//===- ReleaseModeModelRunner.h - Fast, precompiled model runner  ---------===//
 
//
 
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 
// See https://llvm.org/LICENSE.txt for license information.
 
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
//
 
//===----------------------------------------------------------------------===//
 
//
 
// This file implements a model runner wrapping an AOT compiled ML model.
 
// Only inference is supported.
 
//
 
//===----------------------------------------------------------------------===//
 
 
 
#ifndef LLVM_ANALYSIS_RELEASEMODEMODELRUNNER_H
 
#define LLVM_ANALYSIS_RELEASEMODEMODELRUNNER_H
 
 
 
#include "llvm/Analysis/MLModelRunner.h"
 
#include "llvm/Analysis/TensorSpec.h"
 
#include "llvm/Support/ErrorHandling.h"
 
 
 
#include <memory>
 
#include <vector>
 
 
 
namespace llvm {
 
 
 
/// ReleaseModeModelRunner - production mode implementation of the
 
/// MLModelRunner. It uses an AOT-compiled SavedModel for efficient execution.
 
template <class TGen>
 
class ReleaseModeModelRunner final : public MLModelRunner {
 
public:
 
  /// FeatureNames' type should be an indexed collection of std::string, like
 
  /// std::array or std::vector, that has a size() method.
 
  template <class FType>
 
  ReleaseModeModelRunner(LLVMContext &Ctx, const FType &InputSpec,
 
                         StringRef DecisionName, StringRef FeedPrefix = "feed_",
 
                         StringRef FetchPrefix = "fetch_")
 
      : MLModelRunner(Ctx, MLModelRunner::Kind::Release, InputSpec.size()),
 
        CompiledModel(std::make_unique<TGen>()) {
 
    assert(CompiledModel && "The CompiledModel should be valid");
 
 
 
    for (size_t I = 0; I < InputSpec.size(); ++I) {
 
      const int Index =
 
          CompiledModel->LookupArgIndex(FeedPrefix.str() + InputSpec[I].name());
 
      void *Buffer = nullptr;
 
      if (Index >= 0)
 
        Buffer = CompiledModel->arg_data(Index);
 
      setUpBufferForTensor(I, InputSpec[I], Buffer);
 
    }
 
 
 
    ResultIndex = CompiledModel->LookupResultIndex(FetchPrefix.str() +
 
                                                   DecisionName.str());
 
    assert(ResultIndex >= 0 && "Cannot find DecisionName in inlining model");
 
  }
 
 
 
  virtual ~ReleaseModeModelRunner() = default;
 
 
 
  static bool classof(const MLModelRunner *R) {
 
    return R->getKind() == MLModelRunner::Kind::Release;
 
  }
 
 
 
private:
 
  void *evaluateUntyped() override {
 
    CompiledModel->Run();
 
    return CompiledModel->result_data(ResultIndex);
 
  }
 
 
 
  int32_t ResultIndex = -1;
 
  std::unique_ptr<TGen> CompiledModel;
 
};
 
 
 
/// A mock class satisfying the interface expected by ReleaseModeModelRunner for
 
/// its `TGen` parameter. Useful to avoid conditional compilation complexity, as
 
/// a compile-time replacement for a real AOT-ed model.
 
class NoopSavedModelImpl final {
 
#define NOOP_MODEL_ERRMSG                                                      \
 
  "The mock AOT-ed saved model is a compile-time stub and should not be "      \
 
  "called."
 
 
 
public:
 
  NoopSavedModelImpl() = default;
 
  int LookupArgIndex(const std::string &) { llvm_unreachable(NOOP_MODEL_ERRMSG); }
 
  int LookupResultIndex(const std::string &) { llvm_unreachable(NOOP_MODEL_ERRMSG); }
 
  void Run() { llvm_unreachable(NOOP_MODEL_ERRMSG); }
 
  void *result_data(int) { llvm_unreachable(NOOP_MODEL_ERRMSG); }
 
  void *arg_data(int) { llvm_unreachable(NOOP_MODEL_ERRMSG); }
 
#undef NOOP_MODEL_ERRMSG
 
};
 
} // namespace llvm
 
 
 
#endif // LLVM_ANALYSIS_RELEASEMODEMODELRUNNER_H