s390x target seems got slow after update of LLVM.

[HomuHomu833]

Zig Version

0.15.0-dev.263+d8153fa74

Steps to Reproduce and Observed Behavior

As you can see in this run, I made two attempts but hit the GitHub Actions runner limit. However, after downgrading to 0.15.0-dev.137+db7db4802, it compiled fine again. Really strange.

Expected Behavior

New Zig to build faster maybe? idrk

[HomuHomu833]

oh, that also kind of includes hexagon; it's as slow as s390x.

[alexrp]

You're probably hitting: llvm/llvm-project#134714

oh, that also kind of includes hexagon; it's as slow as s390x.

I don't follow; where does Hexagon fit into this?

[HomuHomu833]

I don't follow; where does Hexagon fit into this?

I don't really know, just to make sure I'll run second time and see if that happens again.

[HomuHomu833]

@alexrp yeah hexagon seems does have the same issue as s390x
https://github.com/HomuHomu833-stuff/android-ndk-custom/actions/runs/14420826966
https://github.com/HomuHomu833-stuff/android-ndk-custom/actions/runs/14422762926

[alexrp]

Oh, you're actually trying to build it. Don't bother with that; timeouts aside, there is a bug on the LLVM side that isn't likely to be fixed until LLVM 21, and which prevents building large binaries like zig. The Zig standard library also hasn't been completely ported and tested yet.

[HomuHomu833]

mkay

[HomuHomu833]

@alexrp lil update I compiled through WSL2 and got relocation error ouch

ld.lld: error: lib/libLLVMInstCombine.a(InstCombineVectorOps.cpp.o):(function evaluateInDifferentElementOrder(llvm::Value*, llvm::ArrayRef<int>, llvm::IRBuilderBase&): .text._ZL31evaluateInDifferentElementOrderPN4llvm5ValueENS_8ArrayRefIiEERNS_13IRBuilderBaseE+0xd0c): relocation R_HEX_B22_PCREL out of range: 11479108 is not in [-8388608, 8388607]; references '__ubsan_handle_type_mismatch_v1'
>>> referenced by InstCombineVectorOps.cpp
>>> defined in /home/homura/.cache/zig/o/9fdb2db183d32c7be8cdd99605e4091c/libubsan_rt.a(/home/homura/.cache/zig/o/9fdb2db183d32c7be8cdd99605e4091c/libubsan_rt.a.o)

[HomuHomu833]

maybe we need make seperate issue of this?

[alexrp]

That is the Hexagon issue I was referring to: llvm/llvm-project#111217

[HomuHomu833]

That is the Hexagon issue I was referring to: llvm/llvm-project#111217

oh okay didn't saw that pull request

[alexrp]

The fix for this is in LLVM 20.1.5.