iree-compile segfaults on Pypi out of the box
What happened?
Hi, I followed the Getting started tutorial for CPU, installed the Pypi stable release of the project. with python -m pip install iree-compiler
I tried running iree-compile with various options nothing was happening so I looked a bit into it and found a segFault.
Please see steps below.
Steps to reproduce your issue
python -m pip install iree-compiler
iree-compile --iree-llvmcpu-list-targets
echo $?
245
iree compile calls : lib/python3.9/site-packages/iree/compiler/tools/scripts/ireec/main.py which calls this segFaulting binary. lib/python3.9/site-packages/iree/compiler/_mlir_libs/iree-compile
gdb lib/python3.9/site-packages/iree/compiler/tools/../_mlir_libs/iree-compile -batch -ex 'disassemble main'
Dump of assembler code for function main:
0x00000000002108a4 <+0>: b 0x2109a0 <ireeCompilerRunMain@plt>
End of assembler dump.
ireeCompilerRunMain appears in lib/python3.9/site-packages/iree/compiler/tools/../_mlir_libs/libIREECompiler.so
What component(s) does this issue relate to?
Compiler
Version information
--version fails as well. It is the Pypi version currently available.
Additional context
No response
This is some form of low level failure / basic incompatibility with the binaries we produce and something about your machine. We build our binaries on a rhel7 based system in a way similar to most packages on pypi. While this makes the binaries highly portable, there can still be issues.
Can you please provide full information about your system, minimally OS and exact processor info. Ideally we can get a docker image that shows the problem.
Hi, It is an AArch64 machine with EulerOS.
Versions and exact CPU, please?
I don't know anything about that combination, but EulerOS claims to be a RHEL derivative. Should work barring some basic incompatibility.
We build binaries with this docker image on aarch64: quay.io/pypa/manylinux_2_28_aarch64
EulerOS 2.0 (SP8). Armv8.3-a
Can you please verify on an exact docker image and provide the details of it? We will end up trying that on a cloud instance of some kind as we don't have any more access to arm servers than that.