[Paddle-TRT] Shape sum fix scale (#44394)

* shape sum

* add shape, sum trt layer

paddle/fluid/inference/api/analysis_predictor.cc

@@ -2080,6 +2080,8 @@ USE_TRT_CONVERTER(top_k)
 USE_TRT_CONVERTER(top_k_v2)
 USE_TRT_CONVERTER(squeeze2)
 USE_TRT_CONVERTER(unsqueeze2)
+USE_TRT_CONVERTER(sum)
+USE_TRT_CONVERTER(shape)
 USE_TRT_CONVERTER(fill_constant)
 USE_TRT_CONVERTER(fused_token_prune)
 #if PADDLE_WITH_CUSPARSELT && IS_TRT_VERSION_GE(8000)

paddle/fluid/inference/tensorrt/convert/CMakeLists.txt

@@ -69,6 +69,8 @@ list(
   top_k_op.cc
   squeeze2_op.cc
   unsqueeze2_op.cc
+  sum_op.cc
+  shape_op.cc
   fill_constant_op.cc
   fused_token_prune_op.cc)
 

paddle/fluid/inference/tensorrt/convert/shape_op.cc

@@ -0,0 +1,41 @@
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#include "paddle/fluid/inference/tensorrt/convert/op_converter.h"
+
+namespace paddle {
+namespace inference {
+namespace tensorrt {
+
+class ShapeOpConverter : public OpConverter {
+ public:
+  void operator()(const framework::proto::OpDesc& op,
+                  const framework::Scope& scope,
+                  bool test_mode) override {
+    VLOG(4) << "convert a fluid shape op to tensorrt shape layer";
+
+    framework::OpDesc op_desc(op, nullptr);
+    // Declare inputs
+    auto* input = engine_->GetITensor(op_desc.Input("Input")[0]);
+    nvinfer1::ILayer* layer = TRT_ENGINE_ADD_LAYER(engine_, Shape, *input);
+    auto output_name = op_desc.Output("Out")[0];
+    RreplenishLayerAndOutput(layer, "shape", {output_name}, test_mode);
+  }
+};
+
+}  // namespace tensorrt
+}  // namespace inference
+}  // namespace paddle
+
+REGISTER_TRT_OP_CONVERTER(shape, ShapeOpConverter);

paddle/fluid/inference/tensorrt/convert/sum_op.cc

@@ -0,0 +1,54 @@
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#include "paddle/fluid/inference/tensorrt/convert/op_converter.h"
+
+namespace paddle {
+namespace inference {
+namespace tensorrt {
+
+class SumOpConverter : public OpConverter {
+ public:
+  void operator()(const framework::proto::OpDesc& op,
+                  const framework::Scope& scope,
+                  bool test_mode) override {
+    VLOG(4) << "convert a fluid sum op to tensorrt sum layer";
+
+    framework::OpDesc op_desc(op, nullptr);
+    nvinfer1::ILayer* layer = nullptr;
+    // Declare the first input
+    auto* sum_tmp = engine_->GetITensor(op_desc.Input("X")[0]);
+    if (op_desc.Input("X").size() == 1) {
+      layer = TRT_ENGINE_ADD_LAYER(engine_, Identity, *sum_tmp);
+    } else {
+      for (size_t i = 1; i < op_desc.Input("X").size(); i++) {
+        auto* input_i = engine_->GetITensor(op_desc.Input("X")[i]);
+        layer = TRT_ENGINE_ADD_LAYER(engine_,
+                                     ElementWise,
+                                     *input_i,
+                                     *sum_tmp,
+                                     nvinfer1::ElementWiseOperation::kSUM);
+        sum_tmp = layer->getOutput(0);
+      }
+    }
+    auto output_name = op_desc.Output("Out")[0];
+    RreplenishLayerAndOutput(layer, "sum", {output_name}, test_mode);
+  }
+};
+
+}  // namespace tensorrt
+}  // namespace inference
+}  // namespace paddle
+
+REGISTER_TRT_OP_CONVERTER(sum, SumOpConverter);

paddle/fluid/inference/tensorrt/op_teller.cc

@@ -170,6 +170,8 @@ struct SimpleOpTypeSetTeller : public Teller {
       "recover_padding",
       "remove_padding",
       "fill_constant",
+      "sum",
+      "shape",
       "squeeze2",
       "unsqueeze2"};
   std::unordered_set<std::string> teller_set{
@@ -276,6 +278,8 @@ struct SimpleOpTypeSetTeller : public Teller {
       "recover_padding",
       "remove_padding",
       "fill_constant",
+      "sum",
+      "shape",
       "squeeze2",
       "unsqueeze2",
       "fused_token_prune"};
@@ -1208,6 +1212,11 @@ bool OpTeller::Tell(const framework::ir::Node* node,
       auto x_var_name = desc.Input("X")[0];
       auto* x_var_desc = block->FindVar(x_var_name);
       const auto x_shape = x_var_desc->GetShape();
+      auto dtype = x_var_desc->GetDataType();
+      // At present, only support float32 or float16 into trt.
+      if (!(dtype == 5 || dtype == 4)) {
+        return false;
+      }
       if (!with_dynamic_shape && x_shape.size() == 1) {
         VLOG(3) << "Scale op does not support 1-dimensional input in tensorrt";
         return false;
@@ -1361,6 +1370,14 @@ bool OpTeller::Tell(const framework::ir::Node* node,
         return false;
       }
     }
+    // remember that 1D input in static shape mode is filtered at the beginning
+    if (op_type == "sum") {
+      return true;
+    }
+
+    if (op_type == "shape" && !with_dynamic_shape) {
+      return false;
+    }
 
     if (op_type == "fused_embedding_eltwise_layernorm") {
       if (!with_dynamic_shape) {

python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_shape.py

@@ -0,0 +1,120 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from trt_layer_auto_scan_test import TrtLayerAutoScanTest, SkipReasons
+from program_config import TensorConfig, ProgramConfig
+import numpy as np
+import paddle.inference as paddle_infer
+from functools import partial
+from typing import Optional, List, Callable, Dict, Any, Set
+import unittest
+
+
+class TrtConvertSumTest(TrtLayerAutoScanTest):
+
+    def is_program_valid(self, program_config: ProgramConfig) -> bool:
+        return True
+
+    def sample_program_configs(self):
+
+        def generate_input1(batch):
+            if self.dims == 4:
+                return np.ones([batch, 3, 24, 24]).astype(np.float32)
+            elif self.dims == 3:
+                return np.ones([batch, 3, 24]).astype(np.float32)
+            elif self.dims == 2:
+                return np.ones([batch, 24]).astype(np.float32)
+            elif self.dims == 1:
+                return np.ones([24]).astype(np.float32)
+
+        for dims in [1, 2, 3, 4]:
+            for batch in [1, 4]:
+                self.dims = dims
+                ops_config = [{
+                    "op_type": "shape",
+                    "op_inputs": {
+                        "Input": ["input1"]
+                    },
+                    "op_outputs": {
+                        "Out": ["output"]
+                    },
+                    "op_attrs": {}
+                }]
+                ops = self.generate_op_config(ops_config)
+                program_config = ProgramConfig(
+                    ops=ops,
+                    weights={},
+                    inputs={
+                        "input1":
+                        TensorConfig(data_gen=partial(generate_input1, batch))
+                    },
+                    outputs=["output"])
+
+                yield program_config
+
+    def sample_predictor_configs(
+            self, program_config) -> (paddle_infer.Config, List[int], float):
+
+        def generate_dynamic_shape():
+            if self.dims == 4:
+                self.dynamic_shape.min_input_shape = {"input1": [1, 3, 24, 24]}
+                self.dynamic_shape.max_input_shape = {"input1": [4, 3, 48, 48]}
+                self.dynamic_shape.opt_input_shape = {"input1": [1, 3, 24, 24]}
+            elif self.dims == 3:
+                self.dynamic_shape.min_input_shape = {"input1": [1, 3, 24]}
+                self.dynamic_shape.max_input_shape = {"input1": [4, 3, 48]}
+                self.dynamic_shape.opt_input_shape = {"input1": [1, 3, 24]}
+            elif self.dims == 2:
+                self.dynamic_shape.min_input_shape = {"input1": [1, 24]}
+                self.dynamic_shape.max_input_shape = {"input1": [4, 48]}
+                self.dynamic_shape.opt_input_shape = {"input1": [1, 24]}
+            elif self.dims == 1:
+                self.dynamic_shape.min_input_shape = {"input1": [24]}
+                self.dynamic_shape.max_input_shape = {"input1": [48]}
+                self.dynamic_shape.opt_input_shape = {
+                    "input1": [24],
+                }
+
+        def generate_trt_nodes_num(dynamic_shape):
+            if (not dynamic_shape):
+                return 0, 3
+            return 1, 2
+
+        def clear_dynamic_shape():
+            self.dynamic_shape.min_input_shape = {}
+            self.dynamic_shape.max_input_shape = {}
+            self.dynamic_shape.opt_input_shape = {}
+
+        # for static_shape
+        clear_dynamic_shape()
+        self.trt_param.precision = paddle_infer.PrecisionType.Float32
+        yield self.create_inference_config(), generate_trt_nodes_num(
+            False), 1e-5
+        self.trt_param.precision = paddle_infer.PrecisionType.Half
+        yield self.create_inference_config(), generate_trt_nodes_num(
+            False), 1e-5
+
+        # for dynamic_shape
+        generate_dynamic_shape()
+        self.trt_param.precision = paddle_infer.PrecisionType.Float32
+        yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5
+        self.trt_param.precision = paddle_infer.PrecisionType.Half
+        yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5
+
+    def test(self):
+        self.run_test()
+
+
+if __name__ == "__main__":
+    unittest.main()