fix incorrect in_channel size error messages

mindspore/core/abstract/prim_nn.cc

@@ -333,9 +333,10 @@ AbstractBasePtr InferImplConv2D(const AnalysisEnginePtr &, const PrimitivePtr &p
     w_axis = 2;
   }
   int64_t group = CheckAttrPositiveInt64(op_name, primitive->GetAttr("group"), "group");
-  if ((x_shape[c_axis] != Shape::SHP_ANY) && (x_shape[c_axis] % group != 0)) {
-    MS_LOG(EXCEPTION) << "x_shape[" << c_axis << "] = " << x_shape[c_axis]
-                      << " (channels) must be divisible by group = " << group;
+  if ((x_shape[c_axis] != Shape::SHP_ANY) && (w_shape[c_axis] != Shape::SHP_ANY) &&
+      ((x_shape[c_axis] / group) != w_shape[c_axis])) {
+    MS_LOG(EXCEPTION) << "x_shape[C_in] / group must equal to w_shape[C_in] = " << w_shape[c_axis] << ", but got "
+                      << (x_shape[c_axis] / group);
   }
   int64_t out_channel = CheckAttrPositiveInt64(op_name, primitive->GetAttr("out_channel"), "out_channel");
   if ((w_shape[n_axis] != Shape::SHP_ANY) && (w_shape[n_axis] != out_channel)) {

tests/ut/python/model/test_lenet_core_after_exception.py

@@ -53,5 +53,6 @@ def test_lenet5_exception():
     predict = Tensor(in1)
     label = Tensor(in2)
     net = train_step_with_loss_warp(LeNet5())
-    with pytest.raises(ValueError):
+    with pytest.raises(RuntimeError) as info:
         _executor.compile(net, predict, label)
+    assert "x_shape[C_in] / group must equal to w_shape[C_in] = " in str(info.value)