OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

fix norrow_copy correctness issue for non-contiguous input for cpu path(reland) (#91883)

Browse Source 
This PR is about re-land https://github.com/pytorch/pytorch/pull/91789.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/91883
Approved by: https://github.com/lezcano
This commit is contained in:
XiaobingSuper 2023-01-10 05:35:53 +00:00 committed by PyTorch MergeBot
parent d1cc64b2ac
commit 1892c75a45
3 changed files with 10 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
aten/src/ATen/native/TensorShape.cpp
View File

@ -1217,7 +1217,9 @@ Tensor narrow_copy_dense(const Tensor& self, int64_t dim, int64_t start, int64_t
// Should just use narrow_copy_out, but this API is used internally at Meta: // Should just use narrow_copy_out, but this API is used internally at Meta:
// https://github.com/pytorch/pytorch/pull/87045#issuecomment-1309353561 // https://github.com/pytorch/pytorch/pull/87045#issuecomment-1309353561
Tensor narrow_copy_dense_cpu(const Tensor& self, int64_t dim, int64_t start, int64_t length){ Tensor narrow_copy_dense_cpu(const Tensor& self, int64_t dim, int64_t start, int64_t length){
auto output = at::empty_like(self); // narrow_copy_dense_cpu_out always resize output's size, so there only create
// a zero size tensor.
auto output = at::empty({0}, self.options());
return narrow_copy_dense_cpu_out(self, dim, start, length, output); return narrow_copy_dense_cpu_out(self, dim, start, length, output);
} }

6
test/functorch/test_vmap.py
View File

@ -3542,7 +3542,6 @@ class TestVmapOperatorsOpInfo(TestCase):
xfail('bitwise_left_shift', device_type='cpu'), xfail('bitwise_left_shift', device_type='cpu'),
decorate('bitwise_right_shift', device_type='cpu', decorate('bitwise_right_shift', device_type='cpu',
decorator=expectedFailureIf(not (IS_MACOS and IS_X86))), decorator=expectedFailureIf(not (IS_MACOS and IS_X86))),
xfail('narrow_copy', device_type='cpu'),
# UBSAN: runtime error: shift exponent -1 is negative # UBSAN: runtime error: shift exponent -1 is negative
decorate('bitwise_left_shift', decorator=unittest.skipIf(TEST_WITH_UBSAN, "Fails with above error")), decorate('bitwise_left_shift', decorator=unittest.skipIf(TEST_WITH_UBSAN, "Fails with above error")),
@ -3721,11 +3720,6 @@ class TestVmapOperatorsOpInfo(TestCase):
xfail('le'), xfail('le'),
xfail('lt'), xfail('lt'),
xfail('ne'), xfail('ne'),
# AssertionError
# Mismatched elements: 18 / 20 (90.0%)
# Greatest absolute difference: 14.031710147857666 at index (0, 5) (up to 0.0001 allowed)
# Greatest relative difference: 2.9177700113052603 at index (0, 3) (up to 0.0001 allowed)
xfail('narrow_copy', device_type='cpu'),
# UBSAN: runtime error: 1.27043e+262 is outside the range of representable values of type 'float' # UBSAN: runtime error: 1.27043e+262 is outside the range of representable values of type 'float'
decorate('special.zeta', decorator=unittest.skipIf(TEST_WITH_UBSAN, "Fails with above error")), decorate('special.zeta', decorator=unittest.skipIf(TEST_WITH_UBSAN, "Fails with above error")),
# RuntimeError: Expected all tensors to be on the same device, # RuntimeError: Expected all tensors to be on the same device,

7
test/test_torch.py
View File

@ -2971,6 +2971,13 @@ else:
sz[d] = 0 sz[d] = 0
self.assertEqual(sz, y.size()) 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: move to indexing test suite # FIXME: move to indexing test suite
@parametrize("reduce", ['prod', 'amin', 'amax', 'mean']) @parametrize("reduce", ['prod', 'amin', 'amax', 'mean'])
@dtypes(*all_types_and(torch.half, torch.bfloat16)) @dtypes(*all_types_and(torch.half, torch.bfloat16))