desktop sound spell ssl startup-notification svg test-rust tiff tpm truetype udev udisks uefi unicode upower usb vorbis vulkan wayland wcs widgets wxwidgets x264 xattr xcb xft xml xv xvid zlib" ABI_X86="64" ADA_TARGET="gcc_12" APACHE2_MODULES="authn_core authz_core socache_shmcb unixd actions alias auth_basic authn_anon authn_dbm authn_file authz_dbm authz_groupfile authz_host authz_owner authz_user autoindex cache cgi cgid dav dav_fs dav_lock deflate dir env expires ext_filter file_cache filter headers include info log_config logio mime mime_magic negotiation rewrite setenvif speling status unique_id userdir usertrack vhost_alias" CALLIGRA_FEATURES="karbon sheets words" COLLECTD_PLUGINS="df interface irq load memory rrdtool swap syslog" CPU_FLAGS_X86="mmx mmxext sse sse2 aes avx avx2 f16c fma3 pclmul popcnt rdrand sha sse3 sse4_1 sse4_2 sse4a ssse3" CURL_QUIC="ngtcp2" CURL_SSL="gnutls" ELIBC="glibc" GPSD_PROTOCOLS="ashtech aivdm earthmate evermore fv18 garmin garmintxt gpsclock greis isync itrax navcom oceanserver oncore rtcm104v2 rtcm104v3 sirf skytraq superstar2 tsip tripmate tnt ublox" GUILE_SINGLE_TARGET="3-0" GUILE_TARGETS="3-0" INPUT_DEVICES="libinput" KERNEL="linux" L10N="de" LCD_DEVICES="bayrad cfontz glk hd44780 lb216 lcdm001 mtxorb text" LUA_SINGLE_TARGET="lua5-1" LUA_TARGETS="lua5-1" OFFICE_IMPLEMENTATION="libreoffice" PHP_TARGETS="php8-2" POSTGRES_TARGETS="postgres16" PYTHON_SINGLE_TARGET="python3_12" PYTHON_TARGETS="python3_12" RUBY_TARGETS="ruby32" VIDEO_CARDS="nvidia" XTABLES_ADDONS="quota2 psd pknock lscan length2 ipv4options ipp2p iface geoip fuzzy condition tarpit sysrq proto logmark ipmark dhcpmac delude chaos account" Unset: ADDR2LINE, AR, ARFLAGS, AS, ASFLAGS, CC, CCLD, CONFIG_SHELL, CPP, CPPFLAGS, CTARGET, CXX, CXXFILT, ELFEDIT, EMERGE_DEFAULT_OPTS, EXTRA_ECONF, F77FLAGS, FC, GCOV, GPROF, INSTALL_MASK, LC_ALL, LD, LFLAGS, LIBTOOL, LINGUAS, MAKE, MAKEFLAGS, MAKEOPTS, NM, OBJCOPY, OBJDUMP, PORTAGE_BINHOST, PORTAGE_BUNZIP2_COMMAND, PORTAGE_COMPRESS, PORTAGE_COMPRESS_FLAGS, PORTAGE_RSYNC_EXTRA_OPTS, PYTHONPATH, RANLIB, READELF, RUSTFLAGS, SIZE, STRINGS, STRIP, YACC, YFLAGS
Hi, it appears here is the solution for this: (yes, I have go in my host compiler).... (hm, because go has native parallelism, it would be ideal for gpu, but it is sad that this is what it creates... I will now use the flat -go and then see... https://gcc.gnu.org/legacy-ml/gcc/2018-03/msg00122.html > configure: error: "libffi has not been ported to nvptx-unknown-none." > [...] Right, such things are not supported for a "--target=nvptx" compiler. But, if your desire is to just enable additional languages, without offloading support for them (which doesn't exist, as discussed in my other email), then the solution is to just specify them in "--enable-languages" for your "gcc-8.0.0_host_build" (host compiler) but not for your "gcc-8.0.0_build" (nvptx offloading compiler), keeping the latter at "--enable-languages=c,c++,fortran,lto".
USE="-go" crossdev -t nvptx-none finished without error. This is a bit of a, well, annoying problem, since the only way this then would work regularly, is if one would put the useflag -go in a config file for the offload compiler, but then the gentoo dev teams would have to periodically check whether this is still valid. I will now test some code. Lets see....
Go doesn't support OpenMP AFAIK. I'll fix that issue when I've woken up a bit.
Hi, I am afraid that it won't compile your said test set with the modified make. nvptx mkoffload: severe error: posix_spawnp: File or directory not found. Compilation stopped. lto-wrapper: severe error: /usr/libexec/gcc/x86_64-pc-linux-gnu/14//accel/nvptx-none/mkoffload returned status 1 compilation ended. /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: error: lto-wrapper failed collect2: error: ld returned status 1 make: *** [Makefile:28: testomp] error 1
(In reply to Benjamin Schulz from comment #60) Are you sure you applied the gcc-config patch too?
I tried to compile it with USE="-go -lto" crossdev -t but that did not work too apparently... So, since link time optimization is a good thing, i turned it back on, removed the cross compiler and emerged it again with crossdev and USE="-go" but well i guess there is some problem still... this is my nvidia-drivers [ebuild R ] x11-drivers/nvidia-drivers-565.57.01-r3:0/565::gentoo USE="X modules modules-sign static-libs strip tools wayland -dist-kernel -kernel-open -modules-compress -persistenced -powerd" ABI_X86="(64) -32" 0 KiB this is my cuda dev-util/nvidia-cuda-toolkit-12.4.1:0/12.4.1::gentoo USE="debugger examples nsight profiler rdma sanitizer vis-profiler" 0 KiB Total: 1 package (1 reinstall), Size of downloads: 0 KiB WARNING: One or more updates/rebuilds have been skipped due to a dependency conflict: dev-util/nvidia-cuda-toolkit:0 (dev-util/nvidia-cuda-toolkit-12.6.1:0/12.6.1::gentoo, ebuild scheduled for merge) USE="debugger examples nsight profiler rdma sanitizer vis-profiler" ABI_X86="(64)" conflicts with <dev-util/nvidia-cuda-toolkit-12.5:=[profiler] required by (sci-libs/caffe2-2.5.1-r3:0/0::gentoo, installed) USE="cuda numpy openmp -distributed -fbgemm -flash -gloo -mkl -mpi -nnpack -onednn -openblas -opencl -qnnpack -rocm -xnnpack" ABI_X86="(64)" AMDGPU_TARGETS="gfx1030 gfx1100 gfx906 gfx908 gfx90a gfx942 -gfx1010 -gfx1011 -gfx1012 -gfx1031 -gfx1101 -gfx1102 -gfx803 -gfx900 -gfx940 -gfx941" PYTHON_SINGLE_TARGET="python3_12 -python3_10 -python3_11" ^ ^^^^ <dev-util/nvidia-cuda-toolkit-12.5:0/12.4.1=[profiler] required by (sci-libs/caffe2-2.5.1-r3:0/0::gentoo, installed) USE="cuda numpy openmp -distributed -fbgemm -flash -gloo -mkl -mpi -nnpack -onednn -openblas -opencl -qnnpack -rocm -xnnpack" ABI_X86="(64)" AMDGPU_TARGETS="gfx1030 gfx1100 gfx906 gfx908 gfx90a gfx942 -gfx1010 -gfx1011 -gfx1012 -gfx1031 -gfx1101 -gfx1102 -gfx803 -gfx900 -gfx940 -gfx941" PYTHON_SINGLE_TARGET="python3_12 -python3_10 -python3_11" ^ ^^^^^^^^^^^^^^ Sun Dec 1 23:53:35 2024 +-----------------------------------------------------------------------------------------+ | NVIDIA-SMI 565.57.01 Driver Version: 565.57.01 CUDA Version: 12.7 | |-----------------------------------------+------------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+========================+======================| | 0 NVIDIA GeForce GTX 1660 ... Off | 00000000:2D:00.0 On | N/A | | 31% 40C P8 7W / 125W | 548MiB / 6144MiB | 0% Default | | | | N/A | +-----------------------------------------+------------------------+----------------------+
Hi i have put the gcc patch into the /etc/portage/patches/sys-devel/gcc-config/ folder as a file named gcc-config.patch. then I typed emerge gcc....
(In reply to Benjamin Schulz from comment #63) > Hi i have put the gcc patch into the > /etc/portage/patches/sys-devel/gcc-config/ folder as a file named > gcc-config.patch. > > then I typed emerge gcc.... Yes, you have to re-emerge _gcc-config_ first, not _gcc_, that is emerge -v1 sys-devel/gcc-config first.
ah clearly, perhaps i forgot to emerge gcc-config first...
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=9e258378b3b596b14130a8538fc37193e3abd5e1 commit 9e258378b3b596b14130a8538fc37193e3abd5e1 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-01 23:04:50 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-01 23:04:50 +0000 crossdev: nvptx: disable irrelevant languages and force necessary ones on Link: https://gcc.gnu.org/legacy-ml/gcc/2018-03/msg00122.html Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
hi i now have typed emerge gcc-config, then emerge gcc, then crossdev -C nvptx-none in order to de-install it, then USE="-go" crossdev -t nvptx-none. when i write a simple codefile, i get g++ -O0 -fopenmp -foffload=nvptx-none -fno-stack-protector ./main.cpp still the same errors... nvptx mkoffload: schwerwiegender Fehler: posix_spawnp: Datei oder Verzeichnis nicht gefunden Kompilierung beendet. schwerwiegender Fehler: /usr/libexec/gcc/x86_64-pc-linux-gnu/14//accel/nvptx-none/mkoffload gab Ende-Status 1 zurück Kompilierung beendet. /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: error: lto-wrapper failed I suspect that code should be correct. https://enccs.github.io/openmp-gpu/target/ /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #define NX 102400 int main(void) { double vecA[NX],vecB[NX],vecC[NX]; double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] = pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ #pragma omp target for (int i = 0; i < NX; i++) { vecC[i] = vecA[i] * vecB[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += vecC[i]; } printf("The sum is: %8.6f \n", sum); return 0; } when I type eselect gcc and select the cross compiler, i get: home/benni/projects/openmptestnew/openmpoffloatest # eselect gcc set 1 * Switching cross-compiler to nvptx-none-14 ... * insanity with nvptx-none-x86_64-pc-linux-gnu-accel-nvptx-none-gcc and //usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-x86_64-pc-linux-gnu-accel-nvptx-none-gcc * insanity with nvptx-none-x86_64-pc-linux-gnu-accel-nvptx-none-lto-dump and //usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-x86_64-pc-linux-gnu-accel-nvptx-none-lto-dump so it looks the patch has been installed?
The "insanity" error is exactly the one the patch is supposed to fix, it worked for me but msg e it's incomplete (of course that's part of why I haven't committed it yet).
(In reply to Sam James from comment #68) > The "insanity" error is exactly the one the patch is supposed to fix, it > worked for me but msg e it's incomplete (of course that's part of why I > haven't committed it yet). s/msg e/maybe/
If you are absolutely sure the patch was applied to gcc-config, please try as a workaround manually creating a symlink: /usr/bin/x86_64-pc-linux-gnu-accel-nvptx-none-gcc -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/x86_64-pc-linux-gnu-accel-nvptx-none-gcc.
well at least, my nano -w /etc/portage/patches/sys-devel/gcc-config/gcc-config.patch shows this here: From 57ceb28b266b380411a5e2c84131b04539f8b414 Mon Sep 17 00:00:00 2001 From: Sam James <sam@gentoo.org> Date: Sun, 1 Dec 2024 01:30:31 +0000 Subject: [PATCH] WIP: gcc-config: handle offload/accel compilers properly Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> --- gcc-config | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/gcc-config b/gcc-config index 404ab29..c4cee31 100755 --- a/gcc-config +++ b/gcc-config @@ -318,7 +318,7 @@ update_wrappers() { x=${x%:*} if [[ ! -x ${ref} ]] ; then - if is_cross_compiler ; then + if is_cross_compiler && [[ ${ref} != *-accel-* ]] ; then ewarn "insanity with ${x} and ${ref}" else # For native targets, the ref might not be @@ -332,6 +332,7 @@ update_wrappers() { # Delete stale native symlink. rm -f "${EROOT}usr/bin/${x}" elif [[ -x ${ref} ]] ; then + [[ ${x} == *-accel-* ]] && x="${x#${CTARGET}-}" atomic_ln "${ref#${ROOT}}" "${EROOT}usr/bin" "${x}" else # Make sure we have no stale wrappers Previously, if such a patch was in that directory, then it got applied. I re-emerged gcc-config and i re-emerged gcc-14 with this patch in it.
that symlink, ln -s tells me, file already exists...
We're close, I promise! Can you share (not as a comment, attached) output of: strace -f g++ -O0 -fopenmp -foffload=nvptx-none -fno-stack-protector ./main.cpp I want to see what paths it's trying
ls -hal /usr/bin/*nvptx* too
removing the file and then symlinking, the result is still the same, saying lto-wrapper failed... i remember, recently there was an update to either nvidia-drivers or cuda... i remember a case where they had to be both recent before one can use cuda... Currently, I have not installed the most recent cuda because other packages need the previous version. Perhaps its that?
ls -hal /usr/bin/*nvptx* lrwxrwxrwx 1 root root 12 29. Nov 19:46 /usr/bin/emerge-nvptx-75 -> cross-emerge lrwxrwxrwx 1 root root 12 2. Dez 01:22 /usr/bin/emerge-nvptx-none -> cross-emerge lrwxrwxrwx 1 root root 12 29. Nov 19:46 /usr/bin/nvptx-75-ebuild -> cross-ebuild lrwxrwxrwx 1 root root 12 29. Nov 19:46 /usr/bin/nvptx-75-emerge -> cross-emerge lrwxrwxrwx 1 root root 14 29. Nov 19:46 /usr/bin/nvptx-75-fix-root -> cross-fix-root lrwxrwxrwx 1 root root 16 29. Nov 19:46 /usr/bin/nvptx-75-pkg-config -> cross-pkg-config lrwxrwxrwx 1 root root 11 2. Dez 01:13 /usr/bin/nvptx-none-ar -> /usr/bin/ar -rwxr-xr-x 1 root root 55K 2. Dez 01:13 /usr/bin/nvptx-none-as lrwxrwxrwx 1 root root 14 2. Dez 01:56 /usr/bin/nvptx-none-cc -> nvptx-none-gcc lrwxrwxrwx 1 root root 61 2. Dez 01:17 /usr/bin/nvptx-none-cpp -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-cpp lrwxrwxrwx 1 root root 12 2. Dez 01:22 /usr/bin/nvptx-none-ebuild -> cross-ebuild lrwxrwxrwx 1 root root 12 2. Dez 01:22 /usr/bin/nvptx-none-emerge -> cross-emerge lrwxrwxrwx 1 root root 14 2. Dez 01:22 /usr/bin/nvptx-none-fix-root -> cross-fix-root lrwxrwxrwx 1 root root 61 2. Dez 01:17 /usr/bin/nvptx-none-gcc -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc lrwxrwxrwx 1 root root 64 2. Dez 01:17 /usr/bin/nvptx-none-gcc-ar -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc-ar lrwxrwxrwx 1 root root 64 2. Dez 01:17 /usr/bin/nvptx-none-gcc-nm -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc-nm lrwxrwxrwx 1 root root 68 2. Dez 01:17 /usr/bin/nvptx-none-gcc-ranlib -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc-ranlib lrwxrwxrwx 1 root root 63 2. Dez 01:17 /usr/bin/nvptx-none-gccrs -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gccrs -rwxr-xr-x 1 root root 51K 2. Dez 01:13 /usr/bin/nvptx-none-ld -rwxr-xr-x 1 root root 35K 2. Dez 01:13 /usr/bin/nvptx-none-nm lrwxrwxrwx 1 root root 74 2. Dez 01:17 /usr/bin/nvptx-none-nvptx-none-gccrs -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-nvptx-none-gccrs lrwxrwxrwx 1 root root 16 2. Dez 01:22 /usr/bin/nvptx-none-pkg-config -> cross-pkg-config lrwxrwxrwx 1 root root 15 2. Dez 01:13 /usr/bin/nvptx-none-ranlib -> /usr/bin/ranlib -rwxr-xr-x 1 root root 27K 2. Dez 01:13 /usr/bin/nvptx-none-run lrwxrwxrwx 1 root root 100 2. Dez 01:56 /usr/bin/nvptx-none-x86_64-pc-linux-gnu-accel-nvptx-none-gccrs -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-x86_64-pc-linux-gnu-accel-nvptx-none-gccrs
Created attachment 913013 [details] logfile
i have put the trace file into a log. its rather large...
I don't see x86_64-pc-linux-gnu-accel-nvptx-none-gcc as a symlink in your output...
For reference here is mine that works (ignore the 15, I've tested it with 14 too) $ ls -hal /usr/bin/*nvptx* lrwxrwxrwx 1 root root 12 Nov 30 13:30 /usr/bin/emerge-nvptx-none -> cross-emerge lrwxrwxrwx 1 root root 11 Nov 30 13:31 /usr/bin/nvptx-none-ar -> /usr/bin/ar -rwxr-xr-x 1 root root 51K Nov 30 13:31 /usr/bin/nvptx-none-as lrwxrwxrwx 1 root root 14 Dec 1 16:31 /usr/bin/nvptx-none-cc -> nvptx-none-gcc lrwxrwxrwx 1 root root 61 Nov 30 13:37 /usr/bin/nvptx-none-cpp -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-cpp lrwxrwxrwx 1 root root 12 Nov 30 13:30 /usr/bin/nvptx-none-ebuild -> cross-ebuild lrwxrwxrwx 1 root root 12 Nov 30 13:30 /usr/bin/nvptx-none-emerge -> cross-emerge lrwxrwxrwx 1 root root 14 Nov 30 13:30 /usr/bin/nvptx-none-fix-root -> cross-fix-root lrwxrwxrwx 1 root root 61 Nov 30 13:37 /usr/bin/nvptx-none-gcc -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc lrwxrwxrwx 1 root root 64 Nov 30 13:37 /usr/bin/nvptx-none-gcc-ar -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc-ar lrwxrwxrwx 1 root root 64 Nov 30 13:37 /usr/bin/nvptx-none-gcc-nm -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc-nm lrwxrwxrwx 1 root root 68 Nov 30 13:37 /usr/bin/nvptx-none-gcc-ranlib -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/14/nvptx-none-gcc-ranlib -rwxr-xr-x 1 root root 47K Nov 30 13:31 /usr/bin/nvptx-none-ld -rwxr-xr-x 1 root root 31K Nov 30 13:31 /usr/bin/nvptx-none-nm lrwxrwxrwx 1 root root 16 Nov 30 13:30 /usr/bin/nvptx-none-pkg-config -> cross-pkg-config lrwxrwxrwx 1 root root 15 Nov 30 13:31 /usr/bin/nvptx-none-ranlib -> /usr/bin/ranlib -rwxr-xr-x 1 root root 19K Nov 30 13:31 /usr/bin/nvptx-none-run lrwxrwxrwx 1 root root 87 Dec 1 16:31 /usr/bin/x86_64-pc-linux-gnu-accel-nvptx-none-gcc -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/15/x86_64-pc-linux-gnu-accel-nvptx-none-gcc lrwxrwxrwx 1 root root 92 Dec 1 16:31 /usr/bin/x86_64-pc-linux-gnu-accel-nvptx-none-lto-dump -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/15/x86_64-pc-linux-gnu-accel-nvptx-none-lto-dump $ ls -hal /usr/bin/*accel* lrwxrwxrwx 1 root root 87 Dec 1 16:31 /usr/bin/x86_64-pc-linux-gnu-accel-nvptx-none-gcc -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/15/x86_64-pc-linux-gnu-accel-nvptx-none-gcc lrwxrwxrwx 1 root root 92 Dec 1 16:31 /usr/bin/x86_64-pc-linux-gnu-accel-nvptx-none-lto-dump -> /usr/x86_64-pc-linux-gnu/nvptx-none/gcc-bin/15/x86_64-pc-linux-gnu-accel-nvptx-none-lto-dump
Hi, i found the problem.... I had portage configured to use binary packages... and so it installed gcc-config as a binary and steadfastly ignored your patch. Now i installed it from source... Will re-emerge gcc now...
now its getting strange. i have run emerge gcc-config (with the patch) this was its output: emerge gcc-config * IMPORTANT: 28 news items need reading for repository 'gentoo'. * Use eselect news read to view new items. Calculating dependencies... done! Dependency resolution took 3.69 s (backtrack: 0/20). >>> Verifying ebuild manifests >>> Emerging (1 of 1) sys-devel/gcc-config-2.11::gentoo * gcc-config-2.11.tar.xz BLAKE2B SHA512 size ;-) ... [ ok ] >>> Unpacking source... >>> Unpacking gcc-config-2.11.tar.xz to /var/tmp/portage/sys-devel/gcc-config-2.11/work >>> Source unpacked in /var/tmp/portage/sys-devel/gcc-config-2.11/work >>> Preparing source in /var/tmp/portage/sys-devel/gcc-config-2.11/work/gcc-config-2.11 ... * =================================================================================================================================================================================================== * Applying user patches from /etc/portage/patches ... * Applying gcc-config.patch ... [ ok ] * User patches applied. * =================================================================================================================================================================================================== >>> Source prepared. >>> Configuring source in /var/tmp/portage/sys-devel/gcc-config-2.11/work/gcc-config-2.11 ... >>> Source configured. >>> Compiling source in /var/tmp/portage/sys-devel/gcc-config-2.11/work/gcc-config-2.11 ... make -j24 PV=2.11 SUBLIBDIR=lib64 USE_CC_WRAPPERS=yes USE_NATIVE_LINKS=yes TOOLCHAIN_PREFIX=x86_64-pc-linux-gnu- sed \ -e 's:@GENTOO_EPREFIX@::g' \ -e 's:@GENTOO_LIBDIR@:lib64:g' \ -e 's:@PV@:2.11:g' \ -e 's:@USE_NATIVE_LINKS@:yes:g' \ -e 's:@USE_CC_WRAPPERS@:yes:g' \ gcc-config > .gcc-config sed \ -e '1s:/:/:' \ -e 's:@PV@:2.11:g' \ -e 's:@TOOLCHAIN_PREFIX@:x86_64-pc-linux-gnu-:g' \ c89 > .c89 sed \ -e '1s:/:/:' \ -e 's:@PV@:2.11:g' \ -e 's:@TOOLCHAIN_PREFIX@:x86_64-pc-linux-gnu-:g' \ c99 > .c99 chmod a+rx .gcc-config chmod a+rx .c89 chmod a+rx .c99 >>> Source compiled. >>> Test phase [not enabled]: sys-devel/gcc-config-2.11 >>> Install sys-devel/gcc-config-2.11 into /var/tmp/portage/sys-devel/gcc-config-2.11/image make -j24 PV=2.11 SUBLIBDIR=lib64 USE_CC_WRAPPERS=yes USE_NATIVE_LINKS=yes TOOLCHAIN_PREFIX=x86_64-pc-linux-gnu- DESTDIR=/var/tmp/portage/sys-devel/gcc-config-2.11/image install mkdir -p -m 755 /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/bin /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/share/eselect/modules /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/share/gcc-config /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/share/doc/gcc-config-2.11 install -m 755 .gcc-config /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/bin/gcc-config install -m 755 .c89 /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/share/gcc-config/c89 install -m 755 .c99 /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/share/gcc-config/c99 if [ "yes" = yes ] ; then \ install -m 755 .c89 /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/bin/c89 && \ install -m 755 .c99 /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/bin/c99 ; \ fi install -m 644 gcc.eselect /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/share/eselect/modules install -m 644 README /var/tmp/portage/sys-devel/gcc-config-2.11/image/usr/share/doc/gcc-config-2.11 >>> Completed installing sys-devel/gcc-config-2.11 into /var/tmp/portage/sys-devel/gcc-config-2.11/image * Final size of build directory: 736 KiB * Final size of installed tree: 96 KiB and I saw it getting in. I ran emerge gcc, which installed gcc from source. Afterwards, I ran emerge --sync; emerge crossdev; crossdev -C nvptx-none; USE="-go" crossdev -t nvptx... and now gcc fails to emerge at stage 1 again crossdev -t nvptx-none ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ * crossdev version: 9e258378b3b596b14130a8538fc37193e3abd5e1 * Host Portage ARCH: amd64 * Host Portage System: x86_64-pc-linux-gnu (i686-pc-linux-gnu x86_64-pc-linux-gnu) * Target Portage ARCH: * * Target System: nvptx-none * Stage: 4 (C/C++ compiler) * USE=multilib: yes * Target ABIs: default * binutils: nvptx-tools-[latest] * gcc: gcc-[latest] * libc: newlib-[latest] * CROSSDEV_OVERLAY: /var/db/repos/escpr2 * PORT_LOGDIR: /var/log/portage * PORTAGE_CONFIGROOT: / * Portage flags: _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - * leaving sys-libs/newlib in /var/db/repos/escpr2 * leaving sys-devel/nvptx-tools in /var/db/repos/escpr2 * leaving sys-devel/gcc in /var/db/repos/escpr2 * leaving dev-debug/gdb in /var/db/repos/escpr2 * leaving metadata/layout.conf alone in /var/db/repos/escpr2 _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - * Log: /var/log/portage/cross-nvptx-none-nvptx-tools.log * Emerging cross-nvptx-tools ... [ ok ] * Log: /var/log/portage/cross-nvptx-none-gcc-stage1.log * Emerging cross-gcc-stage1 ... * error: gcc failed :( * * If you file a bug, please attach the following logfiles: * /var/log/portage/cross-nvptx-none-info.log * /var/log/portage/cross-nvptx-none-gcc-stage1.log.xz * /var/tmp/portage/cross-nvptx-none/gcc*/temp/gcc-config.logs.tar.xz
Created attachment 913016 [details] log 1
Created attachment 913017 [details] log2
Created attachment 913018 [details] log3
ar/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/gcc-14-20241116/gcc/config/nvptx/nvptx.cc:325:12: warning: unquoted identifier or keyword ‘sm_’ in format [-Wformat-diag] 325 | error ("PTX version (%<-mptx%>) needs to be at least %s to support selected" I have applied the patch, but did an emerge --sync and emerge crossdev before crosstev -t nvptx-none...
``` configure:3090: checking whether the C compiler works configure:3112: /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build/./gcc/xgcc -B/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241116/work/build/./gcc/ -B/usr/nvptx-none/bin/ -B/usr/nvptx-none/lib/ -isystem /usr/nvptx-none/include -isystem /usr/nvptx-none/sys-include -g -O2 conftest.c >&5 error resolving libg.a collect2: error: ld returned 1 exit status ``` Let's see.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=075277041c223391b2c980d0e05989bfd1adee41 commit 075277041c223391b2c980d0e05989bfd1adee41 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-02 09:42:08 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-02 09:42:30 +0000 crossdev: nvptx: don't force on openmp as it affects stage1 too Bug: https://bugs.gentoo.org/945296 Fixes: 9e258378b3b596b14130a8538fc37193e3abd5e1 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 1 - 1 file changed, 1 deletion(-)
Please try again after -C, emerge --sync, and emerge -v1 crossdev.
Created attachment 913069 [details] 3 logfiles in tar.gz archive
it still fails, after emerge --sync; emerge -v1 crossdev; crossdev -C nvptx-none; crossdev -t nvptx-none. I attached the logfiles...
(In reply to Benjamin Schulz from comment #91) > it still fails, after emerge --sync; emerge -v1 crossdev; crossdev -C > nvptx-none; crossdev -t nvptx-none. I attached the logfiles... ``` Calculating dependencies * IMPORTANT: 28 news items need reading for repository 'gentoo'. * Use eselect news read to view new items. .... done! Dependency resolution took 1.75 s (backtrack: 0/20). emerge: there are no ebuilds to satisfy "cross-nvptx-none/gcc". emerge: searching for similar names... emerge: Maybe you meant cross-nvptx-none/nvptx-tools? ``` Not sure what happened there? Maybe delete all the cross-* stuff from your escpr2 repo (it really shouldn't be writing there anyway), then emerge -vn eselect-repository && eselect repository create crossdev. crossdev should put that stuff in there then which is much better than a random repo where it might get clobbered on syncs.
ah it seems there was just a typo
now it compiled. However, I can still not start anything. gcc -fopenmp -foffload=nvptx-none ./main.cpp would complain about: -fcf-protection=full It would only accept none or check for that flag. But then it would stop with this: gcc -fopenmp -foffload=nvptx-none -fcf-protection=check ./main.cpp lto1: severe error: the compiler does not support zstd-lto compression According to gcc's documentation, I can set -flto-compression-level=0 and that would mean no compression. But that flag should only be meaningful with -flto, which I did not supply. Anyway, this gcc -fopenmp -foffload=nvptx-none -fcf-protection=check -flto-compression-level=0 ./main.cpp and this: gcc -fopenmp -foffload=nvptx-none -fcf-protection=check -flto-compression-level=0 ./main.cpp fail with the same error "the compiler does not support zstd-lto compression" the code is this one: /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #include "omp.h" #define NX 102400 int main(void) { double vecA[NX],vecB[NX],vecC[NX]; double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] = pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ #pragma omp target teams distribute parallel for for (int i = 0; i < NX; i++) { vecC[i] = vecA[i] * vecB[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += vecC[i]; } printf("The sum is: %8.6f \n", sum); return 0; }
ae i meant this gcc -flto -fopenmp -foffload=nvptx-none -fcf-protection=check -flto-compression-level=0 ./main.cpp fails as well as this gcc -fopenmp -foffload=nvptx-none -fcf-protection=check -flto-compression-level=0 ./main.cpp and this gcc -flto -fopenmp -foffload=nvptx-none -fcf-protection=check ./main.cpp saying the compiler does not support lto compression
I'll handle the -fcf-protection issue along with the remaining TODOs later (as it has an easy workaround for now). Can you show me: emerge -pvO sys-devel/gcc:14 cross-nvptx-none/gcc:14 output
my gcc flags are these: sys-devel/gcc-14.2.1_p20241116:14::gentoo USE="cet (cxx) (default-stack-clash-protection) (default-znow) fortran go graphite jit lto (multilib) nls objc objc++ openmp (pie) rust sanitize ssp zstd -ada (-custom-cflags) -d -debug -doc (-fixed-point) -hardened (-ieee-long-double) (-libssp) -modula2 -objc-gc (-pch) -pgo -systemtap -test (-time64) -valgrind -vanilla -vtv" 0 KiB [ebuild R ] cross-nvptx-none/gcc-14.2.1_p20241116:14::escpr2 USE="cxx fortran jit lto (multilib) nls openmp pie (-ada) (-cet) -custom-cflags (-d) -debug (-default-stack-clash-protection) -default-znow -doc -fixed-point (-go) (-graphite) (-hardened) -ieee-long-double -libssp (-modula2) -objc -objc++ -objc-gc (-pch) -pgo (-rust) -sanitize (-ssp) -systemtap -test (-time64) -valgrind -vanilla -vtv -zstd" 0 KiB so there is zstd in there...
But not for the cross compiler! Please hold.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=88d805301c5ca721d66d5102aa6ba2747cb2771e commit 88d805301c5ca721d66d5102aa6ba2747cb2771e Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-02 15:24:29 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-02 15:27:14 +0000 crossdev: nvptx: enable zlib+zstd for LTO LTO is used behind-the-scenes between the host and offload to communicate data. We need to use the same compression for both. It's easiest if we just enable both supported compression formats in the offload compiler: zlib and zstd. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
Try that (again from scratch, sorry).
> escpr2 I see you didn't take my advice yet wrt the other repo -- please do (see https://bugs.gentoo.org/945296#c92).
Hi there, yes thanks for the tip. i have put now crossdev in a separate repository... Now it appears that with -fno-stack-protector and either -fcf-protection=none or -fcf-protection=check, i can compile some things. But it is still a bit strange, unfortunately, depending on the source-code. For example, the following compiles and returns warnings: gcc -O3 -fopenmp -foffload=nvptx-none -fcf-protection=none -fno-stack-protector ./main.cpp -lm /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: /tmp/ccYZZvVP.crtoffloadtable.o: warning: relocation against `__offload_vars_end' in read-only section `.rodata' /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: warning: creating DT_TEXTREL in a PIE /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #include "omp.h" #define NX 102400 int main(void) { double vecA[NX],vecB[NX],vecC[NX]; double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] =pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ #pragma omp target teams distribute parallel for map(vecC) map(vecA,vecB) for (int i = 0; i < NX; i++) { vecC[i] = vecA[i] * vecB[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += vecC[i]; } printf("The sum is: %8.6f \n", sum); return 0; } But does it really work on the graphics card? It seems so. Since it takes a bit long to start, which is typical for gpu offloading. But now let us check that with the first example from https://enccs.github.io/openmp-gpu/target/ /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include "omp.h" int main() { int num_devices = omp_get_num_devices(); printf("Number of available devices %d\n", num_devices); #pragma omp target { if (omp_is_initial_device()) { printf("Running on host\n"); } else { int nteams= omp_get_num_teams(); int nthreads= omp_get_num_threads(); printf("Running on device with %d teams in total and %d threads in each team\n",nteams,nthreads); } } return 0; } gcc -O3 -fopenmp -foffload=nvptx-none -fcf-protection=check -fno-stack-protector ./main2.cpp -lm -lstdc++ unresolved symbol __printf_chk collect2: Fehler: ld gab 1 als Ende-Status zurück nvptx mkoffload: schwerwiegender Fehler: x86_64-pc-linux-gnu-accel-nvptx-none-gcc gab Ende-Status 1 zurück Kompilierung beendet. lto-wrapper: schwerwiegender Fehler: /usr/libexec/gcc/x86_64-pc-linux-gnu/14//accel/nvptx-none/mkoffload gab Ende-Status 1 zurück Kompilierung beendet. /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: error: lto-wrapper failed collect2: Fehler: ld gab 1 als Ende-Status zurück Apparently, it has difficulties finding printf when running on the device. unresolved symbol __printf_chk printf is in libc, I think. But I guess I've linked to libc. Now lets say we want to use a function from math inside the loop on a device: Modify the first example with this: /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #include "omp.h" #include "vector" #define NX 102400 using namespace std; int main(void) { vector<double> vecA(NX),vecB(NX),vecC(NX); double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] =pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ double* A=vecA.data(); double* B=vecB.data(); double* C=vecC.data(); #pragma omp target teams distribute parallel for map(A[0,NX]) map(B[0,NX],C[0,NX]) for (int i = 0; i < NX; i++) { C[i] = (A[i]) * pow((B[i]),1.0); } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += C[i]; } printf("The sum is: %8.6f \n", sum); return 0; } It compiles even. But executing it, leads to libgomp: cuCtxSynchronize error: an illegal memory access was encountered libgomp: cuMemFree_v2 error: an illegal memory access was encountered libgomp: device finalization failed I dont' know but without functions like sqrt, or pow... numerical programming gets a bit hard with just +,-,* and /
Yeah, for now, please do: -U_FORTIFY_SOURCE -fno-stack-protector -fcf-protection=none -Wl,-z,notext These are the issues I was referring to earlier when I said I have to finish off some bits ;) I know that some restrictions apply to what you can and can't do with operations but I have no idea of the specifics. That part at least isn't likely to be our fault.
I can reproduce that crash with sqrt. I will try have a look but no promises on that. Of course, if we find it works on another distro but not Gentoo, then that's obviously more pressing. Maybe you can try in a Docker container?
If I copy the example solution to Exercise 02 from https://enccs.github.io/openmp-gpu/target/, that works. See also https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97203.
Hi apparently it may not be on our/gentoo's sind. openacc has some function that can map std:: math functions to the device: https://stackoverflow.com/questions/15330713/openacc-use-math-h-library Openmp has something like declare target: https://www.ibm.com/docs/de/xl-c-and-cpp-linux/16.1.1?topic=parallelization-pragma-omp-declare-target perhaps one can or has to apply this not only with the simple struct with constructor, as in this example, but also to an stl math function?
Yes, i know, simple examples, where a pointer with a data range is uploaded work.. Its a bit... well... is there some way, one can append #pragma omp declare target in front of stl math functions?
here seems to be some solution? https://stackoverflow.com/questions/49531448/openmp-gpu-offloading-math-library Somebody has tried the same with the sqrt function and got help from ubuntu?
ah now, it seems to be a problem when mapping pointers from vectors... this one here seems to work /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #define NX 102400 int main(void) { double vecA[NX],vecB[NX],vecC[NX]; double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] = pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ #pragma omp target teams distribute parallel for for (int i = 0; i < NX; i++) { vecC[i] = sqrt(vecA[i]) * vecB[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += vecC[i]; } printf("The sum is: %8.6f \n", sum); return 0; }
/* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #define NX 102400 #include <vector> int main(void) { double vecA[NX],vecB[NX],vecC[NX]; double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] = pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ #pragma omp target teams distribute parallel for map (vecA,vecB,vecC) for (int i = 0; i < NX; i++) { vecC[i] = pow(vecA[i],2) * vecB[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += vecC[i]; } printf("The sum is: %8.6f \n", sum); return 0; }
this here seems not to work: /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #define NX 102400 #include <vector> using namespace std; int main(void) { vector<double> vecA(NX); vector<double> vecB(NX); vector<double> vecC(NX); double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] = pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ #pragma omp target teams distribute parallel for map (vecA,vecB,vecC) for (int i = 0; i < NX; i++) { vecC[i] = pow(vecA[i],2) * vecB[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += vecC[i]; } printf("The sum is: %8.6f \n", sum); return 0; }
compiles, but does not work. Vector.data() should have continuous memory. /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #define NX 102400 #include <vector> using namespace std; int main(void) { vector<double> vecA(NX); vector<double> vecB(NX); vector<double> vecC(NX); double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] = pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ double *A=vecA.data(); double *B=vecB.data(); double *C=vecC.data(); #pragma omp target teams distribute parallel for map (to: A[0,NX], B[0,NX]) map (from: C[0,NX]) for (int i = 0; i < NX; i++) { vecC[i] = vecA[i] * vecB[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += vecC[i]; } printf("The sum is: %8.6f \n", sum); return 0; }
sorry it should mean C[i] = A[i] * B[i]; Compiles but creates: libgomp: cuCtxSynchronize error: an illegal memory access was encountered libgomp: cuMemFree_v2 error: an illegal memory access was encountered libgomp: device finalization failed It does not copy that pointer to the device... without openmp offload, the code compiles, and works of course. result should be The sum is: 1.250000 /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #define NX 102400 #include <vector> using namespace std; int main(void) { vector<double> vecA(NX); vector<double> vecB(NX); vector<double> vecC(NX); double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vecA[i] = pow(r, i); vecB[i] = 1.0; } /* dot product of two vectors */ double *A=vecA.data(); double *B=vecB.data(); double *C=vecC.data(); #pragma omp target teams distribute parallel for map (A[0,NX],B[0,NX],C[0,NX]) for (int i = 0; i < NX; i++) { C[i] = A[i] * B[i]; } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += C[i]; } printf("The sum is: %8.6f \n", sum); return 0; }
Apparently, it is not a problem of the stl. This example also compiles but fails. In my book "Using OpenMP, the next step", from van der Pas, Stotzer, Terboven, there are dozens of examples like that, e.g. on p. 274. #include <stdio.h> #include <math.h> #define NX 102400 #include <vector> using namespace std; int main(void) { double *A=(double*)malloc(sizeof(double)*NX); double *B=(double*)malloc(sizeof(double)*NX); double *C=(double*)malloc(sizeof(double)*NX); double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { A[i] = pow(r, i); B[i] = 1.0; } /* dot product of two vectors */ #pragma omp target map (to: A[0,NX], B[0,NX]) map (from: C[0,NX]) { #pragma omp parallel for shared (A,B,C) for (int i = 0; i < NX; i++) { C[i] = A[i] * B[i]; } } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += C[i]; } printf("The sum is: %8.6f \n", sum); free(A); free(B); free(C); return 0; }
turns out that one needs to declare the array bounds in #pragma omp with a : and not a comma. This here works: with these compile options: gcc -O3 -fopenmp -foffload=nvptx-none -fcf-protection=none -fno-stack-protector ./main.cpp -lm -lstdc++ with the following warnings: /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: /tmp/cc782zGt.crtoffloadtable.o: warning: relocation against `__offload_funcs_end' in read-only section `.rodata' /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: warning: creating DT_TEXTREL in a PIE #include <stdio.h> #include <math.h> #define NX 102400 #include <vector> #include "omp.h" using namespace std; int main(void) { vector <double> vA(NX),vB(NX),vC(NX); double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vA[i] = pow(r, i); vB[i] = 1.0; } /* dot product of two vectors */ double*A=vA.data(); double*B=vB.data(); double *C=vC.data(); #pragma omp target map (A[0:NX], B[0:NX], C[0:NX]) { #pragma omp parallel for shared (A,B,C) for (int i = 0; i < NX; i++) { C[i] = A[i] * B[i]; } } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += C[i]; } printf("The sum is: %8.6f \n", sum); return 0; } this here works too, so one can access at least some stl math functions: /* Copyright (c) 2019 CSC Training */ /* Copyright (c) 2021 ENCCS */ #include <stdio.h> #include <math.h> #define NX 102400 #include <vector> #include "omp.h" using namespace std; int main(void) { vector <double> vA(NX),vB(NX),vC(NX); double r=0.2; /* Initialization of vectors */ for (int i = 0; i < NX; i++) { vA[i] = pow(r, i); vB[i] = 1.0; } /* dot product of two vectors */ double*A=vA.data(); double*B=vB.data(); double *C=vC.data(); #pragma omp target map (A[0:NX], B[0:NX], C[0:NX]) { #pragma omp parallel for shared (A,B,C) for (int i = 0; i < NX; i++) { C[i] = sqrt(A[i]) * B[i]; } } double sum = 0.0; /* calculate the sum */ for (int i = 0; i < NX; i++) { sum += C[i]; } printf("The sum is: %8.6f \n", sum); return 0; } Good that its finally ok. Thank you for your help. Best regards, Benjamin I would also like to test clang with this...
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=453e85f4bc2a567710f2d7659504fbdc76d0897e commit 453e85f4bc2a567710f2d7659504fbdc76d0897e Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-02 21:34:42 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-02 21:34:42 +0000 sys-devel/crossdev: add 20241202 Sam James (3): crossdev: nvptx: disable irrelevant languages and force necessary ones on crossdev: nvptx: don't force on openmp as it affects stage1 too crossdev: nvptx: enable zlib+zstd for LTO Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/crossdev/Manifest | 1 + sys-devel/crossdev/crossdev-20241202.ebuild | 40 +++++++++++++++++++++++++++++ 2 files changed, 41 insertions(+)
> Thank you for your help. You're very welcome! > > I would also like to test clang with this... Yeah, mgorny is still working on getting the changes landed, then will backport it, and we can go from there then.
(In reply to Sam James from comment #117) > > Thank you for your help. > > You're very welcome! > I should also say: thank you for your prompt testing and feedback. I wouldn't have had the motivation to carry on without that.
OK, to recap for myself from https://bugs.gentoo.org/945296#c42 with some new bits... TODO: * Fix needing to add -fno-stack-protector (my attempt doesn't *seem* to have worked, need to investigate) * Properly test my gcc-config hack and commit it * Fix compat w/ dash * Propagate the gcc-patches changes to at least 14 once they're all settled * Fix F_S * Fix cet still being added * Do we need to backport the PIE patch from jakub or is -z,notext good enough for now (doubt and more awkward to implement, even)?
Well, If needed, I can test more. Currently, I am always getting this warning during compile: /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: /tmp/ccQdqLfV.crtoffloadtable.o: warning: relocation against `__offload_funcs_end' in read-only section `.rodata' /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: warning: creating DT_TEXTREL in a PIE Which does not seem to affect the behavior at runtime. What i noticed is that, well gcc is not checking the syntax correctly anymore. map(p[a,b]) makes no sense in OpenMP syntax, which requires map(p[a:b]. Yet gcc only complains during runtime. But that is a problem of gcc, but that is clearly something for upstream.
and note: its not only -fno-stack-protector . I need -fcf-protection=check or -fcf-protection=none. With just -fno-stack-protector, i am getting errors like this: nvptx mkoffload: Severe failure x86_64-pc-linux-gnu-accel-nvptx-none-gcc returned status 1. Compilation ended. lto-wrapper: Severe failure: /usr/libexec/gcc/x86_64-pc-linux-gnu/14//accel/nvptx-none/mkoffload returned status 1 Compilation ended /usr/lib/gcc/x86_64-pc-linux-gnu/14/../../../../x86_64-pc-linux-gnu/bin/ld: error: lto-wrapper failed collect2: error: ld returned status 1
when I add -no-pie, then the warnings disappear. So, to recap, that is the full set of options which I need -fcf-protection=none or -fcf-protection=check -fno-stack-protector -no-pie
Interestingly -fcf-protection=check works. According to the documentation: The value check is used for the final link with link-time optimization (LTO). An error is issued if LTO object files are compiled with different -fcf-protection values. The value check is ignored at the compile time. It also works with -lto enabled. -lto affects optimization on several obj files. I now wrote a small program which has two cpp files. I guess that should generate two obj files. it works with -lto and -fcf-protection=check but perhaps that changes if a complicated and large external library is loaded?
I can, by the way, not reproduce that gcc "bug" where somebody called a function inside a simd instruction: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97203 on my system: #pragma omp target map (A[0:NX], B[0:NX], C[0:NX]) { #pragma omp simd for (int i = 0; i < NX; i++) { C[i] = __builtin_sinf(A[i]) * B[i]; } } works. As well as calling an arbitrary own function... so I guess that bug could be closed in gcc... It is interesting, since I did not know that. simd can be much faster than parallel for. Normally one would expect gcc to use simd on its own, by the way. But often it does not do that. stating explicitly #pragma omp simd can speed things up, even with -O3 used, at least on my system...
I'll hopefully get to the remaining TODOs of mine over the next week.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-config.git/commit/?id=491f85516b7903a4893cbd2bd0a3dcf6a7423c93 commit 491f85516b7903a4893cbd2bd0a3dcf6a7423c93 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-01 01:30:31 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-08 03:16:40 +0000 gcc-config: handle offload/accel compilers properly accel/offload compilers are named like 'x86_64-pc-linux-gnu-accel-nvptx-none-gcc'. When building the offload compiler, we'll have ${CHOST}-${CTARGET} where CHOST is the host and CTARGET is whatever the offload target is. gcc-config doesn't expect this 'double tuple'. e.g. when building nvptx-none, we'll have something like 'nvptx-none-x86_64-pc-linux-gnu-accel-nvptx-none-gcc' and we want to strip that to be just x86_64-pc-linux-gnu-accel-nvptx-none-gcc, as that's what we want symlinks to be made for in /usr/bin. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> gcc-config | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=4834ad1c39a6ca6f4012245cfef9d566dace69b6 commit 4834ad1c39a6ca6f4012245cfef9d566dace69b6 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-08 03:25:20 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-08 03:29:12 +0000 sys-devel/gcc-config: add 2.12 Berin Aniesh (1): README: Convert README from rST to markdown Sam James (5): README: grammar tweaks gcc-config: handle offload/accel compilers properly tests/run_tests: adapt to NO_COLOR .gitignore: ignore test temporary directories Makefile: README -> README.md Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/gcc-config/Manifest | 1 + sys-devel/gcc-config/gcc-config-2.12.1.ebuild | 60 +++++++++++++++++++++++++++ 2 files changed, 61 insertions(+)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=35045208b613bd86d21415331da51d235ee5e6e7 commit 35045208b613bd86d21415331da51d235ee5e6e7 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 05:07:05 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 05:07:05 +0000 15.0.0: add nvptx bashism patch Bug: https://gcc.gnu.org/PR117854 Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> ...tx-fix-bashisms-with-gen-copyright.sh-use.patch | 68 ++++++++++++++++++++++ 15.0.0/gentoo/README.history | 1 + 2 files changed, 69 insertions(+)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=f41e17d0b106cc664c051f599c09ce696a3ab66d commit f41e17d0b106cc664c051f599c09ce696a3ab66d Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 05:11:17 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 05:11:17 +0000 15.0.0: update -fstack-clash-protection patch for offload Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> 15.0.0/gentoo/24_all_DEF_GENTOO_SCP-fstack-clash-protection.patch | 2 +- 15.0.0/gentoo/README.history | 1 + 2 files changed, 2 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=ec76a392e824bb1abe5f2d390d4bdbd96ee50040 commit ec76a392e824bb1abe5f2d390d4bdbd96ee50040 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 05:12:51 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 05:13:01 +0000 15.0.0: update -D_GLIBCXX_ASSERTIONS patch for offload Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> 15.0.0/gentoo/15_all_DEF_GENTOO_GLIBCXX_ASSERTIONS.patch | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=95a0fb3364df529328991a746a47ea9bd3ef4e23 commit 95a0fb3364df529328991a746a47ea9bd3ef4e23 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 05:18:49 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 05:19:16 +0000 15.0.0: update -fcf-protection patch for offload Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> 15.0.0/gentoo/26_all_enable-cet.patch | 2 +- 15.0.0/gentoo/README.history | 1 + 2 files changed, 2 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=8883d5d098f3a0b981b04f455748bf13acfc1d32 commit 8883d5d098f3a0b981b04f455748bf13acfc1d32 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 05:24:33 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 05:24:33 +0000 14.2.0: update patches for nvptx/accel Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> 14.2.0/gentoo/04_all_nossp-on-nostdlib.patch | 5 +++-- 14.2.0/gentoo/15_all_DEF_GENTOO_GLIBCXX_ASSERTIONS.patch | 2 +- 14.2.0/gentoo/24_all_DEF_GENTOO_SCP-fstack-clash-protection.patch | 2 +- 14.2.0/gentoo/26_all_enable-cet.patch | 4 ++-- 14.2.0/gentoo/README.history | 7 +++++++ 5 files changed, 14 insertions(+), 6 deletions(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=8b7a28b5604dc83e14f114c97cc465d54c344e75 commit 8b7a28b5604dc83e14f114c97cc465d54c344e75 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 05:27:58 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 05:28:33 +0000 14.2.0: add bashisms patch for nvptx too, cut patchset 6 Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> ...tx-fix-bashisms-with-gen-copyright.sh-use.patch | 71 ++++++++++++++++++++++ 14.2.0/gentoo/README.history | 3 +- 2 files changed, 73 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=97c1985c00e3dace87668528e3d3096ccdbb5304 commit 97c1985c00e3dace87668528e3d3096ccdbb5304 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 05:29:57 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 05:30:11 +0000 sys-devel/gcc: avoid passing hardening flags for nvptx/offload for 14 Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/gcc/Manifest | 1 + sys-devel/gcc/gcc-14.2.1_p20241207.ebuild | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=8a15551bf82f745e3921931ae8843daf27f918df commit 8a15551bf82f745e3921931ae8843daf27f918df Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-10 19:20:52 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-10 19:20:52 +0000 sys-devel/gcc: C23 fixes and nvptx patch tweaks for 15 Bug: https://bugs.gentoo.org/945296 Closes: https://bugs.gentoo.org/944999 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/gcc/Manifest | 1 + sys-devel/gcc/gcc-15.0.0_pre20241208-r1.ebuild | 53 ++++++++++++++++++++++++++ 2 files changed, 54 insertions(+)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=b85da6096998aefba65d432607c12748efccc907 commit b85da6096998aefba65d432607c12748efccc907 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-18 02:45:05 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-18 02:45:05 +0000 toolchain.eclass: handle nvptx multilib changes in 15 GCC trunk recently changed its default multilib for nvptx. Make the `mv` there nonfatal with some comments added. We may want to enable all multilibs in future. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> eclass/toolchain.eclass | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
Updated TODO: * Fix needing to add -fno-stack-protector (**) * Fix F_S (**) * Fix cet still being added (**) * Do we need to backport the PIE patch from jakub or is -z,notext good enough for now (doubt and more awkward to implement, even)? (**) Partially done
Created attachment 914813 [details] log in txt form
Created attachment 914814 [details] stage 1 log.xz
Created attachment 914815 [details] gcc config logs
make[2]: Entering directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' /bin/sh /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-opt.sh "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" \ > tmp-nvptx-gen.opt make[2]: *** [/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/t-nvptx:28: s-nvptx-gen-opt] Error 1 make[2]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' make[2]: *** Waiting for unfinished jobs....
Huh. Can you run: cd /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc && /bin/sh /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-opt.sh "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" ; echo $? and show the output? If it's empty, please run: cd /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc && /bin/bash -x /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-opt.sh "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" ; echo $?
cd /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc && /bin/sh /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-opt.sh "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" ; echo $? ; -*- buffer-read-only: t -*- ; Generated automatically by gen-opt.sh from nvptx-sm.def. Unknown style: "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" 1
the second command yields d /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc && /bin/bash -x /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-opt.sh "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" ; echo $? + nvptx_sm_def=/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/nvptx-sm.def + gen_copyright_sh=/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-copyright.sh ++ grep '^NVPTX_SM' /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/nvptx-sm.def ++ sed 's/.*(//;s/,.*//' + sms='30 35 53 70 75 80' + last= + for sm in $sms + last=30 + for sm in $sms + last=35 + for sm in $sms + last=53 + for sm in $sms + last=70 + for sm in $sms + last=75 + for sm in $sms + last=80 + cat ; -*- buffer-read-only: t -*- ; Generated automatically by gen-opt.sh from nvptx-sm.def. + echo + NVPTX_GEN_COPYRIGHT=opt + . /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-copyright.sh ++ style=/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx ++ case $style in ++ echo 'Unknown style: "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx"' Unknown style: "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" ++ exit 1 1
i guess its that bash problem again.. should I apply some patch? I have removed the gcc-config patch of course, since otherwise it would not compile and because I thought that fix was included by now into portage...
I see the problem, I'll have it fixed shortly
thank you for your fast reply.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=afe723f516bec463e75e08da80cf1b8b093336c7 commit afe723f516bec463e75e08da80cf1b8b093336c7 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-24 20:47:40 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-24 20:47:56 +0000 15.0.0: update bashisms patch Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> ...PR117854-config-nvptx-fix-bashisms-with-gen-copyright.sh-use.patch | 2 +- 15.0.0/gentoo/README.history | 4 ++++ 2 files changed, 5 insertions(+), 1 deletion(-) https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=e1f3439619beb54addc075864c61b15f9d775332 commit e1f3439619beb54addc075864c61b15f9d775332 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-24 20:47:02 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-24 20:47:02 +0000 14.2.0: update bashisms patch Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> ...PR117854-config-nvptx-fix-bashisms-with-gen-copyright.sh-use.patch | 2 +- 14.2.0/gentoo/README.history | 4 ++++ 2 files changed, 5 insertions(+), 1 deletion(-)
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=6a1118133f5c5c13ef564404b96c54c4f8a5f3ca commit 6a1118133f5c5c13ef564404b96c54c4f8a5f3ca Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-24 20:49:15 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-24 20:49:15 +0000 sys-devel/gcc: update nvptx bashisms patch Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> sys-devel/gcc/Manifest | 3 ++- sys-devel/gcc/gcc-14.2.1_p20241221.ebuild | 2 +- sys-devel/gcc/gcc-15.0.0_pre20241222.ebuild | 2 +- 3 files changed, 4 insertions(+), 3 deletions(-)
(In reply to Benjamin Schulz from comment #145) > I have removed the gcc-config patch of course, since otherwise it would not > compile and because I thought that fix was included by now into portage... It is included, just not yet in stable gcc-config. I'll file the stablereq for it soon but it's a critical component so breakage in it can be pretty rough, so didn't want to rush it.
(In reply to Sam James from comment #146) > I see the problem, I'll have it fixed shortly For completeness, the issue was that in 14 vs 15, the semantics are different as of: commit 61da068638779580684c59d6d330f58c016a2268 Author: Thomas Schwinge <tschwinge@baylibre.com> Date: Mon Dec 2 16:42:14 2024 +0100 'gcc/config/nvptx/gen-*.sh': Simplify interface What we currently pass in as '$1' is simply 'dirname "$0"'. gcc/ * config/nvptx/gen-h.sh: Don't pass in '$1'; compute it locally. * config/nvptx/gen-multilib-matches.sh: Likewise. * config/nvptx/gen-omp-device-properties.sh: Likewise. * config/nvptx/gen-opt.sh: Likewise. * config/nvptx/t-nvptx (s-nvptx-gen-h:, s-nvptx-gen-opt:) (t-nvptx-gen-multilib-matches:): Adjust. * config/nvptx/t-omp-device (omp-device-properties-nvptx): Likewise.
by the way, i have set crossdev-9999 into the package.keywords directory. I think crossdev then had installed a file called cross-nvptx-none in the /etc/portage/accept-keywords directory. There, one can find cross-nvptx-none/newlib * ~* -amd64 -~amd64 cross-nvptx-none/nvptx-tools * ~* -amd64 -~amd64 cross-nvptx-none/gcc * ~* -amd64 -~amd64 cross-nvptx-none/gdb * ~* -amd64 -~amd64 which looks ok. but there is also a file with the same name in /etc/portage/package.use/ In /etc/portage/package.use/cross-nvptx-none, i can find cross-nvptx-none/nvptx-tools multilib cross-nvptx-none/gcc cxx openmp fortran -hardened -default-stack-clash-protection -ssp zlib zstd -sanitize -vtv -selinux -d -objc -objc++ -objc-gc -vtv -fortran -go -jit -cxx -openmp -sanitize -zstd -z> I do not know what that means. First openmp and then -openmp, or fortran and then -fortran? I have of course openmp and fortran as + in my make.conf now when I type emerge -uUDNpv world, then I see this: [ebuild U ] cross-nvptx-none/gcc-14.2.1_p20241221:14::crossdev [14.2.1_p20241116:14::crossdev] USE="lto (multilib) nls pie systemtap (-ada) (-cet) -custom-cflags -cxx* (-d) -debug (-default-stack-clash-protection) -default-znow -doc -fixed-point -fortran* (-go) (-graphite) (-hardened) -ieee-long-double -jit* -libssp (-modula2) -objc -objc++ -objc-gc -openmp* (-pch) -pgo (-rust) -sanitize (-ssp) -test (-time64) -valgrind -vanilla -vtv -zstd*" 0 KiB I thought that only crossdev should emerge this, not portage -uUDN world, but why do I see there -openmp and why -fortran? sys-devel gcc would install with this: sys-devel/gcc-14.2.1_p20241221:14::gentoo USE="cet (cxx) d (default-stack-clash-protection) (default-znow) fortran go graphite jit lto modula2 (multilib) nls objc objc++ openmp (pie) rust sanitize ssp systemtap vtv zstd -ada (-custom-cflags) -debug -doc (-fixed-point) -hardened (-ieee-long-double) (-libssp) -objc-gc (-pch) -pgo -test (-time64) -valgrind -vanilla" 0 KiB anyway, crossdev -t nvptx-none would lead to an error described in the following logs. I have of course done an emerge --sync before... twice...
Created attachment 914827 [details] info.log
Created attachment 914828 [details] log.xz
Created attachment 914829 [details] logs.tar.xz
your patch by the way seemed to work: So the last error is due to something else... -opt.sh "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" ; echo $? + nvptx_sm_def=/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/nvptx-sm.def + gen_copyright_sh=/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-copyright.sh ++ grep '^NVPTX_SM' /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/nvptx-sm.def ++ sed 's/.*(//;s/,.*//' + sms='30 35 53 70 75 80' + last= + for sm in $sms + last=30 + for sm in $sms + last=35 + for sm in $sms + last=53 + for sm in $sms + last=70 + for sm in $sms + last=75 + for sm in $sms + last=80 + cat ; -*- buffer-read-only: t -*- ; Generated automatically by gen-opt.sh from nvptx-sm.def. + echo + NVPTX_GEN_COPYRIGHT=opt + . /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-copyright.sh ++ style=opt ++ case $style in ++ cat ++ read line ++ case $style in ++ '[' 'Copyright (C) 2022-2024 Free Software Foundation, Inc.' = '' ']' ++ echo '; Copyright (C) 2022-2024 Free Software Foundation, Inc.' ; Copyright (C) 2022-2024 Free Software Foundation, Inc. ++ read line ++ case $style in ++ '[' '' = '' ']' ++ echo ';' ; ++ read line ++ case $style in ++ '[' 'This file is part of GCC.' = '' ']' ++ echo '; This file is part of GCC.' ; This file is part of GCC. ++ read line ++ case $style in ++ '[' '' = '' ']' ++ echo ';' ; ++ read line ++ case $style in ++ '[' 'GCC is free software; you can redistribute it and/or modify it under' = '' ']' ++ echo '; GCC is free software; you can redistribute it and/or modify it under' ; GCC is free software; you can redistribute it and/or modify it under ++ read line ++ case $style in ++ '[' 'the terms of the GNU General Public License as published by the Free' = '' ']' ++ echo '; the terms of the GNU General Public License as published by the Free' ; the terms of the GNU General Public License as published by the Free ++ read line ++ case $style in ++ '[' 'Software Foundation; either version 3, or (at your option) any later' = '' ']' ++ echo '; Software Foundation; either version 3, or (at your option) any later' ; Software Foundation; either version 3, or (at your option) any later ++ read line ++ case $style in ++ '[' version. = '' ']' ++ echo '; version.' ; version. ++ read line ++ case $style in ++ '[' '' = '' ']' ++ echo ';' ; ++ read line ++ case $style in ++ '[' 'GCC is distributed in the hope that it will be useful, but WITHOUT ANY' = '' ']' ++ echo '; GCC is distributed in the hope that it will be useful, but WITHOUT ANY' ; GCC is distributed in the hope that it will be useful, but WITHOUT ANY ++ read line ++ case $style in ++ '[' 'WARRANTY; without even the implied warranty of MERCHANTABILITY or' = '' ']' ++ echo '; WARRANTY; without even the implied warranty of MERCHANTABILITY or' ; WARRANTY; without even the implied warranty of MERCHANTABILITY or ++ read line ++ case $style in ++ '[' 'FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License' = '' ']' ++ echo '; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License' ; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License ++ read line ++ case $style in ++ '[' 'for more details.' = '' ']' ++ echo '; for more details.' ; for more details. ++ read line ++ case $style in ++ '[' '' = '' ']' ++ echo ';' ; ++ read line ++ case $style in ++ '[' 'You should have received a copy of the GNU General Public License' = '' ']' ++ echo '; You should have received a copy of the GNU General Public License' ; You should have received a copy of the GNU General Public License ++ read line ++ case $style in ++ '[' 'along with GCC; see the file COPYING3. If not see' = '' ']' ++ echo '; along with GCC; see the file COPYING3. If not see' ; along with GCC; see the file COPYING3. If not see ++ read line ++ case $style in ++ '[' '<http://www.gnu.org/licenses/>.' = '' ']' ++ echo '; <http://www.gnu.org/licenses/>.' ; <http://www.gnu.org/licenses/>. ++ read line ++ case $style in + echo + cat Enum Name(ptx_isa) Type(int) Known PTX ISA target architectures (for use with the -misa= option): + echo + for sm in $sms + cat EnumValue Enum(ptx_isa) String(sm_30) Value(PTX_ISA_SM30) + '[' 30 = 80 ']' + echo + for sm in $sms + cat EnumValue Enum(ptx_isa) String(sm_35) Value(PTX_ISA_SM35) + '[' 35 = 80 ']' + echo + for sm in $sms + cat EnumValue Enum(ptx_isa) String(sm_53) Value(PTX_ISA_SM53) + '[' 53 = 80 ']' + echo + for sm in $sms + cat EnumValue Enum(ptx_isa) String(sm_70) Value(PTX_ISA_SM70) + '[' 70 = 80 ']' + echo + for sm in $sms + cat EnumValue Enum(ptx_isa) String(sm_75) Value(PTX_ISA_SM75) + '[' 75 = 80 ']' + echo + for sm in $sms + cat EnumValue Enum(ptx_isa) String(sm_80) Value(PTX_ISA_SM80) + '[' 80 = 80 ']' + continue 0
make[2]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' make[2]: Entering directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' /bin/sh /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/gen-opt.sh "/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx" \ > tmp-nvptx-gen.opt make[2]: *** [/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/t-nvptx:28: s-nvptx-gen-opt] Error 1 make[2]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' make[2]: *** Waiting for unfinished jobs.... make[2]: Entering directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' TARGET_CPU_DEFAULT="" \ HEADERS="auto-host.h ansidecl.h" DEFINES="" \
ah no, wrong file /bin/sh ./libtool --mode=install /usr/lib/portage/python3.12/ebuild-helpers/xattr/install -c libatomic.la '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib' libtool: install: /usr/lib/portage/python3.12/ebuild-helpers/xattr/install -c .libs/libatomic.lai /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/libatomic.la libtool: install: /usr/lib/portage/python3.12/ebuild-helpers/xattr/install -c .libs/libatomic.a /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/libatomic.a libtool: install: chmod 644 /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/libatomic.a libtool: install: /usr/nvptx-none/bin/ranlib /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/libatomic.a libtool: install: warning: remember to run `libtool --finish /usr/nvptx-none/lib' make[4]: Nothing to be done for 'install-data-am'. make[4]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/nvptx-none/libatomic' make[3]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/nvptx-none/libatomic' make[2]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/nvptx-none/libatomic' make[1]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build' mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/mgomp/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/mgomp/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/mptx-3.1/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/mptx-3.1/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/mgomp/mptx-3.1/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/nvptx-none/lib/mgomp/mptx-3.1/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/image/usr/lib/gcc/nvptx-none/14/*.spec': No such file or directory [31;01m*[0m ERROR: cross-nvptx-none/gcc-14.2.1_p20241221::crossdev failed (install phase): [31;01m*[0m (no error message) [31;01m*[0m [31;01m*[0m Call stack: [31;01m*[0m ebuild.sh, line 136: Called src_install [31;01m*[0m environment, line 3728: Called toolchain_src_install [31;01m*[0m environment, line 5502: Called gcc_movelibs [31;01m*[0m environment, line 2215: Called die [31;01m*[0m The specific snippet of code: [31;01m*[0m mv "${ED}"/usr/lib/gcc/nvptx-none/${GCCMAJOR}/*.{a,spec} "${ED}"/usr/lib/gcc/${CHOST}/${GCCMAJOR}/accel/nvptx-none/ || die; [31;01m*[0m [31;01m*[0m If you need support, post the output of `emerge --info '=cross-nvptx-none/gcc-14.2.1_p20241221::crossdev'`, [31;01m*[0m the complete build log and the output of `emerge -pqv '=cross-nvptx-none/gcc-14.2.1_p20241221::crossdev'`. [31;01m*[0m [31;01m*[0m Please include /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-build-logs.tar.xz in your bug report. [31;01m*[0m [31;01m*[0m The complete build log is located at '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/temp/build.log'. [31;01m*[0m The ebuild environment file is located at '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/temp/environment'. [31;01m*[0m Working directory: '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build' [31;01m*[0m S: '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221' >>> Failed to emerge cross-nvptx-none/gcc-14.2.1_p20241221, Log file: >>> '/var/tmp/portage/cross-nvptx-n
Created attachment 914830 [details] info.log
Created attachment 914831 [details] log.xz
Created attachment 914832 [details] tar.xz
86_64-pc-linux-gnu-g++ -fPIE -c -O2 -pipe -DIN_GCC -DCROSS_DIRECTORY_STRUCTURE -fno-exceptions -fno-rtti -fasynchronous-unwind-tables -W -Wall -Wno-narrowing -Wwrite-strings -Wcast-qual -Wmissing-format-attribute -Wconditionally-supported -Woverloaded-virtual -pedantic -Wno-long-long -Wno-variadic-macros -Wno-overlength-strings -DHAVE_CONFIG_H -fPIE -I. -I. -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/. -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/../include -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/../libcpp/include -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/../libcody -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/../libdecnumber -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/../libdecnumber/dpd -I../libdecnumber -I/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/../libbacktrace -D_GENTOO_TIME64_FORCE=0 -o nvptx.o -MT nvptx.o -MMD -MP -MF ./.deps/nvptx.TPo /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/nvptx.cc /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/nvptx.cc: In function ‘void handle_ptx_version_option()’: /var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/gcc-14-20241221/gcc/config/nvptx/nvptx.cc:325:12: warning: unquoted identifier or keyword ‘sm_’ in format [-Wformat-diag] 325 | error ("PTX version (%<-mptx%>) needs to be at least %s to support selected" | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 326 | " %<-misa%> (sm_%s)", ptx_version_to_string (first), | ~~~~~~~~~~~~~~~~~~~~ make[2]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' make[2]: Entering directory '/var/tmp/portage/cross-nvptx-none/gcc-14.2.1_p20241221/work/build/gcc' x86_64-pc-linux-gnu-g++ -fPIE -c -O2 -pipe -DIN_GCC -DCROSS_DIRECTORY_STRUCTURE -fno-exceptions -fno-r
I don't get why you're hitting that now, as nothing's changed there, but this should fix it.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=afc0f03b6e846e1c0aa104a809f86133bad55b6d commit afc0f03b6e846e1c0aa104a809f86133bad55b6d Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-25 00:40:42 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-25 00:42:07 +0000 crossdev: nvptx: force on openmp & fortran Without openmp, various critical files are missing. Offloading makes no sense without OpenMP. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 1 + 1 file changed, 1 insertion(+)
(In reply to Sam James from comment #163) > I don't get why you're hitting that now, as nothing's changed there, but > this should fix it. It's possible you hit it because those things are stabled for *stage1* but need to be re-enabled later, actually, which also means my fix is wrong and needs to be reverted.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=a63705c41afcca7b8cf07fe26076363d4d21e080 commit a63705c41afcca7b8cf07fe26076363d4d21e080 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-25 00:43:14 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-25 00:43:14 +0000 Revert "crossdev: nvptx: force on openmp & fortran" This reverts commit afc0f03b6e846e1c0aa104a809f86133bad55b6d. We have to be able to turn these off for stage1 but we need them on for stage2... need to figure out how to do that. Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 1 - 1 file changed, 1 deletion(-)
(In reply to Sam James from comment #165) > (In reply to Sam James from comment #163) > > I don't get why you're hitting that now, as nothing's changed there, but > > this should fix it. > > It's possible you hit it because those things are stabled for *stage1* but > need to be re-enabled later, actually, which also means my fix is wrong and > needs to be reverted. If you do crossdev -C nvptx-none, then retry, now that stage1 doesn't fail on the bash/nvptx thing, it should be OK.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/proj/crossdev.git/commit/?id=a95616adbecd27e9f88539a588f95d2560eb9819 commit a95616adbecd27e9f88539a588f95d2560eb9819 Author: Sam James <sam@gentoo.org> AuthorDate: 2024-12-25 00:44:52 +0000 Commit: Sam James <sam@gentoo.org> CommitDate: 2024-12-25 00:44:52 +0000 crossdev: nvptx: add comment wrt GFORCE and openmp/fortran Bug: https://bugs.gentoo.org/945296 Signed-off-by: Sam James <sam@gentoo.org> crossdev | 2 ++ 1 file changed, 2 insertions(+)
thanks. after crossdev -C nvptx-none, the command crossdev -t nvptx-none worked and it compiled some test programs.
Thank you too! I'm going to file a bug for the change to crossdev to make it more robust.
by the way, why -graphite? For gpu offload, I guess one would like loop optimizations if they can not be tackeld by openmp parallel directives. can't it handle graphite or is graphite incompatible with openmp?
I don't remember if I tried it and found it broken or if it was pre-emptive to avoid failures. You're free to try it, although really, both graphite (and polly in llvm) don't get much love and the focus these days is usually on getting those kinds of optimisations to be native as part of the usual passes.
ah well, i see the recent useflags are these cross-nvptx-none/nvptx-tools multilib cross-nvptx-none/newlib -selinux -libraries multilib cross-nvptx-none/gcc cxx openmp fortran -hardened -default-stack-clash-protection -ssp zlib zstd -sanitize -vtv -selinux -d -objc -objc++ -objc-gc -vtv -sanitize multilib that looks reasonable to me.
hm for the attached files, gcc claims "alias definitions would not be supported". He points to these, but i do not think that these are "alias" constructors. They have obviously different arguments: #pragma omp begin declare target template<typename T> datastruct<T>::datastruct( T* data, size_t pdatalength, bool rowm, size_t rank, size_t* extents, size_t* strides, bool compute_datalength, bool compute_strides_from_extents ) : pdata(data), pextents(extents), pstrides(strides), pdatalength(pdatalength), prank(rank), prowmayor(rowm) { if(compute_strides_from_extents==true && pextents!=nullptr && pstrides!=nullptr && rank !=0) { fill_strides(pextents,pstrides,rank,rowm); } if(compute_datalength==true && pextents!=nullptr && pstrides!=nullptr && rank !=0) { pdatalength=compute_data_length(pextents,pstrides,rank); } } #pragma omp end declare target #pragma omp begin declare target template<typename T> datastruct<T>::datastruct( T* data, size_t datalength, bool rowm, size_t rows, size_t cols, size_t* extents, size_t* strides, bool compute_datalength, bool compute_strides_from_extents ) : pdata(data), pextents(extents), pstrides(strides), pdatalength(datalength), prank(2), prowmayor(rowm) { if(extents!=nullptr) { pextents[0]=(rowm==true)?rows:cols; pextents[1]=(rowm==true)?cols:rows; } if(pstrides!=nullptr && compute_strides_from_extents) { pstrides[0]=(rowm==true)? cols:1; pstrides[1]=(rowm==true)?1: rows; } if(compute_datalength==true && extents!=nullptr && strides!=nullptr) { pdatalength=(rows-1) * strides[0]+(cols-1)*strides[1]+1; } } #pragma omp end declare target #pragma omp begin declare target template<typename T> datastruct<T>::datastruct( T* data, size_t datalength, bool rowm, bool rowvector, size_t noelements, size_t* extents, size_t* strides, bool compute_datalength, bool compute_strides_from_extents ) : pdata(data), pextents(extents), pstrides(strides), pdatalength(datalength), prank(1), prowmayor(true) { if(extents!=nullptr) { pextents[0]=noelements; } if(pstrides!=nullptr && compute_strides_from_extents) { if(rowvector) pstrides[0]=(rowm==true)? 1:noelements; else pstrides[0]=(rowm==true)? noelements:1; } if(compute_datalength==true && strides!=nullptr) { pdatalength=(noelements-1) * strides[0]+1; } }
Created attachment 916622 [details] arraytestclass
Created attachment 916623 [details] arraytest-main
Created attachment 916624 [details] cmakelists.txt
i just report this because i think i remember have seen a situation where gcc claimed this erroneously. But i am not sure if that is still there as a compiler bug and if that is something for upstream....
I think it'll need reporting upstream to GCC indeed.
because of this gcc 14 regression https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118590 I tried to build gcc 13 as a cross compiler, where this regression is said not to occur by upstream. That failed * crossdev version: 9cc43baef9c8a70213a48815d38cc395084f3902 * Host Portage ARCH: amd64 * Host Portage System: x86_64-pc-linux-gnu (i686-pc-linux-gnu x86_64-pc-linux-gnu) * Target Portage ARCH: * * Target System: nvptx-none * Stage: 4 (C/C++ compiler) * USE=multilib: yes * Target ABIs: default * binutils: nvptx-tools-[latest] * gcc: gcc-13 * libc: newlib-[latest] * CROSSDEV_OVERLAY: /var/db/repos/crossdev * PORT_LOGDIR: /var/log/portage * PORTAGE_CONFIGROOT: / * Portage flags: _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - * leaving sys-libs/newlib in /var/db/repos/crossdev * leaving sys-devel/nvptx-tools in /var/db/repos/crossdev * leaving sys-devel/gcc in /var/db/repos/crossdev * leaving dev-debug/gdb in /var/db/repos/crossdev * leaving metadata/layout.conf alone in /var/db/repos/crossdev _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - ~ - _ - * Log: /var/log/portage/cross-nvptx-none-nvptx-tools.log * Emerging cross-nvptx-tools ... [ ok ] * Log: /var/log/portage/cross-nvptx-none-gcc-stage1.log * Emerging cross-gcc-stage1 ...
Created attachment 917292 [details] info.log
Created attachment 917293 [details] log.gz
Created attachment 917294 [details] gcc-config.log
what i tried was this command crossdev -t nvptx-none --gcc 13 the error seems to be this: checking linker --sysroot support... no checking __stack_chk_fail in target C library... yes checking sys/sdt.h in the target C library... no configure: error: sys/sdt.h was not found make[1]: *** [Makefile:4535: configure-gcc] Error 1 make[1]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-12.5.9999/work/build' make: *** [Makefile:1014: all] Error 2 [31;01m*[0m ERROR: cross-nvptx-none/gcc-12.5.9999::crossdev failed (compile phase): [31;01m*[0m emake failed [31;01m*[0m [31;01m*[0m If you need support, I do not know why it tried to emerge 12.5.9999, as i asked for 13... but well, I do not know if this is possible at all? does eselect gcc support to have various offlload compilers?
Try --gcc "13.4" or so.
Created attachment 917295 [details] stage1 log
Created attachment 917296 [details] config log
Created attachment 917297 [details] info.log
crossdev -t nvptx-none --gcc 13.4 failed at install phase apparently btool: install: /usr/lib/portage/python3.12/ebuild-helpers/xattr/install -c .libs/libatomic.lai /var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/libatomic.la libtool: install: /usr/lib/portage/python3.12/ebuild-helpers/xattr/install -c .libs/libatomic.a /var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/libatomic.a libtool: install: chmod 644 /var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/libatomic.a libtool: install: /usr/nvptx-none/bin/ranlib /var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/libatomic.a libtool: install: warning: remember to run `libtool --finish /usr/nvptx-none/lib' make[4]: Nothing to be done for 'install-data-am'. make[4]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/work/build/nvptx-none/libatomic' make[3]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/work/build/nvptx-none/libatomic' make[2]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/work/build/nvptx-none/libatomic' make[1]: Leaving directory '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/work/build' mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/mgomp/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/mgomp/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/mptx-3.1/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/mptx-3.1/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/mgomp/mptx-3.1/*.a': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/nvptx-none/lib/mgomp/mptx-3.1/*.spec': No such file or directory mv: cannot stat '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/image/usr/lib/gcc/nvptx-none/13/*.spec': No such file or directory [31;01m*[0m ERROR: cross-nvptx-none/gcc-13.3.1_p20250117::crossdev failed (install phase): [31;01m*[0m (no error message) [31;01m*[0m [31;01m*[0m Call stack: [31;01m*[0m ebuild.sh, line 136: Called src_install [31;01m*[0m environment, line 3728: Called toolchain_src_install [31;01m*[0m environment, line 5504: Called gcc_movelibs [31;01m*[0m environment, line 2215: Called die [31;01m*[0m The specific snippet of code: [31;01m*[0m mv "${ED}"/usr/lib/gcc/nvptx-none/${GCCMAJOR}/*.{a,spec} "${ED}"/usr/lib/gcc/${CHOST}/${GCCMAJOR}/accel/nvptx-none/ || die; [31;01m*[0m [31;01m*[0m If you need support, post the output of `emerge --info '=cross-nvptx-none/gcc-13.3.1_p20250117::crossdev'`, [31;01m*[0m the complete build log and the output of `emerge -pqv '=cross-nvptx-none/gcc-13.3.1_p20250117::crossdev'`. [31;01m*[0m [31;01m*[0m Please include /var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/work/gcc-build-logs.tar.xz in your bug report. [31;01m*[0m [31;01m*[0m The complete build log is located at '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/temp/build.log'. [31;01m*[0m The ebuild environment file is located at '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/temp/environment'. [31;01m*[0m Working directory: '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/work/build' [31;01m*[0m S: '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/work/gcc-13-20250117' >>> Failed to emerge cross-nvptx-none/gcc-13.3.1_p20250117, Log file: >>> '/var/tmp/portage/cross-nvptx-none/gcc-13.3.1_p20250117/temp/build.log' * Messages for package cross-nvptx-none/gcc-13.3.1_p20250117: * ERROR: cross-nvptx-none/gcc-13.3.1_p20250117::crossdev failed (install phase): * (no error message) *
Not clear to me why it failed, but note that for offloading, you need the system compiler to be 13 as well (at least for building it), so you need to temporarily `gcc-config` your main compiler to be 13 too.
Hi it appears that before I emerge a new cross compiler, i have to delete the old one. Unfortunately, what is more, gcc 13.4 also appears to have that bug which I reported. Gcc's bugzilla says that 13.2 would be bug free. However, if i eselect gcc 13 and then emerge 13.2 as a cross compiler, and when I then want to compile a sample program, i get this: [ 50%] Linking CXX executable arraytest /usr/bin/cmake -E cmake_link_script CMakeFiles/arraytest.dir/link.txt --verbose=1 lto1: severe error: bydecode stream in file »CMakeFiles/arraytest.dir/main_acc.cpp.o« with LTO-Version 13.1 instead of the expected 13.0 created. compilation stopped. Additionally, by the way, nvc++ and clang can not find the header openacc.h. Do you know where this is? and the associated openacc runtime? And,yes gcc also has problems finding that file when i want to compile something with an offload gcc 14 compiler and call acc_malloc for which i have to #include <openacc.h> So at the moment nothing works on my end..
(In reply to Benjamin Schulz from comment #191) > Hi it appears that before I emerge a new cross compiler, i have to delete > the old one. > > Unfortunately, what is more, gcc 13.4 also appears to have that bug which I > reported. Gcc's bugzilla says that 13.2 would be bug free. I'm not sure where you're seeing that. > > However, if i eselect gcc 13 and then emerge 13.2 as a cross compiler, and > when I then want to compile a sample program, i get this: > > [ 50%] Linking CXX executable arraytest > /usr/bin/cmake -E cmake_link_script CMakeFiles/arraytest.dir/link.txt > --verbose=1 > lto1: severe error: bydecode stream in file > »CMakeFiles/arraytest.dir/main_acc.cpp.o« with LTO-Version 13.1 instead of > the expected 13.0 created. > compilation stopped. You'd have to make sure that the two versions mathc. > > > Additionally, by the way, nvc++ and clang can not find the header openacc.h. > Do you know where this is? and the associated openacc runtime? And,yes gcc > also has problems finding that file when i want to compile something with an > offload gcc 14 compiler and call acc_malloc for which i have to #include > <openacc.h> I've no idea on that.
