!13316 【GraphKernel】Normalize the Reduce nodes' axis

From: @dayschan
Reviewed-by: @gaoxiong1,@dylangeng
Signed-off-by: @gaoxiong1

mindspore/_extends/graph_kernel/expanders/reduce_mean.py

@@ -27,8 +27,9 @@ class ReduceMean(Expander):
         axis = self.attrs['axis']
         keep_dims = self.attrs['keep_dims']
 
-        # cal reduce_mean, when axis is None, reduce all axes.
-        if not axis:
+        if not isinstance(axis, (tuple, list)):
+            axis = (axis,)
+        elif not axis:
             axis = list(range(len(x.shape)))
         reduce_size = 1.0
         for idx in axis:

mindspore/ccsrc/backend/optimizer/graph_kernel/axis_normalizer.cc

@@ -0,0 +1,93 @@
+/**
+ * Copyright 2020 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.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "backend/optimizer/graph_kernel/axis_normalizer.h"
+
+#include <algorithm>
+#include <vector>
+#include "ir/scalar.h"
+#include "backend/optimizer/graph_kernel/graph_kernel_helper.h"
+#include "backend/session/anf_runtime_algorithm.h"
+
+namespace mindspore {
+namespace opt {
+int64_t AxisNormalizer::NormAxis(int64_t x, size_t rank) { return x >= 0 ? x : x + static_cast<int64_t>(rank); }
+
+bool AxisNormalizer::IsReduce(const AnfNodePtr &node) {
+  std::vector<PrimitivePtr> node_with_axis = {prim::kPrimReduceSum, prim::kPrimReduceMax, prim::kPrimReduceMin};
+  return std::any_of(node_with_axis.begin(), node_with_axis.end(),
+                     [&node](PrimitivePtr &p) { return IsPrimitiveCNode(node, p); });
+}
+
+bool AxisNormalizer::Process(const FuncGraphPtr &func_graph) {
+  bool changed = false;
+  auto todos = TopoSort(func_graph->get_return());
+  for (auto node : todos) {
+    if (!IsReduce(node)) {
+      continue;
+    }
+    if (auto primitive = GetCNodePrimitive(node); primitive != nullptr && primitive->HasAttr(kAttrAxis)) {
+      auto axis = primitive->GetAttr(kAttrAxis);
+      size_t rank = AnfAlgo::GetInputDeviceShape(node, 0).size();
+      if (rank == 0) {
+        // scalar tensor
+        rank = 1;
+      }
+      bool diff = false;
+      ShapeVector axis_vec;
+      if (axis->isa<Int32Imm>() || axis->isa<Int64Imm>()) {
+        int64_t v1 = GetValue<int64_t>(axis);
+        int64_t v2 = NormAxis(v1, rank);
+        axis_vec.push_back(v2);
+        diff = diff || (v1 != v2);
+      } else if (axis->isa<ValueList>() || axis->isa<ValueTuple>()) {
+        auto vec = axis->isa<ValueList>() ? axis->cast<ValueListPtr>()->value() : axis->cast<ValueTuplePtr>()->value();
+        if (vec.empty()) {
+          diff = true;
+          for (size_t i = 0; i < rank; i++) {
+            axis_vec.push_back(i);
+          }
+        } else if (vec[0]->isa<Int32Imm>() || vec[0]->isa<Int64Imm>()) {
+          for (auto v : vec) {
+            int64_t v1 = GetValue<int64_t>(v);
+            int64_t v2 = NormAxis(v1, rank);
+            axis_vec.push_back(v2);
+            diff = diff || (v1 != v2);
+          }
+        }
+      }
+      if (diff) {
+        changed = true;
+        SetNodeAttrSafely(kAttrAxis, MakeValue(axis_vec), node);
+      }
+    }
+  }
+  return changed;
+}
+
+bool AxisNormalizer::Run(const FuncGraphPtr &func_graph) {
+  MS_EXCEPTION_IF_NULL(func_graph);
+  bool changed = false;
+  auto todos = TopoSort(func_graph->get_return());
+  for (auto node : todos) {
+    if (AnfAlgo::IsGraphKernel(node)) {
+      auto sub_func_graph = AnfAlgo::GetCNodeFuncGraphPtr(node);
+      changed = Process(sub_func_graph) || changed;
+    }
+  }
+  return changed;
+}
+}  // namespace opt
+}  // namespace mindspore

mindspore/ccsrc/backend/optimizer/graph_kernel/axis_normalizer.h

@@ -0,0 +1,38 @@
+/**
+ * Copyright 2020 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.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_CCSRC_BACKEND_OPTIMIZER_GRAPH_KERNEL_AXIS_NORMALIZER_H_
+#define MINDSPORE_CCSRC_BACKEND_OPTIMIZER_GRAPH_KERNEL_AXIS_NORMALIZER_H_
+
+#include "ir/func_graph.h"
+#include "backend/optimizer/common/optimizer.h"
+
+namespace mindspore {
+namespace opt {
+// change Reduce nodes' axis to non-negative value
+class AxisNormalizer : public Pass {
+ public:
+  AxisNormalizer() : Pass("axis_normalizer") {}
+  ~AxisNormalizer() = default;
+  bool Run(const FuncGraphPtr &func_graph) override;
+
+ private:
+  bool Process(const FuncGraphPtr &func_graph);
+  int64_t NormAxis(int64_t x, size_t rank);
+  bool IsReduce(const AnfNodePtr &node);
+};
+}  // namespace opt
+}  // namespace mindspore
+#endif  // MINDSPORE_CCSRC_BACKEND_OPTIMIZER_GRAPH_KERNEL_AXIS_NORMALIZER_H_

mindspore/ccsrc/backend/optimizer/graph_kernel/graph_kernel_optimization.cc

@@ -41,6 +41,7 @@
 #include "backend/optimizer/graph_kernel/split_assign.h"
 #include "backend/optimizer/graph_kernel/reorder_ops.h"
 #include "backend/optimizer/graph_kernel/update_state_formatter.h"
+#include "backend/optimizer/graph_kernel/axis_normalizer.h"
 #include "backend/optimizer/pass/getitem_tuple.h"
 
 namespace mindspore {
@@ -78,6 +79,9 @@ PassManagerPtr GraphKernelOptimizer::Cluster() {
 
 PassManagerPtr GraphKernelOptimizer::HighLevelOpt1() {
   auto pm = std::make_shared<PassManager>("graphkernel_stage3_highlevelopt1");
+  // normalize the Reduce axis
+  pm->AddPass(std::make_shared<AxisNormalizer>());
+
   // Replace Assign with InplaceAssign, and replace original output with overridden parameters
   pm->AddPass(std::make_shared<OptimizeAssign>());
   pm->AddPass(std::make_shared<EliminateRedundantOutput>());