Improve function call performance

[ksh8281]

• Divide function into user-defined function and user-defined function /w try-catch
• Non try-catch user-defined function should not use c++ try-catch
• Reduce parameter count of Interpreter::interpret

[ksh8281]

test	./work/walrus [s]	origin/walrus [s]
change	3.024 (1.000x)	3.148 (1.041x)
factorial	3.072 (1.000x)	3.296 (1.073x)
fannkuch	2.546 (1.000x)	2.677 (1.051x)
fibonacci	4.335 (1.000x)	4.864 (1.122x)
gregory	3.404 (1.000x)	3.470 (1.019x)
hanoi	3.939 (1.000x)	4.132 (1.049x)
heapsort	2.887 (1.000x)	3.062 (1.060x)
huffman	4.248 (1.000x)	4.035 (0.950x)
k_nucleotide	3.339 (1.000x)	3.356 (1.005x)
mandelbrot	4.450 (1.000x)	4.676 (1.051x)
nbody	3.711 (1.000x)	3.716 (1.001x)
nqueens	3.608 (1.000x)	3.595 (0.997x)
prime	3.921 (1.000x)	4.029 (1.028x)
quick_sort	3.169 (1.000x)	3.093 (0.976x)
red-black	4.139 (1.000x)	4.222 (1.020x)
salesman	3.792 (1.000x)	3.664 (0.966x)

[clover2123]

Why don't you initialize memories as instance->m_memories?

[clover2123]

And is it enough to initialize local memories only?
Other structures like tables, globals don't need local variables since they are not accessed frequently like memories, right?

[ksh8281]

Why don't you initialize memories as instance->m_memories?
-> I tested two way of init memories.
. access instance->m_memories -> needs memory access
. computing address(reinterpret_cast<Memory**>(reinterpret_cast<uintptr_t>(instance) + Instance::alignedSize())) -> needs ALU on CPU

the second way shows better performance when I tested

And is it enough to initialize local memories only?
-> Yes. only memories is used frequently when I tested
but globals and tables are don't. but it needs a space on stack

[clover2123]

LGTM