remove old unused stuff

2026-03-23 20:00:09 +01:00
parent f2d593f749
commit f869925b64
12 changed files with 16 additions and 906 deletions
@@ -1,89 +0,0 @@
-#include "mlir/Transforms/DialectConversion.h"
-
-#include "Conversion/ONNXToSpatial/Patterns.hpp"
-#include "src/Dialect/ONNX/ONNXOps.hpp"
-
-using namespace mlir;
-namespace onnx_mlir {
-
-struct ReduceMeanConversionPattern : public OpConversionPattern<ONNXReduceMeanV13Op> {
-
-  ReduceMeanConversionPattern(MLIRContext* ctx)
-  : OpConversionPattern(ctx) {}
-
-  LogicalResult matchAndRewrite(ONNXReduceMeanV13Op reduceMean,
-                                ONNXReduceMeanV13OpAdaptor adaptor,
-                                ConversionPatternRewriter& rewriter) const final {
-
-    // Get the input tensor.
-    Value inputTensor = adaptor.getData();
-    auto inputTensorType = cast<RankedTensorType>(inputTensor.getType());
-
-    // This pattern will substitute the ONNXReduceMeanV13Op with a
-    // ONNXAveragePoolOp with the same input tensor and an appropriate kernel
-    // shape and strides.
-
-    // To get the stride and shape of the kernel, we need to read the tensor
-    // shape.
-    int image_height = inputTensorType.getShape()[2];
-    int image_width = inputTensorType.getShape()[3];
-
-    // Define the kernel shape and strides.
-    SmallVector<int64_t> kernelShapeVals = {image_height, image_width};
-    SmallVector<int64_t> stridesVals = {image_height, image_width};
-    SmallVector<int64_t> dilationsVals = {1, 1};
-
-    // Set the pads to 0.
-    SmallVector<int64_t> padsVals = {0, 0, 0, 0};
-
-    // Create the ArrayAttrs
-    auto kernelShape = mlir::ArrayAttr::get(
-      rewriter.getContext(), llvm::to_vector(llvm::map_range(kernelShapeVals, [&](int64_t v) -> mlir::Attribute {
-        return rewriter.getI64IntegerAttr(v);
-      })));
-
-    auto strides = mlir::ArrayAttr::get(rewriter.getContext(),
-                                        llvm::to_vector(llvm::map_range(stridesVals, [&](int64_t v) -> mlir::Attribute {
-                                          return rewriter.getI64IntegerAttr(v);
-                                        })));
-
-    auto dilations = mlir::ArrayAttr::get(
-      rewriter.getContext(), llvm::to_vector(llvm::map_range(dilationsVals, [&](int64_t v) -> mlir::Attribute {
-        return rewriter.getI64IntegerAttr(v);
-      })));
-
-    auto pads = mlir::ArrayAttr::get(rewriter.getContext(),
-                                     llvm::to_vector(llvm::map_range(padsVals, [&](int64_t v) -> mlir::Attribute {
-                                       return rewriter.getI64IntegerAttr(v);
-                                     })));
-
-    // Create the resulting tensor type.
-    auto resultType = RankedTensorType::get(
-      /*shape=*/ {inputTensorType.getShape()[0], inputTensorType.getShape()[1], 1, 1},
-      /*elementType=*/inputTensorType.getElementType());
-
-    // Create the ONNXAveragePoolOp.
-    auto averagePool = ONNXAveragePoolOp::create(rewriter,
-                                                 reduceMean.getLoc(),
-                                                 resultType,
-                                                 inputTensor,
-                                                 /*auto_pad=*/"NOTSET",
-                                                 /*ceil_mode=*/0,
-                                                 /*count_include_pad=*/1,
-                                                 dilations,
-                                                 /*kernel_shape=*/kernelShape,
-                                                 /*pads=*/pads,
-                                                 /*strides=*/strides);
-
-    // Replace the ONNXReduceMeanV13Op with the ONNXAveragePoolOp.
-    rewriter.replaceOp(reduceMean, averagePool.getResult());
-
-    return success();
-  }
-};
-
-void populateReduceMeanConversionPattern(RewritePatternSet& patterns, MLIRContext* ctx) {
-  patterns.insert<ReduceMeanConversionPattern>(ctx);
-}
-
-} // namespace onnx_mlir