The upstream meson build system makes the assumption that all arm CPU's have the neon instruction set. Not all do, e.g. tegra20. Reproducible: Always Editing the ebuild with '-D native-arch-optimization=false' fixes the build. But, that turns off optimizations for everyone. I started modifying the ebuild further to use CPU_FLAGS as a proper selection method, but got stuck trying to find a way to propagate those wishes to the meson build system. It appears that upstream does not currently support this. This is being built in a qemu-arm container with a distcc cross compiler. I suspect that '-D native-arch-optimization=true' breaks normal cross compiling, but that is unverified.
Created attachment 638252 [details] emerge --info
Created attachment 638254 [details, diff] native-arch-optimization=false
The native-arch-optimization does three things: 1: set cpu_arch = true, used during evas build 2: use config_h.set10() to define 'BUILD_FLAG' 3: force global CFLAGS Reference, https://git.enlightenment.org/core/efl.git/tree/meson.build?h=v1.24.1#n165
(In reply to Kyle Evans from comment #0) > The upstream meson build system makes the assumption that all arm CPU's have > the neon instruction set. Not all do, e.g. tegra20. > I guess a fast fix would be to disable native-arch-optimization on all arm boards for now. But I'll open an upstream issue and see what they do. > > I suspect that '-D native-arch-optimization=true' breaks normal cross > compiling, but that is unverified. > Good point.
So upstream said this problem only hits a minor userbase of arm sockets, and due to nature of Gentoo is only being visible here, so we should deal with it. I wonder if there's an "easy" way to detect non-neon supporting arm chips? I see you have USE="arm armv5te armv6 armv6t2", would some of these be any help? I'd rather not query info from native CFLAGS, or live /proc.
(In reply to Joonas Niilola from comment #5) > So upstream said this problem only hits a minor userbase of arm sockets, and > due to nature of Gentoo is only being visible here, so we should deal with > it. That is cool if there is a better way to deal with this than what I had in mind. What are your thoughts? > I wonder if there's an "easy" way to detect non-neon supporting arm chips? I > see you have USE="arm armv5te armv6 armv6t2", would some of these be any > help? I'd rather not query info from native CFLAGS, or live /proc. I would say -cpu_flags_neon, but gentoo gets this wrong too in the armv7 profile. I have it masked in my local profile. There is also USE=-neon. I think relying on the user having these set correctly is not unreasonable.
The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=46b1048c257e1074ddaaaaa6ee88ffbac431a6d7 commit 46b1048c257e1074ddaaaaa6ee88ffbac431a6d7 Author: Joonas Niilola <juippis@gentoo.org> AuthorDate: 2020-06-10 05:16:11 +0000 Commit: Joonas Niilola <juippis@gentoo.org> CommitDate: 2020-06-10 05:16:49 +0000 dev-libs/efl: fix arm[-neon] CPU builds, #722552 Closes: https://bugs.gentoo.org/722552 Signed-off-by: Joonas Niilola <juippis@gentoo.org> dev-libs/efl/efl-1.24.2.ebuild | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)
