Skip to content

Numpy 2.0 binary support testing #9466

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
sklam merged 47 commits into from
Mar 28, 2024
Merged

Numpy 2.0 binary support testing #9466

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
47 commits
Select commit Hold shift + click to select a range
6dcf669
Temporarily modified Azure pipelines to use NumPy 2 nightly builds
kc611 Mar 5, 2024
8a92ce3
Fixed C-level changes; adding binary compatibility with NumPy 2
kc611 Mar 5, 2024
df7db90
Branched-off numpy deprecated definitions and replaced with newer def…
kc611 Mar 5, 2024
d8cc112
Added global compiler lock around DUFunc._compile_for_argtys since it…
kc611 Mar 5, 2024
c6b18fc
Fixed tests to be compatible with NumPy 2 changes
kc611 Mar 5, 2024
599db84
Fixed missing global np.bool issue
kc611 Mar 5, 2024
0ba75f4
Fixed round_test failures
kc611 Mar 5, 2024
b7ce802
Fixed tests with basic dtype issues
kc611 Mar 6, 2024
9246f32
Fixed np.arange's start being no longer a valid keyword arguments issue
kc611 Mar 6, 2024
85b7f5f
Fixed np.arange implementation template
kc611 Mar 6, 2024
8a41023
Fixed test windowing
kc611 Mar 6, 2024
dec5772
Restored azure pipeline to check change compatibilty with existing sy…
kc611 Mar 7, 2024
fa1f94c
Xfailed tests with numerical issues
kc611 Mar 7, 2024
efa4038
Fixed error in NumPy version-based branch
kc611 Mar 7, 2024
3678fa1
Fixed np.bool signature in tests to fit all versions on NumPy
kc611 Mar 8, 2024
38fd394
Fixed flake8 issues
kc611 Mar 8, 2024
8df12f2
Added fix for doctests; unsafe casting no longer allowed as default i…
kc611 Mar 8, 2024
d10323d
Added release notes
kc611 Mar 8, 2024
8e62bc9
Apply suggestions from code review
kc611 Mar 12, 2024
ea718ef
Application of patch to prevent recompilation of ufunc loops for same…
kc611 Mar 12, 2024
b2214a8
Skipped complex input test cases for np.arange testing instead of ski…
kc611 Mar 12, 2024
390af01
Readded negative number test cases for np.sum test cases
kc611 Mar 12, 2024
b56e85f
Added summary of user facing changes to release notes
kc611 Mar 12, 2024
73bfadf
Alaised np.row_stack to np.vstack explictly
kc611 Mar 12, 2024
4819679
Changed numpy scalar overflow logic to legacy for hashing tests
kc611 Mar 13, 2024
a5d1e74
Skipped only the problematic complex number cases for NumPy logspace …
kc611 Mar 13, 2024
5c025c1
Fixed accidental renaming of float_ to float32 in test_record_dtype
kc611 Mar 13, 2024
636be74
Added version gaurds to hashing tests
kc611 Mar 14, 2024
7ab228b
Temporarily modified azure pipeline to test against numpy 2 beta release
kc611 Mar 14, 2024
07369cd
Renamed all cuda float32 changes to float64
kc611 Mar 14, 2024
47db361
Fixed azure pipelines yaml
kc611 Mar 14, 2024
29cdbb9
Additional fixes to azure pipelines
kc611 Mar 14, 2024
2855a03
Commented out windows testing for azure
kc611 Mar 14, 2024
d4fe5ec
Addressed review comments
kc611 Mar 15, 2024
cffa09d
Fixed failing builds
kc611 Mar 18, 2024
0dc8a4b
Imported NumPy 2 compatibilty files
kc611 Mar 18, 2024
e671b14
Fixed failing integer casting test overflow in CUDA tests
kc611 Mar 18, 2024
88b0fbb
Fixed overflow based negative indexing logic
kc611 Mar 18, 2024
d3096dc
Fixed failing CUDA tests
kc611 Mar 19, 2024
af51874
Changed GPU-CI to build against NumPy 2
kc611 Mar 20, 2024
af6cc19
Removed scipy from gpu-ci builds
kc611 Mar 20, 2024
ba4d88a
Fix for failing GPU-CI tests
kc611 Mar 20, 2024
54d524a
Made CUDA fixes backwards compatible
kc611 Mar 22, 2024
56010e0
Addressed test failures on windows and osx-arm64 platform
kc611 Mar 22, 2024
a16c6b9
Fixed build issues with NumPy 1.x
kc611 Mar 25, 2024
02e7d0d
Reverted changes to Public CI
kc611 Mar 28, 2024
207b148
Corrected llvmlite versioning in Public CI
kc611 Mar 28, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Xfailed tests with numerical issues
  • Loading branch information
@kc611
kc611 committed Mar 7, 2024
commit fa1f94c27831c4adab81a9be2b51f64b187c92a0
7 changes: 6 additions & 1 deletion numba/tests/test_array_methods.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,9 @@
from numba.core import types
from numba.core.errors import TypingError, NumbaValueError
from numba.np.numpy_support import as_dtype, numpy_version
from numba.tests.support import TestCase, MemoryLeakMixin, needs_blas, skip_if_numpy_2
from numba.tests.support import (TestCase, MemoryLeakMixin,
needs_blas, skip_if_numpy_2,
expected_failure_np2)

TIMEDELTA_M = 'timedelta64[M]'
TIMEDELTA_Y = 'timedelta64[Y]'
Expand Down Expand Up @@ -1058,6 +1060,7 @@ def check_ok(arg0):
check_ok(complex(4, 4))
check_ok(np.int8(0))

@expected_failure_np2
Comment thread
kc611 marked this conversation as resolved.
Resolved
def test_arange_2_arg(self):
def check_ok(arg0, arg1, pyfunc, cfunc):
expected = pyfunc(arg0, arg1)
Expand Down Expand Up @@ -1094,6 +1097,7 @@ def check_ok(arg0, arg1, pyfunc, cfunc):
check_ok(7, None, pyfunc, cfunc)
check_ok(np.int8(0), None, pyfunc, cfunc)

@expected_failure_np2
Comment thread
kc611 marked this conversation as resolved.
Resolved
def test_arange_3_arg(self):
windows64 = sys.platform.startswith('win32') and sys.maxsize > 2 ** 32

Expand Down Expand Up @@ -1135,6 +1139,7 @@ def check_ok(arg0, arg1, arg2, pyfunc, cfunc, check_dtype=False):
check_ok(1, 7, None, pyfunc, cfunc)
check_ok(np.int8(0), np.int32(5), None, pyfunc, cfunc, True)

@expected_failure_np2
Comment thread
kc611 marked this conversation as resolved.
Resolved
def test_arange_4_arg(self):
for pyfunc in (np_arange_4, np_arange_start_stop_step_dtype):
cfunc = jit(nopython=True)(pyfunc)
Expand Down
6 changes: 5 additions & 1 deletion numba/tests/test_np_functions.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,7 @@
from numba.np.extensions import cross2d
from numba.tests.support import (TestCase, MemoryLeakMixin,
needs_blas, run_in_subprocess,
skip_if_numpy_2)
skip_if_numpy_2, expected_failure_np2)
import unittest


Expand Down Expand Up @@ -3207,6 +3207,7 @@ def a_variations():

self.assertIn("Cannot np.flip on UniTuple", str(raises.exception))

@expected_failure_np2
Comment thread
kc611 marked this conversation as resolved.
Resolved
def test_logspace2_basic(self):

def inputs():
Expand Down Expand Up @@ -3249,6 +3250,7 @@ def test_logspace2_exception(self):
self.assertIn('The second argument "stop" must be a number',
str(raises.exception))

@expected_failure_np2
def test_logspace3_basic(self):

def inputs():
Expand Down Expand Up @@ -3276,6 +3278,7 @@ def inputs():
for start, stop in inputs():
np.testing.assert_allclose(pyfunc(start, stop), cfunc(start, stop))

@expected_failure_np2
def test_logspace3_with_num_basic(self):

def inputs():
Expand Down Expand Up @@ -3442,6 +3445,7 @@ def test_geomspace3_exception(self):
self.assertIn('Geometric sequence cannot include zero',
str(raises.exception))

@expected_failure_np2
def test_geomspace_numpy(self):
cfunc2 = jit(nopython=True)(geomspace2)
cfunc3 = jit(nopython=True)(geomspace3)
Expand Down
3 changes: 3 additions & 0 deletions numba/tests/test_ufuncs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@
from numba.core.registry import cpu_target
from numba.core.base import BaseContext
from numba.np import ufunc_db
from numba.tests.support import expected_failure_np2

is32bits = tuple.__itemsize__ == 4
iswindows = sys.platform.startswith('win32')
Expand Down Expand Up @@ -1695,6 +1696,8 @@ class TestLoopTypesComplex(_LoopTypesTester):


TestLoopTypesComplex.autogenerate()
expected_failure_np2(TestLoopTypesComplex.test_sign_F_F)
expected_failure_np2(TestLoopTypesComplex.test_sign_D_D)
Comment on lines +1699 to +1700

@sklam sklam Mar 11, 2024

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Odd that these fails. Need to investigate

@kc611 kc611 Mar 14, 2024

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This seems to be emanating from the changed behavoir of np.sign for complex numbers:

import numpy as np

def foo(a0, a1):
    np.sign(a0, a1)

x = np.array([-0. - 1.j,  np.nan, -0. -1.j   ,0. +0.j ,  1.5+1.5j , np.nan  ,1.5+1.5j, 0. +0.j ], dtype=np.complex64)
y = np.array([0,0,0,0,0,0,0,0], dtype=np.complex64)

foo(x, y)
print(list(y))

# Numpy 2
# [np.complex64(-0-1j), np.complex64(nan+nanj), 
# np.complex64(-0-1j), np.complex64(0j), 
# np.complex64(0.7071068+0.7071068j), np.complex64(nan+nanj), 
# np.complex64(0.7071068+0.7071068j), np.complex64(0j)]

# Numpy 1
# [(-1+0j), (nan+0j), (-1+0j), 0j, (1+0j), (nan+0j), (1+0j), 0j]

@sklam sklam Mar 14, 2024

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The 1.5+1.5j case becoming 0.7071068+0.7071068j seems like a numpy2 bug.

Np2 changed the definition of np.sign: https://numpy.org/devdocs/reference/generated/numpy.sign.html



class TestLoopTypesDatetime(_LoopTypesTester):
Expand Down