What a time to be alive - a human having to prove to an AI that he isn't crazy :)
Jokes aside, your AI friend gave you generic textbook advice without checking the actual linker dependency chain. In `kernels/bvh/bvh.cpp`: ```cpp #if !defined(__AVX__) || !defined(EMBREE_TARGET_SSE2) && !defined(EMBREE_TARGET_SSE42) || defined(__aarch64__) template class BVHN<4>; #endif ``` When building with `-flto` or on builders where `__AVX__` is active during compilation of `bvh.cpp`, that `#if` condition evaluates to false. As a result, explicit instantiation of `template class BVHN<4>;` is skipped. However, `kernels/bvh/bvh_builder_subdiv.cpp` (compiled into the multi- ISA `libembree_avx2.a`) still instantiates `BVHNSubdivPatch1MBlurBuilderSAH<4>`, which directly calls `BVHN<4>::preBuild`, `BVHN<4>::set`, and `BVHN<4>::postBuild`. When linking `libembree4.so`, `ld` failed with 235 undefined reference errors for `embree::BVHN<4>::*` symbols (visible in build log #33012149). Verified in PPA (build #33794027): stripping `-march=x86-64-v3` fixed the SSE4.2 symbols, but the link still failed on `BVHN<4>` until `always-instantiate-bvh4.patch` was added. Including both changes in Debian will ensure clean builds under LTO and future toolchain revisions, and will keep Debian and Ubuntu in sync. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2161642 Title: 4.4.1+dfsg-2 FTBFS in stonking proposed on amd64v3 To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/embree/+bug/2161642/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
