!3657 Add parallel operator for StridedSlice

Merge pull request !3657 from yangzhenzhang/add-stridedslice-op

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

@@ -170,6 +170,8 @@ class ActivationCost : public OperatorCost {
 using ActivationCostPtr = std::shared_ptr<ActivationCost>;
 using TransposeCost = ActivationCost;
 using TransposeCostPtr = std::shared_ptr<TransposeCost>;
+using StridedSliceCost = ActivationCost;
+using StridedSliceCostPtr = std::shared_ptr<StridedSliceCost>;
 
 class SoftmaxCost : public OperatorCost {
  public:

mindspore/ccsrc/frontend/parallel/dynamic_creator.h

@@ -134,6 +134,7 @@ REGISTER(SquareInfo);
 REGISTER(GatherV2PInfo);
 REGISTER(EmbeddingLookupInfo);
 REGISTER(TileInfo);
+REGISTER(StridedSliceInfo);
 }  // namespace parallel
 }  // namespace mindspore
 

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

@@ -38,5 +38,6 @@
 #include "frontend/parallel/ops_info/virtual_dataset_info.h"
 #include "frontend/parallel/ops_info/gather_v2_p_info.h"
 #include "frontend/parallel/ops_info/tile_info.h"
+#include "frontend/parallel/ops_info/strided_slice_info.h"
 
 #endif  // MINDSPORE_CCSRC_FRONTEND_PARALLEL_OPS_INFO_HEAD_FILES_H_

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

@@ -29,6 +29,11 @@ constexpr int32_t NO_SPLIT_STRATEGY = 1;
 constexpr int32_t SPLIT_FLAG = 1;
 constexpr int32_t NO_SPLIT_FLAG = 0;
 constexpr size_t MATMUL_ATTRS_SIZE = 2;
+constexpr size_t STRIDED_SLICE_ATTRS_SIZE = 5;
+constexpr size_t STRIDED_SLICE_INPUTS_SIZE = 4;
+constexpr size_t STRIDED_SLICE_BEGIN_INDEX = 1;
+constexpr size_t STRIDED_SLICE_END_INDEX = 2;
+constexpr size_t STRIDED_SLICE_STRIDES_INDEX = 3;
 constexpr size_t MATMUL_INPUTS_SIZE = 2;
 constexpr size_t MATMUL_OUTPUTS_SIZE = 1;
 constexpr size_t ACTIVATION_ATTR_SIZE = 1;

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

@@ -0,0 +1,72 @@
+/**
+ * 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_FRONTEND_PARALLEL_OPS_INFO_STRIDED_SLICE_INFO_H_
+#define MINDSPORE_CCSRC_FRONTEND_PARALLEL_OPS_INFO_STRIDED_SLICE_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 StridedSliceInfo : public OperatorInfo {
+ public:
+  StridedSliceInfo(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<StridedSliceCost>(false)) {}
+  ~StridedSliceInfo() override = default;
+
+  Status Init(const StrategyPtr &strategy) override;
+  Status InitForCostModel(const StrategyPtr &strategy) override;
+  Status GenerateStrategies(int32_t) override;
+  Status SetCostUnderStrategy(const StrategyPtr &) override;
+  std::shared_ptr<std::vector<std::vector<int32_t>>> GenerateBatchStrategies() override;
+
+ protected:
+  Status GetAttrs() override;
+  Status CheckStrategy(const StrategyPtr &strategy) override;
+  Status InferMirrorOps() override;
+  Status InferForwardCommunication() override { return SUCCESS; }
+  Status InferTensorInfo() override;
+  Status InferDevMatrixShape() override;
+  Status InferTensorMap() override;
+  Status GetMask(const std::string &mask_name, int32_t *mask_value);
+
+ private:
+  std::vector<int32_t> begin_;
+  std::vector<int32_t> end_;
+  std::vector<int32_t> strides_;
+  int32_t begin_mask_ = 0;
+  int32_t end_mask_ = 0;
+  int32_t ellipsis_mask_ = 0;
+  int32_t new_axis_mask_ = 0;
+  int32_t shrink_axis_mask_ = 0;
+  bool has_mask_ = false;
+};
+
+using StridedSliceInfoPtr = std::shared_ptr<StridedSliceInfo>;
+}  // namespace parallel
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_FRONTEND_PARALLEL_OPS_INFO_STRIDED_SLICE_INFO_H_

tests/ut/python/parallel/test_stridedslice.py

@@ -0,0 +1,164 @@
+# 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.
+# ============================================================================
+import numpy as np
+import pytest
+
+import mindspore as ms
+from mindspore import context, Tensor, Parameter
+from mindspore.common.api import _executor
+from mindspore.nn import Cell, TrainOneStepCell, Momentum
+from mindspore.ops import operations as P
+
+
+class Net(Cell):
+    def __init__(self, weight, w2, begin, end, strides, strategy1=None, strategy2=None, is_parameter=True, mask=0):
+        super().__init__()
+        self.mul = P.Mul().set_strategy(strategy1)
+        self.strided_slice = P.StridedSlice(begin_mask=mask).set_strategy(strategy2)
+        if is_parameter:
+            self.weight = Parameter(weight, "w1")
+        else:
+            self.weight = weight
+        self.mul2 = P.Mul()
+        self.weight2 = Parameter(w2, "w2")
+        self.begin = begin
+        self.end = end
+        self.strides = strides
+
+    def construct(self, x, b):
+        out = self.strided_slice(self.weight, self.begin, self.end, self.strides)
+        out = self.mul(x, out)
+        out = self.mul2(out, self.weight2)
+        return out
+
+
+class Net2(Cell):
+    def __init__(self, weight2, begin, end, strides, strategy1=None, strategy2=None):
+        super().__init__()
+        self.mul = P.Mul().set_strategy(strategy1)
+        self.strided_slice = P.StridedSlice().set_strategy(strategy2)
+        self.weight2 = Parameter(weight2, "w2")
+        self.begin = begin
+        self.end = end
+        self.strides = strides
+
+    def construct(self, x, b):
+        out = self.mul(x, self.weight2)
+        out = self.strided_slice(out, self.begin, self.end, self.strides)
+        return out
+
+
+_x = Tensor(np.ones([128, 64, 1]), dtype=ms.float32)
+_w1 = Tensor(np.ones([256, 64, 32]), dtype=ms.float32)
+_w2 = Tensor(np.ones([128, 64, 1]), dtype=ms.float32)
+_b = Tensor(np.ones([128, 64, 32]), dtype=ms.float32)
+
+
+def compile_net(net):
+    context.set_context(save_graphs=True)
+    optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
+    train_net = TrainOneStepCell(net, optimizer)
+    train_net.set_auto_parallel()
+    _executor.compile(train_net, _x, _b)
+    context.reset_auto_parallel_context()
+
+
+def test_stridedslice_no_fully_fetch_split_error():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((2, 2, 2), (2, 2, 2))
+    strategy2 = ((2, 2, 2),)
+    net = Net(_w1, _w2, (0, 0, 0), (128, 64, 32), (1, 1, 1), strategy1, strategy2, is_parameter=True)
+    with pytest.raises(RuntimeError):
+        compile_net(net)
+
+
+def test_stridedslice_strides_no_1_split_error():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((2, 2, 2), (2, 2, 2))
+    strategy2 = ((1, 2, 2),)
+    net = Net(_w1, _w2, (0, 0, 0), (128, 64, 32), (1, 1, 2), strategy1, strategy2, is_parameter=True)
+    with pytest.raises(RuntimeError):
+        compile_net(net)
+
+
+def test_stridedslice_mask_no_0_split_error():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((2, 2, 2), (2, 2, 2))
+    strategy2 = ((1, 2, 2),)
+    net = Net(_w1, _w2, (0, 0, 0), (128, 64, 32), (1, 1, 1), strategy1, strategy2, is_parameter=True, mask=1)
+    with pytest.raises(RuntimeError):
+        compile_net(net)
+
+
+def test_stridedslice_begin_size_smaller():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 4, 1), (1, 4, 2))
+    strategy2 = ((1, 4, 2),)
+    net = Net(_w1, _w2, (0, 0), (128, 64), (1, 1), strategy1, strategy2, is_parameter=True)
+    compile_net(net)
+
+
+def test_stridedslice_parameter():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 4, 1), (1, 4, 2))
+    strategy2 = ((1, 4, 2),)
+    net = Net(_w1, _w2, (0, 0, 0), (128, 64, 32), (1, 1, 1), strategy1, strategy2, is_parameter=True)
+    compile_net(net)
+
+
+def test_stridedslice_tensor():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 4, 1), (1, 4, 2))
+    strategy2 = ((1, 4, 2),)
+    net = Net(_w1, _w2, (0, 0, 0), (128, 64, 32), (1, 1, 1), strategy1, strategy2, is_parameter=False)
+    compile_net(net)
+
+
+def test_stridedslice_parameter_no_full_split():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 4, 1), (1, 4, 2))
+    strategy2 = ((1, 2, 2),)
+    net = Net(_w1, _w2, (0, 0, 0), (128, 64, 32), (1, 1, 1), strategy1, strategy2, is_parameter=True)
+    compile_net(net)
+
+
+def test_stridedslice_output():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 8, 1), (1, 8, 1))
+    strategy2 = ((1, 8, 1),)
+    net = Net2(_w2, (0, 0, 0), (64, 64, 1), (1, 1, 1), strategy1, strategy2)
+    compile_net(net)
+
+
+def test_stridedslice_output_no_full_split():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 8, 1), (1, 8, 1))
+    strategy2 = ((1, 4, 1),)
+    net = Net2(_w2, (0, 0, 0), (64, 64, 1), (1, 1, 1), strategy1, strategy2)
+    compile_net(net)
+
+
+def test_stridedslice_no_strategy():
+    context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 8, 1), (1, 8, 1))
+    strategy2 = None
+    net = Net2(_w2, (0, 0, 0), (128, 64, 1), (1, 1, 1), strategy1, strategy2)
+    compile_net(net)
+
+
+def test_stridedslice_auto_parallel():
+    context.set_auto_parallel_context(parallel_mode="auto_parallel", device_num=8, global_rank=0)
+    net = Net2(_w2, (0, 0, 0), (32, 64, 1), (1, 1, 1))
+    compile_net(net)