numba-dpex
numba-dpex copied to clipboard
IntelPython
Data Parallel Extension for Numba
By AOT I meant also being able to further distribute pre-compiled code in python packages (see [relevant `numba` section](https://numba.readthedocs.io/en/stable/user/pycc.html#compiling-code-ahead-of-time)). Even better if it could be distributed without a dependency to...
The [numba-dpex implementation of the PCA algorithm](https://github.com/adarshyoga/dpbench/blob/main/dpbench/benchmarks/pca/pca_numba_dpex_n.py) has several calls that are currently not supported - dpnp.mean(axis=0), dpnp.linalg.eigh, etc., inside a dpjit decorated function. These functions need to supported to...
```python import dpnp import numba_dpex import numba import numba.cpython.unsafe.tuple as UT @numba_dpex.dpjit def njfiller(i1, i2): return i1 + i1 @numba_dpex.dpjit def do_fill(A, sz, starts): for i in numba.pndindex(sz): arg =...
```python import dpnp import numba_dpex import numba import numba.cpython.unsafe.tuple as UT @numba_dpex.dpjit def do_fill(A, sz, starts): for i in numba.pndindex(sz): arg = sz # Construct the global index...yes I know...
#1236 introduced kernel ref caching at dpex runtime. Currently it is not getting cleaned. We need to provide design how to clean it. Old implementation is using LRU cache.
It prevents from casting fp32 types inside prange loop in dpbench benchmarks for integrated gpu
The following steps need to be completed before we can publish `numba-dppy` on conda-forge - [x] Upstream patches currently inside IDP Numba: - [x] https://github.com/IntelPython/numba/issues/180 - [x] https://github.com/IntelPython/numba/issues/179 - [x]...
https://github.com/IntelPython/numba-dpex/issues/159 was closed as completed after analogue to `np.random` functions were added to `dpnp` but it does not addresses some use cases that are met with `numba.cuda.random`: - generating `float32`...
Currently, for every kernel Numba-dpex generates two functions: a wrapper `spir_kernel` and a `spir_func`. All the kernel code is inside the `spir_kernel`. Enabling debug metadata only adds DWARF symbols to...
At the end of the page https://intelpython.github.io/numba-dppy/latest/user_guides/debugging/local_variables.html, the script ends before the function review() is called: ``` @dppy.func def revive(x): return x @dppy.kernel(debug=True) def data_parallel_sum(a, b, c): i = dppy.get_global_id(0)...
