test_after_aot.py

# Owner(s): ["module: dynamo"]

import io
import os
import shutil
import sys
import tempfile
import unittest

import torch._dynamo.test_case

from torch._dynamo.repro.after_aot import InputReader, InputWriter, save_graph_repro

from torch.fx.experimental.proxy_tensor import make_fx
from torch.testing._internal.common_utils import IS_FBCODE
from torch.utils._traceback import report_compile_source_on_error


def strip_trailing_whitespace(r):
    return "\n".join([l.rstrip() for l in r.split("\n")])


class TestAfterAot(torch._dynamo.test_case.TestCase):
    @unittest.skipIf(IS_FBCODE, "NotImplementedError")
    def test_save_graph_repro(self):
        # TODO: This triggers CUDA context initialization, even though
        # it is CPU only
        buf = io.StringIO()
        args = [torch.randn(4)]

        def f(x):
            return (x * x,)

        gm = make_fx(f)(*args)
        with tempfile.TemporaryDirectory() as d:
            save_graph_repro(buf, gm, args, "inductor_accuracy", save_dir=d)
            r = buf.getvalue()
            with report_compile_source_on_error():
                exec(r, {"__compile_source__": r})

            shutil.rmtree(os.path.join(d, "storages"))

            # Should still work even without the save dir
            with report_compile_source_on_error():
                exec(r, {"__compile_source__": r})

    @unittest.skipIf(sys.byteorder != "little", "checksum depends on endianness")
    def test_dump_tensor(self):
        def test(tensor, expected):
            with tempfile.TemporaryDirectory() as d:
                writer = InputWriter(d, stable_hash=True)
                writer.tensor("x", tensor)
                self.assertExpectedInline("\n".join(writer._lines), expected, skip=1)
                reader = InputReader(d)
                env = {"reader": reader, "torch": torch}
                # TODO: assert no logs
                exec("\n".join(writer._lines), env)
                self.assertEqual(reader.args[0], tensor)

        test(
            torch.zeros(3, 4),
            """\
buf0 = reader.storage('c17fd92682ca5b304ac71074b558dda9e8eb4d66', 48)
reader.tensor(buf0, (3, 4), is_leaf=True)  # x""",
        )
        test(
            torch.ones(3, 4, dtype=torch.int32),
            """\
buf0 = reader.storage('7c221e2da0c58c700cc2996644dd13d042bd552e', 48, dtype_hint=torch.int32)
reader.tensor(buf0, (3, 4), dtype=torch.int32, is_leaf=True)  # x""",
        )
        test(
            torch.empty((3, 4, 5, 6), memory_format=torch.channels_last).fill_(2),
            """\
buf0 = reader.storage('49ebab3961d6221e64c4c72b0aefd976bdd2afc4', 1440)
reader.tensor(buf0, (3, 4, 5, 6), (120, 1, 24, 4), is_leaf=True)  # x""",
        )


if __name__ == "__main__":
    from torch._dynamo.test_case import run_tests

    run_tests()