Add thread pool as function parameters to C++ API

cpp/include/kvikio/defaults.hpp

@@ -111,7 +111,7 @@ std::tuple<std::string_view, T, bool> getenv_or(
  */
 class defaults {
  private:
-  BS_thread_pool _thread_pool{get_num_threads_from_env()};
+  ThreadPool _thread_pool{get_num_threads_from_env()};
   CompatMode _compat_mode;
   std::size_t _task_size;
   std::size_t _gds_threshold;
@@ -212,7 +212,7 @@ class defaults {
    *
    * @return The default thread pool instance.
    */
-  [[nodiscard]] static BS_thread_pool& thread_pool();
+  [[nodiscard]] static ThreadPool& thread_pool();
 
   /**
    * @brief Get the number of threads in the default thread pool.

cpp/include/kvikio/detail/parallel_operation.hpp

@@ -17,6 +17,7 @@
 #include <kvikio/defaults.hpp>
 #include <kvikio/detail/nvtx.hpp>
 #include <kvikio/error.hpp>
+#include <kvikio/threadpool_wrapper.hpp>
 #include <kvikio/utils.hpp>
 
 namespace kvikio {
@@ -75,6 +76,7 @@ std::future<std::size_t> submit_task(F op,
                                      std::size_t size,
                                      std::size_t file_offset,
                                      std::size_t devPtr_offset,
+                                     ThreadPool* thread_pool    = &defaults::thread_pool(),
                                      std::uint64_t nvtx_payload = 0ull,
                                      nvtx_color_type nvtx_color = NvtxManager::default_color())
 {
@@ -85,7 +87,7 @@ std::future<std::size_t> submit_task(F op,
                                       decltype(file_offset),
                                       decltype(devPtr_offset)>);
 
-  return defaults::thread_pool().submit_task([=] {
+  return thread_pool->submit_task([=] {
     KVIKIO_NVTX_SCOPED_RANGE("task", nvtx_payload, nvtx_color);
     return op(buf, size, file_offset, devPtr_offset);
   });
@@ -101,12 +103,13 @@ std::future<std::size_t> submit_task(F op,
 template <typename F>
 std::future<std::size_t> submit_move_only_task(
   F op_move_only,
+  ThreadPool* thread_pool    = &defaults::thread_pool(),
   std::uint64_t nvtx_payload = 0ull,
   nvtx_color_type nvtx_color = NvtxManager::default_color())
 {
   static_assert(std::is_invocable_r_v<std::size_t, F>);
   auto op_copyable = make_copyable_lambda(std::move(op_move_only));
-  return defaults::thread_pool().submit_task([=] {
+  return thread_pool->submit_task([=] {
     KVIKIO_NVTX_SCOPED_RANGE("task", nvtx_payload, nvtx_color);
     return op_copyable();
   });
@@ -124,6 +127,10 @@ std::future<std::size_t> submit_move_only_task(
  * @param size Number of bytes to read or write.
  * @param file_offset Byte offset to the start of the file.
  * @param task_size Size of each task in bytes.
+ * @param devPtr_offset Offset relative to the `devPtr_base` pointer. This parameter should be used
+ * only with registered buffers.
+ * @param thread_pool Thread pool to use for parallel execution. Defaults to the global default
+ * thread pool.
  * @return A future to be used later to check if the operation has finished its execution.
  */
 template <typename F, typename T>
@@ -133,10 +140,12 @@ std::future<std::size_t> parallel_io(F op,
                                      std::size_t file_offset,
                                      std::size_t task_size,
                                      std::size_t devPtr_offset,
+                                     ThreadPool* thread_pool    = &defaults::thread_pool(),
                                      std::uint64_t call_idx     = 0,
                                      nvtx_color_type nvtx_color = NvtxManager::default_color())
 {
   KVIKIO_EXPECT(task_size > 0, "`task_size` must be positive", std::invalid_argument);
+  KVIKIO_EXPECT(thread_pool != nullptr, "The thread pool must not be nullptr");
   static_assert(std::is_invocable_r_v<std::size_t,
                                       decltype(op),
                                       decltype(buf),
@@ -146,16 +155,17 @@ std::future<std::size_t> parallel_io(F op,
 
   // Single-task guard
   if (task_size >= size || get_page_size() >= size) {
-    return detail::submit_task(op, buf, size, file_offset, devPtr_offset, call_idx, nvtx_color);
+    return detail::submit_task(
+      op, buf, size, file_offset, devPtr_offset, thread_pool, call_idx, nvtx_color);
   }
 
   std::vector<std::future<std::size_t>> tasks;
   tasks.reserve(size / task_size);
 
   // 1) Submit all tasks but the last one. These are all `task_size` sized tasks.
   while (size > task_size) {
-    tasks.push_back(
-      detail::submit_task(op, buf, task_size, file_offset, devPtr_offset, call_idx, nvtx_color));
+    tasks.push_back(detail::submit_task(
+      op, buf, task_size, file_offset, devPtr_offset, thread_pool, call_idx, nvtx_color));
     file_offset += task_size;
     devPtr_offset += task_size;
     size -= task_size;
@@ -170,7 +180,7 @@ std::future<std::size_t> parallel_io(F op,
     }
     return ret;
   };
-  return detail::submit_move_only_task(std::move(last_task), call_idx, nvtx_color);
+  return detail::submit_move_only_task(std::move(last_task), thread_pool, call_idx, nvtx_color);
 }
 
 }  // namespace kvikio

cpp/include/kvikio/file_handle.hpp

@@ -20,6 +20,7 @@
 #include <kvikio/shim/cufile.hpp>
 #include <kvikio/shim/cufile_h_wrapper.hpp>
 #include <kvikio/stream.hpp>
+#include <kvikio/threadpool_wrapper.hpp>
 #include <kvikio/utils.hpp>
 
 namespace kvikio {
@@ -228,6 +229,8 @@ class FileHandle {
    * in the null stream. When in KvikIO's compatibility mode or when accessing host memory, the
    * operation is always default stream ordered like the rest of the non-async CUDA API. In this
    * case, the value of `sync_default_stream` is ignored.
+   * @param thread_pool Thread pool to use for parallel execution. Defaults to the global default
+   * thread pool.
    * @return Future that on completion returns the size of bytes that were successfully read.
    *
    * @note The `std::future` object's `wait()` or `get()` should not be called after the lifetime of
@@ -238,7 +241,8 @@ class FileHandle {
                                  std::size_t file_offset   = 0,
                                  std::size_t task_size     = defaults::task_size(),
                                  std::size_t gds_threshold = defaults::gds_threshold(),
-                                 bool sync_default_stream  = true);
+                                 bool sync_default_stream  = true,
+                                 ThreadPool* thread_pool   = &defaults::thread_pool());
 
   /**
    * @brief Writes specified bytes from device or host memory into the file in parallel.
@@ -265,6 +269,8 @@ class FileHandle {
    * in the null stream. When in KvikIO's compatibility mode or when accessing host memory, the
    * operation is always default stream ordered like the rest of the non-async CUDA API. In this
    * case, the value of `sync_default_stream` is ignored.
+   * @param thread_pool Thread pool to use for parallel execution. Defaults to the global default
+   * thread pool.
    * @return Future that on completion returns the size of bytes that were successfully written.
    *
    * @note The `std::future` object's `wait()` or `get()` should not be called after the lifetime of
@@ -275,7 +281,8 @@ class FileHandle {
                                   std::size_t file_offset   = 0,
                                   std::size_t task_size     = defaults::task_size(),
                                   std::size_t gds_threshold = defaults::gds_threshold(),
-                                  bool sync_default_stream  = true);
+                                  bool sync_default_stream  = true,
+                                  ThreadPool* thread_pool   = &defaults::thread_pool());
 
   /**
    * @brief Reads specified bytes from the file into the device memory asynchronously.

cpp/include/kvikio/mmap.hpp

@@ -9,6 +9,7 @@
 
 #include <kvikio/defaults.hpp>
 #include <kvikio/file_handle.hpp>
+#include <kvikio/threadpool_wrapper.hpp>
 #include <optional>
 
 namespace kvikio {
@@ -162,6 +163,8 @@ class MmapHandle {
    * specified, read starts from `offset` to the end of file
    * @param offset File offset
    * @param task_size Size of each task in bytes
+   * @param thread_pool Thread pool to use for parallel execution. Defaults to the global default
+   * thread pool.
    * @return Future that on completion returns the size of bytes that were successfully read.
    *
    * @exception std::out_of_range if the read region specified by `offset` and `size` is
@@ -174,7 +177,8 @@ class MmapHandle {
   std::future<std::size_t> pread(void* buf,
                                  std::optional<std::size_t> size = std::nullopt,
                                  std::size_t offset              = 0,
-                                 std::size_t task_size           = defaults::task_size());
+                                 std::size_t task_size           = defaults::task_size(),
+                                 ThreadPool* thread_pool         = &defaults::thread_pool());
 };
 
 }  // namespace kvikio

cpp/include/kvikio/remote_handle.hpp

@@ -13,6 +13,7 @@
 
 #include <kvikio/defaults.hpp>
 #include <kvikio/error.hpp>
+#include <kvikio/threadpool_wrapper.hpp>
 #include <kvikio/utils.hpp>
 
 struct curl_slist;
@@ -452,12 +453,15 @@ class RemoteHandle {
    * @param size Number of bytes to read.
    * @param file_offset File offset in bytes.
    * @param task_size Size of each task in bytes.
+   * @param thread_pool Thread pool to use for parallel execution. Defaults to the global default
+   * thread pool.
    * @return Future that on completion returns the size of bytes read, which is always `size`.
    */
   std::future<std::size_t> pread(void* buf,
                                  std::size_t size,
                                  std::size_t file_offset = 0,
-                                 std::size_t task_size   = defaults::task_size());
+                                 std::size_t task_size   = defaults::task_size(),
+                                 ThreadPool* thread_pool = &defaults::thread_pool());
 };
 
 }  // namespace kvikio

cpp/include/kvikio/threadpool_wrapper.hpp

@@ -9,24 +9,9 @@
 
 namespace kvikio {
 
-template <typename pool_type>
-class thread_pool_wrapper : public pool_type {
- public:
-  /**
-   * @brief Construct a new thread pool wrapper.
-   *
-   * @param nthreads The number of threads to use.
-   */
-  thread_pool_wrapper(unsigned int nthreads) : pool_type{nthreads} {}
-
-  /**
-   * @brief Reset the number of threads in the thread pool.
-   *
-   * @param nthreads The number of threads to use.
-   */
-  void reset(unsigned int nthreads) { pool_type::reset(nthreads); }
-};
-
-using BS_thread_pool = thread_pool_wrapper<BS::thread_pool>;
+/**
+ * @brief Thread pool type used for parallel I/O operations.
+ */
+using ThreadPool = BS::thread_pool;
 
 }  // namespace kvikio

cpp/src/defaults.cpp

@@ -172,7 +172,7 @@ bool defaults::is_compat_mode_preferred(CompatMode compat_mode) noexcept
 
 bool defaults::is_compat_mode_preferred() { return is_compat_mode_preferred(compat_mode()); }
 
-BS_thread_pool& defaults::thread_pool() { return instance()->_thread_pool; }
+ThreadPool& defaults::thread_pool() { return instance()->_thread_pool; }
 
 unsigned int defaults::thread_pool_nthreads() { return thread_pool().get_thread_count(); }
 

cpp/src/file_handle.cpp

@@ -19,6 +19,7 @@
 #include <kvikio/error.hpp>
 #include <kvikio/file_handle.hpp>
 #include <kvikio/file_utils.hpp>
+#include <kvikio/threadpool_wrapper.hpp>
 
 namespace kvikio {
 
@@ -148,8 +149,10 @@ std::future<std::size_t> FileHandle::pread(void* buf,
                                            std::size_t file_offset,
                                            std::size_t task_size,
                                            std::size_t gds_threshold,
-                                           bool sync_default_stream)
+                                           bool sync_default_stream,
+                                           ThreadPool* thread_pool)
 {
+  KVIKIO_EXPECT(thread_pool != nullptr, "The thread pool must not be nullptr");
   auto& [nvtx_color, call_idx] = detail::get_next_color_and_call_idx();
   KVIKIO_NVTX_FUNC_RANGE(size, nvtx_color);
   if (is_host_memory(buf)) {
@@ -162,7 +165,7 @@ std::future<std::size_t> FileHandle::pread(void* buf,
         _file_direct_off.fd(), buf, size, file_offset, _file_direct_on.fd());
     };
 
-    return parallel_io(op, buf, size, file_offset, task_size, 0, call_idx, nvtx_color);
+    return parallel_io(op, buf, size, file_offset, task_size, 0, thread_pool, call_idx, nvtx_color);
   }
 
   CUcontext ctx = get_context_from_pointer(buf);
@@ -192,17 +195,26 @@ std::future<std::size_t> FileHandle::pread(void* buf,
     return read(devPtr_base, size, file_offset, devPtr_offset, /* sync_default_stream = */ false);
   };
   auto [devPtr_base, base_size, devPtr_offset] = get_alloc_info(buf, &ctx);
-  return parallel_io(
-    task, devPtr_base, size, file_offset, task_size, devPtr_offset, call_idx, nvtx_color);
+  return parallel_io(task,
+                     devPtr_base,
+                     size,
+                     file_offset,
+                     task_size,
+                     devPtr_offset,
+                     thread_pool,
+                     call_idx,
+                     nvtx_color);
 }
 
 std::future<std::size_t> FileHandle::pwrite(void const* buf,
                                             std::size_t size,
                                             std::size_t file_offset,
                                             std::size_t task_size,
                                             std::size_t gds_threshold,
-                                            bool sync_default_stream)
+                                            bool sync_default_stream,
+                                            ThreadPool* thread_pool)
 {
+  KVIKIO_EXPECT(thread_pool != nullptr, "The thread pool must not be nullptr");
   auto& [nvtx_color, call_idx] = detail::get_next_color_and_call_idx();
   KVIKIO_NVTX_FUNC_RANGE(size, nvtx_color);
   if (is_host_memory(buf)) {
@@ -215,7 +227,7 @@ std::future<std::size_t> FileHandle::pwrite(void const* buf,
         _file_direct_off.fd(), buf, size, file_offset, _file_direct_on.fd());
     };
 
-    return parallel_io(op, buf, size, file_offset, task_size, 0, call_idx, nvtx_color);
+    return parallel_io(op, buf, size, file_offset, task_size, 0, thread_pool, call_idx, nvtx_color);
   }
 
   CUcontext ctx = get_context_from_pointer(buf);
@@ -245,8 +257,15 @@ std::future<std::size_t> FileHandle::pwrite(void const* buf,
     return write(devPtr_base, size, file_offset, devPtr_offset, /* sync_default_stream = */ false);
   };
   auto [devPtr_base, base_size, devPtr_offset] = get_alloc_info(buf, &ctx);
-  return parallel_io(
-    op, devPtr_base, size, file_offset, task_size, devPtr_offset, call_idx, nvtx_color);
+  return parallel_io(op,
+                     devPtr_base,
+                     size,
+                     file_offset,
+                     task_size,
+                     devPtr_offset,
+                     thread_pool,
+                     call_idx,
+                     nvtx_color);
 }
 
 void FileHandle::read_async(void* devPtr_base,

cpp/src/mmap.cpp

@@ -415,10 +415,12 @@ std::size_t MmapHandle::read(void* buf, std::optional<std::size_t> size, std::si
 std::future<std::size_t> MmapHandle::pread(void* buf,
                                            std::optional<std::size_t> size,
                                            std::size_t offset,
-                                           std::size_t task_size)
+                                           std::size_t task_size,
+                                           ThreadPool* thread_pool)
 {
   KVIKIO_EXPECT(task_size <= defaults::bounce_buffer_size(),
                 "bounce buffer size cannot be less than task size.");
+  KVIKIO_EXPECT(thread_pool != nullptr, "The thread pool must not be nullptr");
   auto actual_size = validate_and_adjust_read_args(size, offset);
   if (actual_size == 0) { return make_ready_future(actual_size); }
 
@@ -448,6 +450,7 @@ std::future<std::size_t> MmapHandle::pread(void* buf,
                      offset,
                      task_size,
                      0,  // dst buffer offset initial value
+                     thread_pool,
                      call_idx,
                      nvtx_color);
 }

cpp/src/remote_handle.cpp

@@ -809,8 +809,10 @@ std::size_t RemoteHandle::read(void* buf, std::size_t size, std::size_t file_off
 std::future<std::size_t> RemoteHandle::pread(void* buf,
                                              std::size_t size,
                                              std::size_t file_offset,
-                                             std::size_t task_size)
+                                             std::size_t task_size,
+                                             ThreadPool* thread_pool)
 {
+  KVIKIO_EXPECT(thread_pool != nullptr, "The thread pool must not be nullptr");
   auto& [nvtx_color, call_idx] = detail::get_next_color_and_call_idx();
   KVIKIO_NVTX_FUNC_RANGE(size);
   auto task = [this](void* devPtr_base,
@@ -819,7 +821,7 @@ std::future<std::size_t> RemoteHandle::pread(void* buf,
                      std::size_t devPtr_offset) -> std::size_t {
     return read(static_cast<char*>(devPtr_base) + devPtr_offset, size, file_offset);
   };
-  return parallel_io(task, buf, size, file_offset, task_size, 0, call_idx, nvtx_color);
+  return parallel_io(task, buf, size, file_offset, task_size, 0, thread_pool, call_idx, nvtx_color);
 }
 
 }  // namespace kvikio