!26985 fix the negative axis problem of reduce_eliminate

Merge pull request !26985 from huangbingjian/reduce_eliminate

mindspore/ccsrc/frontend/optimizer/irpass/reduce_eliminate.h

@@ -1,5 +1,5 @@
 /**
- * Copyright 2020 Huawei Technologies Co., Ltd
+ * Copyright 2020-2021 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -67,14 +67,20 @@ class ReduceOneEliminater : public AnfVisitor {
       }
 
       // {_Reduce, X, axis} -> {Reshape, X, new_shape}
+      size_t x_shape_size = x_shape_.size();
+      std::vector<int64_t> positive_axis;
+      std::transform(axis_.begin(), axis_.end(), std::back_inserter(positive_axis),
+                     [x_shape_size](int64_t idx) { return idx < 0 ? idx + x_shape_size : idx; });
+
       std::vector<ValuePtr> elements;
-      for (size_t i = 0; i < x_shape_.size(); i++) {
-        auto iter = find(axis_.begin(), axis_.end(), i);
-        if (iter == axis_.end()) {
+      for (size_t i = 0; i < x_shape_size; i++) {
+        auto iter = find(positive_axis.begin(), positive_axis.end(), i);
+        if (iter == positive_axis.end()) {
           ValuePtr s = MakeValue(x_shape_[i]);
           elements.push_back(s);
         }
       }
+
       auto new_shape = std::make_shared<ValueTuple>(elements);
       auto reshape_op = prim::GetPythonOps("reshape", "mindspore.ops.functional")->cast<PrimitivePtr>();
       auto node_abstract = node->abstract();

tests/st/ops/gpu/test_reduce_mean_op.py

@@ -19,6 +19,7 @@ import pytest
 import mindspore.context as context
 import mindspore.nn as nn
 from mindspore import Tensor
+from mindspore import dtype as mstype
 from mindspore.ops import operations as P
 from mindspore.ops.operations import _inner_ops as inner
 
@@ -100,3 +101,30 @@ def test_dynamic_reduce_mean(dtype, shape, axis, keep_dims):
     error = np.ones(shape=expect.shape) * 1.0e-5
     assert np.all(diff < error)
     assert output.shape == expect.shape
+
+
+class ReduceMeanNegativeNet(nn.Cell):
+    def __init__(self):
+        super().__init__()
+        self.mean0 = P.ReduceMean(True)
+        self.mean1 = P.ReduceMean(False)
+
+    def construct(self, x):
+        t = self.mean0(x, ())
+        return self.mean1(t, (-1,))
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_reduce_mean_negative():
+    """
+    Feature: ALL To ALL
+    Description: test cases for ReduceMean with negative axis.
+    Expectation: the result match expectation
+    """
+    context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
+    x = Tensor([[[1, 2, 3,], [3, 2, 1]]], mstype.float32)
+    net = ReduceMeanNegativeNet()
+    out = net(x)
+    assert out.shape == (1, 1)