diff --git a/mindspore/lite/src/runtime/kernel/arm/int8/sub_int8.cc b/mindspore/lite/src/runtime/kernel/arm/int8/sub_int8.cc
index 886aa14f2ed..fffceda0a15 100644
--- a/mindspore/lite/src/runtime/kernel/arm/int8/sub_int8.cc
+++ b/mindspore/lite/src/runtime/kernel/arm/int8/sub_int8.cc
@@ -138,11 +138,28 @@ int SubInt8Run(void *cdata, int task_id, float, float) {
 int SubInt8CPUKernel::Run() {
   if (broadcast_) {
     ArithmeticParameter tile_para;
-    tile_para.ndim_ = out_tensors_.at(0)->shape().size();
-    for (size_t i = 0; i < tile_para.ndim_; i++) {
-      tile_para.in_shape0_[i] = in_tensors_.at(0)->DimensionSize(i);
-      tile_para.in_shape1_[i] = in_tensors_.at(1)->DimensionSize(i);
-      tile_para.out_shape_[i] = out_tensors_.at(0)->DimensionSize(i);
+    auto out_shape = out_tensors_[FIRST_INPUT]->shape();
+    tile_para.ndim_ = out_shape.size();
+    auto in_shape0 = in_tensors_[FIRST_INPUT]->shape();
+    MS_CHECK_TRUE_MSG(out_shape.size() >= in_shape0.size(), RET_ERROR,
+                      "Sub first-input shape size is larger than out.");
+    for (size_t i = 0; i < out_shape.size() - in_shape0.size(); ++i) {
+      tile_para.in_shape0_[i] = 1;
+    }
+    for (size_t i = 0; i < in_shape0.size(); ++i) {
+      tile_para.in_shape0_[i + out_shape.size() - in_shape0.size()] = in_shape0[i];
+    }
+    auto in_shape1 = in_tensors_[SECOND_INPUT]->shape();
+    MS_CHECK_TRUE_MSG(out_shape.size() >= in_shape1.size(), RET_ERROR,
+                      "Sub second-input shape size is larger than out.");
+    for (size_t i = 0; i < out_shape.size() - in_shape1.size(); ++i) {
+      tile_para.in_shape1_[i] = 1;
+    }
+    for (size_t i = 0; i < in_shape1.size(); ++i) {
+      tile_para.in_shape1_[i + out_shape.size() - in_shape1.size()] = in_shape1[i];
+    }
+    for (size_t i = 0; i < out_shape.size(); ++i) {
+      tile_para.out_shape_[i] = out_shape[i];
     }
     tile0_data_ = static_cast<int8_t *>(ms_context_->allocator->Malloc(out_tensors_.at(0)->Size()));
     if (tile0_data_ == nullptr) {