!8431 Add more check in unsorted_segment_sum_fission pass to avoid doing wrong optimize which cause runtime error

From: @irmo
Reviewed-by: 
Signed-off-by:

mindspore/ccsrc/backend/optimizer/ascend/ir_fission/unsorted_segment_sum_fission.cc

@@ -74,6 +74,26 @@ CNodePtr CreateSlice(const FuncGraphPtr &graph, const CNodePtr &unsort_segment_s
   AnfAlgo::SetNodeAttr(kAttrSize, MakeValue(Convert2Long(unsort_segment_sum_shape)), slice);
   return slice;
 }
+
+bool CheckInputs(const CNodePtr &origin_node) {
+  MS_EXCEPTION_IF_NULL(origin_node);
+  if (origin_node->size() != kUnsortedSegmentSumInputNum + 1) {
+    MS_LOG(DEBUG) << "UnsortedSegmentSum has wrong inputs num, not equal " << kUnsortedSegmentSumInputNum
+                  << ". CNode= " << origin_node->DebugString();
+    return false;
+  }
+  auto x_shape = AnfAlgo::GetPrevNodeOutputInferShape(origin_node, 0);
+  auto y_shape = AnfAlgo::GetPrevNodeOutputInferShape(origin_node, 1);
+  if (x_shape.empty() || y_shape.empty()) {
+    return false;
+  }
+  if (x_shape[x_shape.size() - 1] != 1) {
+    MS_LOG(DEBUG) << "UnsortedSegmentSum is not need fission. The last value of input0's shape is "
+                  << x_shape[x_shape.size() - 1];
+    return false;
+  }
+  return x_shape.size() > y_shape.size();
+}
 }  // namespace
 
 const BaseRef UnsortSegmentSumFission::DefinePattern() const {
@@ -88,19 +108,7 @@ const AnfNodePtr UnsortSegmentSumFission::Process(const FuncGraphPtr &graph, con
   MS_EXCEPTION_IF_NULL(node);
   auto origin_node = node->cast<CNodePtr>();
   MS_EXCEPTION_IF_NULL(origin_node);
-  if (origin_node->size() != kUnsortedSegmentSumInputNum + 1) {
-    MS_LOG(INFO) << "UnsortedSegmentSum has wrong inputs num, not equal " << kUnsortedSegmentSumInputNum
-                 << ". CNode= " << origin_node->DebugString();
-    return nullptr;
-  }
-  auto input0_shape = AnfAlgo::GetPrevNodeOutputInferShape(origin_node, 0);
-  if (input0_shape.size() < 2) {
-    MS_LOG(INFO) << "Input0's shape size less than 2, not optimize";
-    return nullptr;
-  }
-  if (input0_shape[input0_shape.size() - 1] != 1) {
-    MS_LOG(INFO) << "UnsortedSegmentSum is not need fission. The last value of input0's shape is "
-                 << input0_shape[input0_shape.size() - 1];
+  if (!CheckInputs(origin_node)) {
     return nullptr;
   }
   size_t pad_dim_size;
@@ -110,7 +118,7 @@ const AnfNodePtr UnsortSegmentSumFission::Process(const FuncGraphPtr &graph, con
   } else if (input_dtype == kNumberTypeFloat16) {
     pad_dim_size = 16;
   } else {
-    MS_LOG(INFO) << "UnsortedSegmentSum data type not in (float21, float16), no need change";
+    MS_LOG(DEBUG) << "UnsortedSegmentSum data type not in (float32, float16), no need change";
     return nullptr;
   }
 

tests/ut/cpp/pre_activate/ascend/ir_fission/unsorted_segment_sum_fission_test.cc

@@ -32,9 +32,11 @@ class TestHWUnsortedSegmentSumFission : public BackendCommon {
 TEST_F(TestHWUnsortedSegmentSumFission, test_fission) {
   FuncGraphPtr g = get_py_fun_.CallAndParseRet("test_unsorted_segment_sum_fission", "before1");
   EXPECT_NE(g, nullptr);
-  std::vector<int64_t> shp_x{16, 1};
+  std::vector<int64_t> shp_x{3, 39, 1};
+  std::vector<int64_t> shp_y{3};
   auto x_abstract = std::make_shared<abstract::AbstractTensor>(kFloat32, shp_x);
-  AbstractBasePtrList args_spec_list{x_abstract, x_abstract};
+  auto y_abstract = std::make_shared<abstract::AbstractTensor>(kInt32, shp_y);
+  AbstractBasePtrList args_spec_list{x_abstract, y_abstract};
   auto kg = GetKernelGraph(g, args_spec_list);
 
   auto optimizer = std::make_shared<opt::GraphOptimizer>();
@@ -50,9 +52,11 @@ TEST_F(TestHWUnsortedSegmentSumFission, test_fission) {
 TEST_F(TestHWUnsortedSegmentSumFission, test_no_fission) {
   FuncGraphPtr g = get_py_fun_.CallAndParseRet("test_unsorted_segment_sum_fission", "before2");
   EXPECT_NE(g, nullptr);
-  std::vector<int64_t> shp_x{16, 2};
+  std::vector<int64_t> shp_x{3, 39, 2};
+  std::vector<int64_t> shp_y{3};
   auto x_abstract = std::make_shared<abstract::AbstractTensor>(kFloat32, shp_x);
-  AbstractBasePtrList args_spec_list{x_abstract, x_abstract};
+  auto y_abstract = std::make_shared<abstract::AbstractTensor>(kInt32, shp_y);
+  AbstractBasePtrList args_spec_list{x_abstract, y_abstract};
   auto kg = GetKernelGraph(g, args_spec_list);
 
   auto optimizer = std::make_shared<opt::GraphOptimizer>();