[ISSUE] Mesa Clover OpenCL crashes on llvm::TargetInstrInfo::ReplaceTailWithBranchTo()

[Mek101 <Mek101@users.noreply.github.com>]

[-- Attachment #1: Type: text/plain, Size: 1302 bytes --]

New issue by Mek101 on void-packages repository

https://github.com/void-linux/void-packages/issues/43921

Description:
### Is this a new report?

Yes

### System Info

Void 6.2.15_1 x86_64 AuthenticAMD notuptodate rrFFFF

### Package(s) Affected

mesa-22.3.5_2 mesa-opencl-22.3.5_2 libllvm15-15.0.7_1

### Does a report exist for this bug with the project's home (upstream) and/or another distro?

_No response_

### Expected behaviour

The system should run the Tensorflow model with OpenCL 1.1

### Actual behaviour

Running some models causes the OpenCL implementation (Mesa's Clover) to call `llvm::TargetInstrInfo::ReplaceTailWithBranchTo()` from `/usr/lib/libLLVM-15.so`, which itself is provided by the `libllvm15` package. This function call causes a segmentation fault (verified via `gdb`).

### Steps to reproduce

- Ensure Mesa uses Clover as an OpenCL implementation
- Donwload an affected TensorFlow Lite model (at least `squeezenet` in my case)
- Compile TensorFlow Lite's benchmark with GPU support.
- Run it with the `--use_gpu=true --gpu_backend="cl"` options.
- Observe the segmentation fault

In my particular case I'm using an old Radeon 8670D, a Terascale 3 iGPU with the `radeon` kernel driver.

Squeezenet: https://tfhub.dev/tensorflow/lite-model/squeezenet/1/default/1