v1.0.0

cuTile v1.0.0

Diff since v0.3.2

Breaking changes:

• ct.cat((a, b), i) has been removed in favor of the Base spellings inside kernels: cat(a, b; dims=i), vcat/hcat, and array construction syntax [a; b] (#289).

Merged pull requests:

• Restore upstream safety conditions on loop-parallel stores (#262) (@AntonOresten)
• Fix soundness holes in the bounds/divisibility dataflow analyses (#265) (@AntonOresten)
• Update FFTW requirement to 1.10.0 (#272) (@dependabot[bot])
• Update CUDA requirement to 6.2.1 (#273) (@dependabot[bot])
• Update Microfloats requirement to 0.2.3 (#274) (@dependabot[bot])
• Update NVTX requirement to 1.0.3 (#275) (@dependabot[bot])
• Update CompilerCaching requirement from 0.3.1 to 0.3.1, 0.4 (#276) (@dependabot[bot])
• Fix analysis of reinterpreted integer values (#277) (@maleadt)
• Sync with cuTile Python (#278) (@maleadt)
• Resolve bounds-check markers and fold constant structured branches (#279) (@maleadt)
• Support positive StepRange TileArray views (#280) (@maleadt)
• Add eachtile for stepped tile windows (#281) (@maleadt)
• Support GatherScatterView through Julia views (#282) (@maleadt)
• Add view-based atomic reductions and ct.@atomic (#283) (@maleadt)
• Add documentation (#284) (@maleadt)
• Support for CUDA 13.4 (#287) (@maleadt)
• Keep Tile memory bounds semantics explicit (#288) (@maleadt)
• Support array construction syntax in kernels (#289) (@AntonOresten)
• Add multi-dim reduction support (#290) (@AntonOresten)
• Improve tile representation (#291) (@maleadt)
• Support large TileArrays with Int64 indexing (#292) (@maleadt)
• Add tileiras remarks to Tile IR reflection (#293) (@maleadt)
• Add explicit rounding modes for float conversions (#294) (@maleadt)
• Sync with cuTile Python (#296) (@maleadt)
• Add support for masked atomics (#297) (@maleadt)
• Unify Tile IR toolchain version handling (#298) (@maleadt)
• Add programmatic dependent launch support (#299) (@maleadt)
• Bump CUDA Compiler JLL, test different compilers. (#300) (@maleadt)
• Untrack CUDA main, set compat to v6.3 (#301) (@AntonOresten)

Closed issues:

• High-level atomic operations (#50)
• TileArray and the AbstractArray interface (#66)
• Support for CUDA 13.3 (#188)
• Deploy documentation to site instead of README (#260)
• hvcat, hvncat, vect for tile construction (#271)