translate grad_add to elementwise_add (#56180)

paddle/fluid/ir_adaptor/translator/op_translator.cc

@@ -1379,6 +1379,19 @@ struct ElementwiseTranscriber : public OpTranscriber {
   }
 };
 
+struct GradAddOpTranscriber : public ElementwiseTranscriber {
+  ir::OpInfo LoopkUpOpInfo(ir::IrContext* ctx, const OpDesc& op_desc) override {
+    const std::string& target_op_name = "pd.add";
+    const auto& op_info = ctx->GetRegisteredOpInfo(target_op_name);
+    if (!op_info) {
+      IR_THROW(
+          "Op assign_value should have corresponding OpInfo pd.assign_value_");
+    }
+
+    return op_info;
+  }
+};
+
 struct ElementwiseGradTranscriber : public OpTranscriber {
   void RecordOpResultMapping(ir::IrContext* ctx,
                              TranslationContext* param_map,
@@ -1450,6 +1463,7 @@ OpTranslator::OpTranslator() {
   special_handlers["feed"] = FeedOpTranscriber();
   special_handlers["data"] = DataOpTranscriber();
   special_handlers["fetch_v2"] = FetchOpTranscriber();
+  special_handlers["grad_add"] = GradAddOpTranscriber();
   special_handlers["increment"] = IncrementOpTranscriber();
   special_handlers["lookup_table_v2"] = EmbeddingOpTranscriber();
   special_handlers["lookup_table_v2_grad"] = EmbeddingGradOpTranscriber();

paddle/phi/api/yaml/op_compat.yaml

@@ -1230,6 +1230,10 @@
     {pre_nms_top_n : pre_nms_topN, post_nms_top_n : post_nms_topN}
 
 - op : grad_add
+  inputs :
+    {x : X, y : Y}
+  outputs :
+    {out : Out}
   extra :
     attrs : [bool use_mkldnn = false, str mkldnn_data_type = "float32",
              bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]

test/ir/new_ir/test_special_op_translator.py

@@ -274,5 +274,31 @@ class TestIndexPutOpTranscriber(unittest.TestCase):
         _ = ir.translate_to_new_ir(main_program.desc)
 
 
+class TestGradAddOpTranscriber(unittest.TestCase):
+    def test_op(self):
+        place = core.Place()
+        place.set_place(paddle.CPUPlace())
+        new_scope = paddle.static.Scope()
+        main_program = paddle.static.Program()
+        with paddle.static.scope_guard(new_scope):
+            with paddle.static.program_guard(main_program):
+                x_data = np.random.rand(100, 2, 3)
+                y_data = np.random.rand(100, 1, 1)
+                x = paddle.to_tensor(x_data, dtype='float32')
+                x.stop_gradient = False
+                y = paddle.to_tensor(y_data, dtype='float32')
+
+                helper = LayerHelper('grad_add')
+                out = helper.create_variable_for_type_inference("float")
+                helper.append_op(
+                    type="grad_add",
+                    inputs={"X": x, "Y": y},
+                    outputs={"Out": out},
+                    attrs={"axis": -1},
+                )
+
+        _ = ir.translate_to_new_ir(main_program.desc)
+
+
 if __name__ == "__main__":
     unittest.main()