Feedback to MLIR

[raviqqe]

• Lack of API to check if dialects are loaded in contexts.
  • But this is possible operation-wise via mlirContextIsRegisteredOperation().
• String refs as inputs need to be null-terminated (C-style) strings.
  • The documentation says both are fine...
• Use after move of operations, regions, and blocks
• LLVM bitcode emission
  • This is possible only with a mlir-translate command right now.
• mlirMemRefTypeGetElementType()
  • Use the ShapedType API.

References

• Remove hacky StringRef cache #54
• Access to dialects not loaded leads to segmentation faults #23

[edg-l]

Access to the datalayout api https://mlir.llvm.org/docs/DataLayout/

[Danacus]

Access to rewrite patterns (https://mlir.llvm.org/docs/PatternRewriter/), but mainly to use PDL modules (https://mlir.llvm.org/doxygen/classmlir_1_1PDLPatternModule.html) as they can be represented as MLIR using the PDL dialect.

See also: https://github.com/beaver-lodge/beaver/blob/main/native/mlir-c/lib/CAPI/Elixir.cpp#L17

[edg-l]

I'm implementing the llvm debug info attributes in the C api: llvm/llvm-project#83992 to achieve nice debug info for mlir programs

[edg-l]

I'm implementing the llvm debug info attributes in the C api: llvm/llvm-project#83992 to achieve nice debug info for mlir programs

Got this merged!

Im doing another one llvm/llvm-project#84453