[BUAA]【Infer Symbolic Shape No.10】Add box_clip op

paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/binary_infer_sym.cc

@@ -91,6 +91,33 @@ bool AllcloseOpInferSymbolicShape(
   return true;
 }
 
+bool BoxClipOpInferSymbolicShape(
+    pir::Operation *op, pir::InferSymbolicShapeContext *infer_context) {
+  const auto &input_shape =
+      infer_context->GetShapeOrDataForValue(op->operand_source(0)).shape();
+  const auto &im_info_shape =
+      infer_context->GetShapeOrDataForValue(op->operand_source(1)).shape();
+
+  // Check rank and dimensions of input tensors
+  const auto &three = symbol::DimExpr{3};
+  const auto &four = symbol::DimExpr{4};
+  infer_context->AddEqualCstr(input_shape[input_shape.size() - 1], four);
+  PADDLE_ENFORCE_EQ(im_info_shape.size(),
+                    2,
+                    common::errors::InvalidArgument(
+                        "The rank of Input(im_info) in BoxClipOp must be 2. "
+                        "But received rank = %d",
+                        im_info_shape.size()));
+  infer_context->AddEqualCstr(im_info_shape[1], three);
+
+  infer_context->SetShapeOrDataForValue(
+      op->result(0),
+      symbol::ShapeOrDataDimExprs{
+          symbol::TensorShapeOrDataDimExprs(input_shape)});
+
+  return true;
+}
+
 bool Atan2OpInferSymbolicShape(pir::Operation *op,
                                pir::InferSymbolicShapeContext *infer_context) {
   const auto x_shape =

paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/binary_infer_sym.h

@@ -22,6 +22,7 @@ OP_DECLARE_INFER_SYMBOLIC_SHAPE(Allclose)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(Atan2)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(BceLoss)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(BceLoss_)
+OP_DECLARE_INFER_SYMBOLIC_SHAPE(BoxClip)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(Binomial)
 OP_DECLARE_INFER_SYMBOLIC_SHAPE(Binomial_)
 // OP_DECLARE_INFER_SYMBOLIC_SHAPE(Bincount)

paddle/phi/ops/yaml/ops.yaml

@@ -652,6 +652,7 @@
     func: BoxClipInferMeta
   kernel:
     func: box_clip
+  interfaces: paddle::dialect::InferSymbolicShapeInterface
 
 - op : box_coder
   args : (Tensor prior_box, Tensor prior_box_var, Tensor target_box, str code_type = "encode_center_size", bool box_normalized = true, int axis = 0, float[] variance = {})