1 files changed, 422 insertions, 0 deletions
diff --git a/compiler/enco/core/src/CppGen/Subnet.cpp b/compiler/enco/core/src/CppGen/Subnet.cpp
new file mode 100644
index 000000000..9a636c6ae
--- /dev/null
+++ b/compiler/enco/core/src/CppGen/Subnet.cpp
@@ -0,0 +1,422 @@
+/*
+ * Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
+ *
+ * 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 "CppGen/Subnet.h"
+
+#include "Dims.h"
+#include "String.h"
+
+#include <pp/LinearDocument.h>
+
+#include <stdex/Memory.h>
+
+#include <sstream>
+
+using stdex::make_unique;
+using enco::concat;
+
+#define S(content) #content
+
+namespace ann
+{
+static std::ostream &operator<<(std::ostream &os, const ann::OperandID &id)
+{
+  os << id.value();
+  return os;
+}
+} // namespace ann
+
+namespace
+{
+
+class SubnetStructImpl final : public enco::SubnetStruct
+{
+public:
+  SubnetStructImpl() : _dtor{pp::LinearDocument::Direction::Reverse}
+  {
+    // DO NOTHING
+  }
+
+public:
+  std::string model(void) const override { return "_model"; }
+  std::string compilation(void) const override { return "_compilation"; }
+
+public:
+  const pp::MultiLineText &def(void) const override { return _def; }
+  pp::LinearDocument *def(void) { return &_def; }
+
+public:
+  const pp::MultiLineText &ctor(void) const override { return _ctor; }
+  pp::LinearDocument *ctor(void) { return &_ctor; }
+
+public:
+  const pp::MultiLineText &dtor(void) const override { return _dtor; }
+  pp::LinearDocument *dtor(void) { return &_dtor; }
+
+private:
+  pp::LinearDocument _def;
+  pp::LinearDocument _ctor;
+  pp::LinearDocument _dtor;
+};
+
+struct CodeFragment
+{
+  virtual ~CodeFragment() = default;
+
+  virtual void dump(pp::LinearDocument *) const = 0;
+};
+
+pp::LinearDocument *operator<<(pp::LinearDocument *doc, const CodeFragment &fragment)
+{
+  fragment.dump(doc);
+  return doc;
+}
+
+const char *scalar_operand_code(const ann::DType &dtype)
+{
+  switch (dtype)
+  {
+    case ann::DType::S32:
+      return "ANEURALNETWORKS_INT32";
+    default:
+      break;
+  };
+
+  throw std::invalid_argument("dtype");
+}
+
+const char *tensor_operand_code(const ann::DType &dtype)
+{
+  switch (dtype)
+  {
+    case ann::DType::S32:
+      return "ANEURALNETWORKS_TENSOR_INT32";
+    case ann::DType::F32:
+      return "ANEURALNETWORKS_TENSOR_FLOAT32";
+    default:
+      break;
+  };
+
+  throw std::invalid_argument("dtype");
+}
+
+class ScalarOperandDecl final : public CodeFragment
+{
+public:
+  ScalarOperandDecl(const std::string &model, const ann::DType &dtype)
+      : _model{model}, _dtype{dtype}
+  {
+    // DO NOTHING
+  }
+
+public:
+  void dump(pp::LinearDocument *doc) const override
+  {
+    doc->append("{");
+    doc->indent();
+    doc->append("ANeuralNetworksOperandType t;");
+    doc->append();
+    doc->append("t.type = ", scalar_operand_code(_dtype), ";");
+    doc->append("t.dimensionCount = 0;");
+    doc->append("t.dimensions = nullptr;");
+    doc->append("t.scale = 1.0f;");
+    doc->append("t.zeroPoint = 0;");
+    doc->append();
+    doc->append("ANeuralNetworksModel_addOperand(", _model, ", &t);");
+    doc->unindent();
+    doc->append("}");
+  }
+
+private:
+  std::string _model;
+  ann::DType _dtype;
+};
+
+class TensorOperandDecl final : public CodeFragment
+{
+public:
+  TensorOperandDecl(const std::string &model, const ann::DType &dtype,
+                    const nncc::core::ADT::tensor::Shape &shape)
+      : _model{model}, _dtype{dtype}, _shape{shape}
+  {
+    // DO NOTHING
+  }
+
+public:
+  void dump(pp::LinearDocument *doc) const override
+  {
+    const auto rank = _shape.rank();
+    const auto dims = as_dims(_shape);
+
+    assert(rank == dims.size());
+
+    doc->append("{");
+    doc->indent();
+    doc->append("uint32_t d[", rank, "] = { ", concat(", ", dims.begin(), dims.end()), " };");
+    doc->append();
+    doc->append("ANeuralNetworksOperandType t;");
+    doc->append();
+    doc->append("t.type = ", tensor_operand_code(_dtype), ";");
+    doc->append("t.dimensionCount = ", rank, ";");
+    doc->append("t.dimensions = d;");
+    doc->append("t.scale = 1.0f;");
+    doc->append("t.zeroPoint = 0;");
+    doc->append();
+    doc->append("ANeuralNetworksModel_addOperand(", _model, ", &t);");
+    doc->unindent();
+    doc->append("}");
+  }
+
+private:
+  std::string _model;
+  ann::DType _dtype;
+  nncc::core::ADT::tensor::Shape _shape;
+};
+
+/**
+ * @brief Code fragment that calls ANeuralNetworksModel_setOperandValue
+ */
+class WeightDecl final : public CodeFragment
+{
+public:
+  WeightDecl(const std::string &model, const ann::OperandID &id, const std::string &base,
+             const std::string &size)
+      : _model{model}, _id{id}, _base{base}, _size{size}
+  {
+    // DO NOTHING
+  }
+
+public:
+  void dump(pp::LinearDocument *doc) const override
+  {
+    doc->append("ANeuralNetworksModel_setOperandValue(", _model, ", ", _id.value(), ", ", _base,
+                ", ", _size, ");");
+  }
+
+private:
+  std::string _model;
+  ann::OperandID _id;
+  std::string _base;
+  std::string _size;
+};
+
+/**
+ * @brief Code fragment that calls ANeuralNetworksModel_addOperation
+ */
+class OperationDecl final : public CodeFragment
+{
+public:
+  OperationDecl(const std::string &model, const ann::Operation *op) : _model{model}, _op{op}
+  {
+    // DO NOTHING
+  }
+
+private:
+  static std::string opcode(const ann::Operation::Code &code)
+  {
+    switch (code)
+    {
+#define ANN_OPERATION(TAG, ENUM)  \
+  case ann::Operation::Code::TAG: \
+    return #ENUM;
+#include "ANN/IR/Operation.def"
+#undef ANN_OPERATION
+      default:
+        throw std::invalid_argument{"code"};
+    };
+  }
+
+public:
+  void dump(pp::LinearDocument *doc) const override
+  {
+    const auto in_count = _op->inputs().size();
+    auto in_beg = _op->inputs().begin();
+    auto in_end = _op->inputs().end();
+
+    const auto out_count = _op->outputs().size();
+    auto out_beg = _op->outputs().begin();
+    auto out_end = _op->outputs().end();
+
+    auto op = opcode(_op->code());
+
+    doc->append("{");
+    doc->indent();
+    doc->append("uint32_t inputs[", in_count, "] = { ", concat(", ", in_beg, in_end), " };");
+    doc->append("uint32_t outputs[", out_count, "] = { ", concat(", ", out_beg, out_end), " };");
+    doc->append();
+    doc->append("ANeuralNetworksModel_addOperation(", _model, ", ", op, ", ", in_count,
+                ", inputs, ", out_count, ", outputs);");
+    doc->unindent();
+    doc->append("}");
+  }
+
+private:
+  std::string _model;
+  const ann::Operation *_op;
+};
+
+/**
+ * @brief Code fragment that calls ANeuralNetworksModel_identifyInputsAndOutputs
+ */
+class ArgumentDecl final : public CodeFragment
+{
+public:
+  ArgumentDecl(const std::string &mname, const ANNBinder *binder) : _mname{mname}, _binder{binder}
+  {
+    // DO NOTHING
+  }
+
+public:
+  void dump(pp::LinearDocument *doc) const override
+  {
+    doc->append("{");
+    doc->indent();
+
+    auto module = _binder->module();
+    const uint32_t input_count = module->input()->size();
+
+    doc->append("uint32_t inputs[", input_count, "];");
+    for (uint32_t n = 0; n < input_count; ++n)
+    {
+      doc->append("inputs[", n, "] = ", module->input()->at(n), ";");
+    }
+
+    const uint32_t output_count = module->output()->size();
+
+    doc->append("uint32_t outputs[", output_count, "];");
+    for (uint32_t n = 0; n < output_count; ++n)
+    {
+      doc->append("outputs[", n, "] = ", module->output()->at(n), ";");
+    }
+
+    doc->append("ANeuralNetworksModel_identifyInputsAndOutputs(", _mname, ", ", input_count,
+                ", inputs, ", output_count, ", outputs);");
+    doc->unindent();
+    doc->append("}");
+  }
+
+private:
+  std::string _mname;
+  const ANNBinder *_binder;
+};
+
+} // namespace
+
+namespace enco
+{
+
+std::unique_ptr<SubnetStruct> SubnetStructBuilder::build(const ANNBinder *binder) const
+{
+  auto res = make_unique<SubnetStructImpl>();
+
+  auto mname = res->model();
+  auto cname = res->compilation();
+
+  res->def()->append("ANeuralNetworksModel *", mname, ";");
+  res->def()->append("ANeuralNetworksCompilation *", cname, ";");
+
+  res->ctor()->append("ANeuralNetworksModel_create(&", mname, ");");
+  res->dtor()->append("ANeuralNetworksModel_free(", mname, ");");
+
+  binder->module()->operand()->each([&](const ann::OperandID &id, const ann::Operand *info) {
+    // TODO Remove dynamic cast
+    if (auto scalar = dynamic_cast<const ann::ScalarOperand *>(info))
+    {
+      res->ctor() << ScalarOperandDecl{mname, scalar->dtype()};
+    }
+    else if (auto tensor = dynamic_cast<const ann::TensorOperand *>(info))
+    {
+      res->ctor() << TensorOperandDecl{mname, tensor->dtype(), tensor->shape()};
+    }
+    else
+    {
+      throw std::runtime_error{"Unsupported"};
+    }
+
+    if (_weighted.find(info) != _weighted.end())
+    {
+      const auto &base_exp = _base_exprs.at(info);
+      const auto &size_exp = _size_exprs.at(info);
+
+      res->ctor() << WeightDecl{mname, id, base_exp, size_exp};
+    }
+  });
+
+  for (unsigned n = 0; n < binder->module()->operation()->count(); ++n)
+  {
+    auto op = binder->module()->operation()->at(n);
+    res->ctor() << OperationDecl{mname, op};
+  }
+
+  // Emit ANeuralNetworksModel_identifyInputsAndOutputs call
+  res->ctor() << ArgumentDecl{mname, binder};
+
+  // Emit ANeuralNetworksModel_finish call
+  res->ctor()->append("ANeuralNetworksModel_finish(", mname, ");");
+
+  // Create compilation
+  res->ctor()->append("ANeuralNetworksCompilation_create(", mname, ", &", cname, ");");
+  res->dtor()->append("ANeuralNetworksCompilation_free(", cname, ");");
+
+  // Finalize compilation
+  res->ctor()->append("ANeuralNetworksCompilation_finish(", cname, ");");
+
+  return std::move(res);
+}
+
+std::unique_ptr<pp::MultiLineText> SubnetBlockCompiler::compile(const ANNBinder *binder) const
+{
+  auto res = make_unique<pp::LinearDocument>();
+
+  const auto compilation = _compilation_ctx.at(binder);
+
+  res->append("ANeuralNetworksExecution *execution;");
+  res->append("ANeuralNetworksEvent *event;");
+  res->append();
+  res->append("ANeuralNetworksExecution_create(", compilation, ", &execution);");
+
+  // Emit ANeuralNetworksExecution_setInput call(s)
+  for (uint32_t n = 0; n < binder->module()->input()->size(); ++n)
+  {
+    auto bag = binder->input(n);
+    auto base = _mem.base(bag);
+    auto size = _mem.size(bag);
+
+    res->append("ANeuralNetworksExecution_setInput(execution, ", n, ", nullptr, ", base, ", ", size,
+                ");");
+  }
+
+  // Emit ANeuralNetworksExecution_setOutput call(s)
+  for (uint32_t n = 0; n < binder->module()->output()->size(); ++n)
+  {
+    auto bag = binder->output(n);
+    auto base = _mem.base(bag);
+    auto size = _mem.size(bag);
+
+    res->append("ANeuralNetworksExecution_setOutput(execution, ", n, ", nullptr, ", base, ", ",
+                size, ");");
+  }
+
+  res->append("ANeuralNetworksExecution_startCompute(execution, &event);");
+  res->append("ANeuralNetworksEvent_wait(event);");
+  res->append("ANeuralNetworksEvent_free(event);");
+
+  res->append("ANeuralNetworksExecution_free(execution);");
+
+  return std::move(res);
+}
+
+} // namespace enco