[dynamo] Missing support for many trivial builtin functions in operator

🐛 Describe the bug

We are missing support for a few operators, operators.iconcat and operator.concat

import functools
import operator
import torch

def fn_works() -> None:
    a = list(range(1, 3))
    b = list(range(3, 6))
    c = [a, b]
    d = functools.reduce(operator.add, c, [])
    e = functools.reduce(operator.iadd, c, [])

def fn() -> None:
    a = list(range(1, 3))
    b = list(range(3, 6))
    c = [a, b]
    d = functools.reduce(operator.concat, c, [])
    e = functools.reduce(operator.iconcat, c, [])

comp_out_works = torch._dynamo.optimize(nopython=True)(fn_works)()
comp_out = torch._dynamo.optimize(nopython=True)(fn)()

Just glancing at operators documentation and the current builtin dynamo code: https://github.com/pytorch/pytorch/blob/c7e9c1510274184b41e408e6409f252bb1717085/torch/_dynamo/variables/builtin.py#L92

Things that work that I did not expect:

• operator.is
• operator.is_not
• operator.__add__ # hurray for aliasing

I think we are also missing support for:

• operator.concat
• operator.iconcat
• operator.contains # almost certainly broken. Be careful this when has reversed operands
• operator.countOf
• operator.delitem # might make sense not to fold?
• operator.setitem # might make sense not to fold?
• operator.abs # surprisingly broken given builtins.abs works
• operator.length_hint # confirmed broken
• operator.call # python 3.11+ only
• operator.attrgetter
• operator.itemgetter
• operator.methodcaller

def fn_abs() -> None:
    a = range(-10, 10)
    b = map(operator.abs, a)

comp_out = torch._dynamo.optimize(nopython=True)(fn_abs)()

@XuehaiPan want to take a look at this? You did a great job fixing some of the already broken ones.

Should probably done as we can just alias it to other builtin calls.

• operator.concat
• operator.iconcat
• operator.abs

Versions

Tried on latest stable, but I am sure it's on master too from looking at the source.

cc @ezyang @msaroufim @wconstab @bdhirsh @anijain2305 @zou3519 @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @aakhundov @kadeng