[Paddle Inference] Support range trt converter and add scalar interface. (#48697)

* add_range

* add_range

paddle/fluid/inference/api/analysis_predictor.cc

@@ -2329,6 +2329,7 @@ USE_TRT_CONVERTER(remove_padding)
 USE_TRT_CONVERTER(equal);
 USE_TRT_CONVERTER(top_k)
 USE_TRT_CONVERTER(top_k_v2)
+USE_TRT_CONVERTER(range)
 USE_TRT_CONVERTER(squeeze2)
 USE_TRT_CONVERTER(unsqueeze2)
 USE_TRT_CONVERTER(sum)

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

@@ -71,6 +71,7 @@ list(
   preln_residual_bias.cc
   c_allreduce_op.cc
   top_k_op.cc
+  range_op.cc
   squeeze2_op.cc
   unsqueeze2_op.cc
   rnn_op.cc

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

+/* Copyright (c) 2022 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 RangeOpConverter : public OpConverter {
+ public:
+  void operator()(const framework::proto::OpDesc& op,
+                  const framework::Scope& scope,
+                  bool test_mode) override {
+    VLOG(3) << "convert a range op to tensorrt layer";
+    framework::OpDesc op_desc(op, nullptr);
+    nvinfer1::ILayer* layer = nullptr;
+    nvinfer1::ITensor* quotient_tensor;
+
+    // Declare inputs
+    auto* start = engine_->GetITensor(op_desc.Input("Start")[0]);
+    auto* end = engine_->GetITensor(op_desc.Input("End")[0]);
+    auto* step = engine_->GetITensor(op_desc.Input("Step")[0]);
+    auto output_name = op_desc.Output("Out")[0];
+
+    auto zero_tensor = Add1DConstantLayer(0, output_name + "_zero_tensor_");
+    auto fquotient_tensor = FloorDiv(Sub(start, end), step);
+    if (start->getType() == nvinfer1::DataType::kFLOAT) {
+      auto* cast_int32_layer =
+          TRT_ENGINE_ADD_LAYER(engine_, Identity, *fquotient_tensor);
+      cast_int32_layer->setOutputType(0, nvinfer1::DataType::kINT32);
+      cast_int32_layer->getOutput(0)->setType(nvinfer1::DataType::kINT32);
+      quotient_tensor = cast_int32_layer->getOutput(0);
+    } else {
+      quotient_tensor = fquotient_tensor;
+    }
+    auto number_tensor = Max(Sub(zero_tensor, quotient_tensor), zero_tensor);
+    auto* start1 = engine_->GetITensor(op_desc.Input("Start")[0], true);
+
+    layer = TRT_ENGINE_ADD_LAYER(
+        engine_, Fill, nvinfer1::Dims{}, nvinfer1::FillOperation::kLINSPACE);
+    layer->setInput(0, *number_tensor);
+    layer->setInput(1, *start1);
+    layer->setInput(2, *step);
+
+    RreplenishLayerAndOutput(layer, "range", {output_name}, test_mode);
+  }
+};
+
+}  // namespace tensorrt
+}  // namespace inference
+}  // namespace paddle
+
+REGISTER_TRT_OP_CONVERTER(range, RangeOpConverter);

paddle/fluid/inference/tensorrt/engine.cc

@@ -451,7 +451,11 @@ void TensorRTEngine::SetITensor(const std::string &name,
   itensor_map_[name] = tensor;
 }
 
-nvinfer1::ITensor *TensorRTEngine::GetITensor(const std::string &name) {
+nvinfer1::ITensor *TensorRTEngine::GetITensor(const std::string &name,
+                                              bool scalar) {
+  if (scalar) {
+    return ConvertWeight2ITensor(name, true);
+  }
   if (itensor_map_.count(name)) {
     return itensor_map_[name];
   } else {
@@ -463,7 +467,7 @@ nvinfer1::ITensor *TensorRTEngine::GetITensor(const std::string &name) {
 // For cases when input is not middle-tensor , but persistable tensor
 // you should call this.
 nvinfer1::ITensor *TensorRTEngine::ConvertWeight2ITensor(
-    const std::string &name) {
+    const std::string &name, bool scalar) {
   auto *var_v = scope_->FindVar(name);
   PADDLE_ENFORCE_NOT_NULL(
       var_v,
@@ -489,9 +493,15 @@ nvinfer1::ITensor *TensorRTEngine::ConvertWeight2ITensor(
       trt_in_shape.d[i] = trt_in_shape.d[i + 1];
     }
   }
+  if (scalar) {
+    trt_in_shape.nbDims = 0;
+    trt_in_shape.d[0] = var_dims[0];
+  }
   nvinfer1::ILayer *layer =
       TRT_ENGINE_ADD_LAYER(this, Constant, trt_in_shape, weight.get());
-  this->SetITensor(name, layer->getOutput(0));
+  if (!scalar) {
+    this->SetITensor(name, layer->getOutput(0));
+  }
   return layer->getOutput(0);
 }
 

paddle/fluid/inference/tensorrt/engine.h

@@ -295,8 +295,9 @@ class TensorRTEngine {
   void DeleteITensor(const std::string& name, nvinfer1::ITensor* tensor);
   void SetITensor(const std::string& name, nvinfer1::ITensor* tensor);
   // Get an ITensor called name.
-  nvinfer1::ITensor* GetITensor(const std::string& name);
-  nvinfer1::ITensor* ConvertWeight2ITensor(const std::string& name);
+  nvinfer1::ITensor* GetITensor(const std::string& name, bool scalar = false);
+  nvinfer1::ITensor* ConvertWeight2ITensor(const std::string& name,
+                                           bool scalar = false);
   std::unordered_map<std::string, nvinfer1::ITensor*>* GetITensorMap();
 
   nvinfer1::ICudaEngine* engine() { return infer_engine_.get(); }

paddle/fluid/inference/tensorrt/op_teller.cc

@@ -337,6 +337,12 @@ struct SimpleOpTypeSetTeller : public Teller {
       }
     }
 
+    if (op_type == "range") {
+      if (!with_dynamic_shape) {
+        return false;
+      }
+    }
+
     if (op_type == "sign") {
 #if IS_TRT_VERSION_GE(8200)
       if (!with_dynamic_shape) {
@@ -2369,6 +2375,7 @@ struct SimpleOpTypeSetTeller : public Teller {
       "matmul",
       "matmul_v2",
       "bmm",
+      "range",
       "conv2d",
       "conv2d_fusion",
       "pool2d",
@@ -2507,6 +2514,7 @@ struct SimpleOpTypeSetTeller : public Teller {
       "matmul",
       "matmul_v2",
       "bmm",
+      "range",
       "conv2d",
       "conv2d_fusion",
       "pool2d",

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

+# Copyright (c) 2022 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.
+
+import unittest
+from functools import partial
+from typing import List
+
+import numpy as np
+from program_config import ProgramConfig, TensorConfig
+from trt_layer_auto_scan_test import TrtLayerAutoScanTest
+
+import paddle.inference as paddle_infer
+
+
+class TrtConvertRangeDynamicTest(TrtLayerAutoScanTest):
+    def is_program_valid(self, program_config: ProgramConfig) -> bool:
+        return True
+
+    def sample_program_configs(self):
+        def generate_input():
+            return np.array([1]).astype(np.int32)
+
+        for in_dtype in [2]:
+            self.in_dtype = in_dtype
+            dics = [{}]
+            ops_config = [
+                {
+                    "op_type": "fill_constant",
+                    "op_inputs": {},
+                    "op_outputs": {"Out": ["start_data"]},
+                    "op_attrs": {
+                        "dtype": self.in_dtype,
+                        "str_value": "7",
+                        "shape": [1],
+                    },
+                },
+                {
+                    "op_type": "fill_constant",
+                    "op_inputs": {},
+                    "op_outputs": {"Out": ["end_data"]},
+                    "op_attrs": {
+                        "dtype": self.in_dtype,
+                        "str_value": "256",
+                        "shape": [1],
+                    },
+                },
+                {
+                    "op_type": "fill_constant",
+                    "op_inputs": {},
+                    "op_outputs": {"Out": ["step_data"]},
+                    "op_attrs": {
+                        "dtype": self.in_dtype,
+                        "str_value": "1",
+                        "shape": [1],
+                    },
+                },
+                {
+                    "op_type": "range",
+                    "op_inputs": {
+                        "Start": ["start_data"],
+                        "End": ["end_data"],
+                        "Step": ["step_data"],
+                    },
+                    "op_outputs": {"Out": ["range_output_data1"]},
+                    "op_attrs": dics[0],
+                },
+                {
+                    "op_type": "cast",
+                    "op_inputs": {"X": ["range_output_data1"]},
+                    "op_outputs": {"Out": ["range_output_data"]},
+                    "op_attrs": {"in_dtype": self.in_dtype, "out_dtype": 5},
+                },
+            ]
+            ops = self.generate_op_config(ops_config)
+
+            program_config = ProgramConfig(
+                ops=ops,
+                weights={},
+                inputs={
+                    "step_data": TensorConfig(data_gen=partial(generate_input)),
+                },
+                outputs=["range_output_data"],
+            )
+
+            yield program_config
+
+    def sample_predictor_configs(
+        self, program_config
+    ) -> (paddle_infer.Config, List[int], float):
+        def generate_dynamic_shape(attrs):
+            self.dynamic_shape.min_input_shape = {
+                "start_data": [1],
+                "end_data": [1],
+                "step_data": [1],
+            }
+            self.dynamic_shape.max_input_shape = {
+                "start_data": [1],
+                "end_data": [1],
+                "step_data": [1],
+            }
+            self.dynamic_shape.opt_input_shape = {
+                "start_data": [1],
+                "end_data": [1],
+                "step_data": [1],
+            }
+
+        def clear_dynamic_shape():
+            self.dynamic_shape.min_input_shape = {}
+            self.dynamic_shape.max_input_shape = {}
+            self.dynamic_shape.opt_input_shape = {}
+
+        def generate_trt_nodes_num(attrs, dynamic_shape):
+            return 1, 2
+
+        attrs = [
+            program_config.ops[i].attrs for i in range(len(program_config.ops))
+        ]
+
+        # for dynamic_shape
+        generate_dynamic_shape(attrs)
+        self.trt_param.precision = paddle_infer.PrecisionType.Float32
+        yield self.create_inference_config(), generate_trt_nodes_num(
+            attrs, True
+        ), 1e-5
+        self.trt_param.precision = paddle_infer.PrecisionType.Half
+        yield self.create_inference_config(), generate_trt_nodes_num(
+            attrs, True
+        ), 1e-2
+
+    def test(self):
+        self.run_test()
+
+
+class TrtConvertRangeStaticTest(TrtLayerAutoScanTest):
+    def is_program_valid(self, program_config: ProgramConfig) -> bool:
+        return True
+
+    def sample_program_configs(self):
+        def generate_input():
+            return np.array([0]).astype(np.int32)
+
+        def generate_input1():
+            return np.array([128]).astype(np.int32)
+
+        def generate_input2():
+            return np.array([1]).astype(np.int32)
+
+        for in_dtype in [2, 5]:
+            self.in_dtype = in_dtype
+            dics = [{}]
+            ops_config = [
+                {
+                    "op_type": "range",
+                    "op_inputs": {
+                        "Start": ["start_data"],
+                        "End": ["end_data"],
+                        "Step": ["step_data"],
+                    },
+                    "op_outputs": {"Out": ["range_output_data1"]},
+                    "op_attrs": dics[0],
+                },
+                {
+                    "op_type": "cast",
+                    "op_inputs": {"X": ["range_output_data1"]},
+                    "op_outputs": {"Out": ["range_output_data"]},
+                    "op_attrs": {"in_dtype": self.in_dtype, "out_dtype": 5},
+                },
+            ]
+            ops = self.generate_op_config(ops_config)
+
+            program_config = ProgramConfig(
+                ops=ops,
+                weights={},
+                inputs={
+                    "start_data": TensorConfig(
+                        data_gen=partial(generate_input)
+                    ),
+                    "end_data": TensorConfig(data_gen=partial(generate_input1)),
+                    "step_data": TensorConfig(
+                        data_gen=partial(generate_input2)
+                    ),
+                },
+                outputs=["range_output_data"],
+            )
+
+            yield program_config
+
+    def sample_predictor_configs(
+        self, program_config
+    ) -> (paddle_infer.Config, List[int], float):
+        def clear_dynamic_shape():
+            self.dynamic_shape.min_input_shape = {}
+            self.dynamic_shape.max_input_shape = {}
+            self.dynamic_shape.opt_input_shape = {}
+
+        def generate_trt_nodes_num(attrs, dynamic_shape):
+            return 0, 6
+
+        attrs = [
+            program_config.ops[i].attrs for i in range(len(program_config.ops))
+        ]
+
+        # for static_shape
+        clear_dynamic_shape()
+        self.trt_param.precision = paddle_infer.PrecisionType.Float32
+        yield self.create_inference_config(), generate_trt_nodes_num(
+            attrs, False
+        ), 1e-5
+        self.trt_param.precision = paddle_infer.PrecisionType.Half
+        yield self.create_inference_config(), generate_trt_nodes_num(
+            attrs, False
+        ), 1e-2
+
+    def test(self):
+        self.run_test()
+
+
+if __name__ == "__main__":
+    unittest.main()