Fix for NumPy 1.23 by takagi · Pull Request #6807 · cupy/cupy

takagi · 2022-06-24T01:48:59Z

toslunar · 2022-06-24T06:22:24Z

For CuPy v11, I'd follow NumPy 1.23's behavior by removing these lines

Lines 259 to 270 in 16b511f

    
           elif isinstance(slices, list): 
        
               slice_list = list(slices)  # copy list 
        
               for s in slice_list: 
        
                   if not isinstance(s, int): 
        
                       warnings.warn( 
        
                           'Using a non-tuple sequence for multidimensional indexing ' 
        
                           'is deprecated; use `arr[tuple(seq)]` instead of ' 
        
                           '`arr[seq]`. In the future this will be interpreted as an ' 
        
                           'array index, `arr[cupy.array(seq)]`, which will result ' 
        
                           'either in an error or a different result.', 
        
                           FutureWarning) 
        
                       break

takagi · 2022-06-24T06:43:05Z

Sure! (I completely had misunderstanding about that...)

takagi · 2022-06-27T02:16:38Z

/test mini

takagi · 2022-06-27T03:23:26Z

/test mini

…y 1.23

takagi · 2022-06-27T12:59:18Z

/test mini

takagi · 2022-06-27T13:01:39Z

tests/cupy_tests/core_tests/test_ndarray_adv_indexing.py

-    {'shape': (0,), 'indexes': True},
-    {'shape': (0,), 'indexes': (True,)},
-    {'shape': (0,), 'indexes': numpy.ones((), dtype=numpy.bool_)},
-    {'shape': (0,), 'indexes': numpy.zeros((), dtype=numpy.bool_)},


These tests were split from TestArrayAdvancedIndexingGetitemParametrized in #1503 for NumPy 1.13 support, so I've made them back there in this PR as numpy<1.13 is already dropped.

takagi · 2022-06-27T13:04:07Z

tests/cupy_tests/core_tests/test_ndarray_adv_indexing.py

-    {'shape': (0,), 'indexes': True, 'value': 1},
-    {'shape': (0,), 'indexes': (True,), 'value': 1},
-    {'shape': (0,), 'indexes': numpy.ones((), dtype=numpy.bool_), 'value': 1},
-    {'shape': (0,), 'indexes': numpy.zeros((), dtype=numpy.bool_), 'value': 1},


These tests were split from TestArrayAdvancedIndexingSetitemScalarValue in #1503 for NumPy 1.13 support, so I've made them back there in this PR as numpy<1.13 is already dropped.

takagi · 2022-06-27T13:05:52Z

cupyx/jit/_cuda_typerules.py

    ast.Sub: arithmetic.subtract,
    ast.Mult: arithmetic.multiply,
    ast.Pow: arithmetic.power,
-    ast.Div: _numpy_scalar_true_divide,


Seems that we don't need _numpy_scalar_true_divide anymore.

It seems so to me, too.

@asi1024 Do you have any comment?

I agree to use arithmetic.true_divide!

kmaehashi · 2022-06-27T16:02:50Z

@toslunar, do you think this can be in the RC release this Thursday? If needed I think it's worth delaying the release, as v11 is a goog chance to bump the API baseline version to v1.23.

toslunar

LGTM

I left some minor comments that seem optional to be fixed in this PR.

toslunar · 2022-06-27T22:32:39Z

cupy/_core/core.pyx

-                DeprecationWarning)
-            axis = 0
        else:
            # Don't mention the deprecated F-contiguous support


nit: This line can be removed, too.

Sure, thanks!

toslunar · 2022-06-27T22:52:42Z

cupy/_core/internal.pyx

    st = 1

+    # For some cases, the strides are all set to zero if the shape is
+    # zero-sized since NumPy 1.23.


Do you know whether NumPy intentionally kept the behavior for no copy reshape? Maybe it's good to report to NumPy.

No, I couldn't search the source enough last night. As you say, it's a kind of weird behavior. I'm going to see it again.

It seems the change in creation (zero strides) was introduced in numpy/numpy#21477 (ctors.c). numpy/numpy#21477 (review)

toslunar · 2022-06-27T22:56:06Z

cupyx/jit/_cuda_typerules.py

    ast.Sub: arithmetic.subtract,
    ast.Mult: arithmetic.multiply,
    ast.Pow: arithmetic.power,
-    ast.Div: _numpy_scalar_true_divide,


It seems so to me, too.

@asi1024 Do you have any comment?

toslunar · 2022-06-28T04:57:36Z

/test mini

toslunar · 2022-06-28T05:00:42Z

v10 fix will be @testing.with_requires('numpy<1.23').

Fix to support NumPy 1.23

5e9008f

takagi marked this pull request as ready for review June 27, 2022 04:09

emcastillo assigned toslunar Jun 27, 2022

emcastillo added cat:numpy-compat Follow the NumPy/SciPy spec changes prio:high to-be-backported Pull-requests to be backported to stable branch blocking Issue/pull-request is mandatory for the upcoming release labels Jun 27, 2022

Set all strides to zero when zero-sized array creation to follow NumP…

1ce34ab

…y 1.23

takagi commented Jun 27, 2022

View reviewed changes

toslunar previously approved these changes Jun 27, 2022

View reviewed changes

Remove unnecessary comment

53b4265

takagi mentioned this pull request Jun 28, 2022

MAINT: Reduce allocation size of empty (0 size) arrays to 1 byte numpy/numpy#21477

Merged

takagi dismissed toslunar’s stale review via 53b4265 June 28, 2022 01:52

kmaehashi mentioned this pull request Jun 28, 2022

Remove CUPY_CUDA_VERSION as much as possible #6810

Merged

toslunar enabled auto-merge June 28, 2022 04:57

toslunar added this to the v11.0.0rc1 milestone Jun 28, 2022

toslunar removed the to-be-backported Pull-requests to be backported to stable branch label Jun 28, 2022

toslunar approved these changes Jun 28, 2022

View reviewed changes

kmaehashi mentioned this pull request Jun 28, 2022

Support NumPy 1.23 #6764

Closed

8 tasks

toslunar merged commit e43e348 into cupy:master Jun 28, 2022

takagi deleted the numpy-1.23 branch June 28, 2022 09:16

takagi restored the numpy-1.23 branch June 28, 2022 11:29

takagi deleted the numpy-1.23 branch June 28, 2022 11:30

Uh oh!

Conversation

takagi commented Jun 24, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

toslunar commented Jun 24, 2022

Uh oh!

takagi commented Jun 24, 2022

Uh oh!

takagi commented Jun 27, 2022

Uh oh!

takagi commented Jun 27, 2022

Uh oh!

takagi commented Jun 27, 2022

Uh oh!

takagi Jun 27, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

takagi Jun 27, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kmaehashi commented Jun 27, 2022

Uh oh!

toslunar left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

takagi Jun 28, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

toslunar commented Jun 28, 2022

Uh oh!

toslunar commented Jun 28, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

takagi commented Jun 24, 2022 •

edited

Loading

takagi Jun 27, 2022 •

edited

Loading

takagi Jun 27, 2022 •

edited

Loading

takagi Jun 28, 2022 •

edited

Loading