/* Copyright 2025 The OpenXLA Authors.

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 XLA_SERVICE_GPU_MODEL_COLLECTIVE_INTERPOLATOR_H_
#define XLA_SERVICE_GPU_MODEL_COLLECTIVE_INTERPOLATOR_H_

#include <cstdint>
#include <memory>
#include <optional>
#include <utility>
#include <variant>

#include "absl/container/flat_hash_map.h"
#include "absl/functional/overload.h"
#include "absl/status/statusor.h"
#include "absl/time/time.h"
#include "xla/hlo/ir/hlo_clone_context.h"
#include "xla/hlo/ir/hlo_instructions.h"
#include "xla/hlo/ir/hlo_opcode.h"
#include "xla/hlo/ir/replica_group.h"
#include "xla/service/gpu/model/gpu_hlo_cost_analysis.h"
#include "xla/service/gpu/model/hlo_op_profile.pb.h"
#include "xla/service/gpu/model/interpolator.h"
#include "xla/service/gpu/transforms/collectives/collective_ops_utils.h"
#include "xla/stream_executor/device_description.h"
#include "xla/xla_data.pb.h"

namespace xla::gpu {

class CollectiveInterpolator {
 public:
  struct FallbackInterpolatorKey {
    HloOpcode opcode;
    GPUCommunicationType communication_type;

    template <typename H>
    friend H AbslHashValue(H h, const FallbackInterpolatorKey& key) {
      return H::combine(std::move(h), key.opcode, key.communication_type);
    }

    bool operator==(const FallbackInterpolatorKey& other) const {
      return opcode == other.opcode &&
             communication_type == other.communication_type;
    }
  };

  struct ExactInterpolatorKey {
    HloOpcode opcode;
    std::variant<CollectiveDeviceList, CollectivePermuteCostModelType>
        collective_params;
    std::optional<PrimitiveType> data_type;

    template <typename H>
    friend H AbslHashValue(H h, const ExactInterpolatorKey& key) {
      h = H::combine(std::move(h), key.opcode, key.data_type);
      std::visit(
          absl::Overload{[&](CollectivePermuteCostModelType permute_type) {
                           h = H::combine(std::move(h), permute_type);
                         },
                         [&](const CollectiveDeviceList& device_list) {
                           h = H::combine(std::move(h),
                                          device_list.ToString(
                                              /*print_full_replica_group_list=*/
                                              true));
                         }},
          key.collective_params);
      return h;
    }

    bool operator==(const ExactInterpolatorKey& other) const {
      if (opcode != other.opcode ||
          collective_params.index() != other.collective_params.index() ||
          data_type != other.data_type) {
        return false;
      }
      return std::visit(
          absl::Overload{
              [&](CollectivePermuteCostModelType permute_type) {
                return permute_type == std::get<CollectivePermuteCostModelType>(
                                           other.collective_params);
              },
              [&](const CollectiveDeviceList& device_list) {
                return device_list.ToString(
                           /*print_full_replica_group_list=*/true) ==
                       std::get<CollectiveDeviceList>(other.collective_params)
                           .ToString(
                               /*print_full_replica_group_list=*/true);
              }},
          collective_params);
    }
  };

  using FallbackInterpolatorMap = std::unique_ptr<absl::flat_hash_map<
      FallbackInterpolatorKey, std::unique_ptr<InterpolatorBase<int64_t, 2>>>>;

  using ExactInterpolatorMap = std::unique_ptr<absl::flat_hash_map<
      ExactInterpolatorKey, std::unique_ptr<InterpolatorBase<int64_t, 1>>>>;

  static absl::StatusOr<std::unique_ptr<CollectiveInterpolator>> Create(
      int num_devices_per_host, const HloInstructionProfileList& profiles,
      const se::DeviceDescription& device_info,
      const GpuHloCostAnalysis* analysis = nullptr);

  static absl::StatusOr<std::unique_ptr<CollectiveInterpolator>> Create(
      int num_devices_per_host, const se::DeviceDescription& device_info,
      const GpuHloCostAnalysis* analysis = nullptr);

  // Constructs the semantically correct module from the profile.
  // Usually the root instruction of the entry computation is of interest and is
  // directly related to the `profile`d information.
  static std::unique_ptr<HloModule> ConstructModule(
      const HloInstructionProfile& profile);

  // Returns the estimated runtime for a supported `collective` or
  // `collective-permute`.
  absl::StatusOr<absl::Duration> EstimatedRuntime(
      const HloInstruction& instr) const;

 private:
  explicit CollectiveInterpolator(
      ExactInterpolatorMap exact_interpolators,
      FallbackInterpolatorMap fallback_interpolators,
      const se::DeviceDescription& device_info, int num_devices_per_host,
      const GpuHloCostAnalysis* analysis)
      : exact_interpolators_(std::move(exact_interpolators)),
        fallback_interpolators_(std::move(fallback_interpolators)),
        device_info_(device_info),
        num_devices_per_host_(num_devices_per_host),
        analysis_(analysis) {}

  ExactInterpolatorMap exact_interpolators_;
  // Fallback interpolators are only necessary for collective with complex
  // dimensions, e.g. async all-reduce, reduce-scatter, etc. Collective-permute
  // doesn't need fallback interpolators because its
  // category is simple and exact interpolation can cover all cases.
  FallbackInterpolatorMap fallback_interpolators_;
  const se::DeviceDescription& device_info_;
  int num_devices_per_host_;
  const GpuHloCostAnalysis* analysis_;
};

}  // namespace xla::gpu

#endif  // XLA_SERVICE_GPU_MODEL_COLLECTIVE_INTERPOLATOR_H_