0.23 Tagged!

[diptorupd]

Fixed

• Array alignment problem for stack arrays allocated for kernel arguments. (Add kernel launcher array allocation alignment #1357)
• Issue Numba-dpex generating incorrect results for sliding window matmul kernel #892, Minimal reproducer for incorrect code generated for multiple barriers followed by multiple conditional ops in a kernel #906 caused by incorrect code generation for indexing (Fix/private array #1377)
• Generation of KernelHasReturnValueError error inside KernelDispatcher. (Fix KernelHasReturnValueError inside KernelDispatcher. #1394)
• Issue Creating slices over dpctl.tensor.usm_ndarray objects inside a kernel cause failure #1390: broken support for slicing into dpctl.tensor.usm_ndarray in kernels (Fix/dpctl slices #1425)
• Support for Wheels package on Windows (fix windows wheels copy cmd #1430)
• Incorrect mangled name for kernel function arguments (Add the mangled_args property to kernel_api types. #1443)
• Remove artifacts from conda/wheel packages residing in root level (Remove MANIFEST.in #1450)
• GDB tests to work properly on Intel Max GPU (Required fixes to run test suite on Max GPU #1451)
• Improper wheels installation on unsupported platforms (update wheel tag with pep600 #1452)
• Ref-counting of Python object temporaries in unboxing code (Fixed ref-counting of Python object temporaries in unboxing code #1454)
• Segfault caused by using malloc to allocate NRT_MemInfo. Replaced with Numba's NRT alloc (Use nrt api to allocate meminfo object #1458)
• Incorrect package name in README.md (Set correct package name in README.md #1463)

Added

• A new overloaded dimensions attribute for all index-space id classes (Feature/overload dimensions attribute for indexers #1359)
• Support for AtomicRef creation using multi-dimensional arrays (Accept multidimentional atomic indexes #1367)
• Support for linearized indexing functions inside a JIT compiled kernel (Feature/add linear indexing overload #1368)
• Improved documentation: overview ([Documentation] Update to overview section of documentation #1341), kernel programming guide ([Documentation] Programming Model, Kernel Programming guide #1388), API docs (Improve docstring for kernel_api functions. #1414), configs options ([Documentation] config options #1415), comparison with SYCL API ([Documentation] Add a comparison between SYCL and numba-dpex. #1417)
• New PrivateArray class in kernel_api to replace dpex.private.array (Add PrivateArray kernel_api #1370, Fix/private array #1377)
• Support for libsycinterface::DPCTLKernelArgType enum for specifying type of kernel args instead of hard coding (Use enum values for kernel arg types if dpctl >= 0.17 #1382)
• New indexing unit tests for kernel_api simulator and JIT compiled modes (Extend indexing function tests to kernel simulator #1378)
• New unit tests to verify all kernel_api features usable inside device_func (Verify global_barrier, indexing, private array inside device_func #1391)
• A sycl::local_accessor-like API (kernel_api.LocalAccessor) for numba-dpex kernel (A sycl::local_accessor-like API for numba-dpex kernel #1331)
• Specialization support for device_func decorator (Feature/specialization device func #1398)
• Support for all kernel_api functions inside the numba_dpex.kernel decorator. (Feature/move experimental decorators to stable #1400)
• Support for dpnp 0.15 (Fix for dpex failure caused by addition of nin, nout and types #1434, Remove versioneer, organize dpnp monkey patch #1464)
• Improvements to pyproject.toml configs to build numba-dpex from source. (Improve pyproject.toml #1449)
• Load the SPV_INTEL_variable_length_array SPIR-V extension to supporting arrays in private address-space on Intel Max GPU. (Required fixes to run test suite on Max GPU #1451)

Changed

• Default inline threshold value set to 2 from None. (Feature/inline threshold #1385)
• Port parfor kernel templates to kernel_api (Port the parfor range kernel template to new API. #1416), (Feature/migrate parfor #1424)
• Use SPIRVKernelDispatcher for parfor kernel dispatch (Migrate parfor to SPIRVKernelDispatcher #1435, Clean up unused code in parfors #1448)
• All examples use the latest dpctl API (Update example to use latest dpctl API. #1431)
• Minimum required dpctl version is now 0.16.1
• Minimum required numba version is now 0.59.0 (Pin numba to 0.59 and use windows-2019 for build #1462)

Removed

• OpenCL-like kernel API functions (Code reorganization: Removes ocl kernel API, moves math, print into kernel_api_impl #1420)
• func decorator (replaced by device_func) (Feature/move experimental decorators to stable #1400)
• numba_dpex.experimental.kernel and numba_dpex.experimental.device_func (Feature/move experimental decorators to stable #1400)