Reduce number of JS <-> WASM jumps in core CPU loop

Similar to mihaip/dingusppc@d0ff52d, the use
of setjmp/longjmp was leading to a lot of js-to-wasm and wasm-to-js due to
invoke_vii. Move the inner loop to a separate non-inlined function.

Reduces the boot times for NeXTSTEP 3.3. Likely some of the longer pauses
are due to network and I/O timeouts, the CPU bound parts to feel a lot
snappier.

                          Before   After
Show ROM Monitor Window   3s       1s
Show boot command         5s       2s
Begin OS boot            14s       7s
Hide OS boot window      50s      42s
Workspace rendered       60s      49s

Author: mihaip

src/cpu/newcpu.c

@@ -6273,23 +6273,18 @@ static void m68k_run_mmu060 (void)
 
 #ifdef CPUEMU_31
 
-/* Aranym MMU 68040  */
-static void m68k_run_mmu040 (void)
-{
-	struct flag_struct f;
-	int halt = 0;
-
-	check_halt();
-#ifdef WINUAE_FOR_HATARI
-	Log_Printf(LOG_DEBUG,  "m68k_run_mmu040\n");
+static
+// Don't inline this function under Emscripten, otherwise we will end up with
+// very inefficient code generation due to the setjmp call in the parent
+// ppc_exec() function.
+#ifdef EMSCRIPTEN
+__attribute__((noinline))
 #endif
-
-	while (!halt) {
-		check_debugger();
-		TRY (prb) {
+void m68k_run_mmu040_inner(struct flag_struct *f)
+{
 	for (;;) {
-				f.cznv = regflags.cznv;
-				f.x = regflags.x;
+		f->cznv = regflags.cznv;
+		f->x = regflags.x;
 		mmu_restart = true;
 		regs.instruction_pc = m68k_getpc ();
 
@@ -6313,6 +6308,24 @@ static void m68k_run_mmu040 (void)
 			}
 		}
 	}
+}
+
+
+/* Aranym MMU 68040  */
+static void m68k_run_mmu040 (void)
+{
+	struct flag_struct f;
+	int halt = 0;
+
+	check_halt();
+#ifdef WINUAE_FOR_HATARI
+	Log_Printf(LOG_DEBUG,  "m68k_run_mmu040\n");
+#endif
+
+	while (!halt) {
+		check_debugger();
+		TRY (prb) {
+			m68k_run_mmu040_inner(&f);
 		} CATCH (prb) {
 
 			if (mmu_restart) {