Albert Sandru
Albert Sandru
@LucQueen I think I've finished the decomposition, or at least come very close to finishing, however the test case still fails.
Yeah, this is my solution so far: https://github.com/llvm/torch-mlir/pull/949/files
Yeah, AtenPowOp is what's needed, however I don't see it emitted anywhere.
The error I've been getting is that when I use floats the error calls for ints and when I use ints it calls for floats.
Just to confirm, the cast would happen with AtenToDtypeOp?
I'm also assuming this new approach would mean reverting the changes I made to TorchOps.cpp.
Alright, I've created a separate PR for aten.div.int, please let me know if it looks good. Since this op depends on aten.div.int, I first want to get aten.div.int merged in,...
``` torch_mlir.compiler_utils.TorchMlirCompilerError: Lower Torch Backend IR -> Linalg-on-Tensors Backend IR failed with the following diagnostics: error: 'torch_c.from_i64' op operand #0 must be 64-bit signless integer, but got 'i32' note: see...