OneToN Dialect Conversion deprecated

[ZenithalHourlyRate]

The upstream MLIR one to N dialect conversion engine has been merged into the default dialect conversion engine, and all related methods for them have been marked as deprecated and will be deleted in April 2025.

Check https://discourse.llvm.org/t/rfc-merging-1-1-and-1-n-dialect-conversions/82513/10

Affected pass in HEIR: lib/Transforms/TensorToScalars/TensorToScalars.cpp

lib/Transforms/TensorToScalars/TensorToScalars.cpp:75:14: warning: 'replaceOp' is deprecated: Use replaceOpWithMultiple() instead [-Wdeprecated-declarations]
    rewriter.replaceOp(op, adaptor.getFlatOperands(),
             ^~~~~~~~~
             replaceOpWithMultiple
lib/Transforms/TensorToScalars/TensorToScalars.cpp:148:28: warning: 'applyPartialOneToNConversion' is deprecated: Use applyPartialConversion() instead [-Wdeprecated-declarations]
    if (mlir::failed(mlir::applyPartialOneToNConversion(
                           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
                           applyPartialConversion

Cc @AlexanderViand-Intel

[j2kun]

We can eagerly fix this, but also part of the benefit of Google ownership is that the Google-internal oncall rotation that will handle any breakages, or at worst disable failing tests, so long as the tests are exercising this section of code.

[AlexanderViand-Intel]

We can eagerly fix this, but also part of the benefit of Google ownership is that the Google-internal oncall rotation that will handle any breakages, or at worst disable failing tests, so long as the tests are exercising this section of code.

Good point! The TensorToScalars pass is only used at the end of the HEIR-to-HERACLES pipeline to remove tensor types and extract/inserts from polynomial/modarith-level IRs (since the HERACLES SDK doesn't have a concept of tensors of polynomials), so it's not exactly "critical" and I'd be happy to see how it's handled and only step in if needed :)