aeubanks wrote: I have heard that x86-64 `-fno-pic` is measurably slower than `-fpie` in large workloads, e.g. having to read RIP when accessing globals RIP-relatively is measurable, as opposed a constant address. There are valid reasons for `-fno-pic`.
It seems unlikely that having to support different layouts for different architectures will impose any noticeable maintenance burden on lld's section sorting code, it's fairly straightforward. > "fail to support -no-pie -mcmodel=medium" is an overstatement. I don't think this is an overstatement, lld is not respecting the fact that large sections should not contribute to 32-bit relocation pressure in `-fno-pic` builds, and this actually causes issues in practice. But again, I'd rather see the current version of this PR go in rather than not go in. https://github.com/llvm/llvm-project/pull/81224 _______________________________________________ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits