support empty tensor (#9338)

* support empty tensor

paddle/fluid/framework/tensor_impl.h

@@ -117,10 +117,10 @@ inline void* Tensor::mutable_data(platform::Place place, std::type_index type) {
   if (holder_ != nullptr) {
     holder_->set_type(type);
   }
-  PADDLE_ENFORCE_GT(
-      numel(), 0,
-      "When calling this method, the Tensor's numel must be larger than zero. "
-      "Please check Tensor::Resize has been called first.");
+  PADDLE_ENFORCE_GE(numel(), 0,
+                    "When calling this method, the Tensor's numel must be "
+                    "equal or larger than zero. "
+                    "Please check Tensor::Resize has been called first.");
   int64_t size = numel() * SizeOfType(type);
   /* some versions of boost::variant don't have operator!= */
   if (holder_ == nullptr || !(holder_->place() == place) ||

paddle/fluid/memory/memory_test.cc

@@ -59,7 +59,7 @@ TEST(BuddyAllocator, CPUMultAlloc) {
   EXPECT_EQ(total_size, 0UL);
 
   for (auto size :
-       {128, 256, 1024, 4096, 16384, 65536, 262144, 1048576, 4194304}) {
+       {0, 128, 256, 1024, 4096, 16384, 65536, 262144, 1048576, 4194304}) {
     ps[paddle::memory::Alloc(cpu, size)] = size;
 
     // Buddy Allocator doesn't manage too large memory chunk
@@ -117,7 +117,7 @@ TEST(BuddyAllocator, GPUMultAlloc) {
   EXPECT_EQ(total_size, 0UL);
 
   for (auto size :
-       {128, 256, 1024, 4096, 16384, 65536, 262144, 1048576, 4194304}) {
+       {0, 128, 256, 1024, 4096, 16384, 65536, 262144, 1048576, 4194304}) {
     ps[paddle::memory::Alloc(gpu, size)] = size;
 
     // Buddy Allocator doesn't manage too large memory chunk

python/paddle/fluid/tests/unittests/test_tensor.py

@@ -126,7 +126,6 @@ class TestTensor(unittest.TestCase):
     def test_lod_tensor_gpu_init(self):
         if not core.is_compiled_with_cuda():
             return
-        scope = core.Scope()
         place = core.CUDAPlace(0)
         lod_py = [[0, 2, 5], [0, 2, 4, 5]]
         lod_tensor = core.LoDTensor()
@@ -144,6 +143,25 @@ class TestTensor(unittest.TestCase):
         self.assertAlmostEqual(2.0, lod_v[0, 0, 0, 1])
         self.assertListEqual(lod_py, lod_tensor.lod())
 
+    def test_empty_tensor(self):
+        place = core.CPUPlace()
+        scope = core.Scope()
+        var = scope.var("test_tensor")
+
+        tensor = var.get_tensor()
+
+        tensor.set_dims([0, 1])
+        tensor.alloc_float(place)
+
+        tensor_array = numpy.array(tensor)
+        self.assertEqual((0, 1), tensor_array.shape)
+
+        if core.is_compiled_with_cuda():
+            gpu_place = core.CUDAPlace(0)
+            tensor.alloc_float(gpu_place)
+            tensor_array = numpy.array(tensor)
+            self.assertEqual((0, 1), tensor_array.shape)
+
 
 if __name__ == '__main__':
     unittest.main()