diff options
| author | Ionen Wolkens <ionen@gentoo.org> | 2024-10-19 08:49:43 -0400 |
|---|---|---|
| committer | Ionen Wolkens <ionen@gentoo.org> | 2024-10-19 09:47:15 -0400 |
| commit | e91760a3bbc3d9267fb0bfd018b8af834e0b4286 (patch) | |
| tree | c32696bcd7e4be3a1bf17bee917e18400a4fb694 /dev-libs/jsoncpp/jsoncpp-1.9.6.ebuild | |
| parent | 44b358c2437431ba2addd455a6f1c6627d5a37e2 (diff) | |
| download | gentoo-e91760a3bbc3d9267fb0bfd018b8af834e0b4286.tar.gz gentoo-e91760a3bbc3d9267fb0bfd018b8af834e0b4286.tar.bz2 gentoo-e91760a3bbc3d9267fb0bfd018b8af834e0b4286.zip | |
qt6-build.eclass: extend -m* flags sanitizing
A user on the forums reported a build failure with:
`-march=znver4 -mavx512vp2intersect`
On its own, there is no issue there (expanded from -march=native) but
when Qt passes `-march=haswell` to build certain parts (e.g. avx2 bits
for runtime detection regardless of what CHOST supports), it overrides
znver4 and keeps -mavx512vp2intersect resulting in an incomplete set
and then qsimd_p.h complains
Do not plan to pursue this myself, but this specific issue could
be improved upstream by passing -mavx2 and others rather than
-march=haswell (this would also allow overriding a -mno-avx2 that
we currently need to change), and/or ideally by making qsimd_p.h
be able to deal with these configurations without #error.
If problems keep piling up, could consider always filtering *all*
-m{,no-}<instr> flags and doing x86-64-v* unconditionally. Albeit
that'd still require maintaining a list of these not to filter
unrelated -m* flags which could be important.
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
Diffstat (limited to 'dev-libs/jsoncpp/jsoncpp-1.9.6.ebuild')
0 files changed, 0 insertions, 0 deletions
