diff --git a/.gitignore b/.gitignore
index 003cb9a1743cb674846ffadc7d850480227e07bf..fc8082d8b494800577d5951d3d104a872ac17a43 100644
--- a/.gitignore
+++ b/.gitignore
@@ -10,3 +10,4 @@ docs/html
/build*
/.vs
/CMakeSettings.json
+.vscode/*
diff --git a/.vscode/launch.json b/.vscode/launch.json
index c86bce66099f93f2a10bf28bb18c8a0b3be623df..da7ac6001e96c9e5fb6c91597bc3db3c5eb0abcd 100644
--- a/.vscode/launch.json
+++ b/.vscode/launch.json
@@ -5,12 +5,16 @@
"version": "0.2.0",
"configurations": [
{
- "name": "g++ - Debug Node",
+ "name": "(gdb) Launch",
"type": "cppdbg",
"request": "launch",
+ "program": "${command:cmake.launchTargetPath}",
"args": [],
"stopAtEntry": false,
"cwd": "${workspaceFolder}/build",
+ "environment": [
+ {"name" : "ASAN_OPTIONS", "value" : "abort_on_error=1,protect_shadow_gap=0"}
+ ],
"externalConsole": false,
"MIMode": "gdb",
"setupCommands": [
@@ -18,15 +22,44 @@
"description": "Enable pretty-printing for gdb",
"text": "-enable-pretty-printing",
"ignoreFailures": true
+ },
+ {
+ "description": "Set Disassembly Flavor to Intel",
+ "text": "-gdb-set disassembly-flavor intel",
+ "ignoreFailures": true
}
],
- "miDebuggerPath": "/usr/bin/gdb",
"sourceFileMap": {
"${workspaceFolder}": {
"editorPath": "${workspaceFolder}",
"useForBreakpoints": "true"
}
- }
+ }
+ },
+ {
+ "name": "(gdb) Attach",
+ "type": "cppdbg",
+ "request": "attach",
+ "program": "${command:cmake.launchTargetPath}",
+ "MIMode": "gdb",
+ "setupCommands": [
+ {
+ "description": "Enable pretty-printing for gdb",
+ "text": "-enable-pretty-printing",
+ "ignoreFailures": true
+ },
+ {
+ "description": "Set Disassembly Flavor to Intel",
+ "text": "-gdb-set disassembly-flavor intel",
+ "ignoreFailures": true
+ }
+ ],
+ "sourceFileMap": {
+ "${workspaceFolder}": {
+ "editorPath": "${workspaceFolder}",
+ "useForBreakpoints": "true"
+ }
+ }
}
]
}
\ No newline at end of file
diff --git a/.vscode/settings.json b/.vscode/settings.json
index 713f547df23addb0c74a2d6fe5b32910b176613b..c02bff1583ead57cd94650c2d90a4118d5c1ee91 100644
--- a/.vscode/settings.json
+++ b/.vscode/settings.json
@@ -66,7 +66,18 @@
"valarray": "cpp",
"variant": "cpp",
"any": "cpp",
- "complex": "cpp"
+ "complex": "cpp",
+ "csignal": "cpp",
+ "compare": "cpp",
+ "concepts": "cpp",
+ "numbers": "cpp",
+ "ranges": "cpp",
+ "span": "cpp",
+ "stop_token": "cpp",
+ "typeindex": "cpp",
+ "semaphore": "cpp",
+ "*.ipp": "cpp"
},
- "cmake.cmakePath": "cmake"
+ "cmake.cmakePath": "cmake",
+ "cmake.configureOnOpen": true
}
\ No newline at end of file
diff --git a/CMakeLists.txt b/CMakeLists.txt
index c7656ec7d0d907f108fc257702f320c557a2c2ee..5b5605d9bbbf9fc16fb36082af27eb4ff1465726 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,6 +2,7 @@ cmake_minimum_required (VERSION 3.18.0)
include (CheckIncludeFile)
include (CheckIncludeFileCXX)
include (CheckFunctionExists)
+include (FetchContent)
if (NOT FTL_VERSION)
set(FTL_VERSION 0.0.1)
@@ -18,6 +19,8 @@ option(WITH_GNUTLS "Enable TLS support" ON)
option(USE_CPPCHECK "Apply cppcheck during build" ON)
option(BUILD_TESTS "Compile all unit and integration tests" ON)
option(BUILD_EXAMPLES "Compile the examples" ON)
+option(ENABLE_PROFILER "Enable builtin performance profiling" OFF)
+option(DEBUG_LOCKS "Enable lock profiling (requires ENABLE_PROFILER)" OFF)
if (NOT WIN32)
option(WITH_PYTHON "Enable python support" ON)
@@ -98,6 +101,7 @@ if (USE_CPPCHECK)
endif()
endif()
+include(ftl_tracy)
include(git_version)
include(ftl_paths)
@@ -177,10 +181,12 @@ target_include_directories(beyond-common PUBLIC
$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
$<INSTALL_INTERFACE:include>)
+target_link_libraries(beyond-common PUBLIC Tracy)
+
add_library(beyond-protocol STATIC
$<TARGET_OBJECTS:beyond-common>
- src/peer.cpp
- src/universe.cpp
+ src/peer.cpp
+ src/universe.cpp
src/socket/socket.cpp
src/protocol/connection.cpp
src/protocol/factory.cpp
@@ -207,7 +213,7 @@ add_library(beyond-protocol STATIC
target_include_directories(beyond-protocol PUBLIC
$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
$<INSTALL_INTERFACE:include>)
-target_link_libraries(beyond-protocol Threads::Threads ${UUID_LIBRARIES})
+target_link_libraries(beyond-protocol Threads::Threads ${UUID_LIBRARIES} Tracy)
if (WITH_GNUTLS)
target_link_libraries(beyond-protocol GnuTLS::GnuTLS)
@@ -224,5 +230,5 @@ if (BUILD_TESTS)
endif()
if (BUILD_EXAMPLES)
- add_subdirectory(examples)
+ add_subdirectory(examples)
endif()
diff --git a/cmake/ftl_tracy.cmake b/cmake/ftl_tracy.cmake
new file mode 100644
index 0000000000000000000000000000000000000000..473c8d30ecda60789d9f9354a654665443ef61c5
--- /dev/null
+++ b/cmake/ftl_tracy.cmake
@@ -0,0 +1,25 @@
+if (ENABLE_PROFILER)
+ set(TRACY_CXX_OPTIONS "-DTRACY_ENABLE -DTRACY_DELAYED_INIT -DTRACY_VERBOSE -DNOMINMAX")
+
+ if (DEBUG_LOCKS)
+ set(TRACY_CXX_OPTIONS "${TRACY_CXX_OPTIONS} -DDEBUG_LOCKS")
+ message(STATUS "Lock profiling enabled")
+ endif()
+
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TRACY_CXX_OPTIONS}")
+
+ FetchContent_Declare(tracy
+ GIT_REPOSITORY https://github.com/wolfpld/tracy.git
+ GIT_TAG 897aec5b062664d2485f4f9a213715d2e527e0ca # tags/v0.9.1
+ GIT_SHALLOW TRUE
+ GIT_PROGRESS TRUE
+ )
+ FetchContent_MakeAvailable(tracy)
+
+ add_library(Tracy ALIAS TracyClient)
+
+ message(STATUS "Profiling (Tracy) enabled")
+
+else()
+ add_library(Tracy INTERFACE)
+endif()
diff --git a/include/ftl/profiler.hpp b/include/ftl/profiler.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..8b6afbbaa912f10c576313860542bc00b818d659
--- /dev/null
+++ b/include/ftl/profiler.hpp
@@ -0,0 +1,40 @@
+/**
+ * @file profiler.hpp
+ * @copyright Copyright (c) 2020-2022 University of Turku, MIT License
+ * @author Nicolas Pope, Sebastian Hahta
+ */
+
+#pragma once
+
+#ifdef TRACY_ENABLE
+
+#include <tracy/Tracy.hpp>
+
+#define FTL_PROFILE_SCOPE(LABEL) ZoneScopedN(LABEL)
+
+// NOTE: Tracy expects Label to be a pointer to same address (this should be the case
+// with GCC and MSVC with string pooling). If not, label has to be defined
+// separately before use (static const char* ...) and exported if necessary
+
+/** Mark (secondary) frame start and stop. Each FTL_PROFILE_FRAME_BEGIN MUST be matched
+ * with FTL_PROFILE_FRAME_END */
+#define FTL_PROFILE_FRAME_BEGIN(LABEL) FrameMarkStart(#LABEL)
+#define FTL_PROFILE_FRAME_END(LABEL) FrameMarkEnd(#LABEL)
+
+/** Mark end of primary frame (main rendering/capture loop etc, if applicable) */
+#define FTL_PROFILE_PRIMARY_FRAME_END() FrameMark
+
+/// deprecated
+#define FTL_Profile(LABEL, LIMIT) FTL_PROFILE_SCOPE(LABEL)
+
+#else
+
+#define FTL_PROFILE_FRAME_BEGIN(LABEL) {}
+#define FTL_PROFILE_FRAME_END(LABEL) {}
+#define FTL_PROFILE_PRIMARY_FRAME_END() {}
+#define FTL_PROFILE_SCOPE(LABEL) {}
+
+/// deprectated
+#define FTL_Profile(LABEL, LIMIT) {}
+
+#endif
diff --git a/include/ftl/threads.hpp b/include/ftl/threads.hpp
index 0d62a76ab0d616dfdb060da269121f64cf032203..92f9bda15f3143d215baeead0dca1b3536e1ad61 100644
--- a/include/ftl/threads.hpp
+++ b/include/ftl/threads.hpp
@@ -12,29 +12,55 @@
#define POOL_SIZE 10
-// #define DEBUG_MUTEX
-#define MUTEX_TIMEOUT 2
+/// consider using DECLARE_MUTEX(name) which allows (optional) profiling
+#define MUTEX std::mutex
+/// consider using DECLARE_RECURSIVE_MUTEX(name) which allows (optional) profiling
+#define RECURSIVE_MUTEX std::recursive_mutex
+/// consider using DECLARE_SHARED_MUTEX(name) which allows (optional) profiling
+#define SHARED_MUTEX std::shared_mutex
+
+#if defined(TRACY_ENABLE)
-#if defined DEBUG_MUTEX
-#include <ftl/lib/loguru.hpp>
-#include <chrono>
#include <type_traits>
+#include <tracy/Tracy.hpp>
-#define MUTEX std::timed_mutex
-#define RECURSIVE_MUTEX std::recursive_timed_mutex
-#define SHARED_MUTEX std::shared_timed_mutex
+#define DECLARE_MUTEX(varname) TracyLockable(MUTEX, varname)
+#define DECLARE_RECURSIVE_MUTEX(varname) TracyLockable(RECURSIVE_MUTEX, varname)
+#define DECLARE_SHARED_MUTEX(varname) TracySharedLockable(SHARED_MUTEX, varname)
-#define UNIQUE_LOCK(M, L) std::unique_lock<std::remove_reference<decltype(M)>::type> L(M, std::chrono::milliseconds(MUTEX_TIMEOUT)); while (!L) { LOG(ERROR) << "Mutex timeout"; L.try_lock_for(std::chrono::milliseconds(MUTEX_TIMEOUT)); };
-#define SHARED_LOCK(M, L) std::shared_lock<std::remove_reference<decltype(M)>::type> L(M, std::chrono::milliseconds(MUTEX_TIMEOUT)); while (!L) { LOG(ERROR) << "Mutex timeout"; L.try_lock_for(std::chrono::milliseconds(MUTEX_TIMEOUT)); };
+#define DECLARE_MUTEX_N(varname, name) TracyLockableN(MUTEX, varname, name)
+#define DECLARE_RECURSIVE_MUTEX_N(varname, name) TracyLockableN(RECURSIVE_MUTEX, varname, name)
+#define DECLARE_SHARED_MUTEX_N(varname, name) TracySharedLockableN(SHARED_MUTEX, varname, name)
+
+/// mark lock acquired (mutex M);
+#define MARK_LOCK_AQUIRED(M) LockMark(M)
+// TODO: should automatic, but requires mutexes to be declared with DECLARE_..._MUTEX macros
+
+#define T_UNIQUE_LOCK(M) std::unique_lock<std::remove_reference<decltype(M)>::type>
+#define UNIQUE_LOCK(M, L) std::unique_lock<std::remove_reference<decltype(M)>::type> L(M)
+#define SHARED_LOCK(M, L) std::shared_lock<std::remove_reference<decltype(M)>::type> L(M)
#else
-#define MUTEX std::mutex
-#define RECURSIVE_MUTEX std::recursive_mutex
-#define SHARED_MUTEX std::shared_mutex
-#define UNIQUE_LOCK(M, L) std::unique_lock<std::remove_reference<decltype(M)>::type> L(M);
-#define SHARED_LOCK(M, L) std::shared_lock<std::remove_reference<decltype(M)>::type> L(M);
-#endif // DEBUG_MUTEX
+/// mutex with optional profiling (and debugging) when built with PROFILE_MUTEX.
+#define DECLARE_MUTEX(varname) MUTEX varname
+/// recursive mutex with optional profiling (and debugging) when built with PROFILE_MUTEX
+#define DECLARE_RECURSIVE_MUTEX(varname) RECURSIVE_MUTEX varname
+/// shared mutex with optional profiling (and debugging) when built with PROFILE_MUTEX
+#define DECLARE_SHARED_MUTEX(varname) SHARED_MUTEX varname
+
+#define DECLARE_MUTEX_N(varname, name) MUTEX varname
+#define DECLARE_RECURSIVE_MUTEX_N(varname, name) RECURSIVE_MUTEX varname
+#define DECLARE_SHARED_MUTEX_N(varname, name) SHARED_MUTEX varname
+
+/// mark lock acquired (mutex M)
+#define MARK_LOCK(M) {}
+
+#define T_UNIQUE_LOCK(M) std::unique_lock<std::remove_reference<decltype(M)>::type>
+#define UNIQUE_LOCK(M, L) std::unique_lock<std::remove_reference<decltype(M)>::type> L(M)
+#define SHARED_LOCK(M, L) std::shared_lock<std::remove_reference<decltype(M)>::type> L(M)
+
+#endif // TRACY_ENABLE
#define SHARED_LOCK_TYPE(M) std::shared_lock<M>
diff --git a/src/ctpl_stl.cpp b/src/ctpl_stl.cpp
index 947560af1898aeec977e707f367959e75aebf701..7d4e8f98cc50d59ef6858510fce2c8f4d7602c89 100644
--- a/src/ctpl_stl.cpp
+++ b/src/ctpl_stl.cpp
@@ -5,9 +5,20 @@
#include <ftl/lib/ctpl_stl.hpp>
+#ifdef TRACY_ENABLE
+#include <tracy/Tracy.hpp>
+#endif
+
void ctpl::thread_pool::set_thread(int i) {
std::shared_ptr<std::atomic<bool>> flag(this->flags[i]); // a copy of the shared ptr to the flag
auto f = [this, i, flag/* a copy of the shared ptr to the flag */]() {
+ #if TRACY_ENABLE
+ {
+ const auto thread_name = "thread_pool/" + std::to_string(i);
+ tracy::SetThreadName(thread_name.c_str());
+ }
+ #endif
+
std::atomic<bool> & _flag = *flag;
std::function<void(int id)> * _f;
bool isPop = this->q.pop(_f);
diff --git a/src/peer.hpp b/src/peer.hpp
index de3fd9da769996dbfa7d3e22a1ea3dc30d2fa3ba..4e7d742e5b794a2e3493cda57a084d35ab3198e2 100644
--- a/src/peer.hpp
+++ b/src/peer.hpp
@@ -260,9 +260,8 @@ class Peer {
// Send buffers
msgpack::vrefbuffer send_buf_;
- RECURSIVE_MUTEX send_mtx_;
-
- RECURSIVE_MUTEX cb_mtx_;
+ DECLARE_RECURSIVE_MUTEX(send_mtx_);
+ DECLARE_RECURSIVE_MUTEX(cb_mtx_);
const bool outgoing_;
unsigned int local_id_;
diff --git a/src/protocol/connection.cpp b/src/protocol/connection.cpp
index 52136952157bba13a680178ba62310ba4e3ea0bc..e1528da309484190402a926a5aadc954a38fbd24 100644
--- a/src/protocol/connection.cpp
+++ b/src/protocol/connection.cpp
@@ -111,7 +111,7 @@ ssize_t SocketConnection::writev(const struct iovec *iov, int iovcnt) {
writev_calls++;
if (sent < 0) {
- DLOG(ERROR) << "writev(): " << sock_.get_error_string();
+ LOG(2) << "writev(): " << sock_.get_error_string();
if (sock_.is_fatal()) {
return sent;
}
@@ -121,7 +121,7 @@ ssize_t SocketConnection::writev(const struct iovec *iov, int iovcnt) {
sent_total += sent;
}
- DLOG(2) << "message required " << writev_calls << " writev() calls";
+ LOG(2) << "message required " << writev_calls << " writev() calls";
/*if (can_increase_sock_buffer_) {
auto send_buf_size = sock_.get_send_buffer_size();
diff --git a/src/universe.cpp b/src/universe.cpp
index 21f6563303883ee68df53b9920947f8af095c761..57af06a1e0721d24abe2c55ac86a73addc0e3c7f 100644
--- a/src/universe.cpp
+++ b/src/universe.cpp
@@ -75,6 +75,7 @@ struct NetImplDetail {
// TODO(Seb): move to ServerSocket and ClientSocket
// Defaults, should be changed in config
+
#define TCP_SEND_BUFFER_SIZE (32*1024*1024)
#define TCP_RECEIVE_BUFFER_SIZE (32*1024*1024) // Perhaps try 24K?
#define WS_SEND_BUFFER_SIZE (32*1024*1024)