Add cublas_handle() to expose cublas_handle to ops (#31157) (#31190)

* add get_cublas_handle() api

* update format

* add unittests

* alter function name

Author: FrostML

paddle/fluid/platform/cuda_helper.h

@@ -96,12 +96,14 @@ class CublasHandleHolder {
 #endif  // CUDA_VERSION >= 9000
   }
 
+  const cublasHandle_t& GetCublasHandle() const { return handle_; }
+
   ~CublasHandleHolder() PADDLE_MAY_THROW {
     PADDLE_RETRY_CUDA_SUCCESS(dynload::cublasDestroy(handle_));
   }
 
   template <typename Callback>
-  inline void Call(Callback &&callback) const {
+  inline void Call(Callback&& callback) const {
     std::lock_guard<std::mutex> guard(mtx_);
     callback(handle_);
   }

paddle/fluid/platform/device_context.cc

@@ -437,6 +437,10 @@ cudnnHandle_t CUDADeviceContext::cudnn_handle() const {
   return context()->CudnnHandle();
 }
 
+cublasHandle_t CUDADeviceContext::cublas_handle() const {
+  return context()->CublasHandle()->GetCublasHandle();
+}
+
 CudnnWorkspaceHandle CUDADeviceContext::cudnn_workspace_handle() const {
   return CudnnWorkspaceHandle(*this, &cudnn_handle_mtx_);
 }

paddle/fluid/platform/device_context.h

@@ -346,6 +346,9 @@ class CUDADeviceContext : public DeviceContext {
   /*! \brief  Return cudnn  handle in the device context. */
   cudnnHandle_t cudnn_handle() const;
 
+  /*! \brief  Return cublas handle in the device context. */
+  cublasHandle_t cublas_handle() const;
+
   /*! \brief  Return a cudnn workspace handle to call multiple cudnn
    *  functions without interrupting by other threads.
    *  Once the first cudnn function is called by the handle, a lock

paddle/fluid/platform/device_context_test.cu

@@ -43,6 +43,8 @@ TEST(Device, CUDADeviceContext) {
     ASSERT_NE(nullptr, gpu_device);
     cudnnHandle_t cudnn_handle = device_context->cudnn_handle();
     ASSERT_NE(nullptr, cudnn_handle);
+    cublasHandle_t cublas_handle = device_context->cublas_handle();
+    ASSERT_NE(nullptr, cublas_handle);
     delete device_context;
   }
 }