[PATCH] D112063: [lld][ELF] Add first bits to support relocation relaxations for AArch64

Eric Christopher via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Dec 14 15:55:18 PST 2021


echristo added inline comments.


================
Comment at: lld/test/ELF/aarch64-adrp-ldr-got.s:97
+.global main
+main:
+  adrp    x1, :got:x
----------------
alexander-shaposhnikov wrote:
> MaskRay wrote:
> > alexander-shaposhnikov wrote:
> > > MaskRay wrote:
> > > > Unlike Mach-O, ELF uses `_start`. Missing `_start` leads to a warning in an executable link.
> > > > 
> > > > If you combine the tests, lots of boilerplate (duplicate `.rodata`, duplicate `main:`, etc) can be avoided.
> > > one day the tests will fail and debugging a small input is much, much easier.
> > > I'd prefer to keep them separate.
> > I may disagree with this. AArch64 does not decrease the number of instructions. If FileCheck reports an error at one line, it is quick to know associated the error location with the source, especially if you use distinct symbols in the first place.
> I'm sorry, but I have to disagree. It's not the problem with FileCheck, a lot of things are happening inside the linker,
> it's easier to reason about (and to debug) small independent inputs rather than bigger binaries (especially when multiple different code paths are triggered intentionally).
FWIW smaller tests in separate files also parallelize more easily during test runs if it helps. Hopefully the boiler plate won't need to be adjusted often or do you have a different concern?



Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112063/new/

https://reviews.llvm.org/D112063



More information about the llvm-commits mailing list