Fix ctypes, don't close fd, use smart pointers

Author: fpetrini15

src/python/library/CMakeLists.txt

@@ -128,7 +128,10 @@ else()
 
   if (${TRITON_ENABLE_PERF_ANALYZER})
     set(perf_analyzer_arg --perf-analyzer ${CMAKE_INSTALL_PREFIX}/bin/perf_analyzer)
-  endif()
+  endif() # TRITON_ENABLE_PERF_ANALYZER
+  if (${TRITON_ENABLE_GPU})
+    set(gpu_arg --include-gpu-libs)
+  endif() # TRITON_ENABLE_GPU
   set(linux_wheel_stamp_file "linux_stamp.whl")
   add_custom_command(
     OUTPUT "${linux_wheel_stamp_file}"
@@ -138,6 +141,7 @@ else()
       --dest-dir "${CMAKE_CURRENT_BINARY_DIR}/linux"
       --linux
       ${perf_analyzer_arg}
+      ${gpu_arg}
     DEPENDS ${LINUX_WHEEL_DEPENDS}
   )
 
@@ -178,7 +182,14 @@ if(${TRITON_ENABLE_PYTHON_GRPC})
   )
 endif() # TRITON_ENABLE_PYTHON_GRPC
 
+# Generic Wheel
+set(WHEEL_DIR "${CMAKE_CURRENT_BINARY_DIR}/generic")
+install(
+  CODE "file(GLOB _Wheel \"${WHEEL_DIR}/triton*.whl\")"
+  CODE "file(INSTALL \${_Wheel} DESTINATION \"${CMAKE_INSTALL_PREFIX}/python\")"
+)
 
+# Platform-specific wheels
 if(WIN32)
   set(WHEEL_DIR "${CMAKE_CURRENT_BINARY_DIR}/windows")
 else()

src/python/library/build_wheel.py

@@ -91,6 +91,12 @@ def sed(pattern, replace, source, dest=None):
         required=False,
         help="Include windows specific artifacts.",
     )
+    parser.add_argument(
+        "--include-gpu-libs",
+        action="store_true",
+        required=False,
+        help="Include gpu specific libraries",
+    )
     parser.add_argument(
         "--perf-analyzer",
         type=str,
@@ -186,10 +192,11 @@ def sed(pattern, replace, source, dest=None):
             "tritonclient/utils/libcshm.so",
             os.path.join(FLAGS.whl_dir, "tritonclient/utils/shared_memory/libcshm.so"),
         )
-        cpdir(
-            "tritonclient/utils/cuda_shared_memory",
-            os.path.join(FLAGS.whl_dir, "tritonclient/utils/cuda_shared_memory"),
-        )
+        if FLAGS.include_gpu_libs:
+            cpdir(
+                "tritonclient/utils/cuda_shared_memory",
+                os.path.join(FLAGS.whl_dir, "tritonclient/utils/cuda_shared_memory"),
+            )
 
         # Copy the pre-compiled perf_analyzer binary
         if FLAGS.perf_analyzer is not None:
@@ -212,10 +219,11 @@ def sed(pattern, replace, source, dest=None):
             os.path.join(FLAGS.whl_dir, "tritonclient/utils/shared_memory/cshm.dll"),
         )
         # FIXME: Enable when Windows supports GPU tensors DLIS-4169
-        # cpdir(
-        #     "tritonclient/utils/cuda_shared_memory",
-        #     os.path.join(FLAGS.whl_dir, "tritonclient/utils/cuda_shared_memory"),
-        # )
+        # if FLAGS.include_gpu_libs:
+        #     cpdir(
+        #         "tritonclient/utils/cuda_shared_memory",
+        #         os.path.join(FLAGS.whl_dir, "tritonclient/utils/cuda_shared_memory"),
+        #     )
 
     shutil.copyfile("LICENSE.txt", os.path.join(FLAGS.whl_dir, "LICENSE.txt"))
     shutil.copyfile("setup.py", os.path.join(FLAGS.whl_dir, "setup.py"))

src/python/library/tritonclient/utils/shared_memory/__init__.py

@@ -205,7 +205,10 @@ def get_contents_as_numpy(shm_handle, datatype, shape, offset=0):
         The numpy array generated using the contents of the specified shared
         memory region.
     """
-    shm_file = c_void_p()
+    # Safe initializer for Unix case where shm_file must be dereferenced to
+    # base in order to store file descriptor.
+    safe_initializer = c_int(-1)
+    shm_file = cast(byref(safe_initializer), c_void_p)
     region_offset = c_uint64()
     byte_size = c_uint64()
     shm_addr = c_char_p()
@@ -284,8 +287,10 @@ def destroy_shared_memory_region(shm_handle):
     SharedMemoryException
         If unable to unlink the shared memory region.
     """
-
-    shm_file = c_void_p()
+    # Safe initializer for Unix case where shm_file must be dereferenced to
+    # base in order to store file descriptor.
+    safe_initializer = c_int(-1)
+    shm_file = cast(byref(safe_initializer), c_void_p)
     offset = c_uint64()
     byte_size = c_uint64()
     shm_addr = c_char_p()

src/python/library/tritonclient/utils/shared_memory/shared_memory.cc

@@ -52,7 +52,7 @@ SharedMemoryHandleCreate(
   handle->triton_shm_name_ = triton_shm_name;
   handle->base_addr_ = shm_addr;
   handle->shm_key_ = shm_key;
-  handle->platform_handle_ = new ShmFile(shm_file);
+  handle->platform_handle_ = std::make_unique<ShmFile>(shm_file);
   handle->offset_ = offset;
   handle->byte_size_ = byte_size;
   return static_cast<void*>(handle);
@@ -97,8 +97,7 @@ SharedMemoryRegionMap(
     return -1;
   }
 
-  // close shared memory descriptor, return 0 if success else return -1
-  return close(fd);
+  return 0;
 #endif
 }
 
@@ -119,29 +118,29 @@ SharedMemoryRegionCreate(
   DWORD high_order_size = (upperbound_size >> 32) & 0xFFFFFFFF;
   DWORD low_order_size = upperbound_size & 0xFFFFFFFF;
 
-  HANDLE local_handle = CreateFileMapping(
+  HANDLE shm_file = CreateFileMapping(
       INVALID_HANDLE_VALUE,  // use paging file
       NULL,                  // default security
       PAGE_READWRITE,        // read/write access
       high_order_size,       // maximum object size (high-order DWORD)
       low_order_size,        // maximum object size (low-order DWORD)
       shm_key);              // name of mapping object
 
-  if (local_handle == NULL) {
+  if (shm_file == NULL) {
     return -7;
   }
 
   // get base address of shared memory region
   void* shm_addr = nullptr;
-  int err = SharedMemoryRegionMap((void*)local_handle, 0, byte_size, &shm_addr);
+  int err = SharedMemoryRegionMap((void*)shm_file, 0, byte_size, &shm_addr);
   if (err == -1) {
     return -4;
   }
 
   // create a handle for the shared memory region
   *shm_handle = SharedMemoryHandleCreate(
       std::string(triton_shm_name), shm_addr, std::string(shm_key),
-      (void*)local_handle, 0, byte_size);
+      (void*)shm_file, 0, byte_size);
 #else
   // get shared memory region descriptor
   int shm_fd = shm_open(shm_key, O_RDWR | O_CREAT, S_IRUSR | S_IWUSR);
@@ -188,12 +187,12 @@ GetSharedMemoryHandleInfo(
 #ifdef _WIN32
   HANDLE* file = static_cast<HANDLE*>(shm_file);
 #else
-  int* file = *static_cast<int**>(shm_file);
+  int* file = *reinterpret_cast<int**>(shm_file);
 #endif  // _WIN32
   SharedMemoryHandle* handle = static_cast<SharedMemoryHandle*>(shm_handle);
   *shm_addr = static_cast<char*>(handle->base_addr_);
   *shm_key = handle->shm_key_.c_str();
-  *file = handle->platform_handle_->shm_file_;
+  *file = *(handle->platform_handle_->GetShmFile());
   *offset = handle->offset_;
   *byte_size = handle->byte_size_;
   return 0;
@@ -213,7 +212,7 @@ SharedMemoryRegionDestroy(void* shm_handle)
   // We keep Windows shared memory handles open until we are done
   // using them. When all handles are closed, the system will free
   // the section of the paging file that the object uses.
-  CloseHandle(handle->platform_handle_->shm_file_);
+  CloseHandle(*(handle->platform_handle_->GetShmFile()));
 #else
   int status = munmap(shm_addr, handle->byte_size_);
   if (status == -1) {
@@ -224,11 +223,11 @@ SharedMemoryRegionDestroy(void* shm_handle)
   if (shm_fd == -1) {
     return -5;
   }
+  close(*(handle->platform_handle_->GetShmFile()));
 #endif  // _WIN32
 
-  // FIXME: Investigate use of smart pointers for these
-  // allocations instead
-  delete handle->platform_handle_;
+  // FIXME: Investigate use of smart pointers for this
+  // allocation instead
   delete handle;
 
   return 0;

src/python/library/tritonclient/utils/shared_memory/shared_memory.h

@@ -32,7 +32,7 @@ extern "C" {
 #ifdef _WIN32
 #define TRITONCLIENT_DECLSPEC __declspec(dllexport)
 #else
-define TRITONCLIENT_DECLSPEC
+#define TRITONCLIENT_DECLSPEC
 #endif
 
 //==============================================================================

src/python/library/tritonclient/utils/shared_memory/shared_memory_handle.h

@@ -33,22 +33,28 @@
 #ifdef _WIN32
 #include <windows.h>
 #endif  // _WIN32
+#include <memory>
 
 struct ShmFile {
 #ifdef _WIN32
   HANDLE shm_file_;
   ShmFile(void* shm_file) { shm_file_ = static_cast<HANDLE>(shm_file); };
+  HANDLE* GetShmFile() { return &shm_file_; };
 #else
-  int shm_file_;
-  ShmFile(int shm_file) { shm_file_ = *static_cast<int*>(shm_file); };
+  std::unique_ptr<int> shm_file_;
+  ShmFile(void* shm_file)
+  {
+    shm_file_ = std::make_unique<int>(*static_cast<int*>(shm_file));
+  };
+  int* GetShmFile() { return shm_file_.get(); }
 #endif  // _WIN32
 };
 
 struct SharedMemoryHandle {
   std::string triton_shm_name_;
   std::string shm_key_;
   void* base_addr_;
-  ShmFile* platform_handle_;
+  std::unique_ptr<ShmFile> platform_handle_;
   size_t offset_;
   size_t byte_size_;
 };