!17758 Fix some bug of ops.

From: @liu_xiao_93
Reviewed-by: @liangchenghui,@wuxuejian,@liangchenghui
Signed-off-by: @liangchenghui,@liangchenghui

mindspore/ccsrc/backend/optimizer/ascend/ir_fusion/avgpool_3d_fusion.cc

@@ -27,6 +27,7 @@
 namespace mindspore {
 namespace opt {
 namespace {
+constexpr size_t kAvgPool3DInputNum = 1;
 constexpr size_t k5DInferDims = 5;
 constexpr size_t kC0 = 16;
 
@@ -229,6 +230,11 @@ const AnfNodePtr AvgPool3DFusion::Process(const FuncGraphPtr &func_graph, const
   MS_EXCEPTION_IF_NULL(node);
   auto avg_pool_3d_node = node->cast<CNodePtr>();
   MS_EXCEPTION_IF_NULL(avg_pool_3d_node);
+  if (avg_pool_3d_node->size() != kAvgPool3DInputNum + 1) {
+    MS_LOG(INFO) << "The node " << avg_pool_3d_node->DebugString() << " is not equal to " << kAvgPool3DInputNum
+                 << " inputs. Can not do fusion.";
+    return nullptr;
+  }
   auto dims_in = AnfAlgo::GetPrevNodeOutputInferShape(avg_pool_3d_node, 0);
   auto dims_out = AnfAlgo::GetOutputInferShape(avg_pool_3d_node, 0);
   if (dims_in.size() < k5DInferDims || dims_out.size() < k5DInferDims) {

mindspore/ccsrc/backend/optimizer/ascend/ir_fusion/avgpool_3d_grad_fusion.cc

@@ -27,6 +27,7 @@
 namespace mindspore {
 namespace opt {
 namespace {
+constexpr size_t kAvgPool3DGradInputNum = 1;
 constexpr size_t k5DInferDims = 5;
 constexpr size_t kKernelDims = 3;
 constexpr size_t kStridesDims = 3;
@@ -208,6 +209,11 @@ const AnfNodePtr AvgPool3DGradFusion::Process(const FuncGraphPtr &func_graph, co
   MS_EXCEPTION_IF_NULL(node);
   auto avg_pool_3d_grad_node = node->cast<CNodePtr>();
   MS_EXCEPTION_IF_NULL(avg_pool_3d_grad_node);
+  if (avg_pool_3d_grad_node->size() != kAvgPool3DGradInputNum + 1) {
+    MS_LOG(INFO) << "The node " << avg_pool_3d_grad_node->DebugString() << " is not equal to " << kAvgPool3DGradInputNum
+                 << " inputs. Can not do fusion.";
+    return nullptr;
+  }
   std::vector<int64_t> kernel_size;
   std::vector<int64_t> strides;
   std::vector<int64_t> pad_list;

mindspore/ops/operations/_grad_ops.py

@@ -1004,7 +1004,7 @@ class MaxPool3DGrad(PrimitiveWithInfer):
     def infer_dtype(self, x_dtype, y_dtype, grad_dtype):
         args = {'x_dtype': x_dtype, 'y_dtype': y_dtype, 'grad_dtype': grad_dtype}
         validator.check_tensors_dtypes_same_and_valid(args, [mstype.float16, mstype.float32], self.name)
-        return mstype.tensor_type(mstype.float32)
+        return x_dtype
 
 
 class MaxPool3DGradGrad(PrimitiveWithInfer):

mindspore/ops/operations/nn_ops.py

@@ -7277,6 +7277,9 @@ class DynamicRNN(PrimitiveWithInfer):
     are learnable weights between the output and the input in the formula. For instance,
     :math:`W_{ix}, b_{ix}` are the weight and bias used to transform from input :math:`x` to :math:`i`.
 
+    Note:
+        The `hidden_size` in shape of inputs must be multiple of 16.
+
     Args:
         cell_type (str): A string identifying the cell type in the op. Default: 'LSTM'.
             Only 'LSTM' is currently supported.
@@ -7785,8 +7788,8 @@ class AvgPool3D(Primitive):
         >>> avg_pool3d = ops.AvgPool3D(kernel_size=2, strides=1, pad_mode="valid")
         >>> output = avg_pool3d(input)
         >>> print(output)
-        [[[[[233.5 248.625]]]
-          [[[233.5 238.125]]]]]
+        [[[[[5. 6.]]]
+          [[[17. 18.]]]]]
     """
     @prim_attr_register
     def __init__(self, kernel_size=1, strides=1, pad_mode="valid", pad=0, ceil_mode=False,