test/distributed/test_c10d_ucc.py

# Owner(s): ["oncall: distributed"]

import copy
import logging
import math
import operator
import os
import random
import sys
import tempfile
from functools import reduce

import torch
import torch.distributed as c10d


if not c10d.is_available() or not c10d.is_ucc_available():
    print("c10d UCC not available, skipping tests", file=sys.stderr)
    sys.exit(0)

import test_c10d_common
from test_c10d_common import (
    gpus_for_rank,
    ModuleForDdpCommHook,
    SparseGradientModule,
    Task,
)

import torch.distributed as dist
import torch.nn.functional as F
import torch.testing._internal.common_utils as common
from torch import nn
from torch.nn.parallel import DistributedDataParallel
from torch.testing._internal.common_distributed import (
    MultiProcessTestCase,
    requires_ucc,
    skip_if_lt_x_gpu,
    verify_ddp_error_logged,
)
from torch.testing._internal.common_utils import (
    retry_on_connect_failures,
    run_tests,
    skip_but_pass_in_sandcastle,
    TestCase,
)


def simple_reduce_tests(rank, world_size):
    tests = [
        (
            c10d.ReduceOp.SUM,
            torch.tensor([rank + 1.0]),
            torch.tensor([float(world_size * (world_size + 1) / 2)]),
        ),
        (
            c10d.ReduceOp.PRODUCT,
            torch.tensor([rank + 1.0]),
            torch.tensor([float(math.factorial(world_size))]),
        ),
        (
            c10d.ReduceOp.MIN,
            torch.tensor([rank + 1.0]),
            torch.tensor([1.0]),
        ),
        (
            c10d.ReduceOp.MAX,
            torch.tensor([rank + 1.0]),
            torch.tensor([world_size]),
        ),
    ]

    # Generate tests for BAND.
    # The bit that is set changes in every iteration to check
    # that the output changes accordingly.
    for i in range(4):
        vin = rank | (1 << i)
        vout = 1 << i
        tests.append(
            (
                c10d.ReduceOp.BAND,
                torch.tensor([vin], dtype=torch.int32),
                torch.tensor([vout], dtype=torch.int32),
            ),
        )

    # Generate tests for BOR.
    # These emulate a larger world size per iteration by having every
    # rank contribute multiple values that are pre-OR'ed.
    for i in range(1, 5):
        vin = reduce(operator.or_, [rank * i + j for j in range(i)])
        vout = reduce(operator.or_, range(world_size * i))
        tests.append(
            (
                c10d.ReduceOp.BOR,
                torch.tensor([vin], dtype=torch.int32),
                torch.tensor([vout], dtype=torch.int32),
            ),
        )

    # Generate tests for XOR.
    # These emulate a larger world size per iteration by having every
    # rank contribute multiple values that are pre-XOR'ed.
    for i in range(1, 5):
        vin = reduce(operator.xor, [rank * i + j for j in range(i)])
        vout = reduce(operator.xor, range(world_size * i))
        tests.append(
            (
                c10d.ReduceOp.BXOR,
                torch.tensor([vin], dtype=torch.int32),
                torch.tensor([vout], dtype=torch.int32),
            ),
        )

    return tests


class RendezvousEnvTest(TestCase):
    @requires_ucc()
    @retry_on_connect_failures
    def test_logging_init(self):
        os.environ["WORLD_SIZE"] = "1"
        os.environ["MASTER_ADDR"] = "127.0.0.1"
        os.environ["MASTER_PORT"] = str(common.find_free_port())
        os.environ["RANK"] = "0"

        previous_handlers = logging.root.handlers

        c10d.init_process_group(backend="ucc", init_method="env://")

        current_handlers = logging.root.handlers
        self.assertEqual(len(previous_handlers), len(current_handlers))
        for current, previous in zip(current_handlers, previous_handlers):
            self.assertEqual(current, previous)

        c10d.destroy_process_group()


class TimeoutTest(test_c10d_common.AbstractTimeoutTest, TestCase):
    @requires_ucc()
    @retry_on_connect_failures
    def test_default_store_timeout_ucc(self):
        self._test_default_store_timeout("ucc")


class ProcessGroupUCCTest(MultiProcessTestCase):
    def _create_process_group_ucc(self):
        store = c10d.FileStore(self.file_name, self.world_size)
        return c10d.ProcessGroupUCC(store, self.rank, self.world_size)

    def setUp(self):
        super().setUp()
        self._spawn_processes()

    def tearDown(self):
        super().tearDown()
        try:
            os.remove(self.file_name)
        except OSError:
            pass

    @requires_ucc()
    def test_empty_tensors(self):
        pg = self._create_process_group_ucc()

        xs = [torch.FloatTensor([])]
        fut = pg.broadcast(xs).get_future()
        fut.wait()
        output = fut.value()
        self.assertEqual(0, output[0].numel())
        self.assertEqual(xs[0], output[0], exact_dtype=False)

    # TODO: add error check testing

    def _test_broadcast_basics(self, fn):
        pg = self._create_process_group_ucc()

        def broadcast(xs, rootRank, rootTensor):
            opts = c10d.BroadcastOptions()
            opts.rootRank = rootRank
            opts.rootTensor = rootTensor
            fut = pg.broadcast(xs, opts).get_future()
            fut.wait()
            return fut.value()

        # Every rank is root once
        for i in range(self.world_size):
            # Run with 1 input tensor
            x = fn(torch.tensor([self.rank]))
            output = broadcast([x], i, 0)
            self.assertEqual(torch.tensor([i]), output[0], exact_dtype=False)

            # TODO: UCC currently does not support multi tensor input

        # Test overloaded convenience function
        x = torch.tensor([self.rank + 1.0])
        fut = pg.broadcast(x, root=0).get_future()
        fut.wait()
        result = fut.value()
        self.assertEqual(torch.tensor([1.0]), result[0])

    @requires_ucc()
    def test_broadcast_basics(self):
        self._test_broadcast_basics(lambda t: t.clone())

    # TODO: test_broadcast_basics_cuda times out locally

    def _test_allreduce_basics(self, fn):
        pg = self._create_process_group_ucc()

        # Single input tests
        tests = simple_reduce_tests(self.rank, self.world_size)
        for op, input, expected in tests:
            opts = c10d.AllreduceOptions()
            opts.reduceOp = op
            tensor = fn(input)
            fut = pg.allreduce([tensor], opts).get_future()
            fut.wait()
            result = fut.value()
            self.assertEqual(expected, result[0], exact_dtype=False)

        # TODO: UCC currently does not support multi tensor input

        # Test overloaded convenience function (defaults to using sum)
        x = fn(torch.tensor([self.rank + 1.0]))
        fut = pg.allreduce(x).get_future()
        fut.wait()
        result = fut.value()
        self.assertEqual(
            torch.tensor([float(self.world_size * (self.world_size + 1) / 2)]),
            result[0],
        )

    @requires_ucc()
    def test_allreduce_basics(self):
        self._test_allreduce_basics(lambda t: t.clone())

    # TODO: test_allreduce_basics_cuda times out locally

    def _test_allgather_basics(self, fn):
        pg = self._create_process_group_ucc()

        # TODO: Run with N input tensor per rank; for now, UCC only supports single tensor input so N=1
        for n in [1]:
            input = [fn(torch.tensor([n * self.rank + i])) for i in range(n)]
            output = [
                [fn(torch.tensor([-1])) for _ in range(n * self.world_size)]
                for _ in range(n)
            ]
            expected_output = [
                [fn(torch.tensor([i])) for i in range(n * self.world_size)]
                for _ in range(n)
            ]
            fut = pg.allgather(output, input).get_future()
            fut.wait()
            result = fut.value()
            if n == 1:
                result = [result]
            self.assertEqual(expected_output, result)

    def test_allgather_basics(self):
        self._test_allgather_basics(lambda t: t.clone())

    def _test_reduce_basics(self, fn):
        pg = self._create_process_group_ucc()
        for op, input, output in simple_reduce_tests(self.rank, self.world_size):
            for root in range(self.world_size):
                opts = c10d.ReduceOptions()
                opts.reduceOp = op
                opts.rootRank = root
                tmp = fn(input)
                fut = pg.reduce([tmp], opts).get_future()
                fut.wait()
                result = fut.value()
                if root == self.rank:
                    self.assertEqual(output, result[0], exact_dtype=False)

    @requires_ucc()
    def test_reduce_basics(self):
        self._test_reduce_basics(lambda t: t.clone())

    # TODO: test_reduce_basics_cuda times out locally

    @requires_ucc()
    def test_send_recv_all_to_all(self):
        pg = self._create_process_group_ucc()

        # Preallocate tensors for input/output
        inputs = [torch.tensor([self.rank]) for _ in range(self.world_size)]
        outputs = [torch.tensor([-1]) for _ in range(self.world_size)]

        # Issue sends
        send_work = []
        for i in range(self.world_size):
            if i == self.rank:
                continue
            send_work.append(pg.send([inputs[i]], i, 0))

        # Issue recvs
        recv_work = []
        for i in range(self.world_size):
            if i == self.rank:
                continue
            recv_work.append(pg.recv([outputs[i]], i, 0))

        # Wait for sends to complete
        for work in send_work:
            work.wait()
            self.assertTrue(work.is_completed())

        # Wait for recvs to complete
        for work in recv_work:
            work.wait()
            self.assertTrue(work.is_completed())

        # Test that every output other than our own contains the respective rank
        for i in range(self.world_size):
            if i == self.rank:
                continue
            self.assertEqual(torch.tensor([i]), outputs[i])

    # TODO: test_barrier_implies_wait fails with numerical mismatch, will investigate later
    @skip_but_pass_in_sandcastle("fails with numerical mismatch, skip for now")
    @requires_ucc()
    def test_barrier_implies_wait(self):
        pg = self._create_process_group_ucc()

        # Kick off allreduce operations
        size = (100, 100)
        num = 16
        tensors = [torch.full(size, float(i)) for i in range(num)]
        for tensor in tensors:
            # Note: leak the returned work handle
            pg.allreduce(tensor)

        # Barrier should ensure all previous work has completed
        pg.barrier().get_future().wait()

        for i, tensor in enumerate(tensors):
            self.assertEqual(torch.full(size, float(i * self.world_size)), tensor)


class DistributedDataParallelTest(
    test_c10d_common.CommonDistributedDataParallelTest, MultiProcessTestCase
):
    def setUp(self):
        super().setUp()
        self._spawn_processes()

    def _get_process_group(self):
        store = self._get_store()
        c10d.init_process_group(
            "ucc", store=store, rank=self.rank, world_size=self.world_size
        )
        return c10d.distributed_c10d._get_default_group()

    def _test_ucc_backend(
        self, devices, device_ids, multi_device=False, gradient_as_bucket_view=False
    ):
        process_group = self._get_process_group()
        self._test_ddp_with_process_group(
            process_group, devices, device_ids, multi_device, gradient_as_bucket_view
        )

    @requires_ucc()
    def test_ucc_backend_cpu_module(self):
        self._test_ucc_backend([torch.device("cpu")], None)

    @requires_ucc()
    def test_ucc_backend_cpu_module_grad_is_view(self):
        self._test_ucc_backend(
            [torch.device("cpu")], None, gradient_as_bucket_view=True
        )

    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_ucc_backend_1gpu_module_device_ids_integer_list(self):
        int_devices = gpus_for_rank(self.world_size)[self.rank][:1]
        devices = [torch.device("cuda:" + str(i)) for i in int_devices]
        self._test_ucc_backend(devices, int_devices)

    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_ucc_backend_1gpu_module_device_ids_torch_device_list(self):
        int_devices = gpus_for_rank(self.world_size)[self.rank][:1]
        devices = [torch.device("cuda:" + str(i)) for i in int_devices]
        self._test_ucc_backend(devices, devices)

    # TODO: test_ucc_backend_2gpu_module and test_ucc_backend_4gpu_module
    # require broadcast_coalesced which is not supported by ucc currently
    @skip_but_pass_in_sandcastle(
        "requires broadcast coalesced, which is not supported by ucc currently"
    )
    @requires_ucc()
    @skip_if_lt_x_gpu(4)
    def test_ucc_backend_2gpu_module(self):
        int_devices = gpus_for_rank(self.world_size)[self.rank][:2]
        devices = [torch.device("cuda:" + str(i)) for i in int_devices]
        self._test_ucc_backend(devices, None, multi_device=True)

    @skip_but_pass_in_sandcastle(
        "requires broadcast coalesced, which is not supported by ucc currently"
    )
    @requires_ucc()
    @skip_if_lt_x_gpu(8)
    def test_ucc_backend_4gpu_module(self):
        int_devices = gpus_for_rank(self.world_size)[self.rank][:4]
        devices = [torch.device("cuda:" + str(i)) for i in int_devices]
        self._test_ucc_backend(devices, None, multi_device=True)

    def _test_global_local_unused_params_grad(
        self, gradient_as_bucket_view=False, static_graph=False
    ):
        """
        By simulating a multi-task training, this test is to make sure:
        1) DDP does not touch the grad of globally unused parameters.
        2) DDP does update the grad of locally unused parameters.
        """

        class GlobalLocalUnusedParamModule(nn.Module):
            def __init__(self) -> None:
                super().__init__()
                self.t0 = Task()
                self.t1 = Task()
                self.task_unused = Task()

            def task_parameters(self):
                return (self.t0.p, self.t1.p, self.task_unused.p)

            def forward(self, x, rank):
                return self.t0(x) if rank == 0 else self.t1(x)

        def run_and_verify_grad(model):
            # Run forward
            output = model(8, self.rank)

            # The grads of all parameters should be None at this point.
            t0_p, t1_p, task_unused_p = model.module.task_parameters()
            self.assertIsNone(t0_p.grad)
            self.assertIsNone(t1_p.grad)
            self.assertIsNone(task_unused_p.grad)

            # Run backward
            output.mean().backward()

            # Now locally unused parameter should have grad updated on all ranks.
            # However the globally unused parameter should still have None grad.
            self.assertIsNotNone(t0_p.grad)
            self.assertIsNotNone(t1_p.grad)
            self.assertIsNone(task_unused_p.grad)

        process_group = self._get_process_group()

        # Test on CPU
        cpu_model = DistributedDataParallel(
            GlobalLocalUnusedParamModule().cpu(),
            process_group=process_group,
            find_unused_parameters=True,
            gradient_as_bucket_view=gradient_as_bucket_view,
            static_graph=static_graph,
        )
        run_and_verify_grad(cpu_model)

        # Test on GPU
        device_id = gpus_for_rank(self.world_size)[self.rank][0]
        gpu_model = DistributedDataParallel(
            GlobalLocalUnusedParamModule().to(device_id),
            device_ids=[device_id],
            process_group=process_group,
            find_unused_parameters=True,
            gradient_as_bucket_view=gradient_as_bucket_view,
            static_graph=static_graph,
        )
        run_and_verify_grad(gpu_model)

    # TODO: times out
    @skip_but_pass_in_sandcastle("times out")
    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_global_local_unused_params_grad(self):
        self._test_global_local_unused_params_grad()

    # TODO: times out
    @skip_but_pass_in_sandcastle("times out")
    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_global_local_unused_params_grad_with_grad_is_view(self):
        self._test_global_local_unused_params_grad(gradient_as_bucket_view=True)

    # TODO: times out
    @skip_but_pass_in_sandcastle("times out")
    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_global_local_unused_params_grad_with_static_graph(self):
        self._test_global_local_unused_params_grad(static_graph=True)

    # TODO: times out
    @skip_but_pass_in_sandcastle("times out")
    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_find_unused_parameters_when_unused_parameters_empty(self):
        """
        An empty unused_parameters array does not imply find_unused_parameters =
        false. This test makes sure that DDP allreduces unused parameters
        accordingly where the forward pass in some process uses all parameters.
        This unit test creates a module that uses all parameters in rank = 0, and
        has unused parameters in other ranks.
        """

        class FindUnusedParamModule(nn.Module):
            def __init__(self) -> None:
                super().__init__()
                self.t0 = Task()
                self.t1 = Task()

            def task_parameters(self):
                return (self.t0.p, self.t1.p)

            def forward(self, x, rank):
                return self.t1(self.t0(x)) if rank == 0 else self.t1(x)

        def run_and_verify_grad(model):
            # Run forward
            output = model(8, self.rank)

            # The grads of all parameters should be None at this point.
            [self.assertIsNone(t_p.grad) for t_p in model.module.task_parameters()]

            # Run backward
            output.mean().backward()

            # Now locally unused parameter should have grad updated on all ranks.
            [self.assertIsNotNone(t_p.grad) for t_p in model.module.task_parameters()]

        process_group = self._get_process_group()

        # Test on CPU
        cpu_model = DistributedDataParallel(
            FindUnusedParamModule().cpu(),
            process_group=process_group,
            find_unused_parameters=True,
        )
        run_and_verify_grad(cpu_model)

        # Test on GPU
        device_id = gpus_for_rank(self.world_size)[self.rank][0]
        gpu_model = DistributedDataParallel(
            FindUnusedParamModule().to(device_id),
            device_ids=[device_id],
            process_group=process_group,
            find_unused_parameters=True,
        )
        run_and_verify_grad(gpu_model)

    @requires_ucc()
    def test_ignored_output(self):
        """
        Test that the output of a model can be ignored and that there is no
        implicit requirement that `backward` gets called.
        """
        process_group = self._get_process_group()

        class IgnoredOutput(nn.Module):
            def __init__(self) -> None:
                super().__init__()
                self.fc1 = nn.Linear(2, 10, bias=False)
                self.fc2 = nn.Linear(10, 4, bias=False)
                self.relu = nn.ReLU()

            def forward(self, x):
                x = self.relu(self.fc1(x))
                x = self.relu(self.fc2(x))
                return F.softmax(x, dim=1)

        model = DistributedDataParallel(
            IgnoredOutput().float(),
            process_group=process_group,
        )

        batch_size = 4
        criterion = nn.CrossEntropyLoss()
        input = torch.rand([batch_size, 2], dtype=torch.float)
        target = torch.LongTensor([random.randrange(4) for _ in range(batch_size)])

        # Run a few iterations where we ignore the output.
        for _ in range(4):
            output = model(input)
            del output

        # Run a few iterations where we use the output.
        for _ in range(4):
            output = model(input)
            loss = criterion(output, target)
            loss.backward()

    @requires_ucc()
    def test_ignored_output_with_unused_parameters(self):
        """
        Test that the output of a model can be ignored and that there is no
        implicit requirement that `backward` gets called, if not all model
        parameters participated in computing the model output.
        """
        process_group = self._get_process_group()

        class IgnoredOutputWithUnusedParameters(nn.Module):
            def __init__(self) -> None:
                super().__init__()
                self.fc1 = nn.Linear(2, 10, bias=False)
                self.fc2 = nn.Linear(10, 4, bias=False)
                self.fc3 = nn.Linear(4, 4, bias=False)
                self.relu = nn.ReLU()

            def forward(self, x):
                x = self.relu(self.fc1(x))
                x = self.relu(self.fc2(x))
                return F.softmax(x, dim=1)

        model = DistributedDataParallel(
            IgnoredOutputWithUnusedParameters().float(),
            process_group=process_group,
            find_unused_parameters=True,
        )

        batch_size = 4
        criterion = nn.CrossEntropyLoss()
        input = torch.rand([batch_size, 2], dtype=torch.float)
        target = torch.LongTensor([random.randrange(4) for _ in range(batch_size)])

        # Run a few iterations where we ignore the output.
        for _ in range(4):
            output = model(input)
            del output

        # Run a few iterations where we use the output.
        for _ in range(4):
            output = model(input)
            loss = criterion(output, target)
            loss.backward()

    def _run_and_verify_sparse_gradients(self, vanilla_model, ddp_model):
        mult = 2
        batch_size = mult * self.world_size
        criterion = nn.CrossEntropyLoss()
        input = torch.randint(0, 10, [batch_size, 2])
        target = torch.randint(0, 10, [batch_size])

        # Run with entire batch against single process version
        criterion(vanilla_model(input), target).backward()

        # Run with partial batch against multi process version
        partial_input = input.split(mult)[self.rank]
        partial_target = target.split(mult)[self.rank]
        criterion(ddp_model(partial_input), partial_target).backward()

        # Check that the gradients are sparse and identical
        vanilla_parameter = next(vanilla_model.parameters())
        ddp_parameter = next(ddp_model.parameters())
        self.assertEqual(
            vanilla_parameter.grad.coalesce(), ddp_parameter.grad.coalesce()
        )

    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_save_load_checkpoint(self):
        dist.init_process_group(
            "ucc",
            init_method=f"file://{self.file_name}",
            world_size=self.world_size,
            rank=self.rank,
        )

        class TestModel(nn.Module):
            def __init__(self) -> None:
                super().__init__()
                self.fc1 = nn.Linear(2, 10, bias=False)
                self.fc2 = nn.Linear(10, 4, bias=False)
                self.relu = nn.ReLU()

            def forward(self, x):
                x = self.relu(self.fc1(x))
                x = self.relu(self.fc2(x))
                return F.softmax(x, dim=1)

        def train_loop(model, optimizer, iterations):
            for _ in range(iterations):
                optimizer.zero_grad()
                output = model(input)
                loss = criterion(output, target)
                loss.backward()
                optimizer.step()

        device_id = gpus_for_rank(self.world_size)[self.rank][0]

        model_withload = TestModel().float().to(device_id)
        model_withoutload = TestModel().float().to(device_id)

        ddp_withload = DistributedDataParallel(
            model_withload,
            device_ids=[device_id],
        )
        ddp_withoutload = DistributedDataParallel(
            model_withoutload,
            device_ids=[device_id],
        )

        # ensure that all the three models start with the same set of parameters. By default they are randomized on construction
        for p in ddp_withload.parameters():
            with torch.no_grad():
                p.zero_()
        for p in model_withload.parameters():
            with torch.no_grad():
                p.zero_()
        for p in ddp_withoutload.parameters():
            with torch.no_grad():
                p.zero_()

        batch_size = 4
        criterion = nn.CrossEntropyLoss()

        optimizer_withload = torch.optim.SGD(ddp_withload.parameters(), lr=0.001)
        optimizer_non_ddp_withload = torch.optim.SGD(
            model_withload.parameters(), lr=0.001
        )
        optimizer_withoutload = torch.optim.SGD(ddp_withoutload.parameters(), lr=0.001)

        input = torch.rand([batch_size, 2], dtype=torch.float).to(device_id)
        target = torch.LongTensor([random.randrange(4) for _ in range(batch_size)]).to(
            device_id
        )

        # run the model for 6 iterations, with a checkpoint in the middle
        train_loop(ddp_withload, optimizer_withload, 3)

        # zero out parameters of both DDP and non-DDP models and reload them from the DDP state dict
        checkpoint_path = tempfile.gettempdir() + "/model.checkpoint"
        if self.rank == 0:
            torch.save(ddp_withload.state_dict(), checkpoint_path)

        dist.barrier()
        map_location = {"cuda:%d" % 0: "cuda:%d" % self.rank}
        ddp_state_dict = torch.load(checkpoint_path, map_location=map_location)

        for model in [ddp_withload, model_withload]:
            for p in ddp_withload.parameters():
                with torch.no_grad():
                    p.zero_()
        ddp_withload.load_state_dict(ddp_state_dict)
        # the non-DDP model needs to first remove the prefix of "module." from the DDP state dict
        torch.nn.modules.utils.consume_prefix_in_state_dict_if_present(
            ddp_state_dict, "module."
        )
        model_withload.load_state_dict(ddp_state_dict)

        train_loop(ddp_withload, optimizer_withload, 3)
        train_loop(model_withload, optimizer_non_ddp_withload, 3)

        # re-run the model with the same inputs for 6 iterations with no checkpoint
        train_loop(ddp_withoutload, optimizer_withoutload, 6)

        for p_withload, p_withoutload, p_non_ddp_withload in zip(
            ddp_withload.parameters(),
            ddp_withoutload.parameters(),
            model_withload.parameters(),
        ):
            self.assertEqual(p_withload, p_withoutload)
            self.assertEqual(p_non_ddp_withload, p_withoutload)

    def _test_sparse_gradients(self, gradient_as_bucket_view=False):
        process_group = self._get_process_group()

        # Ensure initialized weights and inputs are identical across processes
        torch.manual_seed(1337)

        vanilla_model = SparseGradientModule()
        ddp_model = DistributedDataParallel(
            copy.deepcopy(vanilla_model),
            process_group=process_group,
            gradient_as_bucket_view=gradient_as_bucket_view,
        )

        self._run_and_verify_sparse_gradients(vanilla_model, ddp_model)

    # TODO: backward pass: input tensor has to be dense
    @skip_but_pass_in_sandcastle("backward pass: input tensor has to be dense")
    @requires_ucc()
    def test_sparse_gradients(self):
        self._test_sparse_gradients()

    # TODO: backward pass: input tensor has to be dense
    @skip_but_pass_in_sandcastle("backward pass: input tensor has to be dense")
    @requires_ucc()
    def test_sparse_gradients_grad_is_view(self):
        self._test_sparse_gradients(gradient_as_bucket_view=True)

    @requires_ucc()
    def test_ddp_comm_hook_future_passing_cpu(self):
        """
        This unit test verifies whether the Future object is passed properly.
        The callback function creates a Future object and sets a value to it.
        """
        process_group = self._get_process_group()

        # Test on CPU
        cpu_model = DistributedDataParallel(
            ModuleForDdpCommHook().cpu(), process_group=process_group
        )

        # Register DDP Communication Hook
        cpu_model.register_comm_hook(None, self._simple_hook)

        # check whether the grads are equal to what then callback returns.
        # without the comm_hook, result would be 0.25 * torch.ones(2, 2).
        self._run_and_verify_hook(cpu_model, 8, 2 * torch.ones(2, 2))

    def _gpu_model_with_ddp_comm_hook(
        self, process_group, hook=None, gradient_as_bucket_view=False, state=None
    ):
        device_id = gpus_for_rank(self.world_size)[self.rank][0]
        gpu_model = DistributedDataParallel(
            ModuleForDdpCommHook().to(device_id),
            device_ids=[device_id],
            process_group=process_group,
            gradient_as_bucket_view=gradient_as_bucket_view,
        )

        # Register a DDP communication hook if any.
        if hook is not None:
            gpu_model.register_comm_hook(state, hook)

        return gpu_model

    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_ddp_comm_hook_future_passing_gpu_ucc(self):
        """
        This unit test verifies whether the Future object is passed properly using ucc backend.
        The hook callback function creates a Future object and sets a value to it.
        """
        process_group = self._get_process_group()

        # Get GPU model with simple_hook registered.
        gpu_model = self._gpu_model_with_ddp_comm_hook(process_group, self._simple_hook)

        # check whether the grads are equal to what simple_hook's then callback returns.
        # without the comm_hook, result would be 0.25 * torch.ones(2, 2).
        self._run_and_verify_hook(gpu_model, 8, 2 * torch.ones(2, 2))

    @requires_ucc()
    def test_ddp_invalid_comm_hook_init(self):
        """
        This unit test makes sure that register_comm_hook properly checks the format
        of hook defined by user. The Python hook must be callable. This test also
        checks whether bucket annotation checked properly if defined.
        """
        process_group = self._get_process_group()

        model = DistributedDataParallel(
            ModuleForDdpCommHook(), process_group=process_group
        )

        with self.assertRaisesRegex(TypeError, "Communication hook must be callable."):
            model.register_comm_hook(state=None, hook=1)

        with self.assertRaisesRegex(
            ValueError, "bucket annotation should be dist.GradBucket."
        ):

            def comm_hook(
                state: object, bucket: int
            ) -> torch.futures.Future[torch.Tensor]:
                return torch.futures.Future()

            model.register_comm_hook(state=None, hook=comm_hook)

    @requires_ucc()
    def test_ddp_invalid_comm_hook_return_type(self):
        """
        This test checks whether return annotation checked properly if defined. It also
        checks whether an internal error is thrown if return type is incorrect and user
        hasn't specified any return type annotation.
        """
        process_group = self._get_process_group()

        model = DistributedDataParallel(
            ModuleForDdpCommHook(), process_group=process_group
        )

        expected_err = (
            "Communication hook: return annotation should be torch.futures.Future"
        )
        with self.assertRaisesRegex(
            ValueError,
            expected_err,
        ):

            def comm_hook(state: object, bucket: dist.GradBucket) -> int:
                return torch.futures.Future()

            model.register_comm_hook(state=None, hook=comm_hook)

        verify_ddp_error_logged(model, expected_err)

        with self.assertRaisesRegex(
            RuntimeError,
            "callback must return a torch.futures.Future object, but got",
        ):

            def comm_hook(state: object, bucket: dist.GradBucket):
                return 1

            model.register_comm_hook(state=None, hook=comm_hook)

            # Run forward
            output = model(8, self.rank)

            # Run backward
            output.mean().backward()

    @requires_ucc()
    def test_ddp_comm_hook_register_just_once(self):
        """
        DDP communication hook can only be registered once. This test validates whether
        the error is thrown properly when register_comm_hook is called more than once.
        """
        process_group = self._get_process_group()

        model = DistributedDataParallel(
            ModuleForDdpCommHook(), process_group=process_group
        )

        def dummy_hook(state, bucket):
            fut = torch.futures.Future()
            fut.set_result([bucket.buffer()])
            return fut

        model.register_comm_hook(None, dummy_hook)

        with self.assertRaisesRegex(
            RuntimeError,
            "register_comm_hook or register_builtin_comm_hook can only be called once.",
        ):
            model.register_comm_hook(None, dummy_hook)

    # TODO: backward pass: input tensor must be dense
    @skip_but_pass_in_sandcastle("backward pass: input tensor has to be dense")
    @requires_ucc()
    def test_ddp_comm_hook_sparse_gradients(self):
        """
        Runs "test_sparse_gradients" unit test with DDP communication hook. We define a
        simple hook that does allreduce and works with ucc backend for this test.
        """
        process_group = self._get_process_group()

        # Ensure initialized weights and inputs are identical across processes
        torch.manual_seed(1337)

        vanilla_model = SparseGradientModule()
        ddp_model = DistributedDataParallel(
            copy.deepcopy(vanilla_model),
            process_group=process_group,
        )

        def allreduce_hook_ucc(
            state: object, bucket: dist.GradBucket
        ) -> torch.futures.Future[torch.Tensor]:
            def div_by_world_size(fut):
                # Divide the result by 2 * world_size.
                return fut.wait()[0] / self.world_size

            # Prepare allreduced grad bucket tensors by running an async work.
            fut = process_group.allreduce([bucket.buffer()]).get_future()
            return fut.then(div_by_world_size)

        ddp_model.register_comm_hook(None, allreduce_hook_ucc)

        self._run_and_verify_sparse_gradients(vanilla_model, ddp_model)


class CommTest(test_c10d_common.AbstractCommTest, MultiProcessTestCase):
    @property
    def device(self):
        return "cpu"

    def setUp(self):
        super().setUp()
        self._spawn_processes()

    def tearDown(self):
        super().tearDown()
        try:
            os.remove(self.file_name)
        except OSError:
            pass

    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_sequence_num_set_default_pg_ucc(self):
        self._test_sequence_num_set_default_pg(backend="ucc")

    @requires_ucc()
    @skip_if_lt_x_gpu(2)
    def test_sequence_num_set_ucc_new_group(self):
        self._test_sequence_num_set_new_group(backend="ucc")

    @skip_if_lt_x_gpu(2)
    @requires_ucc()
    def test_sequence_num_incremented_ucc_default(self):
        self._test_sequence_num_incremented_default_group("ucc")

    @skip_if_lt_x_gpu(4)
    @requires_ucc()
    def test_sequence_num_incremented_ucc_subgroup(self):
        if self.world_size < 4:
            return skip_but_pass_in_sandcastle("Test requires world_size of at least 4")
        self._test_sequence_num_incremented_subgroup("ucc")

    @skip_but_pass_in_sandcastle("Fails on M60")
    @requires_ucc()
    def test_ucc_barrier_device_ids(self):
        store = c10d.FileStore(self.file_name, self.world_size)
        c10d.init_process_group(
            backend="ucc", rank=self.rank, world_size=self.world_size, store=store
        )

        with self.assertRaisesRegex(RuntimeError, "device_ids not supported"):
            c10d.barrier(device_ids=[self.rank])

    @skip_but_pass_in_sandcastle("Fails on M60")
    @skip_if_lt_x_gpu(2)
    @requires_ucc()
    def test_ucc_warn_not_in_group(self):
        self._test_warn_not_in_group(backend="ucc")

    @skip_if_lt_x_gpu(2)
    @requires_ucc()
    def test_ucc_rank_membership(self):
        self._test_rank_membership(backend="ucc")

    @skip_if_lt_x_gpu(2)
    @requires_ucc()
    def test_tensor_dtype_mismatch(self):
        self._test_tensor_dtype_mismatch(backend="ucc")

    @skip_if_lt_x_gpu(2)
    @requires_ucc()
    def test_tensor_dtype_complex(self):
        self._test_tensor_dtype_complex(backend="ucc")


class UccProcessGroupWithDispatchedCollectivesTests(
    test_c10d_common.ProcessGroupWithDispatchedCollectivesTests
):
    @skip_but_pass_in_sandcastle("Fails on M60")
    @requires_ucc()
    @skip_if_lt_x_gpu(1)
    def test_collectives(self):
        # includes reduce, broadcast, all_reduce, all_gather, reduce_scatter, barrier, all_to_all, scatter
        self._test_collectives(backend="ucc")

    @skip_but_pass_in_sandcastle("Fails on M60")
    @requires_ucc()
    @skip_if_lt_x_gpu(1)
    def test_allgather_base(self):
        store = dist.FileStore(self.file_name, self.world_size)
        dist.init_process_group(
            "ucc",
            world_size=self.world_size,
            rank=self.rank,
            store=store,
        )
        device = "cuda"
        tensor = torch.ones(10, 10, device=torch.device(device))
        output_tensor = torch.zeros(10, 10, device=torch.device(device))
        dist.all_gather_into_tensor(output_tensor, tensor)
        self.assertEqual(output_tensor, tensor)


if __name__ == "__main__":
    assert (
        not torch.cuda._initialized
    ), "test_distributed must not have initialized CUDA context on main process"

    run_tests()