Apply scale should be optionally disabled when lowering via TosaToStandard.
In most cases it should persist until the lowering to specific backend.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
mlir/include/mlir/Conversion/Passes.td | ||
---|---|---|
758 | s/standard/arith/? Or are there other ones along? (I was mentioning to Mehdi I don't know how to rename the ShapeToStandard pass with Standard exploding :)) [you can also defer if this is multiple dialects being lowered to] | |
mlir/lib/Conversion/TosaToStandard/TosaToStandardPass.cpp | ||
41 | Could we add a comment somewhere in code (perhaps in pass description so that it makes it into the generated docs) as to the why. | |
mlir/test/Conversion/TosaToStandard/tosa-to-standard.mlir | ||
1 | Test with it false? (don't have to repeat all CHECK's, just a spot check) |
mlir/include/mlir/Conversion/Passes.td | ||
---|---|---|
714 | Up to you, but I'd just drop ops and have everything on one line (or drop tosa. as this is a tosa-to-arith pass so the dialect is implicit for me) | |
mlir/include/mlir/Conversion/TosaToArith/TosaToArith.h | ||
21 ↗ | (On Diff #419908) | This and addTosaToArithPasses seem to overlap quite a bit, when would one use one vs the other? |
mlir/include/mlir/Conversion/TosaToTensor/TosaToTensor.h | ||
1 ↗ | (On Diff #419908) | optimization pass? TOSA to Tensor legalization pass ? |
mlir/lib/Conversion/TosaToStandard/TosaToStandardPass.cpp | ||
1 | Tensor | |
mlir/lib/Conversion/TosaToTensor/TosaToTensor.cpp | ||
1 ↗ | (On Diff #419908) | Tensor |
9 ↗ | (On Diff #419908) | Tensor |
mlir/test/Conversion/TosaToTensor/tosa-to-tensor.mlir | ||
1 ↗ | (On Diff #419908) | verify-diagnostics doesn't seem needed here |
Up to you, but I'd just drop ops and have everything on one line (or drop tosa. as this is a tosa-to-arith pass so the dialect is implicit for me)