Skip to content

BUG: Fix for npyv__trunc_s32_f32 (VXE)#23147

Merged
charris merged 2 commits intonumpy:maintenance/1.24.xfrom
charris:backport-23077
Feb 2, 2023
Merged

BUG: Fix for npyv__trunc_s32_f32 (VXE)#23147
charris merged 2 commits intonumpy:maintenance/1.24.xfrom
charris:backport-23077

Conversation

@charris
Copy link
Copy Markdown
Member

@charris charris commented Feb 2, 2023

Backport of #23077.

np.sin(), np.cos() are giving erroneous result for float32 (VXE).
This PR is fixing npyv__trunc_s32_f32(npyv_f32 a) to resolve the issue.

np.sin() reproduce (without fix)

>>> c = np.array( [-25.091976, 90.14286], dtype=np.float32)
>>> np.sin(c)
array([0.04075377, 0.5707917 ], dtype=float32)
>>> 

>>> c = np.array( [90.14286], dtype=np.float32)
>>> np.sin(c)
array([0.8210949], dtype=float32)
>>>

>>> c = np.array( [-25.091976,  90.14286,   46.39879], dtype=np.float32)
>>> np.sin(c)
array([ 0.04075377, -0.5707917 ,  0.6632113 ], dtype=float32)

After the fix:

>>> import numpy as np
>>> c = np.array( [-25.091976, 90.14286], dtype=np.float32)
>>> np.sin(c)
array([0.04075377, 0.8210949 ], dtype=float32)

>>> c = np.array( [90.14286], dtype=np.float32)
>>> np.sin(c)
array([0.8210949], dtype=float32)

>>> c = np.array( [-25.091976,  90.14286,   46.39879], dtype=np.float32)
>>> np.sin(c)
array([0.04075377, 0.8210949 , 0.6632113 ], dtype=float32)
>>>

Test failure without fix -

FAILED ../../core/tests/test_umath.py::TestAVXFloat32Transcendental::test_sincos_float32 - AssertionError: Arrays are not almost equal up to 2 ULP (max difference is 2.13071e+09 ULP)
FAILED ../../core/tests/test_umath.py::TestAVXFloat32Transcendental::test_strided_float32 - AssertionError: X and Y are not equal to 2 ULP (max is 6.79208e+06)
FAILED ../../core/tests/test_umath_accuracy.py::TestAccuracy::test_validate_fp16_transcendentals[cos] - AssertionError: Arrays are not almost equal up to 1 ULP (max difference is 30720 ULP)
FAILED ../../core/tests/test_umath_accuracy.py::TestAccuracy::test_validate_fp16_transcendentals[sin] - AssertionE

@pradghos @charris
BUG: Fix for npyv_s32 npyv__trunc_s32_f32 (VXE)
be50440 
np.sin(), np.cos() are giving erroneous result for float32 (VXE)
This PR is fixing `npyv_s32 npyv__trunc_s32_f32(npyv_f32 a)`
to resolve the issue.
@pradghos @charris
replace __builtin_s390_vflls with npyv_doublee as before
289d240
@charris charris added 00 - Bug 06 - Regression 08 - Backport Used to tag backport PRs component: SIMD Issues in SIMD (fast instruction sets) code or machinery labels Feb 2, 2023
@charris charris added this to the 1.24.2 release milestone Feb 2, 2023
@charris charris merged commit 62af62a into numpy:maintenance/1.24.x Feb 2, 2023
@charris charris deleted the backport-23077 branch February 2, 2023 23:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

00 - Bug 06 - Regression 08 - Backport Used to tag backport PRs component: SIMD Issues in SIMD (fast instruction sets) code or machinery

Projects

None yet

Milestone

1.24.2 release

Development

Successfully merging this pull request may close these issues.

2 participants

@charris @pradghos