v0.4.1

What's Changed

• Preserve memref offsets in pointer casts by @tpn in #73
• Expose normalized launch metadata during kernel dispatch by @tpn in #70
• Resolve various CI issues by @brandon-b-miller in #146
• [Lowering] Add support for carray by @atmnp in #140
• [Lowering] Convert to correct integer/float type for power operator by @atmnp in #128
• ci: adopt CCCL check_result aggregator and add doc-only PR mode by @leofang in #142
• ci: disable sccache preprocessor cache mode by @leofang in #141
• Defer CUDA source compilation in linker by @isVoid in #101
• [CI] Strip LLVM .so files at install time by @leofang in #132
• Add full support for big dynamic smem by @ZzEeKkAa in #148
• Delete llvmlite code by @gmarkall in #107
• Support compile-time tuple slicing by @atmnp in #158
• Support namedtuple unpacking in MLIR lowering by @atmnp in #159
• [Debug Info] Fix bitwise logic ops result types by @jiel-nv in #161
• Search for conda supplied LLVM7 build by @brandon-b-miller in #157
• Fix comparators not working with unsigned int types by @acosmicflamingo in #131
• Prioritize MLIR unicode comparison typing by @atmnp in #168
• [CI] Pin numba < 0.66 until numba-cuda is patched by @atmnp in #175
• Add kernel lto to linker before everything else by @ZzEeKkAa in #174
• [Publish] Update workflow to also build/upload Windows wheels by @atmnp in #139
• Improve register-pressure launch diagnostics by @atmnp in #167
• Bump the actions-monthly group across 1 directory with 8 updates by @dependabot[bot] in #177
• Revert "[CI] Pin numba < 0.66 until numba-cuda is patched" by @leofang in #176
• Fix unreachable pyright by @ZzEeKkAa in #104
• Linker Re-Design by @atmnp in #155
• Add native complex32 type support by @atmnp in #166
• Update version to v0.4.1 by @atmnp in #180
• Fix comparisons through coerce_numpy_scalars_for_binary_op by @brandon-b-miller in #116
• Isolate bundled LLVM/MLIR from differently-versioned LLVM in-process by @leofang in #171
• Combined fixes (#204, #205, #179) for #155 by @Jlisowskyy in #206
• Fix Windows miscompilation of floating-point constant vectors by @ZzEeKkAa in #207

Full Changelog: v0.4.0...v0.4.1