diff --git a/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.cu
index 01ee86ceb48a9ef022ba73fe0dbdab4a52324cc6..8e9845183b3fe76f7f4cfd3e69d28c3ead8f852f 100644
--- a/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.cu
@@ -166,7 +166,11 @@ int AnchorGeneratorPlugin::enqueue_impl(int batch_size,
 }
 
 int AnchorGeneratorPlugin::enqueue(int batch_size, const void* const* inputs,
+#if IS_TRT_VERSION_LT(8000)
                                    void** outputs, void* workspace,
+#else
+                                   void* const* outputs, void* workspace,
+#endif
                                    cudaStream_t stream) {
   return enqueue_impl<float>(batch_size, inputs, outputs, workspace, stream);
 }
diff --git a/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.h
index aff0b6a6802f114a25acf32627a39ca42d572d7c..458326d0679ca96df16db1287139de986f2f3cb4 100644
--- a/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/anchor_generator_op_plugin.h
@@ -42,7 +42,11 @@ class AnchorGeneratorPlugin : public nvinfer1::IPluginV2Ext {
   bool supportsFormat(nvinfer1::DataType type,
                       nvinfer1::TensorFormat format) const override;
   size_t getWorkspaceSize(int max_batch_size) const override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batch_size, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
   int initialize() override;
   void terminate() override;
diff --git a/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.cu
index cc17f8aa2481708e3e19c9925a1d83ad06203145..687e564e8a8360c3d94ed4e18761489bfe5f6245 100644
--- a/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.cu
@@ -122,7 +122,11 @@ int ElementWisePlugin::initialize() {
 }
 
 int ElementWisePlugin::enqueue(int batch_size, const void *const *inputs,
+#if IS_TRT_VERSION_LT(8000)
                                void **outputs, void *workspace,
+#else
+                               void *const *outputs, void *workspace,
+#endif
                                cudaStream_t stream) {
   const float *x = reinterpret_cast<const float *>(inputs[0]);
   const float *y = reinterpret_cast<const float *>(inputs[1]);
diff --git a/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.h
index 75a1dd85f0f2c440fdd16beb95144df4127739e6..946e327e355798d61b7907ad98218f0de10fc8c9 100644
--- a/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/elementwise_op_plugin.h
@@ -58,8 +58,11 @@ class ElementWisePlugin : public PluginTensorRT {
 
   int initialize() override;
 
-  // execute the layer
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batch_size, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream);
 
  protected:
diff --git a/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.cu
index deda2e2cc7247f404ff6d11409b665898d550ee1..3d84855bcbddb92c07f8536a40172687adaab7de 100644
--- a/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.cu
@@ -100,7 +100,11 @@ __global__ void no_exact_gelu_kernel(const T a, const T b, const T c, int n,
 }
 
 int GeluPlugin::enqueue(int batch_size, const void* const* inputs,
+#if IS_TRT_VERSION_LT(8000)
                         void** outputs, void*, cudaStream_t stream) {
+#else
+                        void* const* outputs, void*, cudaStream_t stream) {
+#endif
   const auto& input_dims = this->getInputDims(0);
   int num = batch_size;
   for (int i = 0; i < input_dims.nbDims; i++) {
diff --git a/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.h
index 23e507ee477e1a3b85339c7b267b290de19805ab..98c05e9792af4550811ff03588d73a0573943c05 100644
--- a/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.h
@@ -44,7 +44,11 @@ class GeluPlugin : public PluginTensorRT {
                       nvinfer1::PluginFormat format) const override;
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims* inputs,
                                      int nb_input_dims) override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batch_size, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 
  protected:
diff --git a/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.cu
index 8b2d0ac3cf70f77f1ff9ce9a6fe2ed19fdcf9576..df25b5ba927974779389bc38610189397bb70048 100644
--- a/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.cu
@@ -59,7 +59,11 @@ __global__ void hard_swish_kernel(float threshold, float scale, float offset,
 }
 
 int HardSwishPlugin::enqueue(int batch_size, const void* const* inputs,
+#if IS_TRT_VERSION_LT(8000)
                              void** outputs, void*, cudaStream_t stream) {
+#else
+                             void* const* outputs, void*, cudaStream_t stream) {
+#endif
   const auto& input_dims = this->getInputDims(0);
   int num = batch_size;
   for (int i = 0; i < input_dims.nbDims; i++) {
diff --git a/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.h
index 2e1e1d03baf7e1cb046f887f2d799a907f3586d4..ad1952c246a80fbef1744faa813a6d7a8893414a 100644
--- a/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/hard_swish_op_plugin.h
@@ -49,7 +49,11 @@ class HardSwishPlugin : public PluginTensorRT {
   int initialize() override { return 0; }
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims* inputs,
                                      int nbInputDims) override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batchSize, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batchSize, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 
  protected:
diff --git a/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.cu
index a579743ee8ad1a9ae480cebf03380635c3a300c4..af063c61c5a56877f4799c33d61a2a368702f4af 100644
--- a/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.cu
@@ -59,7 +59,11 @@ nvinfer1::Dims InstanceNormPlugin::getOutputDimensions(
 }
 
 int InstanceNormPlugin::enqueue(int batch_size, const void *const *inputs,
+#if IS_TRT_VERSION_LT(8000)
                                 void **outputs, void *workspace,
+#else
+                                void *const *outputs, void *workspace,
+#endif
                                 cudaStream_t stream) {
   const auto &input_dims = this->getInputDims(0);
 
diff --git a/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.h
index 83422708f593d8fef66bb2d3b463ede80f041398..f413505bdf43e9828050a5e3dc6851bd1effcb8d 100644
--- a/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/instance_norm_op_plugin.h
@@ -101,7 +101,11 @@ class InstanceNormPlugin : public PluginTensorRT {
   int getNbOutputs() const override { return 1; }
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims *inputs,
                                      int nbInputDims) override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batchSize, const void *const *inputs, void **outputs,
+#else
+  int enqueue(int batchSize, const void *const *inputs, void *const *outputs,
+#endif
               void *workspace, cudaStream_t stream) override;
 
   bool supportsFormat(nvinfer1::DataType type,
diff --git a/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.cu
index f9341613a0f55e6f4693c5994e369236c0d1080c..4d55aea316a358e617f1d1ee479273cb644b7534 100644
--- a/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.cu
@@ -43,7 +43,11 @@ nvinfer1::Dims LayerNormPlugin::getOutputDimensions(
 }
 
 int LayerNormPlugin::enqueue(int batch_size, const void *const *inputs,
+#if IS_TRT_VERSION_LT(8000)
                              void **outputs, void *workspace,
+#else
+                             void *const *outputs, void *workspace,
+#endif
                              cudaStream_t stream) {
   const auto &input_dims = this->getInputDims(0);
   const float *input = reinterpret_cast<const float *>(inputs[0]);
diff --git a/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.h
index 9c4c31b61e128ddcb8586b2cc82978767a41182d..a16c5191f88644bb80d8d9dd3a430a0c6c004752 100644
--- a/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.h
@@ -100,7 +100,11 @@ class LayerNormPlugin : public PluginTensorRT {
   int getNbOutputs() const override { return 1; }
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims* inputs,
                                      int nbInputDims) override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batchSize, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batchSize, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 };
 
diff --git a/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.cu
index 154f61a2b7cd3f066cc1a671f8277232fde65a9d..fb8043a9d90e4b0274087c7943eebefb5e545bc9 100644
--- a/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.cu
@@ -42,7 +42,12 @@ nvinfer1::Dims PoolPlugin::getOutputDimensions(int index,
 }
 
 int PoolPlugin::enqueue(int batchSize, const void *const *inputs,
+#if IS_TRT_VERSION_LT(8000)
                         void **outputs, void *workspace, cudaStream_t stream) {
+#else
+                        void *const *outputs, void *workspace,
+                        cudaStream_t stream) {
+#endif
   auto const &input_dims = this->getInputDims(0);
   int input_size = 0;
   float const *idata = reinterpret_cast<float const *>(inputs[0]);
diff --git a/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.h
index 6693a1fae4d4304af2f826894b119383ea704727..90ce44e6822565fe22bff5bf18b775ca8b2a7c2f 100644
--- a/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/pool_op_plugin.h
@@ -128,7 +128,11 @@ class PoolPlugin : public PluginTensorRT {
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims* inputs,
                                      int nbInputDims) override;
   int initialize() override { return 0; }
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batchSize, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batchSize, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 
  private:
diff --git a/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.cu
index 00182b87e984fc3c43f46a3fcb2b9d828db4b170..ad3618bc67b0457efbfba52cf6ecfa1c8ee7c398 100644
--- a/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.cu
@@ -57,7 +57,12 @@ nvinfer1::Dims PReluPlugin::getOutputDimensions(int index,
 }
 
 int PReluPlugin::enqueue(int batch_size, const void *const *inputs,
+#if IS_TRT_VERSION_LT(8000)
                          void **outputs, void *workspace, cudaStream_t stream) {
+#else
+                         void *const *outputs, void *workspace,
+                         cudaStream_t stream) {
+#endif
   // input dims is CHW.
   const auto &input_dims = this->getInputDims(0);
   const float *input = reinterpret_cast<const float *>(inputs[0]);
diff --git a/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.h
index a0a24e70a01ef47fa71d9d79f7cc2554a60683d0..313272823d4a6dfc39bf207216aa9d50052307d7 100644
--- a/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/prelu_op_plugin.h
@@ -80,7 +80,11 @@ class PReluPlugin : public PluginTensorRT {
   int getNbOutputs() const override { return 1; }
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims* inputs,
                                      int nbInputDims) override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batchSize, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batchSize, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 };
 
diff --git a/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.cu
index b44b3face92e14fc49732621d5397a6fdcf859a2..42d9018fd057952975ffc572828701662e5ba231 100644
--- a/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.cu
@@ -111,7 +111,12 @@ nvinfer1::Dims SlicePlugin::getOutputDimensions(int index,
 }
 
 int SlicePlugin::enqueue(int batch_size, const void *const *inputs,
+#if IS_TRT_VERSION_LT(8000)
                          void **outputs, void *workspace, cudaStream_t stream) {
+#else
+                         void *const *outputs, void *workspace,
+                         cudaStream_t stream) {
+#endif
   auto input_dims = getInputDims(0);
 
   // notice input dims is [C, H, W], add input batch dim here
diff --git a/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.h
index 9d4f9a35c3b6fe02981853eb3c0a697d5cb3a199..015a6b116f60a198b2c669fa4d357cd962e30cdf 100644
--- a/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/slice_op_plugin.h
@@ -44,7 +44,11 @@ class SlicePlugin : public PluginTensorRT {
                       nvinfer1::PluginFormat format) const override;
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims* inputs,
                                      int nb_input_dims) override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batch_size, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 
  protected:
diff --git a/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.cu
index 1b5c39f8fff855fac4ef8f2ee54faa872023ad05..24d4715e0312dc3200001ae2d37f0fc378fab4b6 100644
--- a/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.cu
@@ -126,7 +126,12 @@ __global__ void split_kernel(int nsegment,
 }
 
 int SplitPlugin::enqueue(int batchSize, const void* const* inputs,
+#if IS_TRT_VERSION_LT(8000)
                          void** outputs, void* workspace, cudaStream_t stream) {
+#else
+                         void* const* outputs, void* workspace,
+                         cudaStream_t stream) {
+#endif
   const int* d_segment_offsets_ptr =
       thrust::raw_pointer_cast(&d_segment_offsets_[0]);
   float const* input_ptr = reinterpret_cast<float const*>(inputs[0]);
diff --git a/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.h
index 1ee895154d6b046c6c18c2e374d3c63f1fcc5d62..a791395f4a3d3824e4c54ed2cfaf97b79859fde4 100644
--- a/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/split_op_plugin.h
@@ -60,7 +60,11 @@ class SplitPlugin : public PluginTensorRTV2Ext {
 
   int initialize() override;
   void terminate() override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batch_size, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 
   void destroy() override { delete this; }
diff --git a/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.cu
index 3847d999446e99dfe0bcdc7abfa06ac6c57e64e2..52e5af01822fac5f563b48f40e039e8297ccd4f0 100644
--- a/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.cu
@@ -85,7 +85,12 @@ __global__ void swish_kernel<half>(int num, const half *input, half *output,
 }
 
 int SwishPlugin::enqueue(int batch_size, const void *const *inputs,
+#if IS_TRT_VERSION_LT(8000)
                          void **outputs, void *workspace, cudaStream_t stream) {
+#else
+                         void *const *outputs, void *workspace,
+                         cudaStream_t stream) {
+#endif
   // input dims is CHW.
   const auto &input_dims = this->getInputDims(0);
   const float *input = reinterpret_cast<const float *>(inputs[0]);
diff --git a/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.h
index 11579aadcc45731123770352ef08b362ff3ef745..2a8b637730b516ef03a7e7d5008110ae839d2fba 100644
--- a/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/swish_op_plugin.h
@@ -67,7 +67,11 @@ class SwishPlugin : public PluginTensorRT {
   int getNbOutputs() const override { return 1; }
   nvinfer1::Dims getOutputDimensions(int index, const nvinfer1::Dims* inputs,
                                      int nbInputDims) override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batchSize, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batchSize, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
 };
 
diff --git a/paddle/fluid/inference/tensorrt/plugin/trt_plugin.h b/paddle/fluid/inference/tensorrt/plugin/trt_plugin.h
index ce3133ae99e94c62c0c8e958065700373d270037..37be06bba3aebd518e8775c7ed8ccd24aa7fa7e1 100644
--- a/paddle/fluid/inference/tensorrt/plugin/trt_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/trt_plugin.h
@@ -82,8 +82,13 @@ class PluginTensorRT : public nvinfer1::IPluginExt {
   int initialize() override { return 0; }
   // Shutdown the layer. This is called when the engine is destroyed
   void terminate() override {}
-  // Execute the layer
+// Execute the layer
+#if IS_TRT_VERSION_LT(8000)
   virtual int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  virtual int enqueue(int batch_size, const void* const* inputs,
+                      void* const* outputs,
+#endif
                       void* workspace, cudaStream_t stream) = 0;
 
   // Find the size of the serialization buffer required
@@ -188,8 +193,13 @@ class PluginTensorRTV2Ext : public nvinfer1::IPluginV2Ext {
   // Find the workspace size required by the layer
   size_t getWorkspaceSize(int) const override { return 0; }
 
-  // Execute the layer
+// Execute the layer
+#if IS_TRT_VERSION_LT(8000)
   virtual int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  virtual int enqueue(int batch_size, const void* const* inputs,
+                      void* const* outputs,
+#endif
                       void* workspace, cudaStream_t stream) = 0;
 
   // Find the size of the serialization buffer required
diff --git a/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.cu
index 13d07e774036a48b0ed6e3c91b168eaab4461df5..f9767f38559482ee8377a37a29b56421dfe35eb7 100644
--- a/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.cu
@@ -243,7 +243,11 @@ int YoloBoxPlugin::enqueue_impl(int batch_size, const void* const* inputs,
 }
 
 int YoloBoxPlugin::enqueue(int batch_size, const void* const* inputs,
+#if IS_TRT_VERSION_LT(8000)
                            void** outputs, void* workspace,
+#else
+                           void* const* outputs, void* workspace,
+#endif
                            cudaStream_t stream) {
   if (data_type_ == nvinfer1::DataType::kFLOAT) {
     return enqueue_impl<float>(batch_size, inputs, outputs, workspace, stream);
diff --git a/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.h b/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.h
index 8ca21da7ae0377164cbb50c502f0abb5ca943058..4cd6a383336e236251b9cbef49c96b18a8fe0537 100644
--- a/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/yolo_box_op_plugin.h
@@ -43,7 +43,11 @@ class YoloBoxPlugin : public nvinfer1::IPluginV2Ext {
   bool supportsFormat(nvinfer1::DataType type,
                       nvinfer1::TensorFormat format) const override;
   size_t getWorkspaceSize(int max_batch_size) const override;
+#if IS_TRT_VERSION_LT(8000)
   int enqueue(int batch_size, const void* const* inputs, void** outputs,
+#else
+  int enqueue(int batch_size, const void* const* inputs, void* const* outputs,
+#endif
               void* workspace, cudaStream_t stream) override;
   template <typename T>
   int enqueue_impl(int batch_size, const void* const* inputs, void** outputs,