GT4Py v1.1.12
Summary of changes since v1.1.11
Cartesian
- Improve DaCe loop scheduling for CPU and CUDA backends.
- Generate cleaner index-bound expressions in the DaCe backend.
- Use descriptive loop names in the DaCe schedule tree for easier profiling and debugging.
- Fix out-of-bounds reads for stencils using K-axis interfaces without extra vertical padding.
All changes
- fix[next]: extend fingerprinting to cover all relevant cache state via structural hashing by @egparedes in #2648
- ci: bump typer version in scripts to fix daily CI errors by @egparedes in #2671
- ci: move daily job into the European night by @havogt in #2672
- fix[next]: scalar dtype in math builtins + implement embedded gamma (#2277) by @havogt in #2658
- fix[next]: Enable the translation cache in compiled backends by @edopao in #2623
- fix[next-dace]: Attempt to achieve deterministic dace auto optimize by @edopao in #2635
- fix[next]: Fix SyntaxWarnings from . in regexes by @msimberg in #2674
- ci: fix daily CI test selection bug by @egparedes in #2676
- fix[next]: Add build configuration to the compilation fingerprint by @edopao in #2650
- ci: update
actions/checkoutv6 -> v7 by @romanc in #2678 - refactor[next]: Split compilation into a actual compilation and loading of the compilation artifact by @havogt in #2587
- fix[next]: symbolic domain union for empty domains (static case) by @tehrengruber in #2677
- feat[cartesian]: DaCe optimal for/map schedule by @FlorianDeconinck in #2628
- test[next]: Fix value type in symbol_mapping (flacky issue in dace tests) by @edopao in #2687
- fix[next-dace]: Set cmake argument for local cuda architecture by @edopao in #2685
- fix[next]: Fix StaticArgs injection by @SF-N in #2659
- feat[next]: add
recursiveoption to CollapseTuple pass by @tehrengruber in #2689 - test[next]: Fix fvm_nabla_setup for daily CI by @edopao in #2693
- refactor[cartesian]: descriptive loop names in dace backend by @romanc in #2683
- perf[cartesian]: avoid
x + 0and0 + xexpression in index bounds by @romanc in #2682 - fix[cartesian]: actually freeze origin & domain in K-axis by @romanc in #2684
- refactor[next]: drop cartesian subscript syntax by @tehrengruber in #2695
- build: Add agent skill to bump gt4py version by @edopao in #2697
- feat[next]: Compilation with process pool by @havogt in #2679
- fix[next]: make OTF build caches crash-consistent by @havogt in #2691
- fix[next]: propagate recursion limit to compilation worker processes by @havogt in #2698
- refactor[next-dace]: Minor code refactor in dace compilation by @edopao in #2699
- test: mark dace backend with
requires_daceby @romanc in #2696 - refactor[cartesian]: update treeir representation of the vertical loop by @romanc in #2688
- tests[cartesian]: add parsing test for ddims size with numpy int type by @romanc in #2703
- fix[next]: deterministic symbol renaming in inline_lambdas by @havogt in #2701
- fix[next]: empty range is absorbing for domain intersection by @havogt in #2702
- refactor: add missing type hint for
externalsby @romanc in #2707 - refactor[cartesian]: use f-strings instead of
.format()by @romanc in #2706 - fix[next-dace]: Ignore
guidin SDFG fingerprint by @edopao in #2704 - feat[next]: Tracer support part 1: tree_map by @SF-N in #2586
- Releasing v1.1.12 by @edopao in #2700
New Contributors
Full Changelog: v1.1.11...v1.1.12