At the moment, accelerated functions are dealt with by checking (at the label do_enter_function_L1) if L1 contains the address of an accelerated function, and calling it if it is. This certainly works, but it does mean that the call to look up the accelerated function (accel_get_func()) occurs every time the call to the function is executed.
However, if the function address is a constant, it ought to be better to move this into Git's JIT logic and directly call the accelerated function. Testing would be needed to determine if this produced a noticable improvement or not. It would also mean that the accelfunc opcode would need to flush the git code cache when called.
At the moment, accelerated functions are dealt with by checking (at the label
do_enter_function_L1) if L1 contains the address of an accelerated function, and calling it if it is. This certainly works, but it does mean that the call to look up the accelerated function (accel_get_func()) occurs every time the call to the function is executed.However, if the function address is a constant, it ought to be better to move this into Git's JIT logic and directly call the accelerated function. Testing would be needed to determine if this produced a noticable improvement or not. It would also mean that the
accelfuncopcode would need to flush the git code cache when called.