tangara-fw/src/tasks/tasks.cpp

/*
 * Copyright 2023 jacqueline <me@jacqueline.id.au>
 *
 * SPDX-License-Identifier: GPL-3.0-only
 */

#include "tasks.hpp"

#include <functional>

#include "esp_heap_caps.h"
#include "freertos/FreeRTOS.h"
#include "freertos/portmacro.h"

#include "memory_resource.hpp"

namespace tasks {

template <Type t>
auto Name() -> std::pmr::string;

template <>
auto Name<Type::kUi>() -> std::pmr::string {
  return "ui";
}
template <>
auto Name<Type::kAudioDecoder>() -> std::pmr::string {
  return "audio_dec";
}
template <>
auto Name<Type::kAudioConverter>() -> std::pmr::string {
  return "audio_conv";
}
template <>
auto Name<Type::kDatabase>() -> std::pmr::string {
  return "db_fg";
}
template <>
auto Name<Type::kDatabaseBackground>() -> std::pmr::string {
  return "db_bg";
}

template <Type t>
auto AllocateStack() -> cpp::span<StackType_t>;

// Decoders often require a very large amount of stack space, since they aren't
// usually written with embedded use cases in mind.
template <>
auto AllocateStack<Type::kAudioDecoder>() -> cpp::span<StackType_t> {
  constexpr std::size_t size = 24 * 1024;
  static StackType_t sStack[size];
  return {sStack, size};
}
// LVGL requires only a relatively small stack. However, it can be allocated in
// PSRAM so we give it a bit of headroom for safety.
template <>
auto AllocateStack<Type::kUi>() -> cpp::span<StackType_t> {
  constexpr std::size_t size = 16 * 1024;
  static StackType_t sStack[size];
  return {sStack, size};
}
template <>
// PCM conversion and resampling uses a very small amount of stack. It works
// entirely with PSRAM-allocated buffers, so no real speed gain from allocating
// it internally.
auto AllocateStack<Type::kAudioConverter>() -> cpp::span<StackType_t> {
  constexpr std::size_t size = 4 * 1024;
  static StackType_t sStack[size];
  return {sStack, size};
}
// Leveldb is designed for non-embedded use cases, where stack space isn't so
// much of a concern. It therefore uses an eye-wateringly large amount of stack.
template <>
auto AllocateStack<Type::kDatabase>() -> cpp::span<StackType_t> {
  std::size_t size = 256 * 1024;
  return {static_cast<StackType_t*>(heap_caps_malloc(size, MALLOC_CAP_SPIRAM)),
          size};
}
template <>
auto AllocateStack<Type::kDatabaseBackground>() -> cpp::span<StackType_t> {
  std::size_t size = 256 * 1024;
  return {static_cast<StackType_t*>(heap_caps_malloc(size, MALLOC_CAP_SPIRAM)),
          size};
}

// 2 KiB in internal ram
// 612 KiB in external ram.

/*
 * Please keep the priorities below in descending order for better readability.
 */

template <Type t>
auto Priority() -> UBaseType_t;

// Realtime audio is the entire point of this device, so give these tasks the
// highest priority.
template <>
auto Priority<Type::kAudioDecoder>() -> UBaseType_t {
  return configMAX_PRIORITIES - 1;
}
template <>
auto Priority<Type::kAudioConverter>() -> UBaseType_t {
  return configMAX_PRIORITIES - 1;
}
// After audio issues, UI jank is the most noticeable kind of scheduling-induced
// slowness that the user is likely to notice or care about. Therefore we place
// this task directly below audio in terms of priority.
template <>
auto Priority<Type::kUi>() -> UBaseType_t {
  return 10;
}
// Database interactions are all inherently async already, due to their
// potential for disk access. The user likely won't notice or care about a
// couple of ms extra delay due to scheduling, so give this task the lowest
// priority.
template <>
auto Priority<Type::kDatabase>() -> UBaseType_t {
  return 2;
}
template <>
auto Priority<Type::kDatabaseBackground>() -> UBaseType_t {
  return 1;
}

template <Type t>
auto WorkerQueueSize() -> std::size_t;

template <>
auto WorkerQueueSize<Type::kDatabase>() -> std::size_t {
  return 8;
}
template <>
auto WorkerQueueSize<Type::kDatabaseBackground>() -> std::size_t {
  return 8;
}

auto PersistentMain(void* fn) -> void {
  auto* function = reinterpret_cast<std::function<void(void)>*>(fn);
  std::invoke(*function);
  assert("persistent task quit!" == 0);
  vTaskDelete(NULL);
}

auto Worker::Main(void* instance) {
  Worker* i = reinterpret_cast<Worker*>(instance);
  while (1) {
    WorkItem item;
    if (xQueueReceive(i->queue_, &item, portMAX_DELAY)) {
      if (item.quit) {
        break;
      } else if (item.fn != nullptr) {
        std::invoke(*item.fn);
        delete item.fn;
      }
    }
  }
  i->is_task_running_.store(false);
  i->is_task_running_.notify_all();
  // Wait for the instance's destructor to delete this task. We do this instead
  // of just deleting ourselves so that it's 100% certain that it's safe to
  // delete or reuse this task's stack.
  while (1) {
    vTaskDelay(portMAX_DELAY);
  }
}

Worker::Worker(const std::pmr::string& name,
               cpp::span<StackType_t> stack,
               std::size_t queue_size,
               UBaseType_t priority)
    : stack_(stack.data()),
      queue_(xQueueCreate(queue_size, sizeof(WorkItem))),
      is_task_running_(true),
      task_buffer_(static_cast<StaticTask_t*>(
          heap_caps_malloc(sizeof(StaticTask_t),
                           MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT))),
      task_(xTaskCreateStatic(&Main,
                              name.c_str(),
                              stack.size(),
                              this,
                              priority,
                              stack_,
                              task_buffer_)) {}

Worker::~Worker() {
  WorkItem item{
      .fn = nullptr,
      .quit = true,
  };
  xQueueSend(queue_, &item, portMAX_DELAY);
  is_task_running_.wait(true);
  vTaskDelete(task_);
  free(stack_);
}

template <>
auto Worker::Dispatch(const std::function<void(void)> fn) -> std::future<void> {
  std::shared_ptr<std::promise<void>> promise =
      std::make_shared<std::promise<void>>();
  WorkItem item{
      .fn = new std::function<void(void)>([=]() {
        std::invoke(fn);
        promise->set_value();
      }),
      .quit = false,
  };
  xQueueSend(queue_, &item, portMAX_DELAY);
  return promise->get_future();
}

}  // namespace tasks