Initial changes for llvm uplift #2568
Conversation
|
The final issues seem to be related to https://discourse.llvm.org/t/psa-bufferization-new-buffer-deallocation-pipeline/73375/11 I'm working through them now |
philass
force-pushed
the
plassen/llvm-update
branch
from
90be6e1 to
aac995d
Compare
There was a problem hiding this comment.
This is quite broken. This just disables the checks and offending ops
There was a problem hiding this comment.
@philass I understand your pain... the same goes for ONNX upgrade ( a lot of issues). Are we planning on just waiting for the next torch-mlir /llvm commit or are you going to work through this still?
I noticed torch-mlir has upgraded to a more recent llvm commit as well: llvm/torch-mlir#2517
There was a problem hiding this comment.
I will land this PR. Than we can figure out how the next bump should go
philass
force-pushed
the
plassen/llvm-update
branch
from
4f464a7 to
9daceb3
Compare
Signed-off-by: philass <[email protected]>
Signed-off-by: philass <[email protected]>
Signed-off-by: philass <[email protected]>
Signed-off-by: philass <[email protected]>
Signed-off-by: philass <[email protected]>
Signed-off-by: philass <[email protected]>
philass
force-pushed
the
plassen/llvm-update
branch
from
9daceb3 to
f152ff5
Compare
Signed-off-by: philass <[email protected]>
Signed-off-by: philass <[email protected]>
|
Hey @AlexandreEichenberger, @tungld, @chentong319 This should be ready for review. Only one lit test has a lingering issue. All other lit tests and end to end tests pass.
Selfishly I would like to either
The test can than be updated at a later point in time once it is fully triaged. Someone familiar with this part of the codebase could triage the issue. The LLVM dialect, memref, and bufferization dialect or not things I regularly work with. So I'm likely not the best person to trouble shoot this lit test. This would allow us to unblock the LLVM uplift which has already slipped (Really my fault there). |
|
@AlexandreEichenberger Thanks for the review
Will do |
|
Jenkins Linux ppc64le Build #12432 [push] Initial changes for llvm... started at 14:11 |
|
Jenkins Linux s390x Build #13438 [push] Initial changes for llvm... started at 14:02 |
|
Jenkins Linux amd64 Build #13411 [push] Initial changes for llvm... started at 13:02 |
|
Jenkins Linux s390x Build #13438 [push] Initial changes for llvm... failed after 31 min |
|
Jenkins Linux amd64 Build #13411 [push] Initial changes for llvm... failed after 1 hr 32 min |
|
Jenkins Linux ppc64le Build #12432 [push] Initial changes for llvm... failed after 2 hr 22 min |
* 'main' of github.ibm.com:zosdev/onnx-mlir: Use dim_params in dynamic dimension analysis (onnx#2620) Update rapidcheck to include the fix for missing <cstdint> include (onnx#2623) Initial changes for llvm uplift (onnx#2568) [build] Add cmake option to enable/disable Java components build (onnx#2613) [StableHLO] Lowers PadOp (constant mode) & GatherElements Op to StableHLO (onnx#2602) [DialectBuilder] add builder funcrions for ONNXSumOp and ONNXConvOp (onnx#2572) Import dim_param for model inputs and outputs (onnx#2616) Parse and set --mcpu in onnx-mlir-opt command (onnx#2614) Layernorm: convert instance norm and group norm to layer norm. (onnx#2595) [NNPA] Use F16 as element type for zTensor (onnx#2611) detect LayerNorm in presence of reciprocal and div of 1 (onnx#2609) # Conflicts: # test/mlir/conversion/onnx_to_krnl/NN/Normalization_O3_SIMD_canonicalize.mlir
The changes in this uplift are mainly related to following the migration guide in this discussion post https://discourse.llvm.org/t/psa-bufferization-new-buffer-deallocation-pipeline/73375/11