Add index validity check and adapt for more conditions of sparse_to_dense ops.

mindspore/lite/nnacl/fp32/sparse_to_dense.c

@@ -0,0 +1,45 @@
+/**
+ * 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 "nnacl/fp32/sparse_to_dense.h"
+
+void SparseToDense(int **sparse_indices, int *output_shape,
+                   float *sparse_values, float default_value, float *output,
+                   bool isScalar, int index_start, int index_end, int out_width) {
+  for (int i = index_start; i < index_end; i++) {
+    for (int j = 0; j < out_width; j++) {
+      output[i * out_width + j] = default_value;
+    }
+  }
+
+  int d1 = output_shape[1] * output_shape[2] * output_shape[3];
+  int d2 = output_shape[2] * output_shape[3];
+  int d3 = output_shape[3];
+
+  int index;
+  if (isScalar == true) {
+    for (int i = index_start; i < index_end; i++) {
+      index = d1 * sparse_indices[i][0] + d2 * sparse_indices[i][1] +
+              d3 * sparse_indices[i][2] + sparse_indices[i][3];
+      output[index] = sparse_values[0];
+    }
+  } else {
+    for (int i = index_start; i < index_end; i++) {
+      index = d1 * sparse_indices[i][0] + d2 * sparse_indices[i][1] +
+              d3 * sparse_indices[i][2] + sparse_indices[i][3];
+      output[index] = sparse_values[i];
+    }
+  }
+}

mindspore/lite/nnacl/fp32/sparse_to_dense.h

@@ -0,0 +1,31 @@
+/**
+ * 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_LITE_NNACL_FP32_SPARSETODENSE_H_
+#define MINDSPORE_LITE_NNACL_FP32_SPARSETODENSE_H_
+
+#include "nnacl/sparse_to_dense_parameter.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+void SparseToDense(int **sparse_indices_vect, int *output_shape,
+                   float *sparse_values, float default_value, float *output,
+                   bool isScalar, int index_start, int index_end, int out_width);
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // MINDSPORE_LITE_NNACL_FP32_SPARSETODENSE_H_

mindspore/lite/nnacl/sparse_to_dense.c

@@ -1,34 +0,0 @@
-/**
- * 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 "nnacl/sparse_to_dense.h"
-
-void SparseToDense(int *input, int *output_shape_, float *snum, float *dnum, int sp_num, float *output,
-                   SparseToDenseParameter *s2d_param_, int task_id) {
-  int m;
-  for (int i = task_id; i < output_shape_[0]; i += s2d_param_->op_parameter_.thread_num_) {
-    for (int j = 0; j < output_shape_[1]; j++) {
-      m = i * output_shape_[1] + j;
-      output[m] = dnum[0];
-    }
-  }
-
-  for (int j = 0; j < sp_num; j++) {
-    int temp = j * 2;
-    int temp1 = j * 2 + 1;
-    int tempout1 = input[temp] * output_shape_[1] + input[temp1];
-    output[tempout1] = snum[j];
-  }
-}

mindspore/lite/nnacl/sparse_to_dense_parameter.h

@@ -13,8 +13,9 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#ifndef MINDSPORE_LITE_NNACL_SPARSETODENSE_H_
-#define MINDSPORE_LITE_NNACL_SPARSETODENSE_H_
+
+#ifndef MINDSPORE_LITE_NNACL_SPARSE_TO_DENSE_PARAMETER_H_
+#define MINDSPORE_LITE_NNACL_SPARSE_TO_DENSE_PARAMETER_H_
 
 #include "nnacl/op_base.h"
 
@@ -22,16 +23,6 @@ typedef struct SparseToDenseParameter {
   OpParameter op_parameter_;
   bool validate_indices_;
   int thread_num_;
-  int count_;
 } SparseToDenseParameter;
 
-#ifdef __cplusplus
-extern "C" {
-#endif
-void SparseToDense(int *input, int *output_shape_, float *snum, float *dnum, int sp_num, float *output,
-                   SparseToDenseParameter *s2d_param_, int task_id);
-#ifdef __cplusplus
-}
-#endif
-
-#endif  // MINDSPORE_LITE_NNACL_SPARSETODENCE_H_
+#endif  // MINDSPORE_LITE_NNACL_SPARSE_TO_DENSE_PARAMETER_H_

mindspore/lite/src/ops/sparse_to_dense.cc

@@ -42,5 +42,34 @@ int SparseToDense::UnPackToFlatBuilder(const schema::Primitive *primitive, flatb
   return RET_OK;
 }
 #endif
+
+int SparseToDense::InferShape(std::vector<Tensor *> inputs_, std::vector<Tensor *> outputs_) {
+  MS_ASSERT(this->primitive_ != nullptr);
+  MS_ASSERT(output_shape != nullptr);
+  auto output = outputs_.front();
+  if (output == nullptr) {
+    MS_LOG(ERROR) << "output null pointer dereferencing.";
+    return RET_ERROR;
+  }
+  auto input2 = inputs_.at(2);
+  outputs_[0]->set_data_type(input2->data_type());
+  outputs_[0]->SetFormat(input2->GetFormat());
+
+  if (!GetInferFlag()) {
+    return RET_OK;
+  }
+  if (this->primitive_ == nullptr) {
+    return RET_NULL_PTR;
+  }
+
+  auto input1 = inputs_.at(1);
+  int *input1_data = reinterpret_cast<int *>(input1->MutableData());
+  std::vector<int> output_shape;
+  for (int i = 0; i < input1->ElementsNum(); i++) {
+    output_shape.push_back(input1_data[i]);
+  }
+  outputs_[0]->set_shape(output_shape);
+  return RET_OK;
+}
 }  // namespace lite
 }  // namespace mindspore

mindspore/lite/src/ops/sparse_to_dense.h

@@ -45,6 +45,7 @@ class SparseToDense : public PrimitiveC {
   std::vector<int> GetSparseValue() const;
   std::vector<int> GetDefaultValue() const;
   bool GetValidateIndices() const;
+  int InferShape(std::vector<lite::Tensor *> inputs_, std::vector<lite::Tensor *> outputs_) override;
 };
 }  // namespace lite
 }  // namespace mindspore

mindspore/lite/src/populate_parameter.cc

@@ -170,7 +170,7 @@
 #include "nnacl/fp32/embedding_lookup.h"
 #include "nnacl/fp32/elu.h"
 #include "nnacl/leaky_relu_parameter.h"
-#include "nnacl/sparse_to_dense.h"
+#include "mindspore/lite/nnacl/fp32/sparse_to_dense.h"
 #include "nnacl/l2_norm_parameter.h"
 #include "nnacl/detection_post_process_parameter.h"
 #include "nnacl/fp32/exp.h"

mindspore/lite/src/runtime/kernel/arm/fp32/sparse_to_dense.cc

@@ -14,12 +14,14 @@
  * limitations under the License.
  */
 #include "src/runtime/kernel/arm/fp32/sparse_to_dense.h"
+
 #include <vector>
+
+#include "include/errorcode.h"
+#include "mindspore/lite/nnacl/fp32/sparse_to_dense.h"
 #include "schema/model_generated.h"
 #include "schema/ops_generated.h"
-#include "nnacl/sparse_to_dense.h"
 #include "src/kernel_registry.h"
-#include "include/errorcode.h"
 #include "src/runtime/runtime_api.h"
 
 using mindspore::kernel::KERNEL_ARCH::kCPU;
@@ -30,12 +32,45 @@ using mindspore::schema::PrimitiveType_SparseToDense;
 
 namespace mindspore::kernel {
 int SparseToDenseCPUKernel::Init() {
-  s2d_param_->op_parameter_.thread_num_ = thread_count_;
+  auto input2 = in_tensors_.at(2);
+  auto input3 = in_tensors_.at(3);
+  sparse_values = reinterpret_cast<float *>(input2->MutableData());
+  default_value = reinterpret_cast<float *>(input3->MutableData())[0];
+  if (input2->ElementsNum() == 1) {
+    isScalar = true;
+  }
+  if (!InferShapeDone()) {
+    return RET_OK;
+  }
+  return ReSize();
+}
+
+int SparseToDenseCPUKernel::ReSize() {
+  auto output0 = out_tensors_.at(0);
+  std::vector<int> out_shape_tensor = output0->shape();
+  auto output_shape_tmp = reinterpret_cast<int *>(out_shape_tensor.data());
+  int output_dim = output0->shape().size();
+  for (int i = 0; i < DIMENSION_4D - output_dim; i++) {
+    output_shape[i] = 1;
+  }
+  for (int i = 0; i < output_dim; i++) {
+    output_shape[i + DIMENSION_4D - output_dim] = output_shape_tmp[i];
+  }
+  output_num = output0->ElementsNum();
   return RET_OK;
 }
 
 int SparseToDenseCPUKernel::DoExcute(int task_id) {
-  SparseToDense(input_data_, output_shape_, snum_, dnum_, sp_num_, output_data, s2d_param_, task_id);
+  int real_dst_count = MSMIN(index_num - task_id * count_unit_, count_unit_);
+  if (real_dst_count <= 0) {
+    return RET_OK;
+  }
+  int index_start = task_id * count_unit_;
+  int index_end = index_start + real_dst_count;
+  int out_width = output_num / index_num;
+  SparseToDense(sparse_indices_vect, output_shape, sparse_values,
+                default_value, output_data, isScalar,
+                index_start, index_end, out_width);
   return RET_OK;
 }
 
@@ -43,38 +78,117 @@ int SparseToDenseRun(void *cdata, int task_id) {
   auto s2ddata = reinterpret_cast<SparseToDenseCPUKernel *>(cdata);
   auto ret = s2ddata->DoExcute(task_id);
   if (ret != RET_OK) {
-    MS_LOG(ERROR) << "SparseToDenseRun error task_id[" << task_id << "] error_code[" << ret << "]";
+    MS_LOG(ERROR) << "SparseToDenseRun error task_id[" << task_id
+                  << "] error_code[" << ret << "]";
     return RET_ERROR;
   }
   return RET_OK;
 }
+
+int SparseToDenseCPUKernel::GenerateIndices() {
+  auto input0 = in_tensors_.at(0);
+  index_dim = input0->shape().size();
+  index_num = input0->shape()[0];
+  int *sparse_indices = reinterpret_cast<int *>(input0->MutableData());
+  sparse_indices_vect = reinterpret_cast<int **>(ctx_->allocator->Malloc(sizeof(int *) * index_num));
+  if (sparse_indices_vect == nullptr) {
+    MS_LOG(ERROR) << "Null pointer reference: sparse_indices_vect.";
+    return RET_ERROR;
+  }
+  switch (index_dim) {
+    case 0:
+    case 1: {
+      for (int i = 0; i < index_num; i++) {
+        sparse_indices_vect[i] = new int[DIMENSION_4D];
+        if (sparse_indices_vect[i] == nullptr) {
+          MS_LOG(ERROR) << "Null pointer reference: sparse_indices_vect[" << i << "].";
+          return RET_ERROR;
+        }
+        for (int j = 0; j < DIMENSION_4D - 1; j++) {
+          sparse_indices_vect[i][j] = 0;
+        }
+        sparse_indices_vect[i][DIMENSION_4D - 1] = sparse_indices[i];
+      }
+      break;
+    }
+    case 2: {
+      int true_dims = input0->shape()[1];
+      MS_ASSERT(true_dims <= DIMENSION_4D);
+      for (int i = 0; i < index_num; i++) {
+        sparse_indices_vect[i] = new int[DIMENSION_4D];
+        if (sparse_indices_vect[i] == nullptr) {
+          MS_LOG(ERROR) << "Null pointer reference: sparse_indices_vect[" << i << "].";
+          return RET_ERROR;
+        }
+        for (int j = 0; j < DIMENSION_4D - true_dims; j++) {
+          sparse_indices_vect[i][j] = 0;
+        }
+        for (int j = 0; j < true_dims; j++) {
+          sparse_indices_vect[i][j + DIMENSION_4D - true_dims] = sparse_indices[i * true_dims + j];
+        }
+      }
+      break;
+    }
+    default: {
+      MS_LOG(ERROR) << "Indices dimensions is " << index_dim << ", which must be 0, 1 or 2";
+      return RET_ERROR;
+    }
+  }
+  return RET_OK;
+}
+
+int SparseToDenseCPUKernel::IndicesValidCheck() {
+  int d1 = output_shape[1] * output_shape[2] * output_shape[3];
+  int d2 = output_shape[2] * output_shape[3];
+  int d3 = output_shape[3];
+  int index_before = -1;
+  for (int i = 0; i < index_num; i++) {
+    int index = d1 * sparse_indices_vect[i][0] + d2 * sparse_indices_vect[i][1] +
+                d3 * sparse_indices_vect[i][2] + sparse_indices_vect[i][3];
+    if (index <= index_before) {
+      return RET_ERROR;
+    }
+    index_before = index;
+  }
+  return RET_OK;
+}
+
 int SparseToDenseCPUKernel::Run() {
   auto ret = Prepare();
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Prepare failed.";
     return RET_ERROR;
   }
-  auto input = in_tensors_.at(0);
-  auto input1 = in_tensors_.at(1);
-  auto input2 = in_tensors_.at(2);
-  auto input3 = in_tensors_.at(3);
-  auto output0 = out_tensors_.at(0);
-
-  input_data_ = reinterpret_cast<int *>(input->MutableData());
-  total_number_ = reinterpret_cast<int *>(input1->MutableData());
-  snum_ = reinterpret_cast<float *>(input2->MutableData());
-  dnum_ = reinterpret_cast<float *>(input3->MutableData());
-  sp_num_ = static_cast<int>(input->ElementsNum() / 2);
-
+  auto ret1 = GenerateIndices();
+  if (ret1 != RET_OK) {
+    MS_LOG(ERROR) << "Generate Indices failed.";
+    return RET_ERROR;
+  }
+  if (s2d_param->validate_indices_ == true) {
+    auto ret2 = IndicesValidCheck();
+    if (ret2 != RET_OK) {
+      MS_LOG(ERROR) << "The sparse indices is not valid.";
+      return RET_ERROR;
+    }
+  }
   output_data = reinterpret_cast<float *>(out_tensors_.at(0)->MutableData());
-  std::vector<int> temp_shape = output0->shape();
-  output_shape_ = reinterpret_cast<int *>(temp_shape.data());
-
-  ret = ParallelLaunch(THREAD_POOL_DEFAULT, SparseToDenseRun, this, s2d_param_->thread_num_);
+  count_unit_ = thread_count_ > 1 ? UP_DIV(index_num, thread_count_) : index_num;
+  ret = ParallelLaunch(THREAD_POOL_DEFAULT, SparseToDenseRun, this,
+                       s2d_param->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "SparseToDenseRun error: error_code[" << ret << "]";
     return RET_ERROR;
   }
+
+  for (int i = 0; i < index_num; i++) {
+    if (sparse_indices_vect[i] != nullptr) {
+      delete sparse_indices_vect[i];
+    }
+  }
+  if (sparse_indices_vect != nullptr) {
+    ctx_->allocator->Free(sparse_indices_vect);
+    sparse_indices_vect = nullptr;
+  }
   return RET_OK;
 }
 
@@ -88,20 +202,25 @@ kernel::LiteKernel *CpuSparseToDenseFp32KernelCreator(const std::vector<lite::Te
     return nullptr;
   }
   MS_ASSERT(desc.type == schema::PrimitiveType_SparseToDense);
-  auto *kernel = new (std::nothrow) SparseToDenseCPUKernel(opParameter, inputs, outputs, ctx, primitive);
+  auto *kernel = new (std::nothrow)
+      SparseToDenseCPUKernel(opParameter, inputs, outputs, ctx, primitive);
   if (kernel == nullptr) {
     MS_LOG(ERROR) << "new SparseToDenseCPUKernel fail!";
     return nullptr;
   }
   auto ret = kernel->Init();
   if (ret != RET_OK) {
-    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
-                  << schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
+    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_
+                  << ", type: "
+                  << schema::EnumNamePrimitiveType(
+                         static_cast<schema::PrimitiveType>(
+                             opParameter->type_));
     delete kernel;
     return nullptr;
   }
   return kernel;
 }
 
-REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_SparseToDense, CpuSparseToDenseFp32KernelCreator)
+REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_SparseToDense,
+           CpuSparseToDenseFp32KernelCreator)
 }  // namespace mindspore::kernel

mindspore/lite/src/runtime/kernel/arm/fp32/sparse_to_dense.h

@@ -20,7 +20,7 @@
 #include "src/lite_kernel.h"
 
 #include "include/context.h"
-#include "nnacl/sparse_to_dense.h"
+#include "mindspore/lite/nnacl/fp32/sparse_to_dense.h"
 #include "src/runtime/kernel/arm/base/layout_transform.h"
 
 using mindspore::lite::Context;
@@ -32,28 +32,34 @@ class SparseToDenseCPUKernel : public LiteKernel {
                          const std::vector<lite::Tensor *> &outputs, const lite::Context *ctx,
                          const mindspore::lite::PrimitiveC *primitive)
       : LiteKernel(parameter, inputs, outputs, ctx, primitive), ctx_(ctx), thread_count_(ctx->thread_num_) {
-    s2d_param_ = (reinterpret_cast<SparseToDenseParameter *>(op_parameter_));
+    s2d_param = (reinterpret_cast<SparseToDenseParameter *>(op_parameter_));
+    s2d_param->thread_num_ = thread_count_;
   }
   ~SparseToDenseCPUKernel() = default;
 
   int Init() override;
-  int ReSize() override { return 0; }
+  int ReSize() override;
   int Run() override;
   int DoExcute(int task_id);
+  int GenerateIndices();
+  int IndicesValidCheck();
 
  protected:
   const Context *ctx_;
   int thread_count_;
-  SparseToDenseParameter *s2d_param_;
+  SparseToDenseParameter *s2d_param;
 
  private:
-  int *input_data_;
-  int *total_number_;
-  int sp_num_;
-  float *snum_;
-  float *dnum_;
-  float *output_data;
-  int *output_shape_;
+  int **sparse_indices_vect = nullptr;
+  float *sparse_values = nullptr;
+  float default_value;
+  bool isScalar = false;
+  int index_num;
+  int index_dim;
+  float *output_data = nullptr;
+  int output_shape[4];
+  int output_num;
+  int64_t count_unit_;
 };
 }  // namespace mindspore::kernel
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_SPARSETODENSE_H_