Use upstream LLVM location

[pbchekin]

There is LLVM for CentOS in upstream again.
Fixes #2738.

[anmyachev]

@pbchekin should we update cache number in https://github.com/intel/intel-xpu-backend-for-triton/blame/229f7a1c10131a6736860e9f985e19c6dc57390f/.github/actions/setup-triton/action.yml#L27?

I noticed that locally in the resulted LLVM IR there is !llvm.dbg.cu = !{!4} instruction, but in LLVM IR from the CI it is not. My best guess is that we have slightly different builds, but I'm not 100% sure.

[pbchekin]

@pbchekin should we update cache number in https://github.com/intel/intel-xpu-backend-for-triton/blame/229f7a1c10131a6736860e9f985e19c6dc57390f/.github/actions/setup-triton/action.yml#L27?

I noticed that locally in the resulted LLVM IR there is !llvm.dbg.cu = !{!4} instruction, but in LLVM IR from the CI it is not. My best guess is that we have slightly different builds, but I'm not 100% sure.

This cache is used only when building LLVM mode is enabled (if: ${{ inputs.build_llvm == 'true' }}). It is disabled by default and we download the LLVM package every time we build Triton. I think the issue somewhere else.