[cmake] upgrade windows paddle inference -> 2.5.0 (#2100)

* [cmake] upgrade windows paddle inference -> 2.5.0 * [cmake] upgrade windows paddle inference -> 2.5.0 * fix paddle custom ops bug on windows
PaddlePaddle · Jul 12, 2023 · 2542a75 · 2542a75
1 parent c95fc1f
commit 2542a75
Show file tree

Hide file tree

Showing 3 changed files with 41 additions and 31 deletions.
diff --git a/cmake/paddle_inference.cmake b/cmake/paddle_inference.cmake
@@ -74,11 +74,11 @@ else()
     set(PADDLEINFERENCE_URL_BASE "https://bj.bcebos.com/fastdeploy/third_libs/")
     if(WIN32)
       if (WITH_GPU)
-        set(PADDLEINFERENCE_FILE "paddle_inference-win-x64-gpu-trt8.5.2.2-mkl-avx-0.0.0.575cafb44b.zip")
-        set(PADDLEINFERENCE_VERSION "0.0.0.575cafb44b")
+        set(PADDLEINFERENCE_FILE "paddle_inference-win-x64-gpu-trt8.5.2.2-mkl-2.5.0.281761089e.zip")
+        set(PADDLEINFERENCE_VERSION "2.5.0.281761089e")
       else()
-        set(PADDLEINFERENCE_FILE "paddle_inference-win-x64-mkl-avx-0.0.0.cbdba50933.zip")
-        set(PADDLEINFERENCE_VERSION "0.0.0.cbdba50933")
+        set(PADDLEINFERENCE_FILE "paddle_inference-win-x64-mkl-2.5.0.281761089e.zip")
+        set(PADDLEINFERENCE_VERSION "2.5.0.281761089e")
       endif()
     elseif(APPLE)
       if(CURRENT_OSX_ARCH MATCHES "arm64")
@@ -310,33 +310,35 @@ if(PADDLEINFERENCE_API_COMPAT_2_5_x)
 endif()
 
 function(set_paddle_custom_ops_compatible_policy)
-  if(PADDLEINFERENCE_API_CUSTOM_OP AND (NOT MSVC))
-    # TODO: add non c++ 14 policy for latest dev
-    if(NOT PADDLEINFERENCE_API_COMPAT_2_5_x)
-      # gcc c++ 14 policy for 2.4.x
-      if(NOT DEFINED CMAKE_CXX_STANDARD)
-        set(CMAKE_CXX_STANDARD 14 PARENT_SCOPE)
-        message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON, but CMAKE_CXX_STANDARD is not defined, use c++ 14 by default!")
-      elseif(NOT (CMAKE_CXX_STANDARD EQUAL 14))
-        set(CMAKE_CXX_STANDARD 14 PARENT_SCOPE)
-        message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON, force use c++ 14!")
-      endif()
-    endif()
-    if(WITH_GPU)
-      # cuda c++ 14 policy for 2.4.x
+  if(PADDLEINFERENCE_API_CUSTOM_OP)
+    if(NOT MSVC)
+      # TODO: add non c++ 14 policy for latest dev
       if(NOT PADDLEINFERENCE_API_COMPAT_2_5_x)
-        if(NOT DEFINED CMAKE_CUDA_STANDARD)
-          set(CMAKE_CUDA_STANDARD 14 PARENT_SCOPE)
-          message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON and WITH_GPU=ON, but CMAKE_CUDA_STANDARD is not defined, use c++ 14 by default!")
-        elseif(NOT (CMAKE_CUDA_STANDARD EQUAL 14))
-          set(CMAKE_CUDA_STANDARD 14 PARENT_SCOPE)
-          message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON and WITH_GPU=ON, force use c++ 14!")
+        # gcc c++ 14 policy for 2.4.x
+        if(NOT DEFINED CMAKE_CXX_STANDARD)
+          set(CMAKE_CXX_STANDARD 14 PARENT_SCOPE)
+          message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON, but CMAKE_CXX_STANDARD is not defined, use c++ 14 by default!")
+        elseif(NOT (CMAKE_CXX_STANDARD EQUAL 14))
+          set(CMAKE_CXX_STANDARD 14 PARENT_SCOPE)
+          message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON, force use c++ 14!")
+        endif()
+      endif()
+      if(WITH_GPU)
+        # cuda c++ 14 policy for 2.4.x
+        if(NOT PADDLEINFERENCE_API_COMPAT_2_5_x)
+          if(NOT DEFINED CMAKE_CUDA_STANDARD)
+            set(CMAKE_CUDA_STANDARD 14 PARENT_SCOPE)
+            message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON and WITH_GPU=ON, but CMAKE_CUDA_STANDARD is not defined, use c++ 14 by default!")
+          elseif(NOT (CMAKE_CUDA_STANDARD EQUAL 14))
+            set(CMAKE_CUDA_STANDARD 14 PARENT_SCOPE)
+            message(WARNING "Found PADDLEINFERENCE_API_CUSTOM_OP=ON and WITH_GPU=ON, force use c++ 14!")
+          endif()
         endif()
       endif()
-      # compile flags for paddle custom ops
-      add_definitions(-DPADDLE_WITH_CUDA)
-      add_definitions(-DPADDLE_ON_INFERENCE)
-      add_definitions(-DPADDLE_NO_PYTHON)
     endif()
+    # common compile flags for paddle custom ops
+    add_definitions(-DPADDLE_WITH_CUDA)
+    add_definitions(-DPADDLE_ON_INFERENCE)
+    add_definitions(-DPADDLE_NO_PYTHON)  
   endif()
 endfunction()
diff --git a/fastdeploy/runtime/backends/paddle/ops/centerpoint_postprocess_op.cu b/fastdeploy/runtime/backends/paddle/ops/centerpoint_postprocess_op.cu
@@ -27,7 +27,7 @@
 
 #define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
 
-const int THREADS_PER_BLOCK_NMS = sizeof(int64_t) * 8;
+static const int THREADS_PER_BLOCK_NMS = sizeof(int64_t) * 8;
 
 void NmsLauncher(const cudaStream_t &stream, const float *bboxes,
                  const int *index, const int64_t *sorted_index,

diff --git a/fastdeploy/runtime/backends/paddle/ops/iou3d_nms_kernel.cu b/fastdeploy/runtime/backends/paddle/ops/iou3d_nms_kernel.cu
@@ -22,8 +22,16 @@ All Rights Reserved 2019-2020.
 #define THREADS_PER_BLOCK 16
 #define DIVUP(m, n) ((m) / (n) + ((m) % (n) > 0))
 
-const int THREADS_PER_BLOCK_NMS = sizeof(int64_t) * 8;
-const float EPS = 1e-8;
+static const int THREADS_PER_BLOCK_NMS = sizeof(int64_t) * 8;
+#if defined(_WIN32)
+#if defined(EPS)
+#undef EPS
+#endif
+#define EPS 1e-8
+#else
+static const float EPS = 1e-8;
+#endif
+
 struct Point {
   float x, y;
   __device__ Point() {}