Upgrade LLVM to 7ac7d418ac2b16fd44789dcf48e2b5d73de3e715 and StableHLO to 2a0595b

[hamptonm1]

For the LLVM and StableHLO upgrade, I checked out commit number 7ac7d418ac2b16fd44789dcf48e2b5d73de3e715 and 2a0595b. One small update for LLVM is found here: llvm/llvm-project#85393.

Changing omp::WsLoopOp -> omp::WsloopOp

I also changed the pass in Sequence_with_dealloc.mlir from --buffer-deallocation-pipeline to --buffer-deallocation based on the following error I received:

+ /Users/meganhampton/zDLC/llvm-project/build/bin/FileCheck /Users/meganhampton/zDLC/onnx-mlir/test/mlir/conversion/onnx_to_krnl/Sequence/Sequence_with_dealloc.mlir
LLVM ERROR: checking for an interface (`mlir::bufferization::BufferDeallocationOpInterface`) that was promised by dialect 'arith' but never implemented. This is generally an indication that the dialect extension implementing the interface was never registered.
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:

[hamptonm1]

@tungld Do you know why this lit test had a pass named --buffer-deallocation-pipeline when I only see --buffer-deallocation pass defined... so I changed it back. Let me know your thoughts please if there is any other changes I should make. Thanks!

[tungld]

I guess this test is to check if memref.dealloc is inserted correctly or not. When you changed it back, I saw that memref.dealloc was missing.

I think --buffer-deallocation is the old mechanism, so we moved to the new one --buffer-deallocation-pipeline from the PR #2568.
It's not recommended to go back to the old mechanism since it will finally be removed in LLVM.

When I blamed this file, @chentong319 is the original author introducing this test. @chentong319 may have a better explanation.

[chentong319]

The sequence bufferization is a special case. You can do anything (including deleting this test) to make your PR pass. I will come back to fix it when onnx-mlir has moved to the new llvm level.

[hamptonm1]

@jenkins-droid test this please

[hamptonm1]

@jenkins-droid test this please

[brnorris03]

Thank you for working on this update!

I noticed that StableHLO just updated to 1e6ce5e284f5, which is from yesterday -- it didn't seem like there were a lot of changes, maybe worth a try?

[brnorris03]

I am not sure if this is related, but buffer-deallocation-pipeline is supposed to be used after one-shot-bufferize.

[hamptonm1]

Thank you for working on this update!

I noticed that StableHLO just updated to 1e6ce5e284f5, which is from yesterday -- it didn't seem like there were a lot of changes, maybe worth a try?

Okay I can try and see how it goes! :)

[chentong319]

LGTM!

[jenkins-droid]

Jenkins Linux s390x Build #14700 [push] Upgrade LLVM to 7ac7d418... started at 11:12

[jenkins-droid]

Jenkins Linux ppc64le Build #13695 [push] Upgrade LLVM to 7ac7d418... started at 11:22

[jenkins-droid]

Jenkins Linux amd64 Build #14670 [push] Upgrade LLVM to 7ac7d418... started at 10:12

[jenkins-droid]

Jenkins Linux amd64 Build #14670 [push] Upgrade LLVM to 7ac7d418... passed after 2 hr 9 min

[jenkins-droid]

Jenkins Linux s390x Build #14700 [push] Upgrade LLVM to 7ac7d418... passed after 2 hr 16 min

[jenkins-droid]

Jenkins Linux ppc64le Build #13695 [push] Upgrade LLVM to 7ac7d418... passed after 3 hr 3 min