[Backend] Bump to llvm/llvm-project@7ec6f1771114

[antiagainst]

• Dropped -Os and -Oz after [Passes] Remove Os and Oz optimization pipelines llvm/llvm-project#191363
• Emit scalar {add,sub,mul}.rn.bf16 ops after [SLP] Reject 2-element vectorization when vector inst count exceeds scalar llvm/llvm-project#190414
• Also changed to avoid hardcoding line numbers in test_line_info.py
• Add identifier/discriminator arguments to DICompositeTypeAttr::get
  calls after [MLIR][LLVM] Add variant part debug info support llvm/llvm-project#195321
• Pass MCRegisterInfo and MCSubtargetInfo by reference when constructing
  MCContext after MCContext/TargetMachine: Take MCRegisterInfo and MCSubtargetInfo by reference. NFC llvm/llvm-project#195032
• Replace NVVM::Barrier0Op and nvvm.barrier0 test expectations with
  NVVM::BarrierOp/nvvm.barrier after [MLIR][NVVM] Remove the nvvm.barrier0 op llvm/llvm-project#195608

[neildhar]

Sorry if the checksum validation is causing problems. I set up the LLVM build job to dump the SHA for each build right before uploading (e.g. https://github.com/triton-lang/triton/actions/runs/26669029128/job/78608320883#step:15:21). I found that pointing codex to that build job was pretty good at fetching and updating the correct checksums.

[antiagainst]

Sorry if the checksum validation is causing problems. I set up the LLVM build job to dump the SHA for each build right before uploading (e.g. https://github.com/triton-lang/triton/actions/runs/26669029128/job/78608320883#step:15:21). I found that pointing codex to that build job was pretty good at fetching and updating the correct checksums.

Thanks for the pointers! I intially thought it's sha256sum from https://github.com/triton-lang/triton/actions/runs/26669029128. Turns out indeed they are different. Would be convenient if we can aggregate and display at the summary page.