| 1 | //===- DialectLinalg.cpp - Pybind module for Linalg dialect API support --===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #include "mlir-c/Dialect/Linalg.h" |
| 10 | #include "mlir-c/IR.h" |
| 11 | #include "mlir/Bindings/Python/Nanobind.h" |
| 12 | #include "mlir/Bindings/Python/NanobindAdaptors.h" |
| 13 | |
| 14 | namespace nb = nanobind; |
| 15 | using namespace mlir::python::nanobind_adaptors; |
| 16 | |
| 17 | static std::optional<MlirLinalgContractionDimensions> |
| 18 | InferContractionDimensions(MlirOperation op) { |
| 19 | MlirLinalgContractionDimensions dims = |
| 20 | mlirLinalgInferContractionDimensions(op); |
| 21 | |
| 22 | // Detect "empty" result. This occurs when `op` is not a contraction op, |
| 23 | // or when `linalg::inferContractionDims` fails. |
| 24 | if (mlirAttributeIsNull(dims.batch) && mlirAttributeIsNull(dims.m) && |
| 25 | mlirAttributeIsNull(dims.n) && mlirAttributeIsNull(dims.k)) { |
| 26 | return std::nullopt; |
| 27 | } |
| 28 | return dims; |
| 29 | } |
| 30 | |
| 31 | static std::optional<MlirLinalgConvolutionDimensions> |
| 32 | InferConvolutionDimensions(MlirOperation op) { |
| 33 | MlirLinalgConvolutionDimensions dims = |
| 34 | mlirLinalgInferConvolutionDimensions(op); |
| 35 | |
| 36 | // Detect "empty" result. This occurs when `op` is not a convolution op, |
| 37 | // or when `linalg::inferConvolutionDims` fails. |
| 38 | if (mlirAttributeIsNull(dims.batch) && |
| 39 | mlirAttributeIsNull(dims.outputImage) && |
| 40 | mlirAttributeIsNull(dims.outputChannel) && |
| 41 | mlirAttributeIsNull(dims.filterLoop) && |
| 42 | mlirAttributeIsNull(dims.inputChannel) && |
| 43 | mlirAttributeIsNull(dims.depth) && mlirAttributeIsNull(dims.strides) && |
| 44 | mlirAttributeIsNull(dims.dilations)) { |
| 45 | return std::nullopt; |
| 46 | } |
| 47 | |
| 48 | return dims; |
| 49 | } |
| 50 | |
| 51 | static void populateDialectLinalgSubmodule(nb::module_ m) { |
| 52 | m.def( |
| 53 | "fill_builtin_region" , |
| 54 | [](MlirOperation op) { mlirLinalgFillBuiltinNamedOpRegion(op); }, |
| 55 | nb::arg("op" ), |
| 56 | "Fill the region for `op`, which is assumed to be a builtin named Linalg " |
| 57 | "op." ); |
| 58 | |
| 59 | m.def("isa_contraction_op" , &mlirLinalgIsAContractionOp, |
| 60 | "Checks if the given operation is a Linalg contraction operation." , |
| 61 | nb::arg("op" )); |
| 62 | |
| 63 | nb::class_<MlirLinalgContractionDimensions>(m, "ContractionDimensions" ) |
| 64 | .def_prop_ro("batch" , |
| 65 | [](const MlirLinalgContractionDimensions &self) { |
| 66 | return self.batch; |
| 67 | }) |
| 68 | .def_prop_ro( |
| 69 | "m" , |
| 70 | [](const MlirLinalgContractionDimensions &self) { return self.m; }) |
| 71 | .def_prop_ro( |
| 72 | "n" , |
| 73 | [](const MlirLinalgContractionDimensions &self) { return self.n; }) |
| 74 | .def_prop_ro("k" , [](const MlirLinalgContractionDimensions &self) { |
| 75 | return self.k; |
| 76 | }); |
| 77 | |
| 78 | m.def("infer_contraction_dimensions" , &InferContractionDimensions, |
| 79 | "Infers contraction dimensions (batch/m/n/k) for a Linalg contraction " |
| 80 | "op." , |
| 81 | nb::arg("op" )); |
| 82 | |
| 83 | m.def("isa_convolution_op" , &mlirLinalgIsAConvolutionOp, |
| 84 | "Checks if the given operation is a Linalg convolution operation." , |
| 85 | nb::arg("op" )); |
| 86 | |
| 87 | nb::class_<MlirLinalgConvolutionDimensions>(m, "ConvolutionDimensions" ) |
| 88 | .def_prop_ro("batch" , |
| 89 | [](const MlirLinalgConvolutionDimensions &self) { |
| 90 | return self.batch; |
| 91 | }) |
| 92 | .def_prop_ro("output_image" , |
| 93 | [](const MlirLinalgConvolutionDimensions &self) { |
| 94 | return self.outputImage; |
| 95 | }) |
| 96 | .def_prop_ro("output_channel" , |
| 97 | [](const MlirLinalgConvolutionDimensions &self) { |
| 98 | return self.outputChannel; |
| 99 | }) |
| 100 | .def_prop_ro("filter_loop" , |
| 101 | [](const MlirLinalgConvolutionDimensions &self) { |
| 102 | return self.filterLoop; |
| 103 | }) |
| 104 | .def_prop_ro("input_channel" , |
| 105 | [](const MlirLinalgConvolutionDimensions &self) { |
| 106 | return self.inputChannel; |
| 107 | }) |
| 108 | .def_prop_ro("depth" , |
| 109 | [](const MlirLinalgConvolutionDimensions &self) { |
| 110 | return self.depth; |
| 111 | }) |
| 112 | .def_prop_ro("strides" , |
| 113 | [](const MlirLinalgConvolutionDimensions &self) { |
| 114 | return self.strides; |
| 115 | }) |
| 116 | .def_prop_ro("dilations" , |
| 117 | [](const MlirLinalgConvolutionDimensions &self) { |
| 118 | return self.dilations; |
| 119 | }); |
| 120 | |
| 121 | m.def("infer_convolution_dimensions" , &InferConvolutionDimensions, |
| 122 | "Infers convolution dimensions" , nb::arg("op" )); |
| 123 | |
| 124 | m.def( |
| 125 | "get_indexing_maps" , |
| 126 | [](MlirOperation op) -> std::optional<MlirAttribute> { |
| 127 | MlirAttribute attr = mlirLinalgGetIndexingMapsAttribute(op); |
| 128 | if (mlirAttributeIsNull(attr)) |
| 129 | return std::nullopt; |
| 130 | return attr; |
| 131 | }, |
| 132 | "Returns the indexing_maps attribute for a linalg op." ); |
| 133 | } |
| 134 | |
| 135 | NB_MODULE(_mlirDialectsLinalg, m) { |
| 136 | m.doc() = "MLIR Linalg dialect." ; |
| 137 | |
| 138 | populateDialectLinalgSubmodule(m); |
| 139 | } |
| 140 | |