numba-dpex
numba-dpex copied to clipboard
IntelPython
Data Parallel Extension for Numba
- [X] Have you provided a meaningful PR description? sklearn-numba-dpex provides an implementation of random number generator functions using numba-dpex. The feature is analogous to similar functionality provided by numba.cuda....
I encountered a segfault when running the dpjit reduction test case on ARC with numba-dpex 0.23.0rc1. The same tests fail with 0.21.1, but do not segfault. ```bash $ ONEAPI_DEVICE_SELECTOR=*:gpu NUMBA_CAPTURED_ERRORS=new_style...
Currently, `numba-dpex` provides no API reference in the documentation page. The lack of API documentation is a serious gap in terms of usability and for new users to explore the...
A meta issue to aggregate various numba-dpex documentation related issues in one place: - [x] #1085 - [x] Ensure no all code snippets in the documentation work and do not...
When two different kernels that call the same overloaded function, *e.g.*, `fetch_add` in the reproducer, are compiled, the extra compilation flags needed for llvm-spirv translation are only applied to the...
I'm trying to run simple dpjit code: ```python import dpnp as dnp import numpy as np import numba as nb from numba_dpex import dpjit @dpjit def _sum_nomask(res): tot = nb.float32(1.0)...
An implementation for the `static_getitem` function for `DpnpNdArray` is missing and that results in the lowering falling back to the generic `lower_getitem` that in turn resolves to the `getitem_arraynd_intp` call...
Do we support dpnp.hypot on GPUs? Is there instruction generated for hypot and correctly lowered to SPIR-V. These code was commented out in mathdecl.py: ``` # class Math_hypot(ConcreteTemplate): # key...
Numba-dpex only has end-to-end tests for kernels. There are no tests to validate the LLVM IR that is generated for a device with corresponding IR generated by dpcpp. There needs...
```python def get_base_slice_internal(sv, arr): t = UTx.build_full_slice3_tuple(arr.ndim) s = _base_offset(sv) e = np.ones(len_base_offset(sv), dtype=ramba_dist_dtype) st = np.ones(len_base_offset(sv), dtype=ramba_dist_dtype) for i, v in enumerate(_axis_map(sv)): if v >= 0: e[v] += (_size(sv)[i]-1)...
