On 9/10/21 11:27 AM, Luboš Luňák wrote:
On Thursday 09 of September 2021, Miklos Vajna wrote:
* tb86 (m1 mac mini) build failures with Illegal Instruction
https://gerrit.libreoffice.org/c/core/+/115675 “Upgrade SSE2 sum to
AVX512 sum with Neumaier 1"
That doesn't make sense. M1 Mac Mini should be ARM64, but build logs for tb86
claim it's "x86_64-apple-darwin20.6.0". And I have M1 Mac Mini here and those
tests pass fine. I also reviewed that gerrit change and AFAICT the extra CPU
code is properly separated and run only when the support is available. Are
tb86 builds running in some kind of x86_64 virtualization?
Those builds are apparently meant to generate x86-64 binaries, see
<https://git.libreoffice.org/lode/+/38ea870f6ca4b14e2ab768a82ef09ba3ec878625%5E%21>
"add host/build flags to autogen job when building on macos-arm64",
though it beats me how that actually works, given that no traces of the
--host=x86_64-apple-macos --build=x86_64-apple-macos
from that commit's bin/macosx_clang_dbgutil.env show up in tb86 build
logs like <https://ci.libreoffice.org/job/gerrit_mac/100772/consoleFull>.
