mindspore/ccsrc/backend/optimizer/ascend/ascend_backend_optimization.cc

@@ -64,6 +64,7 @@
 #include "backend/optimizer/ascend/format_type/rectify_do_mask_kernel_info.h"
 #include "backend/optimizer/ascend/format_type/chang_axis_of_reduce_kernel.h"
 #include "backend/optimizer/ascend/format_type/split_unsupported_transdata.h"
+#include "backend/optimizer/ascend/format_type/insert_reshape_for_extract_image_patches_op.h"
 #include "backend/optimizer/pass/getitem_tuple.h"
 #include "backend/optimizer/pass/optimize_dependence.h"
 #include "backend/optimizer/pass/erase_visit_attr.h"
@@ -231,6 +232,7 @@ void AscendMixPrecision(const std::shared_ptr<session::KernelGraph> &kernel_grap
   auto optimizer = std::make_shared<GraphOptimizer>();
   auto mixed_precision_pm = std::make_shared<PassManager>("cast_pm");
   mixed_precision_pm->AddPass(std::make_shared<InsertCast>());
+  mixed_precision_pm->AddPass(std::make_shared<InsertReshapeForExtractImagePatchesOp>());
   mixed_precision_pm->AddPass(std::make_shared<GetitemTuple>());
   mixed_precision_pm->AddPass(std::make_shared<CommonSubexpressionElimination>());
   mixed_precision_pm->AddPass(std::make_shared<EliminateRedundantOp>());

mindspore/ccsrc/backend/optimizer/ascend/format_type/insert_reshape_for_extract_image_patches_op.cc

@@ -0,0 +1,65 @@
+/**
+ * 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/ascend/format_type/insert_reshape_for_extract_image_patches_op.h"
+#include <memory>
+#include "backend/optimizer/ascend/ascend_helper.h"
+#include "backend/session/anf_runtime_algorithm.h"
+#include "utils/utils.h"
+#include "base/core_ops.h"
+
+namespace mindspore {
+namespace opt {
+const BaseRef InsertReshapeForExtractImagePatchesOp::DefinePattern() const {
+  VarPtr Xs = std::make_shared<SeqVar>();
+  return VectorRef({prim::kPrimExtractImagePatches, Xs});
+}
+
+const AnfNodePtr InsertReshapeForExtractImagePatchesOp::Process(const FuncGraphPtr &func_graph, const AnfNodePtr &node,
+                                                                const EquivPtr &equiv) const {
+  MS_EXCEPTION_IF_NULL(func_graph);
+  MS_EXCEPTION_IF_NULL(equiv);
+  auto extract = CheckAnfNodeIfCNodeAndInputSize(node, 2);
+  MS_EXCEPTION_IF_NULL(extract);
+  auto in_node = extract->input(1);
+  MS_EXCEPTION_IF_NULL(in_node);
+  auto extract_kernel_build_info = AnfAlgo::GetSelectKernelBuildInfo(extract);
+  auto in_node_kernel_build_info = AnfAlgo::GetSelectKernelBuildInfo(in_node);
+  MS_EXCEPTION_IF_NULL(extract_kernel_build_info);
+  MS_EXCEPTION_IF_NULL(in_node_kernel_build_info);
+  std::vector<AnfNodePtr> reshape_inputs = {NewValueNode(std::make_shared<Primitive>(prim::kPrimReshape->name())),
+                                            in_node};
+  auto reshape_builder = std::make_shared<kernel::KernelBuildInfo::KernelBuildInfoBuilder>();
+  reshape_builder->SetInputsFormat({kOpFormat_NC1HWC0});
+  reshape_builder->SetOutputsFormat({kOpFormat_NC1HWC0});
+  reshape_builder->SetInputsDeviceType({AnfAlgo::GetOutputDeviceDataType(in_node, 0)});
+  reshape_builder->SetOutputsDeviceType({AnfAlgo::GetOutputDeviceDataType(in_node, 0)});
+  reshape_builder->SetKernelType(in_node_kernel_build_info->kernel_type());
+  reshape_builder->SetFusionType(in_node_kernel_build_info->fusion_type());
+  reshape_builder->SetProcessor(in_node_kernel_build_info->processor());
+
+  auto reshape = func_graph->NewCNode(reshape_inputs);
+  reshape->set_scope(in_node->scope());
+  auto shape_tmp = AnfAlgo::GetOutputInferShape(in_node, 0);
+  AnfAlgo::SetOutputInferTypeAndShape({AnfAlgo::GetOutputDeviceDataType(in_node, 0)},
+                                      {{shape_tmp[0], shape_tmp[2], shape_tmp[3], shape_tmp[1]}}, reshape.get());
+  AnfAlgo::SetSelectKernelBuildInfo(reshape_builder->Build(), reshape.get());
+  AnfAlgo::SetNodeAttr("nop_op", MakeValue(true), reshape);
+  AnfAlgo::SetNodeInput(extract, reshape, 0);
+  return extract;
+}
+}  // namespace opt
+}  // namespace mindspore

mindspore/ccsrc/backend/optimizer/ascend/format_type/insert_reshape_for_extract_image_patches_op.h

@@ -0,0 +1,41 @@
+/**
+ * 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_ASCEND_FORMAT_TYPE_INSERT_RESHAPE_FOR_EXTRACT_IMAGE_PATCHES_OP_H
+#define MINDSPORE_CCSRC_BACKEND_OPTIMIZER_ASCEND_FORMAT_TYPE_INSERT_RESHAPE_FOR_EXTRACT_IMAGE_PATCHES_OP_H
+#include <vector>
+#include <string>
+#include <utility>
+#include <memory>
+#include "ir/anf.h"
+#include "backend/optimizer/common/pattern_engine.h"
+#include "backend/optimizer/common/helper.h"
+#include "backend/optimizer/common/optimizer.h"
+
+namespace mindspore {
+namespace opt {
+class InsertReshapeForExtractImagePatchesOp : public PatternProcessPass {
+ public:
+  explicit InsertReshapeForExtractImagePatchesOp(bool multigraph = true)
+      : PatternProcessPass("insert_reshape_for_extract_image_patches_op", multigraph) {}
+  ~InsertReshapeForExtractImagePatchesOp() 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_FORMAT_TYPE_INSERT_RESHAPE_FOR_EXTRACT_IMAGE_PATCHES_OP_H

mindspore/ccsrc/backend/session/anf_runtime_algorithm.cc

@@ -516,6 +516,10 @@ std::vector<size_t> AnfRuntimeAlgorithm::GetInputDeviceShape(const AnfNodePtr &n
   if (trans::IsNeedPadding(format, infer_shape.size())) {
     infer_shape = trans::PaddingShapeTo4d(infer_shape, GetInputReshapeType(node, input_idx));
   }
+  if (node->isa<CNode>() && GetCNodeName(node) == kExtractImagePatchesOpName) {
+    auto shape_tmp = {infer_shape[0], infer_shape[3], infer_shape[1], infer_shape[2]};
+    return trans::TransShapeToDevice(shape_tmp, format);
+  }
   return trans::TransShapeToDevice(infer_shape, format);
 }
 

mindspore/core/base/core_ops.h

@@ -104,6 +104,7 @@ inline const PrimitivePtr kPrimPad = std::make_shared<Primitive>("Pad");
 inline const PrimitivePtr kPrimArgMaxWithValue = std::make_shared<Primitive>("ArgMaxWithValue");
 inline const PrimitivePtr kPrimUnique = std::make_shared<Primitive>("Unique");
 inline const PrimitivePtr kPrimUniqueGrad = std::make_shared<Primitive>("UniqueGrad");
+inline const PrimitivePtr kPrimExtractImagePatches = std::make_shared<Primitive>("ExtractImagePatches");
 
 // NN
 inline const PrimitivePtr kPrimFlatten = std::make_shared<Primitive>("Flatten");

mindspore/nn/layer/basic.py

@@ -542,12 +542,16 @@ class Unfold(Cell):
         self.transpose = P.Transpose()
         self.format_NHWC = (0, 2, 3, 1)
         self.format_NCHW = (0, 3, 1, 2)
+        self.is_ge = context.get_context("enable_ge")
 
     def construct(self, input_x):
-        x_transpose = self.transpose(input_x, self.format_NHWC)
-        ret = self.extract_image_patches(x_transpose)
-        ret_transpose = self.transpose(ret, self.format_NCHW)
-        return ret_transpose
+        if self.is_ge:
+            x_transpose = self.transpose(input_x, self.format_NHWC)
+            ret = self.extract_image_patches(x_transpose)
+            result = self.transpose(ret, self.format_NCHW)
+        else:
+            result = self.extract_image_patches(input_x)
+        return result
 
 
 @constexpr

mindspore/ops/_grad/grad_nn_ops.py

@@ -21,6 +21,7 @@ from mindspore.common.tensor import Tensor
 from .grad_base import bprop_getters
 from .. import functional as F
 from .. import operations as P
+from ...common import dtype as mstype
 from ..composite.multitype_ops.zeros_like_impl import zeros_like
 from ..operations import _grad_ops as G
 from ..operations import _inner_ops as inner
@@ -75,11 +76,43 @@ def get_bprop_extract_image_patches(self):
     fill = P.Fill()
     slice_op = P.Slice()
     transpose = P.Transpose()
+    cast = P.Cast()
     matmul = P.MatMul()
 
     _, ksizes_row, ksizes_col, _ = self.ksizes
 
     def bprop(x, out, dout):
+        x_shape = get_shape(x)
+        x_batch, x_depth, x_row, x_col = x_shape
+        x_indices_num = x_row * x_col + 1
+        x_idx = cast(F.tuple_to_array(range(1, x_indices_num)), mstype.float32)
+        x_idx = reshape(x_idx, (1, 1, x_row, x_col))
+        x_idx_patch = cast(extract_image_patches(x_idx), mstype.int32)
+        x_idx_patch = transpose(x_idx_patch, (0, 2, 3, 1))
+
+        out_shape = get_shape(out)
+        _, _, out_row, out_col = out_shape
+        out_indices_num = out_row * out_col * ksizes_row * ksizes_col
+        out_idx = F.tuple_to_array(range(out_indices_num))
+        out_idx = reshape(out_idx, (1, out_row, out_col, ksizes_row * ksizes_col))
+
+        idx_tensor = concat((expand_dims(x_idx_patch, -1), expand_dims(out_idx, -1)))
+        idx_tensor = reshape(idx_tensor, (-1, 2))
+        sp_shape = (x_indices_num, out_indices_num)
+        sp_tensor = scatter_nd(idx_tensor, fill(dtype(dout), (out_indices_num,), 1), sp_shape)
+        sp_tensor = slice_op(sp_tensor, (1, 0), (x_indices_num - 1, out_indices_num))
+
+        grad = transpose(dout, (0, 2, 3, 1))
+        grad = reshape(grad, (x_batch, out_row, out_col, ksizes_row, ksizes_col, x_depth))
+        grad = transpose(grad, (1, 2, 3, 4, 0, 5))
+        grad = reshape(grad, (-1, x_batch * x_depth))
+
+        jac = matmul(sp_tensor, grad)
+        dx = reshape(jac, (x_row, x_col, x_batch, x_depth))
+        dx = transpose(dx, (2, 3, 0, 1))
+        return (dx,)
+
+    def bprop_ge(x, out, dout):
         x_shape = get_shape(x)
         x_batch, x_row, x_col, x_depth = x_shape
         x_indices_num = x_row * x_col + 1
@@ -109,6 +142,9 @@ def get_bprop_extract_image_patches(self):
 
         return (dx,)
 
+    if context.get_context("enable_ge"):
+        return bprop_ge
+
     return bprop
 
 

mindspore/ops/operations/_inner_ops.py

@@ -17,6 +17,7 @@
 
 from ..._checkparam import Rel
 from ..._checkparam import Validator as validator
+from ... import context
 from ...common import dtype as mstype
 from ..primitive import PrimitiveWithInfer, prim_attr_register
 from ..operations.math_ops import _infer_shape_reduce
@@ -200,10 +201,13 @@ class ExtractImagePatches(PrimitiveWithInfer):
         self.padding = validator.check_string('padding', padding.upper(), ['VALID', 'SAME'], self.name)
         self.add_prim_attr("padding", self.padding)
         self.add_prim_attr("io_format", "NHWC")
+        self.is_ge = context.get_context("enable_ge")
 
     def infer_shape(self, input_x):
         """infer shape"""
-        in_batch, in_row, in_col, in_depth = input_x
+        in_batch, in_depth, in_row, in_col = input_x
+        if self.is_ge:
+            in_batch, in_row, in_col, in_depth = input_x
         _, ksize_row, ksize_col, _ = self.ksizes
         _, stride_row, stride_col, _ = self.strides
         _, rate_row, rate_col, _ = self.rates
@@ -223,7 +227,9 @@ class ExtractImagePatches(PrimitiveWithInfer):
             out_row = (in_row - 1) // stride_row + 1
             out_col = (in_col - 1) // stride_col + 1
 
-        out_shape = [out_batch, out_row, out_col, out_depth]
+        out_shape = [out_batch, out_depth, out_row, out_col]
+        if self.is_ge:
+            out_shape = [out_batch, out_row, out_col, out_depth]
         return out_shape
 
     def infer_dtype(self, input_x):