ti.max() behaves differently with built-in function max

[neozhaoliang]

Example code:

import taichi as ti
ti.init(arch=ti.cpu)


# use built-in max 1.
def main1():
    v = ti.Vector([1, 2, 3])
    w = max(v, -3)
    print(w)  # -3

# use built-in max 2.
def main2():
    v = ti.Vector([1, 2, 3])
    w = max(*v, -3)
    print(w)   # 3

# use ti.max 1.
@ti.kernel
def main3():
    v = ti.Vector([1, 2, 3])
    w = ti.max(v, -3)
    print(w)   # [1, 2, 3]

# use ti.max 2.
@ti.kernel
def main4():
    v = ti.Vector([1, 2, 3])
    w = ti.max(*v, -3) # raise error
    print(w)


main1()
#main2()
#main3()
#main4()

[lin-hitonami]

This is indeed a problem.
In python scope,

ti.max(4, ti.Vector([3, 4, 5]))  # [4, 4, 5]
max(4, ti.Vector([3, 4, 5]))  # [3, 4, 5]
max(ti.Vector([3, 4, 5]), 4)  # 4

[lin-hitonami]

ti.max performs broadcast element wise max operation if one of the operand is ti.Vector, and it only receives 2 arguments. Maybe we should replace ti.max with ti.ti_max and let users use it in python scope. cc @strongoier

[lin-hitonami]

If we change ti.max to max in main4, it also raises error because there is a bug when we process starred expressions when the expression is not a list. I will fix it soon.

[lin-hitonami]

For the accordance in Python scope and Taichi scope, we should also deprecate the use of builtin max and min in Taichi scope.