Allocate both ioring buffers and fallback buffer memory at the same time + bug fix

Replacing snprintf(), localtime() and strftime() with custom formating
functions

Reworking progress_thread(), instead of composing the progress printing
with multiple printf() and putchar, we compose it in a buffer and write
it at onse using WriteConsole in Windows and write in Linux

.gitignore

@@ -3,5 +3,13 @@ file_hasher.ilk
 file_hasher.rdi
 file_hasher.exe
 file_hashes.txt
+/Binaries
 file_list.txt
 temp_code.c
+/.cache
+/file_hasher
+/io_uring_test
+/file_hasher
+/io_uring_test
+/compile_commands.json
+/build

CMakeLists.txt

@@ -0,0 +1,284 @@
+cmake_minimum_required(VERSION 3.20)
+project(filehasher
+    VERSION 1.0.0
+    DESCRIPTION "High-performance file hasher with I/O Ring/io_uring support"
+    LANGUAGES C
+)
+
+# ---------------------------------------------------------------------------
+# Force compiler search order
+# ---------------------------------------------------------------------------
+
+# On Windows, prefer clang-cl, then GCC, then Clang
+if(WIN32)
+    # Try to force compiler order if not already set
+    if(NOT CMAKE_C_COMPILER)
+        # Search in preferred order
+        find_program(CLANG_CL_COMPILER NAMES clang-cl)
+        find_program(GCC_COMPILER NAMES gcc)
+        find_program(CLANG_COMPILER NAMES clang)
+        
+        if(CLANG_CL_COMPILER)
+            message(STATUS "Found clang-cl as preferred compiler: ${CLANG_CL_COMPILER}")
+            set(CMAKE_C_COMPILER "${CLANG_CL_COMPILER}" CACHE STRING "" FORCE)
+        elseif(GCC_COMPILER)
+            message(STATUS "Found GCC as fallback compiler: ${GCC_COMPILER}")
+            set(CMAKE_C_COMPILER "${GCC_COMPILER}" CACHE STRING "" FORCE)
+        elseif(CLANG_COMPILER)
+            message(STATUS "Found Clang as last-resort compiler: ${CLANG_COMPILER}")
+            set(CMAKE_C_COMPILER "${CLANG_COMPILER}" CACHE STRING "" FORCE)
+        endif()
+    endif()
+else()
+    # On Linux, prefer GCC, then Clang
+    if(NOT CMAKE_C_COMPILER)
+        find_program(GCC_COMPILER NAMES gcc)
+        find_program(CLANG_COMPILER NAMES clang)
+        
+        if(GCC_COMPILER)
+            message(STATUS "Found GCC as preferred compiler: ${GCC_COMPILER}")
+            set(CMAKE_C_COMPILER "${GCC_COMPILER}" CACHE STRING "" FORCE)
+        elseif(CLANG_COMPILER)
+            message(STATUS "Found Clang as fallback compiler: ${CLANG_COMPILER}")
+            set(CMAKE_C_COMPILER "${CLANG_COMPILER}" CACHE STRING "" FORCE)
+        endif()
+    endif()
+endif()
+
+# Now project() will use the compiler we found
+# However, since we needed project() first to get C support,
+# we check what we actually got
+message(STATUS "Using compiler: ${CMAKE_C_COMPILER} (${CMAKE_C_COMPILER_ID})")
+
+# ---------------------------------------------------------------------------
+# Platform and Compiler Detection
+# ---------------------------------------------------------------------------
+
+if(WIN32)
+    set(PLATFORM_WINDOWS TRUE)
+    set(PLATFORM_NAME "Windows")
+else()
+    set(PLATFORM_LINUX TRUE)
+    set(PLATFORM_NAME "Linux")
+endif()
+
+# Compiler type
+if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+    # Check if it's clang-cl
+    if(CMAKE_C_COMPILER_FRONTEND_VARIANT STREQUAL "MSVC")
+        set(COMPILER_CLANG_CL TRUE)
+        message(STATUS "Detected clang-cl (MSVC-compatible frontend)")
+    else()
+        set(COMPILER_CLANG_GNU TRUE)
+        message(STATUS "Detected Clang (GNU-compatible frontend)")
+    endif()
+elseif(CMAKE_C_COMPILER_ID STREQUAL "GNU")
+    set(COMPILER_GCC TRUE)
+    message(STATUS "Detected GCC")
+elseif(CMAKE_C_COMPILER_ID STREQUAL "MSVC")
+    # We don't want MSVC, but if it's found, warn user
+    message(FATAL_ERROR 
+        "MSVC (cl.exe) detected!\n"
+        "This project requires clang-cl, GCC, or Clang.\n"
+        "Please install one of these compilers or specify manually:\n"
+        "  cmake .. -DCMAKE_C_COMPILER=clang-cl\n"
+        "  cmake .. -DCMAKE_C_COMPILER=gcc\n"
+        "  cmake .. -DCMAKE_C_COMPILER=clang\n"
+    )
+endif()
+
+# ---------------------------------------------------------------------------
+# Build System Selection
+# ---------------------------------------------------------------------------
+
+if(NOT CMAKE_GENERATOR OR CMAKE_GENERATOR STREQUAL "")
+    find_program(NINJA_EXECUTABLE NAMES ninja)
+    if(NINJA_EXECUTABLE)
+        message(STATUS "Using Ninja build system")
+        set(CMAKE_GENERATOR "Ninja")
+    else()
+        message(STATUS "Ninja not found, using default generator: ${CMAKE_GENERATOR}")
+    endif()
+endif()
+
+# ---------------------------------------------------------------------------
+# Source Files
+# ---------------------------------------------------------------------------
+
+set(SOURCES
+    file_hasher.c
+    xxhash.c
+    xxh_x86dispatch.c
+)
+
+# Headers for dependency tracking and IDE
+set(HEADERS
+    arena.h
+    base.h
+    xxhash.h
+    mt_mpmc.h
+)
+
+# ---------------------------------------------------------------------------
+# Create Executable
+# ---------------------------------------------------------------------------
+
+add_executable(${PROJECT_NAME}
+    ${SOURCES}
+    ${HEADERS}
+)
+
+# Include directories
+target_include_directories(${PROJECT_NAME} PRIVATE
+    ${CMAKE_CURRENT_SOURCE_DIR}
+)
+
+# ---------------------------------------------------------------------------
+# Compiler Flags - Exact match to your commands
+# ---------------------------------------------------------------------------
+
+if(PLATFORM_WINDOWS)
+    if(COMPILER_CLANG_CL)
+        # === clang-cl flags ===
+        # Release: /O2
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Release>:/O2>
+        )
+        # Debug: /Zi /Od
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Debug>:/Zi /Od>
+        )
+        # Common warnings
+        target_compile_options(${PROJECT_NAME} PRIVATE /W4)
+        
+    elseif(COMPILER_GCC)
+        # === GCC flags (Windows/MinGW) ===
+        # Release: -O3
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Release>:-O3>
+        )
+        # Debug: -g -O0
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Debug>:-g -O0>
+        )
+        # Common warnings
+        target_compile_options(${PROJECT_NAME} PRIVATE -Wall -Wextra)
+        
+    elseif(COMPILER_CLANG_GNU)
+        # === Clang flags (Windows, GNU frontend) ===
+        # Release: -O3
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Release>:-O3>
+        )
+        # Debug: -g -O0
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Debug>:-g -O0>
+        )
+        # Common warnings
+        target_compile_options(${PROJECT_NAME} PRIVATE -Wall -Wextra)
+    endif()
+    
+    # Windows-specific libraries
+    target_link_libraries(${PROJECT_NAME} PRIVATE
+        kernel32
+    )
+    
+    # Windows-specific defines
+    target_compile_definitions(${PROJECT_NAME} PRIVATE
+        WIN32_LEAN_AND_MEAN
+        _WIN32_WINNT=0x0A00  # Windows 10+
+    )
+    
+    # Set output name with .exe
+    set_target_properties(${PROJECT_NAME} PROPERTIES
+        SUFFIX ".exe"
+    )
+    
+elseif(PLATFORM_LINUX)
+    # === Linux GCC/Clang flags ===
+    if(COMPILER_GCC OR COMPILER_CLANG_GNU)
+        # Release: -O3
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Release>:-O3>
+        )
+        # Debug: -g -O0
+        target_compile_options(${PROJECT_NAME} PRIVATE
+            $<$<CONFIG:Debug>:-g -O0>
+        )
+        # Common warnings
+        target_compile_options(${PROJECT_NAME} PRIVATE -Wall -Wextra)
+    endif()
+    
+    # Linux-specific libraries
+    find_package(Threads REQUIRED)
+    find_library(LIBURING_LIBRARY NAMES uring)
+    
+    if(LIBURING_LIBRARY)
+        message(STATUS "Found liburing: ${LIBURING_LIBRARY}")
+        target_link_libraries(${PROJECT_NAME} PRIVATE
+            Threads::Threads
+            ${LIBURING_LIBRARY}
+        )
+    else()
+        message(FATAL_ERROR "liburing not found! Install liburing-dev or equivalent")
+    endif()
+    
+    # Linux-specific defines
+    target_compile_definitions(${PROJECT_NAME} PRIVATE
+        _GNU_SOURCE
+    )
+endif()
+
+# ---------------------------------------------------------------------------
+# C Standard
+# ---------------------------------------------------------------------------
+
+set_target_properties(${PROJECT_NAME} PROPERTIES
+    C_STANDARD 11
+    C_STANDARD_REQUIRED ON
+    C_EXTENSIONS OFF
+)
+
+# ---------------------------------------------------------------------------
+# Build Configurations
+# ---------------------------------------------------------------------------
+
+# Set default build type if not specified (matching your Release command)
+if(NOT CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE "Release" CACHE STRING 
+        "Choose the type of build: Release or Debug" FORCE)
+    message(STATUS "No build type specified, defaulting to Release")
+endif()
+
+# ---------------------------------------------------------------------------
+# IDE Support
+# ---------------------------------------------------------------------------
+
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+# ---------------------------------------------------------------------------
+# Info Target
+# ---------------------------------------------------------------------------
+
+add_custom_target(info
+    COMMAND ${CMAKE_COMMAND} -E echo "=== Build Configuration ==="
+    COMMAND ${CMAKE_COMMAND} -E echo "Project: ${PROJECT_NAME}"
+    COMMAND ${CMAKE_COMMAND} -E echo "Compiler: ${CMAKE_C_COMPILER} (${CMAKE_C_COMPILER_ID})"
+    COMMAND ${CMAKE_COMMAND} -E echo "Frontend: $<IF:$<BOOL:${COMPILER_CLANG_CL}>,clang-cl,GNU>"
+    COMMAND ${CMAKE_COMMAND} -E echo "Generator: ${CMAKE_GENERATOR}"
+    COMMAND ${CMAKE_COMMAND} -E echo "Build Type: ${CMAKE_BUILD_TYPE}"
+    COMMAND ${CMAKE_COMMAND} -E echo "Platform: ${PLATFORM_NAME}"
+    COMMAND ${CMAKE_COMMAND} -E echo "============================"
+)
+
+# ---------------------------------------------------------------------------
+# Print final configuration
+# ---------------------------------------------------------------------------
+
+message(STATUS "----------------------------------------")
+message(STATUS "Configuration Summary:")
+message(STATUS "  Compiler: ${CMAKE_C_COMPILER}")
+message(STATUS "  Build Type: ${CMAKE_BUILD_TYPE}")
+message(STATUS "  Generator: ${CMAKE_GENERATOR}")
+message(STATUS "  Platform: ${PLATFORM_NAME}")
+message(STATUS "----------------------------------------")

README.md

@@ -1,24 +1,273 @@
 # filehasher
 
-Collects some metadata and hashes files.
+# Presentation  
+Collects some metadata and hashes files. It outputs the path, hash, size, creation and 
+last modification dates and the author in file_hasher.txt.  
+Creation and modification dates and author can be disabled in the config file.  
 
-## Building:  
-### Windows:  
-#### Release:  
-clang-cl /O3 file_hasher.c xxh_x86dispatch.c advapi32.lib  
-clang -O3 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
-gcc -O3 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
+It is a high performance cross platform Windows and Linux compatible program, it uses:  
+  * Multiple threads for scanning and hashing (multi-threading can be disabled in the config file).  
+  * Stores the generated data in thread local configurable arenas that support growing 
+  by committing more memory and chaining blocks.  
+  * Two Multi Producer Multi Consumer queues, one for the scanners and one between the scanners and hashers.  
+  * xxh3_128bits algorithm from xxhash, that supports SIMD instruction sets (SSE2, AVX2, AVX512)
+  and uses a runtime dispatcher to select the best available instruction set.  
+  * IO Ring for asynchronous I/O in Windows and the equivalent io_uring in Linux. 
+  The implementation is event driven, thread local, uses DMA and direct disk I/O, 
+  bypassing the OS cache completely, registered buffers (and registered files in io_uring), 
+  it supports bashing multiple submissions and can handle multiple files at the same time.  
+  It can be disabled in the config file.
+  * Fallback to buffered I/O if there is errors in the IO Ring path.  
 
-#### Debug:  
-clang-cl /Zi /Od file_hasher.c xxh_x86dispatch.c advapi32.lib  
-clang -g -O0 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
-gcc -g -O0 file_hasher.c xxh_x86dispatch.c -ladvapi32 -o file_hasher  
+# Building  
 
-### Linux:  
-#### Release:  
-clang -O3 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
-gcc -O3 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
+## Windows  
+**Requirements**: Make sur to use UCRT64 environment from MSYS2 instead of the standard MinGW environment. 
+UCRT64 uses the modern Universal C Runtime (ucrtbase.dll), which supports the newest APIs, 
+the standard MSYS2 uses the legacy msvcrt.dll and does not support IO Ring.  
+To install:  
+pacman -S mingw-w64-ucrt-x86_64-clang  
+or:  
+pacman -S mingw-w64-ucrt-x86_64-gcc  
 
-#### Debug:  
-clang -g -O0 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
-gcc -g -O0 -pthread file_hasher.c xxh_x86dispatch.c -o file_hasher  
+pacman -Syu  
+And add to path:  
+C:\msys64\ucrt64\bin  
+
+Additionally, to use clang-cl install the latest version of Windows SDK and MSVC, or at least select these in Visual Studio Installer:  
+  * MSVC Build tools fo x64/86.  
+  * C++ Build tools core features.  
+  * MSBuild support for LLVM (clang-cl) toolset.  
+  * Windows Universal C runtime.  
+  * Windows Universal CRT SDK.  
+  * Windows 11 SDK.  
+
+And use the MSVC command prompt or run a script to add MSVC environment variables to current session.  
+Ex: for PowerShell Terminal save as .ps1 (not persistent):
+```ps1
+# Add MS visual studio environment variables
+cmd /c '"C:\Program Files (x86)\Microsoft Visual Studio\18\BuildTools\VC\Auxiliary\Build\vcvarsall.bat" x64 && set' |
+ForEach-Object {
+    if ($_ -match "^(.*?)=(.*)$") {
+        Set-Item -Path "Env:$($matches[1])" -Value $matches[2]
+    }
+}
+```
+
+Optional: to use the build system  
+pacman -S mingw-w64-ucrt-x86_64-cmake  
+The build system uses Ninja and fallsback to make, in Windows it prefers clang-cl > gcc > clang, and in Linux gcc > clang.  
+
+### Using a build system  
+| Command | Description|
+| :--- | :--- |
+| ./build.bat | Build Release with best available compiler |
+| ./build.bat Debug | Build Debug |
+| ./build.bat clean | Clean and build Release |
+| ./build.bat Debug clean | Clean and build Debug |
+
+### Release  
+gcc -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -o filehasher  
+clang -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -o filehasher  
+clang-cl /O2 file_hasher.c xxhash.c xxh_x86dispatch.c  
+
+### Debug  
+gcc -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -o filehasher  
+clang -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -o filehasher  
+clang-cl /Zi /Od file_hasher.c xxhash.c xxh_x86dispatch.c  
+
+## Linux  
+**Requirements**: GCC or clang, optional CMake, Ninja or make.
+
+### Using a build system  
+| Command | Description|
+| :--- | :--- |
+| ./build.sh | Build Release with best available compiler |
+| ./build.sh Debug | Build Debug |
+| ./build.sh clean | Clean and build Release |
+| ./build.sh Debug clean | Clean and build Debug |
+
+### Release  
+gcc -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o filehasher  
+clang -O3 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o filehasher  
+
+### Debug  
+gcc -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o filehasher  
+clang -g -O0 file_hasher.c xxhash.c xxh_x86dispatch.c -pthread -luring -o filehasher  
+
+# Notes about the IO Ring implementations  
+## IO Ring  
+
+### File registration  
+Registering files is a performance optimization that allows the kernel to allocate an array 
+of descriptors/handles to pre-validate and maintain long-term references to file handles. 
+Instead of passing a standard file descriptor/handle with every I/O request, you pass a simple integer 
+index into a pre-registered table.  
+
+The Linux implementation has io_uring_register_files_scarse() to create an empty array of descriptors 
+(initialized with -1) without having to create and initialize it in the user space, and we can 
+use io_uring_register_files_update() to update one or more entries. Windows on the other hand 
+is limited to BuildIoRingRegisterFileHandles() only, so we need to re register the entire array of handles 
+each time. This is why there is a provided macro in config.h to disable or enable it.  
+
+#### *Why Register Files? (The Benefits)*  
+When you use a standard file descriptor in a high-frequency I/O loop, 
+the kernel must perform several "hidden" tasks for every single operation:
+  * Permission Checks: Validating that the process still has the right to read/write 
+  that specific file.  
+  * Reference Counting: Incrementing the file's internal reference count at the start of 
+  the I/O and decrementing it at the end to ensure the file isn't closed while in use.  
+  * Object Lookup: Traversing the internal "file descriptor table" to find the actual 
+  kernel object associated with your integer ID.  
+
+Registering the files performs these checks once at registration time. Subsequent 
+I/O operations skip these steps, significantly reducing CPU overhead and latency, 
+especially when handling thousands of small I/O operations per second.  
+
+#### *Comparison: Linux vs. Windows Implementation*  
+While both systems share the same core concept, their APIs and management styles differ significantly. 
+
+| Feature | Linux (`io_uring`) | Windows (`IoRing`) |
+| :--- | :--- | :--- |
+| **API Call** | `io_uring_register` | `BuildIoRingRegisterFileHandles` |
+| **Registration Method** | Synchronous system call; blocks until the table is set up. | Asynchronous request submitted to the ring like a read/write operation. |
+| **Partial Updates** | Supports `IORING_REGISTER_FILES_UPDATE` to swap specific indices. | No partial updates; a new registration replaces the entire table. |
+| **Scope of Operations** | Extremely broad (files, sockets, timers, signals, etc.). | Primarily focused on file storage (read, write, flush). |
+
+### Completion Wait count and peek  
+To avoid busy waiting when receiving CQEs, we can use io_uring_submit_and_wait() in Linux by entering a wait count, 
+the threads sleep until the count of CQEs are received, in windows the wait_count is present in SubmitIoRing() 
+but is not implemented yet, so we wait with a completion event for a single completion. Another limitation on the completion 
+event is that the kernel will waik up the thread only when receiving the first CQE, after that we need to drain the completion
+queue completely before sleeping again, or we enter an eternal slumber.  
+In the other hand, in Linux we can batch pop completions with io_uring_peek_batch_cqe() + io_uring_cq_advance(), 
+in Windows we can only pop one completion at a time with PopIoRingCompletion() (equivalent to io_uring_peek_cqe() + io_uring_cqe_seen()).  
+To simulate the same behavior as the Linux functions we use a double loop, an outer loop to control how much we wait
+and in inner loop to drain all the available completions.
+
+### Filtering CQEs  
+
+Unlike Linux, The Windows implementation treats buffer and file registration 
+as an asynchronous operation that we submit to the ring, similar to a read or write. 
+Those operations produce CQEs (completion queue entries) that we filter here using 
+cqe.UserData == USERDATA_REGISTER  
+```c
+    if (win_cqe.UserData == USERDATA_REGISTER)
+      continue;
+```
+
+## io_uring  
+
+### Creation flags  
+io_uring provides a lot of configuration flags compared to IO Ring, some 
+of them are at the creation and others during the operations, here what 
+we use in this implementation at creation time and is lacking in the 
+IO Ring implementation.  
+
+  * IORING_SETUP_SINGLE_ISSUER: Since we are using a thread local io_uring, we can 
+  set this flag to remove the atomic operations.  
+  * IORING_SETUP_DEFER_TASKRUN: By default, the kernel sends an interrupts when a CQE 
+  is ready, we use this flag to disable this syscall and wait for a specific number of 
+  CQEs to be ready to group them, this reduces the number of syscall.  
+
+### Memlock limit warning  
+
+```c
+     "WARNING: Buffer registration failed due to memlock limits (ENOMEM).\n"
+     "Increase the limit to solve this warning.\n");
+```
+
+The Memlock limit in Linux restricts the amount of memory that can be
+"locked" into physical RAM using the mlock() family of system calls. This
+prevents the operating system from swapping that memory out to disk.  
+And registering buffers will lock the buffers memory so the hardware
+can access it directly without kernel intervention and prevents the kernel from 
+swapping it to the SSD or HDD.  
+This limit does not apply to a single process, but it applies to what all the runnig processes can lock, so in order 
+to be able to register the buffers, we need to set it to unlimited or increase it to at least:  
+  num_hash_threads * NUM_BUFFERS_PER_THREAD * IORING_BUFFER_SIZE + extra memory reserved for other processes.  
+
+#### *Modifying the Limit*  
+The method for changing the memlock limit depends on whether you are
+managing a user session or a system service.  
+1. For Users and Interactive Sessions  
+To permanently increase the limit for a specific user or group, modify
+the /etc/security/limits.conf file. Add the following lines:  
+
+```conf
+ # Example for a specific user (replace 'username'), unlimited or a custom value in KB
+ username   soft    memlock    unlimited
+ username   hard    memlock    unlimitedhttps://wiki.postgresql.org/wiki/AIO
+```
+```conf
+ # Example for all users
+ *          soft    memlock    unlimited
+ *          hard    memlock    unlimited
+```
+
+Soft Limit: The value the user starts with; can be raised up to the
+hard limit.  
+
+Hard Limit: The absolute maximum; only a privileged user
+(root) can increase this. Values: Can be set in Kilobytes (KB) or as
+unlimited.  
+
+2. For Systemd Services
+Settings in limits.conf do not affect background services managed by
+systemd. To increase the limit for a service, edit its service file
+(e.g., /etc/systemd/system/myservice.service) and add:  
+```conf
+  [Service]
+  LimitMEMLOCK=infinity
+```
+
+#### *Why Register Buffers?*  
+In a standard "unregistered" I/O operation, the kernel must perform several 
+expensive steps for every single read or write:  
+  * Virtual-to-Physical Mapping: The kernel has to translate your application's 
+  virtual memory addresses into physical RAM addresses.  
+  * Page Pinning: The kernel must "pin" the memory pages (using get_user_pages) 
+  to prevent them from being swapped to disk or moved while the hardware 
+  (like your SSD) is writing to them.  
+  * TLB Overhead: Constant mapping and unmapping put pressure on the Translation 
+  Lookaside Buffer (TLB), which can slow down the CPU.  
+
+Registering the buffers performs all of this "pinning" and "mapping" once.  
+
+### Direct I/O: O_DIRECT (Linux) and FILE_FLAG_NO_BUFFERING (Windows)
+
+Modern operating systems normally use a page cache when reading files. This means file 
+data is first loaded into kernel memory and then copied to user space. While this improves 
+performance for many workloads, it introduces extra memory usage and copy overhead.  
+
+Both Linux and Windows provide a way to bypass this cache and perform direct I/O:  
+
+Linux: O_DIRECT
+Windows: FILE_FLAG_NO_BUFFERING
+
+These flags instruct the OS to transfer data directly between disk and user-provided buffers, avoiding the page cache.
+
+#### *Benefits*
+1. Reduced memory overhead
+Avoids polluting the OS page cache
+Especially useful for large sequential reads (e.g. hashing, backups)
+2. Lower CPU usage
+Eliminates extra memory copies between kernel and user space
+3. Predictable performance
+No interference from cache eviction or readahead heuristics
+More consistent throughput for streaming workloads
+4. Better scalability
+Ideal for high-throughput, multi-threaded I/O pipelines
+Prevents cache contention between threads
+5. Avoids double caching
+Important when the application already manages its own buffering
+
+#### *File system compatibility*  
+Not all file systems are compatible with O_DIRECT, if we try to open files residing in an NTFS partition, 
+most of the time it will fail, and some times it opens but the CQEs return with an error code bad 
+descriptor, and it causes some lags.
+
+To address this issue the program falls back to sequential read when the open fails, and falls back to
+buffered sequential hashing if we receive an error in the CQEs. There is also a file system detection 
+that we can enable in the config file, it will enable the collection of the file system in scan_folder() 
+and the file will be opened accordingly, but it costs one additional syscall / directory.

arena.c

@@ -83,8 +83,7 @@ u64 arena_pos_from_ptr(mem_arena *arena, void *ptr) {
 
 void *arena_ptr_from_pos(mem_arena *arena, u64 global_pos) {
   ASSERT(arena);
-  ASSERT(global_pos >= 0);
-  if (!arena || global_pos < 0) {
+  if (!arena) {
     return NULL;
   }
 
@@ -197,7 +196,7 @@ mem_arena *arena_create(arena_params *params) { // mk create
       arena->free_list = arena_create(&(arena_params){
           .reserve_size = MiB(1),
           .commit_size = MiB(1),
-          .align = ARENA_ALIGN,
+          .align = ARENA_CACHE_ALIGN,
           .push_size = sizeof(arena_free_node),
           .allow_free_list = false,
           .free_list = NULL,
@@ -437,12 +436,16 @@ void *arena_push(mem_arena **arena_ptr, u64 size, bool zero) { // mk push
      Commit memory if needed
      ------------------------------------------------------------ */
 
-  if (local_post > selected->commit_pos) {
-    u64 new_commit = ALIGN_UP_POW2(local_post, arena_pagesize());
+  if (local_post > selected->commit_pos -
+                       ALIGN_UP_POW2(sizeof(mem_arena), selected->align)) {
+    u64 new_commit = ALIGN_UP_POW2(
+        local_post + ALIGN_UP_POW2(sizeof(mem_arena), selected->align),
+        arena_pagesize());
     new_commit = MIN(new_commit, selected->reserve_size);
 
     if (!plat_mem_commit((u8 *)selected + selected->commit_pos,
                          new_commit - selected->commit_pos)) {
+      printf("ERROR: Could not commit memory!\n");
       return NULL;
     }
 
@@ -486,7 +489,6 @@ void *arena_free(mem_arena **arena_ptr, u8 **ptr, u64 size) { // mk free
      Find owning block
      ------------------------------------------------------------ */
 
-  mem_arena *selected = arena;
   mem_arena *owner = arena_block_from_ptr(arena, *ptr);
   ASSERT(owner);
   if (!owner) {
@@ -498,7 +500,7 @@ void *arena_free(mem_arena **arena_ptr, u8 **ptr, u64 size) { // mk free
      ------------------------------------------------------------ */
 
   u64 global_offset = arena_pos_from_ptr(arena, *ptr);
-  if (global_offset == -1) {
+  if (global_offset == UINT64_MAX) {
     return NULL;
   }
 
@@ -605,9 +607,6 @@ void *arena_swapback_pop(mem_arena **arena_ptr, u64 index) { // mk swapback
     fprintf(stderr, "ERROR: Swapback pop failed, index out of range");
     return NULL;
   }
-  u8 *owner_base = (u8 *)owner + ALIGN_UP_POW2(sizeof(mem_arena), owner->align);
-  u8 *arena_base = (u8 *)arena + ALIGN_UP_POW2(sizeof(mem_arena), arena->align);
-
   u8 *dst = arena_ptr_from_index(arena, index);
   u8 *src = arena_ptr_from_index(arena, count);
 
@@ -623,6 +622,119 @@ void *arena_swapback_pop(mem_arena **arena_ptr, u64 index) { // mk swapback
 /* ============================================================
    Utilities
    ============================================================ */
+typedef enum arena_trim_flags {
+  ARENA_TRIM_NONE = 0,
+
+  ARENA_TRIM_SPACE = 1 << 0,
+  ARENA_TRIM_TAB = 1 << 1,
+  ARENA_TRIM_LF = 1 << 2,
+  ARENA_TRIM_CR = 1 << 3,
+  ARENA_TRIM_NUL = 1 << 4,
+
+} arena_trim_flags;
+
+u64 arena_trim_string(mem_arena **arena_ptr, char *str, u8 termination_flags) {
+  ASSERT(arena_ptr);
+  ASSERT(*arena_ptr);
+  ASSERT(str);
+
+  if (!arena_ptr || !*arena_ptr || !str) {
+    return 0;
+  }
+
+  mem_arena *arena = *arena_ptr;
+
+  /* ------------------------------------------------------------
+     Find owning block
+     ------------------------------------------------------------ */
+
+  mem_arena *owner = arena_block_from_ptr(arena, (u8 *)str);
+
+  ASSERT(owner);
+  if (!owner) {
+    return 0;
+  }
+
+  /* ------------------------------------------------------------
+     Must be current block
+     ------------------------------------------------------------ */
+
+  if (owner != arena) {
+    fprintf(stderr, "arena_trim_string(): string is not "
+                    "in current arena block.\n");
+    return 0;
+  }
+
+  /* ------------------------------------------------------------
+     Compute string position
+     ------------------------------------------------------------ */
+
+  u64 str_pos = arena_pos_from_ptr(arena, str);
+
+  /* ------------------------------------------------------------
+     Original reserved size
+     ------------------------------------------------------------ */
+
+  u64 allocated_size = arena->pos - str_pos;
+
+  /* ------------------------------------------------------------
+     Compute sizes
+     ------------------------------------------------------------ */
+
+  u64 str_size = strlen(str);
+
+  char *dst = str + str_size;
+  u64 termination_size = 0;
+
+  if (termination_flags & ARENA_TRIM_SPACE) {
+    *dst++ = ' ';
+    termination_size++;
+  }
+
+  if (termination_flags & ARENA_TRIM_TAB) {
+    *dst++ = '\t';
+    termination_size++;
+  }
+
+  if (termination_flags & ARENA_TRIM_CR) {
+    *dst++ = '\r';
+    termination_size++;
+  }
+
+  if (termination_flags & ARENA_TRIM_LF) {
+    *dst++ = '\n';
+    termination_size++;
+  }
+
+  if (termination_flags & ARENA_TRIM_NUL) {
+    *dst++ = '\0';
+    termination_size++;
+  }
+  /* ------------------------------------------------------------
+     Final used size
+     ------------------------------------------------------------ */
+
+  u64 used_size = str_size + termination_size;
+
+  used_size = ALIGN_UP_POW2(used_size, arena->align);
+
+  /* ------------------------------------------------------------
+     Overflow detection
+     ------------------------------------------------------------ */
+
+  if (used_size > allocated_size) {
+    fprintf(stderr, "arena_trim_string(): string overflow "
+                    "detected.\n");
+  }
+
+  /* ------------------------------------------------------------
+     Update arena position
+     ------------------------------------------------------------ */
+
+  arena->pos = str_pos + used_size;
+
+  return used_size;
+}
 
 void *arena_clear(mem_arena **arena_ptr) { // mk clear
 
@@ -804,7 +916,7 @@ mem_arena_temp arena_scratch_get(mem_arena **conflicts, u32 num_conflicts) {
     arena_params params = {
         .reserve_size = MiB(64),
         .commit_size = MiB(1),
-        .align = ARENA_ALIGN,
+        .align = ARENA_CACHE_ALIGN,
         .push_size = 8,
         .allow_free_list = false,
         .allow_swapback = true,

arena.h

@@ -239,7 +239,7 @@ void *arena_ptr_from_index(mem_arena *arena, u64 index);
 */
 
 #define ARENA_HEADER_SIZE (sizeof(mem_arena))
-#define ARENA_ALIGN (sizeof(void *))
+#define ARENA_CACHE_ALIGN (sizeof(void *))
 
 // arena config
 typedef enum arena_growth_policy {

base.h

@@ -1,9 +1,51 @@
 #pragma once
 
+#define _CRT_SECURE_NO_WARNINGS
+
+#if defined(_WIN32) || defined(_WIN64)
+
+#if defined(_MSC_VER)
+#pragma comment(lib, "advapi32.lib")
+#endif
+
+#include <aclapi.h>
+#include <fcntl.h>
+#include <io.h>
+#include <ioringapi.h> // Needs to be included before stdatomic to avoid errors
+#include <ntioring_x.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <windows.h>
+#include <winerror.h>
+
+#elif defined(__linux__)
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+
+#include <dirent.h>
+#include <fcntl.h>
+#include <liburing.h>
+#include <poll.h>
+#include <pthread.h>
+#include <pwd.h>
+#include <sys/eventfd.h>
+#include <sys/resource.h>
+#include <sys/stat.h>
+#include <sys/vfs.h>
+#include <unistd.h>
+#include <sys/mman.h>
+
+#endif
+
 #include <assert.h>
+#include <ctype.h>
 #include <immintrin.h>
+#include <limits.h>
 #include <stdatomic.h>
 #include <stdbool.h>
+#include <stddef.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -11,25 +53,6 @@
 #include <time.h>
 #include <errno.h>
 
-#if defined(_WIN32) || defined(_WIN64)
-#define PLATFORM_WINDOWS 1
-#include <aclapi.h>
-#include <fcntl.h>
-#include <io.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <windows.h>
-
-#define strdup _strdup
-#else
-#include <dirent.h>
-#include <fcntl.h>
-#include <pthread.h>
-#include <pwd.h>
-#include <sys/stat.h>
-#include <unistd.h>
-#endif
-
 /* ------------------------------------------------------------
    Base types
    ------------------------------------------------------------ */
@@ -83,16 +106,16 @@ typedef double f64;
 #define ASSERT(x) assert(x)
 #endif
 
-#define NDEBUG // Comment to enable asserts
+#ifndef NDEBUG
+#define NDEBUG 1 // 0 to enable asserts
+#endif
 
 /* ------------------------------------------------------------
    Some helper functions
    ------------------------------------------------------------ */
 
 #if defined(_WIN32) || defined(_WIN64)
-
 // Memory allocation
-
 static u32 plat_get_pagesize(void) {
   SYSTEM_INFO sysinfo = {0};
   GetSystemInfo(&sysinfo);
@@ -109,58 +132,23 @@ static b32 plat_mem_commit(void *ptr, u64 size) {
   return ret != NULL;
 }
 
-static b32 plat_mem_decommit(void *ptr, u64 size) {
-  return VirtualFree(ptr, size, MEM_DECOMMIT);
-}
+// static b32 plat_mem_decommit(void *ptr, u64 size) { // Comment to prevent warning: unused function
+//   return VirtualFree(ptr, size, MEM_DECOMMIT);
+// }
 
 static b32 plat_mem_release(void *ptr, u64 size) {
   return VirtualFree(ptr, size, MEM_RELEASE);
 }
 
-// Semaphores
-typedef struct plat_sem {
-  HANDLE handle;
-} plat_sem;
-
-static b32 plat_sem_init(plat_sem *s, u32 initial) {
-  s->handle = CreateSemaphore(NULL, initial, LONG_MAX, NULL);
-  return s->handle != NULL;
-}
-
-static void plat_sem_wait(plat_sem *s) {
-  WaitForSingleObject(s->handle, INFINITE);
-}
-
-static b32 plat_sem_trywait(HANDLE sem) {
-  DWORD r = WaitForSingleObject(sem, 0);
-  return r == WAIT_OBJECT_0;
-}
-
-static void plat_sem_post(plat_sem *s, u32 count) {
-  ReleaseSemaphore(s->handle, count, NULL);
-}
-
-static void plat_sem_destroy(plat_sem *s) {
-  if (s->handle) {
-    CloseHandle(s->handle);
-    s->handle = NULL;
-  }
-}
-
 // Sleep
 static void sleep_ms(int ms) { Sleep(ms); }
 
 #elif defined(__linux__)
-
 // Memory allocation
-
 #ifndef _DEFAULT_SOURCE
 #define _DEFAULT_SOURCE
 #endif
 
-#include <sys/mman.h>
-#include <unistd.h>
-
 static u32 plat_get_pagesize(void) { return (u32)sysconf(_SC_PAGESIZE); }
 
 static void *plat_mem_reserve(u64 size) {
@@ -176,46 +164,19 @@ static b32 plat_mem_commit(void *ptr, u64 size) {
   return ret == 0;
 }
 
-static b32 plat_mem_decommit(void *ptr, u64 size) {
-  i32 ret = mprotect(ptr, size, PROT_NONE);
-  if (ret != 0)
-    return false;
-  ret = madvise(ptr, size, MADV_DONTNEED);
-  return ret == 0;
-}
+// static b32 plat_mem_decommit(void *ptr, u64 size) { // Comment to prevent warning: unused function
+//   i32 ret = mprotect(ptr, size, PROT_NONE);
+//   if (ret != 0)
+//     return false;
+//   ret = madvise(ptr, size, MADV_DONTNEED);
+//   return ret == 0;
+// }
 
 static b32 plat_mem_release(void *ptr, u64 size) {
   i32 ret = munmap(ptr, size);
   return ret == 0;
 }
 
-// Semaphores
-#include <semaphore.h>
-
-typedef struct plat_sem {
-  sem_t sem;
-} plat_sem;
-
-static b32 plat_sem_init(plat_sem *s, u32 initial) {
-  return sem_init(&s->sem, 0, initial) == 0;
-}
-
-static void plat_sem_wait(plat_sem *s) {
-  while (sem_wait(&s->sem) == -1 && errno == EINTR) {
-  }
-}
-
-static b32 plat_sem_trywait(sem_t *sem) { return sem_trywait(sem) == 0; }
-
-static void plat_sem_post(plat_sem *s, u32 count) {
-  for (u32 i = 0; i < count; i++) {
-    sem_post(&s->sem);
-  }
-}
-
-static void plat_sem_destroy(plat_sem *s) { sem_destroy(&s->sem); }
-
 // Sleep
 static void sleep_ms(int ms) { usleep(ms * 1000); }
-
 #endif

build.bat

@@ -0,0 +1,172 @@
+@echo off
+setlocal enabledelayedexpansion
+
+set PROJECT_NAME=filehasher
+
+:: ============================================================================
+:: build.bat
+:: ============================================================================
+
+:: Get script directory (project root)
+set SCRIPT_DIR=%~dp0
+set SCRIPT_DIR=%SCRIPT_DIR:~0,-1%
+
+:: ---------------------------------------------------------------------------
+:: Default values
+:: ---------------------------------------------------------------------------
+
+set BUILD_TYPE=Release
+set CLEAN_BUILD=0
+
+:: --------------------------------------------------------------------------
+:: Parse arguments
+:: --------------------------------------------------------------------------
+:parse_args
+if "%~1"=="" goto :main
+
+if /i "%~1"=="Release" (
+    set BUILD_TYPE=Release
+    shift
+    goto :parse_args
+)
+if /i "%~1"=="Debug" (
+    set BUILD_TYPE=Debug
+    shift
+    goto :parse_args
+)
+if /i "%~1"=="clean" (
+    set CLEAN_BUILD=1
+    shift
+    goto :parse_args
+)
+
+echo Unknown argument: %~1
+echo Usage: build [Release^|Debug] [clean]
+exit /b 1
+
+:main
+set BUILD_DIR=%SCRIPT_DIR%\build\windows\%BUILD_TYPE%
+
+echo === Building %PROJECT_NAME% (%BUILD_TYPE%) ===
+
+:: --------------------------------------------------------------------------
+:: Clean if requested
+:: --------------------------------------------------------------------------
+if %CLEAN_BUILD%==1 (
+    echo Cleaning...
+    if exist "%BUILD_DIR%" rmdir /s /q "%BUILD_DIR%" 2>nul
+)
+
+:: --------------------------------------------------------------------------
+:: Create build directory
+:: --------------------------------------------------------------------------
+if not exist "%BUILD_DIR%" mkdir "%BUILD_DIR%"
+pushd "%BUILD_DIR%"
+
+:: --------------------------------------------------------------------------
+:: Compiler selection
+:: --------------------------------------------------------------------------
+set CC=
+
+where clang-cl >nul 2>&1
+if !ERRORLEVEL! equ 0 (
+    echo Compiler: clang-cl ^(preferred^)
+    set "CC=-DCMAKE_C_COMPILER=clang-cl"
+    goto :find_generator
+)
+
+where gcc >nul 2>&1
+if !ERRORLEVEL! equ 0 (
+    echo Compiler: gcc ^(fallback^)
+    set "CC=-DCMAKE_C_COMPILER=gcc"
+    goto :find_generator
+)
+
+where clang >nul 2>&1
+if !ERRORLEVEL! equ 0 (
+    echo Compiler: clang ^(last resort^)
+    set "CC=-DCMAKE_C_COMPILER=clang"
+    goto :find_generator
+)
+
+echo ERROR: No suitable compiler found!
+popd
+exit /b 1
+
+:: --------------------------------------------------------------------------
+:: Generator selection (prefer ninja)
+:: --------------------------------------------------------------------------
+:find_generator
+set GEN=
+where ninja >nul 2>&1
+if !ERRORLEVEL! equ 0 (
+    echo Generator: Ninja
+    set "GEN=-G Ninja"
+) else (
+    echo Generator: Default
+)
+
+:: --------------------------------------------------------------------------
+:: Configure
+:: --------------------------------------------------------------------------
+echo.
+echo Configuring CMake...
+
+:: --------------------------------------------------------------------------
+:: compile_commands.json logic
+:: --------------------------------------------------------------------------
+set EXPORT_COMPILE_COMMANDS=OFF
+
+if /i "%BUILD_TYPE%"=="Release" (
+    if exist "%SCRIPT_DIR%\compile_commands.json" (
+        echo compile_commands.json already exists - skipping generation
+    ) else (
+        echo compile_commands.json will be generated
+        set EXPORT_COMPILE_COMMANDS=ON
+    )
+)
+
+set CMD=cmake "%SCRIPT_DIR%" %GEN% %CC% -DCMAKE_BUILD_TYPE=%BUILD_TYPE% -DCMAKE_EXPORT_COMPILE_COMMANDS=%EXPORT_COMPILE_COMMANDS%
+
+echo !CMD!
+!CMD!
+if !ERRORLEVEL! neq 0 (
+    echo ERROR: Configuration failed
+    popd
+    exit /b 1
+)
+
+:: --------------------------------------------------------------------------
+:: Build
+:: --------------------------------------------------------------------------
+echo.
+echo Building...
+cmake --build . --config %BUILD_TYPE%
+if !ERRORLEVEL! neq 0 (
+    echo ERROR: Build failed
+    popd
+    exit /b 1
+)
+
+:: --------------------------------------------------------------------------
+:: Copy compile_commands.json (only if generated)
+:: --------------------------------------------------------------------------
+if /i "%EXPORT_COMPILE_COMMANDS%"=="ON" (
+    if exist "compile_commands.json" (
+        echo.
+        echo clangd: compile_commands.json generated
+
+        copy /Y "compile_commands.json" "%SCRIPT_DIR%\compile_commands.json" >nul 2>&1
+        if !ERRORLEVEL! equ 0 (
+            echo clangd: Copied to project root
+        ) else (
+            echo clangd: Copy failed
+        )
+    )
+)
+
+popd
+
+echo.
+echo === Build Complete ===
+echo Executable: %BUILD_DIR%\%PROJECT_NAME%.exe

build.sh

@@ -0,0 +1,274 @@
+#!/usr/bin/env bash
+# ============================================================================
+# build.sh - Build script (Linux)
+# Usage: ./build.sh [Release|Debug] [clean]
+#
+# Compiler preference: gcc > clang
+# Build system: Ninja (fallback to Make)
+# ============================================================================
+
+set -euo pipefail
+
+PROJECT_NAME="filehasher"
+
+# ---------------------------------------------------------------------------
+# Colors
+# ---------------------------------------------------------------------------
+readonly RED='\033[0;31m'
+readonly GREEN='\033[0;32m'
+readonly YELLOW='\033[1;33m'
+readonly CYAN='\033[0;36m'
+readonly NC='\033[0m'
+
+# ---------------------------------------------------------------------------
+# Default values
+# ---------------------------------------------------------------------------
+BUILD_TYPE="Release"
+CLEAN_BUILD=0
+
+# ---------------------------------------------------------------------------
+# Parse arguments
+# ---------------------------------------------------------------------------
+while [[ $# -gt 0 ]]; do
+    case "$1" in
+        Release|release)
+            BUILD_TYPE="Release"
+            shift
+            ;;
+        Debug|debug)
+            BUILD_TYPE="Debug"
+            shift
+            ;;
+        clean|-clean|--clean)
+            CLEAN_BUILD=1
+            shift
+            ;;
+        *)
+            echo -e "${RED}Unknown argument: $1${NC}"
+            echo "Usage: $0 [Release|Debug] [clean]"
+            exit 1
+            ;;
+    esac
+done
+
+# ---------------------------------------------------------------------------
+# Setup
+# ---------------------------------------------------------------------------
+readonly BUILD_DIR="build/linux/${BUILD_TYPE}"
+readonly SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+echo -e "${GREEN}=== Building ${PROJECT_NAME} (${BUILD_TYPE}) ===${NC}"
+echo "Project: ${SCRIPT_DIR}"
+
+# ---------------------------------------------------------------------------
+# Clean if requested
+# ---------------------------------------------------------------------------
+if [[ $CLEAN_BUILD -eq 1 ]]; then
+    echo -e "${YELLOW}Cleaning build directory...${NC}"
+    rm -rf "${BUILD_DIR}"
+    echo
+fi
+
+# ---------------------------------------------------------------------------
+# Create build directory
+# ---------------------------------------------------------------------------
+mkdir -p "${BUILD_DIR}"
+cd "${BUILD_DIR}"
+
+# ---------------------------------------------------------------------------
+# Compiler selection (prefer gcc, fallback to clang)
+# ---------------------------------------------------------------------------
+echo -e "${YELLOW}Detecting compiler...${NC}"
+
+CC_BINARY=""
+CC_NAME=""
+
+if command -v gcc &> /dev/null; then
+    CC_BINARY="gcc"
+    CC_VERSION=$(gcc --version | head -n1)
+    CC_NAME="GCC (${CC_VERSION})"
+    echo -e "  ${GREEN}[OK]${NC} Found GCC (preferred): ${CC_VERSION}"
+elif command -v clang &> /dev/null; then
+    CC_BINARY="clang"
+    CC_VERSION=$(clang --version | head -n1)
+    CC_NAME="Clang (${CC_VERSION})"
+    echo -e "  ${YELLOW}[OK]${NC} Found Clang (fallback): ${CC_VERSION}"
+else
+    echo -e "${RED}[FAIL] No suitable compiler found!${NC}"
+    echo "Please install gcc or clang:"
+    echo "  Ubuntu/Debian:  sudo apt install build-essential"
+    echo "  Fedora/RHEL:    sudo dnf install gcc"
+    echo "  Arch:           sudo pacman -S gcc"
+    exit 1
+fi
+
+# ---------------------------------------------------------------------------
+# Check dependencies
+# ---------------------------------------------------------------------------
+echo -e "${YELLOW}Checking dependencies...${NC}"
+
+# Check for liburing
+HAVE_LIBURING=0
+if ldconfig -p | grep -q liburing 2>/dev/null; then
+    HAVE_LIBURING=1
+    echo -e "  ${GREEN}[OK]${NC} Found liburing"
+elif pkg-config --exists liburing 2>/dev/null; then
+    HAVE_LIBURING=1
+    echo -e "  ${GREEN}[OK]${NC} Found liburing (pkg-config)"
+elif [[ -f /usr/lib/liburing.so ]] || [[ -f /usr/lib64/liburing.so ]] || [[ -f /usr/local/lib/liburing.so ]]; then
+    HAVE_LIBURING=1
+    echo -e "  ${GREEN}[OK]${NC} Found liburing (manual detection)"
+else
+    echo -e "${RED}[FAIL] liburing not found!${NC}"
+    echo "Please install liburing-dev:"
+    echo "  Ubuntu/Debian:  sudo apt install liburing-dev"
+    echo "  Fedora/RHEL:    sudo dnf install liburing-devel"
+    echo "  Arch:           sudo pacman -S liburing"
+    exit 1
+fi
+
+# Check for pthreads
+# Check if pthreads is available (either in ldconfig or merged into libc)
+if ldconfig -p | grep -q libpthread 2>/dev/null || ldd --version | grep -qP '2\.(3[4-9]|[4-9][0-9])'; then
+    echo -e "  ${GREEN}[OK]${NC} Found pthreads (merged or standalone)"
+else
+    echo -e "  ${YELLOW}[NOTE]${NC} pthreads not found, attempting link"
+fi
+echo
+
+# ---------------------------------------------------------------------------
+# Generator selection (prefer ninja)
+# ---------------------------------------------------------------------------
+echo -e "${YELLOW}Selecting build system...${NC}"
+
+GENERATOR=""
+GENERATOR_NAME=""
+
+if command -v ninja &> /dev/null; then
+    GENERATOR="Ninja"
+    GENERATOR_NAME="Ninja"
+    echo -e "  ${GREEN}[OK]${NC} Using Ninja"
+elif command -v make &> /dev/null; then
+    GENERATOR="Unix Makefiles"
+    GENERATOR_NAME="Make"
+    echo -e "  ${YELLOW}[OK]${NC} Ninja not found, using Make"
+else
+    echo -e "${RED}[FAIL] No build system found!${NC}"
+    echo "Please install ninja or make:"
+    echo "  Ubuntu/Debian:  sudo apt install ninja-build"
+    echo "  Fedora/RHEL:    sudo dnf install ninja-build"
+    echo "  Arch:           sudo pacman -S ninja"
+    exit 1
+fi
+
+echo
+
+# ---------------------------------------------------------------------------
+# Configure
+# ---------------------------------------------------------------------------
+echo -e "${YELLOW}Configuring CMake...${NC}"
+
+# --------------------------------------------------------------------------
+# compile_commands.json logic
+# --------------------------------------------------------------------------
+
+EXPORT_COMPILE_COMMANDS=OFF
+
+if [[ "$BUILD_TYPE" == "Release" ]]; then
+    if [[ -f "${SCRIPT_DIR}/compile_commands.json" ]]; then
+        echo -e "  compile_commands.json already exists - skipping generation"
+    else
+        echo -e "  compile_commands.json will be generated"
+        EXPORT_COMPILE_COMMANDS=ON
+    fi
+fi
+
+echo -e "  Build type: ${BUILD_TYPE}"
+echo -e "  Compiler:   ${CC_NAME}"
+echo -e "  Generator:  ${GENERATOR_NAME}"
+
+cmake "${SCRIPT_DIR}" \
+    -G "${GENERATOR}" \
+    -DCMAKE_BUILD_TYPE="${BUILD_TYPE}" \
+    -DCMAKE_C_COMPILER="${CC_BINARY}" \
+    -DCMAKE_EXPORT_COMPILE_COMMANDS=${EXPORT_COMPILE_COMMANDS}
+
+if [[ $? -ne 0 ]]; then
+    echo -e "${RED}CMake configuration failed!${NC}"
+    exit 1
+fi
+
+echo -e "${GREEN}Configuration successful!${NC}"
+echo
+
+# ---------------------------------------------------------------------------
+# Build
+# ---------------------------------------------------------------------------
+echo -e "${YELLOW}Building...${NC}"
+
+# Get number of CPU cores
+if command -v nproc &> /dev/null; then
+    CORES=$(nproc)
+else
+    CORES=4
+fi
+
+cmake --build . --config "${BUILD_TYPE}" --parallel "${CORES}"
+
+if [[ $? -ne 0 ]]; then
+    echo -e "${RED}Build failed!${NC}"
+    exit 1
+fi
+
+echo -e "${GREEN}Build successful!${NC}"
+echo
+
+# ---------------------------------------------------------------------------
+# Verify output
+# ---------------------------------------------------------------------------
+cd "${SCRIPT_DIR}"
+
+if [[ -f "${BUILD_DIR}/${PROJECT_NAME}" ]]; then
+    echo -e "${GREEN}Executable: ${BUILD_DIR}/${PROJECT_NAME}${NC}"
+
+    if command -v file &> /dev/null; then
+        echo -e "  Type: $(file -b ${BUILD_DIR}/${PROJECT_NAME})"
+    fi
+
+    if command -v du &> /dev/null; then
+        echo -e "  Size: $(du -h ${BUILD_DIR}/${PROJECT_NAME} | cut -f1)"
+    fi
+elif [[ -f "${BUILD_DIR}/${PROJECT_NAME}.exe" ]]; then
+    echo -e "${GREEN}Executable: ${BUILD_DIR}/${PROJECT_NAME}.exe${NC}"
+else
+    echo -e "${YELLOW}Note: Could not locate executable${NC}"
+    echo "Checking build directory:"
+    find "${BUILD_DIR}" -type f -executable 2>/dev/null || echo "  No executables found"
+fi
+
+# ---------------------------------------------------------------------------
+# Summary
+# ---------------------------------------------------------------------------
+echo
+echo -e "${CYAN}=== Build Complete ===${NC}"
+echo
+echo -e "${YELLOW}Build Information:${NC}"
+echo -e "  Configuration: ${BUILD_TYPE}"
+echo -e "  Compiler:      ${CC_NAME}"
+echo -e "  Generator:     ${GENERATOR_NAME}"
+echo -e "  Output:        ${BUILD_DIR}/"
+
+# ---------------------------------------------------------------------------
+# Copy compile_commands.json for clangd
+# ---------------------------------------------------------------------------
+if [[ "${EXPORT_COMPILE_COMMANDS}" == "ON" ]]; then
+    if [[ -f "${BUILD_DIR}/compile_commands.json" ]]; then
+        echo -e "  clangd:        compile_commands.json generated"
+
+        cp "${BUILD_DIR}/compile_commands.json" "${SCRIPT_DIR}/compile_commands.json"
+        echo -e "  clangd:        Copied to project root"
+    fi
+fi
+
+echo
+echo -e "${GREEN}Ready to run: ./${BUILD_DIR}/${PROJECT_NAME}${NC}"

changelog.txt

@@ -14,7 +14,7 @@ v3.2: Making the lock free MPMC queue growable
 Add padding to avoir false sharing
 Add sleep() and SwitchToThread() to limit spinning
 
-v3.3: Fix bug slots used before initialization,compare and swap is protecting updating committed, but it is not protecting the memory initialization. Adding atomic_flag commit_lock to protect against that
+v3.3: Fix bug slots used before initialization, compare and swap is protecting updating committed, but it is not protecting the memory initialization. Adding atomic_flag commit_lock to protect against that
 Fix bug multiple threads committing at the same time, fixed by using atomic_flag commit_lock and re-checking committed after acquiring the lock
 Reorder helper functions
 
@@ -49,3 +49,10 @@ Fixing user prompt parsing
 4.5: Porting to linux
 Reorganising the code
 Improving the scan function
+
+5.0: Implementing the IO Ring for windows and ui_uring for linux instead of buffered hashing, huge performance gains. The IO Ring is event driven, thread local, uses DMA and direct disk I/O, bypassing the OS cache completely, registered buffers (and registered files in io_uring), it supports bashing multiple submissions and can handle multiple files at the same time.
+Hashing small files using XXH3_128bits() instead of the streaming pipeline(XXH3_128bits_reset(), XXH3_128bits_update(), XXH3_128bits_digest()), this reduses the overhead of creating a state and digest, coupled with the IO Ring it improves the hashing of small files whose size is inferior to the size of IO Ring buffers
+fixing the xxh_x86dispatch warnings
+Updating the progress printing function
+Implementing a config file
+Writing the README file

config.h

@@ -0,0 +1,32 @@
+
+#define FILE_HASHES_TXT "file_hashes.txt"
+
+// Metadata selection
+#define FILE_TIMES 1 // created and modified time
+#define FILE_OWNER 1
+
+#define MULTI_THREADING 1
+#define READ_BLOCK KiB(64)
+
+// -------------------- IO Ring Configuration ----------------------
+#define USE_IORING 1
+
+#if USE_IORING
+#define IORING_BUFFER_SIZE KiB(256)
+#define NUM_BUFFERS_PER_THREAD 32
+#define MAX_ACTIVE_FILES 16
+#define MAX_WAIT_COUNT (NUM_BUFFERS_PER_THREAD / 2)
+
+#define SUBMIT_TIMEOUT_MS 10000
+#define IORING_DEBUG_PRINTS 0
+#define IORING_DEBUG_STATS 0
+
+#if defined(_WIN32) || defined(_WIN64)
+#define USE_REGISTERED_FILES 1
+
+#elif defined(__linux__)
+#define USE_REGISTERED_FILES 1
+#define CHECK_FILE_SYSTEM 0
+
+#endif
+#endif

experiments/io_ring_test.c

@@ -0,0 +1,147 @@
+#pragma once
+
+#include <ioringapi.h>
+#include <ntioring_x.h>
+// #include "ioringapi.c"
+#include <winerror.h>
+
+// Initialize I/O Ring
+HIORING io_ring_init(void) {
+
+  // if (!io_ring_load_functions()) {
+  //   printf("[I/O Ring] Failed to load functions\n");
+  //   return NULL;
+  // }
+
+  IORING_CAPABILITIES caps;
+  ZeroMemory(&caps, sizeof(caps));
+
+  HRESULT hr = QueryIoRingCapabilities(&caps);
+  if (FAILED(hr)) {
+    printf("[I/O Ring] QueryIoRingCapabilities failed: 0x%08lx\n", hr);
+    return NULL;
+  }
+
+  // printf("[I/O Ring] MaxVersion=%d, MaxSubmission=%u, MaxCompletion=%u\n",
+  //        (int)caps.MaxVersion, caps.MaxSubmissionQueueSize,
+  // caps.MaxCompletionQueueSize);
+
+  if (caps.MaxVersion < IORING_VERSION_1) {
+    printf("[I/O Ring] Version too old\n");
+    return NULL;
+  }
+
+  IORING_CREATE_FLAGS flags = {0};
+  HIORING ring = NULL;
+
+  // hr = CreateIoRing(IORING_VERSION_1, flags, 256, 512, &ring);
+  hr = CreateIoRing(caps.MaxVersion, flags, 256, 512, &ring);
+  if (FAILED(hr)) {
+    printf("[I/O Ring] CreateIoRing failed: 0x%08lx\n", hr);
+    return NULL;
+  }
+
+  // printf("[I/O Ring] Created successfully\n");
+
+  // Check if read operation is supported
+
+  // HRESULT io_ring_support = IsIoRingOpSupported(ring, IORING_OP_READ);
+  // if (io_ring_support == S_FALSE) {
+  //   printf("[I/O Ring] Not supported, %ld /n", io_ring_support);
+  // }
+
+  // Get ring info
+  IORING_INFO info;
+  ZeroMemory(&info, sizeof(info));
+  GetIoRingInfo(ring, &info);
+  // printf("[I/O Ring] Submission: %u, Completion: %u\n",
+  // info.SubmissionQueueSize, info.CompletionQueueSize);
+
+  return ring;
+}
+
+void io_ring_cleanup(HIORING ring) {
+  if (ring) {
+    CloseIoRing(ring);
+    // printf("[I/O Ring] Closed\n");
+  }
+}
+
+// Read file using I/O Ring
+int io_ring_read_file(HIORING ring, HANDLE hFile, void *buffer, DWORD size,
+                      UINT64 offset) {
+
+  IORING_HANDLE_REF file_ref = IoRingHandleRefFromHandle(hFile);
+  IORING_BUFFER_REF buf_ref = IoRingBufferRefFromPointer(buffer);
+
+  HRESULT hr = BuildIoRingReadFile(ring, file_ref, buf_ref, size, offset,
+                                   (UINT_PTR)buffer, IOSQE_FLAGS_NONE);
+
+  if (FAILED(hr))
+    return -1;
+
+  UINT32 submitted = 0;
+  hr = SubmitIoRing(ring, 1, INFINITE, &submitted);
+  if (FAILED(hr) || submitted == 0)
+    return -1;
+
+  for (;;) {
+    IORING_CQE cqe;
+    hr = PopIoRingCompletion(ring, &cqe);
+
+    if (FAILED(hr))
+      continue;
+
+    if (cqe.UserData != (UINT_PTR)buffer)
+      continue;
+
+    if (FAILED(cqe.ResultCode))
+      return -1;
+
+    return (int)cqe.Information;
+  }
+}
+
+// Test function
+void test_io_ring(void) {
+  printf("\n=== Testing I/O Ring ===\n");
+
+  HIORING ring = io_ring_init();
+  if (!ring) {
+    printf("I/O Ring not available\n");
+    return;
+  }
+
+  // Create test file
+  HANDLE hFile = CreateFileA("test.txt", GENERIC_READ | GENERIC_WRITE, 0, NULL,
+                             CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);
+  if (hFile != INVALID_HANDLE_VALUE) {
+    char test_data[] =
+        "Hello, I/O Ring! This is a test of the Windows I/O Ring API.";
+    DWORD written;
+    WriteFile(hFile, test_data, sizeof(test_data), &written, NULL);
+    CloseHandle(hFile);
+  }
+
+  // Read using I/O Ring
+  hFile = CreateFileA("test.txt", GENERIC_READ, FILE_SHARE_READ, NULL,
+                      OPEN_EXISTING, FILE_FLAG_OVERLAPPED, NULL);
+  if (hFile != INVALID_HANDLE_VALUE) {
+    char buffer[512] = {0};
+    int bytes = io_ring_read_file(ring, hFile, buffer, sizeof(buffer), 0);
+    if (bytes > 0) {
+      printf("Read %d bytes: %s\n", bytes, buffer);
+    } else {
+      printf("Failed to read file\n");
+    }
+    CloseHandle(hFile);
+  } else {
+    printf("Failed to open test file\n");
+  }
+
+  // Cleanup
+  DeleteFileA("test.txt");
+  io_ring_cleanup(ring);
+
+  printf("=== Test complete ===\n\n");
+}

experiments/io_uring_test.c

@@ -0,0 +1,721 @@
+/*
+# Compile
+gcc -o io_uring_test io_uring_test.c -luring
+
+# Run
+./io_uring_test
+   */
+#include "base.h"
+#include <stdint.h>
+#define _GNU_SOURCE
+#include <errno.h>
+#include <fcntl.h>
+#include <liburing.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+#define TEST_FILE "test_io_uring.txt"
+#define BUFFER_SIZE 4096
+#define NUM_BUFFERS 4
+#define NUM_REGISTERED_FILES 8 // Maximum number of files to register
+
+// Colors for output
+#define COLOR_GREEN "\033[0;32m"
+#define COLOR_RED "\033[0;31m"
+#define COLOR_YELLOW "\033[0;33m"
+#define COLOR_BLUE "\033[0;34m"
+#define COLOR_RESET "\033[0m"
+
+// Test result tracking
+typedef struct {
+  int passed;
+  int failed;
+} TestResults;
+
+static void print_success(const char *step) {
+  printf(COLOR_GREEN "[✓] SUCCESS: %s" COLOR_RESET "\n", step);
+}
+
+static void print_failure(const char *step, const char *error) {
+  printf(COLOR_RED "[✗] FAILED: %s - %s" COLOR_RESET "\n", step, error);
+}
+
+static void print_info(const char *msg) {
+  printf(COLOR_BLUE "[i] INFO: %s" COLOR_RESET "\n", msg);
+}
+
+static void print_step(const char *step) {
+  printf(COLOR_YELLOW "\n>>> Testing: %s" COLOR_RESET "\n", step);
+}
+
+// Create a test file with known content
+static int create_test_file(const char *filename, const char *content) {
+  FILE *f = fopen(filename, "w");
+  if (!f) {
+    perror("Failed to create test file");
+    return -1;
+  }
+
+  fprintf(f, "%s", content);
+  fclose(f);
+
+  printf("  Created test file: %s\n", filename);
+  return 0;
+}
+
+// Test 1: Create io_uring instance
+static int test_io_uring_create(struct io_uring *ring, TestResults *results) {
+  print_step("io_uring creation");
+
+  int ret = io_uring_queue_init(256, ring, 0);
+  if (ret < 0) {
+    print_failure("io_uring_queue_init", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("io_uring instance created");
+  results->passed++;
+  return 0;
+}
+
+// Test 2: Register buffers
+static int test_register_buffers(struct io_uring *ring, void **buffers,
+                                 struct iovec *iovs, TestResults *results) {
+  print_step("Buffer registration");
+
+  // Allocate and prepare buffers
+  size_t total_size = BUFFER_SIZE * NUM_BUFFERS;
+  *buffers = aligned_alloc(4096, total_size); // Page-aligned for O_DIRECT
+  if (!*buffers) {
+    print_failure("Buffer allocation", strerror(errno));
+    results->failed++;
+    return -1;
+  }
+
+  // Initialize iovecs
+  for (int i = 0; i < NUM_BUFFERS; i++) {
+    iovs[i].iov_base = (char *)*buffers + (i * BUFFER_SIZE);
+    iovs[i].iov_len = BUFFER_SIZE;
+    memset(iovs[i].iov_base, 0, BUFFER_SIZE);
+  }
+
+  int ret = io_uring_register_buffers(ring, iovs, NUM_BUFFERS);
+  if (ret < 0) {
+    print_failure("io_uring_register_buffers", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Buffers registered successfully");
+  results->passed++;
+  return 0;
+}
+
+// Test 2b: Register files
+static int test_register_files(struct io_uring *ring, int *fds, int num_fds,
+                               TestResults *results) {
+  print_step("File registration");
+
+  if (num_fds == 0) {
+    print_info("No files to register");
+    results->passed++;
+    return 0;
+  }
+
+  int ret = io_uring_register_files(ring, fds, num_fds);
+  if (ret < 0) {
+    // File registration might not be supported on all kernels
+    if (ret == -EOPNOTSUPP || ret == -EINVAL) {
+      print_info("File registration not supported on this kernel, skipping");
+      results->passed++;
+      return 0;
+    }
+    print_failure("io_uring_register_files", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  printf("  Registered %d files\n", num_fds);
+  print_success("Files registered successfully");
+  results->passed++;
+  return 0;
+}
+
+// Test 3: Open file
+static int test_open_file(const char *filename, int *fd, bool use_direct,
+                          TestResults *results) {
+  print_step("File opening");
+
+  // Get file size
+  struct stat st;
+  if (stat(filename, &st) != 0) {
+    print_failure("stat", strerror(errno));
+    results->failed++;
+    return -1;
+  }
+
+  int page_size = plat_get_pagesize();
+  size_t file_size = st.st_size;
+
+  printf("  File: %s\n", filename);
+  printf("  File size: %zu bytes\n", file_size);
+  printf("  Page size: %d bytes\n", page_size);
+
+  if (file_size % page_size != 0) {
+    printf("  Extending read size from %zu to %zu bytes\n", file_size,
+           ALIGN_UP_POW2(file_size, page_size));
+  }
+
+  // Try to open with specified flags
+  int flags = O_RDONLY;
+  if (use_direct) {
+    flags |= O_DIRECT;
+  }
+
+  *fd = open(filename, flags);
+  if (*fd < 0) {
+    if (use_direct) {
+      print_info("O_DIRECT failed, trying without it");
+      *fd = open(filename, O_RDONLY);
+      if (*fd < 0) {
+        print_failure("open", strerror(errno));
+        results->failed++;
+        return -1;
+      }
+      print_info("Using buffered I/O (O_DIRECT not available)");
+    } else {
+      print_failure("open", strerror(errno));
+      results->failed++;
+      return -1;
+    }
+  } else {
+    const char *io_type = use_direct ? "O_DIRECT" : "buffered I/O";
+    printf("  File opened with %s\n", io_type);
+    print_success("File opened successfully");
+  }
+
+  results->passed++;
+  return 0;
+}
+
+// Test 4: Build and submit read operation (using registered file)
+static int test_submit_read_registered(struct io_uring *ring, int file_index,
+                                       struct iovec *iovs, int buffer_id,
+                                       uint64_t user_data, size_t file_size,
+                                       TestResults *results) {
+  print_step("Building and submitting read operation (registered file)");
+
+  u32 page_size = plat_get_pagesize();
+  size_t read_size = BUFFER_SIZE;
+
+  // For O_DIRECT, ensure read size is sector-aligned
+  if (read_size > file_size) {
+    read_size = ALIGN_UP_POW2(file_size, page_size);
+    printf("  Adjusted read size to %zu bytes for O_DIRECT alignment\n",
+           read_size);
+  }
+
+  struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
+  if (!sqe) {
+    print_failure("io_uring_get_sqe", "No available SQE");
+    results->failed++;
+    return -1;
+  }
+
+  // Use fixed file descriptor
+  io_uring_prep_read_fixed(sqe, file_index, iovs[buffer_id].iov_base, read_size,
+                           0, buffer_id);
+  io_uring_sqe_set_data64(sqe, user_data);
+
+  int ret = io_uring_submit(ring);
+  if (ret < 0) {
+    print_failure("io_uring_submit", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  printf("  Using registered file index: %d\n", file_index);
+  print_success("Read operation submitted successfully (registered file)");
+  results->passed++;
+  return 0;
+}
+
+// Test 4b: Build and submit read operation (using fd directly)
+static int test_submit_read(struct io_uring *ring, int fd, struct iovec *iovs,
+                            int buffer_id, uint64_t user_data,
+                            TestResults *results) {
+  print_step("Building and submitting read operation");
+
+  // Get file size for proper alignment
+  struct stat st;
+  if (fstat(fd, &st) != 0) {
+    print_failure("fstat", strerror(errno));
+    results->failed++;
+    return -1;
+  }
+
+  u32 page_size = plat_get_pagesize();
+  size_t file_size = st.st_size;
+  size_t read_size = BUFFER_SIZE;
+
+  // For O_DIRECT, ensure read size is sector-aligned
+  if (read_size > file_size) {
+    read_size = ALIGN_UP_POW2(file_size, page_size);
+    printf("  Adjusted read size to %zu bytes for O_DIRECT alignment\n",
+           read_size);
+  }
+
+  struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
+  if (!sqe) {
+    print_failure("io_uring_get_sqe", "No available SQE");
+    results->failed++;
+    return -1;
+  }
+
+  // Prepare read operation using registered buffer
+  io_uring_prep_read_fixed(sqe, fd, iovs[buffer_id].iov_base, read_size, 0,
+                           buffer_id);
+  io_uring_sqe_set_data64(sqe, user_data);
+
+  int ret = io_uring_submit(ring);
+  if (ret < 0) {
+    print_failure("io_uring_submit", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Read operation submitted successfully");
+  results->passed++;
+  return 0;
+}
+
+// Test 5: Wait for completion
+static int test_wait_completion(struct io_uring *ring,
+                                struct io_uring_cqe **cqe,
+                                TestResults *results) {
+  print_step("Waiting for completion");
+
+  int ret = io_uring_wait_cqe(ring, cqe);
+  if (ret < 0) {
+    print_failure("io_uring_wait_cqe", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Completion received");
+  results->passed++;
+  return 0;
+}
+
+// Test 6: Process completion
+static int test_process_completion(struct io_uring_cqe *cqe,
+                                   uint64_t expected_user_data,
+                                   TestResults *results) {
+  print_step("Processing completion");
+
+  uint64_t user_data = io_uring_cqe_get_data64(cqe);
+  int res = cqe->res;
+
+  printf("  Completion data:\n");
+  printf("    User data: %lu (expected: %lu)\n", user_data, expected_user_data);
+  printf("    Result: %d bytes read\n", res);
+
+  if (user_data != expected_user_data) {
+    print_failure("User data mismatch",
+                  "User data doesn't match expected value");
+    results->failed++;
+    return -1;
+  }
+
+  if (res < 0) {
+    print_failure("Read operation", strerror(-res));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Completion processed successfully");
+  results->passed++;
+  return res; // Return number of bytes read
+}
+
+// Test 7: Verify read data
+static int test_verify_data(struct iovec *iovs, int buffer_id, int bytes_read,
+                            const char *expected_content,
+                            TestResults *results) {
+  print_step("Data verification");
+
+  char *data = (char *)iovs[buffer_id].iov_base;
+
+  printf("  Read data (first 200 chars):\n");
+  printf("  ---\n");
+  for (int i = 0; i < bytes_read && i < 200; i++) {
+    putchar(data[i]);
+  }
+  if (bytes_read > 200)
+    printf("...");
+  printf("\n  ---\n");
+
+  // Check if data is not empty
+  if (bytes_read == 0) {
+    print_failure("Data verification", "No data read");
+    results->failed++;
+    return -1;
+  }
+
+  // Check if data contains expected content
+  if (expected_content && strstr(data, expected_content) == NULL) {
+    print_failure("Data verification", "Expected content not found");
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Data verified successfully");
+  results->passed++;
+  return 0;
+}
+
+// Test 8: Test multiple concurrent reads
+static int test_concurrent_reads(struct io_uring *ring, int fd,
+                                 struct iovec *iovs, TestResults *results) {
+  print_step("Concurrent reads test");
+
+  int num_reads = 3;
+  int submitted = 0;
+
+  // Submit multiple reads
+  for (int i = 0; i < num_reads; i++) {
+    struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
+    if (!sqe) {
+      print_failure("Getting SQE for concurrent read", "No available SQE");
+      results->failed++;
+      return -1;
+    }
+
+    off_t offset = i * 100; // Read from different offsets
+    io_uring_prep_read_fixed(sqe, fd, iovs[i].iov_base, BUFFER_SIZE, offset, i);
+    io_uring_sqe_set_data64(sqe, i);
+    submitted++;
+  }
+
+  int ret = io_uring_submit(ring);
+  if (ret != submitted) {
+    char msg[64];
+    snprintf(msg, sizeof(msg), "Expected %d, got %d", submitted, ret);
+
+    print_failure("Submitting concurrent reads", msg);
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Concurrent reads submitted");
+
+  // Wait for and process completions
+  for (int i = 0; i < submitted; i++) {
+    struct io_uring_cqe *cqe;
+    ret = io_uring_wait_cqe(ring, &cqe);
+    if (ret < 0) {
+      print_failure("Waiting for concurrent read completion", strerror(-ret));
+      results->failed++;
+      return -1;
+    }
+
+    uint64_t user_data = io_uring_cqe_get_data64(cqe);
+    int res = cqe->res;
+
+    printf("  Concurrent read %lu completed: %d bytes read\n", user_data, res);
+    io_uring_cqe_seen(ring, cqe);
+  }
+
+  print_success("Concurrent reads completed successfully");
+  results->passed++;
+  return 0;
+}
+
+// Test 9: Test file registration with multiple files
+static int test_file_registration(struct io_uring *ring, TestResults *results) {
+  print_step("File registration with multiple files");
+
+  // Create multiple test files
+  const char *filenames[] = {"test_file1.txt", "test_file2.txt",
+                             "test_file3.txt"};
+  const char *contents[] = {"Content of file 1: Hello World!",
+                            "Content of file 2: io_uring is fast!",
+                            "Content of file 3: Registered files test."};
+
+  int fds[3];
+  int num_files = 3;
+
+  // Create and open files
+  for (int i = 0; i < num_files; i++) {
+    if (create_test_file(filenames[i], contents[i]) != 0) {
+      results->failed++;
+      return -1;
+    }
+
+    fds[i] = open(filenames[i], O_RDONLY);
+    if (fds[i] < 0) {
+      print_failure("Opening file for registration", strerror(errno));
+      // Close previously opened files
+      for (int j = 0; j < i; j++)
+        close(fds[j]);
+      results->failed++;
+      return -1;
+    }
+  }
+
+  // Register files
+  int ret = io_uring_register_files(ring, fds, num_files);
+  if (ret < 0) {
+    if (ret == -EOPNOTSUPP || ret == -EINVAL) {
+      print_info("File registration not supported, skipping test");
+      results->passed++;
+    } else {
+      print_failure("io_uring_register_files", strerror(-ret));
+      results->failed++;
+    }
+    // Cleanup
+    for (int i = 0; i < num_files; i++) {
+      close(fds[i]);
+      remove(filenames[i]);
+    }
+    return (ret == -EOPNOTSUPP || ret == -EINVAL) ? 0 : -1;
+  }
+
+  print_success("Multiple files registered successfully");
+
+  // Read from each registered file using fixed operations
+  for (int i = 0; i < num_files; i++) {
+    struct iovec iov;
+    char buf[256] = {0};
+    iov.iov_base = buf;
+    iov.iov_len = sizeof(buf);
+
+    // Register a single buffer for this test
+    ret = io_uring_register_buffers(ring, &iov, 1);
+    if (ret < 0) {
+      print_failure("Registering buffer for file test", strerror(-ret));
+      break;
+    }
+
+    struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
+    if (!sqe) {
+      print_failure("Getting SQE for registered file", "No available SQE");
+      break;
+    }
+
+    // Use fixed file and fixed buffer
+    io_uring_prep_read_fixed(sqe, i, iov.iov_base, strlen(contents[i]), 0, 0);
+    io_uring_sqe_set_data64(sqe, i);
+
+    ret = io_uring_submit(ring);
+    if (ret < 0) {
+      print_failure("Submitting read for registered file", strerror(-ret));
+      break;
+    }
+
+    struct io_uring_cqe *cqe;
+    ret = io_uring_wait_cqe(ring, &cqe);
+    if (ret < 0) {
+      print_failure("Waiting for registered file read", strerror(-ret));
+      break;
+    }
+
+    if (cqe->res < 0) {
+      print_failure("Reading registered file", strerror(-cqe->res));
+      io_uring_cqe_seen(ring, cqe);
+      break;
+    }
+
+    printf("  File %d: Read %d bytes: %.*s\n", i, cqe->res, cqe->res, buf);
+    io_uring_cqe_seen(ring, cqe);
+
+    // Unregister buffer for next iteration
+    io_uring_unregister_buffers(ring);
+  }
+
+  // Cleanup files
+  io_uring_unregister_files(ring);
+  for (int i = 0; i < num_files; i++) {
+    close(fds[i]);
+    remove(filenames[i]);
+  }
+
+  print_success("File registration test completed");
+  results->passed++;
+  return 0;
+}
+
+// Cleanup function
+static void cleanup(struct io_uring *ring, int *fds, int num_fds,
+                    void *buffers) {
+  if (fds) {
+    io_uring_unregister_files(ring);
+    for (int i = 0; i < num_fds; i++) {
+      if (fds[i] >= 0)
+        close(fds[i]);
+    }
+  }
+  if (buffers) {
+    io_uring_unregister_buffers(ring);
+    free(buffers);
+  }
+  io_uring_queue_exit(ring);
+  remove(TEST_FILE);
+}
+
+int main() {
+  TestResults results = {0, 0};
+  struct io_uring ring;
+  int fd = -1;
+  int registered_fds[1] = {-1}; // For registered file test
+  void *buffers = NULL;
+  struct iovec iovs[NUM_BUFFERS];
+
+  printf(COLOR_BLUE "\n========================================\n");
+  printf("    io_uring Test Suite with File Registration\n");
+  printf("========================================\n" COLOR_RESET);
+
+  // Create main test file
+  const char *test_content =
+      "Hello, io_uring! This is a test file for async I/O operations.\n"
+      "Line 2: Testing reads with registered buffers.\n"
+      "Line 3: The quick brown fox jumps over the lazy dog.\n"
+      "Line 4: ABCDEFGHIJKLMNOPQRSTUVWXYZ\n"
+      "Line 5: 0123456789\n";
+
+  if (create_test_file(TEST_FILE, test_content) != 0) {
+    return 1;
+  }
+
+  // Test 1: Create io_uring
+  if (test_io_uring_create(&ring, &results) != 0) {
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+
+  // Test 2: Register buffers
+  if (test_register_buffers(&ring, &buffers, iovs, &results) != 0) {
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+
+  // Test 3: Open file
+  if (test_open_file(TEST_FILE, &fd, true, &results) != 0) {
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+
+  // Test 4: Submit read with direct fd
+  uint64_t test_user_data = 12345;
+  if (test_submit_read(&ring, fd, iovs, 0, test_user_data, &results) != 0) {
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+
+  // Test 5: Wait for completion
+  struct io_uring_cqe *cqe;
+  if (test_wait_completion(&ring, &cqe, &results) != 0) {
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+
+  // Test 6: Process completion
+  int bytes_read = test_process_completion(cqe, test_user_data, &results);
+  if (bytes_read < 0) {
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+  io_uring_cqe_seen(&ring, cqe);
+
+  // Test 7: Verify data
+  if (test_verify_data(iovs, 0, bytes_read, "io_uring", &results) != 0) {
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+
+  // Close the file for file registration test
+  close(fd);
+
+  // Reopen and register the file
+  registered_fds[0] = open(TEST_FILE, O_RDONLY);
+  if (registered_fds[0] < 0) {
+    print_failure("Reopening file for registration", strerror(errno));
+    cleanup(&ring, NULL, 0, buffers);
+    return 1;
+  }
+
+  // Test 2b: Register files
+  if (test_register_files(&ring, registered_fds, 1, &results) != 0) {
+    cleanup(&ring, registered_fds, 1, buffers);
+    return 1;
+  }
+
+  // Get file size for the registered read test
+  struct stat st;
+  stat(TEST_FILE, &st);
+
+  // Test 4b: Submit read using registered file
+  test_user_data = 67890;
+  if (test_submit_read_registered(&ring, 0, iovs, 0, test_user_data, st.st_size,
+                                  &results) != 0) {
+    cleanup(&ring, registered_fds, 1, buffers);
+    return 1;
+  }
+
+  // Wait for and process completion
+  if (test_wait_completion(&ring, &cqe, &results) != 0) {
+    cleanup(&ring, registered_fds, 1, buffers);
+    return 1;
+  }
+
+  bytes_read = test_process_completion(cqe, test_user_data, &results);
+  if (bytes_read < 0) {
+    cleanup(&ring, registered_fds, 1, buffers);
+    return 1;
+  }
+  io_uring_cqe_seen(&ring, cqe);
+
+  // Verify data from registered file read
+  if (test_verify_data(iovs, 0, bytes_read, "io_uring", &results) != 0) {
+    cleanup(&ring, registered_fds, 1, buffers);
+    return 1;
+  }
+
+  // Test 8: Concurrent reads
+  if (test_concurrent_reads(&ring, registered_fds[0], iovs, &results) != 0) {
+    cleanup(&ring, registered_fds, 1, buffers);
+    return 1;
+  }
+
+  // Test 9: File registration with multiple files (requires new ring)
+  cleanup(&ring, registered_fds, 1, buffers);
+  buffers = NULL;
+  registered_fds[0] = -1;
+
+  if (test_io_uring_create(&ring, &results) != 0) {
+    return 1;
+  }
+
+  test_file_registration(&ring, &results);
+
+  // Cleanup the second ring
+  io_uring_queue_exit(&ring);
+
+  // Print summary
+  printf(COLOR_BLUE "\n========================================\n");
+  printf("           TEST SUMMARY\n");
+  printf("========================================\n" COLOR_RESET);
+  printf("  Total tests: %d\n", results.passed + results.failed);
+  printf(COLOR_GREEN "  Passed: %d\n" COLOR_RESET, results.passed);
+  if (results.failed > 0) {
+    printf(COLOR_RED "  Failed: %d\n" COLOR_RESET, results.failed);
+  } else {
+    printf(COLOR_GREEN "  ✓ ALL TESTS PASSED!\n" COLOR_RESET);
+  }
+
+  return results.failed > 0 ? 1 : 0;
+}

experiments/io_uring_test2.c

@@ -0,0 +1,397 @@
+/*
+# Compile
+gcc -o io_uring_test io_uring_test2.c -luring
+
+# Run
+./io_uring_test
+   */
+#include "base.h"
+#include <stdint.h>
+#define _GNU_SOURCE
+#include <errno.h>
+#include <fcntl.h>
+#include <liburing.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+#define BUFFER_SIZE 4096
+#define NUM_BUFFERS 4
+#define NUM_REGISTERED_FILES 3 // Test with 3 files
+
+// Colors for output
+#define COLOR_GREEN "\033[0;32m"
+#define COLOR_RED "\033[0;31m"
+#define COLOR_YELLOW "\033[0;33m"
+#define COLOR_BLUE "\033[0;34m"
+#define COLOR_RESET "\033[0m"
+
+// Test result tracking
+typedef struct {
+  int passed;
+  int failed;
+} TestResults;
+
+static void print_success(const char *step) {
+  printf(COLOR_GREEN "[✓] SUCCESS: %s" COLOR_RESET "\n", step);
+}
+
+static void print_failure(const char *step, const char *error) {
+  printf(COLOR_RED "[✗] FAILED: %s - %s" COLOR_RESET "\n", step, error);
+}
+
+static void print_info(const char *msg) {
+  printf(COLOR_BLUE "[i] INFO: %s" COLOR_RESET "\n", msg);
+}
+
+static void print_step(const char *step) {
+  printf(COLOR_YELLOW "\n>>> Testing: %s" COLOR_RESET "\n", step);
+}
+
+static int create_test_file(const char *filename, const char *content) {
+  FILE *f = fopen(filename, "w");
+  if (!f) {
+    perror("Failed to create test file");
+    return -1;
+  }
+
+  fprintf(f, "%s", content);
+  fclose(f);
+
+  printf("  Created test file: %s\n", filename);
+  return 0;
+}
+
+// Test 1: Create io_uring instance
+static int test_io_uring_create(struct io_uring *ring, TestResults *results) {
+  print_step("io_uring creation");
+
+  int ret = io_uring_queue_init(256, ring, 0);
+  if (ret < 0) {
+    print_failure("io_uring_queue_init", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("io_uring instance created");
+  results->passed++;
+  return 0;
+}
+
+// Test 2: Register buffers
+static int test_register_buffers(struct io_uring *ring, void **buffers,
+                                 struct iovec *iovs, TestResults *results) {
+  print_step("Buffer registration");
+
+  size_t total_size = BUFFER_SIZE * NUM_BUFFERS;
+  *buffers = aligned_alloc(4096, total_size);
+  if (!*buffers) {
+    print_failure("Buffer allocation", strerror(errno));
+    results->failed++;
+    return -1;
+  }
+
+  for (int i = 0; i < NUM_BUFFERS; i++) {
+    iovs[i].iov_base = (char *)*buffers + (i * BUFFER_SIZE);
+    iovs[i].iov_len = BUFFER_SIZE;
+    memset(iovs[i].iov_base, 0, BUFFER_SIZE);
+  }
+
+  int ret = io_uring_register_buffers(ring, iovs, NUM_BUFFERS);
+  if (ret < 0) {
+    print_failure("io_uring_register_buffers", strerror(-ret));
+    results->failed++;
+    return -1;
+  }
+
+  print_success("Buffers registered successfully");
+  results->passed++;
+  return 0;
+}
+
+// Test 3: Register files sparse (empty table)
+static int test_register_files_sparse(struct io_uring *ring, unsigned nr_files,
+                                      TestResults *results) {
+  print_step("Sparse file registration (empty table)");
+
+  int ret = io_uring_register_files_sparse(ring, nr_files);
+  if (ret < 0) {
+    if (ret == -EINVAL) {
+      print_info(
+          "io_uring_register_files_sparse not supported (kernel < 5.19)");
+      print_info("Trying regular file registration with invalid fds...");
+
+      // Fallback: register with invalid fds
+      int *invalid_fds = calloc(nr_files, sizeof(int));
+      if (!invalid_fds) {
+        print_failure("Allocating invalid fds array", "Out of memory");
+        results->failed++;
+        return -1;
+      }
+
+      for (int i = 0; i < nr_files; i++) {
+        invalid_fds[i] = -1; // Mark all as invalid
+      }
+
+      ret = io_uring_register_files(ring, invalid_fds, nr_files);
+      free(invalid_fds);
+
+      if (ret < 0) {
+        print_failure("Regular file registration also failed", strerror(-ret));
+        results->failed++;
+        return -1;
+      }
+      print_success("File table registered (regular, with invalid fds)");
+    } else {
+      print_failure("io_uring_register_files_sparse", strerror(-ret));
+      results->failed++;
+      return -1;
+    }
+  } else {
+    printf("  Registered empty file table with %u slots\n", nr_files);
+    print_success("Sparse file table created");
+  }
+
+  results->passed++;
+  return 0;
+}
+
+// Test 4: Update file slot and read from it
+static int test_file_read_loop(struct io_uring *ring, struct iovec *iovs,
+                               const char **filenames,
+                               const char **expected_contents, int num_files,
+                               TestResults *results) {
+  print_step("File slot update and read loop");
+
+  int *fds = calloc(num_files, sizeof(int));
+  if (!fds) {
+    print_failure("Allocating fd array", "Out of memory");
+    results->failed++;
+    return -1;
+  }
+
+  // Open all files first
+  for (int i = 0; i < num_files; i++) {
+    fds[i] = open(filenames[i], O_RDONLY);
+    if (fds[i] < 0) {
+      print_failure("Opening file", filenames[i]);
+      results->failed++;
+      // Close already opened files
+      for (int j = 0; j < i; j++)
+        close(fds[j]);
+      free(fds);
+      return -1;
+    }
+    printf("  Opened %s (fd=%d)\n", filenames[i], fds[i]);
+  }
+
+  // Test loop: update slot, submit read, verify
+  for (int slot = 0; slot < num_files; slot++) {
+    printf("\n  --- Testing slot %d with file '%s' ---\n", slot,
+           filenames[slot]);
+
+    // Update the file registration for this slot
+    printf("  Updating slot %d with fd %d...\n", slot, fds[slot]);
+    int ret = io_uring_register_files_update(ring, slot, &fds[slot], 1);
+
+    if (ret < 0) {
+      print_failure("File registration update", strerror(-ret));
+      results->failed++;
+      continue;
+    }
+    printf("  Slot update result: %d (expected 1)\n", ret);
+
+    // Get file size for read size calculation
+    struct stat st;
+    if (fstat(fds[slot], &st) != 0) {
+      print_failure("fstat", strerror(errno));
+      results->failed++;
+      continue;
+    }
+
+    size_t file_size = st.st_size;
+    size_t read_size = BUFFER_SIZE;
+
+    // Adjust read size for O_DIRECT if needed
+    int page_size = plat_get_pagesize();
+    if (read_size > file_size) {
+      read_size = ALIGN_UP_POW2(file_size, page_size);
+    }
+
+    printf("  File size: %zu, read size: %zu\n", file_size, read_size);
+
+    // Clear buffer for this test
+    memset(iovs[0].iov_base, 0, BUFFER_SIZE);
+
+    // Submit read using registered file
+    struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
+    if (!sqe) {
+      print_failure("Getting SQE", "No available SQE");
+      results->failed++;
+      continue;
+    }
+
+    // Use slot index with fixed file flag
+    io_uring_prep_read_fixed(sqe, slot, iovs[0].iov_base, read_size, 0, 0);
+    sqe->flags |= IOSQE_FIXED_FILE;
+    io_uring_sqe_set_data64(sqe, 100 + slot); // Unique user_data per slot
+
+    ret = io_uring_submit(ring);
+    if (ret < 0) {
+      print_failure("Submitting read", strerror(-ret));
+      results->failed++;
+      continue;
+    }
+    printf("  Submitted read (1 SQE)\n");
+
+    // Wait for completion
+    struct io_uring_cqe *cqe;
+    ret = io_uring_wait_cqe(ring, &cqe);
+    if (ret < 0) {
+      print_failure("Waiting for completion", strerror(-ret));
+      results->failed++;
+      continue;
+    }
+
+    // Process completion
+    uint64_t user_data = io_uring_cqe_get_data64(cqe);
+    int bytes_read = cqe->res;
+
+    printf("  Completion: user_data=%lu, result=%d\n", (unsigned long)user_data,
+           bytes_read);
+
+    if (bytes_read < 0) {
+      print_failure("Read operation", strerror(-bytes_read));
+      results->failed++;
+      io_uring_cqe_seen(ring, cqe);
+      continue;
+    }
+
+    if (user_data != 100 + slot) {
+      print_failure("User data mismatch", "Wrong user_data value");
+      results->failed++;
+      io_uring_cqe_seen(ring, cqe);
+      continue;
+    }
+
+    // Verify the data
+    char *data = (char *)iovs[0].iov_base;
+    printf("  Data read (%d bytes): %.*s\n", bytes_read,
+           bytes_read < 100 ? bytes_read : 100, data);
+
+    if (strstr(data, expected_contents[slot]) == NULL) {
+      print_failure("Data verification",
+                    "Expected content not found in read data");
+      results->failed++;
+    } else {
+      print_success("Data verified successfully");
+      results->passed++;
+    }
+
+    io_uring_cqe_seen(ring, cqe);
+
+    // Invalidate the slot after use (mark as -1)
+    int invalid_fd = -1;
+    ret = io_uring_register_files_update(ring, slot, &invalid_fd, 1);
+    if (ret < 0) {
+      printf("  Warning: Could not invalidate slot %d: %s\n", slot,
+             strerror(-ret));
+    }
+  }
+
+  // Close all files
+  for (int i = 0; i < num_files; i++) {
+    if (fds[i] >= 0)
+      close(fds[i]);
+  }
+  free(fds);
+
+  return 0;
+}
+
+int main() {
+  TestResults results = {0, 0};
+  struct io_uring ring;
+  void *buffers = NULL;
+  struct iovec iovs[NUM_BUFFERS];
+
+  printf(COLOR_BLUE "\n========================================\n");
+  printf("    io_uring Sparse File Registration Test\n");
+  printf("========================================\n" COLOR_RESET);
+
+  // Define test files and their content
+  const char *filenames[] = {"test_file_0.txt", "test_file_1.txt",
+                             "test_file_2.txt"};
+
+  const char *contents[] = {
+      "This is file 0: Hello World! The quick brown fox jumps over the lazy "
+      "dog.",
+      "This is file 1: io_uring is awesome for async I/O operations!",
+      "This is file 2: Testing sparse file registration with multiple files."};
+
+  const char *expected_substrings[] = {"Hello World", "io_uring is awesome",
+                                       "sparse file registration"};
+
+  int num_files = 3;
+
+  // Create all test files
+  print_info("Creating test files...");
+  for (int i = 0; i < num_files; i++) {
+    if (create_test_file(filenames[i], contents[i]) != 0) {
+      return 1;
+    }
+  }
+
+  // Test 1: Create io_uring
+  if (test_io_uring_create(&ring, &results) != 0) {
+    goto cleanup_files;
+  }
+
+  // Test 2: Register buffers
+  if (test_register_buffers(&ring, &buffers, iovs, &results) != 0) {
+    io_uring_queue_exit(&ring);
+    goto cleanup_files;
+  }
+
+  // Test 3: Register empty file table (sparse)
+  if (test_register_files_sparse(&ring, num_files, &results) != 0) {
+    io_uring_unregister_buffers(&ring);
+    free(buffers);
+    io_uring_queue_exit(&ring);
+    goto cleanup_files;
+  }
+
+  // Test 4: Loop through files, update slots, read and verify
+  test_file_read_loop(&ring, iovs, filenames, expected_substrings, num_files,
+                      &results);
+
+  // Cleanup
+  io_uring_unregister_files(&ring);
+  io_uring_unregister_buffers(&ring);
+  free(buffers);
+  io_uring_queue_exit(&ring);
+
+cleanup_files:
+  // Remove test files
+  for (int i = 0; i < num_files; i++) {
+    remove(filenames[i]);
+  }
+
+  // Print summary
+  int total = results.passed + results.failed;
+  printf(COLOR_BLUE "\n========================================\n");
+  printf("           TEST SUMMARY\n");
+  printf("========================================\n" COLOR_RESET);
+  printf("  Total tests: %d\n", total);
+  printf(COLOR_GREEN "  Passed: %d\n" COLOR_RESET, results.passed);
+  if (results.failed > 0) {
+    printf(COLOR_RED "  Failed: %d\n" COLOR_RESET, results.failed);
+    printf(COLOR_RED "\n  ✗ SOME TESTS FAILED!\n" COLOR_RESET);
+  } else {
+    printf(COLOR_GREEN "\n  ✓ ALL TESTS PASSED!\n" COLOR_RESET);
+  }
+
+  return results.failed > 0 ? 1 : 0;
+}

experiments/ioringapi.c

@@ -0,0 +1,285 @@
+#pragma once
+#include <stdio.h>
+#include <windows.h>
+#include <winnt.h>
+
+// Forward declarations
+typedef struct IORING_HANDLE_REF IORING_HANDLE_REF;
+typedef struct IORING_BUFFER_REF IORING_BUFFER_REF;
+typedef void *HIORING;
+
+/* --------------------- Types declaration --------------------- */
+typedef enum IORING_CREATE_ADVISORY_FLAGS {
+  IORING_CREATE_ADVISORY_FLAGS_NONE,
+  IORING_CREATE_SKIP_BUILDER_PARAM_CHECKS
+} IORING_CREATE_ADVISORY_FLAGS;
+// Specifies advisory flags for creating an I/O ring with a call to
+// CreateIoRing.
+
+typedef enum IORING_CREATE_REQUIRED_FLAGS {
+  IORING_CREATE_REQUIRED_FLAGS_NONE
+} IORING_CREATE_REQUIRED_FLAGS;
+// Specifies required flags for creating an I/O ring with a call to
+// CreateIoRing.
+
+typedef enum IORING_REF_KIND {
+  IORING_REF_RAW = 0,
+  IORING_REF_REGISTERED = 1,
+} IORING_REF_KIND;
+// Specifies the type of an IORING_HANDLE_REF structure.
+
+typedef enum IORING_SQE_FLAGS {
+  IOSQE_FLAGS_NONE,
+  IOSQE_FLAGS_DRAIN_PRECEDING_OPS
+} IORING_SQE_FLAGS;
+// Specifies kernel behavior options for I/O ring submission queue entries
+
+// IORING_REGISTERED_BUFFER structure
+typedef struct IORING_REGISTERED_BUFFER {
+  UINT32 Index;
+  UINT32 Offset;
+} IORING_REGISTERED_BUFFER;
+
+// IORING_HANDLE_REF
+struct IORING_HANDLE_REF {
+  IORING_REF_KIND Kind;
+  union {
+    HANDLE Handle;
+    UINT32 Index;
+  } HandleUnion;
+};
+// Represents a reference to a file handle used in an I/O ring operation
+
+// IORING_BUFFER_REF
+struct IORING_BUFFER_REF {
+  IORING_REF_KIND Kind;
+  union {
+    void *Address;
+    IORING_REGISTERED_BUFFER IndexAndOffset;
+  } BufferUnion;
+};
+
+typedef struct IORING_BUFFER_INFO {
+  void *Address;
+  UINT32 Length;
+} IORING_BUFFER_INFO;
+
+// IORING_BUFFER_REF represents a reference to a buffer used in an I/O ring
+// operation
+
+// IORING_VERSION enumeration
+typedef enum IORING_VERSION {
+  IORING_VERSION_INVALID = 0,
+  IORING_VERSION_1 = 1,
+  IORING_VERSION_2 = 2,
+  IORING_VERSION_3 = 3,
+  IORING_VERSION_4 = 4,
+} IORING_VERSION;
+
+typedef enum IORING_FEATURE_FLAGS {
+  IORING_FEATURE_FLAGS_NONE = 0,
+  IORING_FEATURE_UM_EMULATION = 1
+} IORING_FEATURE_FLAGS;
+
+// IORING_CAPABILITIES structure
+typedef struct IORING_CAPABILITIES {
+  IORING_VERSION MaxVersion;
+  UINT32 MaxSubmissionQueueSize;
+  UINT32 MaxCompletionQueueSize;
+  IORING_FEATURE_FLAGS FeatureFlags;
+} IORING_CAPABILITIES;
+// Represents the IORING API capabilities.
+
+// IORING_CQE structure
+typedef struct IORING_CQE {
+  UINT_PTR UserData;
+  HRESULT ResultCode;
+  ULONG_PTR Information;
+} IORING_CQE;
+// Represents a completed I/O ring queue entry.
+
+// IORING_CREATE_FLAGS structure
+typedef struct IORING_CREATE_FLAGS {
+  IORING_CREATE_REQUIRED_FLAGS Required;
+  IORING_CREATE_ADVISORY_FLAGS Advisory;
+} IORING_CREATE_FLAGS;
+// Specifies flags for creating an I/O ring with a call to CreateIoRing.
+
+// IORING_INFO structure
+typedef struct IORING_INFO {
+  IORING_VERSION IoRingVersion;
+  IORING_CREATE_FLAGS Flags;
+  UINT32 SubmissionQueueSize;
+  UINT32 CompletionQueueSize;
+} IORING_INFO;
+// Represents the shape and version information for the specified I/O ring
+
+// IORING_OP_CODE for IsIoRingOpSupported
+typedef enum IORING_OP_CODE {
+  IORING_OP_NOP = 0,
+  IORING_OP_READ = 1,
+  IORING_OP_WRITE = 2,
+  IORING_OP_FLUSH = 3,
+  IORING_OP_REGISTER_BUFFERS = 4,
+  IORING_OP_REGISTER_FILES = 5,
+  IORING_OP_CANCEL = 6,
+} IORING_OP_CODE;
+
+/* --------------------- Dynamic loader --------------------- */
+// Function pointer types
+typedef BOOL(WINAPI *IsIoRingOpSupported_t)(HIORING, IORING_OP_CODE);
+typedef HRESULT(WINAPI *QueryIoRingCapabilities_t)(IORING_CAPABILITIES *);
+typedef HRESULT(WINAPI *GetIoRingInfo_t)(HIORING, IORING_INFO *);
+typedef HRESULT(WINAPI *CreateIoRing_t)(IORING_VERSION, IORING_CREATE_FLAGS,
+                                        UINT32, UINT32, HIORING *);
+typedef HRESULT(WINAPI *CloseIoRing_t)(HIORING);
+typedef HRESULT(WINAPI *SubmitIoRing_t)(HIORING, UINT32, UINT32, UINT32 *);
+typedef HRESULT(WINAPI *PopIoRingCompletion_t)(HIORING, IORING_CQE *);
+typedef HRESULT(WINAPI *SetIoRingCompletionEvent_t)(HIORING, HANDLE);
+typedef HRESULT(WINAPI *BuildIoRingCancelRequest_t)(HIORING, IORING_HANDLE_REF,
+                                                    UINT_PTR, UINT_PTR);
+typedef HRESULT(WINAPI *BuildIoRingReadFile_t)(HIORING, IORING_HANDLE_REF,
+                                               IORING_BUFFER_REF, UINT32,
+                                               UINT64, UINT_PTR,
+                                               IORING_SQE_FLAGS);
+typedef HRESULT(WINAPI *BuildIoRingRegisterBuffers_t)(
+    HIORING, UINT32, IORING_BUFFER_INFO const[], UINT_PTR);
+
+typedef HRESULT(WINAPI *BuildIoRingRegisterFileHandles_t)(HIORING, UINT32,
+                                                          HANDLE const[],
+                                                          UINT_PTR);
+
+// Core:
+// Queries the support of the specified operation for the specified I/O ring
+static IsIoRingOpSupported_t IsIoRingOpSupported = NULL;
+
+// Queries the OS for the supported capabilities for IORINGs
+static QueryIoRingCapabilities_t QueryIoRingCapabilities = NULL;
+
+// Gets information about the API version and queue sizes of an I/O ring
+static GetIoRingInfo_t GetIoRingInfo = NULL;
+
+// Creates a new instance of an I/O ring submission/completion queue pair and
+// returns a handle for referencing the I/O ring
+static CreateIoRing_t CreateIoRing = NULL;
+
+// Closes an HIORING handle that was previously opened with a call to
+// CreateIoRing
+static CloseIoRing_t CloseIoRing = NULL;
+
+// Submission / completion:
+// Submits all constructed but not yet submitted entries to the kernel’s queue
+// and optionally waits for a set of operations to complete
+static SubmitIoRing_t SubmitIoRing = NULL;
+
+// Pops a single entry from the completion queue, if one is available
+static PopIoRingCompletion_t PopIoRingCompletion = NULL;
+
+// Registers a completion queue event with an IORING
+static SetIoRingCompletionEvent_t SetIoRingCompletionEvent = NULL;
+
+// Operations:
+// Performs an asynchronous read from a file using an I/O ring
+static BuildIoRingReadFile_t BuildIoRingReadFile = NULL;
+
+// Attempts to cancel a previously submitted I/O ring operation
+static BuildIoRingCancelRequest_t BuildIoRingCancelRequest = NULL;
+
+// Registers an array of buffers with the system for future I/O ring operations
+static BuildIoRingRegisterBuffers_t BuildIoRingRegisterBuffers = NULL;
+
+// Registers an array of file handles with the system for future I/O ring
+// operations
+static BuildIoRingRegisterFileHandles_t BuildIoRingRegisterFileHandles = NULL;
+
+static int io_ring_loaded = 0;
+
+static int io_ring_load_functions(void) {
+  if (io_ring_loaded)
+    return 1;
+
+  HMODULE hKernel = GetModuleHandleW(L"kernel32.dll");
+  if (!hKernel)
+    return 0;
+
+  IsIoRingOpSupported =
+      (IsIoRingOpSupported_t)GetProcAddress(hKernel, "IsIoRingOpSupported");
+  QueryIoRingCapabilities = (QueryIoRingCapabilities_t)GetProcAddress(
+      hKernel, "QueryIoRingCapabilities");
+  GetIoRingInfo = (GetIoRingInfo_t)GetProcAddress(hKernel, "GetIoRingInfo");
+  CreateIoRing = (CreateIoRing_t)GetProcAddress(hKernel, "CreateIoRing");
+  CloseIoRing = (CloseIoRing_t)GetProcAddress(hKernel, "CloseIoRing");
+  SubmitIoRing = (SubmitIoRing_t)GetProcAddress(hKernel, "SubmitIoRing");
+  PopIoRingCompletion =
+      (PopIoRingCompletion_t)GetProcAddress(hKernel, "PopIoRingCompletion");
+  SetIoRingCompletionEvent = (SetIoRingCompletionEvent_t)GetProcAddress(
+      hKernel, "SetIoRingCompletionEvent");
+  BuildIoRingReadFile =
+      (BuildIoRingReadFile_t)GetProcAddress(hKernel, "BuildIoRingReadFile");
+  BuildIoRingCancelRequest = (BuildIoRingCancelRequest_t)GetProcAddress(
+      hKernel, "BuildIoRingCancelRequest");
+  BuildIoRingRegisterBuffers = (BuildIoRingRegisterBuffers_t)GetProcAddress(
+      hKernel, "BuildIoRingRegisterBuffers");
+  BuildIoRingRegisterFileHandles =
+      (BuildIoRingRegisterFileHandles_t)GetProcAddress(
+          hKernel, "BuildIoRingRegisterFileHandles");
+
+  io_ring_loaded =
+      (IsIoRingOpSupported && QueryIoRingCapabilities && CreateIoRing &&
+       CloseIoRing && SubmitIoRing && PopIoRingCompletion &&
+       SetIoRingCompletionEvent && BuildIoRingReadFile &&
+       BuildIoRingCancelRequest && BuildIoRingRegisterBuffers &&
+       BuildIoRingRegisterFileHandles);
+
+  if (io_ring_loaded)
+    printf("[I/O Ring] Functions loaded\n");
+  else
+    printf("[I/O Ring] Some functions not available\n");
+
+  return io_ring_loaded;
+}
+
+/* ------------- Standard helper functions definition ------------- */
+// Creates an instance of the IORING_BUFFER_REF structure with the provided
+// buffer index and offset
+static inline IORING_BUFFER_REF
+IoRingBufferRefFromIndexAndOffset(UINT32 index, UINT32 offset) {
+  IORING_BUFFER_REF ref;
+  ref.Kind = IORING_REF_REGISTERED;
+  ref.BufferUnion.IndexAndOffset.Index = index;
+  ref.BufferUnion.IndexAndOffset.Offset = offset;
+  return ref;
+}
+
+// Creates an instance of the IORING_BUFFER_REF structure from the provided
+// pointer
+static IORING_BUFFER_REF IoRingBufferRefFromPointer(void *addr) {
+  IORING_BUFFER_REF ref;
+  ref.Kind = IORING_REF_RAW;
+  ref.BufferUnion.Address = addr;
+  return ref;
+}
+
+// Creates an instance of the IORING_HANDLE_REF structure from the provided file
+// handle
+static IORING_HANDLE_REF IoRingHandleRefFromHandle(HANDLE h) {
+  IORING_HANDLE_REF ref;
+  ref.Kind = IORING_REF_RAW;
+  ref.HandleUnion.Handle = h;
+  return ref;
+}
+
+// Creates an instance of the IORING_HANDLE_REF structure from the provided
+// index
+static inline IORING_HANDLE_REF IoRingHandleRefFromIndex(UINT32 index) {
+  IORING_HANDLE_REF ref;
+  ref.Kind = IORING_REF_REGISTERED; // MUST be registered
+  ref.HandleUnion.Index = index;
+  return ref;
+}
+
+// NOTE: If you are using index-based buffers or handles, make sure you have
+// successfully called BuildIoRingRegisterBuffers or
+// BuildIoRingRegisterFileHandles first so the kernel has a valid table to look
+// into, otherwise the kernel will treat the index as an invalid memory
+// address/handle.

file_hasher.c

@@ -32,7 +32,11 @@ int main(int argc, char **argv) {
     buf[strcspn(buf, "\r\n")] = 0;
 
     if (buf[0] == 0) {
-      strcpy(folders[0], ".");
+      if (!platform_get_current_directory(folders[0], sizeof(folders[0]))) {
+        fprintf(stderr, "Failed to get current directory\n");
+        return 1;
+      }
+      normalize_path(folders[0]);
       folder_count = 1;
     } else {
       folder_count = parse_paths(buf, folders, 64);
@@ -71,24 +75,60 @@ int main(int argc, char **argv) {
       .max_nbre_blocks = 1,
   };
 
-  mem_arena *gp_arena = arena_create(&params);
+  arena_params params_caligned = {
+      .reserve_size = GiB(1),
+      .commit_size = MiB(16),
+      .align = ARENA_CACHE_ALIGN,
+      .push_size = 0,
+      .allow_free_list = true,
+      .allow_swapback = false,
+      .growth_policy = ARENA_GROWTH_NORMAL,
+      .commit_policy = ARENA_COMMIT_LAZY,
+      .max_nbre_blocks = 1,
+  };
+  mem_arena *gp_arena = arena_create(&params_caligned);
 
   // -------------------------------
-  // Detect hardware threads
+  // Detect hardware
   // -------------------------------
   // --- Windows: detect PHYSICAL cores (not logical threads) ---
-  size_t hw_threads = platform_physical_cores();
+  uint8_t cpu_cores = platform_physical_cores();
 
   // Logical threads = CPU cores * 2
-  size_t num_threads = hw_threads * 2;
+  uint8_t cpu_threads = cpu_cores * 2;
 
-  printf("Starting thread pool: %zu threads (CPU cores: %zu)\n", num_threads,
-         hw_threads);
-  printf("  Selected instruction set: %s\n", get_xxhash_instruction_set());
+#if MULTI_THREADING
+  uint8_t num_scan_threads = cpu_threads;
+  uint8_t num_hash_threads = cpu_threads;
+
+  printf("%d cores %d threads CPU detected with %s instruction set\n"
+         "Starting thread pool: %d scanning and %d hashing threads\n",
+         cpu_cores, cpu_threads, get_xxhash_instruction_set(), num_scan_threads,
+         num_hash_threads);
+#else
+  uint32_t num_scan_threads = 1;
+  uint32_t num_hash_threads = 1;
+
+  printf(
+      "%d cores %d threads CPU detected with %s instruction set\n"
+      "Starting thread pool: %d scanning and %d hashing threads(Debug mode)\n",
+      cpu_cores, cpu_threads, get_xxhash_instruction_set(), num_scan_threads,
+      num_hash_threads);
+
+#endif
+
+  // Align IO Ring block size to the system page size
+  g_pagesize = plat_get_pagesize();
+  g_read_block = ALIGN_UP_POW2(READ_BLOCK, g_pagesize);
+#if USE_IORING
+  g_ioring_buffer_size = ALIGN_UP_POW2(IORING_BUFFER_SIZE, g_pagesize);
+#endif
 
   // -------------------------------
   // Scanning and hashing
   // -------------------------------
+
+  // test_io_ring();
   MPMCQueue dir_queue;
   mpmc_init(&dir_queue, MiB(1));
 
@@ -96,19 +136,23 @@ int main(int argc, char **argv) {
   mpmc_init(&file_queue, MiB(1));
 
   // Starting hash threads
-  size_t num_hash_threads = num_threads;
-
-  WorkerContext workers[num_hash_threads];
+  HasherContext workers[num_hash_threads];
   Thread *hash_threads =
       arena_push(&gp_arena, sizeof(Thread) * num_hash_threads, true);
 
-  for (size_t i = 0; i < num_hash_threads; ++i) {
+  for (uint8_t i = 0; i < num_hash_threads; ++i) {
     workers[i].arena = arena_create(&params);
     workers[i].file_queue = &file_queue;
 
+#if USE_IORING
+    if (thread_create(&hash_threads[i], (ThreadFunc)hash_worker_ioring,
+                      &workers[i]) != 0)
+#else
     if (thread_create(&hash_threads[i], (ThreadFunc)hash_worker, &workers[i]) !=
-        0) {
-      fprintf(stderr, "Failed to create hash thread %zu\n", i);
+        0)
+#endif
+    {
+      fprintf(stderr, "Failed to create hash thread %d\n", i);
       exit(1);
     }
   }
@@ -122,22 +166,19 @@ int main(int argc, char **argv) {
   }
 
   // Starting scan threads
-  size_t num_scan_threads = num_threads;
-
   ScannerContext scanners[num_scan_threads];
   Thread *scan_threads =
       arena_push(&gp_arena, sizeof(Thread) * num_scan_threads, true);
 
-  for (size_t i = 0; i < num_scan_threads; i++) {
+  for (uint8_t i = 0; i < num_scan_threads; i++) {
     scanners[i].num_threads = num_scan_threads;
-    scanners[i].path_arena = arena_create(&params);
-    scanners[i].meta_arena = arena_create(&params);
+    scanners[i].meta_arena = arena_create(&params_caligned);
     scanners[i].dir_queue = &dir_queue;
     scanners[i].file_queue = &file_queue;
 
     if (thread_create(&scan_threads[i], (ThreadFunc)scan_worker,
                       &scanners[i]) != 0) {
-      fprintf(stderr, "Failed to create scan thread %zu\n", i);
+      fprintf(stderr, "Failed to create scan thread %d\n", i);
       exit(1);
     }
   }
@@ -145,7 +186,7 @@ int main(int argc, char **argv) {
   // Initial folder push
   for (int i = 0; i < folder_count; i++) {
     size_t len = strlen(folders[i]) + 1;
-    char *path = arena_push(&scanners[0].path_arena, len, false);
+    char *path = arena_push(&scanners[0].meta_arena, len, false);
     memcpy(path, folders[i], len);
     mpmc_push_work(&dir_queue, path);
   }
@@ -153,7 +194,7 @@ int main(int argc, char **argv) {
   // Stop scan threads
   thread_wait_multiple(scan_threads, num_scan_threads);
 
-  for (size_t i = 0; i < num_scan_threads; ++i) {
+  for (uint8_t i = 0; i < num_scan_threads; ++i) {
     thread_close(&scan_threads[i]);
   }
 
@@ -180,7 +221,7 @@ int main(int argc, char **argv) {
   // Stop hashing threads
   thread_wait_multiple(hash_threads, num_hash_threads);
 
-  for (size_t i = 0; i < num_hash_threads; ++i) {
+  for (uint8_t i = 0; i < num_hash_threads; ++i) {
     thread_close(&hash_threads[i]);
   }
 
@@ -197,7 +238,7 @@ int main(int argc, char **argv) {
 
   FILE *f = fopen(FILE_HASHES_TXT, "wb");
 
-  for (int i = 0; i < num_threads; i++) {
+  for (uint8_t i = 0; i < num_hash_threads; i++) {
     mem_arena *arena = workers[i].arena;
     u8 *arena_base =
         (u8 *)arena + ALIGN_UP_POW2(sizeof(mem_arena), arena->align);
@@ -209,6 +250,15 @@ int main(int argc, char **argv) {
   // -------------------------------
   // Print summary
   // -------------------------------
+#if USE_IORING
+  uint64_t incomplete = atomic_load(&g_io_ring_fallbacks);
+  if (incomplete > 0) {
+    printf("\nWARNING: I/O Ring incomplete files: %llu (fallback to buffered "
+           "I/O used)\n",
+           (unsigned long long)incomplete);
+  }
+#endif
+
   double total_seconds = timer_elapsed(&total_timer);
 
   printf("Completed hashing %zu files\n", total_found);

lf_mpmc.h

@@ -23,9 +23,6 @@ static void cpu_pause(void) {
   _mm_pause();
 #endif
 }
-
-typedef struct plat_sem plat_sem;
-
 typedef struct CACHE_ALIGN {
   atomic_size_t seq;
   void *data;
@@ -45,8 +42,6 @@ typedef struct {
   size_t commit_step;
   atomic_flag commit_lock;
 
-  plat_sem items_sem;
-
   MPMCSlot *slots;
 } MPMCQueue;
 
@@ -94,8 +89,6 @@ static void mpmc_init(MPMCQueue *q, size_t max_capacity) {
   atomic_init(&q->head, 0);
   atomic_init(&q->tail, 0);
   atomic_init(&q->work_count, 0);
-
-  plat_sem_init(&q->items_sem, 0);
 }
 
 /* ----------------------------------------------------------- */
@@ -141,7 +134,6 @@ static void mpmc_commit_more(MPMCQueue *q) {
 /* ----------------------------------------------------------- */
 /* PUSH */
 /* ----------------------------------------------------------- */
-// Does not increment work
 static void mpmc_push(MPMCQueue *q, void *item) {
   MPMCSlot *slot;
   size_t pos;
@@ -184,8 +176,6 @@ static void mpmc_push(MPMCQueue *q, void *item) {
   slot->data = item;
 
   atomic_store_explicit(&slot->seq, pos + 1, memory_order_release);
-
-  plat_sem_post(&q->items_sem, 1);
 }
 
 // Increment work
@@ -233,15 +223,11 @@ static void mpmc_push_work(MPMCQueue *q, void *item) {
   atomic_store_explicit(&slot->seq, pos + 1, memory_order_release);
 
   atomic_fetch_add(&q->work_count, 1);
-  plat_sem_post(&q->items_sem, 1);
 }
 /* ----------------------------------------------------------- */
 /* POP */
 /* ----------------------------------------------------------- */
 static void *mpmc_pop(MPMCQueue *q) {
-
-  plat_sem_wait(&q->items_sem);
-
   MPMCSlot *slot;
   size_t pos;
 
@@ -262,9 +248,14 @@ static void *mpmc_pop(MPMCQueue *q) {
                                                 memory_order_relaxed))
         break;
 
+    } else if (diff < 0) { // queue is empty
+
+      Sleep(500);
+
     } else { // slot is still transitioning (written by another thread)
+
       if (++spins > 10) {
-        sleep_ms(0); // yield CPU
+        SwitchToThread(); // yield CPU
         spins = 0;
       } else {
         cpu_pause();
@@ -306,21 +297,19 @@ static void mpmc_task_done(MPMCQueue *q, u8 consumer_count) {
 /* ----------------------------------------------------------- */
 /* MPMC Cleanup */
 /* ----------------------------------------------------------- */
-static void mpmc_finish(MPMCQueue *q) {
-  if (!q)
-    return;
-
-  if (q->slots) {
-    plat_mem_release(q->slots, 0);
-    q->slots = NULL;
-  }
-
-  plat_sem_destroy(&q->items_sem);
-
-  q->capacity = 0;
-  q->mask = 0;
-
-  atomic_store_explicit(&q->head, 0, memory_order_relaxed);
-  atomic_store_explicit(&q->tail, 0, memory_order_relaxed);
-  atomic_store_explicit(&q->committed, 0, memory_order_relaxed);
-}
+// static void mpmc_finish(MPMCQueue *q) { // Comment to prevent warning: unused function
+//   if (!q)
+//     return;
+//
+//   if (q->slots) {
+//     plat_mem_release(q->slots, 0);
+//     q->slots = NULL;
+//   }
+//
+//   q->capacity = 0;
+//   q->mask = 0;
+//
+//   atomic_store_explicit(&q->head, 0, memory_order_relaxed);
+//   atomic_store_explicit(&q->tail, 0, memory_order_relaxed);
+//   atomic_store_explicit(&q->committed, 0, memory_order_relaxed);
+// }

mt_mpmc.h

@@ -0,0 +1,254 @@
+#pragma once
+
+#include "base.h"
+
+// Cache align abstraction
+#define CACHELINE 64
+
+#if defined(_MSC_VER)
+#define CACHE_ALIGN __declspec(align(CACHELINE))
+#else
+#define CACHE_ALIGN __attribute__((aligned(CACHELINE)))
+#endif
+
+// Mutex/Critical section abstraction
+#if defined(_WIN32)
+#include <windows.h>
+typedef CRITICAL_SECTION mtx_t;
+typedef CONDITION_VARIABLE cond_t;
+
+#define mtx_init(m) InitializeCriticalSection(m)
+#define mtx_lock(m) EnterCriticalSection(m)
+#define mtx_unlock(m) LeaveCriticalSection(m)
+#define mtx_destroy(m) DeleteCriticalSection(m)
+
+#define cond_init(c) InitializeConditionVariable(c)
+#define cond_wait(c, m) SleepConditionVariableCS(c, m, INFINITE)
+#define cond_signal(c) WakeConditionVariable(c)
+#define cond_broadcast(c) WakeAllConditionVariable(c)
+
+#else
+#include <pthread.h>
+typedef pthread_mutex_t mtx_t;
+typedef pthread_cond_t cond_t;
+
+#define mtx_init(m) pthread_mutex_init(m, NULL)
+#define mtx_lock(m) pthread_mutex_lock(m)
+#define mtx_unlock(m) pthread_mutex_unlock(m)
+#define mtx_destroy(m) pthread_mutex_destroy(m)
+
+#define cond_init(c) pthread_cond_init(c, NULL)
+#define cond_wait(c, m) pthread_cond_wait(c, m)
+#define cond_signal(c) pthread_cond_signal(c)
+#define cond_broadcast(c) pthread_cond_broadcast(c)
+
+#endif
+
+typedef struct CACHE_ALIGN {
+  void *data;
+  char pad[64 - sizeof(void *)];
+} MPMCSlot;
+
+typedef struct {
+  CACHE_ALIGN size_t head;
+  CACHE_ALIGN size_t tail;
+
+  CACHE_ALIGN size_t count;
+  CACHE_ALIGN size_t work_count;
+
+  size_t capacity;
+  size_t mask;
+
+  size_t committed;
+  size_t commit_step;
+
+  mtx_t lock;
+  cond_t not_empty;
+  cond_t not_full;
+
+  MPMCSlot *slots;
+} MPMCQueue;
+
+/* ----------------------------------------------------------- */
+/* INIT */
+/* ----------------------------------------------------------- */
+static void mpmc_init(MPMCQueue *q, size_t max_capacity) {
+  q->capacity = max_capacity;
+  q->mask = max_capacity - 1;
+
+  size_t pagesize = plat_get_pagesize();
+  size_t bytes = ALIGN_UP_POW2(sizeof(MPMCSlot) * max_capacity, pagesize);
+
+  q->slots = (MPMCSlot *)plat_mem_reserve(bytes);
+  if (!q->slots) {
+    fprintf(stderr, "plat_mem_reserve failed\n");
+    exit(1);
+  }
+
+  size_t commit_bytes = ALIGN_UP_POW2(pagesize, pagesize);
+  q->commit_step = commit_bytes / sizeof(MPMCSlot);
+
+  q->committed = q->commit_step;
+  plat_mem_commit(q->slots, commit_bytes);
+
+  for (size_t i = 0; i < q->committed; i++) {
+    q->slots[i].data = NULL;
+  }
+
+  q->head = 0;
+  q->tail = 0;
+  q->count = 0;
+  q->work_count = 0;
+
+  mtx_init(&q->lock);
+  cond_init(&q->not_empty);
+  cond_init(&q->not_full);
+}
+
+/* ----------------------------------------------------------- */
+/* COMMIT MORE MEMORY */
+/* ----------------------------------------------------------- */
+static void mpmc_commit_more(MPMCQueue *q) {
+  size_t start = q->committed;
+
+  if (start >= q->capacity)
+    return;
+
+  size_t new_commit = start + q->commit_step;
+  if (new_commit > q->capacity)
+    new_commit = q->capacity;
+
+  size_t count = new_commit - start;
+
+  plat_mem_commit(&q->slots[start], count * sizeof(MPMCSlot));
+
+  for (size_t i = start; i < new_commit; i++) {
+    q->slots[i].data = NULL;
+  }
+
+  q->committed = new_commit;
+}
+
+/* ----------------------------------------------------------- */
+/* PUSH */
+/* ----------------------------------------------------------- */
+// Does not increment work
+static void mpmc_push(MPMCQueue *q, void *item) {
+  mtx_lock(&q->lock);
+
+  while (q->count == q->capacity) {
+    cond_wait(&q->not_full, &q->lock);
+  }
+
+  // Ensure committed
+  if (q->tail >= q->committed) {
+    mpmc_commit_more(q);
+  }
+
+  size_t pos = q->tail & q->mask;
+
+  q->slots[pos].data = item;
+  q->tail++;
+  q->count++;
+
+  cond_signal(&q->not_empty);
+  mtx_unlock(&q->lock);
+}
+
+// Increment work
+static void mpmc_push_work(MPMCQueue *q, void *item) {
+  mtx_lock(&q->lock);
+
+  while (q->count == q->capacity) {
+    cond_wait(&q->not_full, &q->lock);
+  }
+
+  if (q->tail >= q->committed) {
+    mpmc_commit_more(q);
+  }
+
+  size_t pos = q->tail & q->mask;
+
+  q->slots[pos].data = item;
+  q->tail++;
+  q->count++;
+  q->work_count++;
+
+  cond_signal(&q->not_empty);
+  mtx_unlock(&q->lock);
+}
+
+/* ----------------------------------------------------------- */
+/* POP */
+/* ----------------------------------------------------------- */
+static void *mpmc_pop(MPMCQueue *q) {
+  mtx_lock(&q->lock);
+
+  while (q->count == 0) {
+    cond_wait(&q->not_empty, &q->lock);
+  }
+
+  size_t pos = q->head & q->mask;
+
+  void *data = q->slots[pos].data;
+
+  q->head++;
+  q->count--;
+
+  cond_signal(&q->not_full);
+  mtx_unlock(&q->lock);
+
+  return data;
+}
+
+/* ----------------------------------------------------------- */
+/* PUSH POISON */
+/* ----------------------------------------------------------- */
+static void mpmc_producers_finished(MPMCQueue *q, u8 consumer_count) {
+  for (u8 i = 0; i < consumer_count; i++) {
+    mpmc_push(q, NULL);
+  }
+}
+
+/* ----------------------------------------------------------- */
+/* Done */
+/* ----------------------------------------------------------- */
+static void mpmc_task_done(MPMCQueue *q, u8 consumer_count) {
+
+  bool finished = false;
+
+  mtx_lock(&q->lock);
+
+  if (--q->work_count == 0) {
+    finished = true;
+  }
+
+  mtx_unlock(&q->lock);
+
+  if (finished) {
+    mpmc_producers_finished(q, consumer_count);
+  }
+}
+
+/* ----------------------------------------------------------- */
+/* MPMC Cleanup */
+/* ----------------------------------------------------------- */
+// static void mpmc_finish(MPMCQueue *q) { // Comment to prevent warning: unused
+// function
+//   if (!q) return;
+//
+//   if (q->slots) {
+//     plat_mem_release(q->slots, 0);
+//     q->slots = NULL;
+//   }
+//
+//   mtx_destroy(&q->lock);
+//
+// #if !defined(_WIN32)
+//   pthread_cond_destroy(&q->not_empty);
+//   pthread_cond_destroy(&q->not_full);
+// #endif
+//
+//   q->capacity = 0;
+//   q->mask = 0;
+// }

sm_mpmc.h

@@ -0,0 +1,388 @@
+#pragma once
+
+#include "base.h"
+
+// Cache align abstraction
+#define CACHELINE 64
+
+#if defined(_MSC_VER)
+#define CACHE_ALIGN __declspec(align(CACHELINE))
+#else
+#define CACHE_ALIGN __attribute__((aligned(CACHELINE)))
+#endif
+
+// Compiler hints
+#if defined(__GNUC__) || defined(__clang__)
+#define likely(x) __builtin_expect((x), 1)
+#define unlikely(x) __builtin_expect((x), 0)
+#else
+#define likely(x) (x)
+#define unlikely(x) (x)
+#endif
+
+static void cpu_pause(void) {
+#if defined(_MSC_VER) || defined(__x86_64__) || defined(__i386__)
+  _mm_pause();
+#endif
+}
+
+// Semaphores
+#if defined(_WIN32) || defined(_WIN64)
+typedef struct plat_sem {
+  HANDLE handle;
+} plat_sem;
+
+static b32 plat_sem_init(plat_sem *s, u32 initial) {
+  s->handle = CreateSemaphore(NULL, initial, LONG_MAX, NULL);
+  return s->handle != NULL;
+}
+
+static void plat_sem_wait(plat_sem *s) {
+  WaitForSingleObject(s->handle, INFINITE);
+}
+
+// static b32 plat_sem_trywait(HANDLE sem) { // Comment to prevent warning: unused function
+//   DWORD r = WaitForSingleObject(sem, 0);
+//   return r == WAIT_OBJECT_0;
+// }
+
+static void plat_sem_post(plat_sem *s, u32 count) {
+  ReleaseSemaphore(s->handle, count, NULL);
+}
+
+// static void plat_sem_destroy(plat_sem *s) { // Comment to prevent warning: unused function
+//   if (s->handle) {
+//     CloseHandle(s->handle);
+//     s->handle = NULL;
+//   }
+// }
+
+#elif defined(__linux__)
+#include <semaphore.h>
+
+typedef struct plat_sem {
+  sem_t sem;
+} plat_sem;
+
+static b32 plat_sem_init(plat_sem *s, u32 initial) {
+  return sem_init(&s->sem, 0, initial) == 0;
+}
+
+static void plat_sem_wait(plat_sem *s) {
+  while (sem_wait(&s->sem) == -1 && errno == EINTR) {
+  }
+}
+
+// static b32 plat_sem_trywait(sem_t *sem) { return sem_trywait(sem) == 0; } // Comment to prevent warning: unused function
+
+static void plat_sem_post(plat_sem *s, u32 count) {
+  for (u32 i = 0; i < count; i++) {
+    sem_post(&s->sem);
+  }
+}
+
+// static void plat_sem_destroy(plat_sem *s) { sem_destroy(&s->sem); } // Comment to prevent warning: unused function
+
+#endif
+
+typedef struct plat_sem plat_sem;
+
+typedef struct CACHE_ALIGN {
+  atomic_size_t seq;
+  void *data;
+  char pad[64 - sizeof(atomic_size_t) - sizeof(void *)];
+} MPMCSlot;
+
+typedef struct {
+  CACHE_ALIGN atomic_size_t head;
+  CACHE_ALIGN atomic_size_t tail;
+
+  CACHE_ALIGN atomic_size_t work_count;
+
+  size_t capacity;
+  size_t mask;
+
+  atomic_size_t committed;
+  size_t commit_step;
+  atomic_flag commit_lock;
+
+  plat_sem items_sem;
+
+  MPMCSlot *slots;
+} MPMCQueue;
+
+// --------------- functions ----------------
+// static: each translation unit gets its own private copy this will solve the
+// error:  Function defined in a header file; function definitions in header
+// files can lead to ODR violations (multiple definition errors if included in
+// more than one file)
+
+/* ----------------------------------------------------------- */
+/* INIT */
+/* ----------------------------------------------------------- */
+static void mpmc_init(MPMCQueue *q, size_t max_capacity) {
+
+  q->capacity = max_capacity;
+  q->mask = max_capacity - 1;
+
+  u32 pagesize = plat_get_pagesize();
+
+  size_t bytes = ALIGN_UP_POW2(sizeof(MPMCSlot) * max_capacity, pagesize);
+
+  q->slots = (MPMCSlot *)plat_mem_reserve(bytes);
+
+  if (!q->slots) {
+    fprintf(stderr, "VirtualAlloc reserve failed\n");
+    exit(1);
+  }
+
+  u64 commit_bytes = pagesize;
+  commit_bytes = ALIGN_UP_POW2(commit_bytes, pagesize);
+
+  q->commit_step = commit_bytes / sizeof(MPMCSlot);
+
+  atomic_flag_clear(&q->commit_lock);
+
+  q->committed = q->commit_step;
+
+  plat_mem_commit(q->slots, commit_bytes);
+
+  for (size_t i = 0; i < q->committed; i++) {
+    atomic_init(&q->slots[i].seq, i);
+    q->slots[i].data = NULL;
+  }
+
+  atomic_init(&q->head, 0);
+  atomic_init(&q->tail, 0);
+  atomic_init(&q->work_count, 0);
+
+  plat_sem_init(&q->items_sem, 0);
+}
+
+/* ----------------------------------------------------------- */
+/* COMMIT MORE MEMORY */
+/* ----------------------------------------------------------- */
+static void mpmc_commit_more(MPMCQueue *q) {
+
+  if (atomic_flag_test_and_set(&q->commit_lock))
+    return;
+
+  size_t start = atomic_load_explicit(&q->committed, memory_order_acquire);
+  size_t tail = atomic_load_explicit(&q->tail, memory_order_relaxed);
+
+  // another thread already committed enough
+  if (tail < start) {
+    atomic_flag_clear(&q->commit_lock);
+    return;
+  }
+
+  if (start >= q->capacity) {
+    atomic_flag_clear(&q->commit_lock);
+    return;
+  }
+
+  size_t new_commit = start + q->commit_step;
+  if (new_commit > q->capacity)
+    new_commit = q->capacity;
+
+  size_t count = new_commit - start;
+
+  plat_mem_commit(&q->slots[start], count * sizeof(MPMCSlot));
+
+  for (size_t i = start; i < new_commit; i++) {
+    atomic_init(&q->slots[i].seq, i);
+    q->slots[i].data = NULL;
+  }
+
+  atomic_store_explicit(&q->committed, new_commit, memory_order_release);
+
+  atomic_flag_clear(&q->commit_lock);
+}
+
+/* ----------------------------------------------------------- */
+/* PUSH */
+/* ----------------------------------------------------------- */
+// Does not increment work
+static void mpmc_push(MPMCQueue *q, void *item) {
+  MPMCSlot *slot;
+  size_t pos;
+
+  for (;;) {
+
+    pos = atomic_load_explicit(&q->tail, memory_order_relaxed);
+
+    // ensure the slot is committed BEFORE accessing it
+    size_t committed =
+        atomic_load_explicit(&q->committed, memory_order_relaxed);
+
+    if (unlikely(pos >= committed)) {
+      mpmc_commit_more(q);
+      continue;
+    }
+
+    slot = &q->slots[pos & q->mask];
+
+    size_t seq = atomic_load_explicit(&slot->seq, memory_order_acquire);
+    intptr_t diff = (intptr_t)seq - (intptr_t)pos;
+
+    if (likely(diff == 0)) {
+
+      if (atomic_compare_exchange_weak_explicit(&q->tail, &pos, pos + 1,
+                                                memory_order_relaxed,
+                                                memory_order_relaxed))
+        break;
+
+    } else if (diff < 0) { // queue actually full
+
+      sleep_ms(1000);
+
+    } else { // waiting to grow
+
+      sleep_ms(0);
+    }
+  }
+
+  slot->data = item;
+
+  atomic_store_explicit(&slot->seq, pos + 1, memory_order_release);
+
+  plat_sem_post(&q->items_sem, 1);
+}
+
+// Increment work
+static void mpmc_push_work(MPMCQueue *q, void *item) {
+  MPMCSlot *slot;
+  size_t pos;
+
+  for (;;) {
+
+    pos = atomic_load_explicit(&q->tail, memory_order_relaxed);
+
+    // ensure the slot is committed BEFORE accessing it
+    size_t committed =
+        atomic_load_explicit(&q->committed, memory_order_relaxed);
+
+    if (unlikely(pos >= committed)) {
+      mpmc_commit_more(q);
+      continue;
+    }
+
+    slot = &q->slots[pos & q->mask];
+
+    size_t seq = atomic_load_explicit(&slot->seq, memory_order_acquire);
+    intptr_t diff = (intptr_t)seq - (intptr_t)pos;
+
+    if (likely(diff == 0)) {
+
+      if (atomic_compare_exchange_weak_explicit(&q->tail, &pos, pos + 1,
+                                                memory_order_relaxed,
+                                                memory_order_relaxed))
+        break;
+
+    } else if (diff < 0) { // queue actually full
+
+      sleep_ms(1000);
+
+    } else { // waiting to grow
+
+      sleep_ms(0);
+    }
+  }
+
+  slot->data = item;
+
+  atomic_store_explicit(&slot->seq, pos + 1, memory_order_release);
+
+  atomic_fetch_add(&q->work_count, 1);
+  plat_sem_post(&q->items_sem, 1);
+}
+
+/* ----------------------------------------------------------- */
+/* POP */
+/* ----------------------------------------------------------- */
+static void *mpmc_pop(MPMCQueue *q) {
+
+  plat_sem_wait(&q->items_sem);
+
+  MPMCSlot *slot;
+  size_t pos;
+
+  int spins = 0;
+
+  for (;;) {
+
+    pos = atomic_load_explicit(&q->head, memory_order_relaxed);
+    slot = &q->slots[pos & q->mask];
+
+    size_t seq = atomic_load_explicit(&slot->seq, memory_order_acquire);
+    intptr_t diff = (intptr_t)seq - (intptr_t)(pos + 1);
+
+    if (likely(diff == 0)) {
+
+      if (atomic_compare_exchange_weak_explicit(&q->head, &pos, pos + 1,
+                                                memory_order_relaxed,
+                                                memory_order_relaxed))
+        break;
+
+    } else { // slot is still transitioning (written by another thread)
+      if (++spins > 10) {
+        sleep_ms(0); // yield CPU
+        spins = 0;
+      } else {
+        cpu_pause();
+      }
+    }
+  }
+
+  void *data = slot->data;
+
+  atomic_store_explicit(&slot->seq, pos + q->capacity, memory_order_release);
+
+  return data;
+}
+
+/* ----------------------------------------------------------- */
+/* PUSH POISON */
+/* ----------------------------------------------------------- */
+/*note:
+ After producers finishes, push N poison pills where N = number of consumer
+threads, this is necessary to stop the consumers.
+*/
+
+static void mpmc_producers_finished(MPMCQueue *q, u8 consumer_count) {
+  for (u8 i = 0; i < consumer_count; i++) {
+    mpmc_push(q, NULL);
+  }
+}
+
+/* ----------------------------------------------------------- */
+/* Done */
+/* ----------------------------------------------------------- */
+static void mpmc_task_done(MPMCQueue *q, u8 consumer_count) {
+  size_t prev = atomic_fetch_sub(&q->work_count, 1);
+  if (prev == 1) {
+    mpmc_producers_finished(q, consumer_count);
+  }
+}
+
+/* ----------------------------------------------------------- */
+/* MPMC Cleanup */
+/* ----------------------------------------------------------- */
+// static void mpmc_finish(MPMCQueue *q) { // Comment to prevent warning: unused function
+//   if (!q)
+//     return;
+//
+//   if (q->slots) {
+//     plat_mem_release(q->slots, 0);
+//     q->slots = NULL;
+//   }
+//
+//   plat_sem_destroy(&q->items_sem);
+//
+//   q->capacity = 0;
+//   q->mask = 0;
+//
+//   atomic_store_explicit(&q->head, 0, memory_order_relaxed);
+//   atomic_store_explicit(&q->tail, 0, memory_order_relaxed);
+//   atomic_store_explicit(&q->committed, 0, memory_order_relaxed);
+// }