Threading: Add lightweight thread wrapper

common/Darwin/DarwinThreads.cpp

@@ -16,13 +16,15 @@
 #if defined(__APPLE__)
 
 #include <sched.h>
+#include <pthread.h>
 #include <unistd.h>
 #include <mach/mach_init.h>
 #include <mach/thread_act.h>
 #include <mach/mach_port.h>
 
 #include "common/PrecompiledHeader.h"
 #include "common/Threading.h"
+#include "common/Assertions.h"
 
 // Note: assuming multicore is safer because it forces the interlocked routines to use
 // the LOCK prefix.  The prefix works on single core CPUs fine (but is slow), but not
@@ -146,6 +148,86 @@ bool Threading::ThreadHandle::SetAffinity(u64 processor_mask) const
 	return false;
 }
 
+Threading::Thread::Thread() = default;
+
+Threading::Thread::Thread(Thread&& thread)
+	: ThreadHandle(thread)
+	, m_stack_size(thread.m_stack_size)
+{
+	thread.m_stack_size = 0;
+}
+
+Threading::Thread::Thread(EntryPoint func)
+	: ThreadHandle()
+{
+	if (!Start(std::move(func)))
+		pxFailRel("Failed to start implicitly started thread.");
+}
+
+Threading::Thread::~Thread()
+{
+	pxAssertRel(!m_native_handle, "Thread should be detached or joined at destruction");
+}
+
+void Threading::Thread::SetStackSize(u32 size)
+{
+	pxAssertRel(!m_native_handle, "Can't change the stack size on a started thread");
+	m_stack_size = size;
+}
+
+void* Threading::Thread::ThreadProc(void* param)
+{
+	std::unique_ptr<EntryPoint> entry(static_cast<EntryPoint*>(param));
+	(*entry.get())();
+	return nullptr;
+}
+
+bool Threading::Thread::Start(EntryPoint func)
+{
+	pxAssertRel(!m_native_handle, "Can't start an already-started thread");
+
+	std::unique_ptr<EntryPoint> func_clone(std::make_unique<EntryPoint>(std::move(func)));
+
+	pthread_attr_t attrs;
+	bool has_attributes = false;
+
+	if (m_stack_size != 0)
+	{
+		has_attributes = true;
+		pthread_attr_init(&attrs);
+	}
+	if (m_stack_size != 0)
+		pthread_attr_setstacksize(&attrs, m_stack_size);
+
+	pthread_t handle;
+	const int res = pthread_create(&handle, has_attributes ? &attrs : nullptr, ThreadProc, func_clone.get());
+	if (res != 0)
+		return false;
+
+	// thread started, it'll release the memory
+	m_native_handle = (void*)handle;
+	func_clone.release();
+	return true;
+}
+
+void Threading::Thread::Detach()
+{
+	pxAssertRel(m_native_handle, "Can't detach without a thread");
+	pthread_detach((pthread_t)m_native_handle);
+	m_native_handle = nullptr;
+}
+
+void Threading::Thread::Join()
+{
+	pxAssertRel(m_native_handle, "Can't join without a thread");
+	void* retval;
+	const int res = pthread_join((pthread_t)m_native_handle, &retval);
+	if (res != 0)
+		pxFailRel("pthread_join() for thread join failed");
+
+	m_native_handle = nullptr;
+}
+
 // name can be up to 16 bytes
 void Threading::SetNameOfCurrentThread(const char* name)
 {

common/Linux/LnxThreads.cpp

@@ -18,6 +18,9 @@
 #define _GNU_SOURCE
 #endif
 
+#include <memory>
+
+#include <pthread.h>
 #include <unistd.h>
 #if defined(__linux__)
 #include <sys/prctl.h>
@@ -35,6 +38,7 @@
 #endif
 
 #include "common/Threading.h"
+#include "common/Assertions.h"
 
 // We wont need this until we actually have this more then just stubbed out, so I'm commenting this out
 // to remove an unneeded dependency.
@@ -204,6 +208,159 @@ bool Threading::ThreadHandle::SetAffinity(u64 processor_mask) const
 #endif
 }
 
+Threading::Thread::Thread() = default;
+
+Threading::Thread::Thread(Thread&& thread)
+	: ThreadHandle(thread)
+	, m_stack_size(thread.m_stack_size)
+{
+	thread.m_stack_size = 0;
+}
+
+Threading::Thread::Thread(EntryPoint func)
+	: ThreadHandle()
+{
+	if (!Start(std::move(func)))
+		pxFailRel("Failed to start implicitly started thread.");
+}
+
+Threading::Thread::~Thread()
+{
+	pxAssertRel(!m_native_handle, "Thread should be detached or joined at destruction");
+}
+
+void Threading::Thread::SetStackSize(u32 size)
+{
+	pxAssertRel(!m_native_handle, "Can't change the stack size on a started thread");
+	m_stack_size = size;
+}
+
+#ifdef __linux__
+// For Linux, we have to do a bit of trickery here to get the thread's ID back from
+// the thread itself, because it's not part of pthreads. We use a semaphore to signal
+// when the thread has started, and filled in thread_id_ptr.
+struct ThreadProcParameters
+{
+	Threading::Thread::EntryPoint func;
+	Threading::KernelSemaphore* start_semaphore;
+	unsigned int* thread_id_ptr;
+};
+
+void* Threading::Thread::ThreadProc(void* param)
+{
+	std::unique_ptr<ThreadProcParameters> entry(static_cast<ThreadProcParameters*>(param));
+	*entry->thread_id_ptr = gettid();
+	entry->start_semaphore->Post();
+	entry->func();
+	return nullptr;
+}
+
+bool Threading::Thread::Start(EntryPoint func)
+{
+	pxAssertRel(!m_native_handle, "Can't start an already-started thread");
+
+	KernelSemaphore start_semaphore;
+	std::unique_ptr<ThreadProcParameters> params(std::make_unique<ThreadProcParameters>());
+	params->func = std::move(func);
+	params->start_semaphore = &start_semaphore;
+	params->thread_id_ptr = &m_native_id;
+
+	pthread_attr_t attrs;
+	bool has_attributes = false;
+
+	if (m_stack_size != 0)
+	{
+		has_attributes = true;
+		pthread_attr_init(&attrs);
+	}
+	if (m_stack_size != 0)
+		pthread_attr_setstacksize(&attrs, m_stack_size);
+
+	pthread_t handle;
+	const int res = pthread_create(&handle, has_attributes ? &attrs : nullptr, ThreadProc, params.get());
+	if (res != 0)
+		return false;
+
+	// wait until it sets our native id
+	start_semaphore.Wait();
+
+	// thread started, it'll release the memory
+	m_native_handle = (void*)handle;
+	params.release();
+	return true;
+}
+
+#else
+
+void* Threading::Thread::ThreadProc(void* param)
+{
+	std::unique_ptr<EntryPoint> entry(static_cast<EntryPoint*>(param));
+	(*entry.get())();
+	return nullptr;
+}
+
+bool Threading::Thread::Start(EntryPoint func)
+{
+	pxAssertRel(!m_native_handle, "Can't start an already-started thread");
+
+	std::unique_ptr<EntryPoint> func_clone(std::make_unique<EntryPoint>(std::move(func)));
+
+	pthread_attr_t attrs;
+	bool has_attributes = false;
+
+	if (m_stack_size != 0)
+	{
+		has_attributes = true;
+		pthread_attr_init(&attrs);
+	}
+	if (m_stack_size != 0)
+		pthread_attr_setstacksize(&attrs, m_stack_size);
+
+	pthread_t handle;
+	const int res = pthread_create(&handle, has_attributes ? &attrs : nullptr, ThreadProc, func_clone.get());
+	if (res != 0)
+		return false;
+
+	// thread started, it'll release the memory
+	m_native_handle = (void*)handle;
+	func_clone.release();
+	return true;
+}
+
+#endif
+
+void Threading::Thread::Detach()
+{
+	pxAssertRel(m_native_handle, "Can't detach without a thread");
+	pthread_detach((pthread_t)m_native_handle);
+	m_native_handle = nullptr;
+#ifdef __linux__
+	m_native_id = 0;
+#endif
+}
+
+void Threading::Thread::Join()
+{
+	pxAssertRel(m_native_handle, "Can't join without a thread");
+	void* retval;
+	const int res = pthread_join((pthread_t)m_native_handle, &retval);
+	if (res != 0)
+		pxFailRel("pthread_join() for thread join failed");
+
+	m_native_handle = nullptr;
+#ifdef __linux__
+	m_native_id = 0;
+#endif
+}
+
+Threading::ThreadHandle& Threading::Thread::operator=(Thread&& thread)
+{
+	ThreadHandle::operator=(thread);
+	m_stack_size = thread.m_stack_size;
+	thread.m_stack_size = 0;
+	return *this;
+}
+
 void Threading::SetNameOfCurrentThread(const char* name)
 {
 #if defined(__linux__)

common/Threading.h

@@ -25,6 +25,7 @@
 #endif
 
 #include <atomic>
+#include <functional>
 
 namespace Threading
 {
@@ -84,7 +85,7 @@ namespace Threading
 		/// Obviously, only works up to 64 processors.
 		bool SetAffinity(u64 processor_mask) const;
 
-	private:
+	protected:
 		void* m_native_handle = nullptr;
 
 		// We need the thread ID for affinity adjustments on Linux.
@@ -93,6 +94,46 @@ namespace Threading
 #endif
 	};
 
+	// --------------------------------------------------------------------------------------
+	//  Thread
+	// --------------------------------------------------------------------------------------
+	// Abstracts a native thread in a lightweight manner. Provides more functionality than
+	// std::thread (allowing stack size adjustments).
+	//
+	class Thread : public ThreadHandle
+	{
+	public:
+		using EntryPoint = std::function<void()>;
+
+		Thread();
+		Thread(Thread&& thread);
+		Thread(const Thread&) = delete;
+		Thread(EntryPoint func);
+		~Thread();
+
+		ThreadHandle& operator=(Thread&& thread);
+		ThreadHandle& operator=(const Thread& handle) = delete;
+
+		__fi bool Joinable() const { return (m_native_handle != nullptr); }
+		__fi u32 GetStackSize() const { return m_stack_size; }
+
+		/// Sets the stack size for the thread. Do not call if the thread has already been started.
+		void SetStackSize(u32 size);
+
+		bool Start(EntryPoint func);
+		void Detach();
+		void Join();
+
+	protected:
+#ifdef _WIN32
+		static unsigned __stdcall ThreadProc(void* param);
+#else
+		static void* ThreadProc(void* param);
+#endif
+
+		u32 m_stack_size = 0;
+	};
+
 	/// A semaphore that may not have a fast userspace path
 	/// (Used in other semaphore-based algorithms where the semaphore is just used for its thread sleep/wake ability)
 	class KernelSemaphore