[Mlir-commits] [mlir] [mlir][MLProgram] Add MLProgram to MemRef bufferization pass (PR #75103)

[Matthias Springer]

matthias-springer wrote:

One-Shot Bufferize skips over ops that do not have tensor semantics. Until now we defined "tensor semantics" as "has tensor operand and/or result" (and some special rules for functions). `ml_program.global` does not have any operands or results, so it was skipped.

#75273 should fix the issue: I added a `hasTensorSemantics` interface method to `BufferizableOpInterface`. The default implementation behaves the same as before. But you can override the behavior for `ml_program.global` and then it should be picked up by the bufferization.


https://github.com/llvm/llvm-project/pull/75103