!34184 fix the bug of StridedSlice op and DynamicStitch op

Merge pull request !34184 from hanhuifeng/asr_bug

mindspore/ccsrc/plugin/device/gpu/kernel/other/dynamic_stitch_gpu_kernel.cc

@@ -40,7 +40,11 @@ bool DynamicStitchKernelMod::Init(const CNodePtr &kernel_node) {
   size_t index_type_size = sizeof(int);
   data_type_size_ = GetDtypeNbyte(TypeIdToString(data_type, true));
   auto first_data_shape = AnfAlgo::GetInputDeviceShapeAdaptively(kernel_node, n_);
-  one_data_ele_num_ = first_data_shape[first_data_shape.size() - 1];
+  auto first_index_dims = AnfAlgo::GetInputDeviceShapeAdaptively(kernel_node, 0).size();
+  one_data_ele_num_ = 1;
+  for (size_t d = first_index_dims; d < first_data_shape.size(); ++d) {
+    one_data_ele_num_ *= first_data_shape[d];
+  }
   for (size_t i = 0; i < n_; i++) {
     auto data_shape = AnfAlgo::GetInputDeviceShapeAdaptively(kernel_node, n_ + i);
     size_t data_size = std::accumulate(data_shape.begin(), data_shape.end(), 1, std::multiplies<size_t>());

mindspore/core/ops/strided_slice.cc

@@ -375,7 +375,7 @@ abstract::ShapePtr StridedSliceInferShape(const PrimitivePtr &primitive,
   }
   ret_in_shape = DynamicComputeInferShape(primitive, begin_v, end_v, strides_v, x_shape, begin_len);
 
-  if (max_shape.empty() || min_shape.empty()) {
+  if (x_is_dyn && (max_shape.empty() || min_shape.empty())) {
     return std::make_shared<abstract::Shape>(ret_in_shape);
   }
 

mindspore/python/mindspore/ops/operations/array_ops.py

@@ -3632,11 +3632,9 @@ class StridedSlice(PrimitiveWithInfer):
             rets = {'shape': ret_shape,
                     'dtype': x['dtype'],
                     'value': None}
-
-            if max_shape is not None and min_shape is not None:
-                rets = self._compute_max_min_shape(rets, x_shape, max_shape, min_shape, ret_shape)
-
-            return rets
+            if -1 in x_shape and (max_shape is None or min_shape is None):
+                return rets
+            return self._compute_max_min_shape(rets, x_shape, max_shape, min_shape, ret_shape)
 
         ret_shape = self._compute_slicing_shape(x_shape, begin_v['value'], end_v['value'], strides_v['value'])
         if all(ret_shape):

tests/st/ops/gpu/test_dynamic_stitch_op.py

@@ -55,3 +55,50 @@ def test_net_int32():
     net = Net()
     output = net(indices, data)
     assert np.array_equal(output.asnumpy(), expected)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_net_1():
+    """
+    Feature: Test dynamicstitch op.
+    Description: An index corresponds to a number
+    Expectation: the result match expected array.
+    """
+    x1 = Tensor(np.array([0, 1]), mindspore.int32)
+    x2 = Tensor([1], mindspore.int32)
+    y1 = Tensor(np.array([1, 3]), mindspore.int32)
+    y2 = Tensor(np.array([2]), mindspore.int32)
+    expected = np.array([1, 2]).astype(np.int32)
+
+    indices = [x1, x2]
+    data = [y1, y2]
+    net = Net()
+    output = net(indices, data)
+    assert np.array_equal(output.asnumpy(), expected)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_net_2():
+    """
+    Feature: Test dynamicstitch op.
+    Description: An index corresponds to a multidimensional array.
+    Expectation: the result match expected array.
+    """
+    x1 = Tensor(np.array([0, 2]), mindspore.int32)
+    x2 = Tensor([1], mindspore.int32)
+    y1 = Tensor(np.array([[[1, 2, 3], [4, 5, 6]],
+                          [[13, 14, 15], [16, 17, 18]]]), mindspore.int32)
+    y2 = Tensor(np.array([[[7, 8, 9], [10, 11, 12]]]), mindspore.int32)
+    expected = np.array([[[1, 2, 3], [4, 5, 6]],
+                         [[7, 8, 9], [10, 11, 12]],
+                         [[13, 14, 15], [16, 17, 18]]]).astype(np.int32)
+
+    indices = [x1, x2]
+    data = [y1, y2]
+    net = Net()
+    output = net(indices, data)
+    assert np.array_equal(output.asnumpy(), expected)