-
Notifications
You must be signed in to change notification settings - Fork 553
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[libunwind] Add v1.8.1 #7959
[libunwind] Add v1.8.1 #7959
Conversation
https://buildkite.com/julialang/yggdrasil/builds/7797#018d136e-5e99-499b-a460-7d62227df9f9/620-2148
May need a newer gcc to compile this asm, I've seen similar assembler error before on aaarch64 |
@giordano, any idea what version of GCC might be needed? Some googling suggests that this |
Seems people building TensorFlow on Linux AArch64 hit the same inline assembly issue we're seeing here and this (plus `-flax-vector-conversion`) seems to fix it for them. This flag seems sus for libunwind but worth a try.
L/LibUnwind/LibUnwind@1.8.1/bundled/patches/libunwind-configure-static-lzma.patch
Outdated
Show resolved
Hide resolved
L/LibUnwind/LibUnwind@1.8.1/bundled/patches/libunwind-configure-static-lzma.patch
Outdated
Show resolved
Hide resolved
Co-authored-by: Mosè Giordano <giordano@users.noreply.github.com>
Also add line breaks for long line
Ah, this is apparently a known issue: libunwind/libunwind#742 |
Should hopefully fix the inline assembly issue on AArch64?
12 seems excessive; 1.7.2 uses 5, let's try that
I've included libunwind/libunwind#748 as a patch, which fixes building on Linux AArch64. An alternative approach would have been to include a patch that reverts libunwind/libunwind#503, which is the approach that the .NET runtime took when upgrading to libunwind 1.8.0 (dotnet/runtime#96969). Evidently Xamarin has their own fix that differs from the what's implemented in the patch I've included: dotnet/android#8857. |
* [libunwind] Add v1.8.0 * [libunwind 1.8.0] Amend patch * [libunwind 1.8.0] Set preferred GCC version to 6 * [libunwind 1.8.0] idk, how about gcc 12 * [libunwind] 1.8.0 -> 1.8.1 * [libunwind@1.8.1] Try `-fomit-frame-pointer` on Linux AArch64 Seems people building TensorFlow on Linux AArch64 hit the same inline assembly issue we're seeing here and this (plus `-flax-vector-conversion`) seems to fix it for them. This flag seems sus for libunwind but worth a try. * Thank you, Mosè Co-authored-by: Mosè Giordano <giordano@users.noreply.github.com> * [libunwind@1.8.1] rm `-fomit-frame-pointer`, prefer GCC 12 Also add line breaks for long line * [libunwind@1.8.1] Include patch for PR 748 Should hopefully fix the inline assembly issue on AArch64? * [libunwind@1.8.1] Try re-lowering the preferred GCC version 12 seems excessive; 1.7.2 uses 5, let's try that * Update L/LibUnwind/LibUnwind@1.8.1/build_tarballs.jl * Revert "Update L/LibUnwind/LibUnwind@1.8.1/build_tarballs.jl" This reverts commit c37f8f9. --------- Co-authored-by: Mosè Giordano <giordano@users.noreply.github.com> Co-authored-by: Mosè Giordano <mose@gnu.org>
List of changes: libunwind/libunwind#698