Kernel: Optimize sse saving/loading

Only save and load sse when new thread is trying to execute sse
instruction. There is no need to do that every time we enter kernel.

kernel/kernel/Scheduler.cpp

@@ -231,6 +231,17 @@ namespace Kernel
 
 		Thread* current = &current_thread();
 
+#if __enable_sse
+		if (current != Thread::sse_thread())
+		{
+			asm volatile(
+				"movq %cr0, %rax;"
+				"orq $(1 << 3), %rax;"
+				"movq %rax, %cr0"
+			);
+		}
+#endif
+
 		while (current->state() == Thread::State::Terminated)
 		{
 			Thread* thread = m_current_thread->thread;

kernel/kernel/Syscall.cpp

@@ -26,10 +26,6 @@ namespace Kernel
 		Thread::current().set_return_rsp(interrupt_stack.rsp);
 		Thread::current().set_return_rip(interrupt_stack.rip);
 
-#if __enable_sse
-		Thread::current().save_sse();
-#endif
-
 		asm volatile("sti");
 
 		(void)arg1;
@@ -230,10 +226,6 @@ namespace Kernel
 
 		ASSERT(Kernel::Thread::current().state() == Kernel::Thread::State::Executing);
 
-#if __enable_sse
-		current_thread.load_sse();
-#endif
-
 		if (ret.is_error())
 			return -ret.error().get_error_code();
 		return ret.value();

kernel/kernel/Thread.cpp

@@ -101,7 +101,16 @@ namespace Kernel
 		: m_tid(tid), m_process(process)
 	{
 #if __enable_sse
+		uintptr_t cr0;
+		asm volatile(
+			"movq %%cr0, %%rax;"
+			"movq %%rax, %%rbx;"
+			"andq $~(1 << 3), %%rax;"
+			"movq %%rax, %%cr0;"
+			: "=b"(cr0)
+		);
 		save_sse();
+		asm volatile("movq %0, %%cr0" :: "r"(cr0));
 #endif
 	}
 
@@ -361,4 +370,24 @@ namespace Kernel
 		ASSERT_NOT_REACHED();
 	}
 
+#if __enable_sse
+	static Thread* s_sse_thread = nullptr;
+
+	void Thread::save_sse()
+	{
+		asm volatile("fxsave %0" :: "m"(m_sse_storage));
+	}
+
+	void Thread::load_sse()
+	{
+		asm volatile("fxrstor %0" :: "m"(m_sse_storage));
+		s_sse_thread = this;
+	}
+
+	Thread* Thread::sse_thread()
+	{
+		return s_sse_thread;
+	}
+#endif
+
 }