[BUG] `cupynumeric.right_shift(1, 64)` produces inconsistent results on RTX 4090 and GTX 1650 GPUs

[apiqwe]

Software versions

Software versions

NVIDIA GeForce RTX 4090:

System info:
  Python      :  3.14.6 | packaged by conda-forge | (main, Aug 11 2026, 10:26:15) [GCC 14.4.0]
  Platform    :  Linux-6.17.0-29-generic-x86_64-with-glibc2.39
  GPU driver  :  595.71.05
  GPU devices :
    GPU 0 :  NVIDIA GeForce RTX 4090

 Error Loading numa symbols, please install libnuma to avoid any performance issues
Package versions:
  realm       :  (failed to detect)
  legion      :  legion-25.12.0-75-g7b770cff5-dirty (commit: 7b770cff5b95a516c3244cc3a0500ce3720967d9)
  legate      :  26.06.01
  cupynumeric :  26.06.01
  numpy       :  2.4.6
  scipy       :  1.16.3
  numba       :  (failed to detect)

Legate build configuration:
  build_type        :  Release
  use_openmp        :  True
  use_cuda          :  True
  networks          :  ucx
  conduit           :  
  configure_options :  --LEGATE_ARCH=arch-conda;--with-python;--with-cc=/tmp/conda-croot/legate/_build_env/bin/x86_64-conda-linux-gnu-cc;--with-cxx=/tmp/conda-croot/legate/_build_env/bin/x86_64-conda-linux-gnu-c++;--build-march=haswell;--cmake-generator=Ninja;--with-openmp;--with-cuda;--build-type=release;--with-ucx

Package details:
  legate      :  legate 26.6.1 (installer: conda, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/legate)
  cupynumeric :  cupynumeric 26.6.1 (installer: conda, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/cupynumeric)
  numpy       :  numpy 2.4.6 (installer: conda, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/numpy)
  scipy       :  scipy 1.16.3 (installer: conda, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/scipy)
  numba       :  (failed to detect)

Conda package details:
  prefix       :  /opt/conda/envs/cupynumeric
  cuda-version :  cuda-version-13.2-he2cc418_3 (https://conda.anaconda.org/conda-forge/noarch)
  legate       :  legate-26.06.01-cuda13_py314_ucx_gpu_g3249a596f_0 (https://conda.anaconda.org/legate/linux-64)
  cupynumeric  :  cupynumeric-26.06.01-cuda13_py314_gpu_g7d7836ce_0 (https://conda.anaconda.org/legate/linux-64)

NVIDIA GeForce GTX 1650:

System info:
  Python      :  3.14.6 | packaged by conda-forge | (main, Aug 11 2026, 10:26:15) [GCC 14.4.0]
  Platform    :  Linux-6.6.87.2-microsoft-standard-WSL2-x86_64-with-glibc2.39
  GPU driver  :  581.57
  GPU devices :
    GPU 0 :  NVIDIA GeForce GTX 1650

Package versions:
  realm       :  (failed to detect)
  legion      :  legion-25.12.0-75-g7b770cff5-dirty (commit: 7b770cff5b95a516c3244cc3a0500ce3720967d9)
  legate      :  26.06.01
  cupynumeric :  26.06.01
  numpy       :  2.4.6
  scipy       :  1.16.3
  numba       :  (failed to detect)

Legate build configuration:
  build_type        :  Release
  use_openmp        :  True
  use_cuda          :  False
  networks          :  ucx
  conduit           :
  configure_options :  --LEGATE_ARCH=arch-conda;--with-python;--with-cc=/tmp/conda-croot/legate/_build_env/bin/x86_64-conda-linux-gnu-cc;--with-cxx=/tmp/conda-croot/legate/_build_env/bin/x86_64-conda-linux-gnu-c++;--build-march=haswell;--cmake-generator=Ninja;--with-openmp;--with-cuda=0;--build-type=release;--with-ucx

Package details:
  legate      :  legate 26.6.1 (installer: conda, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/legate)
  cupynumeric :  cupynumeric 26.6.1 (installer: conda, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/cupynumeric)
  numpy       :  numpy 2.4.6 (installer: conda, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/numpy)
  scipy       :  scipy 1.16.3 (installer: pip, path: /opt/conda/envs/cupynumeric/lib/python3.14/site-packages/scipy)
  numba       :  (failed to detect)

Conda package details:
  prefix       :  /opt/conda/envs/cupynumeric
  cuda-version :  (failed to detect)
  legate       :  legate-26.06.01-py314_ucx_cpu_g3249a596f_0 (https://conda.anaconda.org/legate/linux-64)
  cupynumeric  :  cupynumeric-26.06.01-cuda13_py314_cpu_g7d7836ce_0 (https://conda.anaconda.org/legate/linux-64)

Jupyter notebook / Jupyter Lab version

No response

Expected behavior

cupynumeric.right_shift(1, 64) produces consistent results on RTX 4090 and GTX 1650 GPUs.

Observed behavior

The results of cupynumeric.right_shift(1, 64) are inconsistent between RTX 4090 and GTX 1650 GPUs.

Example code or instructions

import numpy as np
import cupynumeric as cp

print(np.right_shift(1, 64))
print(cp.right_shift(1, 64))

Stack traceback or browser console output

RTX 4090 Result:

0
0

GTX 1650 Result:

0
1