Fix gaussian_random_op compile error

* Should always use `dynload::` for cuda function.
* Fix cublas.h without DSO load.

paddle/operators/gaussian_random_op.cu

@@ -40,8 +40,8 @@ class GaussianRandomKernel : public framework::OpKernel {
         &g, CURAND_RNG_PSEUDO_DEFAULT));
     PADDLE_ENFORCE(
         platform::dynload::curandSetPseudoRandomGeneratorSeed(g, seed));
-    curandGenerateNormal(g, data, framework::product(tensor->dims()), mean,
-                         std);
+    platform::dynload::curandGenerateNormal(
+        g, data, framework::product(tensor->dims()), mean, std);
   }
 };
 

paddle/platform/dynload/cublas.h

@@ -50,8 +50,8 @@ extern void *cublas_dso_handle;
 #else
 #define DECLARE_DYNAMIC_LOAD_CUBLAS_WRAP(__name)     \
   struct DynLoad__##__name {                         \
-    inline template <typename... Args>           \
-    cublasStatus_t operator()(Args... args) {    \
+    template <typename... Args>                      \
+    inline cublasStatus_t operator()(Args... args) { \
       return __name(args...);                        \
     }                                                \
   };                                                 \