!48719 ge add dtype attr for tensorshape, dynamicshape and shape op

Merge pull request !48719 from TuDouNi/ge

mindspore/ccsrc/plugin/device/ascend/optimizer/ge/tensorshape_for_ge.cc

@@ -0,0 +1,63 @@
+/**
+ * Copyright 2023 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 "plugin/device/ascend/optimizer/ge/tensorshape_for_ge.h"
+#include <vector>
+#include <memory>
+#include "include/common/utils/anfalgo.h"
+#include "transform/graph_ir/transform_util.h"
+
+namespace mindspore {
+namespace opt {
+namespace {
+constexpr char kDtypeAttrName[] = "dtype";
+}  // namespace
+
+const BaseRef TensorShapeForGE::DefinePattern() const {
+  VarPtr V = std::make_shared<Var>();
+  VarPtr Xs = std::make_shared<SeqVar>();
+  return VectorRef({V, Xs});
+}
+
+// Set the attr dtype and convert it to ge_dtype
+const AnfNodePtr TensorShapeForGE::Process(const FuncGraphPtr &graph, const AnfNodePtr &node, const EquivPtr &) const {
+  MS_EXCEPTION_IF_NULL(graph);
+  MS_EXCEPTION_IF_NULL(node);
+  auto cnode = node->cast<CNodePtr>();
+  MS_EXCEPTION_IF_NULL(cnode);
+
+  static const PrimitiveSet need_dtype_attr_nodes = {prim::kPrimShape, prim::kPrimTensorShape, prim::kPrimDynamicShape};
+  PrimitivePtr prim = common::AnfAlgo::GetCNodePrimitive(cnode);
+  MS_EXCEPTION_IF_NULL(prim);
+  if (need_dtype_attr_nodes.find(prim) == need_dtype_attr_nodes.end()) {
+    return nullptr;
+  }
+
+  if (common::AnfAlgo::HasNodeAttr(kDtypeAttrName, cnode)) {
+    return nullptr;
+  }
+
+  // get output dtype (ms_dtype)
+  TypeId output_dtype = common::AnfAlgo::GetOutputInferDataType(cnode, 0);
+  // convert to ge_dtype
+  int64_t ge_dtype = static_cast<int64_t>(transform::TransformUtil::ConvertDataType(output_dtype));
+  // update/set attr
+  common::AnfAlgo::SetNodeAttr(kDtypeAttrName, MakeValue(ge_dtype), cnode);
+
+  return node;
+}
+}  // namespace opt
+}  // namespace mindspore

mindspore/ccsrc/plugin/device/ascend/optimizer/ge/tensorshape_for_ge.h

@@ -0,0 +1,34 @@
+/**
+ * Copyright 2023 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_ASCEND_GE_OPTIMIZER_IRPASS_TENSORSHAPE_FOR_GE_H_
+#define MINDSPORE_CCSRC_BACKEND_OPTIMIZER_ASCEND_GE_OPTIMIZER_IRPASS_TENSORSHAPE_FOR_GE_H_
+
+#include "backend/common/optimizer/optimizer.h"
+
+namespace mindspore {
+namespace opt {
+class TensorShapeForGE : public PatternProcessPass {
+ public:
+  explicit TensorShapeForGE(bool multigraph = true) : PatternProcessPass("tensorshape_for_ge", multigraph) {}
+  ~TensorShapeForGE() override = default;
+
+  const BaseRef DefinePattern() const override;
+  const AnfNodePtr Process(const FuncGraphPtr &, const AnfNodePtr &, const EquivPtr &) const override;
+};
+}  // namespace opt
+}  // namespace mindspore
+#endif  // MINDSPORE_CCSRC_BACKEND_OPTIMIZER_ASCEND_GE_OPTIMIZER_IRPASS_TENSORSHAPE_FOR_GE_H_

mindspore/ccsrc/plugin/device/ascend/optimizer/ge_optimization.cc

@@ -35,6 +35,7 @@
 #include "plugin/device/ascend/optimizer/ge/sparse_softmax_cross_entropy_with_logits_split.h"
 #include "plugin/device/ascend/optimizer/enhancer/add_placeholder_for_dynamic_gru.h"
 #include "plugin/device/ascend/optimizer/enhancer/add_placeholder_for_dynamic_rnn.h"
+#include "plugin/device/ascend/optimizer/ge/tensorshape_for_ge.h"
 
 namespace mindspore {
 namespace opt {
@@ -73,6 +74,7 @@ void GeOptimization(const FuncGraphPtr &func_graph) {
   pm->AddPass(std::make_shared<opt::ReduceAxisUpdate>());
   pm->AddPass(std::make_shared<opt::InsertPlaceholderForDynamicGRUV2>());
   pm->AddPass(std::make_shared<opt::InsertPlaceholderForDynamicRNN>());
+  pm->AddPass(std::make_shared<opt::TensorShapeForGE>());
   optimizer->AddPassManager(pm);
 
   (void)optimizer->Optimize(func_graph);

mindspore/ccsrc/transform/graph_ir/convert.cc

@@ -83,6 +83,7 @@ const std::map<std::string, std::vector<std::pair<size_t, TypeId>>> kTransInputD
   {kResizeNearestNeighborGradOpName, {{2, kNumberTypeInt32}}},
   {kTileOpName, {{1, kNumberTypeInt32}}},
   {kConv2DBackpropFilterOpName, {{3, kNumberTypeInt32}}},
+  {kConv2DBackpropInputOpName, {{3, kNumberTypeInt32}}},
   {kOneHotOpName, {{2, kNumberTypeInt32}}},
   {kLinSpaceOpName, {{3, kNumberTypeInt32}}}};
 

mindspore/ccsrc/transform/graph_ir/op_declare/array_ops_declare.cc

@@ -48,7 +48,7 @@ REG_ADPT_DESC(Data, kNameParam, ADPT_DESC(Data))
 
 // Shape
 INPUT_MAP(Shape) = {{1, INPUT_DESC(x)}};
-ATTR_MAP(Shape) = EMPTY_ATTR_MAP;
+ATTR_MAP(Shape) = {{"dtype", ATTR_DESC(dtype, AnyTraits<int64_t>())}};
 OUTPUT_MAP(Shape) = {{0, OUTPUT_DESC(y)}};
 REG_ADPT_DESC(Shape, kNameShape, ADPT_DESC(Shape))