!15227 add parallel operator for gathernd

From: @yangzhenzhang
Reviewed-by: @kisnwang,@stsuteng
Signed-off-by: @stsuteng

mindspore/ccsrc/frontend/parallel/auto_parallel/operator_costmodel.h

@@ -608,6 +608,7 @@ using GreaterCost = SubCost;
 using GreaterEqualCost = SubCost;
 using LessCost = SubCost;
 using LessEqualCost = SubCost;
+using GatherNdCost = SubCost;
 
 class MulCost : public SubCost {
  public:

mindspore/ccsrc/frontend/parallel/dynamic_creator.h

@@ -191,6 +191,7 @@ REGISTER(StackInfo);
 REGISTER(ConcatInfo);
 REGISTER(SplitInfo);
 REGISTER(UniqueInfo);
+REGISTER(GatherNdInfo);
 }  // namespace parallel
 }  // namespace mindspore
 

mindspore/ccsrc/frontend/parallel/ops_info/gathernd_info.cc

@@ -0,0 +1,214 @@
+/**
+ * Copyright 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.
+ * 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 "frontend/parallel/ops_info/gathernd_info.h"
+
+#include <algorithm>
+#include <memory>
+#include <utility>
+#include <vector>
+#include <functional>
+#include <string>
+
+#include "frontend/parallel/device_matrix.h"
+#include "frontend/parallel/strategy.h"
+#include "frontend/parallel/tensor_layout/tensor_redistribution.h"
+#include "pipeline/jit/resource.h"
+
+namespace mindspore {
+namespace parallel {
+// the input can not be split, and the last dimension of indices can not be split
+Status GatherNdInfo::CheckStrategy(const StrategyPtr &strategy) {
+  MS_EXCEPTION_IF_NULL(strategy);
+  if (CheckStrategyValue(strategy, inputs_shape_) != SUCCESS) {
+    MS_LOG(ERROR) << name_ << ": Invalid strategy";
+    return FAILED;
+  }
+
+  std::vector<Dimensions> stra = strategy->GetInputDim();
+  if (stra.size() != 2) {
+    MS_LOG(ERROR) << name_ << ": The size of strategies must be 2";
+    return FAILED;
+  }
+
+  int64_t input_split_size = std::accumulate(stra[0].begin(), stra[0].end(), 1, std::multiplies<int64_t>());
+  if (input_split_size != 1) {
+    MS_LOG(ERROR) << name_ << ": The input can not be split";
+    return FAILED;
+  }
+
+  if (stra[1].empty()) {
+    MS_LOG(ERROR) << name_ << ": The strategy of indices can not be empty";
+    return FAILED;
+  }
+
+  if (stra[1].back() != 1) {
+    MS_LOG(ERROR) << name_ << ": The last dimension of indices can not be split";
+    return FAILED;
+  }
+
+  return SUCCESS;
+}
+
+// the dev matrix is indices_strategy
+Status GatherNdInfo::InferDevMatrixShape() {
+  MS_EXCEPTION_IF_NULL(strategy_);
+  std::vector<Dimensions> stra = strategy_->GetInputDim();
+  if (stra.size() != 2) {
+    MS_LOG(ERROR) << name_ << "The size of strategies must be 2";
+    return FAILED;
+  }
+
+  dev_matrix_shape_ = stra[1];
+  return SUCCESS;
+}
+
+// input shape: [x, y, z], indices shape: [a, b, c, 2], output shape: [a, b, c, z]
+// strategy: ((1, 1, 1), (m, n, o, 1))
+// dev-matrix: [m, n, o, 1]
+// input map: [-1, -1, -1], indices map: [3, 2, 1, 0], output map: [3, 2, 1, -1]
+Status GatherNdInfo::InferTensorMap() {
+  if (inputs_shape_.size() != 2) {
+    MS_LOG(ERROR) << name_ << "The size of input shapes must be 2";
+    return FAILED;
+  }
+
+  if (outputs_shape_.empty() || outputs_shape_[0].size() < (inputs_shape_[1].size() - 1)) {
+    MS_LOG(ERROR) << name_ << "invalid shapes";
+    return FAILED;
+  }
+
+  TensorMap input_tensor_map(inputs_shape_[0].size(), MAP_NONE);  // the input can not split
+
+  // cannot use dev_matrix_shape_ replace inputs_shape_[0], because it may not be fully split in all devices.
+  TensorMap indices_tensor_map;
+  int64_t size = SizeToLong(inputs_shape_[0].size());
+  for (int64_t i = 0; i < size; ++i) {
+    indices_tensor_map.push_back(size - i - 1);
+  }
+
+  TensorMap output_tensor_map(outputs_shape_[0].size(), MAP_NONE);
+  for (size_t i = 0; i < (inputs_shape_[1].size() - 1); ++i) {
+    output_tensor_map[i] = indices_tensor_map[i];
+  }
+
+  inputs_tensor_map_.push_back(input_tensor_map);
+  inputs_tensor_map_.push_back(indices_tensor_map);
+  outputs_tensor_map_.push_back(output_tensor_map);
+  return SUCCESS;
+}
+
+Status GatherNdInfo::InferTensorInfo() {
+  if (inputs_shape_.empty() || outputs_shape_.empty() || inputs_tensor_map_.empty() || outputs_tensor_map_.empty()) {
+    MS_LOG(ERROR) << name_ << ": Invalid args";
+    return FAILED;
+  }
+
+  TensorLayout input_layout, output_layout;
+  for (size_t i = 0; i < inputs_shape_.size(); ++i) {
+    // infer tensor layout
+    if (input_layout.InitFromVector(dev_matrix_shape_, inputs_tensor_map_[i], inputs_shape_[i]) != SUCCESS) {
+      MS_LOG(ERROR) << name_ << ": Infer input tensor layout failed.";
+      return FAILED;
+    }
+    TensorInfo input_tensor_info(input_layout);
+    inputs_tensor_info_.push_back(input_tensor_info);
+  }
+
+  if (output_layout.InitFromVector(dev_matrix_shape_, outputs_tensor_map_[0], outputs_shape_[0]) != SUCCESS) {
+    MS_LOG(ERROR) << name_ << ": Infer output tensor layout failed.";
+    return FAILED;
+  }
+  TensorInfo output_tensor_info(output_layout);
+  outputs_tensor_info_.push_back(output_tensor_info);
+  return SUCCESS;
+}
+
+void GatherNdInfo::ReComputeBatchSplitFlagList() {
+  split_flag_list_[0] = false;
+  split_flag_list_[1] = true;
+}
+
+Status GatherNdInfo::SetCostUnderStrategy(const StrategyPtr &strategy) { return SetCostUnderStrategyBase(strategy); }
+
+Status GatherNdInfo::GenerateStrategies(int64_t stage_id) {
+  if (InferAttrs() != SUCCESS) {
+    MS_LOG(ERROR) << name_ << ": Infer attrs failed";
+    return FAILED;
+  }
+  if (inputs_shape_.empty()) {
+    MS_LOG(ERROR) << name_ << ": The inputs shape is empty";
+    return FAILED;
+  }
+
+  // to generate the indices' strategy
+  Shape input_split(inputs_shape_[1].size(), 1);
+  input_split.back() = 0;
+  Shapes splittable_input = {input_split};
+  Shapes tmp_inputs_shape = {inputs_shape_[1]};
+
+  std::vector<StrategyPtr> sp_vector;
+  if (GenerateStrategiesForIndependentInputs(stage_id, tmp_inputs_shape, splittable_input, &sp_vector) != SUCCESS) {
+    MS_LOG(ERROR) << name_ << ": Generate strategies failed";
+    return FAILED;
+  }
+
+  // the others strategies are equal to the first input's strategy
+  for (auto &sp : sp_vector) {
+    if ((sp == nullptr) || sp->GetInputDim().empty()) {
+      MS_LOG(ERROR) << name_ << ": The strategy is null or empty";
+      return FAILED;
+    }
+    Strategys tmp_strategy;
+    Dimensions indices_strategy = sp->GetInputDim()[0];
+    Dimensions input_strategy(inputs_shape_[0].size(), 1);
+    tmp_strategy.push_back(input_strategy);
+    tmp_strategy.push_back(indices_strategy);
+    sp->ResetInputs(tmp_strategy);
+  }
+
+  size_t success = 0;
+  for (auto &sp : sp_vector) {
+    PrintStrategy(sp);
+    if (SetCostUnderStrategy(sp) == SUCCESS) {
+      success++;
+      MS_LOG(INFO) << name_ << ": Successfully generated " << success << " strategy.";
+      PrintStrategy(sp);
+    }
+  }
+  return SUCCESS;
+}
+
+Status GatherNdInfo::Init(const StrategyPtr &strategy) {
+  if (InitWithAutoRepeatCalc(strategy) != SUCCESS) {
+    MS_LOG(ERROR) << name_ << ": Init failed.";
+    return FAILED;
+  }
+  MS_LOG(INFO) << name_ << ": Init success.";
+  return SUCCESS;
+}
+
+Status GatherNdInfo::InitForCostModel(const StrategyPtr &strategy) {
+  if (InitForCostModelWithAutoRepeatCalc(strategy) != SUCCESS) {
+    MS_LOG(ERROR) << name_ << ": Init for cost model failed.";
+    return FAILED;
+  }
+
+  MS_LOG(INFO) << name_ << ": Init for cost model success.";
+  return SUCCESS;
+}
+}  // namespace parallel
+}  // namespace mindspore

mindspore/ccsrc/frontend/parallel/ops_info/gathernd_info.h

@@ -0,0 +1,58 @@
+/**
+ * Copyright 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.
+ * 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_FRONTEND_PARALLEL_OPS_INFO_GATHERND_INFO_H_
+#define MINDSPORE_CCSRC_FRONTEND_PARALLEL_OPS_INFO_GATHERND_INFO_H_
+
+#include <string>
+#include <memory>
+#include <unordered_map>
+#include <vector>
+
+#include "ir/value.h"
+#include "frontend/parallel/auto_parallel/operator_costmodel.h"
+#include "frontend/parallel/ops_info/operator_info.h"
+#include "frontend/parallel/strategy.h"
+
+namespace mindspore {
+namespace parallel {
+class GatherNdInfo : public OperatorInfo {
+ public:
+  GatherNdInfo(const std::string &operator_name, const Shapes &inputs_shape, const Shapes &outputs_shape,
+               const PrimitiveAttrs &attrs)
+      : OperatorInfo(operator_name, inputs_shape, outputs_shape, attrs, std::make_shared<GatherNdCost>()) {}
+  ~GatherNdInfo() override = default;
+
+  Status Init(const StrategyPtr &strategy) override;
+  Status InitForCostModel(const StrategyPtr &strategy) override;
+  Status GenerateStrategies(int64_t) override;
+  Status SetCostUnderStrategy(const StrategyPtr &) override;
+  void ReComputeBatchSplitFlagList() override;
+
+ protected:
+  Status GetAttrs() override { return SUCCESS; }
+  Status CheckStrategy(const StrategyPtr &strategy) override;
+  Status InferForwardCommunication() override { return SUCCESS; }
+  Status InferTensorInfo() override;
+  Status InferDevMatrixShape() override;
+  Status InferTensorMap() override;
+};
+
+using GatherNdInfoPtr = std::shared_ptr<GatherNdInfo>;
+}  // namespace parallel
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_FRONTEND_PARALLEL_OPS_INFO_GATHERND_INFO_H_

mindspore/ccsrc/frontend/parallel/ops_info/ops_info_head_files.h

@@ -50,5 +50,6 @@
 #include "frontend/parallel/ops_info/unique_info.h"
 #include "frontend/parallel/ops_info/uniform_candidate_sampler_info.h"
 #include "frontend/parallel/ops_info/reluv2_info.h"
+#include "frontend/parallel/ops_info/gathernd_info.h"
 
 #endif  // MINDSPORE_CCSRC_FRONTEND_PARALLEL_OPS_INFO_HEAD_FILES_H_

mindspore/ccsrc/frontend/parallel/ops_info/ops_utils.h

@@ -325,6 +325,7 @@ constexpr char DEPTHWISE_CONV2D[] = "DepthwiseConv2D";
 constexpr char DROPOUT[] = "Dropout";
 constexpr char KStridedSlice[] = "StridedSlice";
 constexpr char UNIQUE[] = "Unique";
+constexpr char GATHERND[] = "GatherNd";
 
 // Parallel don't care
 constexpr char STRING_EQUAL[] = "string_equal";

mindspore/ccsrc/frontend/parallel/step_auto_parallel.cc

@@ -163,7 +163,8 @@ bool IsSplittableOperator(const std::string &op_name) {
      BESSELI0E, BESSELI1E, FLOORMOD, ASSIGN, ASSIGN_ADD, ATAN2, DIVNONAN, LOGICALAND, LOGICALOR, ELU, RELU6, RELUV2,
      SOFTPLUS, SOFTSIGN, GREATEREQUAL, LESSEQUAL, LESS, APPROXIMATEEQUAL, MOD, UNIQUE, UNSORTED_SEGMENT_SUM,
      UNSORTED_SEGMENT_MIN, REPEAT_ELEMENTS, TENSOR_DOT, RANGE, UNIFORM_CANDIDATE_SAMPLER, SLICE,
-     UNSORTED_SEGMENT_MAX};
+     UNSORTED_SEGMENT_MAX, GATHER_ND};
+
   // clang-format on
 
   auto iter = splittable_op.find(op_name);
@@ -492,10 +493,9 @@ Status ConstructCostGraphNodesByUniqueIdTC(const std::vector<AnfNodePtr> &all_no
   std::map<size_t, size_t> loop_to_ops;
   // extract strategy from checkpoint for multi-train
   StrategyMap stra_map;
-  if (StrategyCheckpoint::GetInstance().LoadCheckPointOn()) {
-    if (StrategyCheckpoint::GetInstance().Load(&stra_map) != SUCCESS) {
-      MS_LOG(EXCEPTION) << "Load strategy checkpoint failed";
-    }
+  if (StrategyCheckpoint::GetInstance().LoadCheckPointOn() &&
+      StrategyCheckpoint::GetInstance().Load(&stra_map) != SUCCESS) {
+    MS_LOG(EXCEPTION) << "Load strategy checkpoint failed";
   }
   std::vector<std::string> last_forward_node_ids;
   if (!root->has_flag(TRAINING)) {
@@ -505,8 +505,7 @@ Status ConstructCostGraphNodesByUniqueIdTC(const std::vector<AnfNodePtr> &all_no
   for (auto &node : all_nodes) {
     // NOTE: we only care about splittable Primitive operators
     auto cnode = node->cast<CNodePtr>();
-    bool bool_result = (cnode == nullptr) || (!IsValueNode<Primitive>(cnode->input(0)));
-    if (bool_result) {
+    if ((cnode == nullptr) || (!IsValueNode<Primitive>(cnode->input(0)))) {
       continue;
     }
     ValueNodePtr prim_anf_node = cnode->input(0)->cast<ValueNodePtr>();
@@ -551,9 +550,8 @@ Status ConstructCostGraphNodesByUniqueIdTC(const std::vector<AnfNodePtr> &all_no
       bool is_last_nodes = std::find(last_forward_node_ids.begin(), last_forward_node_ids.end(), cnode->UniqueId()) !=
                            last_forward_node_ids.end();
       auto operator_info = CreateTheOperatorInfo(prim, cnode, is_last_nodes, &stra_map);
-      if (operator_info == nullptr) {
-        return FAILED;
-      }
+      MS_EXCEPTION_IF_NULL(operator_info);
+
       // Needed by rec_parser
       operator_info->set_type(prim->name());
       operator_info->set_last_node_flag(is_last_nodes);
@@ -627,8 +625,7 @@ void ConstructCostGraphEdges(const std::vector<AnfNodePtr> &all_nodes) {
   MS_LOG(INFO) << "Constructing edges for cost graph begins.";
   for (auto &node : all_nodes) {
     auto cnode = node->cast<CNodePtr>();
-    bool bool_result_cnode = (cnode == nullptr) || !IsValueNode<Primitive>(cnode->input(0));
-    if (bool_result_cnode) {
+    if ((cnode == nullptr) || !IsValueNode<Primitive>(cnode->input(0))) {
       continue;
     }
     auto &inputs = cnode->inputs();
@@ -638,7 +635,6 @@ void ConstructCostGraphEdges(const std::vector<AnfNodePtr> &all_nodes) {
     }
     PrimitivePtr prim = GetValueNode<PrimitivePtr>(prim_anf_node);
     size_t edge_count = 0;
-
     auto node_op_info = cnode->user_data<OperatorInfo>();
 
     for (size_t i = 1; i < inputs.size(); ++i) {

tests/ut/python/parallel/test_gathernd.py

@@ -0,0 +1,110 @@
+# Copyright 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.
+# 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.
+# ============================================================================
+import numpy as np
+import pytest
+
+import mindspore as ms
+from mindspore import context, Tensor, Parameter
+from mindspore.nn import Cell, Momentum
+from mindspore.ops import operations as P
+from mindspore.train import Model
+from tests.dataset_mock import MindData
+
+
+class Dataset(MindData):
+    def __init__(self, predict, label, length=3):
+        super(Dataset, self).__init__(size=length)
+        self.predict = predict
+        self.label = label
+        self.index = 0
+        self.length = length
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        if self.index >= self.length:
+            raise StopIteration
+        self.index += 1
+        return self.predict, self.label
+
+    def reset(self):
+        self.index = 0
+
+
+class Net(Cell):
+    def __init__(self, w1, strategy1=None, strategy2=None):
+        super().__init__()
+        self.mul = P.Mul().shard(strategy1)
+        self.w1 = Parameter(w1, "w1")
+        self.indices = Tensor(np.ones([16, 2]), dtype=ms.int32)
+        self.gathernd = P.GatherNd().shard(strategy2)
+
+    def construct(self, x, b):
+        out = self.mul(x, self.w1)
+        out = self.gathernd(out, self.indices)
+        return out
+
+
+_x = Tensor(np.ones([16, 64]), dtype=ms.float32)
+_b = Tensor(np.ones([16, 64]), dtype=ms.float32)
+_w1 = Tensor(np.ones([128, 64]), dtype=ms.float32)
+
+
+def compile_net(net):
+    context.set_context(save_graphs=True)
+    learning_rate = 0.1
+    momentum = 0.9
+    epoch_size = 2
+    dataset = Dataset(_x, _b)
+    opt = Momentum(net.trainable_params(), learning_rate, momentum)
+    model = Model(net, optimizer=opt)
+    model.train(epoch_size, dataset, dataset_sink_mode=False)
+    context.reset_auto_parallel_context()
+
+
+def test_gathernd_data_parallel():
+    context.set_auto_parallel_context(
+        parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((8, 1), (8, 1))
+    strategy2 = ((1, 1), (8, 1))
+    net = Net(_w1, strategy1, strategy2)
+    compile_net(net)
+
+
+def test_gathernd_model_parallel():
+    context.set_auto_parallel_context(
+        parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((2, 4), (2, 4))
+    strategy2 = ((1, 1), (4, 1))
+    net = Net(_w1, strategy1, strategy2)
+    compile_net(net)
+
+
+def test_gathernd_auto_parallel():
+    context.set_auto_parallel_context(
+        parallel_mode="auto_parallel", device_num=8, global_rank=0)
+    net = Net(_w1)
+    compile_net(net)
+
+
+def test_gathernd_strategy_error():
+    context.set_auto_parallel_context(
+        parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((8, 1), (8, 1))
+    strategy2 = ((1, 1), (2, 4))
+    net = Net(_w1, strategy1, strategy2)
+    with pytest.raises(RuntimeError):
+        compile_net(net)