// Copyright 2021 The IREE Authors
//
// Licensed 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

// clang-format off
#cmakedefine IREE_STATIC_LIB_QUERY_FN @IREE_STATIC_LIB_QUERY_FN@
#cmakedefine IREE_STATIC_LIB_HDR @IREE_STATIC_LIB_HDR@
#cmakedefine IREE_MODULE_CREATE_FN @IREE_MODULE_CREATE_FN@
#cmakedefine IREE_MODULE_HDR "@IREE_MODULE_HDR@"
#cmakedefine IREE_EMITC_HDR "@IREE_EMITC_HDR@"
#cmakedefine IREE_MODULE_MAIN_FN @IREE_MODULE_MAIN_FN@
#cmakedefine IREE_INPUT_NUM @IREE_INPUT_NUM@
#cmakedefine IREE_INPUT_DIM_MAX @IREE_INPUT_DIM_MAX@
#cmakedefine IREE_INPUT_DIM_ARR @IREE_INPUT_DIM_ARR@
#cmakedefine IREE_INPUT_TYPE @IREE_INPUT_TYPE@
#cmakedefine IREE_INPUT_SHAPE_ARR @IREE_INPUT_SHAPE_ARR@
#cmakedefine IREE_INPUT_SIZE_ARR @IREE_INPUT_SIZE_ARR@
#cmakedefine IREE_HAL_TYPE @IREE_HAL_TYPE@
#cmakedefine IREE_EXE_NAME @IREE_EXE_NAME@
// clang-format on

#include <stdio.h>

#include IREE_STATIC_LIB_HDR

#include "iree/hal/drivers/local_sync/sync_device.h"
#include "iree/hal/local/loaders/static_library_loader.h"
#include "iree/modules/hal/module.h"
#include "iree/runtime/api.h"

#ifdef EMITC_IMPLEMENTATION
#include IREE_EMITC_HDR
#else
#include IREE_MODULE_HDR
#include "iree/vm/bytecode_module.h"
#endif


// A function to create the bytecode or c module.
iree_status_t create_module(iree_vm_instance_t* instance,
                            iree_vm_module_t** out_module) {
#ifdef EMITC_IMPLEMENTATION
  return module_create(instance, iree_vm_instance_allocator(instance),
                       out_module);
#else
  const struct iree_file_toc_t* module_file_toc =
      IREE_MODULE_CREATE_FN;
  iree_const_byte_span_t module_data =
      iree_make_const_byte_span(module_file_toc->data, module_file_toc->size);
  return iree_vm_bytecode_module_create(
      instance, module_data, iree_allocator_null(),
      iree_vm_instance_allocator(instance), out_module);
#endif
}

iree_status_t create_device_with_static_loader(iree_allocator_t host_allocator,
                                               iree_hal_device_t** out_device) {
  // Set paramters for the device created in the next step.
  iree_hal_sync_device_params_t params;
  iree_hal_sync_device_params_initialize(&params);

  // Register the statically linked executable library.
  const iree_hal_executable_library_query_fn_t libraries[] = {
      IREE_STATIC_LIB_QUERY_FN,
  };
  iree_hal_executable_loader_t* library_loader = NULL;
  iree_status_t status = iree_hal_static_library_loader_create(
      IREE_ARRAYSIZE(libraries), libraries,
      iree_hal_executable_import_provider_null(), host_allocator,
      &library_loader);

  // Use the default host allocator for buffer allocations.
  iree_string_view_t identifier = iree_make_cstring_view("local-sync");
  iree_hal_allocator_t* device_allocator = NULL;
  if (iree_status_is_ok(status)) {
    status = iree_hal_allocator_create_heap(identifier, host_allocator,
                                            host_allocator, &device_allocator);
  }

  // Create the device and release the executor and loader afterwards.
  if (iree_status_is_ok(status)) {
    status = iree_hal_sync_device_create(
        identifier, &params, /*loader_count=*/1, &library_loader,
        device_allocator, host_allocator, out_device);
  }

  iree_hal_allocator_release(device_allocator);
  iree_hal_executable_loader_release(library_loader);
  return status;
}


iree_status_t Run() {
  iree_status_t status = iree_ok_status();

  // Instance configuration (this should be shared across sessions).
  iree_runtime_instance_options_t instance_options;
  iree_runtime_instance_options_initialize(IREE_API_VERSION_LATEST,
                                           &instance_options);
  iree_runtime_instance_options_use_all_available_drivers(&instance_options);
  iree_runtime_instance_t* instance = NULL;

  if (iree_status_is_ok(status)) {
    status = iree_runtime_instance_create(&instance_options,
                                          iree_allocator_system(), &instance);
  }

  // Create local device with static loader.
  iree_hal_device_t* device = NULL;
  if (iree_status_is_ok(status)) {
    status = create_device_with_static_loader(iree_allocator_system(), &device);
  }

  // Session configuration (one per loaded module to hold module state).
  iree_runtime_session_options_t session_options;
  iree_runtime_session_options_initialize(&session_options);
  iree_runtime_session_t* session = NULL;
  if (iree_status_is_ok(status)) {
    status = iree_runtime_session_create_with_device(
        instance, &session_options, device,
        iree_runtime_instance_host_allocator(instance), &session);
  }

  // Load bytecode/c module from the embedded data. Append to the session.
  iree_vm_module_t* module = NULL;

  if (iree_status_is_ok(status)) {
    status =
        create_module(iree_runtime_instance_vm_instance(instance), &module);
  }

  if (iree_status_is_ok(status)) {
    status = iree_runtime_session_append_module(session, module);
  }

  // Lookup the entry point function call.
  const char kMainFunctionName[] = IREE_MODULE_MAIN_FN;
  iree_runtime_call_t call;
  memset(&call, 0, sizeof(call));
  if (iree_status_is_ok(status)) {
    status = iree_runtime_call_initialize_by_name(
        session, iree_make_cstring_view(kMainFunctionName), &call);
  }

  // Prepare input.
  iree_hal_dim_t shape[][IREE_INPUT_DIM_MAX] = IREE_INPUT_SHAPE_ARR;
  iree_hal_dim_t size[] = IREE_INPUT_SIZE_ARR;
  iree_hal_dim_t rank[] = IREE_INPUT_DIM_ARR;
  iree_hal_buffer_view_t* arg_buffer_view = NULL;
  IREE_INPUT_TYPE* arg_buffers[IREE_INPUT_NUM] = {NULL};

  for (int i = 0; i < IREE_INPUT_NUM; ++i) {
    if (iree_status_is_ok(status)) {
      arg_buffers[i] = (IREE_INPUT_TYPE*)aligned_alloc(
          sizeof(iree_device_size_t), sizeof(IREE_INPUT_TYPE) * size[i]);
      if (arg_buffers[i] == NULL) {
        status = iree_make_status(IREE_STATUS_RESOURCE_EXHAUSTED);
      }
    }
    if (iree_status_is_ok(status)) {
      status = iree_hal_buffer_view_allocate_buffer(
          iree_hal_device_allocator(device), rank[i], shape[i],
          IREE_HAL_TYPE, IREE_HAL_ENCODING_TYPE_DENSE_ROW_MAJOR,
          (iree_hal_buffer_params_t){
              .type = IREE_HAL_MEMORY_TYPE_DEVICE_LOCAL,
              .usage = IREE_HAL_BUFFER_USAGE_DEFAULT,
          },
          iree_make_const_byte_span((void*)arg_buffers[i],
                                    sizeof(IREE_INPUT_TYPE) * size[i]),
          &arg_buffer_view);
    }

    // Queue buffer views for input.
    if (iree_status_is_ok(status)) {
      status =
          iree_runtime_call_inputs_push_back_buffer_view(&call,
                                                         arg_buffer_view);
    }
    iree_hal_buffer_view_release(arg_buffer_view);
  }

  // Invoke call.
  if (iree_status_is_ok(status)) {
    status = iree_runtime_call_invoke(&call, /*flags=*/0);
  }

  // Cleanup call and buffers.
  iree_runtime_call_deinitialize(&call);
  for (int i = 0; i < IREE_INPUT_NUM; ++i) {
    if (arg_buffers[i] != NULL) {
      free(arg_buffers[i]);
    }
  }

  // Cleanup session and instance.
  iree_hal_device_release(device);
  iree_runtime_session_release(session);
  iree_runtime_instance_release(instance);
  iree_vm_module_release(module);

  return status;
}


int main() {
  const char* kName = IREE_EXE_NAME;
  const iree_status_t result = Run();
  if (!iree_status_is_ok(result)) {
    fprintf(stderr, "%s encountered error:\n", kName);
    iree_status_fprint(stderr, result);
    iree_status_free(result);
    return -1;
  }
  fprintf(stdout, "%s finished successfully\n", kName);
  return 0;
}