test_torch.py

# mypy: allow-untyped-decorators
# mypy: allow-untyped-defs
# Owner(s): ["module: tests"]

import torch
import torch.utils.data
import numpy as np

import contextlib
import gc
import io
import inspect
import itertools
import math
import random
import re
import copy
import tempfile
import unittest
import warnings
import types
import pickle
import textwrap
import subprocess
import weakref
import sys
import copyreg
from torch import inf, nan
from itertools import product, combinations, permutations, chain
from functools import partial
from torch import multiprocessing as mp
from torch.testing import make_tensor
from torch.testing._internal.common_optimizers import (
    optim_db, optims, _get_optim_inputs_including_global_cliquey_kwargs)

from torch.testing._internal.common_utils import (  # type: ignore[attr-defined]
    MI200_ARCH, TEST_WITH_TORCHINDUCTOR, TEST_WITH_ROCM, run_tests, IS_JETSON,
    IS_FILESYSTEM_UTF8_ENCODING,
    IS_SANDCASTLE, IS_FBCODE, IS_REMOTE_GPU, skipIfRocmArch, skipIfTorchInductor, load_tests, slowTest, slowTestIf,
    skipIfCrossRef, TEST_WITH_CROSSREF, skipIfTorchDynamo, set_default_dtype,
    skipCUDAMemoryLeakCheckIf, BytesIOContext,
    skipIfRocm, skipIfNoSciPy, TemporaryFileName, TemporaryDirectoryName,
    wrapDeterministicFlagAPITest, DeterministicGuard, CudaSyncGuard,
    bytes_to_scalar, parametrize, noncontiguous_like,
    AlwaysWarnTypedStorageRemoval, TEST_WITH_TORCHDYNAMO, xfailIfTorchDynamo,
    xfailIfS390X, set_warn_always_context, decorateIf, isRocmArchAnyOf,
    IS_MACOS,
)
from multiprocessing.reduction import ForkingPickler
from torch.testing._internal.common_device_type import (
    expectedFailureMeta,
    expectedFailureXLA,
    instantiate_device_type_tests,
    onlyCUDA, onlyCPU,
    dtypes, dtypesIfCUDA, dtypesIfCPU, deviceCountAtLeast,
    skipMeta, PYTORCH_CUDA_MEMCHECK, largeTensorTest, onlyNativeDeviceTypes, skipCUDAIfNotRocm,
    get_all_device_types, skipXLA)
import torch.backends.quantized
import torch.testing._internal.data
from torch.testing._internal.common_cuda import (
    tf32_on_and_off, TEST_CUDNN, TEST_MULTIGPU,
    _create_scaling_case, _create_scaling_models_optimizers)
from torch.testing._internal.common_mkldnn import reduced_f32_on_and_off
from torch.testing._internal.common_dtype import (
    floating_types_and, get_all_math_dtypes, all_types_and_complex_and, complex_types,
    all_types_and, floating_types, floating_and_complex_types, integral_types_and,
    get_all_qint_dtypes, all_types_complex_float8_and, all_passthru_types_and,
)
from torch.testing._internal.two_tensor import TwoTensor
from torch.testing._internal.common_utils import IS_WINDOWS

if TEST_WITH_TORCHINDUCTOR:
    from torch._inductor.test_case import TestCase
else:
    from torch.testing._internal.common_utils import TestCase  # type: ignore[assignment]


# Protects against includes accidentally setting the default dtype
if torch.get_default_dtype() is not torch.float32:
    raise AssertionError(f"default dtype should be float32, got {torch.get_default_dtype()}")

# load_tests from torch.testing._internal.common_utils is used to automatically filter tests for
# sharding on sandcastle. This line silences flake warnings
load_tests = load_tests  # noqa: PLW0127

AMPERE_OR_ROCM = TEST_WITH_ROCM or torch.cuda.is_tf32_supported()


is_cuda_sm86 = torch.cuda.is_available() and torch.cuda.get_device_capability(0) == (8, 6)

class TestTorchDeviceType(TestCase):
    exact_dtype = True

    # TODO: move all tensor creation to common ops
    def _rand_shape(self, dim, min_size, max_size):
        shape = []
        for _ in range(dim):
            shape.append(random.randint(min_size, max_size))
        return tuple(shape)

    # Validates that mathematical constants are defined properly, as required by
    # the Python Array API (https://data-apis.org/array-api/latest/API_specification/constants.html)
    @onlyCPU
    def test_constants(self, device):
        self.assertIsInstance(torch.e, float)
        self.assertEqual(torch.e, math.e, atol=0, rtol=0)

        self.assertIsInstance(torch.pi, float)
        self.assertEqual(torch.pi, math.pi, atol=0, rtol=0)

        self.assertIsInstance(torch.nan, float)
        self.assertEqual(torch.nan, math.nan, equal_nan=True)

        self.assertIsInstance(torch.inf, float)
        self.assertEqual(torch.inf, math.inf)

    @onlyNativeDeviceTypes
    @slowTestIf(IS_WINDOWS)
    @dtypes(torch.int8, torch.uint8, torch.int16, torch.int32, torch.int64,
            torch.bool, torch.float32, torch.complex64, torch.float64,
            torch.complex128, torch.uint16, torch.uint32, torch.uint64)
    def test_bytes_to_scalar(self, device, dtype):
        def rand_byte():
            if dtype == torch.bool:
                return torch.randint(0, 2, ()).item()
            else:
                return torch.randint(0, 256, ()).item()

        element_size = torch._utils._element_size(dtype)

        for _ in range(10):
            bytes_list = [rand_byte() for _ in range(element_size)]
            scalar = bytes_to_scalar(bytes_list, dtype, device)
            self.assertEqual(scalar.storage().untyped().tolist(), bytes_list)

    # For testing in64 support in upsample_nearest3d
    @skipIfRocmArch(MI200_ARCH)
    @onlyCUDA
    @largeTensorTest('56GB', device='cuda')
    @dtypes(torch.bfloat16)
    @unittest.skipIf(IS_JETSON, "Large tensor tests are too large for Jetson.")
    @decorateIf(unittest.expectedFailure, lambda params: isRocmArchAnyOf(MI200_ARCH))
    def test_int64_upsample3d(self, device, dtype):
        x = torch.ones((1, 256, 16, 720, 1280), dtype=dtype, device=device)
        try:
            torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest')
        except Exception as e:
            self.fail(f"Unexpected exception raised: {e}")

    @dtypes(torch.int8, torch.uint8, torch.int16, torch.int32, torch.int64,
            torch.bool, torch.float32, torch.complex64, torch.float64,
            torch.complex128, torch.uint16, torch.uint32, torch.uint64)
    @slowTestIf(IS_WINDOWS)
    def test_storage(self, device, dtype):
        v = make_tensor((3, 5), dtype=dtype, device=device, low=-9, high=9)
        self.assertEqual(v.storage()[0], v[0][0])
        self.assertEqual(v.storage()[14], v[2][4])
        v_s = v.storage()

        for el_num in range(v.numel()):
            dim0 = el_num // v.size(1)
            dim1 = el_num % v.size(1)
            self.assertEqual(
                v_s[el_num],
                v[dim0][dim1])

        v_s_byte = v.storage().untyped()
        el_size = v.element_size()

        for el_num in range(v.numel()):
            start = el_num * el_size
            end = start + el_size
            dim0 = el_num // v.size(1)
            dim1 = el_num % v.size(1)
            self.assertEqual(
                bytes_to_scalar(v_s_byte[start:end], dtype, device),
                v[dim0][dim1])

    @onlyNativeDeviceTypes
    @dtypes(torch.int8, torch.uint8, torch.int16, torch.int32, torch.int64,
            torch.bool, torch.float32, torch.complex64, torch.float64,
            torch.complex128, torch.quint8, torch.qint8, torch.qint32,
            torch.quint4x2)
    @slowTestIf(IS_WINDOWS)
    def test_storage_setitem(self, device, dtype):
        # Skip quantized dtypes for CUDA, since they're not supported
        if torch.device(device).type == 'cuda':
            if dtype in [torch.quint8, torch.qint8, torch.qint32, torch.quint4x2]:
                return

        storage_type_name = torch.storage._dtype_to_storage_type_map()[dtype]
        if torch.device(device).type == 'cuda':
            storage_type = eval('torch.cuda.' + storage_type_name)
        else:
            storage_type = eval('torch.' + storage_type_name)

        N = 10

        s = storage_type(N)
        s[:] = 0
        l = [0] * N
        self.assertEqual(s, storage_type(l))

        for i in range(N):
            s[i] = i
            l[i] = i

        self.assertEqual(s, storage_type(l))

        l[2:7] = [1] * 5
        s[2:7] = 1
        self.assertEqual(s, storage_type(l))

    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    @onlyNativeDeviceTypes
    @slowTestIf(IS_WINDOWS)
    def test_storage_use_count(self, device):
        a = torch.randn(10, device=device)
        prev_cf = torch._C._storage_Use_Count(a.untyped_storage()._cdata)
        # Two references: 'a' and the wrapper returned by untyped_storage()
        self.assertEqual(prev_cf, 2)
        b = a.view(2, 5)
        self.assertEqual(torch._C._storage_Use_Count(b.untyped_storage()._cdata), prev_cf + 1)

    @xfailIfTorchDynamo
    @onlyNativeDeviceTypes
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @slowTestIf(IS_WINDOWS)
    def test_tensor_storage_type(self, device, dtype):
        a = make_tensor((10,), dtype=dtype, device=device, low=-9, high=9)

        module = torch.cuda if (torch.device(device).type == 'cuda') else torch
        expected_storage_type = getattr(module, torch.storage._dtype_to_storage_type_map()[dtype])

        self.assertEqual(a.storage_type(), expected_storage_type)

    @onlyNativeDeviceTypes
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16, torch.uint16, torch.uint32, torch.uint64))
    @slowTestIf(IS_WINDOWS)
    def test_tensor_from_storage(self, device, dtype):
        a = make_tensor((4, 5, 3), dtype=dtype, device=device, low=-9, high=9)
        a_s = a.storage()
        b = torch.tensor(a_s, device=device, dtype=dtype).reshape(a.size())
        self.assertEqual(a, b)
        c = torch.tensor(a_s.untyped(), device=device, dtype=dtype).reshape(a.size())
        self.assertEqual(a, c)

        for error_dtype in all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16):
            if error_dtype == dtype:
                continue
            with self.assertRaisesRegex(RuntimeError, r'Expected a Storage of type'):
                error_storage = a.to(error_dtype).storage()
                torch.tensor(error_storage, device=device, dtype=dtype)

    @onlyNativeDeviceTypes
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @slowTestIf(IS_WINDOWS)
    def test_set_storage(self, device, dtype):
        a = make_tensor((4, 5, 3), dtype=dtype, device=device, low=-9, high=9)
        a_s = a.storage()
        b = torch.tensor([], device=device, dtype=dtype).set_(a_s).reshape(a.size())
        self.assertEqual(a, b)
        c = torch.tensor([], device=device, dtype=dtype).set_(a_s.untyped()).reshape(a.size())
        self.assertEqual(a, c)

        for error_dtype in all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16):
            if error_dtype == dtype:
                continue
            with self.assertRaisesRegex(RuntimeError, r'Expected a Storage of type'):
                error_storage = a.to(error_dtype).storage()
                b = torch.tensor([], device=device, dtype=dtype).set_(error_storage)

    def _check_storage_meta(self, s, s_check):
        self.assertTrue(
            isinstance(s, (torch.UntypedStorage, torch.TypedStorage)) and
            isinstance(s_check, type(s)),
            lambda msg: (
                f"{msg}\ns and s_check must both be one of UntypedStorage or "
                "TypedStorage, but got"
                f" {type(s).__name__} and {type(s_check).__name__}"
            ),
        )

        self.assertEqual(s.device.type, 'meta')
        self.assertEqual(s.nbytes(), s_check.nbytes())
        self.assertEqual(s.size(), s_check.size())
        self.assertEqual(s.data_ptr(), 0)

        with self.assertRaisesRegex(NotImplementedError, r'Not available'):
            s[0]

        if isinstance(s, torch.TypedStorage):
            self.assertEqual(s.dtype, s_check.dtype)
            self._check_storage_meta(s.untyped(), s_check.untyped())

    @onlyNativeDeviceTypes
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @slowTestIf(IS_WINDOWS)
    def test_typed_storage_meta(self, device, dtype):
        args_list = [
            [],
            [0],
            [100],
            [[1, 2, 3, 4, 5, 6]],
        ]
        for args in args_list:
            s_check = torch.TypedStorage(*args, dtype=dtype, device=device)
            s = torch.TypedStorage(*args, dtype=dtype, device='meta')
            self._check_storage_meta(s, s_check)

    @onlyNativeDeviceTypes
    @slowTestIf(IS_WINDOWS)
    def test_untyped_storage_meta(self, device):
        args_list = [
            [],
            [0],
            [100],
            [[1, 2, 3, 4, 5, 6]],
        ]
        for args in args_list:
            s_check = torch.UntypedStorage(*args, device=device)
            s = torch.UntypedStorage(*args, device='meta')
            self._check_storage_meta(s, s_check)

    @onlyNativeDeviceTypes
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @slowTestIf(IS_WINDOWS)
    def test_storage_meta_from_tensor(self, device, dtype):
        t_check = make_tensor((4, 5, 3), dtype=dtype, device=device, low=-9, high=9)
        t = t_check.to('meta')

        s_check = t_check.storage()
        s = t.storage()
        self._check_storage_meta(s, s_check)

    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @slowTestIf(IS_WINDOWS)
    def test_storage_meta_errors(self, device, dtype):
        s0 = torch.TypedStorage([1, 2, 3, 4], device='meta', dtype=dtype)

        with self.assertRaisesRegex(NotImplementedError, r'Cannot copy out'):
            s0.cpu()

        with self.assertRaisesRegex(RuntimeError, r'only available on CPU'):
            s0._share_fd_cpu_()

        with self.assertRaisesRegex(RuntimeError, r'only available on CPU'):
            s0._share_filename_cpu_()

        if torch.cuda.is_available():
            with self.assertRaisesRegex(NotImplementedError, r'Cannot copy out'):
                s0.cuda()

            with self.assertRaisesRegex(RuntimeError, r'only available on CUDA'):
                s0._share_cuda_()

            with self.assertRaisesRegex(TypeError, r"cannot pin 'torch.storage.UntypedStorage' only CPU memory can be pinned"):
                s0.pin_memory()

        with self.assertRaisesRegex(RuntimeError, r'only available on CPU'):
            s0.share_memory_()

        self.assertFalse(s0.is_shared())

        with self.assertRaisesRegex(NotImplementedError, r'Not available'):
            s0.tolist()

        with tempfile.NamedTemporaryFile() as f:
            with self.assertRaisesRegex(NotImplementedError, r'Cannot copy out'):
                s0._write_file(f, True, True, s0.element_size())

        for device in ['cpu', 'cuda'] if torch.cuda.is_available() else ['cpu']:
            s1 = torch.TypedStorage([1, 2, 3, 4], device=device, dtype=dtype)

            with self.assertRaisesRegex(NotImplementedError, r'Cannot copy out'):
                s1.copy_(s0)

    @onlyCPU
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @slowTestIf(IS_WINDOWS)
    def test_storage_meta_ok(self, device, dtype):
        s0 = torch.TypedStorage([1, 2, 3, 4], device='meta', dtype=dtype)

        # This is OK, it changes the meta storage size without allocating
        s0.resize_(10)

    @onlyCUDA
    def test_module_share_memory(self):
        # Test fix for issue #80733
        # See https://github.com/pytorch/pytorch/issues/80733
        model = torch.nn.Linear(3, 1)
        _model_cuda = model.to('cuda')
        model.share_memory()

    @dtypes(torch.float32, torch.complex64)
    @slowTestIf(IS_WINDOWS)
    def test_deepcopy(self, device, dtype):
        from copy import deepcopy
        a = torch.randn(5, 5, dtype=dtype, device=device)
        b = torch.randn(5, 5, dtype=dtype, device=device)
        c = a.view(25)
        q = [a, [a.storage(), b.storage()], b, c]
        w = deepcopy(q)
        self.assertEqual(w[0], q[0], atol=0, rtol=0)
        self.assertEqual(w[1][0], q[1][0], atol=0, rtol=0)
        self.assertEqual(w[1][1], q[1][1], atol=0, rtol=0)
        self.assertEqual(w[1], q[1], atol=0, rtol=0)
        self.assertEqual(w[2], q[2], atol=0, rtol=0)

        # Check that deepcopy preserves sharing
        w[0].add_(1)
        for i in range(a.numel()):
            self.assertEqual(w[1][0][i], q[1][0][i] + 1)
        self.assertEqual(w[3], c + 1)
        w[2].sub_(1)
        for i in range(a.numel()):
            self.assertEqual(w[1][1][i], q[1][1][i] - 1)

        # Check that deepcopy preserves attributes
        a.foo = 3
        self.assertEqual(deepcopy(a).foo, 3)

    @dtypes(torch.float32, torch.complex64)
    @slowTestIf(IS_WINDOWS)
    def test_deepcopy_scalar(self, device, dtype):
        from copy import deepcopy
        a = torch.tensor(5, dtype=dtype, device=device)
        self.assertEqual(a.size(), deepcopy(a).size())
        self.assertEqual(a, deepcopy(a))

    def check_internal_mem_overlap(self, inplace_op, num_inputs,
                                   dtype, device,
                                   expected_failure=False):
        if isinstance(inplace_op, str):
            inplace_op = getattr(torch.Tensor, inplace_op)
        input = torch.randn(1, dtype=dtype, device=device).expand(3, 3)
        inputs = [input] + [torch.randn_like(input)
                            for i in range(num_inputs - 1)]
        if not expected_failure:
            with self.assertRaisesRegex(RuntimeError, 'single memory location'):
                inplace_op(*inputs)
        else:
            with self.assertRaises(AssertionError):
                with self.assertRaisesRegex(RuntimeError, 'single memory location'):
                    inplace_op(*inputs)

    def unary_check_input_output_mem_overlap(self, data, sz, op,
                                             expected_failure=False):

        def _test(op, output, input):
            output_exp = torch.empty_like(output)
            op(input, out=output_exp)
            self.assertEqual(op(input, out=output), output_exp, msg=op.__name__)

        # output is identical to input:
        _test(op, output=data[0:sz], input=data[0:sz])
        # output and input are independent:
        _test(op, output=data[0:sz], input=data[sz:2 * sz])
        # output partially overlaps with input:
        if not expected_failure:
            with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
                _test(op, data[0:sz], data[1:sz + 1])
        else:
            with self.assertRaises(AssertionError):
                with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
                    _test(op, data[0:sz], data[1:sz + 1])
        # output is transpose of input:
        length = int(math.sqrt(sz))
        input = data[:length**2].view([length, length])
        out = input.t()
        if not expected_failure:
            with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
                _test(op, out, input)
        else:
            with self.assertRaises(AssertionError):
                with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
                    _test(op, out, input)

    def ternary_check_input_output_mem_overlap(self, op, device,
                                               expected_failure=False):
        sz = 9
        data = torch.randn(2 * sz, device=device)
        other1 = torch.randn(sz, device=device)
        other2 = torch.randn(sz, device=device)

        self.unary_check_input_output_mem_overlap(
            data, sz, lambda input, out:
                op(input, other1.view(input.shape), other2.view(input.shape), out=out),
            expected_failure=expected_failure)

        self.unary_check_input_output_mem_overlap(
            data, sz, lambda input, out:
                op(other1.view(input.shape), input, other2.view(input.shape), out=out),
            expected_failure=expected_failure)

        self.unary_check_input_output_mem_overlap(
            data, sz, lambda input, out:
                op(other1.view(input.shape), other2.view(input.shape), input, out=out),
            expected_failure=expected_failure)

    def _select_broadcastable_dims(self, dims_full=None):
        # select full dimensionality
        if dims_full is None:
            dims_full = []
            ndims = random.randint(1, 4)
            dims_full = [random.randint(1, 8) for _ in range(ndims)]
        else:
            ndims = len(dims_full)

        # select actual dimensions for ops:
        # larger: full ndims, individual sizes may be reduced
        # smaller: possibly reduced ndims, sizes may be reduced
        smaller_ndims = random.randint(1, ndims)
        dims_small = []
        dims_large = []
        for i in range(ndims - 1, -1, -1):
            j = random.randint(1, 3)
            if j == 1:  # no reduced singleton dimension
                ds = dims_full[i]
                dl = dims_full[i]
            elif j == 2:  # larger may have reduced singleton dimension
                ds = dims_full[i]
                dl = 1 if len(dims_small) < smaller_ndims else dims_full[i]
            elif j == 3:  # smaller may have reduced singleton dimension
                ds = 1
                dl = dims_full[i]
            dims_large = [dl] + dims_large
            if len(dims_small) < smaller_ndims:
                dims_small = [ds] + dims_small
        return (dims_small, dims_large, dims_full)

    # collected tests of ops that used scalar_check in Declarations.cwrap for
    # correctness
    def test_scalar_check(self, device):
        zero_d = torch.randn((), device=device)
        one_d = torch.randn((1,), device=device)

        # remainder
        self.assertEqual((), torch.remainder(zero_d, zero_d).shape)
        self.assertEqual((), torch.remainder(zero_d, 2).shape)
        self.assertEqual((1,), torch.remainder(zero_d, one_d).shape)
        self.assertEqual((1,), torch.remainder(one_d, zero_d).shape)

        # fmod
        self.assertEqual((), torch.fmod(zero_d, zero_d).shape)
        self.assertEqual((), torch.fmod(zero_d, 2).shape)
        self.assertEqual((1,), torch.fmod(zero_d, one_d).shape)
        self.assertEqual((1,), torch.fmod(one_d, zero_d).shape)

        # exp, cos, cosh, tan, atan, tanh, erf, erfc, reciprocal
        self.assertEqual((), torch.exp(zero_d).shape)
        self.assertEqual((), torch.cos(zero_d).shape)
        self.assertEqual((), torch.cosh(zero_d).shape)
        self.assertEqual((), torch.tan(zero_d).shape)
        self.assertEqual((), torch.atan(zero_d).shape)
        self.assertEqual((), torch.acosh(zero_d).shape)
        self.assertEqual((), torch.asinh(zero_d).shape)
        self.assertEqual((), torch.atanh(zero_d).shape)
        self.assertEqual((), torch.tanh(zero_d).shape)
        self.assertEqual((), torch.erf(zero_d).shape)
        self.assertEqual((), torch.erfc(zero_d).shape)
        self.assertEqual((), torch.reciprocal(zero_d).shape)
        self.assertEqual((1,), torch.exp(one_d).shape)
        self.assertEqual((1,), torch.cos(one_d).shape)
        self.assertEqual((1,), torch.cosh(one_d).shape)
        self.assertEqual((1,), torch.tan(one_d).shape)
        self.assertEqual((1,), torch.atan(one_d).shape)
        self.assertEqual((1,), torch.acosh(one_d).shape)
        self.assertEqual((1,), torch.asinh(one_d).shape)
        self.assertEqual((1,), torch.atanh(one_d).shape)
        self.assertEqual((1,), torch.tanh(one_d).shape)
        self.assertEqual((1,), torch.erf(one_d).shape)
        self.assertEqual((1,), torch.erfc(one_d).shape)
        self.assertEqual((1,), torch.reciprocal(one_d).shape)

        # clamp
        self.assertEqual((), torch.clamp(zero_d, min=0, max=1).shape)
        self.assertEqual((), torch.clamp(zero_d, min=0).shape)
        self.assertEqual((), torch.clamp(zero_d, max=1).shape)
        self.assertEqual((1,), torch.clamp(one_d, min=0, max=1).shape)
        self.assertEqual((1,), torch.clamp(one_d, min=0).shape)
        self.assertEqual((1,), torch.clamp(one_d, max=1).shape)

        # cumsum, cumprod, cummax, cummin
        self.assertEqual((), torch.logcumsumexp(zero_d, 0).shape)
        self.assertEqual((), torch.cumsum(zero_d, 0).shape)
        self.assertEqual((), torch.cumprod(zero_d, 0).shape)
        self.assertEqual((), torch.cummax(zero_d, 0)[0].shape)
        self.assertEqual((), torch.cummin(zero_d, 0)[0].shape)

        # sort, topk
        self.assertEqual([(), ()], [x.shape for x in torch.sort(zero_d, 0, False)])
        self.assertEqual([(), ()], [x.shape for x in torch.sort(zero_d, 0, True)])
        self.assertEqual([(), ()], [x.shape for x in torch.topk(zero_d, 1, 0, False)])
        self.assertEqual([(), ()], [x.shape for x in torch.topk(zero_d, 1, 0, True)])

        # max, min
        self.assertEqual((), torch.max(zero_d, zero_d).shape)
        self.assertEqual((1,), torch.max(one_d, zero_d).shape)
        self.assertEqual((1,), torch.max(zero_d, one_d).shape)
        self.assertEqual((), torch.min(zero_d, zero_d).shape)
        self.assertEqual((1,), torch.min(one_d, zero_d).shape)
        self.assertEqual((1,), torch.min(zero_d, one_d).shape)

        zero_d_int = torch.tensor(1, device=device)
        one_d_int = torch.tensor([1], device=device)

        # lshift, rshift
        self.assertEqual((), (zero_d_int >> zero_d_int).shape)
        self.assertEqual((), (zero_d_int >> 1).shape)
        self.assertEqual((1,), (one_d_int >> zero_d_int).shape)
        self.assertEqual((1,), (zero_d_int >> one_d_int).shape)
        self.assertEqual((1,), (one_d_int >> 1).shape)

        self.assertEqual((), (zero_d_int << zero_d_int).shape)
        self.assertEqual((), (zero_d_int << 1).shape)
        self.assertEqual((1,), (one_d_int << zero_d_int).shape)
        self.assertEqual((1,), (zero_d_int << one_d_int).shape)
        self.assertEqual((1,), (one_d_int << 1).shape)

        # or
        self.assertEqual((), (zero_d_int | zero_d_int).shape)
        self.assertEqual((), (zero_d_int | 1).shape)
        self.assertEqual((1,), (one_d_int | zero_d_int).shape)
        self.assertEqual((1,), (zero_d_int | one_d_int).shape)
        self.assertEqual((1,), (one_d_int | 1).shape)

        # and
        self.assertEqual((), (zero_d_int & zero_d_int).shape)
        self.assertEqual((), (zero_d_int & 1).shape)
        self.assertEqual((1,), (one_d_int & zero_d_int).shape)
        self.assertEqual((1,), (zero_d_int & one_d_int).shape)
        self.assertEqual((1,), (one_d_int & 1).shape)

        # clone
        self.assertEqual((), zero_d.clone().shape)

        zero_d_bool = torch.tensor(True, device=device)
        one_d_bool = torch.tensor([True], device=device)

        # masked_select
        self.assertEqual((1,), torch.masked_select(zero_d_bool, zero_d_bool).shape)
        self.assertEqual((1,), torch.masked_select(zero_d_bool, one_d_bool).shape)
        self.assertEqual((1,), torch.masked_select(one_d_bool, zero_d_bool).shape)

        torch.tensor(1, dtype=torch.uint8, device=device)
        torch.tensor([1], dtype=torch.uint8, device=device)

        # mode
        self.assertEqual([(), ()], [x.shape for x in torch.mode(zero_d, dim=0, keepdim=True)])
        self.assertEqual([(), ()], [x.shape for x in torch.mode(zero_d, dim=0, keepdim=False)])
        self.assertEqual([(1,), (1,)], [x.shape for x in torch.mode(one_d, dim=0, keepdim=True)])
        self.assertEqual([(), ()], [x.shape for x in torch.mode(one_d, dim=0, keepdim=False)])

        # max
        self.assertEqual([(), ()], [x.shape for x in torch.max(zero_d, dim=0, keepdim=True)])
        self.assertEqual([(), ()], [x.shape for x in torch.max(zero_d, dim=0, keepdim=False)])
        self.assertEqual([(1,), (1,)], [x.shape for x in torch.max(one_d, dim=0, keepdim=True)])
        self.assertEqual([(), ()], [x.shape for x in torch.max(one_d, dim=0, keepdim=False)])

        # amax
        self.assertEqual((), torch.amax(zero_d, dim=0, keepdim=True).shape)
        self.assertEqual((), torch.amax(zero_d, dim=0, keepdim=False).shape)
        self.assertEqual((1,), torch.amax(one_d, dim=0, keepdim=True).shape)
        self.assertEqual((), torch.amax(one_d, dim=0, keepdim=False).shape)

        # min
        self.assertEqual([(), ()], [x.shape for x in torch.min(zero_d, dim=0, keepdim=True)])
        self.assertEqual([(), ()], [x.shape for x in torch.min(zero_d, dim=0, keepdim=False)])
        self.assertEqual([(1,), (1,)], [x.shape for x in torch.min(one_d, dim=0, keepdim=True)])
        self.assertEqual([(), ()], [x.shape for x in torch.min(one_d, dim=0, keepdim=False)])

        # amin
        self.assertEqual((), torch.amin(zero_d, dim=0, keepdim=True).shape)
        self.assertEqual((), torch.amin(zero_d, dim=0, keepdim=False).shape)
        self.assertEqual((1,), torch.amin(one_d, dim=0, keepdim=True).shape)
        self.assertEqual((), torch.amin(one_d, dim=0, keepdim=False).shape)

        # set_
        zero_d_clone = zero_d.clone()
        one_d_clone = one_d.clone()
        self.assertEqual((), zero_d_clone.set_(one_d.storage(), 0, (), ()).shape)
        self.assertEqual((1,), zero_d_clone.set_(one_d.storage(), 0, (1,), (1,)).shape)
        self.assertEqual((), one_d_clone.set_(one_d.storage(), 0, (), ()).shape)
        self.assertEqual((1,), one_d_clone.set_(one_d.storage(), 0, (1,), (1,)).shape)

        self.assertEqual((), zero_d.clone().set_(zero_d).shape)
        self.assertEqual((), one_d.clone().set_(zero_d).shape)
        self.assertEqual((1,), zero_d.clone().set_(one_d).shape)
        self.assertEqual((1,), one_d.clone().set_(one_d).shape)

        # take
        self.assertEqual((), torch.randn((2, 3), device=device).take(zero_d_int).shape)
        self.assertEqual((1,), torch.randn((2, 3), device=device).take(one_d_int).shape)

        # gather
        self.assertEqual((), torch.gather(zero_d, 0, torch.zeros((), dtype=torch.int64, device=device)).shape)
        self.assertEqual((1,), torch.gather(zero_d, 0, torch.zeros((1,), dtype=torch.int64, device=device)).shape)
        self.assertEqual((), torch.gather(one_d, 0, torch.zeros((), dtype=torch.int64, device=device)).shape)
        self.assertEqual((1,), torch.gather(one_d, 0, torch.zeros((1,), dtype=torch.int64, device=device)).shape)

        # normal
        # std must be >= 0
        zero_d_ge_0 = torch.rand((), device=device)
        # documentation says out shape matches shape of mean
        self.assertEqual((), torch.normal(zero_d, zero_d_ge_0).shape)
        self.assertEqual((1,), torch.normal(one_d, zero_d_ge_0).shape)
        self.assertEqual((), torch.normal(1, zero_d_ge_0).shape)
        self.assertEqual((), torch.normal(zero_d, 1).shape)
        self.assertEqual((1,), torch.normal(one_d, 1).shape)
        # TODO: this behavior differs on CPU and GPU, see https://github.com/pytorch/pytorch/issues/30480.
        # self.assertEqual((), torch.normal(zero_d, one_d).shape)
        # self.assertEqual((), torch.normal(1, one_d).shape)

        # convolutions.  Yes, we are testing nn.functional here; seems justified
        # given its similar to the other tests
        w = torch.randn(2, 1, 3, 3, device=device).div_(2).requires_grad_()
        self.assertRaises(RuntimeError, lambda: torch.nn.functional.conv2d(zero_d, w, groups=1))
        self.assertRaises(RuntimeError, lambda: torch.nn.functional.conv2d(zero_d, w, groups=2))

        # nll_loss -- verify input can't be 0-dimensional.
        self.assertRaises(ValueError, lambda: torch.nn.functional.nll_loss(zero_d, zero_d, reduction='none'))
        self.assertRaises(ValueError, lambda: torch.nn.functional.nll_loss(zero_d, one_d, reduction='none'))
        # verify output is 0-dimensional when reduction != 'none'
        for (input, target) in ((torch.randn(1, 1, device=device), torch.tensor([0], device=device)),
                                (torch.randn(1, 1, 1, 1, device=device), torch.tensor([[[0]]], device=device))):
            self.assertEqual((), torch.nn.functional.nll_loss(input, target, reduction='mean').shape)
            self.assertEqual((), torch.nn.functional.nll_loss(input, target, reduction='sum').shape)

    # Test that `torch._check_tensor_all` raises errors in the correct cases
    def test_check_tensor_all(self, device):
        default_message = 'Expected cond to be True'
        check_fn = torch._check_tensor_all
        expected_error = RuntimeError

        # cond must be a tensor
        with self.assertRaisesRegex(TypeError, 'cond must be a tensor'):
            check_fn(True)

        # cond tensor must be boolean
        with self.assertRaisesRegex(TypeError, 'cond tensor must have dtype torch.bool'):
            check_fn(torch.ones(1, device=device))

        test_sizes = [
            (),
            (1,),
            (10,),
            (1, 1),
            (1, 10),
            (10, 1),
            (10, 10),
            (1, 1, 1),
            (10, 1, 1),
            (1, 10, 1),
            (10, 10, 10),
        ]
        for size in test_sizes:
            t_all_true = torch.ones(size, dtype=torch.bool, device=device)
            t_all_false = torch.zeros(size, dtype=torch.bool, device=device)

            # Should not raise error
            check_fn(t_all_true)

            with self.assertRaisesRegex(expected_error, default_message):
                check_fn(t_all_false)

            if t_all_true.numel() > 1:
                t_all_true_but_one = t_all_true.clone()
                # Choose a random element to set to false
                idx = (random.choice(range(dim_size)) for dim_size in size)
                t_all_true_but_one[(..., *idx)] = False

                with self.assertRaisesRegex(expected_error, default_message):
                    check_fn(t_all_true_but_one)

            # Test a simple failure message
            message = 'message'
            with self.assertRaisesRegex(expected_error, message):
                check_fn(t_all_false, lambda: message)

            # Test message with tensor
            def message():
                return torch.arange(4)

            with self.assertRaisesRegex(expected_error, re.escape(str(message()))):
                check_fn(t_all_false, message)

            # Test format string message
            def message():
                return f"{'test'} {[1, 2, 'a', True]} {True} {100} {torch.arange(4)}"

            with self.assertRaisesRegex(expected_error, re.escape(str(message()))):
                check_fn(t_all_false, message)

    # Test that `TORCH_CHECK_TENSOR_ALL` raises errors that propagate from C++ to Python
    def test_check_tensor_internal(self, device):
        test_sizes = [
            (),
            (1,),
            (10,),
            (1, 1),
            (1, 10),
            (10, 1),
            (10, 10),
            (1, 1, 1),
            (10, 1, 1),
            (1, 10, 1),
            (10, 10, 10),
        ]
        for size in test_sizes:
            t_all_true = torch.ones(size, dtype=torch.bool, device=device)
            t_all_false = torch.zeros(size, dtype=torch.bool, device=device)

            # Should not raise error
            torch._test_check_tensor(t_all_true)

            with self.assertRaisesRegex(RuntimeError, "Test message for TORCH_CHECK_TENSOR_ALL"):
                torch._test_check_tensor(t_all_false)

            if t_all_true.numel() > 1:
                t_all_true_but_one = t_all_true.clone()
                # Choose a random element to set to false
                idx = (random.choice(range(dim_size)) for dim_size in size)
                t_all_true_but_one[(..., *idx)] = False

                with self.assertRaisesRegex(RuntimeError, "Test message for TORCH_CHECK_TENSOR_ALL"):
                    torch._test_check_tensor(t_all_true_but_one)

    # Uses mismatched arange out size to trigger a warning
    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    @unittest.skipIf(TEST_WITH_CROSSREF, "crossref perturbs line numbering")
    def test_cpp_warnings_have_python_context(self, device):
        # Creates long string in advance to avoid a too-long Python line
        s = ".+Triggered internally at.+RangeFactories.+"

        def cpp_warn_fn():
            out = torch.empty((5,))
            torch.arange(0, 3, out=out)
            return out

        # Checks eager-mode cpp warning
        with warnings.catch_warnings(record=True) as w:
            # nvfuser deprecation warning filter
            warnings.filterwarnings("ignore", "torch::jit::fuser::cuda", UserWarning)
            cpp_warn_fn()
            frameinfo = inspect.getframeinfo(inspect.currentframe())
            warning = w[0]

            # Checks for cpp context in the warning message
            escaped_warning_message = str(warning.message).encode('unicode_escape')
            self.assertTrue(re.search(s, repr(escaped_warning_message), re.IGNORECASE) is not None)

            # Checks the Python features of the warning
            # Note: the eager mode warning refers to the line in the function
            # that throws the warning.
            self.assertEqual(frameinfo.lineno - 8, warning.lineno)
            self.assertEqual(len(w), 1)

        # Checks jitted cpp warning
        with warnings.catch_warnings(record=True) as w:
            # nvfuser deprecation warning filter
            warnings.filterwarnings("ignore", "torch::jit::fuser::cuda", UserWarning)
            # ignore all deprecation warnings
            warnings.filterwarnings("ignore", category=DeprecationWarning)
            scripted_cpp_warn_fn = torch.jit.script(cpp_warn_fn)
            scripted_cpp_warn_fn()
            warning = w[0]

            # Checks for cpp context in the warning message
            escaped_warning_message = str(warning.message).encode('unicode_escape')
            self.assertTrue(re.search(s, repr(escaped_warning_message), re.IGNORECASE) is not None)

            # Checks the Python features of the warning
            # Note: the jitted warning's lineno refers to the call to the jitted
            # function, which in our test suite has a layer of indirection
            # that makes checking the Python lineno fragile
            self.assertEqual(len(w), 1)

        # Checks jitted Python warning
        def warn_fn():
            warnings.warn("Warning!")

        # The jit mimics an eager-mode Python warning in this case
        with warnings.catch_warnings(record=True) as w:
            # nvfuser deprecation warning filter
            warnings.filterwarnings("ignore", "torch::jit::fuser::cuda", UserWarning)
            # ignore all deprecation warnings
            warnings.filterwarnings("ignore", category=DeprecationWarning)
            scripted_warn_fn = torch.jit.script(warn_fn)
            scripted_warn_fn()
            frameinfo = inspect.getframeinfo(inspect.currentframe())
            warning = w[0]

            self.assertTrue(re.search('Warning!', str(warning.message)) is not None)

            # Checks the Python features of the warning
            self.assertEqual(frameinfo.lineno - 10, warning.lineno)
            self.assertEqual(len(w), 1)

    # FIXME: move to test_testing
    @onlyCPU
    def test_warn_always_caught(self, device):
        # Check that we can catch a TORCH_WARN_ONCE warning twice
        # since assertWarnsOnceRegex uses set_warn_always(True) which changes
        # TORCH_WARN_ONCE to TORCH_WARN
        a = np.arange(10)
        a.flags.writeable = False
        with self.assertWarnsOnceRegex(UserWarning, '.*non-writable.*'):
            torch.from_numpy(a)

        # OK, got it once, now try again
        with self.assertWarnsOnceRegex(UserWarning, '.*non-writable.*'):
            torch.from_numpy(a)

        # Make sure emitting two warnings will pass the assertWarnsOnceRegex
        # context manager
        with self.assertWarnsOnceRegex(UserWarning, '.*non-writable.*'):
            torch.from_numpy(a)
            torch.from_numpy(a)

    @onlyNativeDeviceTypes
    def test_complex_half_experimental_warning(self, device):
        msg = 'ComplexHalf support is experimental'
        with self.assertWarnsOnceRegex(UserWarning, msg):
            t = torch.randn(3, dtype=torch.chalf, device=device)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.rand(3, dtype=torch.chalf, device=device)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.empty(3, dtype=torch.chalf, device=device)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.ones(3, dtype=torch.chalf, device=device)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.zeros(3, dtype=torch.chalf, device=device)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.randn_like(t)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.rand_like(t)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.empty_like(t)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.ones_like(t)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.zeros_like(t)

        with self.assertWarnsOnceRegex(UserWarning, msg):
            # t + 1 allocates a new tensor for result using empty
            t + 1

    @onlyCUDA
    def test_dtypetensor_warnings(self, device):
        msg = 'The torch.cuda.*DtypeTensor constructors are no longer recommended'
        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.cuda.FloatTensor([0])

        with self.assertWarnsOnceRegex(UserWarning, msg):
            torch.cuda.DoubleTensor([0])

    def test_set_default_tensor_type_warnings(self, device):
        msg = '.*is deprecated as of PyTorch 2.1, please use torch.set_default_dtype().*'
        default_type = torch.tensor([]).type()
        try:
            with self.assertWarnsOnceRegex(UserWarning, msg):
                torch.set_default_tensor_type(torch.FloatTensor)

            if torch.cuda.is_available():
                with self.assertWarnsOnceRegex(UserWarning, msg):
                    torch.set_default_tensor_type(torch.cuda.FloatTensor)
        finally:
            torch.set_default_tensor_type(default_type)

    # TODO: this test should be in test_nn.py
    def test_conv_transposed_backward_agnostic_to_memory_format(self, device):
        in_channels = 64
        out_channels = 128
        scale_factor = 8
        batch_size = 8
        length = 16

        conv = torch.nn.ConvTranspose1d(
            in_channels, out_channels, kernel_size=scale_factor * 2, stride=scale_factor).to(device)
        layer_norm = torch.nn.LayerNorm(out_channels).to(device)

        input_ = torch.randn(batch_size, in_channels, length).to(device).contiguous()
        input_ = conv(input_).contiguous()
        input_ = layer_norm(input_.transpose(1, 2).contiguous()).contiguous()
        input_.sum().backward()

        # 3d
        conv = torch.nn.ConvTranspose3d(3, 3, kernel_size=3).to(device)
        input = torch.randn(batch_size, 3, length, length, length, device=device)
        out = conv(input)
        out.backward(torch.ones_like(out).transpose(-2, -1))

    # TODO: this test should be in test_nn.py
    @onlyCUDA
    @largeTensorTest('12GB')
    def test_conv_transposed_large(self, device):
        # ConvTranspose3d works for large input tensors (gh-32866)
        in_channels = 64
        out_channels = 128
        kernel_size = 5

        conv = torch.nn.ConvTranspose3d(
            in_channels, out_channels, kernel_size=kernel_size,
            stride=2, padding=2, output_padding=1).to(device)

        x = torch.rand([1, 64, 8, 128, 172]).to(device)
        conv(x)

    def test_is_set_to(self, device):
        t1 = torch.empty(3, 4, 9, 10, device=device)
        t2 = torch.empty(3, 4, 9, 10, device=device)
        t3 = torch.tensor([], device=device).set_(t1)
        t4 = t3.clone().resize_(12, 90)
        self.assertFalse(t1.is_set_to(t2))
        self.assertTrue(t1.is_set_to(t3))
        self.assertTrue(t3.is_set_to(t1), "is_set_to should be symmetric")
        self.assertFalse(t1.is_set_to(t4))
        self.assertFalse(torch.tensor([]).is_set_to(torch.tensor([])),
                         "Tensors with no storages should not appear to be set "
                         "to each other")

        t1 = torch.tensor([True, True], dtype=torch.bool, device=device)
        t2 = torch.tensor([0], dtype=torch.bool, device=device).set_(t1)
        self.assertTrue(t1.is_set_to(t2))

        # test that sizes must match
        t1 = torch.empty([2, 3, 4], device=device)
        t2 = t1.view(4, 3, 2)
        self.assertFalse(t1.is_set_to(t2))
        self.assertFalse(t2.is_set_to(t1))

        # test that legacy empty size behavior used to be respected (i.e. all
        # empty tensors were logically collapsed to size [0]).
        t1 = torch.empty([2, 5, 0], device=device)
        t2 = t1.view([0])
        self.assertFalse(t1.is_set_to(t2))
        self.assertFalse(t2.is_set_to(t1))

    # See https://github.com/pytorch/pytorch/issues/72650
    @skipMeta
    @parametrize(
        "fn",
        [
            "dist", "atan2", "pow", "lerp", "add", "sub", "mul", "div", "fmod", "remainder", "eq", "ge", "gt", "le",
            "lt", "max", "min", "ne", "addcdiv", "addcmul", "masked_scatter", "masked_select", "masked_fill", "map",
            "map2", "copy",
        ],
    )
    def test_broadcast(self, fn, device):
        # functions with three tensor arguments
        fns_3_args = {"map2"}
        fns_value_kwarg = {"addcdiv", "addcmul"}

        (dims_small, dims_large, dims_full) = self._select_broadcastable_dims()
        full1d = torch.randn(*dims_full, device=device).flatten().float()
        small = torch.randn(*dims_small, device=device).float()
        large = torch.randn(*dims_large, device=device).float()
        small_expanded = small.expand(*dims_full)
        large_expanded = large.expand(*dims_full)
        small2 = None
        small2_expanded = None
        if fn in fns_3_args or fn in fns_value_kwarg:
            # create another smaller tensor
            (dims_small2, _, _) = self._select_broadcastable_dims(dims_full)
            small2 = torch.randn(*dims_small2, device=device).float()
            small2_expanded = small2.expand(*dims_full)

        if small.is_cuda and fn in ['map', 'map2']:
            # map and map2 are not implemented on CUDA tensors
            return

        if hasattr(large_expanded, fn):
            # run through tensor versions of functions
            # and verify fully expanded inputs give same results
            expanded = {large: large_expanded, small: small_expanded, small2: small2_expanded}

            def tensorfn(myfn, t1, t2):
                if fn == "lerp":
                    return myfn(t1, 0.5)
                elif fn == "masked_select":
                    return myfn(t1 < 0)
                elif fn == "masked_scatter":
                    return myfn(t1 < 0.5, full1d)
                elif fn == "masked_fill":
                    return myfn(t1 < 0.5, 1.0)
                elif fn in fns_3_args:
                    return myfn(1, t1, t2)
                elif fn in fns_value_kwarg:
                    return myfn(t1, t2, value=1)
                else:
                    return myfn(t1)

            # test various orders
            for first, second, third in [(large, small, small2), (small, large, small2),
                                         (small2, small, large), (small2, large, small)]:
                if first is None:
                    break  # ignore last iter when small2 is None
                method_expanded = getattr(expanded[first], fn)
                method = getattr(first, fn)
                r1 = tensorfn(method_expanded, expanded[second], expanded[third])
                r2 = tensorfn(method, second, third)
                self.assertEqual(r1, r2)

        # now for torch. versions of functions
        if hasattr(torch, fn):
            fntorch = getattr(torch, fn)
            expanded = {large: large_expanded, small: small_expanded, small2: small2_expanded}

            def torchfn(t1, t2, t3):
                if fn == "lerp":
                    return fntorch(t1, t2, 0.5)
                elif fn == "masked_select":
                    return fntorch(t1, t2 < 0)
                elif fn == "masked_scatter":
                    return fntorch(t1, t2 < 0.5, full1d)
                elif fn == "masked_fill":
                    return fntorch(t1, t2 < 0.5, 1.0)
                elif fn in fns_3_args:
                    return fntorch(t1, 1.0, t2, t3)
                elif fn in fns_value_kwarg:
                    return fntorch(t1, t2, t3, value=1.0)
                else:
                    return fntorch(t1, t2)

            # test various orders
            for first, second, third in [(large, small, small2), (small, large, small2),
                                         (small2, small, large), (small2, large, small)]:
                if first is None:
                    break  # ignore last iter when small2 is None
                r1 = torchfn(expanded[first], expanded[second], expanded[third])
                r2 = torchfn(first, second, third)
                self.assertEqual(r1, r2)

        # now for in place functions
        # in-place tensor is not broadcastable; test only guaranteed
        # to work by broadcasting other argument(s)
        if not hasattr(large_expanded, fn + "_"):
            return

        # need to clone largeExpanded so we can reuse, since functions are in-place
        large_expanded_clone = large_expanded.clone()

        def tensorfn_inplace(t0, t1, t2=None):
            t0_fn = getattr(t0, fn + "_")
            if fn == "lerp":
                return t0_fn(t1, 0.5)
            elif fn == "masked_scatter":
                return t0_fn(t1 < 0.5, full1d)
            elif fn == "masked_fill":
                return t0_fn(t1 < 0.5, 1.0)
            elif fn == "map":
                return t0_fn(t1, lambda x, y: x + y)
            elif fn == "map2":
                return t0_fn(t1, t2, lambda x, y, z: x + y + z)
            elif fn in fns_3_args:
                return t0_fn(1.0, t1, t2)
            elif fn in fns_value_kwarg:
                return t0_fn(t1, t2, value=1.0)
            else:
                return t0_fn(t1)
        # in-place pointwise operations don't actually work if the in-place
        # tensor is 0-strided (numpy has the same issue)
        if (0 not in large_expanded.stride() and 0 not in large_expanded_clone.stride()):
            r1 = tensorfn_inplace(large_expanded, small_expanded, small2_expanded)
            r2 = tensorfn_inplace(large_expanded_clone, small, small2)
            self.assertEqual(r1, r2)

        def broadcastable(t0, t1, t2=None):
            try:
                t1.expand_as(t0)
                if t2 is not None:
                    t2.expand_as(t0)
            except RuntimeError:
                return False
            return True

        def _test_in_place_broadcastable(t0, t1, t2=None):
            if not broadcastable(t0, t1, t2):
                same_size = t0.numel() == t1.numel() and (t0.numel() == t2.numel() if t2 is not None else True)
                if not same_size:
                    # Functionalization converts the inplace to an out-of-place, which causes us to error.
                    # We should fix this, but "error probably on bad inputs" isn't a hi-pri PT2 item.
                    if not TEST_WITH_TORCHINDUCTOR:
                        self.assertRaises(RuntimeError, lambda: tensorfn_inplace(t0, t1, t2))
            else:
                tensorfn_inplace(t0, t1, t2)

        if fn not in fns_3_args and fn not in fns_value_kwarg:
            _test_in_place_broadcastable(small, large_expanded)
            _test_in_place_broadcastable(small, large)
        else:
            _test_in_place_broadcastable(small2, small_expanded, large_expanded)
            _test_in_place_broadcastable(small2, small, large)

    @onlyCPU
    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    @dtypes(*get_all_qint_dtypes())
    def test_nondeterministic_resize_quantized(self, device, dtype):
        a = torch.tensor([-1, 0, 1, 2, 3], dtype=torch.float, device=device)
        b = torch.quantize_per_tensor(a, 0.1, 10, dtype)
        self.check_nondeterministic_alert(
            lambda: b.resize_((10,)),
            'quantized_resize_cpu_')

    @skipXLA
    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16, torch.uint16, torch.uint32, torch.uint64))
    def test_deterministic_resize(self, device, dtype):
        test_cases = [
            # size, stride, resize_size
            ((10,), (1,), (5,)),
            ((10,), (0,), (10,)),
            ((10,), (1,), (20,)),
            ((2, 3, 4), None, (2, 3, 4)),
            ((2, 3, 4), None, (6, 3, 4)),
            ((2, 3, 4), None, (2, 5, 4)),
            ((2, 3, 4), None, (2, 3, 6)),
            ((2, 3, 4), None, (3, 4, 5)),
            ((2, 3, 4), (1, 4, 12), (2, 3, 4)),
            ((2, 3, 4), (1, 4, 12), (4, 3, 4)),
            ((2, 3, 4), (1, 4, 12), (2, 4, 4)),
            ((2, 3, 4), (1, 4, 12), (2, 3, 5)),
            ((2, 3, 4), (1, 4, 12), (3, 4, 5)),
            ((2, 3, 4), (1, 0, 1), (2, 4, 5)),
        ]

        for size, stride, resize_size in test_cases:
            if stride is None:
                a = torch.zeros(size, dtype=dtype, device=device)
            else:
                a = torch.empty_strided(size, stride, dtype=dtype, device=device).fill_(0)
            old_storage = a.untyped_storage().clone()
            with DeterministicGuard(True, fill_uninitialized_memory=True):
                a.resize_(resize_size)

            new_storage = a.untyped_storage()

            # If storage size was increased, check that the new section is
            # filled with NaN/MAX_INT. Otherwise, check that the storages are
            # equal.
            old_tensor = torch.tensor(old_storage, dtype=dtype)
            old_numel = old_tensor.numel()
            new_tensor = torch.tensor(new_storage, dtype=dtype)
            new_numel = new_tensor.numel()

            if new_numel > old_numel:
                self.assertEqual(new_tensor[:old_numel], old_tensor)
                fill_section = new_tensor[old_numel:]

                if dtype.is_floating_point or dtype.is_complex:
                    self.assertTrue(fill_section.isnan().all())
                else:
                    if dtype == torch.bool:
                        max_val = True
                    else:
                        max_val = torch.iinfo(dtype).max
                    self.assertTrue(fill_section.eq(max_val).all())
            else:
                self.assertEqual(old_tensor, new_tensor)

    # When deterministic algorithms are enabled, `torch.empty` should fill floating
    # point tensors with NaN and integer tensors with MAX_INT
    @skipXLA
    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    @dtypes(*all_types_and_complex_and(
        torch.half, torch.bool, torch.bfloat16, torch.uint16, torch.uint32, torch.uint64, torch.complex32))
    def test_deterministic_empty(self, device, dtype):
        gen_fns = [
            lambda: torch.empty(10, 9, device=device, dtype=dtype),
            lambda: torch.empty(10, 9, out=torch.zeros(1, device=device, dtype=dtype)),
            lambda: torch.empty_like(torch.zeros(10, 9, device=device, dtype=dtype)),
            lambda: torch.empty_like(torch.zeros(10, 9, device=device, dtype=dtype), memory_format=torch.contiguous_format),
            lambda: torch.empty_strided((10, 9), (1, 5), device=device, dtype=dtype),
            lambda: torch.empty_permuted((2, 3, 5), (1, 0, 2), device=device, dtype=dtype),
        ]

        for gen_fn in gen_fns:
            with DeterministicGuard(True, fill_uninitialized_memory=True):
                res = gen_fn()

            if dtype.is_floating_point or dtype.is_complex:
                self.assertTrue(res.isnan().all())
            else:
                if dtype == torch.bool:
                    max_val = True
                else:
                    max_val = torch.iinfo(dtype).max
                self.assertTrue(res.eq(max_val).all())

    # FIXME: update OpInfos to support "nondeterministic samples" and port these tests
    #   to that architecture
    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_AvgPool3d(self, device):
        module = torch.nn.AvgPool3d(3)
        input = torch.randn(2, 3, 3, 3, requires_grad=True, device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'avg_pool3d_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_AdaptiveAvgPool2d(self, device):
        module = torch.nn.AdaptiveAvgPool2d(3)
        input = torch.randn(2, 3, 3, requires_grad=True, device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'adaptive_avg_pool2d_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_AdaptiveAvgPool3d(self, device):
        module = torch.nn.AdaptiveAvgPool3d(3)
        input = torch.randn(2, 3, 3, 3, requires_grad=True, device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'adaptive_avg_pool3d_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_MaxPool3d(self, device):
        module = torch.nn.MaxPool3d(3)
        input = torch.randn(2, 3, 3, 3, requires_grad=True, device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'max_pool3d_with_indices_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_AdaptiveMaxPool2d(self, device):
        module = torch.nn.AdaptiveMaxPool2d(3)
        input = torch.randn(2, 3, 3, requires_grad=True, device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'adaptive_max_pool2d_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_FractionalMaxPool2d(self, device):
        module = torch.nn.FractionalMaxPool2d(2, output_ratio=0.5)
        input = torch.randn(2, 3, 3, 3, requires_grad=True, device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'fractional_max_pool2d_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_FractionalMaxPool3d(self, device):
        module = torch.nn.FractionalMaxPool3d(2, output_ratio=0.5)
        input = torch.randn(2, 3, 3, 3, 3, requires_grad=True, device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'fractional_max_pool3d_backward_cuda',
            torch.device(device).type == 'cuda')

    @dtypes(*floating_types_and(torch.half))
    @onlyNativeDeviceTypes
    def test_nondeterministic_alert_MaxUnpool1d(self, device, dtype):
        module = torch.nn.MaxUnpool1d(3, 1)
        input = torch.randn(1, 1, 7, dtype=dtype, device=device)
        indices = torch.zeros_like(input, dtype=torch.long, device=device)

        self.check_nondeterministic_alert(
            lambda: module(input, indices),
            'max_unpooling2d_forward_out')

    @dtypes(*floating_types_and(torch.half))
    @onlyNativeDeviceTypes
    def test_nondeterministic_alert_MaxUnpool2d(self, device, dtype):
        module = torch.nn.MaxUnpool2d(3, 1)
        input = torch.randn(1, 1, 7, 7, dtype=dtype, device=device)
        indices = torch.zeros_like(input, dtype=torch.long, device=device)

        self.check_nondeterministic_alert(
            lambda: module(input, indices),
            'max_unpooling2d_forward_out')

    @dtypes(*floating_types_and(torch.half))
    @onlyNativeDeviceTypes
    def test_nondeterministic_alert_MaxUnpool3d(self, device, dtype):
        module = torch.nn.MaxUnpool3d(3, 1)
        input = torch.randn(1, 1, 7, 7, 7, dtype=dtype, device=device)
        indices = torch.zeros_like(input, dtype=torch.long, device=device)

        self.check_nondeterministic_alert(
            lambda: module(input, indices),
            'max_unpooling3d_forward_out')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_interpolate_linear(self, device):
        input = torch.randn(1, 2, 4, device=device, requires_grad=True)
        res = torch.nn.functional.interpolate(
            input,
            size=12,
            mode='linear',
            align_corners=False)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad),
            'upsample_linear1d_backward_out_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_interpolate_bilinear(self, device):
        input = torch.randn(1, 2, 4, 4, device=device, requires_grad=True)
        res = torch.nn.functional.interpolate(
            input,
            size=12,
            mode='bilinear',
            align_corners=False)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad),
            'upsample_bilinear2d_backward_out_cuda',
            torch.device(device).type == 'cuda')

    def test_no_nondeterministic_alert_interpolate_bilinear(self, device):
        input = torch.randn(1, 2, 4, 4, device=device, requires_grad=True)

        def fn():
            res = torch.nn.functional.interpolate(
                input,
                size=12,
                mode='bilinear',
                align_corners=False)
            grad = torch.ones_like(res)
            return res.backward(grad)

        self.check_nondeterministic_alert(
            fn,
            'upsample_bilinear2d_backward_out_cuda',
            False)

    def test_no_nondeterministic_alert_interpolate_trilinear(self, device):
        input = torch.randn(1, 2, 4, 4, 4, device=device, requires_grad=True)

        def fn():
            res = torch.nn.functional.interpolate(
                input,
                size=12,
                mode='trilinear',
                align_corners=False)
            grad = torch.ones_like(res)
            return res.backward(grad)

        self.check_nondeterministic_alert(
            fn,
            'upsample_trilinear3d_backward_out_cuda',
            False)

    @skipIfTorchInductor("aot-autograd issue")
    def test_deterministic_max_pool3d(self, device):
        test_cases = [
            # size, kernel_size, stride, padding, dilation, ceil_mode
            [(2, 3, 8, 8, 8), 3, 1, 1, 1, False],
            [(2, 3, 8, 8, 8), 3, 2, 1, 1, False],
            [(2, 3, 8, 8, 8), 2, 2, 0, 1, False],
            [(2, 3, 8, 8, 8), 3, 2, 1, 1, True],
            [(3, 8, 8, 8), 3, 1, 1, 1, False],  # unbatched
        ]

        for size, ks, st, pa, di, cm in test_cases:
            input = torch.randn(*size, device=device, requires_grad=True)
            grad = None
            with DeterministicGuard(True):
                for _ in range(5):
                    res, _ = torch.nn.functional.max_pool3d(
                        input, ks, st, pa, di, cm, return_indices=True)
                    res.backward(torch.ones_like(res))
                    if grad is None:
                        grad = input.grad
                    else:
                        self.assertEqual(grad, input.grad, atol=0, rtol=0)
                    input.grad = None

    @skipIfTorchInductor("aot-autograd issue")
    def test_deterministic_replication_pad2d(self, device):
        test_cases = [
            # size, padding
            [(1, 2, 4, 4), (0, 0, 0, 0)],
            [(1, 2, 4, 4), (3, 4, 5, 6)],
            [(3, 8, 7), (0, 0, 0, 0)],
            [(3, 8, 7), (4, 3, 2, 7)],
        ]

        if torch.device(device).type != 'xla':
            test_cases += [
                [(4, 3, 5, 10), (-9, 4, 5, 6)],
                [(3, 8, 7), (-4, -2, -2, -3)],
            ]

        for size, padding in test_cases:
            input = torch.randn(*size, device=device, requires_grad=True)
            grad = None
            with DeterministicGuard(True):
                res = torch.nn.functional.pad(
                    input,
                    padding,
                    mode='replicate')
                res.backward(torch.ones_like(res))
                if grad is None:
                    grad = input.grad
                else:
                    self.assertEqual(grad, input.grad, atol=0, rtol=0)
                input.grad = None

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_deterministic_interpolate_bilinear(self, device):
        input = torch.randn(1, 2, 4, 4, device=device, requires_grad=True)
        grad = None
        with DeterministicGuard(True):
            for _ in range(5):
                res = torch.nn.functional.interpolate(
                    input,
                    size=12,
                    mode='bilinear',
                    align_corners=False)
                res.backward(torch.ones_like(res))
                if grad is None:
                    grad = input.grad
                else:
                    self.assertEqual(grad, input.grad, atol=0, rtol=0)
                input.grad = None

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_interpolate_bicubic(self, device):
        input = torch.randn(1, 2, 4, 4, device=device, requires_grad=True)
        res = torch.nn.functional.interpolate(
            input,
            size=12,
            mode='bicubic',
            align_corners=False)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad),
            'upsample_bicubic2d_backward_out_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_interpolate_trilinear(self, device):
        input = torch.randn(1, 2, 4, 4, 4, device=device, requires_grad=True)
        res = torch.nn.functional.interpolate(
            input,
            size=12,
            mode='trilinear',
            align_corners=False)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad),
            'upsample_trilinear3d_backward_out_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_ReflectionPad1d(self, device):
        module = torch.nn.ReflectionPad1d((1, 2))
        input = torch.randn(2, 3, 8, device=device, requires_grad=True)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'reflection_pad1d_backward_out_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_ReflectionPad3d(self, device):
        module = torch.nn.ReflectionPad3d((1, 2, 3, 4, 5, 6))
        input = torch.randn(2, 3, 8, 8, 8, device=device, requires_grad=True)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'reflection_pad3d_backward_out_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_ReplicationPad1d(self, device):
        module = torch.nn.ReplicationPad1d((1, 2))
        input = torch.randn(2, 3, 4, device=device, requires_grad=True)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'replication_pad1d_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_ReplicationPad2d(self, device):
        module = torch.nn.ReplicationPad2d((1, 2, 3, 4))
        input = torch.randn(2, 3, 4, 4, device=device, requires_grad=True)
        res = module(input)
        grad = torch.ones_like(res)

        # Nondeterministic alert should only be raised if the forward call was
        # nondeterministic
        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'replication_pad2d_backward_cuda',
            torch.device(device).type == 'cuda')

        with DeterministicGuard(True):
            res = module(input)

        grad = torch.ones_like(res)

        # If the forward call was deterministic, nondeterministic alert should
        # not be raised
        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'replication_pad2d_backward_cuda',
            False)

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_ReplicationPad3d(self, device):
        module = torch.nn.ReplicationPad3d((1, 2, 3, 4, 5, 6))
        input = torch.randn(2, 3, 4, 4, 4, device=device, requires_grad=True)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'replication_pad3d_backward_cuda',
            torch.device(device).type == 'cuda')

    @skipIfTorchDynamo("Warning is not raised.")
    def test_nondeterministic_alert_NLLLoss(self, device):
        module = torch.nn.NLLLoss()
        input = torch.randn(2, 3, 5, 5, device=device)
        target = torch.rand(2, 5, 5, device=device).mul(3).floor().long()


        self.check_nondeterministic_alert(
            lambda: module(input, target),
            'nll_loss2d_forward_out_cuda_template',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_CTCLoss(self, device):
        module = torch.nn.CTCLoss()
        input = torch.randn(50, 3, 15, device=device, requires_grad=True)
        target = torch.randint(0, 14, (3, 30), device=device)
        input_lengths = [50, 50, 50]
        target_lengths = [30, 25, 20]
        res = module(input, target, input_lengths, target_lengths)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'ctc_loss_backward_gpu',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_EmbeddingBag_max(self, device):
        module = torch.nn.EmbeddingBag(
            4, 3, None, 2., False, 'max',
            _weight=torch.randn(4, 3, device=device, requires_grad=True))
        input = torch.randint(0, 3, (4, 3), device=device)
        res = module(input)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'embedding_bag_backward_cuda_max',
            torch.device(device).type == 'cuda')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    @onlyCUDA
    def test_deterministic_cumsum(self, device):
        test_cases = [
            # size, dim
            [(1025,), 0],
            [(8193,), 0],
            [(8191,), 0],
            [(128256,), 0],
            [(1282560,), 0],
            [(12825600,), 0],
        ]
        for size, dim in test_cases:
            input = 100 * torch.rand(*size, device=device)
            with DeterministicGuard(True):
                res0 = input.cumsum(dim)
                for _ in range(3):
                    res1 = input.cumsum(dim)
                    self.assertEqual(res0, res1, atol=0, rtol=0)

            res_cpu = input.cpu().cumsum(dim)
            self.assertEqual(res0, res_cpu, atol=1e-3, rtol=1e-2)
            # test double complex that has fewer threads than CTAs
            # that's a problem for large GPUS (H100 with 132 sms)
            # test is very tailored for that and will provide 0 signal
            # on smaller GPUs
            num_sms = 132
            elems_per_cta = 256 * 16
            N = num_sms * elems_per_cta
            input = torch.rand(N, dtype=torch.complex128, device=device)
            with DeterministicGuard(True):
                res0 = input.cumsum(dim)
                for _ in range(3):
                    res1 = input.cumsum(dim)
                    self.assertEqual(res0, res1, atol=0, rtol=0)

            res_cpu = input.cpu().cumsum(dim)
            self.assertEqual(res0, res_cpu, atol=1e-3, rtol=1e-2)

    @onlyCUDA
    @largeTensorTest('49GB')
    def test_cumsum_64bit_indexing(self, device):
        b = torch.ones(2 * 4096 * 8, 100000, dtype=torch.float, device='cuda')
        b /= 100000
        d = b.cumsum(dim=-1)
        chunk = 2**30 // b.shape[-1]
        for i in range(0, b.shape[0], chunk):
            end = min(i + chunk, b.shape[0])
            b[i:end, :].cumsum_(dim=-1)
        # cheat a bit to avoid OOM
        self.assertEqual(b[0, :], d[0, :], atol=3e-5, rtol=3e-5)
        self.assertEqual(b[-1, :], d[-1, :], atol=3e-5, rtol=3e-5)

    @onlyCUDA
    @largeTensorTest('48GB')
    def test_cumsum_outer_dim_64bit_indexing(self, device):
        x = torch.zeros(309504, 1, 16384, device=device)
        torch.exp(x)
        cumsum = torch.cumsum(x, dim=1)
        self.assertEqual(cumsum.max().item(), 0., atol=0., rtol=0.)

    @expectedFailureMeta  # expected a non-determinitic error, but it was not raised
    @onlyNativeDeviceTypes
    def test_nondeterministic_alert_put(self, device):
        a = torch.randn(10, device=device)
        indices = torch.tensor([0, 0], device=device)
        values = torch.tensor([0., 1.], device=device)

        for op_call in [torch.Tensor.put, torch.Tensor.put_]:
            self.check_nondeterministic_alert(
                lambda: op_call(a, indices, values, accumulate=False),
                'put_')

    # warn_only=False correctly raises RuntimeError: put_ does not have a deterministic implementation
    # warn_only=True logs warning from the FallbackKernel: torch.ops.aten.put_.default, instead of as UserWarning:
    # [W Context.cpp:%(lineno)] Warning: put_ does not have a deterministic implementation
    @skipIfTorchInductor("warning is logged from the FallbackKernel: torch.ops.aten.put_.default when warn_only=True")
    def test_nondeterministic_alert_put_accumulate(self, device):
        a = torch.randn(10, device=device)
        indices = torch.tensor([0, 0], device=device)
        values = torch.tensor([0., 1.], device=device)

        for op_call in [torch.Tensor.put, torch.Tensor.put_]:
            self.check_nondeterministic_alert(
                lambda: op_call(a, indices, values, accumulate=True),
                'put_',
                torch.device(device).type == 'cuda')

    @dtypes(torch.float32)
    @dtypesIfCUDA(torch.float32, torch.int32)
    def test_nondeterministic_alert_histc(self, device, dtype):
        a = torch.tensor([], device=device, dtype=dtype)
        for op_call in [torch.histc, torch.Tensor.histc]:
            self.check_nondeterministic_alert(
                lambda: op_call(a, min=0, max=3),
                '_histc_cuda with floating point input',
                torch.device(device).type == 'cuda' and dtype.is_floating_point)

    def test_nondeterministic_alert_bincount(self, device):
        a = torch.tensor([], device=device, dtype=torch.long)
        weights = torch.tensor([], device=device)

        for op_call in [torch.bincount, torch.Tensor.bincount]:
            # Error should only be raised when device is CUDA and weights are
            # given
            self.check_nondeterministic_alert(
                lambda: op_call(a, weights),
                '_bincount_cuda',
                torch.device(device).type == 'cuda')

            self.check_nondeterministic_alert(
                lambda: op_call(a),
                '_bincount_cuda',
                False)

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_grid_sample_2d(self, device):
        input = torch.empty(1, 1, 2, 2, device=device, requires_grad=True)
        grid = torch.empty(1, 1, 1, 2, device=device)
        res = torch.nn.functional.grid_sample(input, grid, align_corners=False)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'grid_sampler_2d_backward_cuda',
            torch.device(device).type == 'cuda')

    @unittest.skipIf(not TEST_CUDNN, "CUDNN not available")
    @skipIfRocm
    @onlyCUDA
    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_grid_sample_2d_cudnn(self, device):
        def fn():
            input = torch.empty(1, 1, 2, 2, device=device, requires_grad=True)
            grid = torch.empty(1, 1, 1, 2, device=device)
            with torch.backends.cudnn.flags(enabled=True):
                res = torch.nn.functional.grid_sample(input, grid, align_corners=True)
                res.backward(torch.ones_like(res))

        self.check_nondeterministic_alert(
            fn,
            'cudnn_grid_sampler_backward')

    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/113707")
    def test_nondeterministic_alert_grid_sample_3d(self, device):
        input = torch.empty(1, 1, 2, 2, 2, device=device, requires_grad=True)
        grid = torch.empty(1, 1, 1, 2, 3, device=device)
        res = torch.nn.functional.grid_sample(input, grid, align_corners=False)
        grad = torch.ones_like(res)

        self.check_nondeterministic_alert(
            lambda: res.backward(grad, retain_graph=True),
            'grid_sampler_3d_backward_cuda',
            torch.device(device).type == 'cuda')

    def test_invalid_shapes_grid_sampler(self, device):
        make_arg = partial(
            make_tensor, device=device, dtype=torch.float64, requires_grad=True)

        inputs = (
            # input, grid
            ((5, 5, 5, 5, 5,), (1, 1, 1, 4, 4,)),  # 3d
            ((5, 5, 5, 5,), (1, 1, 4, 4,)),  # 2d
        )

        interpolation_mode = 0
        padding_mode = 0
        align_corners = True

        err = "expected grid and input to have same batch size"

        for input, grid in inputs:
            input = make_arg(input)
            grid = make_arg(grid, low=-1, high=1)

            # Wrapper for the 2d, 3d, and cuDNN functions listed below.
            with self.assertRaisesRegex(RuntimeError, err):
                torch.grid_sampler(
                    input, grid, interpolation_mode, padding_mode,
                    align_corners)

            # Expects 2d input.
            with self.assertRaisesRegex(RuntimeError, err):
                torch.grid_sampler_2d(
                    input, grid, interpolation_mode, padding_mode,
                    align_corners)

            # Expects 3d input.
            with self.assertRaisesRegex(RuntimeError, err):
                torch.grid_sampler_3d(
                    input, grid, interpolation_mode, padding_mode,
                    align_corners)

            # Expects 2d input.
            with self.assertRaisesRegex(RuntimeError, err):
                torch._grid_sampler_2d_cpu_fallback(
                    input, grid, interpolation_mode, padding_mode,
                    align_corners)

            # Expects 2d input, on CUDA.
            # Doesn't work on CPU and ROCm.
            if device != 'cpu' and TEST_CUDNN and not TEST_WITH_ROCM:
                with self.assertRaisesRegex(RuntimeError, err):
                    torch.cudnn_grid_sampler(input, grid)

    def test_dist(self, device):
        def run_test(x, y):
            for p in [0, 1, 2, 3, 4, inf, -inf]:
                dist_xy = torch.dist(x, y, p)
                dist_xy_norm = torch.norm(x - y, p)
                self.assertEqual(dist_xy, dist_xy_norm)

        run_test(torch.randn(5, device=device), torch.randn(5, device=device))

        x = torch.zeros(3, device=device)
        y = torch.zeros(3, device=device)
        y[1] = 1.
        run_test(x, y)

    # Ensures that median throws nondeterministic alerts in the correct cases
    @dtypes(torch.double)
    def test_nondeterministic_alert_median(self, device, dtype):
        def test_func(call_type):
            S = 10
            a = torch.randn(S, device=device)
            if call_type == 'function':
                torch.median(a)
            elif call_type == 'function with indices':
                torch.median(a, 0)
            elif call_type == 'method':
                a.median()
            elif call_type == 'method with indices':
                a.median(0)
            elif call_type == 'out with indices':
                result = torch.empty_like(a)
                indices = torch.empty((), dtype=torch.long, device=device)
                torch.median(a, 0, out=(result, indices))
            else:
                self.fail(f"'{call_type}' is not a valid call type")

        def test_func_expect_error(call_type, should_error):
            self.check_nondeterministic_alert(
                lambda: test_func(call_type),
                'median CUDA with indices output',
                should_error)

        is_cuda = torch.device(device).type == 'cuda'

        test_func_expect_error('function', False)
        test_func_expect_error('function with indices', is_cuda)
        test_func_expect_error('method', False)
        test_func_expect_error('method with indices', is_cuda)
        test_func_expect_error('out with indices', is_cuda)

    # FIXME: move to test_scatter_gather_ops
    def _test_gather_backward_one_dim(self, device, deterministic: bool = False) -> None:
        with DeterministicGuard(deterministic):
            m = random.randint(2000, 3000)
            elems = random.randint(10 * m, 20 * m)
            dim = 0
            src = torch.randn(m, device=device, requires_grad=True)
            idx = torch.randint(m, (elems,), device=device)
            res = torch.gather(src, dim, idx)
            weight = torch.rand_like(res, device=device) * 10 ** 6
            res.backward(weight)
            if src.grad is None:
                raise AssertionError("expected src.grad to be not None")
            grad = src.grad.detach().clone()

            if torch.device(device).type == 'cuda' or torch.device(device).type == 'mtia':
                for _ in range(2):
                    src.grad.data.zero_()
                    res = torch.gather(src, dim, idx)
                    res.backward(weight)
                    self.assertEqual(src.grad, grad, atol=0, rtol=0)
            else:
                expected = torch.zeros_like(src, device=device)
                for i in range(elems):
                    expected[idx[i]] += weight[i]
                self.assertEqual(grad, expected, atol=0, rtol=0)

    # FIXME: move to test_scatter_gather_ops
    @onlyNativeDeviceTypes
    def test_gather_backward_deterministic_path(self, device) -> None:
        self._test_gather_backward_one_dim(device, True)

    # FIXME: move to test_scatter_gather_ops
    @onlyCPU
    def test_gather_backward_one_dim(self, device) -> None:
        self._test_gather_backward_one_dim(device, False)

    # FIXME: move to test_scatter_gather_ops
    @onlyNativeDeviceTypes
    @dtypes(torch.float, torch.int)
    def test_scatter_add_one_dim_deterministic(self, device, dtype) -> None:
        with DeterministicGuard(True):
            m = random.randint(20, 30)
            elems = random.randint(2000 * m, 3000 * m)
            dim = 0
            # initialize in a way that would produce
            # different ints
            src = (torch.rand(elems, device=device) * 10).to(dtype)
            idx = torch.randint(m, (elems,), device=device)

            x = torch.zeros(m, device=device, dtype=dtype)
            res = x.scatter_add(dim, idx, src)

            # Checking if scatter_add is deterministic
            for _ in range(5):
                res_next = x.scatter_add(dim, idx, src)
                self.assertEqual(res, res_next, atol=0, rtol=0)
                res = res_next

            expected = torch.zeros(m, device=device, dtype=dtype)
            for i in range(elems):
                expected[idx[i]] += src[i]

            self.assertEqual(res, expected, atol=1e-4, rtol=1e-5)

    # FIXME: move to test_scatter_gather_ops
    @onlyNativeDeviceTypes
    def test_scatter_zero_size_index(self, device) -> None:
        null_index = torch.zeros((0, 4), dtype=torch.int64)
        null_arr = torch.zeros((0, 4))
        original = torch.arange(4, dtype=torch.float32)
        result = original.scatter(0, null_index, null_arr)
        self.assertEqual(result, original, atol=0, rtol=0)

    @onlyCUDA
    @skipIfTorchInductor("FIXME")
    def test_sync_warning(self, device):

        def _sync_raises_helper(f, level):
            with CudaSyncGuard(level):
                if level == 1:
                    with self.assertWarnsRegex(UserWarning, "called a synchronizing "):
                        f()
                elif level == 2:
                    with self.assertRaisesRegex(RuntimeError, "called a synchronizing "):
                        f()

        def _no_sync_helper(f, level):
            with CudaSyncGuard(level):
                f()

        def _ind_put_fn(x, ind, val):
            x[ind] = val
            return x

        def _ind_get_fn(x, ind):
            return x[ind]

        def _cond_fn(x):
            if x:  # taking boolean value of a tensor synchronizes
                return x
            else:
                return 2 * x

        # prepare inputs for subsequent ops
        size = 4
        x = torch.rand(size, device=device)
        y = torch.rand((), device=device)
        ind = torch.randint(size, (3,), device=device)
        ind_cpu = ind.cpu()
        repeats = torch.full((1,), 2, device=device)
        mask = torch.randint(2, (size,), device=device, dtype=bool)
        mask_cpu = mask.cpu()
        expect_no_sync = (lambda: _ind_put_fn(x, mask, 1.),
                          lambda: _ind_put_fn(x, mask_cpu, y),
                          lambda: _ind_put_fn(x, ind, y),
                          lambda: _ind_get_fn(x, mask_cpu),
                          lambda: _ind_get_fn(x, ind),
                          lambda: torch.nn.functional.one_hot(ind, num_classes=size),
                          lambda: torch.randperm(20000, device=device),
                          lambda: torch.repeat_interleave(x, 2, output_size=2 * size),
                          lambda: torch.repeat_interleave(x, repeats, output_size=2 * size),
                          lambda: torch.any(y),
                          lambda: torch.normal(x, x))
        expect_sync = (lambda: _ind_put_fn(x, mask, y),
                       lambda: _ind_put_fn(x, ind_cpu, y),
                       lambda: _ind_get_fn(x, mask),
                       lambda: _ind_get_fn(x, ind_cpu),
                       lambda: x.nonzero(),
                       lambda: _cond_fn(y),
                       lambda: torch.nn.functional.one_hot(ind),
                       lambda: torch.repeat_interleave(x, repeats))
        for f, level in product(expect_no_sync, (1, 2)):
            _no_sync_helper(f, level)
        for f, level in product(expect_sync, (1, 2)):
            _sync_raises_helper(f, level)


    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_log_normal(self, device, dtype):
        a = torch.tensor([10], dtype=dtype, device=device).log_normal_()
        self.assertEqual(a.dtype, dtype)
        self.assertEqual(a.size(), torch.Size([1]))

    @dtypes(*all_types_and(torch.half, torch.bfloat16))
    def test_geometric(self, device, dtype):
        a = torch.tensor([10], dtype=dtype, device=device).geometric_(0.5)
        self.assertEqual(a.dtype, dtype)
        self.assertEqual(a.size(), torch.Size([1]))

    def test_repeat_interleave(self, device):
        y = torch.tensor([[1, 2], [3, 4]], device=device)
        # exercise single argument function signature
        temp = y.repeat_interleave(2)
        self.assertEqual(torch.Size([8]), temp.size())

        repeat_dtypes = [torch.int, torch.long]
        if device == "cpu":
            repeat_dtypes.extend([torch.int8, torch.uint8, torch.int16])

        for dtype in repeat_dtypes:
            lengths = torch.tensor([1, 2], dtype=dtype, device=device)
            output_size = torch.sum(lengths)
            a = torch.repeat_interleave(
                y,
                lengths,
                dim=0,
            )
            self.assertEqual(a.dtype, y.dtype)
            self.assertEqual(a.size(), torch.Size([3, 2]))

            a_with_output = torch.repeat_interleave(
                y,
                lengths,
                dim=0,
                output_size=output_size,
            )
            self.assertEqual(a_with_output.dtype, y.dtype)
            self.assertEqual(a_with_output.size(), torch.Size([3, 2]))

    @dtypes(*floating_types())
    @dtypesIfCPU(*floating_types_and(torch.bfloat16, torch.half))
    @dtypesIfCUDA(*floating_types_and(torch.half))
    def test_bernoulli_p(self, device, dtype):
        for trivial_p in ([0, 1], [1, 0, 1, 1, 0, 1]):
            x = torch.tensor(trivial_p, dtype=dtype, device=device)
            self.assertEqual(x.bernoulli().tolist(), trivial_p)

        def isBinary(t):
            return torch.ne(t, 0).mul_(torch.ne(t, 1)).sum().item() == 0

        p = torch.rand(5, 5, dtype=dtype, device=device)
        self.assertTrue(isBinary(p.bernoulli()))

        p = torch.rand(5, dtype=dtype, device=device).expand(5, 5)
        self.assertTrue(isBinary(p.bernoulli()))

        p = torch.rand(5, 5, dtype=dtype, device=device)
        torch.bernoulli(torch.rand_like(p), out=p)
        self.assertTrue(isBinary(p))

    # RngUniform not implemented for Integral type in XLA test
    @dtypes(*floating_types())
    @dtypesIfCPU(*all_types_and(torch.bool, torch.half))
    @dtypesIfCUDA(*all_types_and(torch.bool, torch.half))
    def test_bernoulli_self(self, device, dtype):

        def isBinary(t):
            return torch.ne(t, 0).mul_(torch.ne(t, 1)).sum().item() == 0

        t = torch.empty(10, 10, dtype=dtype, device=device)

        t.fill_(2)
        t.bernoulli_(0.5)
        self.assertTrue(isBinary(t))

        for p_dtype in floating_types_and(*[torch.half] if device.startswith('cuda') else []):
            p = torch.rand(10, dtype=p_dtype, device=device).expand(10, 10)
            t.fill_(2)
            t.bernoulli_(p)
            self.assertTrue(isBinary(t))

            t.fill_(2)
            torch.bernoulli(torch.rand_like(t, dtype=p_dtype), out=t)
            self.assertTrue(isBinary(t))

            t.fill_(2)
            t.bernoulli_(torch.rand_like(t, dtype=p_dtype))
            self.assertTrue(isBinary(t))

    @slowTest
    @dtypes(*floating_types_and(torch.half))
    @dtypesIfCUDA(*floating_types_and(torch.half))
    def test_bernoulli_edge_cases(self, device, dtype):
        # Need to draw a lot of samples to cover every random floating point number.
        a = torch.zeros(10000, 10000, dtype=dtype, device=device)  # probability of drawing "1" is 0
        num_ones = (torch.bernoulli(a) == 1).sum()
        self.assertEqual(num_ones, 0)

        b = torch.ones(10000, 10000, dtype=dtype, device=device)  # probability of drawing "1" is 1
        num_zeros = (torch.bernoulli(b) == 0).sum()
        self.assertEqual(num_zeros, 0)

    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_exponential(self, device, dtype):
        a = torch.tensor([10], dtype=dtype, device=device).exponential_(0.5)
        self.assertEqual(a.dtype, dtype)
        self.assertEqual(a.size(), torch.Size([1]))

        # Tests extremal behavior
        t = torch.empty((1,), device=device, dtype=dtype).exponential_(float('inf'))
        self.assertTrue(t.item() == 0)

        # Tests that negative lambda fails
        with self.assertRaises(RuntimeError):
            torch.empty((1,), device=device, dtype=dtype).exponential_(-0.5)

    @onlyCUDA
    @dtypes(torch.half, torch.float)
    def test_exponential_no_zero(self, device, dtype):
        # naively, 0 in exponential can be generated with probability 2^-24
        # so we need more samples to check if it's not generated
        # instead of doing one
        # don't test CPU, that would be a long test
        x = torch.empty(50000000, device=device, dtype=dtype).exponential_()
        self.assertTrue(x.min() > 0)

    def _generate_correlation_tensors(self, device, dtype):
        yield make_tensor((0, 0), dtype=dtype, device=device)
        yield make_tensor((1, 0), dtype=dtype, device=device)
        yield make_tensor((0, 1), dtype=dtype, device=device)
        yield make_tensor((2,), dtype=dtype, device=device)
        yield make_tensor((2, 1), dtype=dtype, device=device)
        yield make_tensor((2, 2), dtype=dtype, device=device)
        yield make_tensor((2, 3), dtype=dtype, device=device)
        yield make_tensor((5, 10), dtype=dtype, device=device)
        yield make_tensor((5, 10), dtype=dtype, device=device, noncontiguous=True)
        if dtype != torch.int:
            yield torch.tensor([0, -2, nan, 10.2, inf], dtype=dtype, device=device)

    @tf32_on_and_off(0.005)
    @onlyNativeDeviceTypes
    @dtypes(torch.int, torch.float, torch.cfloat)
    def test_corrcoef(self, device, dtype):
        for x in self._generate_correlation_tensors(device, dtype):
            res = torch.corrcoef(x)
            ref = np.corrcoef(x.cpu().numpy())
            # complex cov can produce rounding errors, and when the norm factor is 0,
            # the imag portion of the answer could be +/- inf instead of nan, causing
            # the numpy answer to differ from the torch answer. We always check the real
            # part, but only check the imag part if it contains some normal values
            # (indicating that the norm factor was not 0).
            if res.is_complex():
                self.assertEqual(res.real, ref.real, atol=1e-05, rtol=1e-05, exact_dtype=False)
                if not (torch.isnan(res.imag) | torch.isinf(res.imag)).all():
                    self.assertEqual(res.imag, ref.imag, atol=1e-05, rtol=1e-05, exact_dtype=False)
            else:
                self.assertEqual(res, ref, atol=1e-05, rtol=1e-05, exact_dtype=False)

    @dtypes(torch.int, torch.float, torch.cfloat)
    def test_cov(self, device, dtype):
        def check(t, correction=1, fweights=None, aweights=None):
            res = torch.cov(t, correction=correction, fweights=fweights, aweights=aweights)
            t = t.cpu().numpy()
            fweights = fweights.cpu().numpy() if fweights is not None else None
            aweights = aweights.cpu().numpy() if aweights is not None else None
            ref = np.cov(t, ddof=correction, fweights=fweights, aweights=aweights)
            # complex cov can produce rounding errors, and when the norm factor is 0,
            # the imag portion of the answer could be +/- inf instead of nan, causing
            # the numpy answer to differ from the torch answer. We always check the real
            # part, but only check the imag part if it contains some normal values
            # (indicating that the norm factor was not 0).
            if res.is_complex():
                self.assertEqual(res.real, ref.real, atol=1e-05, rtol=1e-05, exact_dtype=False)
                if not (torch.isnan(res.imag) | torch.isinf(res.imag)).all():
                    self.assertEqual(res.imag, ref.imag, atol=1e-05, rtol=1e-05, exact_dtype=False)
            else:
                self.assertEqual(res, ref, atol=1e-05, rtol=1e-05, exact_dtype=False)

        for x in self._generate_correlation_tensors(device, dtype):
            check(x)
            num_observations = x.numel() if x.ndim < 2 else x.size(1)
            if num_observations > 0:
                fweights = torch.randint(1, 10, (num_observations,), device=device)
                aweights = make_tensor((num_observations,), dtype=torch.float, device=device, low=1)
                for correction, fw, aw in product([0, 1, 2], [None, fweights], [None, aweights]):
                    check(x, correction, fw, aw)

    @skipIfNoSciPy
    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_uniform_kstest(self, device, dtype):
        from scipy import stats
        size = 1000
        for from_ in [-42, 0, 4.2]:
            for to_ in [-4.2, 0, 42]:
                if to_ > from_:
                    t = torch.empty(size, dtype=dtype, device=device).uniform_(from_, to_)
                    res = stats.kstest(t.cpu().to(torch.double), 'uniform', args=(from_, (to_ - from_)))
                    self.assertTrue(res.statistic < 0.1)

    @skipIfNoSciPy
    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_normal_kstest(self, device, dtype):
        from scipy import stats
        size = 1000
        for mean in [-10, 0, 50]:
            for std in [1, 5, 10]:
                t = torch.empty(size, dtype=dtype, device=device).normal_(mean=mean, std=std)
                res = stats.kstest(t.cpu().to(torch.double), 'norm', args=(mean, std))
                self.assertTrue(res.statistic < 0.1)

    @skipIfNoSciPy
    @dtypes(torch.half, torch.bfloat16)
    def test_normal_small_std_kstest(self, device, dtype):
        from scipy import stats
        size = 50000
        for std in [0.005, 0.01]:
            t = torch.empty(size, dtype=dtype, device=device).normal_(mean=0, std=std)
            res = stats.kstest(t.cpu().to(torch.double), 'norm', args=(0, std))
            self.assertTrue(
                res.statistic < 0.01,
                msg=lambda msg: f"{msg}\nKS statistic {res.statistic:.4f} for {dtype} std={std}",
            )

    @skipIfNoSciPy
    @dtypes(torch.half, torch.bfloat16)
    def test_normal_kurtosis(self, device, dtype):
        from scipy import stats
        size = 50000
        for std in [0.01, 0.1, 1.0]:
            t = torch.empty(size, dtype=dtype, device=device).normal_(mean=0, std=std)
            kurtosis = stats.kurtosis(t.cpu().to(torch.double).numpy())
            self.assertAlmostEqual(
                kurtosis, 0.0, delta=0.15,
                msg=lambda msg: f"{msg}\nExcess kurtosis {kurtosis:.3f} too far from 0 for {dtype} std={std}",
            )

    @dtypes(torch.bfloat16)
    def test_normal_tail_reach(self, device, dtype):
        # With the broken bf16 Box-Muller (256 uniform inputs), the maximum
        # output is capped at sqrt(-2*log(1/256)) ≈ 3.33 sigma. A correct
        # implementation generates in float and reaches 4+ sigma.
        size = 1000000
        t = torch.empty(size, dtype=dtype, device=device).normal_(mean=0, std=1.0)
        max_sigma = t.float().abs().max().item()
        self.assertGreater(
            max_sigma, 3.5,
            msg=lambda msg: f"{msg}\nMax |z| = {max_sigma:.2f} sigma, expected > 3.5 (tail truncation)",
        )

    @dtypes(torch.half, torch.bfloat16)
    def test_normal_unique_values(self, device, dtype):
        size = 10000
        t = torch.empty(size, dtype=dtype, device=device).normal_(mean=0, std=0.01)
        n_unique = t.unique().numel()
        # bf16 has ~200 representable values in [-0.03, 0.03] (3-sigma range).
        # fp16 has ~60 representable values in [-0.03, 0.03].
        # A correct implementation generates in fp32 and casts, so we should
        # hit most representable values. The broken implementation produces
        # far fewer due to quantized Box-Muller inputs.
        if dtype == torch.bfloat16:
            self.assertGreater(n_unique, 100)
        else:
            self.assertGreater(n_unique, 50)

    @dtypes(torch.half, torch.bfloat16)
    def test_uniform_unique_values(self, device, dtype):
        size = 10000
        t = torch.empty(size, dtype=dtype, device=device).uniform_(-100, 100)
        n_unique = t.unique().numel()
        # With a correct implementation (generate in float, cast to dtype),
        # bf16 gives ~1400 unique and fp16 gives ~5200 for this range/size.
        # The broken implementation (generate directly in dtype) gives only
        # 256 for bf16 and ~2000 for fp16 due to using too few random bits.
        if dtype == torch.bfloat16:
            self.assertGreater(n_unique, 1000)
        else:
            self.assertGreater(n_unique, 4000)

    @skipIfNoSciPy
    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_lognormal_kstest(self, device, dtype):
        from scipy import stats
        size = 1000
        for mean in [-3, 0, 7]:
            for std in [1, 5, 7]:
                t = torch.empty(size, dtype=dtype, device=device).log_normal_(mean=mean, std=std)
                res = stats.kstest(t.cpu().to(torch.double), 'lognorm', args=(std, 0, math.exp(mean)))
                if dtype == torch.half:
                    self.assertTrue(res.statistic < 0.3)
                else:
                    self.assertTrue(res.statistic < 0.1)

    @skipIfNoSciPy
    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_exponential_kstest(self, device, dtype):
        from scipy import stats
        size = 1000
        for lambd in [0.5, 1.0, 5.0]:
            t = torch.empty(size, dtype=dtype, device=device).exponential_(lambd=lambd)
            res = stats.kstest(t.cpu().to(torch.double), 'expon', args=(0, 1 / lambd,))
            self.assertTrue(res.statistic < 0.1)

    @skipIfNoSciPy
    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_cauchy_kstest(self, device, dtype):
        from scipy import stats
        size = 1000
        for median in [-10, 0, 50]:
            for sigma in [0.5, 1.0, 10.0]:
                t = torch.empty(size, dtype=dtype, device=device).cauchy_(median=median, sigma=sigma)
                res = stats.kstest(t.cpu().to(torch.double), 'cauchy', args=(median, sigma))
                self.assertTrue(res.statistic < 0.1)

    @slowTest
    @onlyCUDA
    @dtypes(torch.bfloat16, torch.float32)
    def test_cauchy_no_inf(self, device, dtype):
        # torch.float16 will have `inf` because of its smaller range.
        for _ in range((2**16) * 2):
            x = torch.empty((2**16), dtype=dtype, device=device)
            x.cauchy_()
            self.assertFalse(x.isinf().sum())

    @dtypes(*floating_types_and(torch.half, torch.bfloat16))
    def test_cauchy(self, device, dtype):
        a = torch.tensor([10], dtype=dtype, device=device).cauchy_(0.0, 0.5)
        self.assertEqual(a.dtype, dtype)
        self.assertEqual(a.size(), torch.Size([1]))

        # Tests extremal behavior
        t = torch.empty((1,), device=device, dtype=dtype).cauchy_(float('inf'), 0.5)
        self.assertTrue(t.item() == float('inf'))

        # Tests non-positive rate fails
        with self.assertRaises(RuntimeError):
            torch.empty((1,), device=device, dtype=dtype).cauchy_(0.0, 0.0)

    @skipIfNoSciPy
    @skipIfTorchDynamo("FIXME: Skip due to dynamo failure on scipy.stats: https://github.com/pytorch/pytorch/issues/170509")
    @dtypes(*all_types_and(torch.half, torch.bfloat16))
    def test_geometric_kstest(self, device, dtype):
        from scipy import stats
        size = 1000
        for p in [0.2, 0.5, 0.8]:
            t = torch.empty(size, dtype=dtype, device=device).geometric_(p=p)
            actual = np.histogram(t.cpu().to(torch.double), np.arange(1, 100))[0]
            expected = stats.geom(p).pmf(np.arange(1, 99)) * size
            res = stats.chisquare(actual, expected)
            self.assertEqual(res.pvalue, 1.0, atol=0.1, rtol=0)

    # FIXME: find test suite for pdist and cdist
    def test_pairwise_distance_empty(self, device):
        shape = (2, 0)
        x = torch.randn(shape, device=device)
        y = torch.randn(shape, device=device)

        self.assertEqual(torch.zeros(2, device=device), torch.pairwise_distance(x, y))
        self.assertEqual(torch.zeros((2, 1), device=device), torch.pairwise_distance(x, y, keepdim=True))

        shape = (0, 2)
        x = torch.randn(shape, device=device)
        y = torch.randn(shape, device=device)
        self.assertEqual(torch.zeros(0, device=device), torch.pairwise_distance(x, y))
        self.assertEqual(torch.zeros((0, 1), device=device), torch.pairwise_distance(x, y, keepdim=True))

    def test_pdist_empty(self, device):
        shape = (0, 2)
        x = torch.randn(shape, device=device)
        self.assertEqual(torch.empty(0, device=device), torch.pdist(x))

        shape = (1, 2)
        x = torch.randn(shape, device=device)
        self.assertEqual(torch.empty(0, device=device), torch.pdist(x))

        shape = (3, 0)
        x = torch.randn(shape, device=device)
        self.assertEqual(torch.zeros(3, device=device), torch.pdist(x))

    def test_cdist_empty(self, device):
        x = torch.randn((0, 5), device=device)
        y = torch.randn((4, 5), device=device)
        self.assertEqual(torch.empty(0, 4, device=device), torch.cdist(x, y))

        x = torch.randn((2, 5), device=device)
        y = torch.randn((0, 5), device=device)
        self.assertEqual(torch.empty(2, 0, device=device), torch.cdist(x, y))

        x = torch.randn((2, 0), device=device)
        y = torch.randn((3, 0), device=device)
        self.assertEqual(torch.zeros(2, 3, device=device), torch.cdist(x, y))

        x = torch.randn((2, 0), device=device)
        y = torch.randn((0, 0), device=device)
        self.assertEqual(torch.empty(2, 0, device=device), torch.cdist(x, y))

    def _brute_cdist(self, x, y, p=2):
        r1 = x.shape[-2]
        r2 = y.shape[-2]
        if r1 == 0 or r2 == 0:
            return torch.empty(r1, r2, device=x.device)
        return torch.norm(x[..., None, :] - y[..., None, :, :], p=p, dim=-1)

    def test_cdist_norm(self, device):
        for r1 in [3, 4, 5, 6]:
            for m in [2, 3, 4, 10]:
                for r2 in [4, 6, 7, 8]:
                    for p in [0, 1, 2, 3, 1.5, 2.5, float('inf')]:
                        x = torch.randn(r1, m, device=device)
                        y = torch.randn(r2, m, device=device)
                        if p == 2:
                            for cm in ['use_mm_for_euclid_dist', 'donot_use_mm_for_euclid_dist']:
                                actual = torch.cdist(x, y, p=2, compute_mode=cm)
                                expected = self._brute_cdist(x, y, p=2)
                                self.assertEqual(expected, actual, rtol=0, atol=0.02)
                        else:
                            actual = torch.cdist(x, y, p=p)
                            expected = self._brute_cdist(x, y, p=p)
                            self.assertEqual(expected, actual)

    def test_cdist_norm_batch(self, device):
        for r1 in [3, 4, 5, 6]:
            for m in [2, 3, 4, 10]:
                for r2 in [4, 6, 7, 8]:
                    for p in [0, 1, 2, 3, 1.5, 2.5, float('inf')]:
                        x = torch.randn(2, 3, 6, r1, m, device=device)
                        y = torch.randn(2, 3, 6, r2, m, device=device)
                        if p == 2:
                            for cm in ['use_mm_for_euclid_dist', 'donot_use_mm_for_euclid_dist']:
                                actual = torch.cdist(x, y, p=2, compute_mode=cm)
                                expected = self._brute_cdist(x, y, p=2)
                                self.assertEqual(expected, actual, rtol=0, atol=0.02)
                        else:
                            actual = torch.cdist(x, y, p=p)
                            expected = self._brute_cdist(x, y, p=p)
                            self.assertEqual(expected, actual)

    @onlyCUDA
    def test_cdist_cuda_backward(self, device):
        for l1 in [1, 511, 513]:
            for l2 in [1, 511, 513]:
                for p in [0, 1, 2, 3, 1.5, 2.5, float('inf')]:
                    x1 = torch.randn(4, l1, 32, device=device, requires_grad=True)
                    x2 = x1.clone().detach_().requires_grad_()
                    y1 = torch.randn(4, l2, 32, device=device, requires_grad=True)
                    y2 = y1.clone().detach_().requires_grad_()
                    if p == 2:
                        for cm in ['use_mm_for_euclid_dist', 'donot_use_mm_for_euclid_dist']:
                            z1 = torch.cdist(x1, y1, p=2, compute_mode=cm).mean()
                            z2 = self._brute_cdist(x2, y2, p=2).mean()
                            z1.backward()
                            z2.backward()
                            self.assertEqual(x1.grad, x2.grad, rtol=0, atol=0.001)
                            self.assertEqual(y1.grad, y2.grad, rtol=0, atol=0.001)
                    else:
                        z1 = torch.cdist(x1, y1, p=p).mean()
                        z2 = self._brute_cdist(x2, y2, p=p).mean()
                        self.assertEqual(x1.grad, x2.grad, rtol=0, atol=0.001)
                        self.assertEqual(y1.grad, y2.grad, rtol=0, atol=0.001)

    # On ROCm, AMD XF32 round-down bias in the K=10 mm-based cdist path
    # (||a-b||² via a·b) pushes max_abs to ~5.8e-3, exceeding 0.005; ideal
    # NVIDIA TF32 stays under. See https://github.com/jeffdaily/tf32_analysis.
    @tf32_on_and_off(0.01 if TEST_WITH_ROCM else 0.005)
    @reduced_f32_on_and_off(0.08)
    def test_cdist_large(self, device):
        for cm in ['use_mm_for_euclid_dist_if_necessary', 'use_mm_for_euclid_dist', 'donot_use_mm_for_euclid_dist']:
            x = torch.randn(1000, 10, device=device)
            y = torch.randn(1000, 10, device=device)
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertEqual(expected, actual)

    @slowTest
    @tf32_on_and_off(0.01)
    @reduced_f32_on_and_off(0.08)
    def test_cdist_large_batch(self, device):
        for cm in ['use_mm_for_euclid_dist_if_necessary', 'use_mm_for_euclid_dist', 'donot_use_mm_for_euclid_dist']:
            x = torch.randn(4, 3, 1000, 10, device=device)
            y = torch.randn(4, 3, 1000, 10, device=device)
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertEqual(expected, actual)

    @tf32_on_and_off(0.005)
    @reduced_f32_on_and_off(0.04)
    def test_cdist_non_contiguous(self, device):
        for cm in ['use_mm_for_euclid_dist', 'donot_use_mm_for_euclid_dist']:
            x = torch.randn(5, 7, device=device).mT
            y = torch.randn(5, 3, device=device).mT
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertFalse(x.is_contiguous())
            self.assertFalse(y.is_contiguous())
            self.assertEqual(expected, actual)

            x = torch.randn(7, 5, device=device)
            y = torch.randn(5, 3, device=device).t()
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertTrue(x.is_contiguous())
            self.assertFalse(y.is_contiguous())
            self.assertEqual(expected, actual)

            x = torch.randn(5, 7, device=device).t()
            y = torch.randn(3, 5, device=device)
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertFalse(x.is_contiguous())
            self.assertTrue(y.is_contiguous())
            self.assertEqual(expected, actual)

    @tf32_on_and_off(0.005)
    @reduced_f32_on_and_off(0.04)
    def test_cdist_non_contiguous_batch(self, device):
        for cm in ['use_mm_for_euclid_dist', 'donot_use_mm_for_euclid_dist']:
            x = torch.randn(4, 3, 2, 5, 7, device=device).mT
            y = torch.randn(4, 3, 2, 5, 3, device=device).mT
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertFalse(x.is_contiguous())
            self.assertFalse(y.is_contiguous())
            self.assertEqual(expected, actual)

            x = torch.randn(7, 2, 7, 5, device=device)
            y = torch.randn(7, 2, 5, 3, device=device).mT
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertTrue(x.is_contiguous())
            self.assertFalse(y.is_contiguous())
            self.assertEqual(expected, actual)

            x = torch.randn(4, 5, 7, device=device).mT
            y = torch.randn(4, 3, 5, device=device)
            actual = torch.cdist(x, y, p=2, compute_mode=cm)
            expected = self._brute_cdist(x, y, p=2)
            self.assertFalse(x.is_contiguous())
            self.assertTrue(y.is_contiguous())
            self.assertEqual(expected, actual)

    # Maybe merge into OpInfo?
    def test_cdist_euclidean_large(self, device):
        def _test_euclidean_large_cdist(sizex, sizey=None):
            if sizey is None:
                sizey = sizex
            x = torch.randn(sizex, device=device, dtype=torch.float)
            y = torch.randn(sizey, device=device, dtype=torch.float)
            eps = 1e-6
            # to avoid extremum
            x = x - (((x - y) < eps).float() * 2 * eps)
            x.requires_grad = True
            y.requires_grad = True
            dist = torch.cdist(x, y, p=2)
            # Do a backward pass to check that it is valid for large
            # matrices
            loss = dist.sum()
            loss.backward()

        _test_euclidean_large_cdist((2000, 5))

    # Ensure that cdist backward with p<1 does not produce NaNs
    def test_cdist_grad_p_lt_1_no_nan(self, device):
        for p in [0.99, 0.7, 0.5, 0.1, 0.01]:
            x = torch.randn(1, 2, device=device)
            y = x.detach().clone() + torch.tensor([[1., 0.]], device=device)
            x.requires_grad = True
            y.requires_grad = True
            result = torch.cdist(x, y, p=p)
            result.backward(torch.ones_like(result))
            self.assertFalse(torch.isnan(x.grad).any())
            self.assertFalse(torch.isnan(y.grad).any())

    def test_cdist_same_inputs(self, device):
        # Test to detect issues in cdist gradient calculation
        # When the distances are 0
        sizex = (1, 27, 32)
        for p in [0, 1, 2, 3, 1.5, 2.5, float('inf')]:
            x = torch.randn(sizex, device=device, dtype=torch.float)
            dist_grad = torch.randn((1, 27, 27), device=device, dtype=torch.float)
            y = x.clone()
            x.requires_grad = True
            d = torch.cdist(x, y, p=p)
            d.backward(dist_grad)
            # Check that the backward pass does not contain invalid
            # values such as nan or inf
            if not torch.isfinite(x.grad).all():
                raise AssertionError("expected x.grad to be finite")

    def test_cumsum(self, device):
        x = torch.rand(100, 100, device=device)
        res1 = torch.cumsum(x, 1)
        res2 = torch.tensor([]).to(device)
        torch.cumsum(x, 1, out=res2)
        self.assertEqual(res1, res2)
        x.cumsum_(1)
        self.assertEqual(res1, x)

        a = torch.tensor([[True, False, True],
                          [False, False, False],
                          [True, True, True]], device=device)
        b = a.byte()
        aRes = torch.cumsum(a, 0)
        bRes = torch.cumsum(b, 0)
        self.assertEqual(aRes, bRes)
        self.assertEqual(aRes, torch.tensor([[1, 0, 1],
                                             [1, 0, 1],
                                             [2, 1, 2]]))

        aRes = torch.cumsum(a, 1)
        bRes = torch.cumsum(b, 1)
        self.assertEqual(aRes, bRes)
        self.assertEqual(aRes, torch.tensor([[1, 1, 2],
                                             [0, 0, 0],
                                             [1, 2, 3]]))

        # Check that cumulative sum over a zero length dimension doesn't crash on backprop.
        # Also check that cumsum over other dimensions in a tensor with a zero-length
        # dimensiuon also works
        # Also include a basic suite of similar tests for other bases cases.
        shapes = [[2, 0], [2, 1, 4], [0, 2, 3], [1], [5]]
        for shape in shapes:
            for dim in range(len(shape)):
                raw_tensor = torch.zeros(*shape, requires_grad=True)
                integrated = raw_tensor.cumsum(dim=dim)
                # Check that backward does not crash
                integrated.sum().backward()
                # Check that output maintained correct shape
                self.assertEqual(raw_tensor.shape, raw_tensor.grad.shape)

        # Check a scalar example
        raw_tensor = torch.tensor(3., requires_grad=True)
        integrated = raw_tensor.cumsum(dim=-1)
        self.assertEqual(raw_tensor, integrated)
        # Check that backward does not crash
        integrated.sum().backward()
        # Check that output maintained correct shape
        self.assertEqual(raw_tensor.shape, raw_tensor.grad.shape)

    def test_cumprod(self, device):
        x = torch.rand(100, 100, device=device)
        res1 = torch.cumprod(x, 1)
        res2 = torch.tensor([]).to(device)
        if not TEST_WITH_TORCHINDUCTOR:
            torch.cumprod(x, 1, out=res2)
            self.assertEqual(res1, res2)
        x.cumprod_(1)
        self.assertEqual(res1, x)

        a = torch.tensor([[True, False, True],
                          [False, False, False],
                          [True, True, True]], dtype=torch.bool, device=device)
        b = a.byte()
        aRes = torch.cumprod(a, 0)
        bRes = torch.cumprod(b, 0)
        self.assertEqual(aRes, bRes)
        self.assertEqual(aRes, torch.tensor([[1, 0, 1],
                                             [0, 0, 0],
                                             [0, 0, 0]]))

        aRes = torch.cumprod(a, 1)
        bRes = torch.cumprod(b, 1)
        self.assertEqual(aRes, bRes)
        self.assertEqual(aRes, torch.tensor([[1, 0, 0],
                                             [0, 0, 0],
                                             [1, 1, 1]]))

        # Check that cumulative prod over a zero length dimension doesn't crash on backprop.
        # Also check that cumprod over other dimensions in a tensor with a zero-length
        # dimensiuon also works
        # Also include a basic suite of similar tests for other bases cases.
        shapes = [[2, 0], [2, 1, 4], [0, 2, 3], [1], [5]]
        for shape in shapes:
            for dim in range(len(shape)):
                raw_tensor = torch.zeros(*shape, requires_grad=True)
                integrated = raw_tensor.cumprod(dim=dim)
                # Check that backward does not crash
                integrated.sum().backward()
                # Check that output maintained correct shape
                self.assertEqual(raw_tensor.shape, raw_tensor.grad.shape)

        # Check a scalar example
        raw_tensor = torch.tensor(3., requires_grad=True)
        integrated = raw_tensor.cumprod(dim=-1)
        self.assertEqual(raw_tensor, integrated)
        # Check that backward does not crash
        integrated.sum().backward()
        # Check that output maintained correct shape
        self.assertEqual(raw_tensor.shape, raw_tensor.grad.shape)

    def test_cummax_cummin(self, device):
        def test_ops(op, string_of_function_name, expected_output1, expected_output2):
            x = torch.rand(100, 100, device=device)
            out1 = op(x, 1)
            res2 = torch.empty(0, device=device)
            indices2 = torch.empty(0, dtype=torch.int64, device=device)
            op(x, 1, out=(res2, indices2))
            self.assertEqual(out1[0], res2)
            self.assertEqual(out1[1], indices2)

            a = torch.tensor([[True, False, True],
                              [False, False, False],
                              [True, True, True]], dtype=torch.bool, device=device)
            b = a.byte()
            aRes = op(a, 0)
            bRes = op(b, 0)
            self.assertEqual(aRes[0], bRes[0].bool())
            self.assertEqual(aRes[0], expected_output1.bool())

            # test inf and nan input
            x = torch.tensor([4, inf, 1.5, -inf, 0, nan, 1])
            xRes = op(x, 0)[0]
            self.assertEqual(xRes, expected_output2)

            # op shouldn't support values, indices with a dtype, device type or layout
            # different from that of input tensor
            t = torch.randn(10)
            values = torch.empty(0, dtype=torch.int16)
            indices = torch.empty(0, dtype=torch.int64)
            with self.assertRaisesRegex(
                    RuntimeError,
                    'expected scalar_type Float but found Short'):
                op(t, 0, out=(values, indices))

            # Range-check 0-d tensors
            x = torch.rand([])
            dim = 100
            with self.assertRaisesRegex(
                    IndexError,
                    'Expected reduction dim -1 or 0 for scalar but got 100'):
                op(x, dim)

            # Check that zero-size tensors with invalid dims raise IndexError
            # consistently, matching non-zero tensor behavior
            x = torch.tensor([], device=device)
            with self.assertRaisesRegex(
                    IndexError,
                    'Dimension out of range'):
                op(x, -46)

            # Check that op over a zero length dimension doesn't crash on backprop.
            # Also check that op over other dimensions in a tensor with a zero-length
            # dimension also works
            # Also include a basic suite of similar tests for other bases cases.
            shapes = [[2, 0], [2, 1, 4], [0, 2, 3], [1], [5]]
            for shape in shapes:
                for dim in range(len(shape)):
                    raw_tensor = torch.zeros(*shape, requires_grad=True)
                    integrated = getattr(raw_tensor, string_of_function_name)(dim=dim)
                    # Check that backward does not crash
                    integrated[0].sum().backward()
                    # Check that output maintained correct shape
                    self.assertEqual(raw_tensor.shape, raw_tensor.grad.shape)

            # Check a scalar example
            raw_tensor = torch.tensor(3., requires_grad=True)
            integrated = getattr(raw_tensor, string_of_function_name)(dim=-1)
            # Check that backward does not crash
            integrated[0].sum().backward()
            # Check that output maintained correct shape
            self.assertEqual(raw_tensor.shape, raw_tensor.grad.shape)

        expected_out = torch.tensor([4, inf, inf, inf, inf, nan, nan])
        test_ops(torch.cummax, "cummax", torch.tensor([[1, 0, 1],
                                                       [1, 0, 1],
                                                       [1, 1, 1]]), expected_out)

        expected_out = torch.tensor([4, 4, 1.5, -inf, -inf, nan, nan])
        test_ops(torch.cummin, "cummin", torch.tensor([[1, 0, 1],
                                                       [0, 0, 0],
                                                       [0, 0, 0]]), expected_out)

        for op in [torch.cummax, torch.cummin]:
            x = torch.randn(5, dtype=torch.complex64, device=device)
            with self.assertRaisesRegex(RuntimeError, "not implemented for 'ComplexFloat'"):
                op(x, 0)

    def test_logcumsumexp(self, device):
        def logcumsumexp(a, axis):
            return torch.cumsum(a.exp(), axis=axis).log_()

        axis = -1
        a = torch.randn(100, 100, device=device)

        actual = a.logcumsumexp(axis)
        expected = logcumsumexp(a, axis)
        self.assertEqual(a.dtype, actual.dtype)
        self.assertEqual(expected.shape, actual.shape)
        self.assertEqual(expected, actual)

        # check -inf and nan handling
        x = torch.tensor([-float('inf'), -float('inf'), 1.0, 1.0, float('inf'),
                         float('inf'), float('nan'), 1.0, 1.0], device=device)
        x2d = x.unsqueeze(0).expand(2, -1)

        for inp in (x, x2d):
            actual = inp.logcumsumexp(axis)
            expected = logcumsumexp(inp, axis)
            self.assertEqual(expected, actual)

        # Check that out is actually inplace
        b = torch.randn(5, 2, device=device)
        inplace_out = torch.zeros(5, 2, device=device)

        expected = logcumsumexp(b, axis)
        torch.logcumsumexp(b, axis=axis, out=inplace_out)

        self.assertEqual(inplace_out, expected)

        # Check input and inplace_output type mismatch
        b = torch.randn(5, 2, device=device, dtype=torch.float64)
        inplace_out = torch.zeros(5, 2, device=device, dtype=torch.float32)
        with self.assertRaisesRegex(
                RuntimeError,
                'expected scalar_type Double but found Float'):
            torch.logcumsumexp(b, axis, out=inplace_out)

    def _test_diff_numpy(self, t, dims=None):
        # Helper for test_diff to compare with NumPy reference implementation
        def to_np(t):
            if t.dtype == torch.bfloat16:
                return t.to(dtype=torch.float, device="cpu").numpy()
            else:
                return t.cpu().numpy()

        for dim in dims if dims else range(t.dim()):
            prepend = t.narrow(dim, 0, 1)
            append = t.narrow(dim, 0, 1)
            np_t = to_np(t)

            # test when no prepend and append
            for n in range(t.size(dim)):
                actual = torch.diff(t, dim=dim, n=n)
                expected = torch.from_numpy(np.diff(np_t, axis=dim, n=n))
                self.assertEqual(actual, expected.to(t.dtype))

            # test when prepend and append's size along dim is 1
            for n in range(1, t.size(dim) + 4):
                actual = torch.diff(t, dim=dim, n=n, prepend=prepend, append=append)
                expected = torch.from_numpy(np.diff(np_t, axis=dim, n=n, prepend=to_np(prepend), append=to_np(append)))
                self.assertEqual(actual, expected.to(t.dtype))

            # test when prepend and append's size along dim != 1
            for n in range(1, t.size(dim) * 3):
                actual = torch.diff(t, dim=dim, n=n, prepend=t, append=t)
                expected = torch.from_numpy(np.diff(np_t, axis=dim, n=n, prepend=np_t, append=np_t))
                self.assertEqual(actual, expected.to(t.dtype))

    # All tensors appear contiguous on XLA
    @onlyNativeDeviceTypes
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool))
    def test_diff_noncontig(self, device, dtype):
        shapes = (
            (1,),
            (1, 5),
            (3, 5),
            (1, 5, 1),
            (2, 3, 5))

        for shape in shapes:
            contig = make_tensor(shape, dtype=dtype, device=device, low=-9, high=9)

            non_contig = torch.empty(shape + (2, 2), device=device, dtype=dtype)[..., 0]
            non_contig = non_contig.select(-1, -1)
            non_contig.copy_(contig)
            self.assertTrue(not non_contig.is_contiguous() or shape == (1,))

            self._test_diff_numpy(non_contig)

    # RngNormal not implemented for type f16 for XLA
    @dtypes(*all_types_and_complex_and(torch.bool))
    @dtypesIfCPU(*all_types_and_complex_and(torch.half, torch.bool))
    @dtypesIfCUDA(*all_types_and_complex_and(torch.half, torch.bool))
    def test_diff(self, device, dtype):
        shapes = (
            (1,),
            (1, 5),
            (3, 5),
            (1, 5, 1),
            (2, 3, 5))

        for shape in shapes:
            contig = make_tensor(shape, dtype=dtype, device=device, low=-9, high=9)
            self._test_diff_numpy(contig)

        t = torch.ones(2, 3)

        with self.assertRaisesRegex(
                RuntimeError, 'diff expects prepend or append to be the same dimension as input'):
            invalid_prepend = torch.tensor([1, 2, 3], device=device, dtype=dtype)
            t.diff(dim=0, prepend=invalid_prepend)

        with self.assertRaisesRegex(
                RuntimeError, 'diff expects the shape of tensor to prepend or append to match that of input'):
            invalid_prepend = torch.tensor([[0, 1]], device=device, dtype=dtype)
            t.diff(dim=0, prepend=invalid_prepend)

        with self.assertRaisesRegex(
                RuntimeError, 'diff expects input to be at least one-dimensional'):
            scalar = torch.tensor(2, device=device, dtype=dtype)
            torch.diff(scalar)

    # if the given input arg is not a list, it returns a list of single element: [arg]
    def _wrap_to_list(self, input_array):
        return list(input_array) if isinstance(input_array, (list, tuple)) else [input_array]

    # To ensure inf, -inf, and nan values do not cause divergence between Numpy and PyTorch.
    # There are two types of possible divergence:
    # 1. When we compute a,b both real numbers and has very small absolute values (i.e. very near to 0.0)
    # then, result of a/b be inf, -inf and nan, and this cause divergence.
    # 2. When we are dividing complex numbers by zero. For example, when a = torch.tensor(3+5j) we have
    # a/0 to be equal to nan + nan*j in PyTorch and inf + inf*j in Numpy.
    def _inf_nan_preprocess(self, actual, expected):
        for i in range(len(expected)):
            expected[i] = np.nan_to_num(expected[i], nan=nan, posinf=nan, neginf=nan)
            # nan_to_num is not defined for complex tensors in PyTorch.
            if actual[i].dtype == torch.complex64 :
                actual[i].real = torch.nan_to_num(actual[i].real, nan=nan, posinf=nan, neginf=nan)
                actual[i].imag = torch.nan_to_num(actual[i].imag, nan=nan, posinf=nan, neginf=nan)
            else:
                actual[i] = torch.nan_to_num(actual[i], nan=nan, posinf=nan, neginf=nan)

        return actual, expected

    @onlyNativeDeviceTypes
    @dtypes(torch.long, torch.float32, torch.complex64)
    def test_gradient_all(self, device, dtype):
        def create_scalar(shape):
            return make_tensor((1,), device='cpu', dtype=dtype, low=1.).item()

        def create_list(shape):
            return make_tensor((len(shape),), device='cpu', dtype=dtype, low=1.).tolist()

        def create_coordinate_tensors(shape):
            tensor_list = []
            for i in range(len(shape)):
                tensor_list.append(make_tensor((shape[i],), device=device, dtype=dtype))
            return tensor_list

        def filter_shape(shape, dim):
            filtered_shape = []
            for i in range(len(dim)):
                filtered_shape.append(shape[dim[i]])
            return filtered_shape

        # shape, dims format
        test_cases = (
            ((5,), (0,)),
            ((4, 4), (0, 1)),
            ((3, 3, 3), (-1, 0)),
            ((4, 4, 4), (2,)),
            ((4, 4, 4), (0, 1)),
            ((4, 4, 4, 3), (0, 2, 3)),
            ((4, 5, 3, 4, 3), (1, 2)),
            ((4, 3, 6, 5, 3), (2, 4)),
            ((4, 3, 3, 5, 3), (0, 1, 2, 3, 4)),
            ((1, 3, 3), (1, 2)),
            ((1, 5), (1,)),
        )

        for case, contig, edge_order, space_fn in product(test_cases, [True, False], [1, 2],
                                                          (create_scalar, create_list, create_coordinate_tensors)):
            shape, dims = case
            # filter shape by dims before passing filtered shape to create_* functions
            filtered_shape = filter_shape(shape, dims)

            spacing = space_fn(filtered_shape)
            t = make_tensor(shape, device=device, dtype=dtype, noncontiguous=not contig)
            t_np = t.cpu().numpy()

            actual = torch.gradient(t, spacing=spacing, dim=dims, edge_order=edge_order)
            if space_fn is create_coordinate_tensors and spacing[0].device != 'cpu':
                spacing = [space.cpu().detach().numpy() for space in spacing]
            expected = np.gradient(t_np, *self._wrap_to_list(spacing), axis=dims, edge_order=edge_order)
            actual, expected = self._inf_nan_preprocess(list(actual), self._wrap_to_list(expected))
            self.assertEqual(actual, expected, equal_nan=True, atol=1e-4, rtol=0, exact_dtype=False)

    @onlyNativeDeviceTypes
    @slowTestIf(TEST_WITH_TORCHINDUCTOR)
    @dtypes(torch.long, torch.float32, torch.complex64)
    def test_gradient_extreme_cases(self, device, dtype):
        # Test behaviour for inf and nan values
        actual = torch.gradient(torch.tensor([2, -2, inf, inf, -inf, -inf, inf, 3, -inf, 2, nan, nan, 3, inf, nan]))
        expected = np.gradient(np.array([2, -2, inf, inf, -inf, -inf, inf, 3, -inf, 2, nan, nan, 3, inf, nan]))
        self.assertEqual(actual, self._wrap_to_list(expected), exact_dtype=False)

        # Test behaviour in very big tensors
        large_size = 100000
        t = make_tensor((large_size,), dtype=dtype, device=device)
        t_np = t.cpu().numpy()
        coordinates_np = np.random.randn(large_size)
        coordinates = [torch.tensor(coordinates_np, device=device)]
        actual = torch.gradient(t, spacing=coordinates, dim=0, edge_order=1)
        expected = [np.gradient(t_np, coordinates_np, axis=0, edge_order=1)]
        self.assertEqual(actual, expected, exact_dtype=False)

        actual = torch.gradient(t, spacing=coordinates, dim=0, edge_order=2)
        expected = [np.gradient(t_np, coordinates_np, axis=0, edge_order=2)]
        self.assertEqual(actual, expected, exact_dtype=False)

    @onlyNativeDeviceTypes
    def test_gradient_type_promotion(self, device):
        inputs = (
            make_tensor((4, 4), device=device, dtype=torch.float32),
            make_tensor((4, 4), device=device, dtype=torch.complex64),
            make_tensor((4, 4), device=device, dtype=torch.int64),
        )

        spacing = (
            make_tensor((1,), device='cpu', dtype=torch.float32).item(),
            make_tensor((1,), device='cpu', dtype=torch.int64).item(),
            make_tensor((1,), device='cpu', dtype=torch.complex64).item(),
            make_tensor((2,), device='cpu', dtype=torch.float32, low=0.1).tolist(),
            make_tensor((2,), device='cpu', dtype=torch.int64, low=1).tolist(),
            make_tensor((2,), device='cpu', dtype=torch.complex64).tolist(),
            [make_tensor((4,), device=device, dtype=torch.float32),
             make_tensor((4,), device=device, dtype=torch.float32)],
            [make_tensor((4,), device=device, dtype=torch.int64),
             make_tensor((4,), device=device, dtype=torch.int64)],
            [make_tensor((4,), device=device, dtype=torch.complex64),
             make_tensor((4,), device=device, dtype=torch.complex64)],
        )

        for input, spacing_or_coord, edge_order in product(inputs, spacing, [1, 2]):
            input_np = input.cpu().numpy()
            input_np = input.cpu().numpy()
            actual = torch.gradient(input, spacing=spacing_or_coord, dim=(0, 1), edge_order=edge_order)
            spacing_or_coord_wrapped = self._wrap_to_list(spacing_or_coord)
            spacing_or_coord_np = []
            if torch.is_tensor(spacing_or_coord_wrapped[0]) and torch.device(spacing_or_coord_wrapped[0].device).type != 'cpu':
                for i in range(len(spacing_or_coord_wrapped)):
                    spacing_or_coord_np.append(spacing_or_coord_wrapped[i].detach().clone().cpu().numpy())
            else:
                spacing_or_coord_np = spacing_or_coord_wrapped
            expected = np.gradient(input_np, *spacing_or_coord_np, axis=(0, 1), edge_order=edge_order)
            if actual[0].dtype == torch.complex64 and input.dtype != torch.complex64:
                for i in range(len(actual)):
                    self.assertEqual(actual[i].real, expected[i].real, exact_dtype=False)
                    # Type promotion fails on Numpy when spacing is given as complex number and input is given as real.
                    # Result is given just as real number and all the imaginary parts to be equal to zero.
                    self.assertEqual(expected[i].imag, torch.zeros(actual[i].shape), exact_dtype=False)
            else:
                actual, expected = self._inf_nan_preprocess(list(actual), list(expected))
                self.assertEqual(actual, expected, equal_nan=True, exact_dtype=False)

    @onlyNativeDeviceTypes
    @dtypes(torch.long, torch.float32, torch.complex64)
    def test_gradient_spacing_list_length_error(self, device, dtype):
        t = make_tensor((2, 2), device=device, dtype=dtype)

        spacing = (make_tensor((2,), device=device, dtype=dtype),)
        with self.assertRaisesRegex(RuntimeError, r'expected spacing to be'):
            torch.gradient(t, spacing=spacing)

        spacing = (make_tensor((2,), device=device, dtype=dtype),) * 2
        torch.gradient(t, spacing=spacing)

        spacing = (make_tensor((2,), device=device, dtype=dtype),) * 3
        with self.assertRaisesRegex(RuntimeError, r'expected spacing to be'):
            torch.gradient(t, spacing=spacing)

        spacing = (2,)
        with self.assertRaisesRegex(RuntimeError, r'expected spacing to be'):
            torch.gradient(t, spacing=spacing)

        spacing = (2, 2)
        torch.gradient(t, spacing=spacing)

        spacing = (2, 2, 2)
        with self.assertRaisesRegex(RuntimeError, r'expected spacing to be'):
            torch.gradient(t, spacing=spacing)

    def _test_large_cum_fn_helper(self, x, fn):
        expected = fn(x.cpu().float())
        actual = fn(x).cpu().float()
        # Avoid self.assertEqual to save memory.
        torch.testing.assert_close(expected, actual)

    @unittest.skipIf(IS_FBCODE and IS_REMOTE_GPU, "sandcastle OOM with current tpx gpu/re configuration")
    @unittest.skipIf(IS_JETSON, "psutil issue for largeTensorTest. Too large for Jetson.")
    @onlyCUDA
    @dtypes(torch.half)  # only small dtype not to get oom
    @largeTensorTest('25GB', device='cpu')
    @largeTensorTest('4GB', device='cuda')
    def test_large_cumsum(self, device, dtype):
        # initialization to avoid overflow and half caveats
        x = torch.empty(2**30 + 200, device=device, dtype=dtype)
        x[::3] = -3
        x[1::3] = 2
        x[2::3] = 1
        self._test_large_cum_fn_helper(x, lambda x: torch.cumsum(x, 0))

    @onlyCUDA
    @dtypes(torch.half)  # only small dtype not to get oom
    @largeTensorTest('25GB', device='cpu')
    @largeTensorTest('4GB', device='cuda')
    @unittest.skipIf(IS_JETSON, "psutil issue for largeTensorTest. Too large for Jetson.")
    def test_large_cumprod(self, device, dtype):
        # initialization to avoid overflow and half caveats
        x = torch.empty(2**30 + 200, device=device, dtype=dtype)
        x[::3] = 8
        x[1::3] = .25
        x[2::3] = .5
        self._test_large_cum_fn_helper(x, lambda x: torch.cumprod(x, 0))

    @skipIfTorchDynamo("Torchdynamo fails with unknown reason")
    def test_discontiguous_out_cumsum(self, device):
        x = torch.randn(4, 8, device=device)
        y = torch.empty(4, 16, device=device)[:, ::2]
        out = torch.cumsum(x, 0)
        torch.cumsum(x, 0, out=y)
        self.assertFalse(y.is_contiguous())
        self.assertEqual(out, y, atol=0., rtol=0.)

    def _test_cumminmax_helper(self, x, fn, expected_val, expected_ind):
        val, ind = fn(x, -1)
        self.assertEqual(val, expected_val, atol=0, rtol=0)
        self.assertEqual(ind, expected_ind, atol=0, rtol=0)
        out_val = torch.empty_like(val).t().contiguous().t()
        out_ind = torch.empty_like(ind).t().contiguous().t()
        fn(x, -1, out=(out_val, out_ind))
        # TODO: Fix this. It reproduces with aot_eager too, and looks like a functionalization bug.
        # (the problematic case seems rare, as we're calling an out= op directly from user code,
        # where the passed-in out tensors are non-contiguous).
        if not TEST_WITH_TORCHINDUCTOR:
            self.assertFalse(out_val.is_contiguous())
            self.assertFalse(out_ind.is_contiguous())
        self.assertEqual(out_val, expected_val, atol=0, rtol=0)
        self.assertEqual(out_ind, expected_ind, atol=0, rtol=0)

    def test_cummax_discontiguous(self, device):
        x = torch.tensor([[0, 1, 2, 3, 2, 1], [4, 5, 6, 5, 6, 7]], device=device, dtype=torch.float).t().contiguous().t()
        expected_val = torch.tensor([[0, 1, 2, 3, 3, 3], [4, 5, 6, 6, 6, 7]], device=device, dtype=torch.float)
        expected_ind = torch.tensor([[0, 1, 2, 3, 3, 3], [0, 1, 2, 2, 4, 5]], device=device, dtype=torch.long)
        self._test_cumminmax_helper(x, torch.cummax, expected_val, expected_ind)

    def test_cummin_discontiguous(self, device):
        x = torch.tensor([[3, 2, 1, 0, 1, 2], [7, 6, 5, 4, 5, 2]], device=device, dtype=torch.float).t().contiguous().t()
        expected_val = torch.tensor([[3, 2, 1, 0, 0, 0], [7, 6, 5, 4, 4, 2]], device=device, dtype=torch.float)
        expected_ind = torch.tensor([[0, 1, 2, 3, 3, 3], [0, 1, 2, 3, 3, 5]], device=device, dtype=torch.long)
        self._test_cumminmax_helper(x, torch.cummin, expected_val, expected_ind)

    def test_bool_tensor_value_change(self, device):
        x = torch.tensor([True, False], dtype=torch.bool, device=device)
        x[0] = False
        x[1] = True
        self.assertEqual(x, torch.tensor([False, True], dtype=torch.bool, device=device))

    # FIXME: move to data movement test suite
    def test_copy_all_dtypes_and_devices(self, device):
        from copy import copy
        for dt in all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16):
            x = torch.tensor([1, 2, 3, 4], dtype=dt, device=device)
            _x_clone = x.clone()
            y = copy(x)
            y.fill_(1)
            # copy is a shallow copy, only copies the tensor view,
            # not the data
            self.assertEqual(x, y)

    @onlyCPU
    def test_bfloat16_neg_abs(self, device):
        src = torch.randn(256)
        src[0] = torch.nan
        src[1] = -torch.nan
        src[2] = torch.inf
        src[3] = -torch.inf
        src_bf16 = src.bfloat16()
        self.assertEqual(src.neg().bfloat16(), src_bf16.neg())
        self.assertEqual(src.abs().bfloat16(), src_bf16.abs())

    @onlyCPU
    @dtypes(torch.bfloat16, torch.half)
    def test_reduced_type_float_copy(self, device, dtype):
        for shape in [(20, 7), (249, 137), (1029, 917), (1, 7, 19, 17), (3, 77, 1091)]:
            input = torch.randn(shape, dtype=torch.float, device=device)
            out1 = input.to(dtype=dtype)
            self.assertEqual(input, out1, atol=None, rtol=None, exact_dtype=False)
            out2 = out1.to(torch.float)
            self.assertEqual(out2, out1, atol=0, rtol=0, exact_dtype=False)

            input_s = input[..., ::2, :]
            out1 = input_s.to(dtype=dtype)
            self.assertEqual(input_s, out1, atol=None, rtol=None, exact_dtype=False)
            out2 = out1.to(torch.float)
            self.assertEqual(out2, out1, atol=0, rtol=0, exact_dtype=False)

    # FIXME: move to data movement test suite
    @onlyNativeDeviceTypes
    def test_copy_math_view(self, device):
        for dst_dtype, src_dtype in [
                (torch.float32, torch.float32),
                (torch.float64, torch.float32),
                (torch.int64, torch.int32),
                (torch.complex128, torch.complex64),
        ]:
            src = make_tensor((100,), dtype=src_dtype, device=device)
            dst = torch.empty(100, dtype=dst_dtype, device=device)

            dst.copy_(src)
            self.assertEqual(dst, src, exact_dtype=False)

            dst.copy_(src._neg_view())
            self.assertEqual(dst, src.neg(), exact_dtype=False)

            dst._neg_view().copy_(torch._neg_view(src))
            self.assertEqual(dst, src, exact_dtype=False)

            dst._neg_view().copy_(src)
            self.assertEqual(dst, src.neg(), exact_dtype=False)

            # issue: https://github.com/pytorch/pytorch/issues/106051
            dst._neg_view().copy_(dst)
            self.assertEqual(dst, src, exact_dtype=False)

        for dst_dtype, src_dtype in [
                (torch.complex64, torch.complex64),
                (torch.complex128, torch.complex64),
        ]:
            src = make_tensor((100,), dtype=src_dtype, device=device)
            dst = torch.empty(100, dtype=dst_dtype, device=device)

            dst.conj().copy_(src)
            self.assertEqual(dst, src.conj_physical(), exact_dtype=False)

            dst.conj().copy_(src._neg_view())
            self.assertEqual(dst, src.neg().conj_physical(), exact_dtype=False)

    # FIXME: move to data movement test suite
    @onlyNativeDeviceTypes
    @dtypes(torch.int64, torch.float32, torch.complex64)
    def test_copy_transpose_math_view(self, device, dtype):
        src = make_tensor((100, 100), dtype=dtype, device=device).transpose(0, 1)
        dst = torch.empty((100, 100), dtype=dtype, device=device)

        dst._neg_view().copy_(src)
        self.assertEqual(dst, -src)
        dst._neg_view().copy_(src._neg_view())
        self.assertEqual(dst, src)
        dst.copy_(src._neg_view())
        self.assertEqual(dst, -src)

        if dtype.is_complex:
            dst.conj().copy_(src)
            self.assertEqual(dst, src.conj_physical())
            dst.conj().copy_(src.conj())
            self.assertEqual(dst, src)
            dst.copy_(src.conj())
            self.assertEqual(dst, src.conj_physical())

    def test_clone_all_dtypes_and_devices(self, device):
        for dt in all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16):
            x = torch.tensor((1, 1), dtype=dt, device=device)
            y = x.clone()
            self.assertEqual(x, y)

    def test_clone_zero_stride_dim(self, device):
        # stride zero, size 1 axis, not contiguous
        x = torch.randn(10)
        y = x.as_strided([2, 1, 5], [1, 0, 2])
        self.assertEqual(y, y.clone())

    def test_clone_not_memory_dense(self):
        # github issue: https://github.com/pytorch/pytorch/issues/64176
        x = torch.randn(10, 8).t()[::2, ::2]
        y = x.clone()
        # should retain permutation after densification
        self.assertTrue(y.stride() == (1, 4))

    # FIXME: move to elementwise ternary test suite
    @parametrize("use_cpu_scalar", [True, False])
    @dtypesIfCUDA(*set(get_all_math_dtypes('cuda')))
    @dtypes(*set(get_all_math_dtypes('cpu')))
    def test_addcmul(self, device, dtype, use_cpu_scalar):
        # Returns floating or integral scalar corresponding to dtype
        def _number(floating, integer, dtype):
            if dtype in [torch.half, torch.float, torch.double, torch.bfloat16]:
                return floating
            elif dtype in [torch.cfloat, torch.cdouble]:
                return floating * (1 + 1j)
            else:
                return integer

        def rand_tensor(size, dtype, device):
            if dtype.is_floating_point or dtype.is_complex:
                return torch.rand(size=size, dtype=dtype, device=device)
            if dtype == torch.uint8:
                return torch.randint(1, 5, size=size, dtype=dtype, device=device)
            else:
                return torch.randint(-5, 5, size=size, dtype=dtype, device=device)

        a = rand_tensor((2, 2), dtype=dtype, device=device)
        b = rand_tensor((2, 2), dtype=dtype, device=device)
        if use_cpu_scalar:
            c = rand_tensor([], device="cpu", dtype=dtype)
        else:
            c = rand_tensor((2, 2), dtype=dtype, device=device)

        alpha = _number(0.5, 3, dtype)

        actual = torch.addcmul(a, b, c, value=alpha)
        expected = a + alpha * b * c

        self.assertEqual(expected, actual)

        with self.assertWarnsOnceRegex(
                UserWarning, "This overload of addcmul is deprecated"):
            self.assertEqual(actual, torch.addcmul(a, alpha, b, c))

        if self.device_type == 'cuda' and dtype == torch.half:
            a = torch.tensor([60000.0], device=device, dtype=dtype)
            b = torch.tensor([60000.0], device=device, dtype=dtype)
            c = torch.tensor([2.0], device=device, dtype=dtype)
            out = torch.addcmul(a, b, c, value=-1)
            self.assertTrue(not (out.isnan() or out.isinf()))

    @onlyCUDA
    def test_addcmul_cuda_errors_with_cpu_scalars(self, device):
        # Logic is dtype agnostic, so dtype isn't tested
        alpha = 0.5

        a = torch.rand((2, 2), device=device)
        b = torch.rand((2, 2), device=device)
        c = torch.rand((2, 2), device=device)
        scalar = torch.rand([], device="cpu")

        with self.assertRaisesRegex(RuntimeError, r'CPU Scalar support for tensor1 argument'):
            torch.addcmul(a, scalar, c, value=alpha)
        with self.assertRaisesRegex(RuntimeError, r'CPU Scalar support for self argument'):
            torch.addcmul(scalar, b, c, value=alpha)

    # FIXME: move to shape ops test suite
    def test_narrow_empty(self, device):
        x = torch.randn(2, 3, 4, device=device)
        for d in range(x.dim()):
            y = x.narrow(d, x.size(d), 0)
            sz = list(x.size())
            sz[d] = 0
            self.assertEqual(sz, y.size())

    def test_narrow_copy_non_contiguous(self, device):
        # see https://github.com/pytorch/pytorch/issues/91690.
        inp = torch.randn(10, 2, device=device).movedim(-1, 0)
        expected = torch.narrow_copy(inp.contiguous(), 1, 0, 10)
        actual = torch.narrow_copy(inp, 1, 0, 10)
        self.assertEqual(expected, actual)

    # FIXME: find a test suite for the take operator
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @slowTestIf(IS_WINDOWS)
    def test_take(self, device, dtype):
        idx_size = (4,)

        make_arg = partial(make_tensor, device=device, dtype=dtype)
        make_idx = partial(make_tensor, low=0, device=device, dtype=torch.int64)

        def ref_take(src, idx):
            if dtype == torch.bfloat16:
                src = src.half()
            src = src.cpu().numpy()
            idx = idx.cpu().numpy()
            out = torch.from_numpy(np.take(src, idx)).to(device=device, dtype=dtype)
            return out

        for src_contig, idx_contig, idx_reshape in product([True, False], repeat=3):
            for src_size in ((5,), (4, 5)):
                src = make_arg(src_size, noncontiguous=not src_contig)
                idx = make_idx(idx_size, high=src.numel(), noncontiguous=not idx_contig)
                if idx_reshape:
                    idx = idx.reshape(2, 2)
                out = torch.take(src, idx)
                out2 = ref_take(src, idx)
                self.assertEqual(out, out2)

        # Create the 4 possible combinations of scalar sizes for source / index
        for size_s, size_i in product([(), (1,)], repeat=2):
            source = make_arg(size_s)
            idx = make_idx(size_i, high=1)
            out = source.take(idx)
            self.assertEqual(out.item(), source.item())

    # FIXME: find a test suite for the put operator
    # The bool instance does not work on GPU. See
    # https://github.com/pytorch/pytorch/issues/54317
    @dtypes(*all_types_and_complex_and(torch.half, torch.bfloat16))
    def test_put(self, device, dtype):
        src_size = (4,)

        make_arg = partial(make_tensor, device=device, dtype=dtype)
        make_idx = partial(make_tensor, low=0, device=device, dtype=torch.int64)

        def ref_put(dst, idx, src, accumulate):
            new_dst = dst.clone(memory_format=torch.contiguous_format).view(-1)
            new_idx = idx.contiguous().view(-1)
            new_src = src.contiguous().view(-1)
            method = new_dst.index_add_ if accumulate else new_dst.index_copy_
            return method(0, new_idx, new_src).view_as(dst)

        for dst_contig, src_contig, idx_contig, idx_reshape, accumulate in product([True, False], repeat=5):
            for dst_size in ((5,), (4, 5)):
                dst = make_arg(dst_size, noncontiguous=not dst_contig)
                src = make_arg(src_size, noncontiguous=not src_contig)

                # If accumulate=True, `put_` should be deterministic regardless of the inputs on CPU
                # On CUDA it may not be, but the test has enough tolerance to account for this
                if accumulate:
                    idx = make_idx(src_size, high=dst.numel())
                else:
                    idx = torch.randperm(dst.numel(), dtype=torch.int64, device=device)[:src_size[0]]
                if not idx_contig:
                    idx = torch.repeat_interleave(idx, 2, dim=-1)[..., ::2]
                if idx_reshape:
                    idx = idx.reshape(2, 2)
                out = torch.put(dst, idx, src, accumulate)
                # out-place
                reference = ref_put(dst, idx, src, accumulate)
                self.assertEqual(out, reference)

                # in-place
                dst.put_(idx, src, accumulate)
                self.assertEqual(dst, reference)


        # Create the 8 possible combinations of scalar sizes for target / index / source
        scalars = ((make_arg(size_t),
                    make_idx(size_i, high=1),
                    make_arg(size_s))
                   for size_t, size_i, size_s in product([(), (1,)], repeat=3))
        for (dest, idx, source), accumulate in product(scalars, [True, False]):
            dest_init = dest.clone()
            # out-place
            out = torch.put(dest, idx, source, accumulate=accumulate)
            # in-place
            dest1 = dest.clone()
            dest1.put_(idx, source, accumulate=accumulate)
            for d in [out, dest1]:
                if accumulate:
                    self.assertEqual(d.item(), (dest_init + source).item())
                else:
                    self.assertEqual(d.item(), source.item())

        # Empty case
        dest = make_arg((3, 2))
        reference = dest.clone()
        idx = make_idx((0,), high=1)
        source = make_arg((0,))
        for accumulate in [True, False]:
            out = torch.put(dest, idx, source, accumulate=accumulate)
            self.assertEqual(out, reference)
            dest.put_(idx, source, accumulate=accumulate)
            self.assertEqual(dest, reference)

    # FIXME: find a test suite for the put operator
    # The bool instance does not work on GPU. See
    # https://github.com/pytorch/pytorch/issues/54317
    @dtypes(*all_types_and_complex_and(torch.half, torch.bfloat16))
    def test_put_accumulate(self, device, dtype):
        # Test for parallel adds with accumulate == True
        low_precision = dtype == torch.half or dtype == torch.bfloat16
        # Less numbers to avoid overflow with low_precision
        # Grainsize is 3000 for the for_loop to be parallelized on CPU
        sizes = ((100,)) if low_precision else ((200,), (3002,))
        # Bfloat16 has a particularly bad performance here
        # This operation is nondeterministic on GPU, so we are generous with the rtol
        rtol, atol = (1e-1, 1e-2) if low_precision else (1e-3, 1e-4)

        make_arg = partial(make_tensor, low=-2, high=3, device=device, dtype=dtype)
        # Dump everything into the 0-th position
        make_idx = partial(torch.zeros, device=device, dtype=torch.int64)
        args = ((make_idx(size), make_arg(size)) for size in sizes)

        for idx, source in args:
            orig = make_arg((1,))
            out = orig.put(idx, source, accumulate=True)
            self.assertEqual(out, orig + source.sum(), rtol=rtol, atol=atol)

    # FIXME: find a test suite for the take operator
    def test_take_empty(self, device):
        for input_shape in [(0,), (0, 1, 2, 0), (1, 2, 3)]:
            for indices_shape in [(0,), (0, 1, 2, 0)]:
                input = torch.empty(input_shape, device=device)
                indices = torch.empty(indices_shape, dtype=torch.int64, device=device)
                self.assertEqual(indices, torch.take(input, indices), exact_dtype=False)

    # FIXME: find a test suite for the put operator
    def test_put_empty(self, device):
        for dst_shape in [(0,), (0, 1, 2, 0), (1, 2, 3)]:
            for indices_shape in [(0,), (0, 1, 2, 0)]:
                for accumulate in [False, True]:
                    dst = torch.randn(dst_shape, device=device)
                    indices = torch.empty(indices_shape, dtype=torch.int64, device=device)
                    src = torch.randn(indices_shape, device=device)
                    self.assertEqual(dst, dst.put_(indices, src, accumulate=accumulate))

    # FIXME: port to test_scatter_gather_ops.py
    def scatter_allow_reduce(self, device, dtype, reduceop):
        device_type = torch.device(device).type
        return device_type != 'cuda' or (reduceop == 'multiply' and dtype.is_floating_point)

    @dtypes(*floating_and_complex_types())
    @dtypesIfCPU(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @dtypesIfCUDA(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_scatter_reduce_operations_to_large_input(self, device, dtype):
        index = torch.tensor([[1], [2]], device=device, dtype=torch.long)
        test_data = [
            (torch.zeros(4, 4, device=device, dtype=dtype),
             torch.ones(2, 2, device=device, dtype=dtype),
             torch.tensor([[0, 0, 0, 0],
                           [1, 0, 0, 0],
                           [1, 0, 0, 0],
                           [0, 0, 0, 0]],
                          device=device, dtype=dtype), "add"),
            (torch.tensor([2], device=device, dtype=dtype).repeat(4, 4),
             torch.tensor([6], device=device, dtype=dtype).repeat(2, 2),
             torch.tensor([[2, 2, 2, 2],
                           [12, 2, 2, 2],
                           [12, 2, 2, 2],
                           [2, 2, 2, 2]], device=device, dtype=dtype), "multiply"),
        ]

        for input, src, result, operation in test_data:
            if not self.scatter_allow_reduce(device, dtype, operation):
                continue
            input.scatter_(0, index, src, reduce=operation)
            self.assertEqual(input, result)

    @dtypes(*floating_and_complex_types())
    @dtypesIfCPU(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @dtypesIfCUDA(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_scatter_reduce_scalar(self, device, dtype):
        index = torch.tensor([[1], [2]], device=device, dtype=torch.long)
        test_data = [
            (torch.zeros(4, 4, device=device, dtype=dtype), 1,
             torch.tensor([[0, 0, 0, 0],
                           [1, 0, 0, 0],
                           [1, 0, 0, 0],
                           [0, 0, 0, 0]],
                          device=device, dtype=dtype), "add"),
            (torch.tensor([2], device=device, dtype=dtype).repeat(4, 4), 2,
             torch.tensor([[2, 2, 2, 2],
                           [4, 2, 2, 2],
                           [4, 2, 2, 2],
                           [2, 2, 2, 2]], device=device, dtype=dtype), "multiply"),
        ]

        for input, src, result, operation in test_data:
            if not self.scatter_allow_reduce(device, dtype, operation):
                continue
            input.scatter_(0, index, src, reduce=operation)
            self.assertEqual(input, result)

    # FIXME: port to test_scatter_gather_ops.py
    # TODO: remove this after scatter_add_ is deprecated.
    def test_scatter_add_non_unique_index(self, device):
        height = 2
        width = 65536
        input = torch.ones(height, width, device=device)
        index = torch.zeros(height, width, dtype=torch.long, device=device)
        src = torch.ones(height, width, device=device)
        input.scatter_add_(0, index, src)

        self.assertEqual(input,
                         torch.tensor([[3], [1]], device=device,
                                      dtype=torch.float32).repeat(1, width))

    @dtypes(*floating_and_complex_types())
    @dtypesIfCPU(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    @dtypesIfCUDA(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_scatter_reduce_non_unique_index(self, device, dtype):
        height = 2
        width = 2
        index = torch.zeros(height, width, dtype=torch.long, device=device)
        test_data = [
            (torch.ones(height, width, device=device, dtype=dtype),
             torch.ones(height, width, device=device, dtype=dtype),
             torch.tensor([[3], [1]], device=device, dtype=dtype).repeat(1, width), "add"),
            (torch.tensor([2], device=device, dtype=dtype).repeat(height, width),
             torch.tensor([2], device=device, dtype=dtype).repeat(height, width),
             torch.tensor([[8], [2]], device=device,
                          dtype=dtype).repeat(1, width), "multiply"),
        ]

        for input, src, result, operation in test_data:
            if not self.scatter_allow_reduce(device, dtype, operation):
                continue
            input.scatter_(0, index, src, reduce=operation)
            self.assertEqual(input, result, msg=lambda msg: f"{msg}\nresult: {result} input: {input} method: {str(operation)}")

    @onlyCUDA
    @dtypes(*complex_types())
    def test_scatter_reduce_multiply_unsupported_dtypes(self, device, dtype):
        height = 2
        width = 2
        index = torch.zeros(height, width, dtype=torch.long, device=device)
        input = torch.ones(height, width, device=device, dtype=dtype)
        src = torch.ones(height, width, device=device, dtype=dtype)
        with self.assertRaises(RuntimeError):
            input.scatter_(0, index, src, reduce="multiply")

    # FIXME: port to test_scatter_gather_ops.py
    def test_scatter_to_large_input(self, device):
        input = torch.zeros(4, 4, device=device)
        src = torch.ones(2, 2, device=device)
        index = torch.tensor([[1], [2]], device=device, dtype=torch.long)
        input.scatter_(0, index, src)
        self.assertEqual(input, torch.tensor([[0, 0, 0, 0],
                                              [1, 0, 0, 0],
                                              [1, 0, 0, 0],
                                              [0, 0, 0, 0]], device=device, dtype=torch.float32))

    # FIXME: port to test_scatter_gather_ops.py
    def test_scatter_add_to_large_input(self, device):
        input = torch.zeros(4, 4, device=device)
        src = torch.ones(2, 2, device=device)
        index = torch.tensor([[1], [2]], device=device, dtype=torch.long)
        input.scatter_add_(0, index, src)
        self.assertEqual(input, torch.tensor([[0, 0, 0, 0],
                                              [1, 0, 0, 0],
                                              [1, 0, 0, 0],
                                              [0, 0, 0, 0]], device=device, dtype=torch.float32))

    # FIXME: port to test_scatter_gather_ops.py
    def test_scatter_bool(self, device):
        x = torch.tensor([[True, True, True], [True, True, True]], device=device)
        res = torch.zeros(3, 3, dtype=torch.bool, device=device)
        res = res.scatter_(0, torch.tensor([[0, 1, 2], [0, 1, 2]], device=device), x)
        self.assertEqual(res, torch.tensor([[True, False, False],
                                            [False, True, False],
                                            [False, False, True]], device=device))

    # FIXME: port to test_scatter_gather_ops.py
    def test_scatter_add_bool(self, device):
        x = torch.tensor([[True, True, True, True, True], [True, True, True, True, True]], device=device)
        res = torch.zeros(3, 5, dtype=torch.bool, device=device)
        res = res.scatter_add_(0, torch.tensor([[0, 1, 2, 0, 0], [2, 0, 0, 1, 2]], device=device), x)
        self.assertEqual(res, torch.tensor([[True, True, True, True, True],
                                            [False, True, False, True, False],
                                            [True, False, True, False, True]], device=device))

    # FIXME: find a test suite for the masked scatter operator
    @onlyNativeDeviceTypes
    @dtypes(*all_types_and_complex_and(torch.half, torch.bfloat16))
    def test_masked_scatter(self, device, dtype):
        dt = dtype
        num_copy, num_dest = 3, 10
        dest = torch.tensor([1, 2, 3, 4, 5, 6, 7, 8, 9, 10], dtype=dt, device=device)
        dest2 = dest.clone()
        dest_ones = dest.clone()
        dest_ones_expected = dest.clone()
        src = torch.tensor([0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=dt, device=device)
        src_ones = torch.tensor([1, 1, 1, 1, 1, 1, 1, 1, 1, 1], dtype=dt, device=device)
        mask = torch.tensor((0, 0, 0, 0, 1, 0, 1, 0, 1, 0), dtype=torch.bool, device=device)

        dest.masked_scatter_(mask, src)
        j = 0
        for i in range(num_dest):
            if mask[i]:
                dest2[i] = src[j]
                dest_ones_expected[i] = src_ones[j]
                j += 1
        self.assertEqual(dest, dest2, atol=0, rtol=0)

        dest_ones.masked_scatter_(mask, src_ones)
        self.assertEqual(dest_ones, dest_ones_expected, atol=0, rtol=0)

        # Bound checking in CUDA is done inside a kernel
        # in order to avoid synchronization, but this means
        # we can not clear the failures. So there is no way
        # to test it then recover.
        if self.device_type != 'cuda':
            # make src smaller. this should fail
            src = torch.zeros(num_copy - 1, dtype=dt, device=device)
            with self.assertRaises(RuntimeError):
                dest.masked_scatter_(mask, src)

        # empty tensor
        dest = torch.empty((5, 0, 5), dtype=dt, device=device)
        mask = torch.ones_like(dest, dtype=torch.bool, device=device)
        src = torch.empty((0,), dtype=dt, device=device)
        dest.masked_scatter_(mask, src)

        dest = torch.empty((5, 0, 5), dtype=dt, device=device)
        mask = torch.ones((5, 1, 5), dtype=torch.bool, device=device)
        src = torch.empty((0,), dtype=dt, device=device)
        dest.masked_scatter_(mask, src)

    # FIXME: find a test suite for the masked scatter operator
    def test_masked_scatter_bool_tensor(self, device):
        src = torch.tensor([True, True, True], device=device)
        dst = torch.tensor([False, False, False], device=device)
        mask = torch.tensor([False, True, False], device=device)

        dst.masked_scatter_(mask, src)
        self.assertEqual(dst, torch.tensor([False, True, False], device=device))

        mask = torch.tensor([True, False, True], device=device)
        dst = dst.masked_scatter(mask, src)
        self.assertEqual(dst, torch.tensor([True, True, True], device=device))

    # FIXME: find a test suite for the masked scatter operator
    #   test_scatter_gather_ops or test_masked_ops?
    @onlyCUDA
    @largeTensorTest('30GB')
    def test_masked_scatter_large_tensor(self, device):
        t_cpu = torch.empty(2**31 + 1, dtype=torch.bool).random_()
        t = t_cpu.to(device)
        result_cpu = t_cpu.masked_scatter(t_cpu, t_cpu)
        result = t.masked_scatter(t, t)
        self.assertEqual(result, result_cpu)

    # FIXME: find a test suite for the masked select operator
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_masked_select(self, device, dtype):
        for maskType in integral_types_and(torch.bool):
            num_src = 10
            src = torch.tensor([0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=dtype, device=device)
            mask = torch.randint(2, (num_src,), device=device, dtype=maskType)

            if maskType is not torch.bool:
                with self.assertRaisesRegex(RuntimeError, r'expected BoolTensor for mask'):
                    dst = src.masked_select(mask)
                continue
            else:
                dst = src.masked_select(mask)
            dst2 = []
            for i in range(num_src):
                if mask[i]:
                    dst2 += [src[i]]
            self.assertEqual(dst, torch.tensor(dst2), atol=0, rtol=0)

            dst3 = torch.empty(0, device=device, dtype=dtype)
            torch.masked_select(src, mask, out=dst3)
            self.assertEqual(dst3, torch.tensor(dst2, dtype=dst3.dtype), atol=0, rtol=0)

        # Ensure that masks are expanded to match tensor properly
        a = torch.rand(100, 100, device=device).mul(100).to(dtype)
        mask_first_el_each_row = torch.zeros(100, device=device, dtype=torch.bool)
        mask_first_el_each_row[0] = True
        a_masked = a.masked_select(mask_first_el_each_row)
        self.assertEqual(a_masked, a[:, 0])

        mask_first_row = torch.zeros(100, 1, device=device, dtype=torch.bool)
        mask_first_row[0][0] = True
        a_masked = a.masked_select(mask_first_row)
        self.assertEqual(a_masked, a[0, :])

        # Ensure that tensor is expanded to match mask properly
        a = torch.rand(100, device=device).mul(100).to(dtype)
        mask_copy_3_times = torch.tensor([[True], [True], [False], [True]], device=device)
        a_masked = a.masked_select(mask_copy_3_times)
        self.assertEqual(a_masked, a.unsqueeze(0).expand(3, 100).flatten())

    # FIXME: find a test suite for the masked select operator
    def test_masked_select_discontiguous(self, device):
        for size in (10, 200):
            vals = torch.rand(size, size, device=device)
            mask = torch.full((size, size), False, dtype=torch.bool, device=device)
            mask[:, ::2] = True
            vals_list = (vals, vals.t())
            mask_list = (mask, mask.t())
            out_dc = torch.empty(size * size, device=device)[::2]
            for v, m in product(vals_list, mask_list):
                if m.is_contiguous():
                    expected = v[:, ::2].clone().reshape((-1, ))
                else:
                    expected = v[::2].clone().reshape((-1, ))
                out = torch.masked_select(v, m)
                self.assertEqual(out, expected, atol=0, rtol=0)
                torch.masked_select(v, m, out=out_dc)
                self.assertEqual(out_dc, expected, atol=0, rtol=0)

    # FIXME: find a test suite for the masked fill operator
    @dtypes(*product(all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16), (torch.uint8, torch.bool)))
    def test_masked_fill(self, device, dtypes):
        dtype = dtypes[0]
        mask_dtype = dtypes[1]

        num_dest = 10
        dst = torch.zeros(num_dest, dtype=dtype)
        mask = torch.randint(2, (num_dest,), dtype=mask_dtype)
        val = random.random()
        dst2 = dst.clone()

        if mask_dtype is not torch.bool:
            with self.assertRaisesRegex(RuntimeError, 'only supports boolean masks'):
                dst.masked_fill_(mask, val)
            return

        dst.masked_fill_(mask, val)
        for i in range(num_dest):
            if mask[i]:
                dst2[i] = val
        self.assertEqual(dst, dst2, atol=0, rtol=0)

        # test non-contiguous case
        dst = make_tensor(
            (num_dest, num_dest, num_dest), dtype=dtype, device=device, low=-30, high=30
        ).permute((2, 0, 1))
        dst2 = dst.contiguous()
        if dtype.is_complex:
            mask = dst.abs() > 0
        else:
            mask = dst > 0
        self.assertTrue(not dst.is_contiguous())
        self.assertTrue(dst2.is_contiguous())
        dst.masked_fill_(mask.to(mask_dtype), val)
        dst2.masked_fill_(mask.to(mask_dtype), val)
        self.assertEqual(dst, dst2, atol=0, rtol=0)

    # FIXME: find a test suite for the masked fill operator
    def test_masked_fill_bool_tensor(self, device):
        dst = torch.tensor([True, False, True], device=device)
        mask = torch.tensor([False, True, False], device=device)

        dst.masked_fill_(mask, True)
        self.assertEqual(dst, torch.tensor([True, True, True], device=device))

        dst = dst.masked_fill(mask, False)
        self.assertEqual(dst, torch.tensor([True, False, True], device=device))

    def test_tensor_shape_empty(self, device):
        x = torch.randn((0, 1, 3, 0), device=device)
        # flatten
        self.assertEqual((0,), torch.flatten(x, 0, 3).shape)
        self.assertEqual((0, 0), torch.flatten(x, 0, 2).shape)
        self.assertEqual((0, 3, 0), torch.flatten(x, 1, 2).shape)

        # squeeze, unsqueeze
        self.assertEqual((0, 1, 1, 3, 0), torch.unsqueeze(x, 1).shape)
        self.assertEqual((0, 3, 0), torch.squeeze(x, 1).shape)
        self.assertEqual((0, 3, 0), torch.squeeze(x).shape)

        # transpose, t
        self.assertEqual((0, 0, 3, 1), torch.transpose(x, 1, 3).shape)
        y = torch.randn((5, 0), device=device)
        self.assertEqual((0, 5), y.t().shape)

        # select
        self.assertEqual((0, 1, 0), torch.select(x, 2, 2).shape)

        # repeat, permute
        self.assertEqual((9, 0, 5, 6, 0), x.repeat(9, 7, 5, 2, 3).shape)
        self.assertEqual((3, 0, 0, 1), x.permute(2, 3, 0, 1).shape)

        # diagonal, diagflat
        self.assertEqual((0,), torch.diagonal(torch.randn((5, 0), device=device)).shape)
        self.assertEqual((0,), torch.diagonal(torch.randn((0, 5), device=device)).shape)
        # off the end offsets are valid
        self.assertEqual((0,), torch.diagonal(torch.randn((5, 0), device=device), offset=1).shape)
        self.assertEqual((0,), torch.diagonal(torch.randn((0, 5), device=device), offset=1).shape)
        # check non-zero sized offsets off the end
        self.assertEqual((5, 6, 0), torch.diagonal(torch.randn((3, 4, 5, 6), device=device), offset=45252).shape)
        self.assertEqual((5, 6, 0), torch.diagonal(torch.randn((3, 4, 5, 6), device=device), offset=-45252).shape)

        self.assertEqual((0, 0), torch.diagflat(torch.tensor([], device=device)).shape)
        self.assertEqual(torch.zeros(1, 1), torch.diagflat(torch.tensor([], device=device), offset=1))
        self.assertEqual((0, 0), torch.diagflat(torch.tensor([[]], device=device)).shape)
        self.assertEqual(torch.zeros(1, 1), torch.diagflat(torch.tensor([[]], device=device), offset=1))

        # stack, split, chunk
        self.assertEqual((4, 0, 1, 3, 0), torch.stack((x, x, x, x)).shape)
        self.assertEqual([(0, 1, 3, 0)],
                         [z.shape for z in torch.chunk(x, 1, dim=0)])

        self.assertEqual([(0, 1, 3, 0), ] * 3, [z.shape for z in torch.chunk(x, 3, dim=0)])
        self.assertEqual([(0, 1, 1, 0), ] * 3, [z.shape for z in torch.chunk(x, 3, dim=2)])

        # NOTE: split_with_sizes behaves differently than NumPy in that it
        # takes sizes rather than offsets
        self.assertEqual([(0, 1, 0, 0), (0, 1, 1, 0), (0, 1, 2, 0)],
                         [z.shape for z in torch.split(x, (0, 1, 2), dim=2)])

        self.assertRaises(RuntimeError, lambda: torch.split(x, 0, dim=1))
        # This is strange because the split size is larger than the dim size, but consistent with
        # how split handles that case generally (when no 0s are involved).
        self.assertEqual([(0, 1, 3, 0)], [z.shape for z in torch.split(x, 1, dim=0)])
        self.assertEqual([(0, 1, 3, 0)], [z.shape for z in torch.split(x, 0, dim=0)])

    # functions that operate over a dimension but don't reduce.
    def test_dim_function_empty(self, device):
        shape = (0, 1, 2, 0)
        x = torch.randn(shape, device=device)

        # size stride
        self.assertEqual(0, x.size(3))
        self.assertEqual(2, x.size(2))
        self.assertEqual(2, x.stride(0))
        self.assertEqual(1, x.stride(2))

        self.assertEqual(x, torch.nn.functional.glu(x, 0))
        self.assertEqual((0, 1, 1, 0), torch.nn.functional.glu(x, 2).shape)

        # softmax, logsoftmax
        self.assertEqual(x, torch.nn.functional.softmax(x, 0))
        self.assertEqual(x, torch.nn.functional.softmax(x, 2))
        self.assertEqual(x, torch.nn.functional.softmax(x, 3))

        self.assertEqual(x, torch.nn.functional.log_softmax(x, 0))
        self.assertEqual(x, torch.nn.functional.log_softmax(x, 2))
        self.assertEqual(x, torch.nn.functional.log_softmax(x, 3))

        # cumsum, cumprod, cummax, cummin
        self.assertEqual(shape, torch.cumsum(x, 0).shape)
        self.assertEqual(shape, torch.cumsum(x, 2).shape)
        self.assertEqual(shape, torch.cumprod(x, 0).shape)
        self.assertEqual(shape, torch.cumprod(x, 2).shape)
        self.assertEqual(shape, torch.cummax(x, 0)[0].shape)
        self.assertEqual(shape, torch.cummax(x, 2)[0].shape)
        self.assertEqual(shape, torch.cummin(x, 0)[0].shape)
        self.assertEqual(shape, torch.cummin(x, 2)[0].shape)
        self.assertEqual(shape, torch.logcumsumexp(x, 0).shape)
        self.assertEqual(shape, torch.logcumsumexp(x, 2).shape)

        # flip
        self.assertEqual(x, x.flip(0))
        self.assertEqual(x, x.flip(2))

        # roll
        self.assertEqual(x, x.roll(0, 1).roll(0, -1))
        self.assertEqual(x, x.roll(1, x.size(1)))
        self.assertEqual(x, x.roll(1))
        self.assertEqual(x, x.roll((1, 1), (3, 1)))

        # unbind
        self.assertEqual((), x.unbind(0))
        self.assertEqual((torch.empty((0, 1, 0), device=device), torch.empty((0, 1, 0), device=device)),
                         x.unbind(2))

        # cross
        y = torch.randn((0, 1, 3, 0), device=device)
        self.assertEqual(y.shape, torch.cross(y, y).shape)

        # renorm
        self.assertEqual(shape, torch.renorm(x, 1, 0, 5).shape)
        self.assertEqual(shape, torch.renorm(x, 1, 2, 5).shape)

        # sort
        self.assertEqual([shape, shape], [z.shape for z in torch.sort(x, dim=0)])
        self.assertEqual([shape, shape], [z.shape for z in torch.sort(x, dim=2)])

        # topk
        self.assertEqual([shape, shape], [z.shape for z in torch.topk(x, 0, dim=0)])
        self.assertEqual([(0, 1, 1, 0), (0, 1, 1, 0)], [z.shape for z in torch.topk(x, 1, dim=2)])

        y = torch.randn((2, 3, 4), device=device)
        self.assertEqual([(2, 3, 0), (2, 3, 0)], [z.shape for z in torch.topk(y, 0)])

        # gather
        self.assertEqual(shape, torch.gather(x, 0, torch.empty(shape, dtype=torch.int64, device=device)).shape)
        self.assertEqual(shape, torch.gather(x, 2, torch.empty(shape, dtype=torch.int64, device=device)).shape)
        larger_shape = torch.empty((0, 1, 3, 0), dtype=torch.int64, device=device)
        self.assertEqual(larger_shape.shape, torch.gather(x, 2, larger_shape).shape)
        smaller_shape = torch.empty((0, 1, 0, 0), dtype=torch.int64, device=device)
        self.assertEqual(smaller_shape.shape, torch.gather(x, 2, smaller_shape).shape)
        y = torch.randn((2, 3, 4), device=device)
        self.assertEqual((0, 3, 4),
                         torch.gather(y, 0, torch.empty((0, 3, 4), dtype=torch.int64, device=device)).shape)

        # scatter, scatter_add
        for dim in [0, 2]:
            y = torch.randn(shape, device=device)
            y_src = torch.randn(shape, device=device)
            ind = torch.empty(shape, dtype=torch.int64, device=device)
            self.assertEqual(shape, y.scatter_(dim, ind, y_src).shape)
            self.assertEqual(shape, y.scatter_add_(dim, ind, y_src).shape)

        z = torch.randn((2, 3, 4), device=device)
        z_src = torch.randn((2, 3, 4), device=device)
        self.assertEqual(z, z.scatter_(2, torch.empty((2, 3, 0), dtype=torch.int64, device=device), z_src))
        self.assertEqual(z, z.scatter_add_(2, torch.empty((2, 3, 0), dtype=torch.int64, device=device), z_src))

        # index_fill, index_copy, index_add
        c = x.clone()
        c_clone = c.clone()
        ind_empty = torch.tensor([], dtype=torch.int64, device=device)
        ind_01 = torch.tensor([0, 1], dtype=torch.int64, device=device)
        self.assertEqual(c_clone, c.index_fill_(0, ind_empty, -1))
        self.assertEqual(c_clone, c.index_fill_(2, ind_empty, -1))
        self.assertEqual(c_clone, c.index_fill_(2, ind_01, -1))
        self.assertEqual(c_clone, c.index_copy_(0, ind_empty, torch.empty((0, 1, 2, 0), device=device)))
        self.assertEqual(c_clone, c.index_copy_(2, ind_empty, torch.empty((0, 1, 0, 0), device=device)))
        self.assertEqual(c_clone, c.index_copy_(2, ind_01, torch.empty((0, 1, 2, 0), device=device)))
        self.assertEqual(c_clone, c.index_add_(0, ind_empty, torch.empty((0, 1, 2, 0), device=device)))
        self.assertEqual(c_clone, c.index_add_(2, ind_empty, torch.empty((0, 1, 0, 0), device=device)))
        self.assertEqual(c_clone, c.index_add_(2, ind_01, torch.empty((0, 1, 2, 0), device=device)))

        c = torch.randn((0, 1, 2), device=device)
        c_clone = c.clone()
        self.assertEqual(c_clone, c.index_fill_(0, ind_empty, -1))
        self.assertEqual(c_clone, c.index_copy_(0, ind_empty, torch.empty((0, 1, 2), device=device)))
        self.assertEqual(c_clone, c.index_add_(0, ind_empty, torch.empty((0, 1, 2), device=device)))
        self.assertEqual(c_clone, c.index_fill_(0, ind_empty, -1))
        self.assertEqual(c_clone, c.index_copy_(0, ind_empty, torch.empty((0, 1, 2), device=device)))
        self.assertEqual(c_clone, c.index_add_(0, ind_empty, torch.empty((0, 1, 2), device=device)))

        # index fill/copy/add non-empty
        z = torch.randn((2, 3, 4), device=device)
        self.assertEqual(z, z.index_fill_(0, ind_empty, -1))
        z = torch.randn((2, 3, 4), device=device)
        self.assertEqual(z, z.index_copy_(0, ind_empty, torch.empty((0, 3, 4), device=device)))
        z = torch.randn((2, 3, 4), device=device)
        self.assertEqual(z, z.index_add_(0, ind_empty, torch.empty((0, 3, 4), device=device)))

        # index_select
        self.assertEqual(x, x.index_select(0, ind_empty))
        self.assertEqual((0, 1, 0, 0), x.index_select(2, ind_empty).shape)
        self.assertEqual(x, x.index_select(2, ind_01))
        z = torch.randn((2, 3, 4), device=device)  # non-empty
        self.assertEqual((0, 3, 4), z.index_select(0, ind_empty).shape)
        c = torch.randn((0, 1, 2), device=device)
        self.assertEqual(c, c.index_select(0, ind_empty))
        c = torch.randn((0, 1, 2), device=device)
        self.assertEqual(c, c.index_select(0, ind_empty))
        w = torch.randn((0, 3), device=device)
        self.assertEqual((0, 2), w.index_select(1, ind_01).shape)
        w = torch.randn((3, 0), device=device)
        self.assertEqual((2, 0), w.index_select(0, ind_01).shape)
        ind_01_int32 = torch.tensor([0, 1], dtype=torch.int32, device=device)
        self.assertEqual((2, 0), w.index_select(0, ind_01_int32).shape)
        s = torch.randn([], device=device)
        ind_0 = torch.tensor([0], dtype=torch.int32, device=device)
        self.assertEqual([], s.index_select(0, ind_0).shape)
        if device == 'cpu':
            w = torch.randn((0, 3), device=device)
            with self.assertRaisesRegex(RuntimeError, "self indexing axis dim should be positive"):
                torch.index_select(w, 0, ind_01)
            ind_05 = torch.tensor([0, 5], dtype=torch.int64, device=device)
            with self.assertRaisesRegex(RuntimeError, "INDICES element is out of DATA bounds"):
                torch.index_select(w, 1, ind_05)
            with self.assertRaisesRegex(RuntimeError, "Index to scalar can have only 1 value"):
                torch.index_select(s, 0, ind_empty)
        with self.assertRaisesRegex(RuntimeError, "Index to scalar can have only 1 value"):
            torch.ones([]).index_select(0, torch.Tensor([0, 0]).int())

    # FIXME: find a test suite for the pdist operator
    @unittest.skipIf(IS_FBCODE and IS_REMOTE_GPU, "sandcastle OOM with current tpx gpu/re configuration")
    @skipIfRocm
    @onlyCUDA
    @largeTensorTest('32GB', device='cpu')
    @largeTensorTest('5GB', device='cuda')
    def test_pdist_norm_large(self, device):
        # use dim0>=46342 for forward, see:
        # https://github.com/pytorch/pytorch/issues/30583
        # Compare output using GPU with the CPU implementation
        x = torch.randn(50000, 1, dtype=torch.float32)      # 50k * 4 bytes = 200 KB
        # Will require 1249975000 float32s
        expected_cpu = torch.pdist(x, p=2)                  # ~1250M * 4 bytes = 5 GB on CPU
        actual_cpu = torch.pdist(x.to(device), p=2).cpu()         # 5 GB on GPU + 5GB on CPU
        # Workaround for large memory overhead of self.assertTrue (see #84944)
        self.assertTrue(torch.allclose(expected_cpu, actual_cpu))  # ~20GB in allclose

    # FIXME: move to elementwise ternary test suite
    @onlyNativeDeviceTypes
    @dtypesIfCUDA(*set(get_all_math_dtypes('cuda')))
    @dtypes(*set(get_all_math_dtypes('cpu')))
    def test_addcdiv(self, device, dtype):
        # Returns floating or integral scalar corresponding to dtype
        def _number(floating, integer, dtype):
            if dtype in [torch.half, torch.float, torch.double, torch.bfloat16]:
                return floating
            elif dtype in [torch.cfloat, torch.cdouble]:
                return floating * (1 + 1j)
            else:
                return integer

        def non_zero_rand(size, dtype, device):
            if dtype.is_floating_point or dtype.is_complex:
                a = torch.rand(size=size, dtype=dtype, device=device)
            elif dtype == torch.uint8:
                a = torch.randint(1, 5, size=size, dtype=dtype, device=device)
            else:
                a = torch.randint(-5, 5, size=size, dtype=dtype, device=device)
            return a + (a == 0).to(dtype)

        def _test_addcdiv():
            a = non_zero_rand((2, 2), dtype=dtype, device=device)
            b = non_zero_rand((2, 2), dtype=dtype, device=device)
            c = non_zero_rand((2, 2), dtype=dtype, device=device)
            alpha = _number(0.5, 3, dtype)

            expected = a + (alpha * b) / c
            actual = torch.addcdiv(a, b, c, value=alpha)
            self.assertEqual(expected, actual)

            with self.assertWarnsOnceRegex(
                    UserWarning, "This overload of addcdiv is deprecated"):
                self.assertEqual(actual, torch.addcdiv(a, alpha, b, c))

        if not (dtype.is_floating_point or dtype.is_complex):
            # Integer division with addcdiv is prohibited
            with self.assertRaises(RuntimeError):
                _test_addcdiv()
        else:
            _test_addcdiv()

        if self.device_type == 'cuda' and dtype == torch.half:
            a = torch.tensor([60000.0], device=device, dtype=dtype)
            b = torch.tensor([60000.0], device=device, dtype=dtype)
            c = torch.tensor([1.0], device=device, dtype=dtype)
            out = torch.addcmul(a, b, c, value=-2)
            self.assertTrue(not (out.isnan() or out.isinf()))

    def test_nullary_op_mem_overlap(self, device):
        ops = (
            ("random_", ()),
            ("uniform_", ()),
            ("cauchy_", ()),
            ("log_normal_", ()),
            ("exponential_", ()),
            ("geometric_", (0.5,)),
            ("normal_", ()),
        )

        x = torch.rand((1, 3)).expand((3, 3))
        for op, args in ops:
            with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
                getattr(x, op)(*args)

    # FIXME: move to an elementwise ternary test suite and make this an OpInfo test
    # https://github.com/pytorch/pytorch/issues/126474
    @xfailIfTorchDynamo
    @skipIfTorchInductor("https://github.com/pytorch/pytorch/issues/126474")
    @dtypes(torch.double)
    def test_ternary_op_mem_overlap(self, device, dtype):
        if device == "cpu" and TEST_WITH_TORCHINDUCTOR:
            self.skipTest("Failing on cpu")

        ops = [
            ("addcmul", True, True, 'cpu'),
            ("addcmul", True, True, 'cuda'),
            ("addcdiv", True, True, 'cpu'),
            ("addcdiv", True, True, 'cuda'),
            ("lerp", True, True, 'cpu'),
            ("lerp", True, True, 'cuda')
        ]

        for (fn, has_input_output_mem_overlap_check,
             has_internal_mem_overlap_check, dev) in ops:
            if dev != device:
                continue
            out_op = getattr(torch, fn)
            inplace_op = getattr(torch.Tensor, fn + '_')
            self.check_internal_mem_overlap(
                inplace_op, 3, dtype, device,
                expected_failure=not has_internal_mem_overlap_check)
            self.ternary_check_input_output_mem_overlap(out_op, dev,
                                                        expected_failure=not has_input_output_mem_overlap_check)

    @expectedFailureMeta  # RuntimeError not raised
    @dtypes(torch.double)
    @onlyNativeDeviceTypes
    def test_copy_mem_overlap(self, device, dtype):
        self.check_internal_mem_overlap(
            torch.Tensor.copy_, num_inputs=2, dtype=dtype, device=device)
        sz = 9
        doubles = torch.randn(2 * sz, dtype=dtype, device=device)
        self.unary_check_input_output_mem_overlap(
            doubles, sz, lambda input, out: out.copy_(input))

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @onlyNativeDeviceTypes
    def test_index_add_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        y = torch.rand((6,), device=device)
        ind = torch.tensor([2, 1, 0], device=device)
        value = torch.rand((3,), device=device)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.index_add_(0, ind, value)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            y.index_add_(0, ind, y[:3])
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_add_(0, ind, ind.clone())
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_add_(0, ind.clone(), ind)

    @onlyCUDA
    @skipCUDAIfNotRocm  # This UT throws an OOM error on CUDA
    def test_index_add_large_inputs(self, device):
        D = 6144
        x = torch.zeros([16384, D], device=device, dtype=torch.bfloat16)
        index = torch.randint(0, 16384, (1, 32, 16384), device=device, dtype=torch.int64)
        output = torch.ones([1, 32, 16384, D], device=device, dtype=torch.bfloat16)  # Use random values for test

        x_before = x.clone()
        # Manually update x_before to generate expected values
        for batch in range(output.shape[1]):  # Loop over batch size (32 in this case)
            for idx in range(output.shape[2]):  # Loop over index
                idx_val = index[0, batch, idx].item()
                x_before[idx_val] += output[0, batch, idx]

        # Run index_add to get actual values
        x.index_add_(0, index.view(-1), output.view(-1, D))

        self.assertEqual(x_before, x)

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @onlyNativeDeviceTypes
    def test_index_copy_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        y = torch.rand((6,), device=device)
        ind = torch.tensor([2, 1, 0], device=device)
        value = torch.rand((3,), device=device)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.index_copy_(0, ind, value)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            y.index_copy_(0, ind, y[:3])
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_copy_(0, ind, ind.clone())
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_copy_(0, ind.clone(), ind)

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @expectedFailureMeta  # Warning not triggered
    @onlyNativeDeviceTypes
    def test_index_fill_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        ind = torch.tensor([2, 1, 0], device=device)

        with self.assertWarnsRegex(UserWarning, "index_fill_ on expanded tensors"):
            x.index_fill_(0, ind, 1.0)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_fill_(0, ind, 0)

    # FIXME: convert to ErrorInputs
    @expectedFailureMeta  # RuntimeError not raised
    @onlyNativeDeviceTypes
    def test_shift_mem_overlap(self, device):
        x = torch.rand(3, device=device)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x[:-1] <<= x[1:]
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x[:-1] >>= x[1:]

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors)
    @expectedFailureMeta  # RuntimeError not raised
    @onlyNativeDeviceTypes
    def test_bernoulli_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))

        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.bernoulli_()
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.bernoulli_(p=0.1)
        p = torch.rand(6, device=device)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.bernoulli_(p=p)

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @expectedFailureMeta  # RuntimeError not raised
    @onlyNativeDeviceTypes
    def test_put_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        y = torch.rand((6,), device=device)
        ind = torch.tensor([2, 1, 0], device=device)
        value = torch.rand((3,), device=device)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.put_(ind, value)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            y.put_(ind[0], y[0])
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.put_(ind, ind)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            y.put_(ind, y[:3])
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.put_(ind, ind.clone())
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.put_(ind.clone(), ind)

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @expectedFailureMeta  # UserWarning not triggered
    @onlyNativeDeviceTypes
    def test_index_put_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        y = torch.rand((6,), device=device)
        ind = torch.tensor([2, 1, 0], device=device)
        value = torch.rand((3,), device=device)
        with self.assertWarnsRegex(UserWarning, 'expanded tensors'):
            x.index_put_((ind,), value)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            y.index_put_((ind,), y[0])
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_put_((ind,), ind)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            y.index_put_((ind,), y[:3])
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_put_((ind,), ind.clone())
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.index_put_((ind.clone(),), ind)

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @expectedFailureMeta  # UserWarning not triggered
    @onlyNativeDeviceTypes
    def test_masked_fill_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        mask = torch.tensor([True, False, True, True, False, False], device=device)
        with self.assertWarnsRegex(UserWarning, 'expanded tensors'):
            x.masked_fill_(mask, 0.)

        fill_val = torch.tensor(0., device=device)
        with self.assertWarnsRegex(UserWarning, 'expanded tensors'):
            x.masked_fill_(mask, fill_val)

        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            mask[1:].masked_fill_(mask[:-1], False)

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @expectedFailureMeta  # RuntimeError not raised
    @onlyNativeDeviceTypes
    def test_masked_scatter_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        src = torch.rand((3,), device=device)
        mask = torch.tensor([True, False, True, True, False, False], device=device)

        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.masked_scatter_(mask, src)

    # FIXME: convert to ErrorInputs
    # (but have to extend ErrorInputs to handle inplace-only errors!)
    @onlyNativeDeviceTypes
    def test_scatter_mem_overlap(self, device):
        x = torch.rand((1,), device=device).expand((6,))
        src = torch.rand((3,), device=device)
        ind = torch.tensor([2, 1, 0], device=device, dtype=torch.int64)

        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            x.scatter_(0, ind, src)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            src.scatter_(0, ind, src)
        with self.assertRaisesRegex(RuntimeError, 'unsupported operation'):
            ind.scatter_(0, ind, ind.clone())

    # FIXME: move to test distributions
    @onlyCUDA
    def test_multinomial_device_constrain(self, device):
        x = torch.empty(3, device="cpu")
        y = torch.empty(3, device=device)
        self.assertRaisesRegex(
            RuntimeError, "Expected all tensors to be on the same device",
            lambda: torch.multinomial(x, 2, out=y))

    # FIXME: move to test distributions
    @deviceCountAtLeast(2)
    @onlyCUDA
    @skipIfTorchInductor("FIXME: error not thrown")
    def test_multinomial_gpu_device_constrain(self, devices):
        x = torch.empty(3, device=devices[0])
        y = torch.empty(3, device=devices[1], dtype=torch.long)
        self.assertRaisesRegex(
            RuntimeError, "Expected all tensors to be on the same device",
            lambda: torch.multinomial(x, 2, out=y))

    # FIXME: convert this to an automated OpInfo test
    @deviceCountAtLeast(2)
    @onlyCUDA
    def test_device_guard(self, devices):
        # verify that all operators with `device_guard: False` behave properly with multiple devices.
        # TODO: if we had operator introspection we could figure out this set of operators automatically...
        x = torch.randn((1, 2, 3), device=devices[1])
        y = torch.zeros((1, 3, 2), device=devices[1])
        scalar = torch.tensor(5, device=devices[1])

        # property ops
        torch.cudnn_is_acceptable(x)
        x.is_distributed()
        x.is_floating_point()
        x.is_complex()
        x.is_same_size(y)
        x.is_signed()
        x.size(0)
        x.stride(0)
        x.numel()
        x.is_set_to(y)
        x.data_ptr()
        scalar.is_nonzero()

        # sparse property ops
        y[0][1] = 5
        y_sparse = y.to_sparse()
        y_sparse.sparse_dim()
        y_sparse._dimI()
        y_sparse.dense_dim()
        y_sparse._dimV()
        y_sparse._nnz()
        y_sparse.is_coalesced()
        y_sparse._indices()
        y_sparse._values()
        y_sparse.indices()
        y_sparse.values()

        # in-place ops
        def inplace():
            return torch.randn((1, 2, 3), device=devices[1])
        inplace().as_strided_(y.size(), y.stride())
        inplace().resize_(y.size())
        inplace().squeeze_()
        inplace().squeeze_(0)
        inplace().unsqueeze_(2)
        inplace().transpose_(1, 2)
        inplace().squeeze_().t_()
        inplace().set_(x.storage())
        inplace().set_(x.storage(), x.storage_offset(), x.size(), x.stride())
        inplace().set_(x)
        inplace().set_()
        y_sparse._coalesced_(True)

        # shape modification
        x.as_strided(y.size(), y.stride())
        x.expand((5, 2, 3))
        x.expand_as(x)
        x.sum_to_size((1,))
        torch.broadcast_tensors(x , x)
        x.reshape((1, 3, 2))
        x.reshape_as(y)
        x.squeeze()
        x.squeeze(0)
        x.squeeze().t()
        x.transpose(1, 2)
        x.unsqueeze(2)
        x.view((1, 3, 2))
        x.view_as(y)

        # chunk, split, etc.
        x.chunk(2, dim=1)
        x.split(1, dim=2)
        x.split_with_sizes([1, 2], dim=2)
        x.unfold(dimension=2, size=1, step=1)

        x.narrow(1, 1, 1)
        x.select(1, 1)
        torch.isnan(x)

        torch.empty((1, 3, 2), out=y)
        torch.empty_like(x)
        torch.empty_like(x, dtype=torch.int64)

        # to
        x.to(x)
        x.to(y)
        x.to(x, copy=True)

    def test_is_signed(self, device):
        self.assertEqual(torch.IntTensor(5).to(device).is_signed(), True)
        self.assertEqual(torch.ByteTensor(5).to(device).is_signed(), False)
        self.assertEqual(torch.CharTensor(5).to(device).is_signed(), True)
        self.assertEqual(torch.FloatTensor(5).to(device).is_signed(), True)
        self.assertEqual(torch.HalfTensor(10).to(device).is_signed(), True)

    def test_tensor_type(self):
        for t in torch._tensor_classes:
            if 'cuda' in t.__module__:
                self.assertEqual(t.is_cuda, True)
            else:
                self.assertEqual(t.is_cuda, False)
            if 'xpu' in t.__module__:
                self.assertEqual(t.is_xpu, True)
            else:
                self.assertEqual(t.is_xpu, False)

    # Note - reports a leak of 512 bytes on CUDA device 1
    @deviceCountAtLeast(2)
    @skipCUDAMemoryLeakCheckIf(True)
    @onlyCUDA
    def test_tensor_set_errors_multigpu(self, devices):
        f_cuda0 = torch.randn((2, 3), dtype=torch.float32, device=devices[0])
        f_cuda1 = torch.randn((2, 3), dtype=torch.float32, device=devices[1])

        self.assertRaises(RuntimeError, lambda: f_cuda0.set_(f_cuda1.storage()))
        self.assertRaises(RuntimeError,
                          lambda: f_cuda0.set_(f_cuda1.storage(), 0, f_cuda1.size(), f_cuda1.stride()))
        self.assertRaises(RuntimeError, lambda: f_cuda0.set_(f_cuda1))

    # FIXME: move to test_serialization
    @onlyCUDA
    @deviceCountAtLeast(1)  # Note: Tests works with one but prefers more devices
    def test_serialization(self, devices):
        def _test_serialization(filecontext_lambda):
            t0 = torch.cuda.FloatTensor(5).fill_(1)
            with torch.cuda.device(devices[-1]):
                tn = torch.cuda.FloatTensor(3).fill_(2)
            torch.cuda.set_device(devices[0])
            b = (t0, tn)
            with filecontext_lambda() as f:
                torch.save(b, f)
                f.seek(0)
                c = torch.load(f)
                self.assertEqual(b, c, atol=0, rtol=0)
                u0, un = c
                self.assertEqual(str(u0.device), devices[0])
                self.assertEqual(str(un.device), devices[-1])

        _test_serialization(tempfile.NamedTemporaryFile)
        _test_serialization(BytesIOContext)

    # FIXME: move memory format tests to their own test class/suite
    def test_memory_format_preserved_after_permute(self, device):
        x = torch.randn(4, 3, 8, 8, device=device)
        nhwc = x.contiguous(memory_format=torch.channels_last)
        y = nhwc.permute(0, 1, 3, 2).permute(0, 1, 3, 2)
        self.assertTrue(y.is_contiguous(memory_format=torch.channels_last))

        x = torch.randn(4, 3, 8, 8, 8, device=device)
        ndhwc = x.contiguous(memory_format=torch.channels_last_3d)
        y = ndhwc.permute(0, 1, 4, 3, 2).permute(0, 1, 4, 3, 2)
        self.assertTrue(y.is_contiguous(memory_format=torch.channels_last_3d))

    def test_memory_format_propagation_rules(self, device):

        contiguous = torch.rand(10, 3, 5, 5, device=device)
        cl = torch.rand(10, 3, 5, 5, device=device).contiguous(memory_format=torch.channels_last)
        ambiguous = torch.rand(10, 3, 1, 1, device=device).contiguous(memory_format=torch.channels_last)
        self.assertTrue(ambiguous.is_contiguous(memory_format=torch.channels_last))
        self.assertTrue(ambiguous.is_contiguous(memory_format=torch.contiguous_format))
        bias = torch.rand(1, 1, 1, 1, device=device).contiguous(memory_format=torch.channels_last)

        def _test_propagation_rules(self, contiguous, cl, ambiguous, bias):
            options = ((ambiguous, contiguous, torch.contiguous_format),
                       (ambiguous, cl, torch.channels_last),
                       (contiguous, ambiguous, torch.contiguous_format),
                       (contiguous, cl, torch.contiguous_format),
                       (cl, ambiguous, torch.channels_last),
                       (cl, contiguous, torch.channels_last),
                       (bias, cl, torch.channels_last),
                       (cl, bias, torch.channels_last),)

            for a, b, mf in options:
                result = a + b
                self.assertTrue(result.is_contiguous(memory_format=mf))

        _test_propagation_rules(self, contiguous, cl, ambiguous, bias)

        cl = cl.to(memory_format=torch.channels_last)
        ambiguous = ambiguous.to(memory_format=torch.channels_last)
        bias = bias.to(memory_format=torch.channels_last)

        _test_propagation_rules(self, contiguous, cl, ambiguous, bias)

        # test cases when strides matter in ambiguous tensors
        for mf in (torch.channels_last, torch.contiguous_format):
            ambiguous = torch.rand(10, 3, 1, 1, device=device).to(memory_format=mf)
            bias = torch.rand(3, 1, 1, device=device)
            result = ambiguous + bias
            self.assertEqual(ambiguous.stride(), result.stride())
            result = bias + ambiguous
            self.assertEqual(ambiguous.stride(), result.stride())
            result = ambiguous * 5
            self.assertEqual(ambiguous.stride(), result.stride())

    def test_memory_format_empty_like(self, device):
        def test_helper(x, memory_format):
            xc = x.contiguous(memory_format=memory_format)

            like = torch.empty_like(xc, memory_format=torch.preserve_format)
            self.assertFalse(like.is_contiguous())
            self.assertTrue(like.is_contiguous(memory_format=memory_format))

            like_x = torch.empty_like(x, memory_format=torch.preserve_format)
            self.assertTrue(like_x.is_contiguous())
            self.assertFalse(like_x.is_contiguous(memory_format=memory_format))

            like = torch.empty_like(x, memory_format=memory_format)
            self.assertFalse(like.is_contiguous())
            self.assertTrue(like.is_contiguous(memory_format=memory_format))

            like = torch.empty_like(xc, memory_format=torch.contiguous_format)
            self.assertTrue(like.is_contiguous())
            self.assertFalse(like.is_contiguous(memory_format=memory_format))

            like = torch.empty_like(xc)
            self.assertFalse(like.is_contiguous())
            self.assertTrue(like.is_contiguous(memory_format=memory_format))

            sparse = x.to_sparse()
            with self.assertRaises(RuntimeError):
                torch.empty_like(sparse, memory_format=torch.preserve_format)

        test_helper(torch.randn(4, 3, 8, 8, device=device), torch.channels_last)
        test_helper(torch.randn(4, 3, 8, 8, 8, device=device), torch.channels_last_3d)

    def test_memory_format_consistency(self, device):
        x = torch.randn(10, 3, 1, 1, device=device)
        x_rep = x.as_strided(x.size(), x.stride())
        self.assertEqual(x.size(), x_rep.size())
        self.assertEqual(x.stride(), x_rep.stride())
        self.assertEqual(x.is_contiguous(), x_rep.is_contiguous())
        self.assertEqual(x.is_contiguous(memory_format=torch.channels_last), x_rep.is_contiguous(memory_format=torch.channels_last))
        self.assertEqual(
            x.is_contiguous(memory_format=torch.channels_last_3d), x_rep.is_contiguous(memory_format=torch.channels_last_3d))

    # FIXME: make this a elementwise unary and elementwise binary OpInfo test
    def test_memory_format_operators(self, device):
        def _chunk_op(x, y):
            x1, x2 = x.chunk(2, dim=1)
            return x1 + x2

        def _unsqueeze_op_add(x, y):
            return x[0].unsqueeze(0) + 3

        def _unsqueeze_op_clone(x, y):
            return x[0].unsqueeze(0).clone()

        def _test_helper(x, y, bias, memory_format):
            return_contig_fns = [
                lambda x, y: y + x,
                lambda x, y: y * x,
                lambda x, y: y.addcdiv(x, y, value=2),
                lambda x, y: y.addcmul(x, y, value=2),
            ]
            bias_fns = [
                lambda x, b: x + b,
                lambda x, b: b + x,
            ]
            fns = [
                lambda x, y: x.clone(),
                lambda x, y: x + 3,
                lambda x, y: 3 * x,
                lambda x, y: x + y,
                lambda x, y: x * y,
                lambda x, y: abs(x),
                lambda x, y: x.abs(),
                lambda x, y: x.abs_(),
                lambda x, y: x.acos(),
                lambda x, y: x.acos_(),
                lambda x, y: x.add(y, alpha=3),
                lambda x, y: x.add_(y, alpha=3),
                lambda x, y: x.addcdiv(y, y, value=2),
                lambda x, y: x.addcdiv_(y, y, value=2),
                lambda x, y: x.addcmul(y, y, value=2),
                lambda x, y: x.addcmul_(y, y, value=2),
                lambda x, y: x.acosh(),
                lambda x, y: x.acosh_(),
                lambda x, y: x.asinh(),
                lambda x, y: x.asinh_(),
                lambda x, y: x.atanh(),
                lambda x, y: x.atanh_(),
                lambda x, y: x.asin(),
                lambda x, y: x.asin_(),
                lambda x, y: x.atan(),
                lambda x, y: x.atan2(y),
                lambda x, y: x.atan2_(y),
                lambda x, y: x.ceil(),
                lambda x, y: x.ceil_(),
                lambda x, y: x.clamp(-1, 1),
                lambda x, y: x.cos(),
                lambda x, y: x.cosh(),
                lambda x, y: x.div(0.5),
                lambda x, y: x.div_(0.5),
                lambda x, y: x.div(y),
                lambda x, y: x.div_(y),
                lambda x, y: x.digamma(),
                lambda x, y: x.digamma_(),
                lambda x, y: x.erf(),
                lambda x, y: x.erfc(),
                lambda x, y: x.erfinv(),
                lambda x, y: x.erfinv_(),
                lambda x, y: x.exp(),
                lambda x, y: x.expm1(),
                lambda x, y: x.expm1_(),
                lambda x, y: x.floor(),
                lambda x, y: x.floor_(),
                lambda x, y: x.fmod(2),
                lambda x, y: x.frac(),
                lambda x, y: x.hypot(y),
                lambda x, y: x.hypot_(y),
                lambda x, y: x.i0(),
                lambda x, y: x.i0_(),
                lambda x, y: x.lerp(y, 0.5),
                lambda x, y: x.log(),
                lambda x, y: x.log_(),
                lambda x, y: x.log10(),
                lambda x, y: x.log10_(),
                lambda x, y: x.log1p(),
                lambda x, y: x.log1p_(),
                lambda x, y: x.log2(),
                lambda x, y: x.log2_(),
                lambda x, y: x.mul(3),
                lambda x, y: x.mul_(3),
                lambda x, y: x.neg(),
                lambda x, y: x.neg_(),
                lambda x, y: x.pow(3),
                lambda x, y: x.pow_(3),
                lambda x, y: x.pow(0.0),
                lambda x, y: x.pow(1.0),
                lambda x, y: x.reciprocal(),
                lambda x, y: x.remainder(2),
                lambda x, y: x.round(),
                lambda x, y: x.round_(),
                lambda x, y: x.rsqrt(),
                lambda x, y: x.rsqrt_(),
                lambda x, y: x.sigmoid(),
                lambda x, y: x.sigmoid_(),
                lambda x, y: x.logit(),
                lambda x, y: x.logit_(),
                lambda x, y: x.logit(1e-6),
                lambda x, y: x.logit_(1e-6),
                lambda x, y: x.sign(),
                lambda x, y: x.sign_(),
                lambda x, y: x.sgn(),
                lambda x, y: x.sgn_(),
                lambda x, y: x.sin(),
                lambda x, y: x.sin_(),
                lambda x, y: x.sinh(),
                lambda x, y: x.sinh_(),
                lambda x, y: x.sqrt(),
                lambda x, y: x.sqrt_(),
                lambda x, y: x.tan(),
                lambda x, y: x.tanh(),
                lambda x, y: x.trunc(),
                lambda x, y: x.trunc_(),
                _chunk_op,
                _unsqueeze_op_add,
                _unsqueeze_op_clone,
            ]
            x_c = x.contiguous()
            y_c = y.contiguous()
            b_c = bias.contiguous()
            for fn in fns:
                is_inplace = '_(' in inspect.getsource(fn)
                x_clone = x.clone() if is_inplace else x
                x_c_clone = x_c.clone() if is_inplace else x_c
                result_c = fn(x_c_clone, y_c)
                result = fn(x_clone, y)
                self.assertEqual(result, result_c, lambda msg: f"{msg}\nFailed for '{inspect.getsource(fn).strip()}'")
                self.assertTrue(
                    result.is_contiguous(memory_format=memory_format),
                    lambda msg: f"{msg}\nresult of the '{inspect.getsource(fn).strip()}' is not in '{memory_format}' format")

            for fn in bias_fns:
                result_c = fn(x_c, b_c)
                result = fn(x, bias)
                self.assertEqual(result, result_c, lambda msg: f"{msg}\nFailed for '{inspect.getsource(fn).strip()}'")
                self.assertTrue(
                    result.is_contiguous(memory_format=memory_format),
                    lambda msg: f"{msg}\nresult of the '{inspect.getsource(fn).strip()}' is not in '{memory_format}' format")

            for fn in return_contig_fns:
                result_c = fn(x_c, y_c)
                result = fn(x, y)
                self.assertEqual(result, result_c, lambda msg: f"{msg}\nFailed for '{inspect.getsource(fn).strip()}'")
                self.assertTrue(
                    result.is_contiguous(memory_format=torch.contiguous_format),
                    lambda msg: f"{msg}\nresult of the '{inspect.getsource(fn).strip()}' is not in '{torch.contiguous_format}' format")

        _test_helper(
            torch.randn((4, 3, 8, 8), device=device).contiguous(memory_format=torch.channels_last),
            abs(torch.randn((4, 3, 8, 8), device=device)) + 1,
            torch.randn((1, 3, 1, 1), device=device).contiguous(memory_format=torch.channels_last),
            torch.channels_last)
        _test_helper(
            torch.randn((4, 3, 8, 8, 8), device=device).contiguous(memory_format=torch.channels_last_3d),
            abs(torch.randn((4, 3, 8, 8, 8), device=device)) + 1,
            torch.randn((1, 3, 1, 1, 1), device=device).contiguous(memory_format=torch.channels_last_3d),
            torch.channels_last_3d)

    # FIXME: make this a elementwise unary and elementwise binary OpInfo test
    def test_strides_propagation(self, device):
        def _test_helper(x, op, unary=False):
            def compare_strides(s1, s2, div):
                sdiv = [s // div for s in s1]
                self.assertEqual(sdiv, s2)

            dim = x.dim()
            # we produce memory dense outputs, so when input is strided on the last dimension
            # we need to divide by that dimension stride to compare input and result strides
            div = x.stride(-1)
            for p in permutations(range(dim)):
                xp = x.permute(p)
                if not unary:
                    y = torch.randn(xp.size(-1), device=x.device, dtype=x.dtype)
                    for inputs in ((xp, xp), (xp, y), (y, xp)):
                        res = op(*inputs)
                        compare_strides(xp.stride(), res.stride(), div)
                        self.assertEqual(xp.size(), res.size())
                        out = torch.empty(0, device=xp.device, dtype=res.dtype)
                        res = op(*inputs, out=out)
                        compare_strides(xp.stride(), res.stride(), div)
                        self.assertEqual(xp.size(), res.size())
                else:
                    res = op(xp)
                    compare_strides(xp.stride(), res.stride(), div)
                    self.assertEqual(xp.size(), res.size())
                    out = torch.empty(0, device=xp.device, dtype=res.dtype)
                    res = op(xp, out=out)
                    compare_strides(xp.stride(), res.stride(), div)
                    self.assertEqual(xp.size(), res.size())

        # torch.eq by default calls TensorIterator with defined output, torch.add with undefined
        binary_ops = (torch.eq, torch.add)
        unary_ops = (torch.exp,)
        # memory dense, sliced and ambiguous sliced (ambiguous dense loses permutation information)
        xs = (torch.randn(2, 3, 4, device=device), torch.randn(2, 3, 8, device=device)[:, :, ::2],
              torch.randn(1, 1, 4, 12, device=device)[:, :, :, ::2])
        for op in binary_ops:
            for x in xs:
                _test_helper(x, op)
        for op in unary_ops:
            for x in xs:
                _test_helper(x, op, unary=True)

    @onlyCUDA
    @unittest.skipIf(PYTORCH_CUDA_MEMCHECK, "is_pinned uses failure to detect pointer property")
    @skipIfTorchDynamo("NotImplementedError: PrimTorch does not support pinned memory")
    def test_pin_memory_from_constructor(self, device):
        def _get_like(t, **kwargs):
            return [
                torch.rand_like(t, **kwargs),
                torch.randn_like(t, **kwargs),
                torch.empty_like(t, **kwargs),
                torch.full_like(t, 4, **kwargs),
                torch.zeros_like(t, **kwargs),
                torch.ones_like(t, **kwargs),
            ]

        def _get_tensors(**kwargs):
            return [
                torch.tensor([10, 11], **kwargs),
                torch.randn(3, 5, **kwargs),
                torch.rand(3, **kwargs),
                # torch.randint(3, 5, **kwargs), // unsupported
                torch.zeros(3, **kwargs),
                torch.randperm(3, **kwargs),
                torch.empty(6, **kwargs),
                torch.ones(6, **kwargs),
                torch.eye(6, **kwargs),
                torch.arange(3, 5, **kwargs)]

        pinned_tensors = _get_tensors(pin_memory=True) + _get_like(torch.empty(5, dtype=torch.float64), pin_memory=True)
        for x in pinned_tensors:
            self.assertTrue(x.is_pinned())

        tensors = _get_tensors() + _get_like(torch.empty(5, dtype=torch.float64, pin_memory=True))
        for x in tensors:
            self.assertFalse(x.is_pinned())

    @deviceCountAtLeast(1)
    @onlyCUDA
    @parametrize("non_blocking", (True, False))
    def test_storage_all_devices(self, devices, non_blocking):
        for device in devices:
            t = torch.randn(6, device=device)
            self.assertEqual(t.dtype, t.storage().dtype)
            s = t.untyped_storage()
            s_cpu = s.to(device='cpu', non_blocking=non_blocking)
            if non_blocking:
                torch.cuda.synchronize()
                self.assertTrue(s_cpu.is_pinned())
            else:
                self.assertFalse(s_cpu.is_pinned())
            t_cpu = torch.empty(()).set_(s_cpu)
            self.assertEqual(t.cpu(), t_cpu)

    # Note [lazy_clone_ tests with inductor enabled]
    # These `lazy_clone_` tests are written in a way that makes them pass in
    # both eager mode and compiled mode (`PYTORCH_TEST_WITH_INDUCTOR=1`). There
    # are cases where COW tensors can materialize at different times and in
    # different ways in compiled mode versus eager mode, and those cases need to
    # be avoided. There are two main wrinkles the be aware of.
    #
    # The first wrinkle is that these tests have to check the internal
    # properties of tensors to make sure they materialize in the expected way,
    # and those checks cause dynamo graph breaks. Depending on the situation, a
    # graph break in-between two compiled graphs that operate on the same COW
    # tensor can make the tensor materialize when it would not materialize in
    # eager mode, causing the checks to fail. The strategy for avoiding this is
    # to make all the operations on COW tensors get compiled into the same
    # graph, by not doing any checks between the operations, and just do all the
    # checks at the end of the test. If we really do want to perform checks
    # between two operations, `op1` and `op2`, the solution is to create two
    # different tests. One test performs just `op1` and then checks. The other
    # test performs `op1` followed immediately by `op2` and then checks.
    #
    # The second wrinkle is that in eager mode, if we perform writes on two COW
    # tensors where one is a lazy clone of the other, the first tensor to be
    # written will be materialized with a new data pointer, and the second
    # tensor will just reuse the original data pointer when it is materialized.
    # But in compiled mode, if these writes happen in the same graph, the order
    # in which the tensors materialize can be different than in eager mode. So
    # in this case the strategy is to purposefully cause a graph break to happen
    # in-between the two write operations, by adding checks between them, so
    # that they have to materialize in the expected order.
    @skipXLA
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_const_data_ptr(self, device, dtype):
        t = torch.tensor([[0, 1], [2, 3]], device=device, dtype=dtype)

        # For a regular tensor, const_data_ptr and data_ptr return the same address
        self.assertEqual(t.const_data_ptr(), t.data_ptr())

        clone = t._lazy_clone()

        self.assertTrue(torch._C._is_cow_tensor(t))
        self.assertTrue(torch._C._is_cow_tensor(clone))

        # const_data_ptr should not trigger COW materialization
        addr = clone.const_data_ptr()
        self.assertEqual(addr, t.const_data_ptr())

        self.assertTrue(torch._C._is_cow_tensor(t))
        self.assertTrue(torch._C._is_cow_tensor(clone))

    # See Note [lazy_clone_ tests with inductor enabled]
    @skipXLA
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_lazy_clone(self, device, dtype):
        t = torch.tensor([[0, 1], [2, 3]], device=device, dtype=dtype)
        t_orig_storage_addr = torch._C._storage_address(t)
        orig_data_ptr = torch._C._data_address(t)
        clone = t._lazy_clone()

        # Lazy cloning a tensor should cause both it and its clone to become COW
        # tensors. They should have different storages, but the same data
        # pointer.

        self.assertTrue(torch._C._is_cow_tensor(clone))
        self.assertTrue(torch._C._is_cow_tensor(t))

        self.assertTrue(torch._C._storage_address(t) == t_orig_storage_addr)
        self.assertTrue(torch._C._storage_address(clone) != t_orig_storage_addr)

        self.assertTrue(torch._C._data_address(t) == orig_data_ptr)
        self.assertTrue(torch._C._data_address(clone) == orig_data_ptr)

    # See Note [lazy_clone_ tests with inductor enabled]
    @skipXLA
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_lazy_clone_view(self, device, dtype):
        t = torch.tensor([[0, 1], [2, 3]], device=device, dtype=dtype)
        t_orig_storage_addr = torch._C._storage_address(t)
        orig_data_ptr = torch._C._data_address(t)
        clone = t._lazy_clone()
        view = t.view([4])

        # Viewing `t` should not cause a copy (materialize) to happen. All the
        # tensors should still be COW and have the same data pointer. `view` and
        # `t` should have the same storage, and `clone` should have a different
        # storage.

        self.assertTrue(torch._C._is_cow_tensor(t))
        self.assertTrue(torch._C._is_cow_tensor(view))
        self.assertTrue(torch._C._is_cow_tensor(clone))

        self.assertTrue(torch._C._storage_address(t) == t_orig_storage_addr)
        self.assertTrue(torch._C._storage_address(view) == t_orig_storage_addr)
        self.assertTrue(torch._C._storage_address(clone) != t_orig_storage_addr)

        self.assertTrue(torch._C._data_address(t) == orig_data_ptr)
        self.assertTrue(torch._C._data_address(clone) == orig_data_ptr)
        self.assertTrue(torch._C._data_address(view) == orig_data_ptr)

    # See Note [lazy_clone_ tests with inductor enabled]
    @skipXLA
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_lazy_clone_view_materialize(self, device, dtype):
        t = torch.tensor([[0, 1], [2, 3]], device=device, dtype=dtype)
        t_orig_storage_addr = torch._C._storage_address(t)
        orig_data_ptr = torch._C._data_address(t)
        clone = t._lazy_clone()
        view = t.view([4])
        view += torch.ones(1, device=device, dtype=dtype)

        # Writing to `t` should cause the storage under `t` and `view` to be
        # copied (materialized), but should not affect `clone`.

        self.assertFalse(torch._C._is_cow_tensor(t))
        self.assertFalse(torch._C._is_cow_tensor(view))
        self.assertTrue(torch._C._is_cow_tensor(clone))

        self.assertTrue(torch._C._storage_address(t) == t_orig_storage_addr)
        self.assertTrue(torch._C._storage_address(view) == t_orig_storage_addr)
        self.assertTrue(torch._C._storage_address(clone) != t_orig_storage_addr)

        t_new_data_addr = torch._C._data_address(t)
        self.assertTrue(t_new_data_addr != orig_data_ptr)
        self.assertTrue(torch._C._data_address(view) == t_new_data_addr)
        self.assertTrue(torch._C._data_address(clone) == orig_data_ptr)

        clone += torch.ones(1, device=device, dtype=dtype)

        # Writing to `clone` should materialize it, so it should no longer
        # be COW. However, since `clone`'s storage is the only COW storage
        # left that holds a reference to the original data pointer, this
        # materialization should not actually cause a copy--it should
        # just reuse the original data pointer.

        self.assertFalse(torch._C._is_cow_tensor(t))
        self.assertFalse(torch._C._is_cow_tensor(view))
        self.assertFalse(torch._C._is_cow_tensor(clone))

        self.assertTrue(torch._C._storage_address(t) == t_orig_storage_addr)
        self.assertTrue(torch._C._storage_address(view) == t_orig_storage_addr)
        self.assertTrue(torch._C._storage_address(clone) != t_orig_storage_addr)

        self.assertTrue(torch._C._data_address(t) == t_new_data_addr)
        self.assertTrue(torch._C._data_address(view) == t_new_data_addr)
        self.assertTrue(torch._C._data_address(clone) == orig_data_ptr)

    @skipXLA
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_lazy_clone_binary_op_no_materialize(self, device, dtype):
        t = torch.tensor([[0, 1], [2, 3]], device=device, dtype=dtype)
        clone = t._lazy_clone()
        t + clone
        self.assertTrue(torch._C._is_cow_tensor(t))
        self.assertTrue(torch._C._is_cow_tensor(clone))

    # This tests that if a COW materialization is attempted inside an
    # `at::parallel_for` loop function, then an error is raised. This test is
    # implemented in Python rather than C++ because the C++ tests are built
    # without multithreading support in `at::parallel_for`.
    @skipXLA
    @skipIfTorchDynamo("Torchdynamo fails and we do not need to test it here anyway")
    @dtypes(*all_types_and_complex_and(torch.half, torch.bool, torch.bfloat16))
    def test_parallel_cow_materialize_error(self, device, dtype):

        def run(num_threads, num_parallel, skip_first, should_error):
            orig_num_threads = torch.get_num_threads()

            try:
                torch.set_num_threads(num_threads)

                a = torch.tensor([[0, 1], [2, 3]], device=device, dtype=dtype)._lazy_clone()

                if should_error:
                    with self.assertRaisesRegex(RuntimeError, r'Materializing a storage'):
                        torch._test_parallel_materialize(
                            a, num_parallel, skip_first)
                else:
                    torch._test_parallel_materialize(a, num_parallel, skip_first)

                # Error should not raise in any case if the tensor is not COW
                b = torch.tensor([[0, 1], [2, 3]], device=device, dtype=dtype)
                torch._test_parallel_materialize(b, num_parallel, skip_first)

            finally:
                torch.set_num_threads(orig_num_threads)

        run(1, 1, False, True)
        run(1, 1, True, False)
        run(1, 10, False, True)
        run(1, 10, True, True)
        run(10, 1, False, True)
        run(10, 1, True, False)
        run(10, 10, False, True)
        run(10, 10, True, True)
        run(10, 2, False, True)
        run(10, 2, True, True)

    # FIXME: move to test distributions
    @dtypesIfCUDA(torch.float, torch.double, torch.half)
    @dtypes(torch.float, torch.double, torch.half)
    def test_multinomial(self, device, dtype):
        def make_prob_dist(shape, is_contiguous):
            if is_contiguous:
                if dtype == torch.half:
                    return torch.zeros(shape, device=device).uniform_().to(dtype=torch.half)
                return torch.zeros(shape, device=device, dtype=dtype).uniform_()
            elif len(shape) == 1:
                if dtype == torch.half:
                    return torch.zeros((shape + [5]), device=device).uniform_().to(dtype=torch.half)[:, 2]
                return torch.zeros((shape + [5]), device=device, dtype=dtype).uniform_()[:, 2]
            else:
                # num dim = 2
                new_shape = [2, shape[1], 7, 1, shape[0], 1, 10]
                if dtype == torch.half:
                    prob_dist = torch.zeros(new_shape, device=device).uniform_().to(dtype=torch.half)
                else:
                    prob_dist = torch.zeros(new_shape, device=device, dtype=dtype).uniform_()
                prob_dist = prob_dist.transpose(1, 4)
                prob_dist = prob_dist[1, :, 5, 0, :, 0, 4]
                if prob_dist.is_contiguous():
                    raise AssertionError("expected prob_dist to be non-contiguous")
                return prob_dist

        for is_contiguous in (True, False):
            # with replacement
            n_row = 3
            for n_col in range(4, 5 + 1):
                prob_dist = make_prob_dist([n_row, n_col], is_contiguous)
                # indices that shouldn't be sampled (<0 means none)
                zero_prob_indices = torch.LongTensor(n_row).random_(-2, n_col).tolist()
                for i, j in enumerate(zero_prob_indices):
                    if j >= 0:
                        prob_dist[i, j] = 0
                n_sample = n_col * 3
                sample_indices = torch.multinomial(prob_dist, n_sample, True)
                self.assertEqual(prob_dist.dim(), 2)
                self.assertEqual(sample_indices.size(1), n_sample)
                for i in range(n_row):
                    zero_prob_idx = zero_prob_indices[i]
                    if zero_prob_idx < 0:
                        continue
                    for j in range(n_sample):
                        self.assertNotEqual(sample_indices[i, j], zero_prob_idx,
                                            msg="sampled an index with zero probability")

            # without replacement
            n_row = 3
            for n_col in range(2, 10 + 1, 2):
                prob_dist = make_prob_dist([n_row, n_col], is_contiguous)
                # indices that shouldn't be sampled (<0 means none)
                zero_prob_indices = torch.LongTensor(n_row).random_(-1, n_col).tolist()
                for i, j in enumerate(zero_prob_indices):
                    if j >= 0:
                        prob_dist[i, j] = 0
                n_sample = max(1, n_col - 2)
                sample_indices = torch.multinomial(prob_dist, n_sample, False)
                self.assertEqual(prob_dist.dim(), 2)
                self.assertEqual(sample_indices.size(1), n_sample)
                for i in range(n_row):
                    row_samples = {}
                    zero_prob_idx = zero_prob_indices[i]
                    for j in range(n_sample):
                        sample_idx = sample_indices[i, j]
                        if zero_prob_idx >= 0:
                            self.assertNotEqual(sample_idx, zero_prob_idx,
                                                msg="sampled an index with zero probability")
                        self.assertNotIn(sample_idx, row_samples, "sampled an index twice")
                        row_samples[sample_idx] = True

            # vector
            n_col = 4
            prob_dist = make_prob_dist([n_col], is_contiguous).fill_(1)
            zero_prob_idx = 1  # index that shouldn't be sampled
            prob_dist[zero_prob_idx] = 0
            n_sample = 20
            sample_indices = torch.multinomial(prob_dist, n_sample, True)
            for sample_index in sample_indices:
                self.assertNotEqual(sample_index, zero_prob_idx, msg="sampled an index with zero probability")
            self.assertEqual(sample_indices.dim(), 1, msg="wrong number of dimensions")
            self.assertEqual(prob_dist.dim(), 1, msg="wrong number of prob_dist dimensions")
            self.assertEqual(sample_indices.size(0), n_sample, msg="wrong number of samples")

        # CUDA misalignment issue (#46702)
        n_row, n_col = 2, 3
        prob_dist = make_prob_dist([n_row, n_col], True)
        n_sample = 1
        sample_indices = torch.multinomial(prob_dist, n_sample, True)
        self.assertEqual(sample_indices.dim(), 2, msg="wrong number of dimensions")
        self.assertEqual(sample_indices.size(1), n_sample, msg="wrong number of samples")

    # FIXME: move to test distributions
    @onlyCUDA
    @dtypes(torch.float, torch.double, torch.half)
    def test_multinomial_deterministic(self, device, dtype):
        gen = torch.Generator(device=device)

        trials = 5
        seed = 0
        prob_dist = torch.rand(10000, 1000, device=device, dtype=dtype)
        n_sample = 1

        for _ in range(trials):
            gen.manual_seed(seed)
            samples_1 = torch.multinomial(prob_dist, n_sample, True, generator=gen)

            gen.manual_seed(seed)
            samples_2 = torch.multinomial(prob_dist, n_sample, True, generator=gen)

            self.assertEqual(samples_1, samples_2)
            self.assertEqual(samples_1.dim(), 2, msg="wrong number of dimensions")
            self.assertEqual(samples_1.size(1), n_sample, msg="wrong number of samples")

    # FIXME: move to test distributions
    @slowTest
    @dtypes(torch.float)
    def test_multinomial_rng_state_advance(self, device, dtype):
        corpus_size = 100000
        freqs = torch.ones(corpus_size, dtype=torch.float, device=device)
        n_sample = 100
        samples1 = torch.multinomial(freqs, n_sample, replacement=True)
        samples2 = torch.multinomial(freqs, n_sample, replacement=True)
        samples = torch.cat([samples1, samples2])
        # expect no more than 1 repeating elements generated in 2 attempts
        # the probability of at least element being repeated is surprisingly large, 18%
        self.assertLessEqual(2 * n_sample - samples.unique().size(0), 2)
        samples1 = torch.multinomial(freqs, n_sample, replacement=False)
        samples2 = torch.multinomial(freqs, n_sample, replacement=False)
        samples = torch.cat([samples1, samples2])
        # expect no more than 1 repeating elements generated in 2 attempts
        self.assertLessEqual(2 * n_sample - samples.unique().size(0), 1)

    def _test_memory_format_transformations(self, device, input_generator_fn, transformation_fn,
                                            memory_format, compare_data=True, default_is_preserve=False):

        if memory_format not in (torch.channels_last, torch.channels_last_3d):
            raise AssertionError(f"unexpected memory_format: {memory_format}")

        # xc is a channels last tensor
        xc = input_generator_fn(device)
        # xc is not memory dense, but looks like channels last
        # We don't preserve non-dense striding
        if not TEST_WITH_TORCHINDUCTOR:
            if memory_format == torch.channels_last:
                xc = xc[..., ::2, ::2]
            else:
                xc = xc[..., ::2, ::2, ::2]

        clone = transformation_fn(xc, memory_format=torch.preserve_format)


        self.assertFalse(clone.is_contiguous())
        self.assertTrue(clone.is_contiguous(memory_format=memory_format))
        if not TEST_WITH_TORCHINDUCTOR:
            self.assertFalse(xc.is_contiguous())
            self.assertFalse(xc.is_contiguous(memory_format=memory_format))
        if compare_data:
            self.assertEqual(xc, clone.to(xc))

        xc = input_generator_fn(device)
        clone = transformation_fn(xc, memory_format=torch.contiguous_format)
        self.assertTrue(clone.is_contiguous())
        self.assertFalse(clone.is_contiguous(memory_format=memory_format))
        if compare_data:
            self.assertEqual(xc, clone.to(xc))

        xc = input_generator_fn(device)
        clone = transformation_fn(xc)

        if default_is_preserve:
            self.assertFalse(clone.is_contiguous())
            self.assertTrue(clone.is_contiguous(memory_format=memory_format))
        else:
            self.assertTrue(clone.is_contiguous())
            self.assertFalse(clone.is_contiguous(memory_format=memory_format))
        if compare_data:
            self.assertEqual(xc, clone.to(xc))

        # TODO copy _like constructors to stride permutation instead of just layout
        if not TEST_WITH_TORCHINDUCTOR:
            x = torch.randn((3, 4, 5, 6, 7, 8, 9), device=device)
            for _ in range(10):
                permutation = list(range(len(x.shape)))
                random.shuffle(permutation)
                x = x.permute(permutation)
                self.assertEqual(x.stride(), transformation_fn(x, memory_format=torch.preserve_format).stride())

    def test_memory_format_to(self, device):
        def get_generator(memory_format, shape):
            def input_generator_fn(device):
                return torch.randn(shape, device=device, dtype=torch.float32).contiguous(memory_format=memory_format)
            return input_generator_fn

        def transformation_fn(tensor, **kwargs):
            return tensor.to(dtype=torch.float64, **kwargs)

        formats_shapes = (
            (torch.channels_last, (4, 3, 8, 8)),
            (torch.channels_last_3d, (4, 3, 8, 8, 8)))

        for mf, shape in formats_shapes:
            self._test_memory_format_transformations(
                device, get_generator(mf, shape), transformation_fn, mf, default_is_preserve=True)

    def test_memory_format_type(self, device):
        def get_generator(memory_format, shape):
            def input_generator_fn(device):
                return torch.randn(shape, device=device, dtype=torch.float32).contiguous(memory_format=memory_format)
            return input_generator_fn

        def transformation_fn(tensor, **kwargs):
            return tensor.to(torch.float64, **kwargs)

        formats_shapes = (
            (torch.channels_last, (4, 3, 8, 8)),
            (torch.channels_last_3d, (4, 3, 8, 8, 8)))

        for mf, shape in formats_shapes:
            self._test_memory_format_transformations(
                device, get_generator(mf, shape), transformation_fn, mf, default_is_preserve=True)

    def test_memory_format_clone(self, device):
        def get_generator(memory_format, shape):
            def input_generator_fn(device):
                return torch.randn(shape, device=device, dtype=torch.float32).contiguous(memory_format=memory_format)
            return input_generator_fn

        def transformation_fn(tensor, **kwargs):
            return tensor.clone(**kwargs)

        formats_shapes = (
            (torch.channels_last, (4, 3, 8, 8)),
            (torch.channels_last_3d, (4, 3, 8, 8, 8)))

        for mf, shape in formats_shapes:
            self._test_memory_format_transformations(
                device, get_generator(mf, shape), transformation_fn, mf, True, default_is_preserve=True)

    def test_memory_format_factory_like_functions_preserve(self, device):
        def get_generator(memory_format, shape):
            def input_generator_fn(device):
                return torch.randn(shape, device=device, dtype=torch.float32).contiguous(memory_format=memory_format)
            return input_generator_fn

        transformation_fns = [
            lambda t, **kwargs: torch.zeros_like(t, **kwargs),
            lambda t, **kwargs: torch.ones_like(t, **kwargs),
            lambda t, **kwargs: torch.randint_like(t, 10, 100, **kwargs),
            lambda t, **kwargs: torch.randint_like(t, 100, **kwargs),
            lambda t, **kwargs: torch.randn_like(t, **kwargs),
            lambda t, **kwargs: torch.rand_like(t, **kwargs),
            lambda t, **kwargs: torch.full_like(t, 7, **kwargs),
            lambda t, **kwargs: torch.empty_like(t, **kwargs)]

        formats_shapes = (
            (torch.channels_last, (4, 3, 8, 8)),
            (torch.channels_last_3d, (4, 3, 8, 8, 8)))

        for mf, shape, in formats_shapes:
            for transformation_fn in transformation_fns:
                self._test_memory_format_transformations(
                    device, get_generator(mf, shape), transformation_fn, mf, compare_data=False, default_is_preserve=True)

    def test_memory_format_type_shortcuts(self, device):
        def get_generator(memory_format, shape, dtype):
            def input_generator_fn(device):
                return torch.randn(shape, device=device, dtype=dtype).clamp(0, 1) \
                    .round().contiguous(memory_format=memory_format)
            return input_generator_fn


        def get_fn(fn_name):
            def transformation_fn(tensor, **kwargs):
                fn = getattr(tensor, fn_name)
                return fn(**kwargs)
            return transformation_fn

        shortcuts = ['byte', 'char', 'double', 'bool', 'half', 'int', 'long', 'short']
        if device == 'cpu':
            shortcuts += ['bfloat16']

        formats_shapes = (
            (torch.channels_last, (4, 3, 8, 8)),
            (torch.channels_last_3d, (4, 3, 8, 8, 8)))

        for mf, shape in formats_shapes:
            for fn_name in shortcuts:
                self._test_memory_format_transformations(
                    device, get_generator(mf, shape, torch.float32), get_fn(fn_name), mf, default_is_preserve=True)

        # Test 'float' separately to avoid float->float no-op.
        for mf, shape in formats_shapes:
            self._test_memory_format_transformations(
                device, get_generator(mf, shape, torch.float64), get_fn('float'), mf, default_is_preserve=True)

    @onlyCUDA
    def test_memory_format_cpu_and_cuda_ops(self, device):
        def get_generator(memory_format, shape):
            def input_generator_fn(device):
                return torch.randn(shape, device=device, dtype=torch.float32).contiguous(memory_format=memory_format)
            return input_generator_fn

        def transformation_cpu_fn(tensor, **kwargs):
            return tensor.cpu(**kwargs)

        def transformation_cuda_fn(tensor, **kwargs):
            return tensor.cuda(**kwargs)

        formats_shapes = (
            (torch.channels_last, (4, 3, 8, 8)),
            (torch.channels_last_3d, (4, 3, 8, 8, 8)))

        for mf, shape in formats_shapes:
            self._test_memory_format_transformations(
                'cuda', get_generator(mf, shape), transformation_cpu_fn, mf, default_is_preserve=True)
            self._test_memory_format_transformations(
                'cpu', get_generator(mf, shape), transformation_cuda_fn, mf, default_is_preserve=True)

    # FIXME: move to test_serialization
    @onlyNativeDeviceTypes
    def test_pickle_gradscaler(self, device):
        # This test should pass in 3 cases for cuda:
        #  1. cuda is not available.
        #  2. cuda is available but device is not cuda.
        #  3. cuda is available and device is cuda.
        # In case 1, a and b disable themselves on construction and shouldn't try to pickle workhorse attributes.
        # In case 2, a and b are enabled.  Workhorse attributes participate in pickling, but none are lazy-inited
        # to cuda Tensors, because I don't want to do cuda things if device is not cuda.
        # In case 3, a and b are enabled and we may also try lazy-initing _scale to a cuda tensor.
        device = torch.device(device)
        try_lazy_inits = (True, False)
        GradScaler = partial(torch.GradScaler, device=device.type)
        for lazy_init_scale in try_lazy_inits:
            a = GradScaler(init_scale=3., growth_factor=4., backoff_factor=.5, growth_interval=2)
            if device.type == "cuda":
                self.assertTrue(not a.is_enabled() if torch.cuda.amp.common.amp_definitely_not_available() else a.is_enabled())
            else:
                self.assertTrue(a.is_enabled())
            if lazy_init_scale:
                # Dummy a.scale() call lazy-inits a._scale Tensor.
                a.scale(torch.tensor([4.0], dtype=torch.float32, device=device))
                self.assertTrue(a._scale.device.type == device.type)
            # The following three lines should work whether or not cuda is available.
            serialized = pickle.dumps(a)
            b = pickle.loads(serialized)
            self.assertEqual(b.is_enabled(), a.is_enabled())
            if a.is_enabled():
                self.assertEqual(b.get_scale(), 3.)
                self.assertEqual(b.get_growth_factor(), 4.)
                self.assertEqual(b.get_backoff_factor(), .5)
                self.assertEqual(b.get_growth_interval(), 2)
                self.assertEqual(b._init_growth_tracker, 0)
                # supplies a dummy key to test the defaultdict's default_factory
                self.assertEqual(b._per_optimizer_states["fdsa"],
                                 torch.amp.grad_scaler._refresh_per_optimizer_state())
                if lazy_init_scale:
                    self.assertEqual(b.scale(torch.tensor([4.0], dtype=torch.float32, device=device)), 12.0)

    # FIXME: move to test distributions
    def _test_multinomial_empty(self, device, replacement, num_samples):
        probs = torch.ones(0, 3, device=device)
        expected = torch.empty(0, num_samples, dtype=torch.int64)
        out = torch.multinomial(probs, num_samples=num_samples, replacement=replacement)
        self.assertEqual(out, expected)

    # FIXME: move to test distributions
    def test_multinomial_empty_w_replacement(self, device):
        self._test_multinomial_empty(device, True, 1)
        self._test_multinomial_empty(device, True, 2)

    # FIXME: move to test distributions
    def test_multinomial_empty_wo_replacement(self, device):
        self._test_multinomial_empty(device, False, 1)
        self._test_multinomial_empty(device, False, 2)

    @onlyNativeDeviceTypes
    @dtypes(torch.float, torch.double)
    def test_grad_scaling_unscale(self, device, dtype):
        device = torch.device(device)
        device0 = "cuda:0" if device.type == "cuda" else "cpu"
        inv_scale = torch.full((1,), 0.25, dtype=torch.float, device=device0)
        found_inf = torch.full((1,), 0.0, dtype=torch.float, device=device0)

        size = 20
        g = torch.full((size, size), 4.0, dtype=dtype, device=device0)
        ginf = g.clone()
        ginf[2, 2] = float('inf')
        gnan = g.clone()
        gnan[2, 2] = float('nan')

        # Tries selected combinations of
        #  - contiguous grads
        #  - g.clone().t() which is not contiguous but still non overlapping and dense
        #  - variants of g.clone()[:, :5] which are not non overlapping and dense
        # Non overlapping and dense grads route into a multi tensor apply kernel,
        # others use a fallback per-tensor kernel, so we should try both.
        cases = (
            ([g.clone(), g.clone()], False),
            ([g.clone(), g.clone().t()], False),
            ([g.clone(), g.clone()[:, :5]], False),
            ([g.clone()[:, :5], g.clone()[:, :5]], False),
            ([g.clone(), ginf.clone()], True),
            ([g.clone(), gnan.clone()], True),
            ([g.clone(), ginf.clone()[:, :5]], True),
            ([g.clone(), gnan.clone()[:, :5]], True),
            ([ginf.clone(), g.clone()[:, :5]], True),
            ([ginf.clone()[:, :5], g.clone()[:, :5]], True),
        )

        for grads, has_inf in cases:
            found_inf.zero_()
            torch._amp_foreach_non_finite_check_and_unscale_(grads, found_inf, inv_scale)
            if has_inf:
                self.assertEqual(found_inf, 1.0)
            else:
                self.assertEqual(found_inf, 0.0)
                for grad in grads:
                    self.assertEqual(grad, torch.ones_like(grad), rtol=1e-5, atol=1e-7)

        # When passing lists with mismatched dtypes to a raw
        # _amp_foreach_non_finite_check_and_unscale_ call on CUDA,
        # it's expected to fall back to single-tensor TensorIterator kernel.
        grads = [g.clone(), g.to(dtype=torch.float16)]
        torch._amp_foreach_non_finite_check_and_unscale_(grads, found_inf, inv_scale)
        for grad in grads:
            self.assertEqual(grad, torch.ones_like(grad), rtol=1e-5, atol=1e-7)

        # Passing lists with mismatched devices to a raw
        # _amp_foreach_non_finite_check_and_unscale_ call should raise errors.
        if device.type == "cuda" and TEST_MULTIGPU:
            with self.assertRaisesRegex(RuntimeError, r"Expected all tensors to be on the same device"):
                torch._amp_foreach_non_finite_check_and_unscale_([g.clone(), g.to(device="cuda:1")],
                                                                 found_inf,
                                                                 inv_scale)

        # Creates a list of grads with mismatched dtypes and devices, to ensure
        # scaler._unscale_grads_ organizes grads by dtype and device before calling
        # _amp_foreach_non_finite_check_and_unscale_ on each set.
        # If inject_inf >= 0, writes an inf into one grad for _unscale_grads_ to find.
        def perfect_storm_grads(inject_inf):
            grads = [g.clone(), g.clone()[:, :5], g.to(dtype=torch.float16), g.to(dtype=torch.float16)]
            if device.type == "cuda" and TEST_MULTIGPU:
                grads += [g.to(device="cuda:1"),
                          g.to(device="cuda:1")[:, :5],
                          g.to(device="cuda:1", dtype=torch.float16),
                          g.to(device="cuda:1", dtype=torch.float16)]
            if inject_inf >= 0:
                grads[inject_inf][2, 2] = float('inf')
            return grads

        GradScaler = partial(torch.GradScaler, device=device.type)
        scaler = GradScaler()
        dummy_params = [torch.empty_like(g) for g in perfect_storm_grads(-1)]
        dummy_opt = torch.optim.SGD(dummy_params, lr=1.)

        # Ensures the inf/nan checking can find an inf injected onto any grad in the perfect storm.
        for inject_inf in range(-1, len(dummy_params)):
            found_inf = torch.full((1,), 0.0, dtype=torch.float, device=device0)
            grads = perfect_storm_grads(inject_inf)
            for i, p in enumerate(dummy_params):
                p.grad = grads[i]
            found_inf_per_device = scaler._unscale_grads_(dummy_opt, inv_scale, found_inf, True)
            if inject_inf < 0:
                # No inf was injected, ensures unscaling worked normally.
                self.assertTrue(sum(v.item() for v in found_inf_per_device.values()) == 0)
                for grad in grads:
                    self.assertEqual(grad, torch.ones_like(grad), rtol=1e-5, atol=1e-7)
            else:
                # inf was injected, ensures inf was found.
                self.assertTrue(sum(v.item() for v in found_inf_per_device.values()) == 1)

    @onlyNativeDeviceTypes
    @dtypes(torch.float)
    def test_grad_scaling_update_scale(self, device, dtype):
        growth = 2.0
        backoff = 0.25
        growth_interval = 2
        scale = torch.full((1,), 4.0, dtype=dtype, device=device)
        growth_tracker = torch.full((1,), 0.0, dtype=torch.int32, device=device)
        found_inf = torch.full((1,), 0.0, dtype=torch.float, device=device)

        # Simulates 2 consecutive unskipped iterations
        torch._amp_update_scale_(scale, growth_tracker, found_inf, growth, backoff, growth_interval)
        self.assertEqual(growth_tracker, 1)
        self.assertEqual(scale, 4.0)
        torch._amp_update_scale_(scale, growth_tracker, found_inf, growth, backoff, growth_interval)
        self.assertEqual(growth_tracker, 0)
        self.assertEqual(scale, 8.0)

        # Simulates a skipped iteration
        found_inf.fill_(1.0)
        torch._amp_update_scale_(scale, growth_tracker, found_inf, growth, backoff, growth_interval)
        self.assertEqual(growth_tracker, 0)
        self.assertEqual(scale, 2.0)

    @skipIfTorchDynamo("Failed running call_function for sparse_coo_tensor. See https://github.com/pytorch/pytorch/issues/118856")
    @onlyNativeDeviceTypes
    @dtypes(torch.float)
    def test_grad_scaling_unscale_sparse(self, device, dtype):
        device = torch.device(device)
        scaler = torch.GradScaler(device=device.type)

        inv_scale = torch.full((1,), 0.25, dtype=dtype, device=device)
        found_inf = torch.empty((1,), dtype=dtype, device=device)
        cur = found_inf.device

        i = torch.tensor([[0, 1, 1],
                          [2, 0, 2]], device=device, dtype=torch.int64)
        v = torch.tensor([16., 32., 64.], device=device, dtype=torch.float)
        s = torch.sparse_coo_tensor(i, v, torch.Size([2, 3]), device=device, dtype=dtype)

        p = s.clone()
        if not p.is_sparse:
            raise AssertionError("expected p to be sparse")
        opt = torch.optim.SGD([p], lr=1.)

        p.grad = s.clone()
        found_inf.zero_()
        found_inf = scaler._unscale_grads_(opt, inv_scale, found_inf, False)[cur]
        self.assertEqual(found_inf, 0.0)
        self.assertEqual(p.grad.to_dense(), (s / 4).to_dense())

        v = torch.FloatTensor([16., 32., float('inf')])
        p.grad = torch.sparse_coo_tensor(i, v, torch.Size([2, 3]), device=device, dtype=dtype)
        found_inf.zero_()
        found_inf = scaler._unscale_grads_(opt, inv_scale, found_inf, False)[cur]
        self.assertEqual(found_inf, 1.0)

        v = torch.FloatTensor([16., 32., float('nan')])
        p.grad = torch.sparse_coo_tensor(i, v, torch.Size([2, 3]), device=device, dtype=dtype)
        found_inf.zero_()
        found_inf = scaler._unscale_grads_(opt, inv_scale, found_inf, False)[cur]
        self.assertEqual(found_inf, 1.0)

        p = s.clone().half()
        if not p.is_sparse:
            raise AssertionError("expected p to be sparse")
        opt = torch.optim.SGD([p], lr=1.)

        p.grad = s.clone().half()
        found_inf.zero_()
        found_inf = scaler._unscale_grads_(opt, inv_scale, found_inf, True)[cur]
        self.assertEqual(found_inf, 0.0)
        self.assertEqual(p.grad.to_dense(), (s.half() / 4).to_dense())

        # Creates fp16 sparse tensor with duplicated indices (uncoalesced).  The uncoalesced representation
        # does not overflow in fp16, but the coalesced representation would, because 64000 + 64000 > fp16 max.
        # _amp_non_finite_check_and_unscale_ should report an overflow here.
        i = torch.LongTensor([[0, 1, 0],
                              [2, 0, 2]])
        v = torch.FloatTensor([64000., 32., 64000.])
        p.grad = torch.sparse_coo_tensor(i, v, torch.Size([2, 3]), device=device, dtype=torch.float16)
        found_inf.zero_()
        found_inf = scaler._unscale_grads_(opt, inv_scale, found_inf, True)[cur]
        self.assertEqual(found_inf, 1.0)

    @onlyNativeDeviceTypes
    def test_grad_scaling_state_dict(self, device):
        device = torch.device(device)
        GradScaler = partial(torch.GradScaler, device=device.type)
        for lazy_init_scale in True, False:
            s0 = GradScaler(init_scale=3., growth_factor=4., backoff_factor=.5, growth_interval=2)
            s1 = GradScaler(init_scale=6., growth_factor=7., backoff_factor=.8, growth_interval=1)

            # sets a random value for load_state_dict to overwrite
            s1._init_growth_tracker = 7

            if lazy_init_scale:
                # Dummy scale() call to ensure the scale tensor is lazily initialized.
                s1.scale(torch.full((1,), 4.0, dtype=torch.float32, device=device))
                if "cuda" == device.type:
                    self.assertTrue(isinstance(s1._scale, torch.cuda.FloatTensor))
                else:
                    self.assertTrue(isinstance(s1._scale, torch.FloatTensor))

            s1.load_state_dict(s0.state_dict())

            self.assertEqual(s1.get_scale(), 3.)
            self.assertEqual(s1.get_growth_factor(), 4.)
            self.assertEqual(s1.get_backoff_factor(), .5)
            self.assertEqual(s1.get_growth_interval(), 2)
            self.assertEqual(s1._init_growth_tracker, 0)

    # _run_scaling_case generalizes some single-optimizer test logic to avoid too much copy-pasting below.
    def _run_scaling_case(self, device, run, unskipped, skipped, atol=1e-7, optimizer_ctor=torch.optim.SGD, optimizer_kwargs=None):
        # Ensure scaling can be disabled without changing user control flow.
        for enabled in True, False:
            (
                mod_control, mod_scaling, opt_control, opt_scaling, data, loss_fn, skip_iter,
            ) = _create_scaling_case(device=device, optimizer_ctor=optimizer_ctor, optimizer_kwargs=optimizer_kwargs)

            # For functionality, test with a modest initial scale, and an unrealistically-large growth factor
            # so any potential errors with the growth factor handling will be magnified.
            GradScaler = partial(torch.GradScaler, device=device)
            scaler = GradScaler(init_scale=128., growth_factor=2.0, enabled=enabled, growth_interval=1)

            _ = run(device, data, mod_control, opt_control, scaler, loss_fn, skip_iter, False)
            ret = run(device, data, mod_scaling, opt_scaling, scaler, loss_fn, skip_iter, True)

            # Allows run() to optionally return a different scaler instance.
            scaler = ret if ret else scaler

            # If scaling was enabled, the scale factor should have been multiplied by the growth factor
            # len(data) - skipped times and the backoff factor "skipped" times.
            if enabled:
                net_growth = scaler.get_growth_factor()**unskipped if unskipped > 0 else 1.0
                net_backoff = scaler.get_backoff_factor()**skipped if skipped > 0 else 1.0
                self.assertTrue(scaler.get_scale() == (128. * net_growth * net_backoff))
            else:
                self.assertTrue(scaler.get_scale() == 1.0)

            for c, s in zip(mod_control.parameters(), mod_scaling.parameters()):
                self.assertEqual(c.grad, s.grad, atol=atol, rtol=1e-05)

                c_state, s_state = opt_control.state[c], opt_scaling.state[s]
                for k in c_state:
                    self.assertEqual(c_state[k], s_state[k], atol=atol, rtol=1e-05, msg=k)

                self.assertEqual(c, s, atol=atol, rtol=1e-05)

    @onlyNativeDeviceTypes
    @parametrize("foreach, fused", [(None, None), (True, None), (None, True)])
    @optims(
        [optim for optim in optim_db if optim.optim_cls in [torch.optim.AdamW, torch.optim.Adam, torch.optim.SGD]],
        dtypes=[torch.float32]
    )
    def test_grad_scaling_autocast(self, device, dtype, optim_info, foreach, fused):
        try_pickle = False

        def run(device, data, model, optimizer, scaler, loss_fn, skip_iter, try_scaling_api):
            for i, (input, target) in enumerate(data):
                optimizer.zero_grad()
                with torch.autocast(device_type=device, dtype=torch.half, enabled=try_scaling_api):
                    output = model(input)
                    loss = loss_fn(output, target)
                if try_scaling_api:
                    scaler.scale(loss).backward()
                    if i == skip_iter and scaler.is_enabled():
                        with torch.no_grad():
                            model[1].weight.grad.fill_(float('inf'))
                    scaler.step(optimizer)
                    scaler.update()
                    if try_pickle:
                        scaler = pickle.loads(pickle.dumps(scaler))
                else:
                    loss.backward()
                    if (not scaler.is_enabled()) or (i != skip_iter):
                        optimizer.step()
            return scaler

        optimizer_ctor = optim_info.optim_cls

        # Compares no scaling + no autocasting against scaling + autocasting.
        # NOTE(mkozuki): With current way of testing, `torch.optim.Adam` is failing in spite of `foreach` and `fused`.
        #   Giving some flexibility to this test might help.
        context = contextlib.nullcontext
        if optimizer_ctor in (torch.optim.Adam, torch.optim.AdamW):
            from functools import partial
            context = partial(self.assertRaises, AssertionError)
        with context():
            # sets atol=1e-3 because we're comparing pure fp32 arithmetic vs a mixture of fp16 and fp32
            self._run_scaling_case(
                device, run, unskipped=3, skipped=1, atol=1e-3,
                optimizer_ctor=optimizer_ctor, optimizer_kwargs={"foreach": foreach, "fused": fused},
            )
            # this will be picked up by try_pickle within run():
            try_pickle = True
            self._run_scaling_case(
                device, run, unskipped=3, skipped=1, atol=1e-3,
                optimizer_ctor=optimizer_ctor, optimizer_kwargs={"foreach": foreach, "fused": fused},
            )

    # Make sure that the parameters become nonsense when scaled gradients are finite
    # but they get invalidated before `optimizer.step`, after `GradScaler.unscale_`

    def _test_params_invalidated_with_grads_invalidated_between_unscale_and_step(self, device, dtype, optim_info):
        optimizer_ctor = optim_info.optim_cls
        all_optim_inputs = _get_optim_inputs_including_global_cliquey_kwargs(
            device, dtype, optim_info, skip=("differentiable",))

        for optim_input in all_optim_inputs:
            model, _, optimizer, _, data, loss_fn, _ = _create_scaling_case(
                device, optimizer_ctor=optimizer_ctor, optimizer_kwargs=optim_input.kwargs,
            )
            scaler = torch.GradScaler(device=device, init_scale=128.0)

            for input, target in data:
                optimizer.zero_grad()
                with torch.autocast(device_type=device, dtype=torch.half):
                    output = model(input)
                    loss = loss_fn(output, target)
                scaler.scale(loss).backward()
                scaler.unscale_(optimizer)

                # deliberately break grads
                for j, param in enumerate(model.parameters()):
                    param.grad.copy_(torch.inf if j % 2 else torch.nan)

                scaler.step(optimizer)
                scaler.update()

            self.assertTrue(all((p.isnan().any() or p.isinf().any()) for p in model.parameters()))

    @onlyNativeDeviceTypes
    @optims(
        [optim for optim in optim_db if optim.optim_cls in [torch.optim.AdamW, torch.optim.Adam, torch.optim.SGD]],
        dtypes=[torch.float32]
    )
    def test_params_invalidated_with_grads_invalidated_between_unscale_and_step(self, device, dtype, optim_info):
        self._test_params_invalidated_with_grads_invalidated_between_unscale_and_step(device, dtype, optim_info)

    @onlyNativeDeviceTypes
    @optims(
        [optim for optim in optim_db if optim.optim_cls in [torch.optim.AdamW, torch.optim.Adam, torch.optim.SGD]],
        dtypes=[torch.float32]
    )
    @torch._inductor.config.patch("graph_partition", True)
    def test_params_invalidated_with_grads_invalidated_and_graph_partition(self, device, dtype, optim_info):
        self._test_params_invalidated_with_grads_invalidated_between_unscale_and_step(device, dtype, optim_info)

    @onlyNativeDeviceTypes
    def test_grad_scale_will_not_overflow(self, device):
        device = torch.device(device)
        model = torch.nn.Linear(5, 1).to(device)
        optimizer = torch.optim.Adam(model.parameters())
        scaler = torch.GradScaler(device=device.type, growth_interval=1, growth_factor=2**4, init_scale=1e38)
        optimizer.zero_grad()
        x = torch.randn(1, 5).to(device)
        y = 1e-30 * torch.randn(1, 1).to(device)
        l = ((model(x) - y) ** 2).mean()
        scaler.scale(l).backward()
        scaler.step(optimizer)
        scaler.update()
        if scaler._scale == float("inf") or scaler._scale != scaler._scale:
            raise AssertionError(f"scaler._scale should not be inf or nan, got {scaler._scale}")

    @onlyNativeDeviceTypes
    def test_grad_scaling_clipping(self, device):
        device = torch.device(device)

        def run(device, data, model, optimizer, scaler, loss_fn, skip_iter, try_scaling_api):
            max_norm = 0.2  # A reasonable value that actually has an effect, based on printouts of grads
            for i, (input, target) in enumerate(data):
                optimizer.zero_grad()
                output = model(input)
                loss = loss_fn(output, target)
                if try_scaling_api:
                    scaler.scale(loss).backward()
                    torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm * scaler.get_scale())
                    if i == skip_iter and scaler.is_enabled():
                        model[1].weight.grad.data.fill_(float('inf'))
                    scaler.step(optimizer)
                    scaler.update()
                else:
                    loss.backward()
                    torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm)
                    if (not scaler.is_enabled()) or (i != skip_iter):
                        optimizer.step()

        self._run_scaling_case(device.type, run, unskipped=3, skipped=1, atol=1e-5)

    @onlyNativeDeviceTypes
    def test_grad_scaling_clipping_separate_unscale(self, device):
        device = torch.device(device)

        def run(device, data, model, optimizer, scaler, loss_fn, skip_iter, try_scaling_api):
            max_norm = 0.2  # A reasonable value that actually has an effect, based on printouts of grads
            for i, (input, target) in enumerate(data):
                optimizer.zero_grad()
                output = model(input)
                loss = loss_fn(output, target)
                if try_scaling_api:
                    scaler.scale(loss).backward()
                    if i == skip_iter and scaler.is_enabled():
                        model[1].weight.grad.data.fill_(float('inf'))
                    scaler.unscale_(optimizer)
                    torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm, error_if_nonfinite=False)
                    scaler.step(optimizer)
                    scaler.update()
                else:
                    loss.backward()
                    torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm)
                    if (not scaler.is_enabled()) or (i != skip_iter):
                        optimizer.step()

        self._run_scaling_case(device.type, run, unskipped=3, skipped=1)

    @onlyNativeDeviceTypes
    def test_grad_scaling_penalty(self, device):
        device = torch.device(device)

        def run(device, data, model, optimizer, scaler, loss_fn, skip_iter, try_scaling_api):
            for i, (input, target) in enumerate(data):
                optimizer.zero_grad()
                output = model(input)
                loss = loss_fn(output, target)

                if try_scaling_api:
                    grad_params = torch.autograd.grad(scaler.scale(loss),
                                                      model.parameters(), create_graph=True)
                    inv_scale = 1. / scaler.get_scale()
                    grad_params = [p * inv_scale for p in grad_params]
                else:
                    grad_params = torch.autograd.grad(loss, model.parameters(), create_graph=True)

                grad_norm = 0
                for grad in grad_params:
                    grad_norm += grad.pow(2).sum()
                grad_norm = grad_norm.sqrt()
                loss = loss + grad_norm

                if try_scaling_api:
                    scaler.scale(loss).backward()
                    if i == skip_iter and scaler.is_enabled():
                        model[1].weight.grad.data.fill_(float('inf'))
                    scaler.step(optimizer)
                    scaler.update()
                else:
                    loss.backward()
                    if (not scaler.is_enabled()) or (i != skip_iter):
                        optimizer.step()

        self._run_scaling_case(device.type, run, unskipped=3, skipped=1)

    @onlyNativeDeviceTypes
    def test_grad_scaling_accumulation(self, device):
        device = torch.device(device)

        def run(device, data, model, optimizer, scaler, loss_fn, skip_iter, try_scaling_api):
            iters_to_accumulate = 2
            for i, (input, target) in enumerate(data):
                output = model(input)
                loss = loss_fn(output, target)
                loss = loss / iters_to_accumulate
                if try_scaling_api:
                    scaler.scale(loss).backward()
                else:
                    loss.backward()
                if (i + 1) % iters_to_accumulate == 0:
                    if try_scaling_api:
                        scaler.step(optimizer)
                        scaler.update()
                        optimizer.zero_grad()
                    else:
                        optimizer.step()
                        optimizer.zero_grad()

        self._run_scaling_case(device.type, run, unskipped=2, skipped=0)

    @onlyNativeDeviceTypes
    def test_grad_scaling_multiple(self, device):
        device = torch.device(device)
        # Tests gradient scaling with 2 models and 2 optimizers that both receive gradients from 2 losses.
        # Some of the logic here cannot reuse the generic helper functions created for the 1-optimizer cases.
        for enabled in True, False:
            mod_control0, mod_scaling0, opt_control0, opt_scaling0, data, loss_fn, skip_iter = \
                _create_scaling_case(device.type)
            mod_control1, mod_scaling1, opt_control1, opt_scaling1 = \
                _create_scaling_models_optimizers(device.type)

            GradScaler = partial(torch.GradScaler, device=device.type)
            scaler = GradScaler(init_scale=128., growth_factor=2.0, enabled=enabled, growth_interval=1)

            def run(model0, model1, optimizer0, optimizer1, try_scaling_api):
                for i, (input, target) in enumerate(data):
                    optimizer0.zero_grad()
                    optimizer1.zero_grad()
                    output0 = model0(input)
                    output1 = model1(input)
                    loss0 = loss_fn(0.3 * output0 + 0.7 * output1, target)
                    loss1 = loss_fn(0.6 * output0 - 0.4 * output1, target)

                    if try_scaling_api:
                        scaler.scale(loss0).backward(retain_graph=True)
                        scaler.scale(loss1).backward()
                        if i == skip_iter and scaler.is_enabled():
                            model1[1].weight.grad.data.fill_(float('inf'))

                        # As an additional stress test, separately unscale for one of the optimizers.
                        scaler.unscale_(optimizer0)

                        scaler.step(optimizer0)
                        scaler.step(optimizer1)
                        scaler.update()
                    else:
                        loss0.backward(retain_graph=True)
                        loss1.backward()
                        optimizer0.step()
                        if (not scaler.is_enabled()) or (i != skip_iter):
                            optimizer1.step()

            run(mod_control0, mod_control1, opt_control0, opt_control1, False)
            run(mod_scaling0, mod_scaling1, opt_scaling0, opt_scaling1, True)

            # The loss scale should have been multiplied by the growth factor 3 times and the backoff factor once.
            self.assertTrue(scaler.get_scale() == (128. * scaler.get_growth_factor()**3 *
                                                   scaler.get_backoff_factor()**1) if enabled else 1.0)

            for c, s in zip(chain(mod_control0.parameters(), mod_control1.parameters()),
                            chain(mod_scaling0.parameters(), mod_scaling1.parameters())):
                self.assertEqual(c, s, rtol=1e-5, atol=1e-7)

    @onlyNativeDeviceTypes
    def test_grad_scaler_pass_itself(self, device):
        device = torch.device(device)
        GradScaler = partial(torch.amp.GradScaler, device=device.type)

        class _PlaceHolderOptimizer(torch.optim.Optimizer):
            tester = self

            def __init__(self, params, defaults=None):
                if defaults is None:
                    defaults = {}
                super().__init__(params, defaults)
                self._step_supports_amp_scaling = True

        class Optimizer1(_PlaceHolderOptimizer):
            def step(self, closure=None, *, grad_scaler=None):
                self.tester.assertTrue(isinstance(grad_scaler, torch.amp.GradScaler))
                self.tester.assertFalse(hasattr(self, "grad_scale"))
                self.tester.assertFalse(hasattr(self, "found_inf"))

        class Optimizer2(_PlaceHolderOptimizer):
            def step(self, closure=None):
                self.tester.assertTrue(isinstance(self.grad_scale, torch.Tensor))
                self.tester.assertTrue(isinstance(self.found_inf, torch.Tensor))

        x = torch.randn(4, 4).to(device)
        m = torch.nn.Linear(4, 1).to(device)
        o1 = Optimizer1(m.parameters())
        o2 = Optimizer2(m.parameters())
        scaler = GradScaler(init_scale=2.0)

        with torch.autocast(device_type=device.type, dtype=torch.half):
            y = m(x)
            loss = y.mean()
        scaler.scale(loss).backward()
        with self.assertWarns(FutureWarning):
            scaler.step(o1)
        scaler.step(o2)
        scaler.update()

    @onlyNativeDeviceTypes
    def test_grad_scaler_deprecated_warning(self, device):
        device = torch.device(device)
        GradScaler = torch.cuda.amp.GradScaler if "cuda" == device.type else torch.cpu.amp.GradScaler

        with self.assertWarnsRegex(
            FutureWarning,
            rf"`torch.{device.type}.amp.GradScaler\(args...\)` is deprecated.",
        ):
            _ = GradScaler(init_scale=2.0)

    @dtypesIfCUDA(torch.float, torch.double, torch.half)
    @dtypesIfCPU(torch.float, torch.double, torch.bfloat16, torch.half)
    @dtypes(torch.float, torch.double)
    def test_multinomial_cpu(self, device, dtype):
        def make_prob_dist(shape, is_contiguous):
            if is_contiguous:
                if dtype == torch.half or dtype == torch.bfloat16:
                    return torch.zeros(shape, device=device).uniform_().to(dtype=dtype)
                return torch.zeros(shape, device=device, dtype=dtype).uniform_()
            elif len(shape) == 1:
                if dtype == torch.half or dtype == torch.bfloat16:
                    return torch.zeros((shape + [5]), device=device).uniform_().to(dtype=dtype)[:, 2]
                return torch.zeros((shape + [5]), device=device, dtype=dtype).uniform_()[:, 2]
            else:
                # num dim = 2
                new_shape = [2, shape[1], 7, 1, shape[0], 1, 10]
                if dtype == torch.half or dtype == torch.bfloat16:
                    prob_dist = torch.zeros(new_shape, device=device).uniform_().to(dtype=dtype)
                else:
                    prob_dist = torch.zeros(new_shape, device=device, dtype=dtype).uniform_()
                prob_dist = prob_dist.transpose(1, 4)
                prob_dist = prob_dist[1, :, 5, 0, :, 0, 4]
                if prob_dist.is_contiguous():
                    raise AssertionError("expected prob_dist to be non-contiguous")
                return prob_dist

    # FIXME: move to elementwise ternary test suite
    # As the test fails with Runtime Error not raised on XLA
    @onlyNativeDeviceTypes
    def test_where_scalar_handcrafted_values(self, device):
        # Tests ScalarxScalar, ScalarxTensor and TensorxScalar
        # variant of `where` against NumPy version with
        # handcrafted values.
        condition_shape = (5, 5)
        dtypes = (
            torch.bool, torch.uint8, torch.int8, torch.int16, torch.int64,
            torch.float16, torch.float32, torch.float64,
            torch.complex64, torch.complex128,
        )
        shapes = ((), (5,), (1, 5),)

        with torch.no_grad():
            tensors = (torch.empty(shape, dtype=dtype, device=device).fill_(17)
                       for shape, dtype in product(shapes, dtypes))

        # Use different values for `x` and `y`
        # as they are the output values which are compared.
        x_vals = (True, 3, 7.0, 1 + 0.5j)
        y_vals = itertools.chain((False, 4, 8.0, 2 + 0.5j), tensors)
        for x in x_vals:
            for y in y_vals:
                condition = torch.empty(*condition_shape, dtype=torch.bool, device=device).bernoulli_()
                common_dtype = torch.result_type(x, y)

                def check_equal(condition, x, y):
                    condition_np = condition.cpu().numpy()
                    x_np = x.cpu().numpy() if isinstance(x, torch.Tensor) else x
                    y_np = y.cpu().numpy() if isinstance(y, torch.Tensor) else y

                    # NumPy aggressively promotes to double, hence cast to output to correct dtype
                    expected = torch.from_numpy(np.where(condition_np, x_np, y_np)).to(common_dtype)
                    result = torch.where(condition, x, y)
                    self.assertEqual(expected, result)

                check_equal(condition, x, y)
                check_equal(condition, y, x)
                if self.device_type == "cuda":
                    check_equal(condition, torch.tensor(x), y)
                    check_equal(condition, y, torch.tensor(x))
                    if not isinstance(y, torch.Tensor):
                        check_equal(condition, torch.tensor(y), torch.tensor(x))
                    if isinstance(y, torch.Tensor) and y.ndim > 0:
                        check_equal(torch.tensor(True), x, y)
                        check_equal(torch.tensor(True), y, x)

    @onlyNativeDeviceTypes
    @dtypes(torch.uint16, torch.uint32, torch.uint64)
    def test_fill_barebones_unsigned(self, device, dtype):
        # Both FillKernel.cpp and FillKernel.cu dispatch over barebones
        # unsigned via AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES), but the fill
        # OpInfo does not enumerate them. Cover that dispatch directly.
        for shape in ((5,), (4, 5), (2, 3, 4)):
            t = torch.empty(shape, dtype=dtype, device=device).fill_(7)
            self.assertEqual(t.cpu(), torch.full(shape, 7, dtype=dtype))

        # Non-contiguous tensor (every other row of a larger allocation).
        big = torch.empty((8, 5), dtype=dtype, device=device)
        nc = big[::2]
        nc.fill_(11)
        self.assertEqual(nc.cpu(), torch.full((4, 5), 11, dtype=dtype))

        # Boundary fill value: max of each dtype.
        max_val = torch.iinfo(dtype).max
        t = torch.empty(8, dtype=dtype, device=device).fill_(max_val)
        self.assertEqual(t.cpu().tolist(), [max_val] * 8)

    @onlyNativeDeviceTypes
    @dtypes(torch.uint16, torch.uint32, torch.uint64)
    def test_where_barebones_unsigned(self, device, dtype):
        # The barebones unsigned dtypes are excluded from the broader
        # test_where_scalar_handcrafted_values because torch.result_type
        # does not support promoting them. Cover the same-dtype path
        # for where() against a CPU reference here.
        for shape in ((5,), (1, 5), (4, 5)):
            a = make_tensor(shape, dtype=dtype, device=device)
            b = make_tensor(shape, dtype=dtype, device=device)
            cond = torch.randint(0, 2, shape, dtype=torch.bool, device=device)
            out = torch.where(cond, a, b)
            expected = torch.where(cond.cpu(), a.cpu(), b.cpu()).to(device)
            self.assertEqual(out, expected)

        # Non-contiguous inputs.
        big = make_tensor((8, 10), dtype=dtype, device=device)
        a = big[::2]
        b = big[1::2]
        cond = torch.randint(0, 2, a.shape, dtype=torch.bool, device=device)
        out = torch.where(cond, a, b)
        expected = torch.where(cond.cpu(), a.cpu(), b.cpu()).to(device)
        self.assertEqual(out, expected)

    @onlyNativeDeviceTypes
    @dtypes(torch.uint16, torch.uint32, torch.uint64)
    def test_eq_ne_barebones_unsigned(self, device, dtype):
        # The eq/ne kernels dispatch over uint16/uint32/uint64 via
        # AT_DISPATCH_V2(... AT_EXPAND(AT_BAREBONES_UNSIGNED_TYPES)),
        # but the standard test_torch.py comparison sweeps don't list
        # those dtypes. Cover the same-dtype path here.
        for shape in ((5,), (4, 5), (2, 3, 4)):
            a = make_tensor(shape, dtype=dtype, device=device)
            b = a.clone()
            # Differ exactly one element so eq/ne both have True and False
            # to report. XOR with 1 toggles the low bit, so the perturbed
            # value is always in range and differs from old.
            old = int(a.view(-1)[0].item())
            b.view(-1)[0] = old ^ 1
            self.assertEqual(torch.eq(a, b), (a.cpu() == b.cpu()).to(device))
            self.assertEqual(torch.ne(a, b), (a.cpu() != b.cpu()).to(device))
            self.assertEqual(torch.eq(a, a), torch.ones_like(a, dtype=torch.bool))
            self.assertEqual(torch.ne(a, a), torch.zeros_like(a, dtype=torch.bool))

        # Non-contiguous: every-other-row slice.
        big = make_tensor((8, 4), dtype=dtype, device=device)
        a = big[::2]
        b = big[1::2]
        self.assertEqual(torch.eq(a, b), (a.cpu() == b.cpu()).to(device))
        self.assertEqual(torch.ne(a, b), (a.cpu() != b.cpu()).to(device))

        # Boundary value: max-int comparison.
        max_val = torch.iinfo(dtype).max
        a = torch.tensor([0, max_val, max_val - 1], dtype=dtype, device=device)
        b = torch.tensor([0, max_val, max_val], dtype=dtype, device=device)
        self.assertEqual(torch.eq(a, b).tolist(), [True, True, False])
        self.assertEqual(torch.ne(a, b).tolist(), [False, False, True])


    @skipIfTorchInductor("FIXME")
    def test_hook_remove(self, device):
        # Reference: https://github.com/pytorch/pytorch/issues/58354
        def _test_helper(remove_hook):
            def install_hook(tensor):
                handle = None

                def hook(tensor):
                    if remove_hook:
                        handle.remove()
                    return torch.zeros_like(tensor)
                handle = tensor.register_hook(hook)

            t = torch.ones((1, 5), device=device, requires_grad=True)
            install_hook(t)

            # First call to backward
            t.mean().backward()
            self.assertEqual(t.grad, torch.zeros_like(t))

            # Second call to backward
            t.mean().backward()
            if remove_hook:
                # After removing the hook, make sure the usual gradient is returned
                self.assertEqual(t.grad, 0.2 * torch.ones_like(t))
            else:
                self.assertEqual(t.grad, torch.zeros_like(t))

        _test_helper(remove_hook=True)
        _test_helper(remove_hook=False)

    # FIXME: get PyTorch/XLA to run test_testing
    # This test should ideally be in test_testing.py,
    # but since pytorch/xla runs tests from test_torch.py, we have it here.
    @skipXLA
    def test_skip_xla(self, device):
        if self.device_type == 'xla':
            # Should not reach here!
            self.assertTrue(False)

    # FIXME: get PyTorch/XLA to run test_testing
    # This test should ideally be in test_testing.py,
    # but since pytorch/xla runs tests from test_torch.py, we have it here.
    @expectedFailureXLA
    def test_expected_failure_xla(self, device):
        if self.device_type == 'xla':
            self.assertTrue(False)

    # FIXME: get PyTorch/XLA to run test_testing
    # This test should ideally be in test_testing.py,
    # but since pytorch/xla runs tests from test_torch.py, we have it here.
    def test_assertRaisesRegex_ignore_msg_non_native_device(self, device):
        # Verify that self.assertRaisesRegex only checks the Error and ignores
        # message for non-native devices.
        x = torch.randn((10, 3), device=device)
        t = torch.empty(10, dtype=torch.int64, device=device).random_(0, 3)
        invalid_weight = torch.randn(4, device=device)
        msg = "weight tensor should be defined either for all 3 classes or no classes"

        # XLA raises RuntimeError with a different message.
        with self.assertRaisesRegex(RuntimeError, msg):
            torch.nn.functional.nll_loss(x, t, weight=invalid_weight)

    @dtypes(*all_passthru_types_and(torch.chalf))
    def test_copy_(self, device, dtype):
        def can_cast(src_dtype, dst_dtype):
            # torch.can_cast(torch.int16, torch.uint8) returns True which
            # isn't actually safe-cast. Override for unsigned destinations:
            # only a bool source, or an unsigned source no wider than the
            # destination, round-trips losslessly. Narrowing (uint16 ->
            # uint8) is rejected.
            unsigned = {torch.uint8, torch.uint16, torch.uint32, torch.uint64}

            if dst_dtype in unsigned:
                return (src_dtype is torch.bool
                        or (src_dtype in unsigned
                            and src_dtype.itemsize <= dst_dtype.itemsize))
            return torch.can_cast(src_dtype, dst_dtype)

        def make_tensor_wrapper(shape, dtype):
            if dtype is not torch.complex32:
                # Make tensor does not support generating
                # complex32 tensor
                return make_tensor(shape, device=device, dtype=dtype)
            return torch.randn(shape, device=device, dtype=dtype)

        t = make_tensor_wrapper((50,), dtype)
        src_dtypes = all_passthru_types_and(torch.chalf)
        for src_dtype in src_dtypes:
            src = make_tensor_wrapper((50,), dtype=src_dtype)
            t.copy_(src)
            dst = make_tensor_wrapper((50, ), dtype=src_dtype)
            if can_cast(src_dtype, dtype):
                rtol = None
                atol = None
                if dtype in (torch.half, torch.complex32):
                    rtol = 1e-3
                    atol = 1e-3
                if dtype == torch.bfloat16:
                    rtol = 1e-2
                    atol = 1e-2
                self.assertEqual(src, dst.copy_(t), rtol=rtol, atol=atol)

    @dtypes(*all_types_complex_float8_and(
        torch.bool, torch.half, torch.bfloat16, torch.complex32,
        torch.uint16, torch.uint32, torch.uint64))
    def test_item(self, device, dtype):
        xla_unsupported_dtypes = [
            torch.uint16,
            torch.uint32,
            torch.uint64,
            torch.float8_e4m3fn,
            torch.float8_e5m2,
            torch.float8_e4m3fnuz,
            torch.float8_e5m2fnuz,
        ]
        if torch.device(device).type == 'xla' and dtype in xla_unsupported_dtypes:
            self.skipTest('uint16,32,64,float8 not implemented on XLA')
        t = torch.ones((), device=device, dtype=dtype)
        self.assertEqual(1, t.item())

    def test__local_scalar_dense_with_empty_tensor(self, device):
        input = torch.randn(0, device=device)
        with self.assertRaisesRegex(RuntimeError, "Empty tensor not supported"):
            torch.ops.aten._local_scalar_dense(input)

    @onlyNativeDeviceTypes
    def test_masked_scatter_inplace_noncontiguous(self, device):
        t = torch.zeros(5, 2, dtype=torch.long, device=device)
        t_non_contig = t.transpose(0, 1)
        t_contig = t_non_contig.contiguous()

        if not t_contig.is_contiguous():
            raise AssertionError("expected t_contig to be contiguous")
        if t_non_contig.is_contiguous():
            raise AssertionError("expected t_non_contig to be non-contiguous")

        mask = torch.tensor([[False, True], [False, True], [False, False], [True, True], [True, True]], device=device)
        mask_non_contig = mask.transpose(0, 1)
        mask_contig = mask_non_contig.contiguous()

        if not mask_contig.is_contiguous():
            raise AssertionError("expected mask_contig to be contiguous")
        if mask_non_contig.is_contiguous():
            raise AssertionError("expected mask_non_contig to be non-contiguous")

        # source is always converted to contiguous by the op.
        source = torch.tensor([[1, 2, 3, 4, 5], [6, 7, 8, 9, 9]], device=device)

        # t: contig, mask: contig
        expected = t_contig.masked_scatter_(mask_contig, source)

        # t: non-contig, mask: non-contig
        actual = t_non_contig.masked_scatter_(mask_non_contig, source)
        self.assertEqual(actual, expected)

        # t: contig, mask: non-contig
        actual = t_contig.masked_scatter_(mask_non_contig, source)
        self.assertEqual(actual, expected)

        # t: non-contig, mask: contig
        actual = t_non_contig.masked_scatter_(mask_contig, source)
        self.assertEqual(actual, expected)

    def test_tolist_with_grad(self, device):
        def f(x):
            result = x.tolist()
            for l, r in zip(result, [1.0, 2.0, 3.0]):
                if l != r:
                    raise AssertionError(f"expected {l} == {r}")
            return sum(x)

        x = torch.tensor([1.0, 2.0, 3.0], requires_grad=True, device=device)
        grad_f = torch.func.grad(f)
        result = grad_f(x)
        self.assertEqual(result.tolist() , [1.0, 1.0, 1.0])

    def test_swap_data_ptr_with_empty(self, device):
        x = torch.randn(4, device=device)
        y = torch.randn(4, device=device)
        y_data = y.clone()
        x_storage = x.untyped_storage()
        y_storage = y.untyped_storage()
        x_storage.resize_(0)
        self.assertEqual(x_storage.nbytes(), 0)
        x_storage._swap_data_ptr_(y_storage)
        self.assertEqual(x, y_data)
        self.assertEqual(y_storage.nbytes(), 0)

    def test_swap_data_ptr_non_empty(self, device):
        x = torch.randn(4, device=device)
        y = torch.randn(4, device=device)
        x_data = x.clone()
        y_data = y.clone()
        x_storage = x.untyped_storage()
        y_storage = y.untyped_storage()
        x_storage._swap_data_ptr_(y_storage)
        self.assertEqual(x, y_data)
        self.assertEqual(y, x_data)

    def test_swap_data_ptr_preserves_views(self, device):
        x = torch.randn(4, 4, device=device)
        x_view = x[1:3]
        y = torch.randn(4, 4, device=device)
        y_data = y.clone()
        x_storage = x.untyped_storage()
        y_storage = y.untyped_storage()
        x_storage.resize_(0)
        x_storage._swap_data_ptr_(y_storage)
        self.assertEqual(x, y_data)
        self.assertEqual(x_view, y_data[1:3])

    def test_swap_data_ptr_self(self, device):
        x = torch.randn(4, device=device)
        x_data = x.clone()
        s = x.untyped_storage()
        s._swap_data_ptr_(s)
        self.assertEqual(x, x_data)

    def test_swap_data_ptr_device_mismatch(self, device):
        x = torch.randn(4, device=device)
        y = torch.randn(4, device="meta")
        with self.assertRaisesRegex(RuntimeError, "same device"):
            x.untyped_storage()._swap_data_ptr_(y.untyped_storage())

    def test_swap_data_ptr_nbytes_mismatch(self, device):
        x = torch.randn(4, device=device)
        y = torch.randn(8, device=device)
        with self.assertRaisesRegex(RuntimeError, "same nbytes"):
            x.untyped_storage()._swap_data_ptr_(y.untyped_storage())


# Tests that compare a device's computation with the (gold-standard) CPU's.
class TestDevicePrecision(TestCase):
    exact_dtype = True

    # FIXME: move to indexing test suite
    @onlyCUDA
    def test_index_add_bfloat16(self, device):
        inp_tensor = torch.randn(5, 3, device='cpu').bfloat16()
        t = torch.tensor([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=torch.bfloat16, device='cpu')
        index = torch.tensor([0, 4, 2], device='cpu')
        out_cpu = inp_tensor.index_add(0, index, t)

        inp_tensor = inp_tensor.to(device=device)
        t = t.to(device=device)
        index = index.to(device=device)
        out_gpu = inp_tensor.index_add(0, index, t)

        self.assertEqual(out_cpu, out_gpu, atol=1e-2, rtol=0)

    # FIXME: move to serialization test suite
    def test_device_serialization(self, device):
        x = torch.randn(4, 4, device=device)

        with tempfile.NamedTemporaryFile() as f:
            torch.save(x, f)
            f.seek(0)
            x_copy = torch.load(f)

        self.assertEqual(x_copy, x)
        self.assertIs(type(x_copy), type(x))
        self.assertEqual(x_copy.device, x.device)

    # FIXME: move to serialization test suite
    @deviceCountAtLeast(2)
    def test_multidevice_serialization(self, devices):
        x = [torch.randn(4, 4, device=devices[0]),
             torch.randn(4, 4, device=devices[1])]

        with tempfile.NamedTemporaryFile() as f:
            torch.save(x, f)
            f.seek(0)
            x_copy = torch.load(f)

        for original, cp in zip(x, x_copy):
            self.assertEqual(cp, original)
            self.assertIs(type(cp), type(original))
            self.assertEqual(cp.device, original.device)

    # FIXME: move to data movement test suite
    @deviceCountAtLeast(1)
    def test_copy_noncontig(self, devices):
        def do_test(d0, d1):
            x = torch.tensor([1.5, 2.5, 3.5, 4.5, 5.5, 6.5], device=d0)
            y = torch.tensor([0, 0, 0, 0, 0, 0], device=d1)
            self.assertNotEqual(x.dtype, y.dtype)

            y[::2].copy_(x[::2])
            self.assertEqual(y, [1, 0, 3, 0, 5, 0])

        do_test('cpu', devices[0])
        do_test(devices[0], 'cpu')

        if len(devices) > 1:
            do_test(devices[0], devices[1])

    @deviceCountAtLeast(2)
    def test_type_conversions_same_device(self, devices):
        x = torch.randn(5, 5, device=devices[1])
        self.assertEqual(x.int().device, torch.device(devices[1]))
        self.assertEqual(x.type(torch.int).device, torch.device(devices[1]))
        self.assertEqual(x.to(torch.int).device, torch.device(devices[1]))

    @dtypesIfCUDA(torch.half, torch.float, torch.double,
                  torch.int8, torch.short, torch.int, torch.long,
                  torch.uint8)
    @dtypes(torch.float, torch.double,
            torch.int8, torch.short, torch.int, torch.long,
            torch.uint8)
    def test_from_sequence(self, device, dtype):
        seq = [list(range(i * 4, i * 4 + 4)) for i in range(5)]
        reference = torch.arange(0, 20).resize_(5, 4)
        self.assertEqual(torch.tensor(seq, dtype=dtype, device=device), reference, exact_dtype=False)

    # FIXME: moved to indexing test suite
    @deviceCountAtLeast(1)
    def test_advancedindex_mixed_cpu_devices(self, devices) -> None:
        def test(x: torch.Tensor, ia: torch.Tensor, ib: torch.Tensor) -> None:
            # test getitem
            self.assertEqual(x[:, ia, None, ib, 0].cpu(),
                             x.cpu()[:, ia.cpu(), None, ib.cpu(), 0])
            self.assertEqual(x[ia], x.cpu()[ia.cpu()])
            # test setitem
            x_clone1 = x.clone()
            x_clone2 = x.clone()
            first_shape = x[:, ia, None, ib, 0].shape
            second_shape = x[ia].shape
            x_clone1[:, ia, None, ib, 0] = torch.randn(first_shape).to(x_clone1)
            x_clone2[ia] = torch.randn(second_shape).to(x_clone2)

        cpu = torch.device('cpu')
        for device in devices:
            x = torch.randn(3, 4, 4, 4, 3)
            ia = torch.tensor([0, 2, 1])
            ib = torch.tensor([0, 2, 1])

            # Index device tensor with cpu tensor
            x = x.to(device)
            ia = ia.to(cpu)
            ib = ib.to(cpu)
            test(x, ia, ib)

            # Index device tensor with mixed cpu, device tensors
            x = x.to(device)
            ia = ia.to(cpu)
            ib = ib.to(device)
            test(x, ia, ib)

    @deviceCountAtLeast(1)
    def test_advancedindex_mixed_devices_error(self, devices) -> None:
        def test(x: torch.Tensor, ia: torch.Tensor, ib: torch.Tensor) -> None:
            # test getitem
            with self.assertRaisesRegex(RuntimeError, fr"indices should be either .* \({x.device}\)"):
                x[:, ia, None, ib, 0]
            with self.assertRaisesRegex(RuntimeError, fr"indices should be either .* \({x.device}\)"):
                x[ib]

        cpu = torch.device('cpu')
        for device in devices:
            # Index cpu tensor with device tensor
            x = torch.randn(3, 4, 4, 4, 3)
            ia = torch.tensor([0, 2, 1]).to(device)
            ib = torch.tensor([0, 2, 1]).to(device)
            test(x, ia, ib)

            # Index cpu tensor with mixed cpu, device tensors
            x = x.to(cpu)
            ia = ia.to(cpu)
            ib = ib.to(device)
            test(x, ia, ib)

            if len(devices) > 1:
                other_device = devices[0] if device == devices[1] else devices[1]

                # Index device tensor with mixed cpu, device tensors on different devices
                x = x.to(device)
                ia = ia.to(cpu)
                ib = ib.to(other_device)
                test(x, ia, ib)

    # FIXME: move to data movement test suite
    def test_copy_broadcast(self, device) -> None:
        x = torch.randn(10, 5)
        y = torch.randn(5, device=device)
        x.copy_(y)
        self.assertEqual(x[3], y)

        x = torch.randn(10, 5, device=device)
        y = torch.randn(5)
        x.copy_(y)
        self.assertEqual(x[3], y)

    # FIXME: move to an elementwise ternary test suite
    @dtypes(torch.int64, torch.float32, torch.float64)
    def test_clamp(self, device, dtype):
        test_args = [
            *product(
                [(100, 50), (10, 64), (97,)],  # shape
                (True, False),  # non-contiguous
            )
        ]

        for shape, noncontig in test_args:
            x = make_tensor(shape, device=device, dtype=dtype,
                            noncontiguous=noncontig)
            ub = make_tensor(shape, device=device, dtype=dtype,
                             noncontiguous=noncontig)
            lb = make_tensor(shape, device=device, dtype=dtype,
                             noncontiguous=noncontig)

            expect = x.max(lb).min(ub)
            actual = x.clamp(lb, ub)
            self.assertEqual(expect, actual)

            expect = np.clip(x.cpu().numpy(), lb.cpu().numpy(), ub.cpu().numpy())
            self.assertEqual(expect, actual)

            expect = x.max(lb)
            actual = x.clamp(min=lb)
            self.assertEqual(expect, actual)

            expect = x.min(ub)
            actual = x.clamp(max=ub)
            self.assertEqual(expect, actual)

            # Test broadcasting min & max
            expect = x.max(lb[0]).min(ub[..., :1])
            actual = x.clamp(lb[0], ub[..., :1])
            self.assertEqual(expect, actual)

            # Test broadcasting x
            expect = x[..., :1].max(lb).min(ub)
            actual = x[..., :1].clamp(lb, ub)
            self.assertEqual(expect, actual)

    def test_cuda_device_idx(self, device):
        x = torch.zeros(3, device=device)
        y = torch._efficientzerotensor(3, device=device)
        self.assertEqual(x.device, y.device)

# we implemented custom deallocation for subclasses, so it behooves
# us to make sure all of these bits work.  We'll use __del__ to
# track if objects die or not
class Tracker:
    def __init__(self, marker):
        self.marker = marker

    @staticmethod
    def make():
        marker = [False]
        return marker, Tracker(marker)

    def __del__(self):
        self.marker[0] = True

@contextlib.contextmanager
def disable_gc():
    if gc.isenabled():
        try:
            gc.disable()
            yield
        finally:
            gc.enable()
    else:
        yield

class TestTorch(TestCase):
    exact_dtype = True

    def test_dir(self):
        dir(torch)

    def test_wildcard_import(self):
        exec('from torch import *')

    def test_newaxis_numpy_comparison(self):
        def run_test(tensor, *idx):
            npt = tensor.numpy()
            self.assertEqual(tensor[idx], npt[idx])

        # 1D Tensor Tests
        x = torch.arange(0, 10)
        cases = [
            [None],
            [None, None],
            [Ellipsis, None],
            [None, Ellipsis],
            [2, None],
            [None, 2],
            [Ellipsis, None, 2],
            [Ellipsis, 2, None],
            [2, Ellipsis, None],
            [2, None, Ellipsis],
            [None, 2, Ellipsis],
            [None, Ellipsis, 2],
        ]

        for case in cases:
            run_test(x, *case)

        # 2D Tensor Tests
        x = torch.arange(0, 12).view(3, 4)
        cases = [
            [None],
            [None, None],
            [None, None, None],
            [Ellipsis, None],
            [Ellipsis, None, None],
            [None, Ellipsis],
            [None, Ellipsis, None],
            [None, None, Ellipsis],
            [2, None],
            [2, None, Ellipsis],
            [2, Ellipsis, None],
            [None, 2, Ellipsis],
            [Ellipsis, 2, None],
            [Ellipsis, None, 2],
            [None, Ellipsis, 2],
            [1, 2, None],
            [1, 2, Ellipsis, None],
            [1, Ellipsis, 2, None],
            [Ellipsis, 1, None, 2],
            [Ellipsis, 1, 2, None],
            [1, None, 2, Ellipsis],
            [None, 1, Ellipsis, 2],
            [None, 1, 2, Ellipsis],
        ]

        for case in cases:
            run_test(x, *case)

    def _consecutive(self, size, start=1):
        sequence = torch.ones(torch.tensor(size).prod(0)).cumsum(0)
        sequence.add_(start - 1)
        return sequence.resize_(*size)

    def test_newindex(self):
        reference = self._consecutive((3, 3, 3))
        # This relies on __index__() being correct - but we have separate tests for that

        def checkPartialAssign(index):
            reference = torch.zeros(3, 3, 3)
            reference[index] = self._consecutive((3, 3, 3))[index]
            self.assertEqual(reference[index], self._consecutive((3, 3, 3))[index], atol=0, rtol=0)
            reference[index] = 0
            self.assertEqual(reference, torch.zeros(3, 3, 3), atol=0, rtol=0)

        checkPartialAssign(0)
        checkPartialAssign(1)
        checkPartialAssign(2)
        checkPartialAssign((0, 1))
        checkPartialAssign((1, 2))
        checkPartialAssign((0, 2))
        checkPartialAssign(torch.LongTensor((0, 2)))

        with self.assertRaises(IndexError):
            reference[1, 1, 1, 1] = 1
        with self.assertRaises(IndexError):
            reference[1, 1, 1, (1, 1)] = 1
        with self.assertRaises(IndexError):
            reference[3, 3, 3, 3, 3, 3, 3, 3] = 1
        with self.assertRaises(IndexError):
            reference[0.0] = 1
        with self.assertRaises(TypeError):
            reference[0.0:2.0] = 1
        with self.assertRaises(IndexError):
            reference[0.0, 0.0:2.0] = 1
        with self.assertRaises(IndexError):
            reference[0.0, :, 0.0:2.0] = 1
        with self.assertRaises(IndexError):
            reference[0.0, ..., 0.0:2.0] = 1
        with self.assertRaises(IndexError):
            reference[0.0, :, 0.0] = 1

    # Test `torch._check*` functions
    def test_check(self):
        test_cases = [
            # check function, expected error
            (torch._check, RuntimeError),
            (torch._check_index, IndexError),
            (torch._check_value, ValueError),
            (torch._check_type, TypeError),
            (torch._check_not_implemented, NotImplementedError),
        ]

        for check_fn, expected_error in test_cases:
            # cond=True should not raise an error
            check_fn(True)

            # Test default failure message for cond=False
            default_message = 'Expected cond to be True'
            with self.assertRaisesRegex(expected_error, default_message):
                check_fn(False)

            # Test a simple failure message
            message = 'message'
            with self.assertRaisesRegex(expected_error, message):
                check_fn(False, lambda: message)

            # A plain string message (not wrapped in a callable) should be
            # accepted and used directly as the error message.
            with self.assertRaisesRegex(expected_error, 'plain string error text'):
                check_fn(False, 'plain string error text')

            # Test message with tensor
            def message():
                return torch.arange(4)

            with self.assertRaisesRegex(expected_error, re.escape(str(message()))):
                check_fn(False, message)

            # Test format string message
            def message():
                return f"{'test'} {[1, 2, 'a', True]} {True} {100} {torch.arange(4)}"

            with self.assertRaisesRegex(expected_error, re.escape(str(message()))):
                check_fn(False, message)

            # Test incorrect `cond` arg type
            with self.assertRaisesRegex(TypeError, 'cond must be a bool'):
                check_fn('wrong type')

            with self.assertRaisesRegex(
                TypeError, r'cond must be a bool.*torch[.]_check_tensor_all'
            ):
                check_fn(torch.tensor(True))

    # FIXME: move to indexing test suite
    def test_index_add(self):
        for device in get_all_device_types():
            for dest_contig, src_contig, index_contig in product([True, False], repeat=3):
                for other_sizes in ((), (4, 5)):
                    for dtype in [torch.int, torch.long]:
                        num_copy, num_dest = 3, 3
                        dest = torch.randn(num_dest, *other_sizes, device=device)
                        if not dest_contig:
                            dest = make_tensor(dest.shape, device=device, dtype=dest.dtype, noncontiguous=True)
                        src = torch.randn(num_copy, *other_sizes, device=device)
                        if not src_contig:
                            src = noncontiguous_like(src)
                        idx = torch.randperm(num_dest, dtype=dtype, device=device).narrow(0, 0, num_copy)
                        if not index_contig:
                            idx = noncontiguous_like(idx)
                        # index_add_ without alpha argument
                        dest2 = dest.clone()
                        dest.index_add_(0, idx, src)
                        for i in range(idx.size(0)):
                            dest2[idx[i]] += src[i]
                        self.assertEqual(dest, dest2)
                        # index_add_ with alpha argument
                        dest2 = dest.clone()
                        dest.index_add_(0, idx, src, alpha=2)
                        for i in range(idx.size(0)):
                            dest2[idx[i]] += src[i] * 2
                        self.assertEqual(dest, dest2)

    # FIXME: resolve comment below and move this to indexing test suite
    # add coverage for issue with atomic add that appeared only for
    # specific dtypes on cuda:
    # https://github.com/pytorch/pytorch/issues/29153
    def test_index_add_all_dtypes(self):
        for device in get_all_device_types():
            for dtype in get_all_math_dtypes(device):
                for idx_dtype in [torch.int, torch.long]:
                    size = [5, 5]
                    if dtype.is_floating_point or dtype.is_complex:
                        tensor = torch.rand(size, dtype=dtype, device=device)
                    elif dtype.is_signed:
                        tensor = torch.randint(-5, 15, size, dtype=dtype, device=device)
                    else:
                        tensor = torch.randint(0, 10, size, dtype=dtype, device=device)

                    # index_add calls atomicAdd on cuda.
                    zeros = torch.zeros(size, dtype=dtype, device=device)

                    added = zeros.index_add(0, torch.arange(0, size[0], dtype=idx_dtype, device=device), tensor)
                    self.assertEqual(added, tensor)

                    added = zeros.index_add(0, torch.arange(0, size[0], dtype=idx_dtype, device=device), tensor, alpha=-1)
                    self.assertEqual(added, -tensor)

    @unittest.mock.patch.object(torch._dynamo.config, "suppress_errors", False)
    @set_default_dtype(torch.double)
    def test_index_add_correctness(self):
        # Check whether index_add can get correct result when
        # alpha is 1, and dtype of index is torch.long,
        # i.e., using scatter_add
        def helper(dim, dtype, device, size_result, size_source):
            tensor = torch.zeros(size_result, dtype=dtype, device=device)
            index = torch.randint(0, size_result[dim], (size_source[dim],),
                                  dtype=torch.long, device=device)
            if dtype.is_floating_point or dtype.is_complex:
                source = torch.rand(size_source, dtype=dtype, device=device)
            elif dtype.is_signed:
                source = torch.randint(-2, 5, size_source, dtype=dtype, device=device)
            else:
                source = torch.randint(0, 5, size_source, dtype=dtype, device=device)

            ref_out = tensor.index_add(dim, index, source, alpha=2.) / 2.
            ref_out = ref_out.to(dtype=dtype)
            out = tensor.index_add(dim, index, source)

            # Determine tolerances based on dtype and device
            # Low-precision types (float16, bfloat16) on GPU have non-deterministic
            # accumulation order, leading to larger rounding differences.
            # See: https://github.com/pytorch/pytorch/issues/91184
            if device == 'cuda' and dtype in (torch.half, torch.bfloat16):
                # Relaxed tolerance for low-precision GPU accumulation
                atol, rtol = 1e-1, 1e-1
            elif device == 'cuda':
                atol, rtol = 1e-2, 1e-2
            else:
                # scatter_add uses fp32 as accumulate type, while index_add doesn't.
                atol, rtol = 1e-2, 1e-2

            self.assertEqual(out, ref_out.to(dtype=dtype), atol=atol, rtol=rtol)

        for dim in [-1, -2, -3]:
            for dtype in all_types_and_complex_and(torch.half, torch.bfloat16):
                for device in get_all_device_types():
                    for size in [(2, 512, 256), (5, 256, 256)]:
                        helper(dim, dtype, device, size, size)

                # Check bound
                result = torch.zeros(1, 512, 256, dtype=dtype)
                source = torch.ones(1, 512, 256, dtype=dtype)
                index = torch.ones(257).to(dtype=torch.long)
                self.assertRaises(RuntimeError, lambda: result.index_add_(dim, index, source))
                index = (torch.ones(256) * 257).to(dtype=torch.long)
                self.assertRaises(RuntimeError, lambda: result.index_add_(dim, index, source))

    def test_index_add_cornercase(self):
        for device in get_all_device_types():
            dest = torch.randn((), device=device)
            index = torch.tensor([0], device=device)
            source = torch.randn(1, 1, 1, device=device)
            with self.assertRaisesRegex(
                RuntimeError,
                r"source tensor shape must match self tensor shape, excluding the specified dimension",
            ):
                dest.index_add(0, index, source)

    def test_linspace_logspace(self):
        # Ensure the output does not require grad regardless of inputs requiring guard or not.
        # The output of factory functions should not be part of any computational graph.
        start = 0.0
        end = 3.0

        for step in [0, 1, 2]:
            self.assertFalse(
                torch.linspace(
                    torch.tensor(start, requires_grad=True),
                    torch.tensor(end, requires_grad=True), step
                ).requires_grad
            )
            self.assertFalse(torch.linspace(torch.tensor(start, requires_grad=True), end, step).requires_grad)
            self.assertFalse(torch.linspace(start, torch.tensor(end, requires_grad=True), step).requires_grad)
            self.assertFalse(
                torch.logspace(
                    torch.tensor(start, requires_grad=True),
                    torch.tensor(end, requires_grad=True), step
                ).requires_grad
            )
            self.assertFalse(torch.logspace(torch.tensor(start, requires_grad=True), end, step).requires_grad)
            self.assertFalse(torch.logspace(start, torch.tensor(end, requires_grad=True), step).requires_grad)

    # FIXME: move to shape ops test suite
    def test_unflatten(self):
        # test args: tensor, int, sizes
        self.assertEqual(torch.tensor([]).unflatten(0, (0, 1)), torch.empty(0, 1))
        self.assertEqual(torch.tensor([1]).unflatten(0, (1, 1)), torch.tensor([[1]]))
        self.assertEqual(torch.tensor([1, 2, 3, 4]).unflatten(0, (2, 2)), torch.tensor([[1, 2], [3, 4]]))
        self.assertEqual(torch.tensor([1, 2, 3, 4]).unflatten(0, [2, 2]), torch.tensor([[1, 2], [3, 4]]))
        self.assertEqual(torch.tensor([1, 2, 3, 4]).unflatten(0, torch.Size([2, 2])), torch.tensor([[1, 2], [3, 4]]))
        self.assertEqual(torch.ones(2, 10).unflatten(1, (5, 2)), torch.ones(2, 5, 2))
        self.assertEqual(torch.tensor([1, 2, 3, 4]).unflatten(0, (-1, 2)),
                         torch.tensor([[1, 2], [3, 4]]))
        self.assertEqual(torch.ones(2, 10).unflatten(1, (5, -1)),
                         torch.ones(2, 5, 2))
        self.assertEqual(torch.ones(2, 10).unflatten(1, (-1,)),
                         torch.ones(2, 10))
        self.assertEqual(torch.ones(2, 3 * 4 * 5 * 6).unflatten(1, (3, 4, -1, 6)),
                         torch.ones(2, 3, 4, 5, 6))
        self.assertEqual(torch.ones(2, 0, 2).unflatten(1, (3, -1, 4, 5)),
                         torch.ones(2, 3, 0, 4, 5, 2))

        # test invalid args: tensor, str, sizes
        with self.assertRaisesRegex(TypeError, r"unflatten\(\): argument 'dim' \(position 1\) must be int, not str"):
            torch.tensor([1]).unflatten('A', (1, 1))

        # test other invalid arguments
        with self.assertRaisesRegex(RuntimeError, r"sizes must be non-empty"):
            torch.tensor([1]).unflatten(0, [])
        with self.assertRaisesRegex(RuntimeError, r"Provided sizes \[2, 2\] don't multiply up to the size of dim 0 \(1\)"):
            torch.tensor([1]).unflatten(0, [2, 2])
        with self.assertRaisesRegex(RuntimeError, r".*Dimension specified as 0 but tensor has no dimensions.*"):
            torch.tensor(1).unflatten(0, [0])
        with self.assertRaisesRegex(RuntimeError, r"only one dimension can be inferred"):
            torch.randn(5, 10).unflatten(1, (-1, -1))
        with self.assertRaisesRegex(RuntimeError,
                                    r"Provided sizes \[-1, 4\] don't multiply up to the size of dim 1 \(10\)"):
            torch.randn(5, 10).unflatten(1, (-1, 4))
        with self.assertRaisesRegex(RuntimeError,
                                    r"the unspecified dimension size -1 can be any value and is ambiguous"):
            torch.randn(2, 0).unflatten(1, (2, -1, 0))

    # Test that warnings generated from C++ are translated to the correct type
    def test_warn_types(self):
        test_cases = [
            # function, warning type, message
            (torch._C._warn, UserWarning, r"Test message for TORCH_WARN"),
            (torch._C._warn_deprecation, DeprecationWarning, r"Test message for TORCH_WARN_DEPRECATION"),
        ]

        for fn, warning_type, message in test_cases:
            with warnings.catch_warnings(record=True) as w:
                warnings.resetwarnings()
                warnings.filterwarnings('always', category=warning_type)
                fn()

                self.assertEqual(len(w), 1, msg=lambda msg: f'{msg}\n{warning_type} not raised')
                warning = w[0].message
                self.assertTrue(isinstance(warning, warning_type), msg=lambda msg: f'{msg}\n{warning_type} not raised')
                self.assertTrue(re.search(
                    message,
                    str(warning)))

    def test_structseq_repr(self):
        a = torch.arange(250).reshape(5, 5, 10)
        expected = """
        torch.return_types.max(
        values=tensor([[ 40,  41,  42,  43,  44,  45,  46,  47,  48,  49],
                [ 90,  91,  92,  93,  94,  95,  96,  97,  98,  99],
                [140, 141, 142, 143, 144, 145, 146, 147, 148, 149],
                [190, 191, 192, 193, 194, 195, 196, 197, 198, 199],
                [240, 241, 242, 243, 244, 245, 246, 247, 248, 249]]),
        indices=tensor([[4, 4, 4, 4, 4, 4, 4, 4, 4, 4],
                [4, 4, 4, 4, 4, 4, 4, 4, 4, 4],
                [4, 4, 4, 4, 4, 4, 4, 4, 4, 4],
                [4, 4, 4, 4, 4, 4, 4, 4, 4, 4],
                [4, 4, 4, 4, 4, 4, 4, 4, 4, 4]]))"""
        self.assertEqual(repr(a.max(1)), textwrap.dedent(expected).strip())

    def test_is_same_size(self):
        t1 = torch.empty(3, 4, 9, 10)
        t2 = torch.empty(3, 4)
        t3 = torch.empty(1, 9, 3, 3)
        t4 = torch.empty(3, 4, 9, 10)

        self.assertFalse(t1.is_same_size(t2))
        self.assertFalse(t1.is_same_size(t3))
        self.assertTrue(t1.is_same_size(t4))

        nt1 = torch.nested.nested_tensor([torch.ones(2, 4), torch.ones(3, 4), torch.ones(5, 4)])
        nt2 = torch.nested.nested_tensor([torch.ones(2, 4), torch.ones(2, 4), torch.ones(2, 4)])
        nt3 = torch.nested.nested_tensor([torch.ones(2, 4, 5), torch.ones(2, 6, 5)])
        nt4 = torch.nested.nested_tensor([torch.ones(2, 4), torch.ones(3, 4), torch.ones(5, 4)])

        self.assertFalse(nt1.is_same_size(nt2))
        self.assertFalse(nt1.is_same_size(nt3))
        self.assertTrue(nt1.is_same_size(nt4))
        with self.assertRaisesRegex(RuntimeError, "Expected both self and other to be nested tensors."):
            t1.is_same_size(nt1)

        with self.assertRaisesRegex(RuntimeError, "Expected both self and other to be nested tensors."):
            nt1.is_same_size(t1)

    def test_tensor_set(self):
        t1 = torch.tensor([])
        t2 = torch.empty(3, 4, 9, 10).uniform_()
        t1.set_(t2)
        self.assertEqual(t1.storage()._cdata, t2.storage()._cdata)
        size = torch.Size([9, 3, 4, 10])
        t1.set_(t2.storage(), 0, size)
        self.assertEqual(t1.size(), size)
        t1.set_(t2.storage(), 0, tuple(size))
        self.assertEqual(t1.size(), size)
        self.assertEqual(t1.stride(), (120, 40, 10, 1))
        stride = (10, 360, 90, 1)
        t1.set_(t2.storage(), 0, size, stride)
        self.assertEqual(t1.stride(), stride)
        t1.set_(t2.storage(), 0, size=size, stride=stride)
        self.assertEqual(t1.size(), size)
        self.assertEqual(t1.stride(), stride)

        # test argument names
        t1 = torch.tensor([])
        # 1. case when source is tensor
        t1.set_(source=t2)
        self.assertEqual(t1.storage()._cdata, t2.storage()._cdata)
        # 2. case when source is storage
        t1.set_(source=t2.storage())
        self.assertEqual(t1.storage()._cdata, t2.storage()._cdata)
        # 3. case when source is storage, and other args also specified
        t1.set_(source=t2.storage(), storage_offset=0, size=size, stride=stride)
        self.assertEqual(t1.size(), size)
        self.assertEqual(t1.stride(), stride)

        t1 = torch.tensor([True, True], dtype=torch.bool)
        t2 = torch.tensor([False, False], dtype=torch.bool)
        t1.set_(t2)
        self.assertEqual(t1.storage()._cdata, t2.storage()._cdata)

    def test_tensor_set_errors(self):
        f_cpu = torch.randn((2, 3), dtype=torch.float32)
        d_cpu = torch.randn((2, 3), dtype=torch.float64)

        storage_offset = 0x41414141
        with self.assertRaisesRegex(RuntimeError, "out of bounds for storage of size"):
            t = torch.randn(1)
            t.set_(t.untyped_storage(), storage_offset, t.size())

        # if size changes, set_ will resize the storage inplace
        t = torch.randn(1)
        size = torch.Size([2, 3])
        t.set_(t.untyped_storage(), storage_offset, size)
        self.assertEqual(t.storage_offset(), storage_offset)
        self.assertEqual(t.untyped_storage().nbytes(), (storage_offset + size[0] * size[1]) * 4)

        # change dtype
        self.assertRaises(RuntimeError, lambda: f_cpu.set_(d_cpu.storage()))
        self.assertRaises(RuntimeError,
                          lambda: f_cpu.set_(d_cpu.storage(), 0, d_cpu.size(), d_cpu.stride()))
        self.assertRaises(RuntimeError, lambda: f_cpu.set_(d_cpu))

        # change device
        if torch.cuda.is_available():
            f_cuda = torch.randn((2, 3), dtype=torch.float32, device='cuda')

            # cpu -> cuda
            self.assertRaises(RuntimeError, lambda: f_cpu.set_(f_cuda.storage()))
            self.assertRaises(RuntimeError,
                              lambda: f_cpu.set_(f_cuda.storage(), 0, f_cuda.size(), f_cuda.stride()))
            self.assertRaises(RuntimeError, lambda: f_cpu.set_(f_cuda))

            # cuda -> cpu
            self.assertRaises(RuntimeError, lambda: f_cuda.set_(f_cpu.storage()))
            self.assertRaises(RuntimeError,
                              lambda: f_cuda.set_(f_cpu.storage(), 0, f_cpu.size(), f_cpu.stride()))
            self.assertRaises(RuntimeError, lambda: f_cuda.set_(f_cpu))

    # FIXME: move this test test_testing.py (along with allclose testing)
    # NOTE: test_equal will be deprecated in favor of torch.testing.assert_close
    #   once torch.testing is out of beta
    def test_equal(self):
        for device in ["cpu", "cuda"]:
            if device == "cuda" and not torch.cuda.is_available():
                continue

            # Contiguous, 1D
            t1 = torch.tensor((3., 4., 9., 10.), device=device)
            t2 = t1.contiguous()
            t3 = torch.tensor((1., 9., 3., 10.), device=device)
            t4 = torch.tensor((3., 4., 9.), device=device)
            t5 = torch.tensor([], device=device)
            self.assertTrue(t1.equal(t2))
            self.assertFalse(t1.equal(t3))
            self.assertFalse(t1.equal(t4))
            self.assertFalse(t1.equal(t5))
            self.assertTrue(torch.equal(t1, t2))
            self.assertFalse(torch.equal(t1, t3))
            self.assertFalse(torch.equal(t1, t4))
            self.assertFalse(torch.equal(t1, t5))

            # Non contiguous, 2D
            s = torch.tensor(((1, 2, 3, 4), (5, 6, 7, 8)), device=device)
            s1 = s[:, 1:3]
            s2 = s1.clone()
            s3 = torch.tensor(((2, 3), (6, 7)), device=device)
            s4 = torch.tensor(((0, 0), (0, 0)), device=device)

            self.assertFalse(s1.is_contiguous())
            self.assertTrue(s1.equal(s2))
            self.assertTrue(s1.equal(s3))
            self.assertFalse(s1.equal(s4))
            self.assertTrue(torch.equal(s1, s2))
            self.assertTrue(torch.equal(s1, s3))
            self.assertFalse(torch.equal(s1, s4))

            # Different dtypes
            x = torch.tensor((1, 2, 3), dtype=torch.float, device=device)
            y = torch.tensor((1, 2, 3), dtype=torch.int, device=device)
            z = torch.tensor((1, -1), dtype=torch.int, device=device)
            self.assertTrue(torch.equal(x, y))
            self.assertFalse(torch.equal(z, x))

            # Fast path test: tensor flags, like neg and conj
            neg_0 = torch.tensor((1, 2, 3), dtype=torch.float, device=device)
            neg_1 = neg_0._neg_view()
            self.assertTrue(neg_1.is_neg())
            self.assertEqual(neg_0.data_ptr(), neg_1.data_ptr())
            self.assertEqual(neg_0.storage_offset(), neg_1.storage_offset())
            self.assertEqual(neg_0.stride(), neg_1.stride())
            self.assertEqual(neg_0.size(), neg_1.size())
            self.assertFalse(torch.equal(neg_0, neg_1))
            # FIXME: Disable the following check due to the inductor failure
            # See https://github.com/pytorch/pytorch/issues/100340 and
            # https://github.com/pytorch/pytorch/issues/98175
            if not TEST_WITH_TORCHINDUCTOR:
                self.assertTrue(torch.equal(neg_0, neg_1._neg_view()))

            conj_0 = torch.tensor([1.0 + 2.0j, 2.0 + 1.0j], device=device)
            conj_1 = conj_0.conj()
            self.assertTrue(conj_1.is_conj())
            self.assertEqual(conj_0.data_ptr(), conj_1.data_ptr())
            self.assertEqual(conj_0.storage_offset(), conj_1.storage_offset())
            self.assertEqual(conj_0.stride(), conj_1.stride())
            self.assertEqual(conj_0.size(), conj_1.size())
            self.assertFalse(torch.equal(conj_0, conj_1))
            # FIXME: Disable the following check due to the inductor failure
            # See https://github.com/pytorch/pytorch/issues/100340 and
            # https://github.com/pytorch/pytorch/issues/98175
            if not TEST_WITH_TORCHINDUCTOR:
                self.assertTrue(torch.equal(conj_0, conj_1.conj()))

            # Fast path test: two tensors share the same storage, but different dtype
            s_0 = torch.rand((2, 3), dtype=torch.float, device=device)
            s_1 = s_0.view(dtype=torch.int32)
            self.assertEqual(s_0.data_ptr(), s_1.data_ptr())
            self.assertEqual(s_0.storage_offset(), s_1.storage_offset())
            self.assertEqual(s_0.stride(), s_1.stride())
            self.assertEqual(s_0.size(), s_1.size())
            self.assertFalse(torch.equal(s_0, s_1))

            # Fast path test: two tensors share the same storage, but different strides
            t_0 = torch.rand((2, 3), dtype=torch.float, device=device)
            t_1 = t_0.t()
            self.assertEqual(t_0.data_ptr(), t_1.data_ptr())
            self.assertEqual(t_0.storage_offset(), t_1.storage_offset())
            self.assertNotEqual(t_0.stride(), t_1.stride())
            self.assertNotEqual(t_0.size(), t_1.size())
            self.assertFalse(torch.equal(t_0, t_1))

            # Fast path: tensor containing `nan` is not equal to self
            for dtype in floating_and_complex_types():
                t = torch.tensor([1., float('nan')], dtype=dtype)
                self.assertFalse(torch.equal(t, t))

    def test_element_size(self):
        byte = torch.ByteStorage().element_size()
        char = torch.CharStorage().element_size()
        short = torch.ShortStorage().element_size()
        int = torch.IntStorage().element_size()
        long = torch.LongStorage().element_size()
        float = torch.FloatStorage().element_size()
        double = torch.DoubleStorage().element_size()
        bool = torch.BoolStorage().element_size()
        bfloat16 = torch.BFloat16Storage().element_size()
        complexfloat = torch.ComplexFloatStorage().element_size()
        complexdouble = torch.ComplexDoubleStorage().element_size()

        self.assertEqual(byte, torch.ByteTensor().element_size())
        self.assertEqual(byte, torch.ByteTensor().itemsize)
        self.assertEqual(char, torch.CharTensor().element_size())
        self.assertEqual(char, torch.CharTensor().itemsize)
        self.assertEqual(short, torch.ShortTensor().element_size())
        self.assertEqual(short, torch.ShortTensor().itemsize)
        self.assertEqual(int, torch.IntTensor().element_size())
        self.assertEqual(int, torch.IntTensor().itemsize)
        self.assertEqual(long, torch.LongTensor().element_size())
        self.assertEqual(long, torch.LongTensor().itemsize)
        self.assertEqual(float, torch.FloatTensor().element_size())
        self.assertEqual(float, torch.FloatTensor().itemsize)
        self.assertEqual(double, torch.DoubleTensor().element_size())
        self.assertEqual(double, torch.DoubleTensor().itemsize)
        self.assertEqual(bool, torch.BoolTensor().element_size())
        self.assertEqual(bool, torch.BoolTensor().itemsize)
        self.assertEqual(bfloat16, torch.tensor([], dtype=torch.bfloat16).element_size())
        self.assertEqual(bfloat16, torch.tensor([], dtype=torch.bfloat16).itemsize)
        self.assertEqual(complexfloat, torch.tensor([], dtype=torch.complex64).element_size())
        self.assertEqual(complexfloat, torch.tensor([], dtype=torch.complex64).itemsize)
        self.assertEqual(complexdouble, torch.tensor([], dtype=torch.complex128).element_size())
        self.assertEqual(complexdouble, torch.tensor([], dtype=torch.complex128).itemsize)

        self.assertGreater(byte, 0)
        self.assertGreater(char, 0)
        self.assertGreater(short, 0)
        self.assertGreater(int, 0)
        self.assertGreater(long, 0)
        self.assertGreater(float, 0)
        self.assertGreater(double, 0)
        self.assertGreater(bool, 0)
        self.assertGreater(bfloat16, 0)
        self.assertGreater(complexfloat, 0)
        self.assertGreater(complexdouble, 0)

        # These tests are portable, not necessarily strict for your system.
        self.assertEqual(byte, 1)
        self.assertEqual(char, 1)
        self.assertEqual(bool, 1)
        self.assertGreaterEqual(short, 2)
        self.assertGreaterEqual(int, 2)
        self.assertGreaterEqual(int, short)
        self.assertGreaterEqual(long, 4)
        self.assertGreaterEqual(long, int)
        self.assertGreaterEqual(double, float)

    def test_permute(self):
        orig = [1, 2, 3, 4, 5, 6, 7]
        perm = torch.randperm(7).tolist()
        x = torch.empty(*orig).fill_(0)
        new = [i - 1 for i in x.permute(*perm).size()]
        self.assertEqual(perm, new)
        self.assertEqual(x.size(), orig)

    @skipIfTorchDynamo("TorchDynamo fails with unknown reason")
    def test_reversed(self):
        val = torch.arange(0, 10)
        self.assertEqual(reversed(val), torch.arange(9, -1, -1))

        val = torch.arange(1, 10).view(3, 3)
        self.assertEqual(reversed(val), torch.tensor([[7, 8, 9], [4, 5, 6], [1, 2, 3]]))

        val = torch.tensor(42)
        self.assertEqual(reversed(val), torch.tensor(42))

    def test_contains(self):
        x = torch.arange(0, 10)
        self.assertEqual(4 in x, True)
        self.assertEqual(12 in x, False)

        x = torch.arange(1, 10).view(3, 3)
        val = torch.arange(1, 4)
        self.assertEqual(val in x, True)
        val += 10
        self.assertEqual(val in x, False)

        self.assertRaisesRegex(
            RuntimeError,
            f"Tensor.__contains__ only supports Tensor or scalar, but you passed in a {str}.",
            lambda: "foo" in x)
        self.assertRaisesRegex(
            RuntimeError,
            f"Tensor.__contains__ only supports Tensor or scalar, but you passed in a {type([1, 2])}.",
            lambda: [1, 2] in x)

    @skipIfTorchDynamo("TorchDynamo fails with unknown reason")
    def test_deepcopy_parameter(self):
        from copy import deepcopy
        l = torch.nn.Linear(10, 1)
        s = l.state_dict(keep_vars=True)
        self.assertEqual(torch.nn.Parameter, type(s['weight']))
        self.assertEqual(torch.nn.Parameter, type(s['bias']))

        s2 = deepcopy(s)
        self.assertEqual(torch.nn.Parameter, type(s2['weight']))
        self.assertEqual(torch.nn.Parameter, type(s2['bias']))

    def test_pickle(self):
        import pickle
        a = torch.randn(5, 5)
        serialized = pickle.dumps(a)
        b = pickle.loads(serialized)
        self.assertEqual(a, b)

    @skipIfTorchDynamo("TorchDynamo fails with unknown reason")
    def test_pickle_parameter(self):
        import pickle
        a = torch.nn.Parameter(torch.randn(5, 5))
        serialized = pickle.dumps(a)
        b = pickle.loads(serialized)
        self.assertTrue(isinstance(b, torch.nn.Parameter))
        self.assertEqual(a.requires_grad, b.requires_grad)
        self.assertEqual(a, b)

    @skipIfTorchDynamo("TorchDynamo fails with unknown reason")
    def test_pickle_parameter_no_requires_grad(self):
        import pickle
        a = torch.nn.Parameter(torch.randn(5, 5), requires_grad=False)
        serialized = pickle.dumps(a)
        b = pickle.loads(serialized)
        self.assertTrue(isinstance(b, torch.nn.Parameter))
        self.assertEqual(a.requires_grad, b.requires_grad)
        self.assertEqual(a, b)

    def test_pickle_dtype(self):
        t = torch.float32
        serialized = pickle.dumps(t)
        b = pickle.loads(serialized)
        self.assertTrue(isinstance(b, torch.dtype))
        self.assertEqual(id(b), id(t))

    def test_pickle_size(self):
        a = torch.rand(10).size()
        serialized = pickle.dumps(a)
        b = pickle.loads(serialized)
        self.assertTrue(isinstance(b, torch.Size))
        self.assertEqual(a, b)

    def test_pickle_function(self):
        # https://github.com/pytorch/pytorch/issues/37703
        a = torch.tanh
        serialized = pickle.dumps(a)
        b = pickle.loads(serialized)
        self.assertEqual(a, b)

    def test_generator_cpu(self):
        # test default generators are equal
        self.assertEqual(torch.default_generator, torch.default_generator)

        # tests Generator API
        # manual_seed, seed, initial_seed, get_state, set_state
        g1 = torch.Generator()
        g2 = torch.Generator()
        g1.manual_seed(12345)
        g2.manual_seed(12345)
        self.assertEqual(g1.initial_seed(), g2.initial_seed())

        g1.seed()
        g2.seed()
        self.assertNotEqual(g1.initial_seed(), g2.initial_seed())

        g1 = torch.Generator()
        g2_state = g2.get_state()
        g2_randn = torch.randn(1, generator=g2)
        g1.set_state(g2_state)
        g1_randn = torch.randn(1, generator=g1)
        self.assertEqual(g1_randn, g2_randn)

        default_state = torch.default_generator.get_state()
        q = torch.empty(100)
        g1_normal = q.normal_()
        g2 = torch.Generator()
        g2.set_state(default_state)
        g2_normal = q.normal_(generator=g2)
        self.assertEqual(g1_normal, g2_normal)

    def test_invalid_generator_raises(self):
        self.assertRaises(RuntimeError, lambda: torch.Generator('opengl'))

    def test_pickle_generator(self) -> None:
        devices = ['cpu']
        if torch.cuda.is_available():
            devices += ['cuda']

        for device in devices:
            with self.subTest(device=device):
                generator = torch.Generator(device=device).manual_seed(12345)
                if device != "cpu":
                    generator.set_offset(100)
                torch.randn((100, 100), generator=generator, device=device)  # progress the RNG state

                reserialized: torch.Generator = pickle.loads(pickle.dumps(generator))

                self.assertEqual(generator.device, reserialized.device)
                self.assertEqual(generator.initial_seed(), reserialized.initial_seed())
                if device != "cpu":
                    self.assertEqual(generator.get_offset(), reserialized.get_offset())
                torch.testing.assert_close(generator.get_state(), reserialized.get_state())

    def _sobol_reference_samples(self, scramble: bool) -> torch.Tensor:
        if not scramble:
            # theoretical values from Joe Kuo 2010
            return torch.tensor(
                [
                    [0., 0.],
                    [0.5, 0.5],
                    [0.75, 0.25],
                    [0.25, 0.75],
                    [0.375, 0.375],
                    [0.875, 0.875],
                    [0.625, 0.125],
                    [0.125, 0.625],
                ],
            )
        else:
            # theoretical values unknown: convergence properties checked
            return torch.tensor(
                [
                    [0.50860737, 0.29320504],
                    [0.07116939, 0.89594537],
                    [0.49354145, 0.11524881],
                    [0.93097717, 0.70244044],
                    [0.87266153, 0.23887917],
                    [0.31021884, 0.57600391],
                    [0.13687253, 0.42054182],
                    [0.69931293, 0.77336788],
                ],
            )

    def test_sobolengine_bounds(self, scramble: bool = False):
        engine = torch.quasirandom.SobolEngine(100, scramble=scramble, seed=123456)
        sample = engine.draw(512)
        self.assertTrue(torch.all(sample >= 0))
        self.assertTrue(torch.all(sample <= 1))

    def test_sobolengine_bounds_scrambled(self):
        self.test_sobolengine_bounds(scramble=True)

    def test_sobolengine_draw(self, scramble: bool = False):
        ref_sample = self._sobol_reference_samples(scramble=scramble)
        engine = torch.quasirandom.SobolEngine(2, scramble=scramble, seed=123456)
        sample = engine.draw(n=len(ref_sample))
        self.assertEqual(sample, ref_sample)
        self.assertEqual(engine.num_generated, len(ref_sample))

    def test_sobolengine_draw_scrambled(self):
        self.test_sobolengine_draw(scramble=True)

    def test_sobolengine_first_point(self):
        for dtype in (torch.float, torch.double):
            engine = torch.quasirandom.SobolEngine(2, scramble=False)
            sample = engine.draw(1, dtype=dtype)
            self.assertTrue(torch.all(sample == 0))
            self.assertEqual(sample.dtype, dtype)
        for dtype in (torch.float, torch.double):
            engine = torch.quasirandom.SobolEngine(2, scramble=True, seed=123456)
            sample = engine.draw(1, dtype=dtype)
            self.assertTrue(torch.all(sample != 0))
            self.assertEqual(sample.dtype, dtype)

    def test_sobolengine_continuing(self, scramble: bool = False):
        ref_sample = self._sobol_reference_samples(scramble=scramble)
        engine = torch.quasirandom.SobolEngine(2, scramble=scramble, seed=123456)
        n_half = len(ref_sample) // 2
        _ = engine.draw(n=n_half)
        sample = engine.draw(n=n_half)
        torch.testing.assert_close(sample, ref_sample[n_half:])

    def test_sobolengine_continuing_scrambled(self):
        self.test_sobolengine_continuing(scramble=True)

    def test_sobolengine_reset(self, scramble: bool = False):
        ref_sample = self._sobol_reference_samples(scramble=scramble)
        engine = torch.quasirandom.SobolEngine(2, scramble=scramble, seed=123456)
        _ = engine.draw(n=len(ref_sample) // 2)
        engine.reset()
        self.assertEqual(engine.num_generated, 0)
        sample = engine.draw(n=len(ref_sample))
        torch.testing.assert_close(sample, ref_sample)

    def test_sobolengine_reset_scrambled(self):
        self.test_sobolengine_reset(scramble=True)

    def test_sobolengine_fast_forward(self, scramble: bool = False):
        ref_sample = self._sobol_reference_samples(scramble=scramble)
        engine = torch.quasirandom.SobolEngine(2, scramble=scramble, seed=123456)
        engine.fast_forward(4)
        sample = engine.draw(n=4)
        torch.testing.assert_close(sample, ref_sample[4:])
        # alternate fast forwarding with sampling
        engine.reset()
        even_draws = []
        for i in range(8):
            if i % 2 == 0:
                even_draws.append(engine.draw())
            else:
                engine.fast_forward(1)
        torch.testing.assert_close(
            ref_sample[[i for i in range(8) if i % 2 == 0]],
            torch.from_numpy(np.concatenate(even_draws)),
        )

    def test_sobolengine_fast_forward_scrambled(self):
        self.test_sobolengine_fast_forward(scramble=True)

    def test_sobolengine_default_dtype(self):
        engine = torch.quasirandom.SobolEngine(dimension=3, scramble=True, seed=123456)
        # Check that default dtype is correctly handled
        self.assertEqual(engine.draw(n=5).dtype, torch.float32)
        with set_default_dtype(torch.float64):
            engine = torch.quasirandom.SobolEngine(dimension=3, scramble=True, seed=123456)
            # Check that default dtype is correctly handled (when set to float64)
            self.assertEqual(engine.draw(n=5).dtype, torch.float64)
            # Check that explicitly passed dtype is adhered to
            self.assertEqual(engine.draw(n=5, dtype=torch.float32).dtype, torch.float32)
            # Reinitialize the engine and check that first draw dtype is correctly handled
            engine = torch.quasirandom.SobolEngine(dimension=3, scramble=True, seed=123456)
            self.assertEqual(engine.draw(n=5, dtype=torch.float32).dtype, torch.float32)

    @skipIfTorchDynamo("np.float64 restored as float32 after graph break.")
    def test_sobolengine_distribution(self, scramble=False):
        d = 50
        engine = torch.quasirandom.SobolEngine(d, scramble=scramble, seed=123456)
        sample = engine.draw(1024)
        torch.testing.assert_close(
            torch.mean(sample, dim=0), torch.full((d,), 0.5), atol=2, rtol=2
        )
        torch.testing.assert_close(
            np.percentile(sample, 25, axis=0).astype(np.float64), np.repeat(0.25, d), atol=2, rtol=2
        )
        torch.testing.assert_close(
            np.percentile(sample, 75, axis=0).astype(np.float64), np.repeat(0.75, d), atol=2, rtol=2
        )

    @skipIfTorchDynamo("np.float64 restored as float32 after graph break.")
    def test_sobolengine_distribution_scrambled(self):
        self.test_sobolengine_distribution(scramble=True)

    def test_sobolengine_draw_base2(self, scramble=False):
        ref_sample = self._sobol_reference_samples(scramble=scramble)
        engine = torch.quasirandom.SobolEngine(2, scramble=scramble, seed=123456)
        sample = engine.draw_base2(2)
        self.assertEqual(ref_sample[:4], sample)
        # resampling still having N=2**n
        sample = engine.draw_base2(2)
        self.assertEqual(ref_sample[4:8], sample)

    def test_sobolengine_draw_base2_scrambled(self):
        self.test_sobolengine_draw_base2(scramble=True)

    def test_sobolengine_raise(self):
        maxdim = torch.quasirandom.SobolEngine.MAXDIM
        with self.assertRaises(ValueError):
            torch.quasirandom.SobolEngine(maxdim + 1)

    def test_sobolengine_high_dim(self):
        engine = torch.quasirandom.SobolEngine(1111, scramble=False, seed=123456)
        samples1 = engine.draw()
        vals1, counts1 = torch.unique(samples1, return_counts=True)
        samples2 = engine.draw()
        vals2, counts2 = torch.unique(samples2, return_counts=True)
        self.assertEqual(vals1.item(), 0.0)
        self.assertEqual(counts1.item(), 1111)
        self.assertEqual(vals2.item(), 0.5)
        self.assertEqual(counts1.item(), 1111)

    def test_parsing_int64(self):
        # accepts integer arguments
        x = torch.cumsum(torch.ones(5, 5), 0)
        self.assertEqual(x, torch.cumsum(torch.ones(5, 5), torch.tensor(0)))
        # doesn't accept floating point variables
        self.assertRaises(TypeError, lambda: torch.cumsum(torch.ones(5, 5), torch.tensor(0.)))

    def test_parsing_double(self):
        # accepts floating point and integer arguments
        x = torch.randn(2, 3)
        torch.isclose(x, x, 1, 1)
        self.assertTrue(torch.isclose(x, x, 1, 1).all())
        self.assertTrue(torch.isclose(x, x, 1.5, 1.).all())
        # accepts floating point and integer tensors
        self.assertTrue(torch.isclose(x, x, torch.tensor(1), torch.tensor(1)).all())
        self.assertTrue(torch.isclose(x, x, torch.tensor(1.5), torch.tensor(1.)).all())
        # doesn't accept variables with requires_grad
        self.assertRaises(TypeError,
                          lambda: torch.isclose(x, x, torch.tensor(1.5), torch.tensor(1., requires_grad=True)).all())

    def test_parsing_intlist(self):
        #  parse with integer variables
        self.assertEqual(torch.Size([3, 4]), torch.ones((torch.tensor(3), torch.tensor(4))).shape)
        self.assertEqual(torch.Size([3, 4]), torch.ones(torch.tensor(3), torch.tensor(4)).shape)
        # parse with numpy integers
        self.assertEqual(torch.Size([3, 4]), torch.ones((np.array(3), np.int64(4))).shape)
        self.assertEqual(torch.Size([3, 4]), torch.ones(np.array(3), np.int64(4)).shape)
        self.assertEqual(torch.Size([3, 4]), torch.ones((np.int64(3), np.array(4))).shape)
        self.assertEqual(torch.Size([3, 4]), torch.ones(np.int64(3), np.array(4)).shape)

        # fail parse with float variables
        self.assertRaises(TypeError, lambda: torch.ones((torch.tensor(3.), torch.tensor(4))))
        # fail parse with numpy floats
        self.assertRaises(TypeError, lambda: torch.ones((3., torch.tensor(4))))
        self.assertRaises(TypeError, lambda: torch.ones((np.array(3.), torch.tensor(4))))

        # fail parse with > 1 element variables
        self.assertRaises(TypeError, lambda: torch.ones(torch.tensor(3, 3)))
        self.assertRaises(TypeError, lambda: torch.ones(torch.tensor(3, 3)))
        self.assertRaises(TypeError, lambda: torch.ones(np.array(3, 3)))
        self.assertRaises(TypeError, lambda: torch.ones(np.array(3, 3)))

        # fail parse with additional positional args after intlist arg
        self.assertRaisesRegex(TypeError,
                               "received an invalid combination of arguments",
                               lambda: torch.LongTensor((6, 0), 1, 1, 0))
        self.assertRaisesRegex(TypeError,
                               "missing 1 required positional arguments",
                               lambda: torch.tensor().new_zeros((5, 5), 0))

        # ensure ones() throws an error when extra positional (non-keyword) arguments are given.
        self.assertRaises(TypeError, lambda: torch.ones((3, 3), torch.float32))

    def test_from_buffer(self):
        a = bytearray([1, 2, 3, 4])
        self.assertEqual(torch.ByteStorage.from_buffer(a).tolist(), [1, 2, 3, 4])
        shorts = torch.ShortStorage.from_buffer(a, 'big')
        self.assertEqual(shorts.size(), 2)
        self.assertEqual(shorts.tolist(), [258, 772])
        ints = torch.IntStorage.from_buffer(a, 'little')
        self.assertEqual(ints.size(), 1)
        self.assertEqual(ints[0], 67305985)
        f = bytearray([0x40, 0x10, 0x00, 0x00])
        floats = torch.FloatStorage.from_buffer(f, 'big')
        self.assertEqual(floats.size(), 1)
        self.assertEqual(floats[0], 2.25)

        f = bytearray([0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x10, 0x40])
        bools = torch.BoolStorage.from_buffer(f, 'big')
        self.assertEqual(bools.size(), 8)
        self.assertEqual(bools.tolist(), [False, True, True, True, True, True, True, True])
        self.assertEqual(bools.type(), 'torch.BoolStorage')
        self.assertTrue(isinstance(bools, torch.BoolStorage))

        f = bytearray(b'\x80\x02\x8a\nl\xfc\x9cF\xf9 j\xa8P\x19.\x80\x02M\xe9')
        bools = torch.BoolStorage.from_buffer(f, 'big')
        self.assertEqual(bools.size(), 19)

        f = bytearray(b'\0x4A')
        bools = torch.BoolStorage.from_buffer(f, 'big')
        self.assertEqual(bools.size(), 4)
        self.assertEqual(bools.tolist(), [False, True, True, True])
        bytes = torch.ByteStorage.from_buffer(a)
        self.assertEqual(bytes.nbytes(), 4)
        self.assertEqual(bytes.tolist(), [1, 2, 3, 4])
        self.assertTrue(isinstance(bytes, torch.ByteStorage))

    def test_storage_error(self):
        quantized_storages = [
            torch.QInt32Storage,
            torch.QInt8Storage,
            torch.QUInt2x4Storage,
            torch.QUInt4x2Storage,
            torch.QUInt8Storage,
        ]

        with self.assertRaisesRegex(RuntimeError, r"Only child classes of _LegacyStorage can be instantiated"):
            torch.storage._LegacyStorage()

        for storage_class in torch._storage_classes:
            if storage_class in [torch.UntypedStorage, torch.TypedStorage]:
                continue

            device = 'cuda' if storage_class.__module__ == 'torch.cuda' else 'cpu'
            dtype = storage_class.dtype

            if device == 'cuda' and not torch.cuda.is_available():
                continue

            # Legacy <type>Storage constructor errors
            with self.assertRaisesRegex(RuntimeError, r"'device' cannot be specified"):
                storage_class(device='cpu')

            with self.assertRaisesRegex(RuntimeError, r"'dtype' cannot be specified"):
                storage_class(dtype=torch.float)

            with self.assertRaisesRegex(TypeError, r"got an unexpected keyword"):
                storage_class(sdlkjf=torch.float)

            with self.assertRaisesRegex(RuntimeError, r"Too many positional arguments"):
                storage_class(0, 0)

            with self.assertRaisesRegex(TypeError, r"invalid data type"):
                storage_class('string')

            with self.assertRaisesRegex(TypeError, r"Argument type not recognized"):
                storage_class(torch.tensor([]))

            s = storage_class()

            with self.assertRaisesRegex(RuntimeError, r"No positional arguments"):
                storage_class(0, wrap_storage=s.untyped())

            with self.assertRaisesRegex(TypeError, r"must be UntypedStorage"):
                storage_class(wrap_storage=s)

            if torch.cuda.is_available():
                if storage_class in quantized_storages:
                    with self.assertRaisesRegex(RuntimeError, r"Cannot create CUDA storage with quantized dtype"):
                        s.cuda()

                else:

                    if s.is_cuda:
                        s_other_device = s.cpu()
                    else:
                        s_other_device = s.cuda()

                    with self.assertRaisesRegex(RuntimeError, r"Device of 'wrap_storage' must be"):
                        storage_class(wrap_storage=s_other_device.untyped())

            # TypedStorage constructor errors
            with self.assertRaisesRegex(RuntimeError, r"No positional arguments"):
                torch.TypedStorage(0, wrap_storage=s.untyped(), dtype=dtype)

            with self.assertRaisesRegex(RuntimeError, r"Argument 'dtype' must be specified"):
                torch.TypedStorage(wrap_storage=s.untyped())

            with self.assertRaisesRegex(TypeError, r"Argument 'dtype' must be torch.dtype"):
                torch.TypedStorage(wrap_storage=s.untyped(), dtype=0)

            with self.assertRaisesRegex(RuntimeError, r"Argument 'device' should not be specified"):
                torch.TypedStorage(wrap_storage=s.untyped(), dtype=dtype, device=device)

            with self.assertRaisesRegex(TypeError, r"Argument 'wrap_storage' must be UntypedStorage"):
                torch.TypedStorage(wrap_storage=s, dtype=dtype)

            with self.assertRaisesRegex(RuntimeError, r"Storage device not recognized"):
                torch.TypedStorage(dtype=dtype, device='xla')

            if torch.cuda.is_available():
                if storage_class in quantized_storages:
                    with self.assertRaisesRegex(RuntimeError, r"Cannot create CUDA storage with quantized dtype"):
                        torch.TypedStorage(dtype=dtype, device='cuda')

            with self.assertRaisesRegex(TypeError, r"Argument type not recognized"):
                torch.TypedStorage(torch.tensor([]), dtype=dtype, device=device)

            with self.assertRaisesRegex(RuntimeError, r"Too many positional arguments"):
                torch.TypedStorage(0, 0, dtype=dtype, device=device)

            if isinstance(s, torch.TypedStorage):
                s_other = torch.TypedStorage([1, 2, 3, 4], device=device, dtype=dtype)

                with self.assertRaisesRegex(RuntimeError, r'cannot set item'):
                    s.fill_(s_other)

    def test_storage_error_no_attribute(self):
        storage_classes = [
            torch.cuda.ByteStorage,
            torch.cuda.FloatStorage,
        ]
        for storage_class in storage_classes:
            with self.assertRaisesRegex(RuntimeError, r'Not available for CUDA storage'):
                storage_class.from_buffer()

            with self.assertRaisesRegex(RuntimeError, r'Not available for CUDA storage'):
                storage_class._new_with_weak_ptr()

            with self.assertRaisesRegex(RuntimeError, r'Not available for CUDA storage'):
                storage_class._new_shared_filename(0, 0, 0)

    def test_storage_casts(self):
        storage = torch.IntStorage([-1, 0, 1, 2, 3, 4])
        self.assertEqual(storage.size(), 6)
        self.assertEqual(storage.tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertEqual(storage.type(), 'torch.IntStorage')
        self.assertIs(storage.dtype, torch.int32)

        floatStorage = storage.float()
        self.assertEqual(floatStorage.size(), 6)
        self.assertEqual(floatStorage.tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertEqual(floatStorage.type(), 'torch.FloatStorage')
        self.assertEqual(floatStorage.int().tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertIs(floatStorage.dtype, torch.float32)

        halfStorage = storage.half()
        self.assertEqual(halfStorage.size(), 6)
        self.assertEqual(halfStorage.tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertEqual(halfStorage.type(), 'torch.HalfStorage')
        self.assertEqual(halfStorage.int().tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertIs(halfStorage.dtype, torch.float16)

        bfloat16Storage = storage.bfloat16()
        self.assertEqual(bfloat16Storage.size(), 6)
        self.assertEqual(bfloat16Storage.tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertEqual(bfloat16Storage.type(), 'torch.BFloat16Storage')
        self.assertEqual(bfloat16Storage.int().tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertIs(bfloat16Storage.dtype, torch.bfloat16)

        longStorage = storage.long()
        self.assertEqual(longStorage.size(), 6)
        self.assertEqual(longStorage.tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertEqual(longStorage.type(), 'torch.LongStorage')
        self.assertEqual(longStorage.int().tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertIs(longStorage.dtype, torch.int64)

        shortStorage = storage.short()
        self.assertEqual(shortStorage.size(), 6)
        self.assertEqual(shortStorage.tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertEqual(shortStorage.type(), 'torch.ShortStorage')
        self.assertEqual(shortStorage.int().tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertIs(shortStorage.dtype, torch.int16)

        doubleStorage = storage.double()
        self.assertEqual(doubleStorage.size(), 6)
        self.assertEqual(doubleStorage.tolist(), [-1.0, 0.0, 1.0, 2.0, 3.0, 4.0])
        self.assertEqual(doubleStorage.type(), 'torch.DoubleStorage')
        self.assertEqual(doubleStorage.int().tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertIs(doubleStorage.dtype, torch.float64)

        charStorage = storage.char()
        self.assertEqual(charStorage.size(), 6)
        self.assertEqual(charStorage.tolist(), [-1.0, 0.0, 1.0, 2.0, 3.0, 4.0])
        self.assertEqual(charStorage.type(), 'torch.CharStorage')
        self.assertEqual(charStorage.int().tolist(), [-1, 0, 1, 2, 3, 4])
        self.assertIs(charStorage.dtype, torch.int8)

        byteStorage = storage.byte()
        self.assertEqual(byteStorage.size(), 6)
        self.assertEqual(byteStorage.tolist(), [255, 0, 1, 2, 3, 4])
        self.assertEqual(byteStorage.type(), 'torch.ByteStorage')
        self.assertEqual(byteStorage.int().tolist(), [255, 0, 1, 2, 3, 4])
        self.assertIs(byteStorage.dtype, torch.uint8)

        boolStorage = storage.bool()
        self.assertEqual(boolStorage.size(), 6)
        self.assertEqual(boolStorage.tolist(), [True, False, True, True, True, True])
        self.assertEqual(boolStorage.type(), 'torch.BoolStorage')
        self.assertEqual(boolStorage.int().tolist(), [1, 0, 1, 1, 1, 1])
        self.assertIs(boolStorage.dtype, torch.bool)

        complexfloat_storage = torch.ComplexFloatStorage([-1, 0, 1 + 2j, 2.5j, 3.5, 4 - 2j])
        self.assertEqual(complexfloat_storage.size(), 6)
        self.assertEqual(complexfloat_storage.tolist(), [-1, 0, 1 + 2j, 2.5j, 3.5, 4 - 2j])
        self.assertEqual(complexfloat_storage.type(), 'torch.ComplexFloatStorage')
        self.assertIs(complexfloat_storage.dtype, torch.complex64)

        complexdouble_storage = complexfloat_storage.complex_double()
        self.assertEqual(complexdouble_storage.size(), 6)
        self.assertEqual(complexdouble_storage.tolist(), [-1, 0, 1 + 2j, 2.5j, 3.5, 4 - 2j])
        self.assertEqual(complexdouble_storage.type(), 'torch.ComplexDoubleStorage')
        self.assertIs(complexdouble_storage.dtype, torch.complex128)

    def test_storage_byteswap(self):
        input = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
        swapped_8bytes = [7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8]
        swapped_4bytes = [3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12]
        swapped_2bytes = [1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14]
        swapped_1byte = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]

        storage = torch.storage.TypedStorage(input, dtype=torch.uint8)._untyped_storage

        storage_f64 = storage.__copy__()
        storage_f64.byteswap(torch.float64)
        self.assertEqual(storage_f64.tolist(), swapped_8bytes)

        storage_f32 = storage.__copy__()
        storage_f32.byteswap(torch.float32)
        self.assertEqual(storage_f32.tolist(), swapped_4bytes)

        storage_f16 = storage.__copy__()
        storage_f16.byteswap(torch.float16)
        self.assertEqual(storage_f16.tolist(), swapped_2bytes)

        storage_bf16 = storage.__copy__()
        storage_bf16.byteswap(torch.bfloat16)
        self.assertEqual(storage_bf16.tolist(), swapped_2bytes)

        storage_i64 = storage.__copy__()
        storage_i64.byteswap(torch.int64)
        self.assertEqual(storage_i64.tolist(), swapped_8bytes)

        storage_i32 = storage.__copy__()
        storage_i32.byteswap(torch.int32)
        self.assertEqual(storage_i32.tolist(), swapped_4bytes)

        storage_i16 = storage.__copy__()
        storage_i16.byteswap(torch.int16)
        self.assertEqual(storage_i16.tolist(), swapped_2bytes)

        storage_i8 = storage.__copy__()
        storage_i8.byteswap(torch.int8)
        self.assertEqual(storage_i8.tolist(), swapped_1byte)

        storage_ui8 = storage.__copy__()
        storage_ui8.byteswap(torch.uint8)
        self.assertEqual(storage_ui8.tolist(), swapped_1byte)

        storage_bool = storage.__copy__()
        storage_bool.byteswap(torch.bool)
        self.assertEqual(storage_bool.tolist(), swapped_1byte)

        storage_c128 = storage.__copy__()
        storage_c128.byteswap(torch.complex128)
        self.assertEqual(storage_c128.tolist(), swapped_8bytes)

        storage_c64 = storage.__copy__()
        storage_c64.byteswap(torch.complex64)
        self.assertEqual(storage_c64.tolist(), swapped_4bytes)

    # Test that internal versions of functions related to TypedStorage do not
    # produce a deprecation warning
    def test_typed_storage_internal_no_warning(self):
        s0 = torch.FloatStorage(10)
        s0_untyped = s0.untyped()
        t0 = torch.randn(10)

        funcs = [
            lambda: torch.FloatStorage(_internal=True),
            lambda: torch.TypedStorage(
                dtype=torch.float,
                device='cpu',
                _internal=True),
            lambda: torch.TypedStorage(
                wrap_storage=s0_untyped,
                dtype=s0.dtype,
                _internal=True),
            lambda: torch.FloatStorage._dtype,
            lambda: s0._resize_(20),
            lambda: s0._size(),
            lambda: s0._untyped_storage,
            lambda: s0._is_shared(),
            lambda: s0._share_memory_(),
            lambda: s0._pickle_storage_type(),
            lambda: s0._setitem(slice(0, s0._size()), 1),
            lambda: s0._element_size(),
            lambda: s0._deepcopy({}),
            lambda: s0._data_ptr(),
            lambda: s0._nbytes(),
            lambda: t0._typed_storage(),
        ]

        if torch.cuda.is_available():
            s1 = torch.cuda.FloatStorage(10)
            s1_untyped = s1.untyped()
            t1 = torch.randn(10, device='cuda')

            funcs += [
                lambda: torch.cuda.FloatStorage(_internal=True),
                lambda: torch.TypedStorage(
                    dtype=torch.float,
                    device='cuda',
                    _internal=True),
                lambda: torch.TypedStorage(
                    wrap_storage=s1_untyped,
                    dtype=s1.dtype,
                    _internal=True),
                lambda: torch.cuda.FloatStorage._dtype,
                lambda: s1._resize_(20),
                lambda: s1._size(),
                lambda: s1._untyped_storage,
                lambda: s1._is_shared(),
                lambda: s1._share_memory_(),
                lambda: s1._pickle_storage_type(),
                lambda: s1._setitem(slice(0, s1._size()), 1),
                lambda: s1._element_size(),
                lambda: s1._deepcopy({}),
                lambda: s1._data_ptr(),
                lambda: s1._nbytes(),
                lambda: t1._typed_storage(),
            ]

        # Check that each of the TypedStorage internal function calls do not
        # produce a deprecation warning
        for f in funcs:
            with warnings.catch_warnings():
                warnings.filterwarnings('error', "TypedStorage is deprecated")
                f()

    # Test that public functions related to TypedStorage produce a deprecation
    # warning
    @skipIfTorchInductor("FIXME")
    def test_typed_storage_deprecation_warning(self):
        s0 = torch.FloatStorage(10)
        funcs = [
            lambda: torch.FloatStorage(),
            lambda: torch.FloatStorage.dtype,
            lambda: s0.fill_(0),
            lambda: s0.is_cuda,
            lambda: s0.untyped(),
            lambda: len(s0),
            lambda: s0[0],
        ]

        if torch.cuda.is_available():
            s1 = torch.cuda.FloatStorage(10)
            funcs += [
                lambda: torch.cuda.FloatStorage(),
                lambda: torch.cuda.FloatStorage.dtype,
                lambda: s1.fill_(0),
                lambda: s1.is_cuda,
                lambda: s1.untyped(),
                lambda: len(s1),
                lambda: s1[0],
            ]

        # Check that each of the TypedStorage function calls produce a warning
        # if warnings are reset between each
        for f in funcs:
            with AlwaysWarnTypedStorageRemoval(True):
                with warnings.catch_warnings(record=True) as w:
                    warnings.resetwarnings()
                    f()
                    self.assertEqual(len(w), 1, msg=str([str(a) for a in w]))
                    warning = w[0].message
                    self.assertTrue(warning, DeprecationWarning)
                    self.assertTrue(re.search(
                        '^TypedStorage is deprecated',
                        str(warning)))

        # Test that only the first warning is raised by default
        torch.storage._reset_warn_typed_storage_removal()
        with warnings.catch_warnings(record=True) as w:
            warnings.resetwarnings()
            torch.FloatStorage()
            torch.randn(10).storage()
            self.assertEqual(len(w), 1, msg=str([str(a) for a in w]))
            warning = w[0].message
            self.assertTrue(re.search(
                '^TypedStorage is deprecated',
                str(warning)))
            # Check the line of code from the warning's stack
            with open(w[0].filename, encoding="utf-8") as f:
                code_line = f.readlines()[w[0].lineno - 1]
            self.assertTrue(re.search(re.escape('torch.FloatStorage()'), code_line))

        # Check that warnings are not emitted if it happened in the past
        with warnings.catch_warnings(record=True) as w:
            warnings.resetwarnings()
            torch.FloatStorage()
            torch.randn(10).storage()
            self.assertEqual(len(w), 0, msg=str([str(a) for a in w]))

    def test_from_file(self):
        def assert_with_filename(filename):
            size = 10000
            s1 = torch.FloatStorage.from_file(filename, True, size)
            t1 = torch.FloatTensor(s1).copy_(torch.randn(size))
            self.assertEqual(s1.data_ptr(), torch.FloatTensor(s1).data_ptr())

            # check mapping
            s2 = torch.FloatStorage.from_file(filename, True, size)
            t2 = torch.FloatTensor(s2)
            self.assertEqual(t1, t2, atol=0, rtol=0)

            # check changes to t1 from t2
            rnum = random.uniform(-1, 1)
            t1.fill_(rnum)
            self.assertEqual(t1, t2, atol=0, rtol=0)

            # check changes to t2 from t1
            rnum = random.uniform(-1, 1)
            t2.fill_(rnum)
            self.assertEqual(t1, t2, atol=0, rtol=0)

            # release the tensors
            del s1, t1, s2, t2

        with TemporaryFileName() as fname:
            assert_with_filename(fname)

        if IS_FILESYSTEM_UTF8_ENCODING:
            with TemporaryDirectoryName(suffix='\u4e2d\u6587') as dname, TemporaryFileName(dir=dname) as fname:
                assert_with_filename(fname)

    def test_torch_from_file(self):
        def assert_with_filename(filename):
            size = 10000
            s1 = torch.from_file(filename, True, size, dtype=torch.float)
            t1 = torch.FloatTensor(s1).copy_(torch.randn(size))

            # check mapping
            s2 = torch.from_file(filename, True, size, dtype=torch.float)
            t2 = torch.FloatTensor(s2)
            self.assertEqual(t1, t2, atol=0, rtol=0)

            # check changes to t1 from t2
            rnum = random.uniform(-1, 1)
            t1.fill_(rnum)
            self.assertEqual(t1, t2, atol=0, rtol=0)

            # check changes to t2 from t1
            rnum = random.uniform(-1, 1)
            t2.fill_(rnum)
            self.assertEqual(t1, t2, atol=0, rtol=0)

            # release the tensors
            del s1, t1, s2, t2

        with TemporaryFileName() as fname:
            assert_with_filename(fname)

        if IS_FILESYSTEM_UTF8_ENCODING:
            with TemporaryDirectoryName(suffix='\u4e2d\u6587') as dname, TemporaryFileName(dir=dname) as fname:
                assert_with_filename(fname)

    def test_print(self):
        default_type = torch.tensor([]).type()
        for t in torch._tensor_classes:
            if t == torch.HalfTensor:
                continue  # HalfTensor does not support fill
            if t.is_sparse:
                continue
            if t.is_cuda and not torch.cuda.is_available():
                continue
            obj = t(100, 100).fill_(1)
            obj.__repr__()
            str(obj)
        # test half tensor
        obj = torch.rand(100, 100, device='cpu').half()
        obj.__repr__()
        str(obj)
        for t in torch._storage_classes:
            if t == torch.BFloat16Storage:
                continue  # Fix once fill is enabled for bfloat16
            if t.is_cuda and not torch.cuda.is_available():
                continue
            if t == torch.BoolStorage or t == torch.cuda.BoolStorage:
                obj = t(100).fill_(True)
            else:
                obj = t(100).fill_(1)
            obj.__repr__()
            str(obj)

        # test complex tensor
        # complex tensor print uses two formatters, one for real values
        # and the other for imag values. this is consistent with numpy
        x = torch.tensor([2.3 + 4j, 7 + 6j])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([2.3000+4.j, 7.0000+6.j])''')

        # test complex half tensor
        x = torch.tensor([1.25 + 4j, -7. + 6j], dtype=torch.chalf)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([ 1.2500+4.j, -7.0000+6.j], dtype=torch.complex32)''')

        # test scientific notation for complex tensors
        x = torch.tensor([1e28 + 2j , -1e-28j])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1.0000e+28+2.0000e+00j, -0.0000e+00-1.0000e-28j])''')

        # test big integer
        x = torch.tensor(2341234123412341)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor(2341234123412341)''')

        # test scientific notation
        x = torch.tensor([1e28, 1e-28])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1.0000e+28, 1.0000e-28])''')

        # test scientific notation using set_printoptions
        x = torch.tensor([1e2, 1e-2])
        torch.set_printoptions(sci_mode=True)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1.0000e+02, 1.0000e-02])''')
        torch.set_printoptions(sci_mode=False)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([100.0000,   0.0100])''')
        torch.set_printoptions(sci_mode=None)  # reset to the default value

        # test no leading space if all elements positive
        x = torch.tensor([1, 2])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1, 2])''')

        # test for leading space if there are negative elements
        x = torch.tensor([1, -2])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([ 1, -2])''')

        # test inf and nan
        x = torch.tensor([4, inf, 1.5, -inf, 0, nan, 1])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([4.0000,    inf, 1.5000,   -inf, 0.0000,    nan, 1.0000])''')

        y = torch.tensor([4, inf, complex(1.5, inf), complex(-inf, 4), 0, complex(nan, inf), complex(3, nan)])
        self.assertEqual(y.__repr__(), str(y))
        expected_str = '''\
tensor([4.0000+0.j,    inf+0.j, 1.5000+infj,   -inf+4.j, 0.0000+0.j,    nan+infj,
        3.0000+nanj])'''
        self.assertExpectedInline(str(y), expected_str)

        # test dtype
        with set_default_dtype(torch.float):
            x = torch.tensor([1e-324, 1e-323, 1e-322, 1e307, 1e308, 1e309], dtype=torch.float64)
            self.assertEqual(x.__repr__(), str(x))
            expected_str = '''\
tensor([ 0.0000e+00, 9.8813e-324, 9.8813e-323, 1.0000e+307, 1.0000e+308,
                inf], dtype=torch.float64)'''
            self.assertExpectedInline(str(x), expected_str)

        # test changing default dtype
        with set_default_dtype(torch.float64):
            self.assertEqual(x.__repr__(), str(x))
            expected_str = '''\
tensor([ 0.0000e+00, 9.8813e-324, 9.8813e-323, 1.0000e+307, 1.0000e+308,
                inf])'''
            self.assertExpectedInline(str(x), expected_str)

        # test summary
        x = torch.zeros(10000)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([0., 0., 0.,  ..., 0., 0., 0.])''')

        # test internal summary function
        x = torch.rand(1, 20, 5, 30)
        summary = torch._tensor_str.get_summarized_data(x)
        self.assertEqual(summary.shape, (1, 6, 5, 6))
        first_and_last = [0, 1, 2, -3, -2, -1]
        self.assertEqual(summary, x[:, first_and_last][..., first_and_last])

        # test device
        if torch.cuda.is_available():
            x = torch.tensor([123], device='cuda:0')
            self.assertEqual(x.__repr__(), str(x))
            self.assertExpectedInline(str(x), '''tensor([123], device='cuda:0')''')

            # test changing default to cuda
            torch.set_default_tensor_type(torch.cuda.FloatTensor)
            self.assertEqual(x.__repr__(), str(x))
            self.assertExpectedInline(str(x), '''tensor([123])''')

            # test printing a tensor on a different gpu than current one.
            if torch.cuda.device_count() >= 2:
                with torch.cuda.device(1):
                    self.assertEqual(x.__repr__(), str(x))
                    self.assertExpectedInline(str(x), '''tensor([123], device='cuda:0')''')

            # test printing cpu tensor when default device is cuda
            y = torch.tensor([123], device='cpu')
            self.assertEqual(y.__repr__(), str(y))
            self.assertExpectedInline(str(y), '''tensor([123], device='cpu')''')
        torch.set_default_tensor_type(default_type)


        # test integral floats and requires_grad
        x = torch.tensor([123.], requires_grad=True)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([123.], requires_grad=True)''')

        # test non-contiguous print
        # sliced tensor should have > PRINT_OPTS.threshold elements
        x = torch.ones(100, 2, 2, 10)
        y = x.as_strided(size=(100, 2, 10), stride=(2 * 2 * 10, 2 * 10, 1))
        self.assertEqual(str(y), y.__repr__())
        expected_str = '''\
tensor([[[1., 1., 1.,  ..., 1., 1., 1.],
         [1., 1., 1.,  ..., 1., 1., 1.]],

        [[1., 1., 1.,  ..., 1., 1., 1.],
         [1., 1., 1.,  ..., 1., 1., 1.]],

        [[1., 1., 1.,  ..., 1., 1., 1.],
         [1., 1., 1.,  ..., 1., 1., 1.]],

        ...,

        [[1., 1., 1.,  ..., 1., 1., 1.],
         [1., 1., 1.,  ..., 1., 1., 1.]],

        [[1., 1., 1.,  ..., 1., 1., 1.],
         [1., 1., 1.,  ..., 1., 1., 1.]],

        [[1., 1., 1.,  ..., 1., 1., 1.],
         [1., 1., 1.,  ..., 1., 1., 1.]]])\
'''

        self.assertExpectedInline(str(y), expected_str)

        x = torch.ones(100, 2, 2, 10) * (1 + 1j)
        y = x.as_strided(size=(100, 2, 10), stride=(2 * 2 * 10, 2 * 10, 1))
        self.assertEqual(str(y), y.__repr__())
        expected_str = '''\
tensor([[[1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j],
         [1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j]],

        [[1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j],
         [1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j]],

        [[1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j],
         [1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j]],

        ...,

        [[1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j],
         [1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j]],

        [[1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j],
         [1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j]],

        [[1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j],
         [1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j]]])\
'''
        self.assertExpectedInline(str(y), expected_str)

        # test print 0-dim tensor: there's no 0-dim in Numpy, we match arrayprint style
        x = torch.tensor(0.00002)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor(2.0000e-05)''')

        # test print boolean tensor
        x = torch.tensor([True])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([True])''')

        x = torch.tensor(True)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor(True)''')

        # [Numpy] test print float in sci_mode when min < 0.0001.
        x = torch.tensor([0.00002])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([2.0000e-05])''')

        # [Numpy] test print complex in sci_mode when real_min < 0.0001 and (or) imag_min < 0.0001.
        x = torch.tensor([0.00002]) * (1 + 1j)
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([2.0000e-05+2.0000e-05j])''')

        # [Numpy] test print float in sci_mode when max > 1e8.
        # TODO: Pytorch uses fixed precision to print, while Numpy uses dragon4_scientific
        # to do automatic trimming and padding.
        x = torch.tensor([123456789.])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1.2346e+08])''')

        # [Numpy] test print float in sci_mode when max / min > 1000.
        x = torch.tensor([0.01, 11])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1.0000e-02, 1.1000e+01])''')

        # [Numpy] test print int max / min > 1000, no sci_mode
        x = torch.tensor([1, 1010])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([   1, 1010])''')

        # [Numpy] test print int > 1e8, no sci_mode
        x = torch.tensor([1000000000])  # 1e9
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1000000000])''')

        # [Numpy] test printing float in int_mode
        x = torch.tensor([1., 1000.])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([   1., 1000.])''')

        # [Numpy] test printing float in int_mode in sci format when max / min > 1000.
        x = torch.tensor([1., 1010.])
        self.assertEqual(x.__repr__(), str(x))
        self.assertExpectedInline(str(x), '''tensor([1.0000e+00, 1.0100e+03])''')

    def test_sizeof(self) -> None:
        sizeof_empty = torch.randn(0).storage().__sizeof__()
        sizeof_10 = torch.randn(10).storage().__sizeof__()
        sizeof_100 = torch.randn(100).storage().__sizeof__()
        self.assertEqual((sizeof_100 - sizeof_empty) // (sizeof_10 - sizeof_empty), 10)
        self.assertEqual((sizeof_100 - sizeof_empty) % (sizeof_10 - sizeof_empty), 0)

        sizeof_empty = torch.randn(0).to(torch.uint8).storage().__sizeof__()
        sizeof_10 = torch.randn(10).to(torch.uint8).storage().__sizeof__()
        sizeof_100 = torch.randn(100).to(torch.uint8).storage().__sizeof__()
        self.assertEqual((sizeof_100 - sizeof_empty) // (sizeof_10 - sizeof_empty), 10)
        self.assertEqual((sizeof_100 - sizeof_empty) % (sizeof_10 - sizeof_empty), 0)

    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    def test_resizable(self) -> None:
        x = torch.randn(5)
        self.assertTrue(x.storage().resizable())
        x.numpy()
        self.assertFalse(x.storage().resizable())

    def test_iter(self) -> None:
        x = torch.randn(5, 5)
        for i, sub in enumerate(x):
            self.assertEqual(sub, x[i])  # noqa: PLR1736

        x = torch.tensor([])
        self.assertEqual(list(x), [])

    def test_new(self) -> None:
        x = torch.autograd.Variable(torch.tensor([]))
        y = torch.autograd.Variable(torch.randn(4, 4))
        z = torch.autograd.Variable(torch.IntTensor([1, 2, 3]))
        self.assertEqual(x.new().shape, [0])
        self.assertEqual(x.new(), x)
        self.assertEqual(x.new(1, 2).shape, [1, 2])
        self.assertEqual(x.new(torch.Size([3, 4])).shape, [3, 4])
        self.assertEqual(x.new([3, 4]).shape, [2])
        self.assertEqual(x.new([3, 4]).tolist(), [3, 4])
        self.assertEqual(x.new((3, 4)).tolist(), [3, 4])
        self.assertEqual(x.new([np.int32(3), np.float64(4)]).tolist(), [3, 4])
        self.assertEqual(x.new(np.array((3, 4))).tolist(), [3, 4])
        self.assertEqual(x.new([z[2], z[0] + 3]).tolist(), [3, 4])
        self.assertEqual(x.new(size=(3, 4)).shape, [3, 4])
        self.assertEqual(x.new(()).shape, [0])
        self.assertEqual(x.new(y.storage()).data_ptr(), y.data_ptr())
        self.assertEqual(x.new(y).data_ptr(), y.data_ptr())
        self.assertIsNot(x.new(y), y)

        self.assertRaises(TypeError, lambda: x.new(z))
        # TypeError would be better
        self.assertRaises(RuntimeError, lambda: x.new(z.storage()))

    @unittest.skipIf(PYTORCH_CUDA_MEMCHECK, "is_pinned uses failure to detect pointer property")
    def test_pin_memory(self):
        x = torch.randn(3, 5)
        self.assertFalse(x.is_pinned())
        if torch.cuda.is_available():
            pinned = x.pin_memory()
            self.assertTrue(pinned.is_pinned())
            self.assertEqual(pinned, x)
            self.assertNotEqual(pinned.data_ptr(), x.data_ptr())
            # test that pin_memory on already pinned tensor has no effect
            self.assertIs(pinned, pinned.pin_memory())
            self.assertEqual(pinned.data_ptr(), pinned.pin_memory().data_ptr())

    def test_error_msg_type_translation(self):
        with self.assertRaisesRegex(
                RuntimeError,
                # message includes both Double and Long
                '(?=.*Double)(?=.*Long)'):

            # Calls model with a LongTensor input but DoubleTensor weights
            input = torch.zeros(1, 1, 1, 6, dtype=torch.long)
            weight = torch.nn.Parameter(torch.zeros(1, 1, 1, 3, dtype=torch.double))
            model = torch.nn.Conv2d(1, 1, (1, 3), stride=1, padding=0, bias=False)
            model.weight = weight
            model(input)

    def test_apply(self):
        x = torch.arange(1, 6)
        res = x.clone().apply_(lambda k: k + k)
        self.assertEqual(res, x * 2)
        self.assertRaises(TypeError, lambda: x.apply_(lambda k: "str"))

    def test_map(self):
        x = torch.autograd.Variable(torch.randn(3, 3))
        y = torch.autograd.Variable(torch.randn(3))
        res = x.clone()
        res.map_(y, lambda a, b: a + b)
        self.assertEqual(res, x + y)
        self.assertRaisesRegex(TypeError, "not callable", lambda: res.map_(y, "str"))

    def test_map2(self):
        x = torch.autograd.Variable(torch.randn(3, 3))
        y = torch.autograd.Variable(torch.randn(3))
        z = torch.autograd.Variable(torch.randn(1, 3))
        res = x.clone()
        res.map2_(y, z, lambda a, b, c: a + b * c)
        self.assertEqual(res, x + y * z)
        z.requires_grad = True
        self.assertRaisesRegex(
            RuntimeError, "requires grad",
            lambda: res.map2_(y, z, lambda a, b, c: a + b * c))

    def test_Size(self):
        # expects iterable of int, not Tensor
        self.assertRaises(TypeError, lambda: torch.Size(torch.ones(3)))
        # initialization
        empty_size = torch.Size([])
        size = torch.Size([1, 2, 3])
        self.assertIsInstance(empty_size, tuple)
        self.assertIsInstance(size, tuple)
        # value check __len__
        self.assertEqual(len(empty_size), 0)
        self.assertEqual(len(size), 3)
        # type check __getitem__[int]
        self.assertIsInstance(size[0], int)
        self.assertIsInstance(size[1], int)
        self.assertIsInstance(size[2], int)
        # value check __getitem__[int]
        self.assertEqual(size[0], 1)
        self.assertEqual(size[1], 2)
        self.assertEqual(size[2], 3)
        # type check __getitem__[slice]
        self.assertIsInstance(size[:], torch.Size)
        self.assertIsInstance(size[:-1], torch.Size)
        self.assertIsInstance(size[0:0], torch.Size)
        # value check __getitem__[slice]
        self.assertEqual(size[:], (1, 2, 3))
        self.assertEqual(size[:-1], (1, 2))
        self.assertEqual(size[0:0], ())
        # type check __add__
        self.assertIsInstance(empty_size + (), torch.Size)
        self.assertIsInstance(size + (), torch.Size)
        self.assertIsInstance(size + (4, 5), torch.Size)
        self.assertIsInstance(size + size, torch.Size)
        # value check __add__
        self.assertEqual(empty_size + (), ())
        self.assertEqual(size + (), (1, 2, 3))
        self.assertEqual(size + (4, 5), (1, 2, 3, 4, 5))
        self.assertEqual(size + size, (1, 2, 3, 1, 2, 3))
        # type check __radd__
        self.assertIsInstance(() + empty_size, torch.Size)
        self.assertIsInstance((4, 5) + size, torch.Size)
        # value check __radd__
        self.assertEqual(() + size, (1, 2, 3))
        self.assertEqual((4, 5) + size, (4, 5, 1, 2, 3))
        # type check __mul__
        self.assertIsInstance(empty_size * 0, torch.Size)
        self.assertIsInstance(size * 0, torch.Size)
        self.assertIsInstance(size * 1, torch.Size)
        self.assertIsInstance(size * 2, torch.Size)
        # value check __mul__
        self.assertEqual(empty_size * 0, ())
        self.assertEqual(size * 0, ())
        self.assertEqual(size * 1, (1, 2, 3))
        self.assertEqual(size * 2, (1, 2, 3, 1, 2, 3))
        # type check __rmul__
        self.assertIsInstance(0 * empty_size, torch.Size)
        self.assertIsInstance(0 * size, torch.Size)
        self.assertIsInstance(1 * size, torch.Size)
        self.assertIsInstance(2 * size, torch.Size)
        # value check __rmul__
        self.assertEqual(0 * empty_size, ())
        self.assertEqual(0 * size, ())
        self.assertEqual(1 * size, (1, 2, 3))
        self.assertEqual(2 * size, (1, 2, 3, 1, 2, 3))

    def test_Size_concat_non_tuple_sequence(self):
        # check that TypeError gets raised on adding non-tuple sequences.
        from collections.abc import Sequence

        class DummySequence(Sequence):
            vals = list(range(5))
            def __len__(self): return len(self.vals)
            def __getitem__(self, i): return self.vals[i]
            def __iter__(self): return iter(self.vals)

        size = torch.Size([1, 2, 3])
        seq = DummySequence()
        msg = r"can only concatenate tuple \(not \w+\) to torch.Size"
        self.assertRaisesRegex(TypeError, msg, lambda: size + seq)
        msg = r"unsupported operand type"
        self.assertRaisesRegex(TypeError, msg, lambda: seq + size)

    def test_Size_concat_wildcard(self):
        # check that 3rd party classes can support addition with torch.Size
        class Wildcard:
            def __add__(self, other): return 42
            def __radd__(self, other): return 42

        size = torch.Size([1, 2, 3])
        wildcard = Wildcard()
        self.assertEqual(wildcard + size, 42)
        self.assertEqual(size + wildcard, 42)

    def test_Size_scalar(self):
        three = torch.tensor(3)
        two = torch.tensor(2)
        x = torch.Size([0, 1, two, three, 4])
        for i in range(1, 5):
            self.assertEqual(x[i], i)

    def test_Size_iter(self):
        for sizes in [iter([1, 2, 3, 4, 5]), range(1, 6)]:
            x = torch.Size(sizes)
            for i in range(5):
                self.assertEqual(x[i], i + 1)

    def test_t_not_2d_error(self):
        self.assertRaises(RuntimeError, lambda: torch.randn(2, 3, 4).t())
        self.assertRaises(RuntimeError, lambda: torch.randn(2, 3, 4).t_())

    # skip this test for now as it affects all tests
    @unittest.skipIf(True, "flush_denormal not supported")
    def test_set_flush_denormal(self):
        tiny_float = 1e-42
        tiny_double = 1e-320
        float_tensor = torch.FloatTensor([1.0, tiny_float])
        double_tensor = torch.DoubleTensor([1.0, tiny_float, tiny_double])

        self.assertEqual(float_tensor[0], 1.0, atol=0.0, rtol=0)
        self.assertEqual(float_tensor[1], tiny_float, atol=tiny_float / 16, rtol=0)
        self.assertEqual(double_tensor[0], 1.0, atol=0.0, rtol=0)
        self.assertEqual(double_tensor[1], tiny_float, atol=0.0, rtol=0)
        self.assertEqual(double_tensor[2], tiny_double, atol=0.0, rtol=0)

        torch.set_flush_denormal(True)
        self.assertEqual(float_tensor[0], 1.0, atol=0.0, rtol=0)
        self.assertEqual(float_tensor[1], 0.0, atol=0.0, rtol=0)  # tiny_float to zero
        self.assertEqual(double_tensor[0], 1.0, atol=0.0, rtol=0)
        # tiny_float is not converted to zero in double type
        self.assertEqual(double_tensor[1], tiny_float, atol=0.0, rtol=0)
        self.assertEqual(double_tensor[2], 0.0, atol=0.0, rtol=0)  # tiny_double to zero
        torch.set_flush_denormal(False)

    def test_show_config(self):
        # We can't usefully test the output; just make sure this doesn't crash
        torch.__config__.show()

    @unittest.skipIf(IS_FBCODE, "CXX_FLAGS is only for OSS build.")
    def test_cxx_flags(self):
        torch.__config__._cxx_flags()

    def test_parallel_info(self):
        torch.__config__.parallel_info()

    def test_get_cpu_capability(self):
        # This method is primarily exposed for torchvision's resize
        torch.backends.cpu.get_cpu_capability()

        # We have to ensure that method is torchscriptable as torchvision's resize
        # should be torchscriptable
        torch.jit.script(torch.backends.cpu.get_cpu_capability)

    @slowTest
    def test_slow_test(self):
        # Just a smoketest to make sure our slowTest decorator works.
        pass

    def test_is_nonzero(self):
        with self.assertRaisesRegex(RuntimeError, "Boolean value of Tensor with no values is ambiguous"):
            torch.tensor([]).is_nonzero()
        with self.assertRaisesRegex(RuntimeError, "Boolean value of Tensor with more than one value is ambiguous"):
            torch.tensor([0, 0]).is_nonzero()
        self.assertFalse(torch.tensor(0).is_nonzero())
        self.assertTrue(torch.tensor(1).is_nonzero())
        self.assertFalse(torch.tensor([0]).is_nonzero())
        self.assertTrue(torch.tensor([1]).is_nonzero())
        self.assertFalse(torch.tensor([[0]]).is_nonzero())
        self.assertTrue(torch.tensor([[1]]).is_nonzero())
        self.assertTrue(torch.tensor(0.1).is_nonzero())
        self.assertTrue(torch.tensor(-0.1).is_nonzero())
        self.assertFalse(torch.tensor(0.0).is_nonzero())
        self.assertTrue(torch.tensor(True).is_nonzero())
        self.assertFalse(torch.tensor(False).is_nonzero())
        self.assertFalse(torch.tensor(0 + 0j).is_nonzero())
        self.assertTrue(torch.tensor(0 + 0.1j).is_nonzero())

    def test_assert_async(self):
        with self.assertRaisesRegex(RuntimeError, "Boolean value of Tensor with no values is ambiguous"):
            torch._assert_async(torch.tensor([]))
        with self.assertRaisesRegex(RuntimeError, "Boolean value of Tensor with more than one value is ambiguous"):
            torch._assert_async(torch.tensor([0, 0]))
        with self.assertRaisesRegex(RuntimeError, "Expected Tensor with single nonzero value, but got zero"):
            torch._assert_async(torch.tensor(0))
        torch._assert_async(torch.tensor(1))
        torch._assert_async(torch.tensor(0.1))
        torch._assert_async(torch.tensor(-0.1))
        with self.assertRaisesRegex(RuntimeError, "Expected Tensor with single nonzero value, but got zero"):
            torch._assert_async(torch.tensor(0.0))
        torch._assert_async(torch.tensor(True))
        with self.assertRaisesRegex(RuntimeError, "Expected Tensor with single nonzero value, but got zero"):
            torch._assert_async(torch.tensor(False))
        torch._assert_async(torch.tensor(0 + 0.1j))
        with self.assertRaisesRegex(RuntimeError, "Expected Tensor with single nonzero value, but got zero"):
            torch._assert_async(torch.tensor(0 + 0j))

    # NB: we must not be built with CUDA; if we are built with CUDA but no CUDA
    # is available, we get a different error.
    @unittest.skipIf(torch.backends.cuda.is_built() or IS_SANDCASTLE, "CUDA is built, can't test CUDA not built error")
    def test_cuda_not_built(self):
        msg = "Torch not compiled with CUDA enabled"
        self.assertRaisesRegex(AssertionError, msg, lambda: torch.cuda.current_device())
        self.assertRaisesRegex(AssertionError, msg, lambda: torch.tensor([1], device="cuda"))
        self.assertRaisesRegex(AssertionError, msg, lambda: torch.tensor([1]).cuda())
        self.assertRaisesRegex(TypeError, msg, lambda: torch.cuda.FloatTensor())
        self.assertRaisesRegex(TypeError, msg, lambda: torch.set_default_tensor_type(torch.cuda.FloatTensor))
        self.assertRaisesRegex(AssertionError, msg, lambda: torch.tensor([1]).to(device="cuda"))

    def test_has_internal_overlap(self):
        OVERLAP_NO = 0
        OVERLAP_YES = 1
        OVERLAP_TOO_HARD = 2

        # Check for contiguous tensors
        a = torch.randn(3, 3)
        self.assertEqual(torch._debug_has_internal_overlap(a), OVERLAP_NO)

        # Checks for zero strides
        b = torch.randn(1, 3)
        b_expanded = b.expand(4, 3)
        self.assertEqual(torch._debug_has_internal_overlap(b_expanded), OVERLAP_YES)

        # Check for zero strided, size 1 axis, in non-contiguous storage (gh-33812)
        c = torch.randn(10).as_strided([2, 1, 5], [1, 0, 2])
        self.assertEqual(torch._debug_has_internal_overlap(c), OVERLAP_NO)
        c = torch.randn(2, 1, 10)[::2].as_strided((2, 1, 5), (10, 0, 2))
        self.assertEqual(torch._debug_has_internal_overlap(c), OVERLAP_TOO_HARD)

    def test_allow_tensor_metadata_change(self):
        torch.ones(2, 3)
        # Metadata changes are allowed on view tensors that are created from detach().

    def test_memory_format(self):
        def test_helper(x, memory_format):
            y = x.contiguous(memory_format=memory_format)
            self.assertFalse(y.is_contiguous())
            self.assertTrue(y.is_contiguous(memory_format=memory_format))
            self.assertEqual(y, x)

        test_helper(torch.randn(4, 3, 8, 8), torch.channels_last)
        test_helper(torch.randn(4, 3, 8, 8, 8), torch.channels_last_3d)

    def test_memory_format_contiguous_returns_same_tensor_if_already_satisfies(self):
        def test_helper(x, memory_format):
            alias = x.contiguous(memory_format=memory_format)
            alias.fill_(7)
            self.assertEqual(x, alias)

        test_helper(torch.randn(4, 8, 8, 3).permute(0, 3, 1, 2), torch.channels_last)
        test_helper(torch.randn(4, 8, 8, 8, 3).permute(0, 4, 1, 2, 3), torch.channels_last_3d)

    def test_memory_format_empty(self):
        def test_helper(dim1, dim2, memory_format):
            with self.assertRaises(RuntimeError):
                x = torch.empty(dim1, memory_format=memory_format)
            x = torch.empty(dim2, memory_format=memory_format)
            self.assertTrue(x.is_contiguous(memory_format=memory_format))

        test_helper((3, 3), (3, 3, 3, 3), torch.channels_last)
        test_helper((3, 3, 3), (3, 3, 3, 3, 3), torch.channels_last_3d)

    @skipIfCrossRef
    def test_dim_order(self):
        shape = (2, 3, 5, 7)

        t = torch.empty(shape)
        self.assertSequenceEqual(t.dim_order(), (0, 1, 2, 3), seq_type=tuple)
        self.assertSequenceEqual(t.dim_order(ambiguity_check=True), (0, 1, 2, 3), seq_type=tuple)
        # transpose doesn't really change the underlying physical memory
        # so expecting dim_order change to reflect that (like strides)
        self.assertSequenceEqual(t.transpose(0, 1).dim_order(), (1, 0, 2, 3))

        t = torch.empty(shape, memory_format=torch.channels_last)
        self.assertSequenceEqual(t.dim_order(), (0, 2, 3, 1))

        t = torch.empty((2, 3, 5, 7, 8), memory_format=torch.channels_last_3d)
        self.assertSequenceEqual(t.dim_order(), (0, 2, 3, 4, 1))

        for dim_order in itertools.permutations(range(4)):
            self.assertSequenceEqual(
                dim_order, torch.empty_permuted(shape, dim_order).dim_order()
            )

        target_shapes = [[2, 2, 1, 2], [1, 2, 2, 2], [2, 2, 2, 1], [1, 2, 2, 1], [1, 2, 1, 2]]

        for shape in target_shapes:
            for memory_format in (torch.contiguous_format, torch.channels_last):
                t = torch.empty(shape).to(memory_format=memory_format)
                with self.assertRaises(RuntimeError):
                    t.dim_order(ambiguity_check=True)

                if memory_format == torch.contiguous_format:
                    dim_order_target = list(range(len(shape)))
                elif memory_format == torch.channels_last:
                    dim_order_target = [0, *list(range(2, len(shape))), 1]

                self.assertSequenceEqual(
                    dim_order_target, t.dim_order(ambiguity_check=[torch.contiguous_format, torch.channels_last])
                )


        ambiguous_shapes = [[2, 1, 2, 2], [2, 2, 1, 1], [1, 2, 1, 1], [2, 1, 1, 2], [2, 1, 2, 1],
                            [1, 1, 1, 2], [1, 1, 2, 2], [1, 1, 1, 1], [2, 1, 1, 1], [1, 1, 2, 1]]

        for shape in ambiguous_shapes:
            for memory_format in (torch.contiguous_format, torch.channels_last):
                t = torch.empty(shape).to(memory_format=memory_format)
                with self.assertRaises(RuntimeError):
                    t.dim_order(ambiguity_check=True)
                    t.dim_order(ambiguity_check=[torch.contiguous_format, torch.channels_last])

        with self.assertRaises(TypeError):
            torch.empty((1, 2, 3, 4)).dim_order(ambiguity_check="ILLEGAL_STR")

        # sparse tensor does not support dim order
        with self.assertRaises(AttributeError):
            indices = torch.tensor([[0, 1, 2], [0, 1, 2]])  # (row, column) indices
            values = torch.tensor([1.0, 2.0, 3.0])  # values at those indices
            sparse_tensor = torch.sparse_coo_tensor(indices, values, size=(3, 3))
            sparse_tensor.dim_order()

    def test_subclass_tensors(self):
        # raise an error when trying to subclass FloatTensor
        with self.assertRaisesRegex(TypeError, "type 'torch.FloatTensor' is not an acceptable base type"):
            class Foo1(torch.FloatTensor):
                pass

        # but allow subclassing Tensor:
        class Foo2(torch.Tensor):
            def foo(self):
                return 5
        f = Foo2()
        self.assertEqual(f.foo(), 5)

    def test_ndim(self):
        a = torch.randn(1, 2, 3)
        self.assertEqual(3, a.ndim)
        b = torch.randn(())
        self.assertEqual(0, b.ndim)
        c = torch.randn(1, 0)
        self.assertEqual(2, c.ndim)

    def test_nbytes(self):
        a = torch.randn(1, 2, 3, dtype=torch.float64)
        self.assertEqual(a.numel() * a.element_size(), a.nbytes)
        b = torch.randn(())
        self.assertEqual(b.numel() * b.element_size(), b.nbytes)
        c = torch.randn(1, 0)
        self.assertEqual(c.numel() * c.element_size(), c.nbytes)

    def test_fill_diagonal(self):
        a1 = torch.randn(7, 3)
        a2 = a1.clone()
        v = 1
        for i in range(3):
            a2[i][i] = v
        a1.fill_diagonal_(v)
        self.assertEqual(a1, a2)

        b1 = torch.randn(7, 3)
        b2 = b1.clone()
        for i in range(3):
            b2[i][i] = v
            b2[i + 4][i] = v
        b1.fill_diagonal_(v, wrap=True)
        self.assertEqual(b1, b2)

        c1 = torch.rand(3, 3, 3)
        c2 = c1.clone()
        for i in range(3):
            c2[i][i][i] = v
        c1.fill_diagonal_(v)
        self.assertEqual(c1, c2)

        # non-contiguous tensor
        d1 = torch.rand(3, 3, 3)[:, 1, ...]
        d2 = d1.clone()
        for i in range(3):
            d2[i][i] = v
        d1.fill_diagonal_(v)
        self.assertEqual(d1, d2)

        e1 = torch.rand(7, 3, 3)[:, 1, ...]
        e2 = e1.clone()
        for i in range(3):
            e2[i][i] = v
            e2[i + 4][i] = v
        e1.fill_diagonal_(v, wrap=True)
        self.assertEqual(e1, e2)

    def test_setting_real_imag_to_a_number(self):
        x = torch.randn(4, dtype=torch.cfloat)
        x.real = 0
        x.imag = 0
        zeros = torch.zeros(4)
        self.assertEqual(x.real, zeros)
        self.assertEqual(x.imag, zeros)

    def test_batch_norm_cpu_inference(self):
        # input nchw in (2,1,1,1), (2,2,2,2)
        inputs = [
            torch.tensor([[[[-0.5000]]], [[[0.5000]]]]),
            torch.tensor([
                [
                    [[-0.5000, 0.5000], [-1.0000, 1.0000]],
                    [[-0.2500, -0.5000], [0.2500, 0.5000]]
                ],
                [
                    [[0.1000, 1.0000], [1.0000, 0.1000]],
                    [[1.0000, 0.5000], [1.5000, -1.5000]]
                ]])]
        # output nchw in (2,1,1,1), (2,2,2,2)
        outputs = [
            torch.tensor([
                [[[-0.499997496604919433593750000]]],
                [[[0.499997496604919433593750000]]]]),
            torch.tensor([
                [[[-0.499997496604919433593750000, 0.499997496604919433593750000],
                  [-0.999994993209838867187500000, 0.999994993209838867187500000]],
                 [[-0.249998748302459716796875000, -0.499997496604919433593750000],
                  [0.249998748302459716796875000, 0.499997496604919433593750000]]],
                [[[0.099999502301216125488281250, 0.999994993209838867187500000],
                  [0.999994993209838867187500000, 0.099999502301216125488281250]],
                 [[0.999994993209838867187500000, 0.499997496604919433593750000],
                  [1.499992489814758300781250000, -1.499992489814758300781250000]]]])]


        for i in range(len(inputs)):
            for affine in [False, True]:
                m = torch.nn.BatchNorm2d(inputs[i].size()[1], 1e-05, 0.1, affine=affine)
                m.eval()
                # contiguous case
                input1 = inputs[i].contiguous()
                output1 = m(input1)
                # non-contiguous case
                input2 = input1.permute(0, 1, 3, 2)
                output2 = m(input2).permute(0, 1, 3, 2)
                # channels last case
                input3 = input1.contiguous(memory_format=torch.channels_last)
                output3 = m(input3)
                self.assertEqual(output3, outputs[i])
                self.assertEqual(output3, output1)
                self.assertEqual(output3, output2)

    # FIXME: move these meta tests to their own test suite/class or
    #   distribute them among the appropriate test suites for their ops
    @skipIfTorchDynamo("Fails after Triton update, see https://github.com/pytorch/pytorch/issues/94687")
    def test_empty_meta(self):
        x = torch.empty(2 ** 20, 2 ** 20, device='meta')
        y = torch.empty(2 ** 20, device='meta')
        z = x + y
        self.assertEqual(z.size(), (2 ** 20, 2 ** 20))
        self.assertRaises(RuntimeError, lambda: z[0][0].item())

    @skipIfTorchDynamo("Fails after Triton update, see https://github.com/pytorch/pytorch/issues/94687")
    def test_format_scalar_meta(self):
        x = torch.empty((), device='meta')
        self.assertEqual(format(x), repr(x))

    def test_upsample_nearest1d_meta(self):
        # TODO: this test should be triggered by test_nn.py but right
        # now meta is not enabled (and even if it was, we are probably
        # missing too many meta functions to get through the test unmolested)

        # NB: Can't make the exponent too big, or it will overflow
        # signed 64-bit integer
        x = torch.empty(2 * 10 ** 8, 3, 2 * 10 ** 8, device='meta')
        z = torch.nn.functional.interpolate(x, scale_factor=2)
        self.assertEqual(z.size(), (2 * 10 ** 8, 3, 4 * 10 ** 8))
        self.assertRaises(RuntimeError, lambda: z[0][0][0].item())

        # TODO: the out tests cannot be triggered by test_nn.py because
        # we don't actually do out= arguments for nn functions, so there
        # is no public API by which to get the out version

        # interpolate doesn't seem to support out=
        # (not sure why passing None here doesn't work? How strange...)
        z = torch.empty(0, device='meta')
        torch._C._nn.upsample_nearest1d(x, (4 * 10 ** 8,), 2, out=z)
        self.assertEqual(z.size(), (2 * 10 ** 8, 3, 4 * 10 ** 8))
        self.assertRaises(RuntimeError, lambda: z[0][0][0].item())

    def test_upsample_nearest2d_meta(self):
        # TODO: the out tests cannot be triggered by test_nn.py because
        # we don't actually do out= arguments for nn functions, so there
        # is no public API by which to get the out version

        # Make sure we don't clobber strides of out tensor.  NB: this
        # test must be done on 2d/3d, because 1d doesn't have any meaningful
        # layout support
        x = torch.empty(4, 3, 8, 8, device='meta')
        out = torch.empty(4, 3, 16, 16, device='meta', memory_format=torch.channels_last)
        torch._C._nn.upsample_nearest2d(x, (16, 16), out=out)
        self.assertTrue(out.is_contiguous(memory_format=torch.channels_last))

        x = torch.empty(4, 3, 8, 8, device='meta', memory_format=torch.channels_last)
        out = torch.empty(4, 3, 16, 16, device='meta')
        torch._C._nn.upsample_nearest2d(x, (16, 16), out=out)
        self.assertTrue(out.is_contiguous())

        # But if resize occurs, do clobber
        x = torch.empty(4, 3, 8, 8, device='meta', memory_format=torch.channels_last)
        out = torch.empty(0, device='meta')
        torch._C._nn.upsample_nearest2d(x, (16, 16), out=out)
        self.assertTrue(out.is_contiguous(memory_format=torch.channels_last))

        # Complain if out dtype mismatch
        x = torch.empty(4, 3, 8, 8, device='meta', dtype=torch.float)
        out = torch.empty(4, 3, 16, 16, device='meta', dtype=torch.double)
        self.assertExpectedRaisesInline(
            RuntimeError, lambda: torch._C._nn.upsample_nearest2d(x, (16, 16), out=out),
            """Expected out tensor to have dtype torch.float32 but got torch.float64 instead"""
        )

        # Complain if out device mismatch
        x = torch.empty(0, 3, 8, 8, device='meta')
        out = torch.empty(0, 3, 16, 16, device='cpu')
        # FIXME: compiling should properly error with a device mismatch.
        if not TEST_WITH_TORCHINDUCTOR:
            self.assertExpectedRaisesInline(
                RuntimeError, lambda: torch._C._nn.upsample_nearest2d(x, (16, 16), out=out),
                """Attempting to copy from device meta to device cpu, but cross-device copies are not allowed!"""
            )

    def test_add_meta_scalar(self):
        # From https://github.com/pytorch/pytorch/issues/53815
        x = torch.empty(2, device='meta')
        y = x + 2
        self.assertEqual(y.size(), x.size())

    def test_normal_shape(self):
        for device in get_all_device_types():
            tensor1 = torch.rand(1, device=device)
            tensor4 = torch.rand(4, device=device)
            tensor120 = torch.rand(120, device=device)
            tensor2145 = torch.rand(2, 1, 4, 5, device=device)
            tensor2345 = torch.rand(2, 3, 4, 5, device=device)
            tensor2345_non_contiguous = torch.rand(2, 4, 3, 5, device=device).permute(0, 2, 1, 3)
            tensor2345_channels_last = tensor2345.contiguous(memory_format=torch.channels_last)
            output2345 = torch.zeros(2, 3, 4, 5, device=device)
            output345 = torch.zeros(3, 4, 5, device=device)

            # inputs have same size
            self.assertEqual(torch.normal(tensor2345, tensor2345).size(), (2, 3, 4, 5))
            self.assertEqual(torch.normal(tensor2345_non_contiguous, tensor2345).size(), (2, 3, 4, 5))
            self.assertEqual(torch.normal(tensor2345, tensor2345_channels_last).size(), (2, 3, 4, 5))
            self.assertEqual(torch.normal(tensor2345_non_contiguous, tensor2345_channels_last).size(), (2, 3, 4, 5))

            # scalar case
            self.assertEqual(torch.normal(tensor2345, 2).size(), (2, 3, 4, 5))
            self.assertEqual(torch.normal(2, tensor2345).size(), (2, 3, 4, 5))

            # inputs are expandable tensors
            self.assertEqual(torch.normal(tensor2345, tensor1).size(), (2, 3, 4, 5))
            self.assertEqual(torch.normal(tensor2145, tensor2345).size(), (2, 3, 4, 5))

            # inputs are non-expandable tensors, but they have same number of elements
            with self.assertRaisesRegex(
                    RuntimeError,
                    r"The size of tensor a \(120\) must match the size of "
                    r"tensor b \(5\) at non-singleton dimension 3"):
                self.assertEqual(torch.normal(tensor120, tensor2345).size(), (120,))
            with self.assertRaisesRegex(
                    RuntimeError,
                    r"The size of tensor a \(5\) must match the size of "
                    r"tensor b \(120\) at non-singleton dimension 3"):
                self.assertEqual(torch.normal(tensor2345, tensor120).size(), (2, 3, 4, 5))

            # inputs are non-expandable tensors and they don't have same number of elements
            with self.assertRaisesRegex(
                    RuntimeError,
                    r"The size of tensor a \(5\) must match the size of "
                    r"tensor b \(4\) at non-singleton dimension 3"):
                torch.normal(tensor2345, tensor4)

            # output and inputs are size compatible
            self.assertEqual(torch.normal(tensor2345, tensor2345, out=output2345).size(), (2, 3, 4, 5))

            # output and inputs are not size compatible
            with self.assertWarnsRegex(
                    UserWarning,
                    "This behavior is deprecated, and in a future PyTorch "
                    "release outputs will not be resized unless they have "
                    "zero elements"):
                self.assertEqual(torch.normal(tensor2345, tensor2145, out=output345).size(), (2, 3, 4, 5))
            with self.assertRaisesRegex(
                    RuntimeError,
                    r"The size of tensor a \(5\) must match the size of "
                    r"tensor b \(120\) at non-singleton dimension 3"):
                # inputs are not expandable, output size is not the same as mean
                torch.normal(tensor2345, tensor120, out=output345)

    @unittest.skipIf(IS_MACOS, "https://github.com/pytorch/pytorch/issues/157246")
    def test_tensoriterator_output_setup(self):
        # Test whether the output's memory layout is correct
        def test_memory_layout(x, y, scale, zero_point, out):
            self.assertEqual(x.dim(), 4)
            self.assertEqual(x.size(), y.size())
            self.assertEqual(y.size(), out.size())

            shape = x.size()
            for n in range(shape[0]):
                for c in range(shape[1]):
                    for h in range(shape[2]):
                        for w in range(shape[3]):
                            if scale is not None and zero_point is not None:
                                self.assertEqual(
                                    out[n][c][h][w],
                                    torch.ops.quantized.add(x[n][c][h][w], y[n][c][h][w], scale, zero_point))
                            else:
                                self.assertEqual(out[n][c][h][w], x[n][c][h][w] + y[n][c][h][w])

        xraw = torch.rand(2, 3, 4, 4)
        yraw = torch.rand(2, 3, 4, 4)
        qxraw = torch.quantize_per_tensor(xraw, 0.1, 5, torch.quint8)
        qyraw = torch.quantize_per_tensor(yraw, 0.1, 5, torch.quint8)

        # contiguous case fast setup
        test_memory_layout(xraw, yraw, None, None, xraw + yraw)
        test_memory_layout(qxraw, qyraw, 0.1, 5, torch.ops.quantized.add(qxraw, qyraw, 0.1, 5))

        # channels last case fast setup
        x = xraw.contiguous(memory_format=torch.channels_last)
        y = yraw.contiguous(memory_format=torch.channels_last)
        test_memory_layout(x, y, None, None, x + y)
        qx = qxraw.contiguous(memory_format=torch.channels_last)
        qy = qyraw.contiguous(memory_format=torch.channels_last)
        test_memory_layout(qx, qy, 0.1, 5, torch.ops.quantized.add(qx, qy, 0.1, 5))

        # non contiguous case fast setup (dense, non-overlapping, same shape and strides)
        x = xraw.permute(0, 2, 3, 1)
        y = yraw.permute(0, 2, 3, 1)
        test_memory_layout(x, y, None, None, x + y)
        qx = qxraw.permute(0, 2, 3, 1)
        qy = qyraw.permute(0, 2, 3, 1)
        test_memory_layout(qx, qy, 0.1, 5, torch.ops.quantized.add(qx, qy, 0.1, 5))

        # non contiguous case fast setup (dense, non-overlapping)
        # input tensors have same shape and strides
        # output tensor have same shape as input tensors but different stride
        # output tensor should preserve its strides in this case
        x = xraw.permute(0, 2, 3, 1)
        y = yraw.permute(0, 2, 3, 1)
        out = torch.empty_like(xraw)
        out = out.permute(0, 3, 2, 1)
        expected_stride = out.stride()
        test_memory_layout(x, y, None, None, torch.add(x, y, out=out))
        self.assertEqual(expected_stride, out.stride())

        # non contiguous case non fast setup
        x = xraw.permute(0, 2, 3, 1)
        y = yraw.permute(0, 3, 2, 1)
        test_memory_layout(x, y, None, None, x + y)
        qx = qxraw.permute(0, 2, 3, 1)
        qy = qyraw.permute(0, 3, 2, 1)
        test_memory_layout(qx, qy, 0.1, 5, torch.ops.quantized.add(qx, qy, 0.1, 5))

    def test_conj_physical_meta_stride(self):
        a = torch.zeros((5, 3, 6), dtype=torch.complex128, device='meta')
        b = torch._fft_c2c(a, [1], 1, True)
        c = torch.conj_physical(b)
        self.assertEqual(b.stride(), c.stride())

    # Tests to make sure we still handle .data properly until it is removed
    def test_dot_data_use(self):
        # .data allows to change the Tensors types inplace, check that we still
        # raise a nice error.
        with self.assertRaisesRegex(
                RuntimeError,
                # message includes both Double and ComplexFloat
                '(?=.*Double)(?=.*ComplexFloat)'):

            # Calls model with a LongTensor input but DoubleTensor weights
            input = torch.randn(1, 1, 1, 6, dtype=torch.double)
            weight = torch.zeros(1, 1, 1, 3, dtype=torch.complex64)
            model = torch.nn.Conv2d(1, 1, (1, 3), stride=1, padding=0, bias=False)
            model.weight.data = weight
            model(input)

    def test_empty_storage_view(self):
        # we should be able to "modify" slices of a 0-element
        # array without an error being raised due to
        # trying to resize its storage
        t = torch.from_numpy(np.empty((0, 4)))
        t[:, 1::2] *= 1

    def test_has_storage(self):
        self.assertIsNotNone(torch.tensor([]).storage())
        self.assertIsNotNone(torch.empty(0).storage())
        self.assertIsNotNone(torch.tensor([]).clone().storage())
        self.assertIsNotNone(torch.tensor([0, 0, 0]).nonzero().storage())
        self.assertIsNotNone(torch.tensor([]).new().storage())

    # FIXME: Extend this test and put in a TensorProperties test class
    def test_numel(self):
        b = torch.ByteTensor(3, 100, 100)
        self.assertEqual(b.nelement(), 3 * 100 * 100)
        self.assertEqual(b.numel(), 3 * 100 * 100)

    # Verifies that (deep)copies of dtypes are the same objects
    def test_copy_dtypes(self):
        for dtype in all_types_and_complex_and(torch.half, torch.bfloat16, torch.bool):
            copied_dtype = copy.deepcopy(dtype)
            self.assertIs(dtype, copied_dtype)

    def test_dtype_is_signed(self):
        for dtype in all_types_and_complex_and(torch.half, torch.bfloat16, torch.half):
            self.assertEqual(dtype.is_signed, torch.is_signed(torch.tensor(0, dtype=dtype)))

        self.assertRaisesRegex(RuntimeError, 'not supported for quantized', lambda: torch.quint8.is_signed)
        self.assertRaisesRegex(RuntimeError, 'not supported for quantized', lambda: torch.qint8.is_signed)
        self.assertRaisesRegex(RuntimeError, 'not supported for quantized', lambda: torch.qint32.is_signed)

    def test_dtype_abbr(self):
        expected = {
            torch.float64: "f64",
            torch.float32: "f32",
            torch.float16: "f16",
            torch.bfloat16: "bf16",
            torch.float8_e4m3fn: "f8e4m3fn",
            torch.float8_e5m2: "f8e5m2",
            torch.float8_e4m3fnuz: "f8e4m3fnuz",
            torch.float8_e5m2fnuz: "f8e5m2fnuz",
            torch.float8_e8m0fnu: "f8e8m0fnu",
            torch.float4_e2m1fn_x2: "f4e2m1fnx2",
            torch.complex32: "c32",
            torch.bcomplex32: "bc32",
            torch.complex64: "c64",
            torch.complex128: "c128",
            torch.int8: "i8",
            torch.int16: "i16",
            torch.int32: "i32",
            torch.int64: "i64",
            torch.bool: "b8",
            torch.uint8: "u8",
            torch.uint16: "u16",
            torch.uint32: "u32",
            torch.uint64: "u64",
            torch.bits16: "b16x1",
            torch.bits1x8: "b1x8",
            torch.bits2x4: "b2x4",
            torch.bits4x2: "b4x2",
            torch.bits8: "b8x1",
        }
        for dtype, abbr in expected.items():
            self.assertEqual(dtype.abbr, abbr)
        self.assertEqual(set(torch._C._get_all_dtypes()), set(expected.keys()))

    def test_get_all_dtypes(self):
        # The C++ implementation iterates AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_AND_QINTS
        # and filters out quantized and sub-byte placeholder dtypes; pin the
        # result against a manually computed list so regressions are caught.
        expected = {
            torch.uint8, torch.uint16, torch.uint32, torch.uint64,
            torch.int8, torch.int16, torch.int32, torch.int64,
            torch.float16, torch.float32, torch.float64, torch.bfloat16,
            torch.complex32, torch.bcomplex32, torch.complex64, torch.complex128,
            torch.bool,
            torch.float8_e5m2, torch.float8_e4m3fn,
            torch.float8_e5m2fnuz, torch.float8_e4m3fnuz,
            torch.float8_e8m0fnu, torch.float4_e2m1fn_x2,
            torch.bits1x8, torch.bits2x4, torch.bits4x2, torch.bits8, torch.bits16,
        }
        self.assertEqual(set(torch._C._get_all_dtypes()), expected)

    # FIXME: Put the following random tests into their own test class or test suite
    @skipIfTorchDynamo("requires https://github.com/pytorch/torchdynamo/pull/1098")
    def test_RNGState(self):
        state = torch.get_rng_state()
        stateCloned = state.clone()
        before = torch.rand(1000)

        self.assertEqual(state.ne(stateCloned).long().sum(), 0, atol=0, rtol=0)

        torch.set_rng_state(state)
        after = torch.rand(1000)
        self.assertEqual(before, after, atol=0, rtol=0)

    @skipIfTorchDynamo("requires https://github.com/pytorch/torchdynamo/pull/1098")
    def test_RNGStateAliasing(self):
        # Fork the random number stream at this point
        gen = torch.Generator()
        gen.set_state(torch.get_rng_state())
        self.assertEqual(gen.get_state(), torch.get_rng_state())

        target_value = torch.rand(1000)
        # Dramatically alter the internal state of the main generator
        _ = torch.rand(100000)
        forked_value = torch.rand(1000, generator=gen)
        self.assertEqual(target_value, forked_value, atol=0, rtol=0, msg="RNG has not forked correctly.")

    @skipIfTorchDynamo("requires https://github.com/pytorch/torchdynamo/pull/1098")
    def test_RNG_after_pickle(self):
        torch.random.manual_seed(100)
        before = torch.rand(10)

        torch.random.manual_seed(100)
        buf = io.BytesIO()
        tensor = torch.tensor([1, 2, 3])
        ForkingPickler(buf, pickle.HIGHEST_PROTOCOL).dump(tensor)
        after = torch.rand(10)

        self.assertEqual(before, after, atol=0, rtol=0)

    @skipIfTorchDynamo("requires https://github.com/pytorch/torchdynamo/pull/1098")
    def test_boxMullerState(self):
        torch.manual_seed(123)
        odd_number = 101
        seeded = torch.randn(odd_number)
        state = torch.get_rng_state()
        midstream = torch.randn(odd_number)
        torch.set_rng_state(state)
        repeat_midstream = torch.randn(odd_number)
        torch.manual_seed(123)
        reseeded = torch.randn(odd_number)
        self.assertEqual(midstream, repeat_midstream, atol=0, rtol=0,
                         msg='get_rng_state/set_rng_state not generating same sequence of normally distributed numbers')
        self.assertEqual(seeded, reseeded, atol=0, rtol=0,
                         msg='repeated calls to manual_seed not generating same sequence of normally distributed numbers')

    @skipIfTorchDynamo("requires https://github.com/pytorch/torchdynamo/pull/1098")
    def test_manual_seed(self):
        rng_state = torch.get_rng_state()
        torch.manual_seed(2)
        x = torch.randn(100)
        self.assertEqual(torch.initial_seed(), 2)
        torch.manual_seed(2)
        y = torch.randn(100)
        self.assertEqual(x, y)

        max_int64 = 0x7fff_ffff_ffff_ffff
        min_int64 = -max_int64 - 1
        max_uint64 = 0xffff_ffff_ffff_ffff
        # Check all boundary cases of valid seed value inputs
        test_cases = [
            # (seed, expected_initial_seed)
            # Positive seeds should be unchanged
            (max_int64, max_int64),
            (max_int64 + 1, max_int64 + 1),
            (max_uint64, max_uint64),
            (0, 0),
            # Negative seeds wrap around starting from the largest seed value
            (-1, max_uint64),
            (min_int64, max_int64 + 1)
        ]
        for seed, expected_initial_seed in test_cases:
            torch.manual_seed(seed)
            actual_initial_seed = torch.initial_seed()
            msg = (f"expected initial_seed() = {expected_initial_seed:x} "
                   f"after calling manual_seed({seed:x}), but got {actual_initial_seed:x} instead")
            self.assertEqual(expected_initial_seed, actual_initial_seed, msg=msg)
        for invalid_seed in [min_int64 - 1, max_uint64 + 1]:
            with self.assertRaisesRegex(ValueError, r'Overflow when unpacking long long'):
                torch.manual_seed(invalid_seed)

        torch.set_rng_state(rng_state)

    # FIXME: Describe this test and port to the generic device framework in a more
    #   appropriate test suite for the copy operation
    def test_copy_transpose(self):
        x = torch.arange(100 * 100, dtype=torch.float).reshape(100, 100).t()
        y = torch.empty(100, 100, dtype=torch.float)
        y.copy_(x)
        self.assertEqual(y[:, 0], range(100))
        self.assertEqual(y[:, 40], range(4000, 4100))

        y = torch.empty(100, 100, dtype=torch.double)
        y.copy_(x)
        self.assertEqual(y[:, 0], range(100))
        self.assertEqual(y[:, 40], range(4000, 4100))

        # Validates regression reported in https://github.com/pytorch/pytorch/issues/45269
        x = torch.arange(100 * 100).reshape(100, 100).to(dtype=torch.cfloat).t()
        y = torch.empty(100, 100, dtype=torch.cfloat)
        y.copy_(x)
        self.assertEqual(y[:, 0], range(100))
        self.assertEqual(y[:, 40], range(4000, 4100))

        x = torch.arange(100 * 100).reshape(100, 100).to(dtype=torch.complex32).t()
        y = torch.empty(100, 100, dtype=torch.complex32)
        y.copy_(x)
        self.assertEqual(y[:, 0], range(100))
        self.assertEqual(y[:, 40], range(4000, 4100))

    # FIXME: Port to a more appropriate test suite
    def test_copy_broadcast(self):
        torch.zeros(5, 6).copy_(torch.zeros(6))
        self.assertRaises(RuntimeError, lambda: torch.zeros(5, 6).copy_(torch.zeros(30)))

    # FIXME: Port to a more appropriate test suite
    # Fails with inductor (and aot_eager) because functionalization replaces copy_ with copy,
    # which doesn't properly error on bad inputs.
    def test_copy_many_to_one(self):
        # Testing in-place copy where it attempt to write from many memory
        # storage to a single storage would cause RuntimeError to be thrown
        self.assertRaises(RuntimeError, lambda: torch.zeros(1, 6).expand(5, 6).copy_(torch.zeros(5, 6)))

    def test_copy_float16(self):
        # Check that fbgemm code no longer reads memory out of bounds, see
        # copy_impl and fbgemm::Float16ToFloat_ref.
        # https://github.com/pytorch/pytorch/issues/88543

        # Types to test different code paths in copy_impl.
        dtypes = (
            # out_dtype, src_dtype
            (torch.float32, torch.float16),  # fbgemm
            (torch.float16, torch.float32),  # fbgemm
            (torch.float32, torch.float32),  # TensorIterator
        )

        cases = (
            # out_shape, src_shape, is_ok
            # These cases used to crash with fbgemm, make sure these also raise
            # exceptions with TensorIterator.
            ((1, 2, 3), (0, 2, 3), False),  # same strides, not allowed by TI
            ((1, 5, 6), (4, 5, 6), False),  # same strides, not allowed by TI
            (1, (0, 2, 3), False),  # different strides
            ((4, 5, 6), (0, 2, 3), False),  # different strides
            ((4, 5, 6), (1, 2, 3), False),  # different strides
            ((4, 5, 6), (6, 5, 4), False),  # same numel

            # These cases should pass with fbgemm and TensorIterator.
            ((4, 5, 6), (1, 5, 6), True),  # same strides
            ((4, 5, 6), (4, 5, 6), True),  # same strides
            ((0, 2, 3), 1, True),  # different strides, allowed by TI
            ((4, 5, 6), (4, 5, 1), True),  # different strides, allowed by TI
        )

        for (out_shape, src_shape, is_ok), (out_dtype, src_dtype) in itertools.product(cases, dtypes):
            out = torch.zeros(out_shape, dtype=out_dtype, device=torch.device('cpu'))
            src = torch.ones(src_shape, dtype=src_dtype, device=torch.device('cpu'))
            if is_ok:
                if torch.cuda.is_available():
                    out_cuda = out.cuda()
                    src_cuda = src.cuda()
                res = out.copy_(src)
                if torch.cuda.is_available():
                    res_cuda = out_cuda.copy_(src_cuda)
                    self.assertEqual(res, res_cuda)
            else:
                self.assertRaises(RuntimeError, lambda: out.copy_(src))

    # FIXME: Port to a more appropriate test suite
    def _test_to_with_layout(self, layout):
        def test_copy_behavior(t, non_blocking=False):
            self.assertIs(t, t.to(t, non_blocking=non_blocking))
            self.assertIs(t, t.to(t.dtype, non_blocking=non_blocking))
            self.assertIs(t, t.to(torch.empty_like(t), non_blocking=non_blocking))
            self.assertIsNot(t, t.to(t, non_blocking=non_blocking, copy=True))
            self.assertIsNot(t, t.to(t.dtype, non_blocking=non_blocking, copy=True))
            self.assertIsNot(t, t.to(torch.empty_like(t), non_blocking=non_blocking, copy=True))

            devices = [t.device]
            if t.device.type == 'cuda':
                if t.device.index == -1:
                    devices.append(f'cuda:{torch.cuda.current_device()}')
                elif t.device.index == torch.cuda.current_device():
                    devices.append('cuda')
            for device in devices:
                self.assertIs(t, t.to(device, non_blocking=non_blocking))
                self.assertIs(t, t.to(device, t.dtype, non_blocking=non_blocking))
                self.assertIsNot(t, t.to(device, non_blocking=non_blocking, copy=True))
                self.assertIsNot(t, t.to(device, t.dtype, non_blocking=non_blocking, copy=True))

        a = torch.tensor(5)
        if layout == torch.sparse_csr:
            a = torch.tensor([[0, 1, 2], [2, 0, 3]]).to_sparse_csr()
        test_copy_behavior(a)
        self.assertEqual(a.device, a.to('cpu').device)
        self.assertEqual(a.device, a.to('cpu', dtype=torch.float32).device)
        self.assertIs(torch.float32, a.to('cpu', dtype=torch.float32).dtype)
        self.assertEqual(a.device, a.to(torch.float32).device)
        self.assertIs(torch.float32, a.to(dtype=torch.float32).dtype)

        def test_data_ptr(getter):
            self.assertEqual(getter(a), getter(a.to('cpu')))
            self.assertEqual(getter(a), getter(a.to(dtype=a.dtype, device=a.device, copy=False)))
            self.assertEqual(getter(a), getter(a.to('cpu', copy=False)))
            self.assertNotEqual(getter(a), getter(a.to('cpu', copy=True)))
        if layout == torch.sparse_csr:
            # TODO: compressed sparse tensors currently don't support data_ptr.
            # Exercising failure will allow us to widen coverage of this test once it does.
            with self.assertRaisesRegex(RuntimeError, "Cannot access data pointer of Tensor that doesn't have storage"):
                a.data_ptr()
            # While compressed sparse tensors don't have a concept of data_ptr
            # the underlying tensors do. The implementation of to appropriately forwards
            # the call to the components, which is what we're test here.
            test_data_ptr(lambda a: a.values().data_ptr())
            test_data_ptr(lambda a: a.crow_indices().data_ptr())
            test_data_ptr(lambda a: a.col_indices().data_ptr())
        else:
            test_data_ptr(lambda a: a.data_ptr())

        if torch.cuda.is_available():
            for non_blocking in [True, False]:
                for cuda in ['cuda', 'cuda:0' if torch.cuda.device_count() == 1 else 'cuda:1']:
                    b = torch.tensor(5., device=cuda)
                    test_copy_behavior(b, non_blocking)
                    self.assertEqual(b.device, b.to(cuda, non_blocking=non_blocking).device)
                    self.assertEqual(a.device, b.to('cpu', non_blocking=non_blocking).device)
                    self.assertEqual(b.device, a.to(cuda, non_blocking=non_blocking).device)
                    self.assertIs(torch.int32, b.to('cpu', dtype=torch.int32, non_blocking=non_blocking).dtype)
                    self.assertEqual(a.device, b.to('cpu', dtype=torch.int32, non_blocking=non_blocking).device)
                    self.assertIs(torch.int32, b.to(dtype=torch.int32).dtype)
                    self.assertEqual(b.device, b.to(dtype=torch.int32).device)

    def test_to(self):
        self._test_to_with_layout(torch.strided)
        if torch.version.cuda is not None:
            self._test_to_with_layout(torch.sparse_csr)

    # FIXME: describe this test
    def test_as_subclass(self):
        class SubTensor(torch.Tensor):
            member_var = object()

        t0 = torch.tensor(0)
        t1 = torch.tensor([1, 2])
        t2 = torch.tensor([[3, 4], [5, 6]])

        s0 = t0.as_subclass(SubTensor)
        s1 = t1.as_subclass(SubTensor)
        s2 = t2.as_subclass(SubTensor)

        # Check that the correct type is returned.
        self.assertTrue(type(s0) is SubTensor)
        self.assertTrue(type(s1) is SubTensor)
        self.assertTrue(type(s2) is SubTensor)

        # Check that the data is equal.
        self.assertEqual(t0, s0)
        self.assertEqual(t1, s1)
        self.assertEqual(t2, s2)

        t0[()] = 1
        t1[1] = 3
        t2[1, 1] = 7

        # Check that the data is equal even after modification.
        self.assertEqual(t0, s0)
        self.assertEqual(t1, s1)
        self.assertEqual(t2, s2)

        # Check that member variables are passed through.
        self.assertTrue(s0.member_var is SubTensor.member_var)
        self.assertTrue(s1.member_var is SubTensor.member_var)
        self.assertTrue(s2.member_var is SubTensor.member_var)

        # Test that autograd is propagated.
        t = torch.tensor(5, dtype=torch.float32, requires_grad=True)

        # Run a calculation on the tensor.
        exp_t = torch.exp(t)

        # Cast exp_t to a subclass.
        exp_s = exp_t.as_subclass(SubTensor)

        # Make sure that t.grad was initially None
        self.assertTrue(t.grad is None)

        # Run the autograd calculation.
        exp_s.backward()

        # Make sure autograd was propagated to the original tensor
        # declared with requires_grad.
        self.assertTrue(t.grad is not None)

        # Make sure invalid subclasses raise nice errors
        class BadSubTensor:
            member_var = object()

        err_msg = "Creating a Tensor subclass from a class that does not inherit from Tensor"
        with self.assertRaisesRegex(TypeError, err_msg):
            s0 = t0.as_subclass(BadSubTensor)

    # FIXME: Port to a test suite that better fits slicing
    def test_slice(self):
        empty = torch.empty(0, 4)
        x = torch.arange(0., 16).view(4, 4)
        self.assertEqual(x[:], x)
        self.assertEqual(x[:4], x)
        # start and stop are clamped to the size of dim
        self.assertEqual(x[:5], x)
        # if start >= stop then the result is empty
        self.assertEqual(x[2:1], empty)
        self.assertEqual(x[2:2], empty)
        # out of bounds is also empty
        self.assertEqual(x[10:12], empty)
        # additional correctness checks
        self.assertEqual(x[:1].tolist(), [[0, 1, 2, 3]])
        self.assertEqual(x[:-3].tolist(), [[0, 1, 2, 3]])
        self.assertEqual(x[:, -2:3].tolist(), [[2], [6], [10], [14]])
        self.assertEqual(x[0:-1:2].tolist(), [[0, 1, 2, 3], [8, 9, 10, 11]])

    def test_split_with_sizes_copy_out(self):
        device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
        shape = (30, 40, 50)
        x = torch.rand(*shape, device=device)
        cases = [
            (0, [3, 7, 8, 12]),
            (1, [3, 7, 10, 20]),
            (-2, [3, 7, 10, 20]),
            (2, [3, 7, 10, 12, 18]),
            (-1, [3, 7, 10, 12, 18]),
            (2, [3, 7, 10, 0, 30]),
        ]
        for dim, split_sizes in cases:
            views = x.split_with_sizes(split_sizes, dim=dim)
            expects = [v.clone() for v in views]
            out = [torch.zeros_like(v) for v in views]
            for expect, t in zip(expects, out):
                if expect.numel() != 0:
                    self.assertFalse(expect.eq(t).all().item())

            torch.split_with_sizes_copy(x, split_sizes, dim=dim, out=out)
            for expect, t in zip(expects, out):
                self.assertTrue(expect.eq(t).all().item())

            if not torch.cuda.is_available():
                continue

            # Test with cuda graph
            out = [torch.zeros_like(v) for v in views]
            for expect, t in zip(expects, out):
                if expect.numel() != 0:
                    self.assertFalse(expect.eq(t).all().item())

            g = torch.cuda.CUDAGraph()
            with torch.cuda.graph(g):
                torch.split_with_sizes_copy(x, split_sizes, dim=dim, out=out)

            g.replay()
            for expect, t in zip(expects, out):
                self.assertTrue(expect.eq(t).all().item())

    def test_type(self):
        x = torch.randn(3, 3).double()
        self.assertEqual(x.type('torch.FloatTensor').dtype, torch.float32)
        self.assertEqual(x.type(torch.FloatTensor).dtype, torch.float32)
        self.assertEqual(x.int().type(torch.Tensor).dtype, torch.get_default_dtype())
        self.assertEqual(x.type(torch.int32).dtype, torch.int32)

    # FIXME: port to a quantization test suite
    @unittest.skipIf(IS_MACOS, "https://github.com/pytorch/pytorch/issues/157245")
    @xfailIfS390X
    def test_qengine(self):
        qengines = torch.backends.quantized.supported_engines
        original_qe = torch.backends.quantized.engine
        for qe in qengines:
            torch.backends.quantized.engine = qe
            if torch.backends.quantized.engine != qe:
                raise AssertionError(f"qengine not set successfully: expected {qe}, got {torch.backends.quantized.engine}")
        torch.backends.quantized.engine = original_qe

    def test_terminate_handler_on_crash(self):
        cmd = [sys.executable, '-c', "import os; os.environ[\"TORCH_CUSTOM_TERMINATE\"] ='1'; \
               import torch; import torch._C; torch._C._abort()"]
        with self.assertRaises(subprocess.CalledProcessError) as cm:
            subprocess.check_output(cmd, shell=False)
        e = cm.exception
        output = e.stdout.decode("utf-8")
        self.assertNotEqual(e.returncode, 0)
        self.assertNotEqual(output, None)
        self.assertIn('Unhandled exception caught in c10/util/AbortHandler.h', output)

    # FIXME: port to a distributed test suite
    @slowTest
    def test_multinomial_invalid_probs(self):
        def _spawn_method(self, method, arg):
            try:
                mp.set_start_method('spawn')
            except RuntimeError:
                pass
            with mp.Pool(1) as pool:
                out = pool.map(method, [arg])
                self.assertTrue(out[0])

        def _test_multinomial_invalid_probs(probs):
            try:
                # n_sample = 1 is a special case, test n_sample=2 which is more general
                torch.multinomial(probs.to('cpu'), 2)
                return False  # Should not be reached
            except RuntimeError as e:
                return 'probability tensor contains either `inf`, `nan` or element < 0' in str(e)

            _spawn_method(_test_multinomial_invalid_probs, torch.tensor([1., -1., 1.]))
            _spawn_method(_test_multinomial_invalid_probs, torch.tensor([1., inf, 1.]))
            _spawn_method(_test_multinomial_invalid_probs, torch.tensor([1., -inf, 1.]))
            _spawn_method(_test_multinomial_invalid_probs, torch.tensor([1., 1., nan]))

    # FIXME: port to more appropriate test suite
    def test_to_with_tensor(self):
        a = torch.tensor(5)
        self.assertEqual(a.device, a.to(a).device)

        if torch.cuda.is_available():
            for non_blocking in [True, False]:
                for cuda in ['cuda', 'cuda:0' if torch.cuda.device_count() == 1 else 'cuda:1']:
                    b = torch.tensor(5., device=cuda)
                    self.assertEqual(b.device, b.to(b, non_blocking=non_blocking).device)
                    self.assertEqual(a.device, b.to(a, non_blocking=non_blocking).device)
                    self.assertEqual(b.device, a.to(b, non_blocking=non_blocking).device)

    def test_device(self):
        cpu = torch.device('cpu')
        self.assertEqual('cpu', str(cpu))
        self.assertEqual('cpu', cpu.type)
        self.assertEqual(None, cpu.index)

        cpu0 = torch.device('cpu:0')
        self.assertEqual('cpu:0', str(cpu0))
        self.assertEqual('cpu', cpu0.type)
        self.assertEqual(0, cpu0.index)

        cpu0 = torch.device('cpu', 0)
        self.assertEqual('cpu:0', str(cpu0))
        self.assertEqual('cpu', cpu0.type)
        self.assertEqual(0, cpu0.index)

        cuda = torch.device('cuda')
        self.assertEqual('cuda', str(cuda))
        self.assertEqual('cuda', cuda.type)
        self.assertEqual(None, cuda.index)

        cuda1 = torch.device('cuda:1')
        self.assertEqual('cuda:1', str(cuda1))
        self.assertEqual('cuda', cuda1.type)
        self.assertEqual(1, cuda1.index)

        cuda1 = torch.device('cuda', 1)
        self.assertEqual('cuda:1', str(cuda1))
        self.assertEqual('cuda', cuda1.type)
        self.assertEqual(1, cuda1.index)

        cuda90 = torch.device('cuda', 90)
        self.assertEqual('cuda:90', str(cuda90))
        self.assertEqual('cuda', cuda90.type)
        self.assertEqual(90, cuda90.index)

        self.assertRaises(RuntimeError, lambda: torch.device('cpu:-1'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:-1'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2 '))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda: 2'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2 2'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2.'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2?'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:?2'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2.232'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2 cuda:3'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2+cuda:3'))
        self.assertRaises(RuntimeError, lambda: torch.device('cuda:2cuda:3'))
        self.assertRaises(RuntimeError, lambda: torch.device(-1))

        self.assertRaises(RuntimeError, lambda: torch.device('other'))
        self.assertRaises(RuntimeError, lambda: torch.device('other:0'))

        device_set = {'cpu', 'cpu:0', 'cuda', 'cuda:0', 'cuda:1', 'cuda:10', 'cuda:100'}
        device_hash_set = set()
        device_hash_set.update(hash(torch.device(device)) for device in device_set)
        self.assertEqual(len(device_set), len(device_hash_set))

        def get_expected_device_repr(device):
            if device.index is not None:
                return f"device(type='{device.type}', index={device.index})"

            return f"device(type='{device.type}')"

        for device in device_set:
            dev = torch.device(device)
            self.assertEqual(repr(dev), get_expected_device_repr(dev))

    # Tests that the use_deterministic_flag can be set as expected
    @wrapDeterministicFlagAPITest
    def test_deterministic_flag(self):
        for deterministic, warn_only in product([True, False], [True, False]):
            torch.use_deterministic_algorithms(deterministic, warn_only=warn_only)
            self.assertEqual(deterministic, torch.are_deterministic_algorithms_enabled())
            self.assertEqual(warn_only, torch.is_deterministic_algorithms_warn_only_enabled())

            if deterministic:
                if warn_only:
                    debug_mode = 1
                else:
                    debug_mode = 2
            else:
                debug_mode = 0

            self.assertEqual(debug_mode, torch.get_deterministic_debug_mode())

        for debug_mode in [0, 1, 2]:
            torch.set_deterministic_debug_mode(debug_mode)
            self.assertEqual(debug_mode, torch.get_deterministic_debug_mode())
            deterministic = debug_mode in [1, 2]
            warn_only = debug_mode == 1

            self.assertEqual(deterministic, torch.are_deterministic_algorithms_enabled())
            self.assertEqual(warn_only, torch.is_deterministic_algorithms_warn_only_enabled())

        for debug_mode, debug_mode_str in [(0, 'default'), (1, 'warn'), (2, 'error')]:
            torch.set_deterministic_debug_mode(debug_mode_str)
            self.assertEqual(debug_mode, torch.get_deterministic_debug_mode())

        with self.assertRaisesRegex(
                TypeError,
                r"_set_deterministic_algorithms\(\): argument 'mode' \(position 1\) must be bool, not int"):
            torch.use_deterministic_algorithms(1)

        with self.assertRaisesRegex(
                TypeError,
                r"_set_deterministic_algorithms\(\): argument 'warn_only' must be bool, not int"):
            torch.use_deterministic_algorithms(False, warn_only=1)

    # Tests that torch.utils.deterministic.fill_uninitialized_memory can be set as expected
    def test_deterministic_fill_uninitialized_memory(self):
        with DeterministicGuard(True, fill_uninitialized_memory=False):
            self.assertFalse(torch.utils.deterministic.fill_uninitialized_memory)
            self.assertFalse(torch._C._get_deterministic_fill_uninitialized_memory())

            with DeterministicGuard(True, fill_uninitialized_memory=True):
                self.assertTrue(torch.utils.deterministic.fill_uninitialized_memory)
                self.assertTrue(torch._C._get_deterministic_fill_uninitialized_memory())

            self.assertFalse(torch.utils.deterministic.fill_uninitialized_memory)
            self.assertFalse(torch._C._get_deterministic_fill_uninitialized_memory())

            torch.utils.deterministic.fill_uninitialized_memory = False
            self.assertFalse(torch.utils.deterministic.fill_uninitialized_memory)
            self.assertFalse(torch._C._get_deterministic_fill_uninitialized_memory())

            torch.utils.deterministic.fill_uninitialized_memory = True
            self.assertTrue(torch.utils.deterministic.fill_uninitialized_memory)
            self.assertTrue(torch._C._get_deterministic_fill_uninitialized_memory())

            torch._C._set_deterministic_fill_uninitialized_memory(False)
            self.assertFalse(torch.utils.deterministic.fill_uninitialized_memory)
            self.assertFalse(torch._C._get_deterministic_fill_uninitialized_memory())

            torch._C._set_deterministic_fill_uninitialized_memory(True)
            self.assertTrue(torch.utils.deterministic.fill_uninitialized_memory)
            self.assertTrue(torch._C._get_deterministic_fill_uninitialized_memory())

            with self.assertRaisesRegex(RuntimeError, r"expected a bool, but got int"):
                torch.utils.deterministic.fill_uninitialized_memory = 1

    def test_type_conversion_via_dtype_name(self):
        x = torch.tensor([1])
        self.assertEqual(x.byte().dtype, torch.uint8)
        self.assertEqual(x.bool().dtype, torch.bool)
        self.assertEqual(x.char().dtype, torch.int8)
        self.assertEqual(x.double().dtype, torch.float64)
        self.assertEqual(x.float().dtype, torch.float32)
        self.assertEqual(x.half().dtype, torch.float16)
        self.assertEqual(x.int().dtype, torch.int32)
        self.assertEqual(x.bfloat16().dtype, torch.bfloat16)
        cfloat = x.cfloat()
        self.assertEqual(cfloat.dtype, torch.complex64)
        self.assertEqual(cfloat.real, x.float())
        self.assertEqual(cfloat.imag, torch.zeros_like(cfloat.imag))
        cdouble = x.cdouble()
        self.assertEqual(cdouble.dtype, torch.complex128)
        self.assertEqual(cdouble.real, x.double())
        self.assertEqual(cdouble.imag, torch.zeros_like(cdouble.imag))
        chalf = x.chalf()
        self.assertEqual(chalf.dtype, torch.complex32)
        self.assertEqual(chalf.real, x.half())
        self.assertEqual(chalf.imag, torch.zeros_like(chalf.imag))

    def test_type_alias(self):
        type_alias_map = {torch.float64: torch.double,
                          torch.float32: torch.float,
                          torch.int32: torch.int,
                          torch.int64: torch.long,
                          torch.int16: torch.short,
                          torch.float16: torch.half,
                          torch.complex32: torch.chalf,
                          torch.complex64: torch.cfloat}
        for dtype, alias in type_alias_map.items():
            self.assertIs(alias, dtype)

    def test_doc_template(self) -> None:
        """
        Test that all public API doc strings use the same standard template for
        all common arguments such as tensor or dim
        """
        from torch._torch_docs import __file__ as doc_file
        from torch._torch_docs import multi_dim_common, single_dim_common, factory_common_args, factory_like_common_args

        with open(doc_file, encoding="utf-8") as f:
            doc_strs = f.read()

        matches = re.findall(
            r'add_docstr\(([^,]+?),[^"\']*?(?:"""|\'\'\')(.*?)(?:"""|\'\'\')(?:\.|,?[^,\)]*?\))',
            doc_strs,
            re.MULTILINE | re.DOTALL,
        )
        self.assertTrue(matches)

        for m in matches:
            func = m[0].strip()
            desc = m[1].strip()

            for common_args in [multi_dim_common, single_dim_common, factory_common_args, factory_like_common_args]:
                for k, v in common_args.items():
                    self.assertNotIn(v, desc, lambda msg: f'{msg}\nThe argument description "{v}" in {func} can be replaced by {{{k}}}')

    def test_doc(self):
        checked_types = (types.MethodType, types.FunctionType,
                         types.BuiltinFunctionType, types.BuiltinMethodType)

        def _test_namespace(ns, *skips):
            if isinstance(ns, object):
                ns_name = ns.__class__.__name__
            else:
                ns_name = ns.__name__
            skip_regexes = []
            for r in skips:
                if isinstance(r, str):
                    skip_regexes.append(re.compile(f'^{re.escape(r)}$'))
                else:
                    skip_regexes.append(r)

            for name in dir(ns):
                if name.startswith('_'):
                    continue
                if name in ['real', 'imag']:
                    y = torch.randn(1, dtype=torch.cfloat)
                    var = getattr(y, name)
                elif name in ["H", "mT", "mH"]:
                    y = torch.randn(1, 1)
                    var = getattr(y, name)
                else:
                    var = getattr(ns, name)
                if not isinstance(var, checked_types):
                    continue
                doc = var.__doc__
                has_doc = doc is not None and len(doc.strip()) > 0
                full_name = ns_name + '.' + name
                if any(r.match(name) for r in skip_regexes):
                    self.assertFalse(has_doc,
                                     lambda msg: f'{msg}\nNew docs have been added for {full_name}, please remove '
                                     'it from the skipped list in TestTorch.test_doc')
                else:
                    self.assertTrue(has_doc, lambda msg: f'{msg}\n{full_name} is missing documentation')

            # FIXME: All of the following should be marked as expected failures
            # so that it is easier to tell when missing has been added.
            # FIXME: fix all the skipped ones below!
            test_namespace(torch.randn(1),  # noqa: F821
                           'as_strided_',
                           re.compile('^clamp_(min|max)_?$'),
                           'is_distributed',
                           'is_nonzero',
                           'is_same_size',
                           'log_softmax',
                           'map2_',
                           'new',
                           'reinforce',
                           'relu',
                           'relu_',
                           'prelu',
                           'resize',
                           'resize_as',
                           'softmax',
                           'split_with_sizes',
                           'unsafe_split_with_sizes',
                           '_autocast_to_fp16',
                           '_autocast_to_fp32',
                           )

            test_namespace(torch.nn)  # noqa: F821
            test_namespace(torch.nn.functional, 'assert_int_or_pair')  # noqa: F821
            # TODO: add torch.* tests when we have proper namespacing on ATen functions
            # test_namespace(torch)

    # FIXME: deprecate torch.Tensor constructor
    def test_tensor_ctor_scalar(self):
        x = torch.Tensor(torch.tensor(1.0))
        self.assertEqual(x, torch.tensor(1.0))

    def test_deepcopy_gradient(self):
        from copy import deepcopy
        a = torch.zeros(10)
        a.grad = torch.ones(10)
        self.assertEqual(a.grad, deepcopy(a).grad)
        s = torch.zeros(10).to_sparse()
        s.grad = torch.ones(10).to_sparse()
        self.assertEqual(s.grad, deepcopy(s).grad)

        # ensure sharing is not broken
        c = deepcopy([a, a.grad])
        self.assertTrue(c[0].grad is c[1])

    def test_tensor_base_init(self):
        # Direct construction not OK
        self.assertRaises(RuntimeError, lambda: torch._C.TensorBase())

        # Subclassing it directly no OK
        with self.assertRaisesRegex(RuntimeError, "Cannot subclass"):
            class Tfail(torch._C.TensorBase):
                pass

        # Doing so with Tensor is ok though
        class T(torch.Tensor):
            pass

        T()

    def test_storage_base_init(self):
        # Direct construction not OK
        self.assertRaises(RuntimeError, lambda: torch._C.StorageBase())

        # But construction of subclass is OK
        class T(torch._C.StorageBase):
            pass

        T()

    def test_tensor_base_new(self):

        # OK to call super().__new__, see
        # https://github.com/pytorch/pytorch/issues/57421
        class TestTensor(torch.Tensor):
            @staticmethod
            def __new__(cls, x, *args, **kwargs):
                return super().__new__(cls, x, *args, **kwargs)

        x = torch.ones(5)
        TestTensor(x)

    def test_storage_base_new(self):

        # OK to call super().__new__, see
        # https://github.com/pytorch/pytorch/issues/57421
        class TestStorage(torch._C.StorageBase):
            @staticmethod
            def __new__(cls, x, *args, **kwargs):
                return super().__new__(cls, x, *args, **kwargs)

        x = torch.UntypedStorage(5)
        TestStorage(x)

    def test_pyobj_preserved(self):
        x = torch.empty(2)
        x.foo = 2  # put something on __dict__
        y = torch.empty(2)
        y.grad = x
        del x  # x is dead in Python
        self.assertEqual(y.grad.foo, 2)
        z = y.grad  # it's live
        del z  # it's dead again
        self.assertEqual(y.grad.foo, 2)

    def test_subclass_preserved(self):
        class MyTensor(torch.Tensor):
            pass

        x = MyTensor(torch.empty(2))
        y = torch.empty(2)
        y.grad = x
        del x  # x is dead in Python
        self.assertEqual(type(y.grad), MyTensor)
        z = y.grad  # it's live
        del z  # it's dead again
        self.assertEqual(type(y.grad), MyTensor)

    @skipIfTorchDynamo("Tracker hook does not work in TorchDynamo")
    def test_storage_dealloc(self):
        m, t = Tracker.make()
        s0 = torch.UntypedStorage(10)
        s1 = s0
        s0._tracker = t
        del t

        self.assertFalse(m[0])
        del s0
        self.assertFalse(m[0])
        del s1
        self.assertTrue(m[0])

    @skipIfTorchDynamo("Tracker hook does not work in TorchDynamo")
    def test_storage_from_tensor_dealloc(self):
        m, t = Tracker.make()
        a = torch.randn(10)
        s0 = a.untyped_storage()
        s0._tracker = t
        del t

        s1 = a.untyped_storage()
        self.assertTrue(s0 is s1)
        self.assertTrue(hasattr(s1, '_tracker'))

        del a

        self.assertFalse(m[0])
        del s0
        self.assertFalse(m[0])
        del s1
        self.assertTrue(m[0])

    @skipIfTorchDynamo("Tracker hook does not work in TorchDynamo")
    def test_storage_from_tensor_dealloc_zombie(self):
        m, t = Tracker.make()
        a = torch.randn(10)
        s0 = a.untyped_storage()
        s0._tracker = t
        del t

        s1 = a.untyped_storage()
        self.assertTrue(s0 is s1)
        self.assertTrue(hasattr(s1, '_tracker'))

        self.assertFalse(m[0])
        del s0
        self.assertFalse(m[0])
        del s1
        self.assertFalse(m[0])
        del a
        self.assertTrue(m[0])

    @skipIfTorchDynamo("Tracker hook does not work in TorchDynamo")
    def test_storage_from_tensor_dealloc_resurrected(self):
        m, t = Tracker.make()
        a = torch.randn(10)
        s0 = a.untyped_storage()
        s0._tracker = t
        del t

        s1 = a.untyped_storage()
        self.assertTrue(s0 is s1)
        self.assertTrue(hasattr(s1, '_tracker'))

        self.assertFalse(m[0])
        del s0
        self.assertFalse(m[0])
        del s1
        self.assertFalse(m[0])

        s0 = a.untyped_storage()
        self.assertTrue(isinstance(s0, torch.UntypedStorage))

        del a
        self.assertFalse(m[0])
        del s0
        self.assertTrue(m[0])

    @skipIfTorchDynamo("Tracker hook does not work in TorchDynamo")
    def test_storage_dealloc_resurrected(self):
        m, t = Tracker.make()
        s = torch.UntypedStorage(10)
        s._tracker = t
        del t

        a = torch.tensor(s)
        self.assertFalse(m[0])
        del s

        self.assertFalse(m[0])

        s = a.untyped_storage()
        self.assertTrue(isinstance(s, torch.UntypedStorage))

        del a
        self.assertFalse(m[0])
        del s
        self.assertTrue(m[0])

    @skipIfTorchDynamo("Tracker hook does not work in TorchDynamo")
    def test_storage_dealloc_subclass_zombie(self):
        class MyStorage(torch.UntypedStorage):
            finalized_count = 0

            def __del__(self):
                MyStorage.finalized_count += 1

        m, t = Tracker.make()
        s = MyStorage(10)
        s._tracker = t
        del t

        a = torch.tensor(s)
        self.assertFalse(m[0])
        del s

        self.assertEqual(MyStorage.finalized_count, 0)
        self.assertFalse(m[0])

        del a
        self.assertEqual(MyStorage.finalized_count, 1)
        self.assertTrue(m[0])

    @skipIfTorchDynamo("Tracker hook does not work in TorchDynamo")
    def test_storage_dealloc_subclass_resurrected(self):
        class MyStorage(torch.UntypedStorage):
            finalized_count = 0

            def __del__(self):
                MyStorage.finalized_count += 1

        m, t = Tracker.make()
        s = MyStorage(10)
        s._tracker = t
        del t

        a = torch.tensor(s)
        self.assertFalse(m[0])
        del s

        self.assertEqual(MyStorage.finalized_count, 0)
        self.assertFalse(m[0])

        s = a.untyped_storage()
        del a
        self.assertFalse(m[0])
        self.assertEqual(MyStorage.finalized_count, 0)
        self.assertTrue(isinstance(s, MyStorage))
        del s
        self.assertEqual(MyStorage.finalized_count, 1)
        self.assertTrue(m[0])

    @skipIfTorchDynamo("cannot trace t2.grad_fn._saved_self")
    def test_tensor_ressurecting_clear(self):
        # Regression test for https://github.com/pytorch/pytorch/issues/136358
        # A Tensor with custom __dict__
        # Autograd here is for the c++ reference later
        t = torch.rand(2, requires_grad=True).clone()
        t.foo = 2

        # that is part of a cycle
        l = []
        l.append(42)
        l.append(t)

        # Keep the Tensor alive from c++
        # Using autograd graph here (any other mean would work)
        t2 = t ** 2
        self.assertIs(t2.grad_fn._saved_self, t)

        # Clear all python references and trigger the gc
        del t, l
        gc.collect()

        # We used to loose the dict!
        self.assertTrue(hasattr(t2.grad_fn._saved_self, "foo"))

    def test_tensor_slot_dealloc(self):

        class SlotTensor1(torch.Tensor):
            __slots__ = ['slot1']

        class SlotTensor2(SlotTensor1):
            __slots__ = ['slot2']

        m1, t1 = Tracker.make()
        m2, t2 = Tracker.make()
        slot_tensor = SlotTensor2(torch.empty(2))
        slot_tensor.slot1 = t1
        slot_tensor.slot2 = t2
        del t1
        del t2
        self.assertFalse(m1[0])
        self.assertFalse(m2[0])
        del slot_tensor
        self.assertTrue(m1[0])
        self.assertTrue(m2[0])

    def test_storage_slot_dealloc(self):

        class SlotStorage1(torch._C.StorageBase):
            __slots__ = ['slot1']

        class SlotStorage2(SlotStorage1):
            __slots__ = ['slot2']

        m1, t1 = Tracker.make()
        m2, t2 = Tracker.make()
        slot_storage = SlotStorage2(torch.UntypedStorage(2))
        slot_storage.slot1 = t1
        slot_storage.slot2 = t2
        del t1
        del t2
        self.assertFalse(m1[0])
        self.assertFalse(m2[0])
        del slot_storage
        self.assertTrue(m1[0])
        self.assertTrue(m2[0])

    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    def test_tensor_dict_dealloc(self):
        m, t = Tracker.make()
        x = torch.empty(2)
        x.arf = t
        del t
        self.assertFalse(m[0])
        del x
        self.assertTrue(m[0])

    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    def test_storage_dict_dealloc(self):
        m, t = Tracker.make()
        x = torch.UntypedStorage(2)
        x.arf = t
        del t
        self.assertFalse(m[0])
        del x
        self.assertTrue(m[0])

    def test_tensor_finalizer_dealloc(self):
        m = [False]

        class FinalizerTensor(torch.Tensor):
            def __del__(self):
                m[0] = True

        fin_tensor = FinalizerTensor(torch.empty(2))
        self.assertFalse(m[0])
        del fin_tensor
        self.assertTrue(m[0])

    def test_storage_finalizer_dealloc(self):
        m = [False]

        class FinalizerStorage(torch._C.StorageBase):
            def __del__(self):
                m[0] = True

        fin_storage = FinalizerStorage(torch.UntypedStorage(2))
        self.assertFalse(m[0])
        del fin_storage
        self.assertTrue(m[0])

    @skipIfTorchDynamo("https://github.com/pytorch/torchdynamo/issues/1993")
    def test_tensor_weakref_dealloc(self):
        x = torch.empty(2)
        m = [False]

        def cb(r):
            m[0] = True

        wref = weakref.ref(x, cb)
        del x
        self.assertTrue(m[0])
        self.assertEqual(wref(), None)

    @skipIfTorchDynamo("https://github.com/pytorch/torchdynamo/issues/1993")
    def test_storage_weakref_dealloc(self):

        x = torch.UntypedStorage(2)
        m = [False]

        def cb(r):
            m[0] = True

        wref = weakref.ref(x, cb)
        del x
        self.assertTrue(m[0])
        self.assertEqual(wref(), None)

    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    def test_tensor_cycle_via_dict(self):
        m1, t1 = Tracker.make()
        x = torch.empty(2)
        x._tracker = t1
        del t1

        m2, t2 = Tracker.make()
        y = torch.empty(2)
        y._tracker = t2
        del t2

        x._loop = y
        y._loop = x

        # C++ reference should keep the cycle live!
        # This exercise THPVariable_subtype_traverse
        # NB: Because z.grad is a reference done entirely in C++, cycles
        # involving it directly are NOT broken by Python GC; you've
        # set up a good old C++ reference cycle which we cannot safely
        # break (because C++ references are allowed to be accessed
        # multithreaded-ly) (TODO: except maybe if you can prove that
        # only Python has access to the C++ object, in which case you can
        # also prove that no multithreaded access occurs)
        z = torch.empty(2)
        z.grad = x

        del x
        del y

        gc.collect()
        self.assertFalse(m1[0])
        self.assertFalse(m2[0])

        with disable_gc():
            del z
            self.assertFalse(m1[0])
            self.assertFalse(m2[0])

        gc.collect()
        self.assertTrue(m1[0])
        self.assertTrue(m2[0])

    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    def test_storage_cycle_via_dict(self):
        m1, t1 = Tracker.make()
        x = torch.UntypedStorage(2)
        x._tracker = t1
        del t1

        m2, t2 = Tracker.make()
        y = torch.UntypedStorage(2)
        y._tracker = t2
        del t2

        x._loop = y
        y._loop = x

        # C++ reference should keep the cycle live!
        # This exercise THPVariable_subtype_traverse
        # NB: Because z.grad is a reference done entirely in C++, cycles
        # involving it directly are NOT broken by Python GC; you've
        # set up a good old C++ reference cycle which we cannot safely
        # break (because C++ references are allowed to be accessed
        # multithreaded-ly) (TODO: except maybe if you can prove that
        # only Python has access to the C++ object, in which case you can
        # also prove that no multithreaded access occurs)
        z = torch.UntypedStorage(2)
        z.grad = x

        del x
        del y

        gc.collect()
        self.assertFalse(m1[0])
        self.assertFalse(m2[0])

        with disable_gc():
            del z
            self.assertFalse(m1[0])
            self.assertFalse(m2[0])

        gc.collect()
        self.assertTrue(m1[0])
        self.assertTrue(m2[0])

    def test_tensor_cycle_via_slots(self):
        m1 = [False]
        m2 = [False]

        class SlotTensor1(torch.Tensor):
            __slots__ = ['slot1']

            def __del__(self):
                m1[0] = True

        class SlotTensor2(SlotTensor1):
            __slots__ = ['slot2']

            def __del__(self):
                m2[0] = True

        x = SlotTensor1(torch.empty(2))
        x_ref = weakref.ref(x)
        y = SlotTensor2(torch.empty(2))

        x.slot1 = y
        y.slot2 = x

        del x
        with disable_gc():
            del y
            self.assertFalse(m1[0])
            self.assertFalse(m2[0])

        gc.collect()
        self.assertTrue(m1[0])
        self.assertTrue(m2[0])

        self.assertIsNone(x_ref())

        # At this point, we know the finalizer ran and the weakref
        # was cleared. But is the object really gone?
        self.assertFalse(any(isinstance(o, SlotTensor1) for o in gc.get_objects()))


    def test_storage_cycle_via_slots(self):
        m1 = [False]
        m2 = [False]

        class SlotStorage1(torch._C.StorageBase):
            __slots__ = ['slot1']

            def __del__(self):
                m1[0] = True

        class SlotStorage2(SlotStorage1):
            __slots__ = ['slot2']

            def __del__(self):
                m2[0] = True

        x = SlotStorage1(torch.UntypedStorage(2))
        y = SlotStorage2(torch.UntypedStorage(2))

        x.slot1 = y
        y.slot2 = x

        del x
        with disable_gc():
            del y
            self.assertFalse(m1[0])
            self.assertFalse(m2[0])

        gc.collect()
        self.assertTrue(m1[0])
        self.assertTrue(m2[0])

    @skipIfTorchDynamo("Not a suitable test for TorchDynamo")
    def test_storage_preserve_nonhermetic_in_hermetic_context(self):
        from torch.library import _scoped_library, impl
        global _my_storage

        with _scoped_library("my_lib", "DEF") as my_lib:
            my_lib.define('my_func() -> None')

            a = torch.tensor([1.])
            _my_storage = a.untyped_storage()

            m, t = Tracker.make()
            _my_storage._tracker = t
            del t

            @impl(my_lib, 'my_func', '')
            def my_func():
                global _my_storage
                del _my_storage

            self.assertFalse(m[0])
            torch.ops.my_lib.my_func()
            self.assertFalse(m[0])

            s = a.untyped_storage()
            del a
            del s
            self.assertTrue(m[0])

    # FIXME: move to test_autograd?
    @skipIfTorchDynamo("TorchDynamo does not work well with hooks")
    def test_backward_hooks_traverse(self):
        m1, t1 = Tracker.make()
        m2, t2 = Tracker.make()
        x = torch.empty(2, requires_grad=True)
        x._tracker = t1
        y = torch.empty(2, requires_grad=True)
        y._tracker = t2
        del t1
        del t2

        # this hits a special setter, it's not just a __dict__ entry
        x._backward_hooks = y
        y._backward_hooks = x

        del x
        with disable_gc():
            del y
            self.assertFalse(m1[0])
            self.assertFalse(m2[0])

        gc.collect()

        self.assertTrue(m1[0])
        self.assertTrue(m2[0])

    @skipIfTorchDynamo("https://github.com/pytorch/torchdynamo/issues/1993")
    def test_tensor_dead_weak_ref(self):
        x = torch.ones(2)
        w_x = weakref.ref(x)
        y = torch.ones(2)
        y.grad = x
        del x

        x = w_x()
        # x should keep the tensor live. This didn't happen in earlier PyTorch
        # versions.
        del y

        self.assertEqual(2, x.sum())

        del x
        self.assertIsNone(w_x())

    @skipIfTorchDynamo("https://github.com/pytorch/torchdynamo/issues/1993")
    def test_storage_dead_weak_ref(self):
        x = torch.UntypedStorage(2)
        w_x = weakref.ref(x)
        y = torch.tensor(x)
        del x
        self.assertIsNotNone(w_x())
        del y
        self.assertIsNone(w_x())

    def test_tensor_resurrected_weak_ref(self):
        x = torch.empty(2)
        w_x = weakref.ref(x)
        y = torch.empty(2)
        y.grad = x
        del x

        x = w_x()
        # Use this to manually fix weak references after dereferencing them
        x._fix_weakref()
        del y
        x.sigmoid()

    def test_storage_resurrected_weak_ref(self):
        x = torch.UntypedStorage(2)
        w_x = weakref.ref(x)
        y = torch.tensor(x)
        del x

        x = w_x()
        # Use this to manually fix weak reference after dereferencing them
        x._fix_weakref()
        del y
        x.float()

    @skipIfTorchDynamo("https://github.com/pytorch/torchdynamo/issues/1993")
    def test_tensor_fix_weakref_no_leak(self):
        import weakref

        called = False

        a = torch.randn(1)

        def callback(w):
            nonlocal called
            called = True
        _wa = weakref.ref(a, callback)
        a._fix_weakref()
        del a

        self.assertTrue(called)

    @skipIfTorchDynamo("https://github.com/pytorch/torchdynamo/issues/1993")
    def test_storage_fix_weakref_no_leak(self):
        import weakref

        called = False

        a = torch.UntypedStorage(1)

        def callback(w):
            nonlocal called
            called = True
        _wa = weakref.ref(a, callback)
        a._fix_weakref()
        del a

        self.assertTrue(called)

    # FIXME: move to test_linalg
    @torch.inference_mode()
    def test_bmm_multithreaded(self):
        device = 'cpu'
        num_threads = torch.get_num_threads()

        torch.set_num_threads(4)
        batch_sizes = [1, 10]
        M, N, O = 23, 8, 12
        dtype = torch.float32
        numpy_dtype = dtype

        def invert_perm(p):
            d = {x: i for i, x in enumerate(p)}
            return (d[0], d[1], d[2])

        def generate_inputs(num_batches):
            # transposed tensors
            for perm1, perm2 in itertools.product(itertools.permutations((0, 1, 2)), repeat=2):
                b1 = make_tensor((num_batches, M, N), dtype=dtype, device=device, low=-1, high=1)
                b2 = make_tensor((num_batches, N, O), dtype=dtype, device=device, low=-1, high=1)
                b1 = b1.permute(perm1).contiguous().permute(invert_perm(perm1))
                b2 = b2.permute(perm2).contiguous().permute(invert_perm(perm2))
                yield b1, b2
            # broadcasting tensors
            for b1, b2, b3, b4, b5, b6 in itertools.product((True, False), repeat=6):
                shape1 = (num_batches if b1 else 1, M if b2 else 1, N if b3 else 1)
                shape2 = (num_batches if b4 else 1, N if b5 else 1, O if b6 else 1)
                b1 = make_tensor(shape1, dtype=dtype, device=device, low=-1, high=1).expand(num_batches, M, N)
                b2 = make_tensor(shape2, dtype=dtype, device=device, low=-1, high=1).expand(num_batches, N, O)
                yield b1, b2
            # zero-sized tensors
            for z1, z2, z3, z4 in itertools.product((True, False), repeat=4):
                shape1 = (num_batches if z1 else 0, M if z2 else 0, N if z3 else 0)
                shape2 = (num_batches if z1 else 0, N if z3 else 0, O if z4 else 0)
                b1 = torch.randn(shape1, dtype=dtype, device=device)
                b2 = torch.randn(shape2, dtype=dtype, device=device)
                yield b1, b2

        try:
            for num_batches in batch_sizes:
                for (b1, b2), perm3 in itertools.product(generate_inputs(num_batches), itertools.permutations((0, 1, 2))):
                    res1 = torch.bmm(b1, b2)
                    res2 = torch.full((num_batches, M, O), math.nan, dtype=dtype, device=device) \
                        .permute(perm3).contiguous().permute(invert_perm(perm3))
                    torch.bmm(b1, b2, out=res2)
                    expect = torch.from_numpy(
                        b1.to(numpy_dtype).cpu().numpy() @ b2.to(numpy_dtype).cpu().numpy()).to(device=device, dtype=dtype)
                    self.assertEqual(expect, res1)
                    self.assertEqual(expect, res2)
        finally:
            torch.set_num_threads(num_threads)

    @unittest.skipIf(not torch.cuda.is_available(), "CUDA not available")
    def test_bmm_matmul_mixed_dtype_error(self):
        a = torch.randn(2, 8, 8, device="cuda", dtype=torch.float16)
        b = torch.randn(2, 8, 64, device="cuda", dtype=torch.float32)

        with self.assertRaisesRegex(RuntimeError, "expected scalar type .* but found"):
            torch.bmm(a, b)

        with self.assertRaisesRegex(RuntimeError, "expected scalar type .* but found"):
            torch.compile(lambda x, y: torch.bmm(x, y), fullgraph=True)(a, b)

        with self.assertRaisesRegex(RuntimeError, "expected scalar type .* but found"):
            torch.matmul(a, b)

        with self.assertRaisesRegex(RuntimeError, "expected scalar type .* but found"):
            torch.compile(lambda x, y: torch.matmul(x, y), fullgraph=True)(a, b)

    def test_conj_neg_tolist(self):
        x = torch.randn(2, dtype=torch.cfloat)
        y1 = x.conj()
        y1_expect = x.conj_physical()
        y2 = y1.imag
        self.assertEqual(y1, y1_expect.tolist())
        self.assertEqual(y2, y1_expect.imag.tolist())

    @unittest.skipIf(torch.backends.cuda.is_built(), "Skipped for cuda-enabled build")
    def test_no_cuda_monkeypatch(self):
        # Note that this is not in test_cuda.py as this whole file is skipped when cuda
        # is not available.
        with self.assertRaisesRegex(RuntimeError, "torch.cuda.Stream requires CUDA support"):
            torch.cuda.Stream()

        with self.assertRaisesRegex(RuntimeError, "Tried to instantiate dummy base class Event"):
            torch.cuda.Event()

        with self.assertRaisesRegex(RuntimeError, "Tried to instantiate dummy base class CUDAGraph"):
            torch.cuda.graphs.CUDAGraph()

    def test_tensor_where_scalar(self):

        a = torch.arange(4.0)
        not_zero = 0.001

        # b is generated through torch.where function with not_zero being a scalar parameter
        b = torch.where(a != 0, a, not_zero)
        # c is generated through Tensor.where method with not_zero being a scalar parameter
        c = a.where(a != 0, not_zero)

        self.assertEqual(b, c)

    def test_data_ptr_of_empty_tensor_with_storage(self):
        t = torch.empty((2, 2))
        self.assertNotEqual(t.data_ptr(), 0)
        t.resize_((0, 2))
        self.assertEqual(t.data_ptr(), 0)

    def test_data_ptr_of_empty_view_with_storage(self):
        t = torch.empty((2, 2))
        self.assertNotEqual(t.data_ptr(), 0)
        t2 = t[0:0].view(0, 1)
        self.assertEqual(t2.data_ptr(), 0)

    def test_size_stride(self) -> None:
        t = torch.rand(2, 3, dtype=torch.float32)
        self.assertEqual(t.size(0), 2)
        self.assertEqual(t.size(dim=None), torch.Size([2, 3]))
        self.assertEqual(t.stride(dim=None), torch.Size([3, 1]))
        self.assertEqual(t.t().stride(), torch.Size([1, 3]))

    def test_invalid_arg_error_handling(self) -> None:
        """ Tests that errors from old TH functions are propagated back """
        for invalid_val in [-1, 2**65]:
            self.assertRaises((ValueError, RuntimeError), lambda: torch.set_num_threads(invalid_val))
            self.assertRaises((ValueError, RuntimeError), lambda: torch.set_num_interop_threads(invalid_val))

    def test_intlist_arg_with_misbehaving_list_subclass(self) -> None:
        # Regression test for https://github.com/pytorch/pytorch/issues/184877:
        # an INTERNAL ASSERT fired when the arg parser tried to format a type
        # error for a list subclass whose __len__/__iter__ raised.
        class Foo(list):
            def __iter__(self):
                raise RuntimeError("should not be called")

            def __len__(self):
                raise RuntimeError("should not be called")

        x = Foo()
        x.append("a")
        with self.assertRaisesRegex(
            TypeError, r"found element of type str at pos 0"
        ):
            torch.randn(x)

    def _get_tensor_prop(self, t):
        preserved = (
            id(t),
            # Refcount values get modified by Dynamo resume frames
            0 if TEST_WITH_TORCHDYNAMO else sys.getrefcount(t),
        )
        slotnames = copyreg._slotnames(t.__class__)
        moved = (
            slotnames,
            id(t.__dict__),
            tuple(t.__dict__.keys()),
            [getattr(t, name, None) for name in slotnames]
        )
        return preserved, moved

    def _checked_swap(self, t1, t2):
        t1_pres, t1_moved = self._get_tensor_prop(t1)
        t2_pres, t2_moved = self._get_tensor_prop(t2)

        torch.utils.swap_tensors(t1, t2)

        new_t1_pres, new_t1_moved = self._get_tensor_prop(t1)
        new_t2_pres, new_t2_moved = self._get_tensor_prop(t2)
        self.assertEqual(t1_pres, new_t1_pres)
        self.assertEqual(t2_pres, new_t2_pres)
        self.assertEqual(t1_moved, new_t2_moved)
        self.assertEqual(t2_moved, new_t1_moved)

        # tests that PyObject slots on TensorImpl are correctly swapped by
        # checking that when the function applied on a swapped tensor is
        # returns doesn't change the TensorImpl, the returned value (which is
        # given by returning the reference to the PyObject in the TensorImpl's
        # PyObjectSlot) is still correct
        self.assertEqual(id(t1.fill_(0.5)), id(t1))
        self.assertEqual(id(t2.fill_(0.5)), id(t2))

    @unittest.skipIf(TEST_WITH_TORCHDYNAMO, "Dynamo adds weakrefs")
    def test_swap_basic(self):
        ts = [
            torch.rand(2),
            torch.rand(3, 3),
            torch.empty(3, dtype=torch.int),
            TwoTensor(torch.rand(4), torch.rand(4))
        ]

        for t1, t2 in itertools.combinations(ts, 2):
            t1 = t1.clone()
            t2 = t2.clone()
            t2.foo = "bar"
            holder = []
            holder.append(t1)

            self._checked_swap(t1, t2)

            self.assertIs(holder[0], t1)
            self.assertEqual(t1.foo, "bar")

            if t1.is_floating_point():
                t3 = t1.detach().clone().requires_grad_(True)
                out = t3 * 2
                torch.utils.swap_tensors(t3, t2)
                with self.assertRaisesRegex(RuntimeError, "AccumulateGrad node that was poisoned by swap_tensors"):
                    out.sum().backward()

            _wr = weakref.ref(t1)
            with self.assertRaisesRegex(RuntimeError, "has weakref"):
                torch.utils.swap_tensors(t1, t2)


    @unittest.skipIf(TEST_WITH_TORCHDYNAMO, "Dynamo adds weakrefs")
    def test_swap_fail_slots(self):
        class MyTwoTensor(TwoTensor):
            __slots__ = ("a", "b")

        class MyTwoTensor2(TwoTensor):
            __slots__ = ("b", "a")

        class MyTwoTensor3(TwoTensor):
            __slots__ = ("a", "b", "c", "d")

        class MyTwoTensor4(TwoTensor):
            __slots__ = ("a", "c")


        t1 = torch.rand(4)
        t2 = TwoTensor(torch.rand(4), torch.rand(4))
        t3 = MyTwoTensor(torch.rand(4), torch.rand(4))
        t4 = MyTwoTensor(torch.rand(4), torch.rand(4))
        t5 = MyTwoTensor2(torch.rand(4), torch.rand(4))
        t6 = MyTwoTensor3(torch.rand(4), torch.rand(4))
        t7 = MyTwoTensor3(torch.rand(4), torch.rand(4))
        t8 = MyTwoTensor4(torch.rand(4), torch.rand(4))

        self._checked_swap(t1, t2)
        with self.assertRaisesRegex(RuntimeError, "Cannot swap t1 and t2 if they have different slots"):
            torch.utils.swap_tensors(t1, t3)
        with self.assertRaisesRegex(RuntimeError, "Cannot swap t1 and t2 if they have different slots"):
            torch.utils.swap_tensors(t2, t3)
        with self.assertRaisesRegex(RuntimeError, "Cannot swap t1 and t2 if they have different slots"):
            torch.utils.swap_tensors(t2, t8)
        self._checked_swap(t3, t4)
        self._checked_swap(t3, t5)
        with self.assertRaisesRegex(RuntimeError, "Cannot swap t1 and t2 if they have different slots"):
            torch.utils.swap_tensors(t3, t6)
        t3.c = "foo"
        t4.d = "bar"
        self._checked_swap(t3, t4)
        self.assertEqual(t4.c, "foo")
        self.assertEqual(t3.d, "bar")
        t6.c = "cat"
        t7.d = "dog"
        self._checked_swap(t6, t7)

    @unittest.skipIf(torch.cuda.is_available(), "Test specific for CPU")
    def test_bf16_supported_on_cpu(self):
        self.assertFalse(torch.cuda.is_bf16_supported())

    def test_tensor_with_grad_to_scalar_warning(self) -> None:
        with (warnings.catch_warnings(record=True) as w,
                set_warn_always_context(True)):
            warnings.simplefilter("always")

            x = torch.tensor(2.0, requires_grad=True)
            math.pow(x, 3)  # calling this results in a warning

            self.assertEqual(len(w), 1)
            self.assertTrue(issubclass(w[0].category, UserWarning))
            self.assertIn(
                "Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.",
                str(w[0].message)
            )

    def test_tensor_item_no_warning(self):
        with (warnings.catch_warnings(record=True) as w,
                set_warn_always_context(True)):
            warnings.simplefilter("always")

            x = torch.tensor(2.0, requires_grad=True)
            max(x, 3)  # No warning
            x.item()  # No warning

            self.assertEqual(len(w), 0)

# The following block extends TestTorch with negative dim wrapping tests
# FIXME: replace these with OpInfo sample inputs or systemic OpInfo tests
# Functions to test negative dimension wrapping
METHOD = 1
INPLACE_METHOD = 2
FUNCTIONAL = 4
DIM_ARG: None = None

def make_neg_dim_test(name, tensor_arg, arg_constr, types, extra_dim=0):
    def neg_dim_test(self):
        if isinstance(tensor_arg, list):
            if METHOD in types or INPLACE_METHOD in types:
                raise AssertionError("METHOD and INPLACE_METHOD should not be in types for list tensor_arg")
            x = [torch.randn(arg) for arg in tensor_arg]
            ndim = len(tensor_arg[-1])
        else:
            x = torch.randn(*tensor_arg)
            ndim = len(tensor_arg)
        ndim += extra_dim

        n_dim_to_test = sum(e is DIM_ARG for e in arg_constr())

        for dims_val in combinations(range(ndim), n_dim_to_test):
            arg = arg_constr()
            arg_neg = copy.deepcopy(arg)
            idx = 0
            for i, v in enumerate(arg):
                if v is DIM_ARG:
                    arg[i] = dims_val[idx]
                    arg_neg[i] = dims_val[idx] - ndim
                    idx += 1

            if METHOD in types:
                a = getattr(x, name)(*arg)
                b = getattr(x, name)(*arg_neg)
                self.assertEqual(a, b)

            if INPLACE_METHOD in types:
                a = x.clone()
                getattr(a, name + '_')(*arg)
                b = x.clone()
                getattr(b, name + '_')(*arg_neg)
                self.assertEqual(a, b)

            if FUNCTIONAL in types:
                a = getattr(torch, name)(x, *arg)
                b = getattr(torch, name)(x, *arg_neg)
                self.assertEqual(a, b)

    return neg_dim_test

def idx_tensor(size, max_val):
    return torch.LongTensor(*size).random_(0, max_val - 1)

def add_neg_dim_tests():
    neg_dim_tests = [
        ('narrow', (10, 20, 30), lambda: [DIM_ARG, 0, 5], [METHOD]),
        ('transpose', (10, 20, 30), lambda: [DIM_ARG, DIM_ARG], [METHOD, INPLACE_METHOD, FUNCTIONAL]),
        ('size', (10, 20, 30), lambda: [DIM_ARG], [METHOD]),
        ('cat', [(2, 3, 4), (2, 3, 4)], lambda: [DIM_ARG], [FUNCTIONAL]),
        ('chunk', (10, 20, 30), lambda: [5, DIM_ARG], [METHOD, FUNCTIONAL]),
        ('gather', (10, 20), lambda: [DIM_ARG, idx_tensor((10, 20), 10)], [METHOD, FUNCTIONAL]),
        ('index_select', (10, 10), lambda: [DIM_ARG, idx_tensor((10,), 10)], [METHOD, FUNCTIONAL]),
        ('split', (10, 20), lambda: [5, DIM_ARG], [METHOD, FUNCTIONAL]),
        ('squeeze', (10, 1, 20, 1), lambda: [DIM_ARG], [METHOD, INPLACE_METHOD, FUNCTIONAL]),
        ('unbind', (2, 3, 4), lambda: [DIM_ARG], [FUNCTIONAL]),
        ('unsqueeze', (10, 20), lambda: [DIM_ARG], [METHOD, INPLACE_METHOD, FUNCTIONAL], 1),
        ('logcumsumexp', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('cumprod', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('cumsum', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('cummax', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('cummin', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('mean', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('median', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('nanmedian', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('mode', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('norm', (10, 20), lambda: [2, DIM_ARG], [METHOD, FUNCTIONAL]),
        ('prod', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('std', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('sum', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('var', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('kthvalue', (10, 20), lambda: [3, DIM_ARG], [METHOD, FUNCTIONAL]),
        ('max', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('min', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('sort', (10, 20), lambda: [DIM_ARG], [METHOD, FUNCTIONAL]),
        ('topk', (10, 20), lambda: [5, DIM_ARG], [METHOD, FUNCTIONAL]),
        ('renorm', (10, 20), lambda: [2, DIM_ARG, 1], [METHOD, INPLACE_METHOD, FUNCTIONAL]),
        ('index_add', (10, 10), lambda: [DIM_ARG, idx_tensor((10,), 10), torch.randn(10, 10)], [INPLACE_METHOD]),
        ('index_copy', (10, 10), lambda: [DIM_ARG, idx_tensor((10,), 10), torch.randn(10, 10)], [INPLACE_METHOD]),
        ('index_fill', (10, 10), lambda: [DIM_ARG, idx_tensor((10,), 10), 12], [INPLACE_METHOD]),
        ('scatter', (10, 10), lambda: [DIM_ARG, idx_tensor((10, 10), 10), torch.randn(10, 10)], [INPLACE_METHOD]),
        ('select', (10, 20), lambda: [DIM_ARG, 3], [METHOD]),
        ('unfold', (10, 20), lambda: [DIM_ARG, 5, 2], [METHOD]),
    ]

    for decl in neg_dim_tests:
        if len(decl) == 4:
            name, tensor_arg, arg_constr, types = decl
            extra_dim = 0
        elif len(decl) == 5:
            name, tensor_arg, arg_constr, types, extra_dim = decl

        test_name = 'test_' + name + '_neg_dim'

        if hasattr(TestTorch, test_name):
            raise AssertionError("Duplicated test name: " + test_name)
        setattr(TestTorch, test_name, make_neg_dim_test(name, tensor_arg, arg_constr, types, extra_dim))

# TODO: these empty classes are temporarily instantiated for XLA compatibility
#   once XLA updates their test suite it should be removed
class TestViewOps(TestCase):
    pass

class TestTensorDeviceOps(TestCase):
    pass

# Generates tests
# Note: test generation must be done at file scope, not within main, or
# pytest will fail.
add_neg_dim_tests()
instantiate_device_type_tests(TestViewOps, globals())
instantiate_device_type_tests(TestTensorDeviceOps, globals())
instantiate_device_type_tests(TestTorchDeviceType, globals())
instantiate_device_type_tests(TestDevicePrecision, globals(), except_for='cpu')

if __name__ == '__main__':
    TestCase._default_dtype_check_enabled = True
    run_tests()