GT4Py v1.1.9

Summary of changes since v1.1.8

Cartesian

• Fix loop re-ordering in schedule tree.
• Fix scalarization of temporaries
• Redundant region syntax removed.

All changes

• fix[next]: Fix bug in literal values when using Enums by @egparedes in #2551
• ci: Fix daily CI job with explicit uv lock upgrades by @egparedes in #2552
• feat[cartesian]: Remove redundant region syntax by @twicki in #2554
• ci: Set Kubernetes memory request and limit to 64Gi on beverin by @havogt in #2556
• fix[cartesian]: loop reordering in schedule tree by @romanc in #2546
• fix[next-dace]: Suppress dace progress both in transformations and code generation by @edopao in #2557
• build: update frozen dependencies by @havogt in #2548
• fix[next]: Don't format non-visible Python code by @havogt in #2560
• fix[cartesian]: Scalarization of temporaries by @romanc in #2558
• build: add extra cuda13, rocm7; remove cuda11, rocm4,5 by @havogt in #2561
• Releasing v1.1.9 by @havogt in #2562

Full Changelog: v1.1.8...v1.1.9

GT4Py v1.1.8

Summary of changes since v1.1.7

Cartesian

• Fix default compile flags for for various compilers.
• Generate numpy >= 2.0 compatible code in the debug backend.
• Support for data dims of size one in dace backends.

All changes

• fix|build[cartesian]: Fix NVCC default flags for -O0 and refactor configuration out of config.py by @FlorianDeconinck in #2524
• feat[next]: support Enums as constant namespaces for value inlining by @egparedes in #2515
• ci: update github actions dependencies by @romanc in #2525
• build: fix versioningit complaints in shallow git clones by @egparedes in #2518
• fix[cartesian]: GCC 12/13 cxx default compile flag fix by @FlorianDeconinck in #2528
• ci[cartesian]: restore OpenMP for macos on daily ci by @romanc in #2530
• fix[next]: fix metrics source key context handlers by @egparedes in #2533
• feat[cartesian]: Update debug backend to generate numpy >= 2.0 compatible code by @twicki in #2526
• fix[next]: Reduce type ignores in client code by @DropD in #2484
• docs[cartesian]: Update ADRs with recent development by @romanc in #2512
• refactor[cartesian]: Separate horizontal and vertical interval parsers by @twicki in #2510
• feat[cartesian] Add icpx to default compilers + distutils better imports by @FlorianDeconinck in #2542
• fix[cartesian]: In the debug backend: Itemize lower dimensional fields also on the LHS for numpy >= 2.0 by @twicki in #2543
• feat[next-dace]: Disable dace trace for SDFG transformation progress by @edopao in #2540
• feat[next]: extend and refactor node fingerprinting utils by @egparedes in #2535
• fix[next]: properly ignore ndarray embedded caches by @egparedes in #2536
• feat[next-dace]: Introduce backend option to enable horizontal unit stride by @edopao in #2539
• fix[eve]: lru_cache should not call eq when a key is provided by @tehrengruber in #2529
• feat[next]: Embedded domain construction from dimension comparison by @havogt in #2532
• feat[next-dace]: Less Verbose Warnings by @philip-paul-mueller in #2544
• fix[next-dace]: Add entry-point synchronization by @edopao in #2527
• fix[cartesian]: Support for data dims of size one in dace backends by @romanc in #2547
• feat[next]: warn in case python is run without -O for non-embedded by @havogt in #2538
• feat[next]: Expose lru_cache cache_clear functionality by @tehrengruber in #2549
• Releasing v1.1.8 by @havogt in #2550

Full Changelog: v1.1.7...v1.1.8

GT4Py v1.1.7

Summary of changes since v1.1.6

Cartesian

• Leverage unrolling of integer power calls, in dace backends, for exponents 1, 2, and 3.
• Faster SDFG construction in dace backends.
• Introduce default compiler flags.

All changes

• feat[cartesian]: leverage integer power in dace backend by @romanc in #2502
• test[cartesian]: attempt to stabilize test_ij_field_reset by @romanc in #2505
• ci: don't format uv.lock file by @romanc in #2513
• fix[next]: fix metrics collection when compiling the same program multiple times by @egparedes in #2504
• fix[next]: Fix where & concat_where with named collections by @tehrengruber in #2511
• perf[next-dace]: Enhance MoveDataflowIntoIfBody transformation by @iomaganaris in #2514
• perf[cartesian]: reduce SDFG construction time in dace backends by @romanc in #2519
• feat[cartesian]: Introduce default compiler flags by @FlorianDeconinck in #2520
• perf[next-dace]: Allow more fusion of ConditionalBlocks by @iomaganaris in #2517
• Releasing v1.1.7 by @edopao in #2523

Full Changelog: v1.1.6...v1.1.7

GT4Py v1.1.6

Summary of changes since v1.1.5

Cartesian

• Added a GT4PY_CARTESIAN_ENABLE_OPENMP environment variable to disable OpenMP, which allows to support compilers (e.g. apple-clang) not shipping with OpenMP by default.
• Fixed issue in the numpy backend related to K-query expressions, where an internal variable was shadowing any user-provided k_mask variable.

All changes

• fix[cartesian]: Disable OpenMP multithreading for DaCe backends by @FlorianDeconinck in #2491
• perf[next-dace]: Only write to global on scan last level by @edopao in #2497
• fix[next-dace]: Add debug information to dace build depending on config by @edopao in #2499
• fix[cartesian]: Protect k_mask user variable by renaming internal mask by @FlorianDeconinck in #2501
• Releasing v1.1.6 by @edopao in #2500

Full Changelog: v1.1.5...v1.1.6

GT4Py v1.1.5

Summary of changes since v1.1.4

Cartesian

• Switch to DaCe main development branch, which is going to be released as DaCe v2.x.
• Changes in dace:X backends to reduce the size of SDFGs saved to disk.
• Fix casting of arguments of power function.

All changes

• feat[next]: Compile time domains by @SF-N in #2173
• build[cartesian]: update DaCe to get ScheduleTree based on mainline DaCe by @romanc in #2458
• refactor[next]: rename and simplify the programming language concept in the toolchain by @egparedes in #2478
• feat[next]: add atexit handler to dump the performance metrics at exit by @egparedes in #2481
• feat[cartesian]: save compressed SDFGs by @romanc in #2485
• fix[cartesian]: minified SDFG wihout duplication by @romanc in #2486
• fix[cartesian]: don't upcast arguments of power function by @romanc in #2489
• feat[next-dace]: Added gt_replace_concat_where_node() by @philip-paul-mueller in #2482
• feat[next]: add support for array_namespace allocation by @havogt in #2442
• feat[next-dace]: Support scan with single level output by @edopao in #2490
• fix[next]: Fix default value for use_max_domain_range_on_unstructured_shift by @tehrengruber in #2493
• fix[next-dace]: Emit warning in concat_where_mapper only in debug mode by @edopao in #2495
• feat[next]: Integrate jax.numpy in testing by @havogt in #2488
• Releasing v1.1.5 by @edopao in #2494

Full Changelog: v1.1.4...v1.1.5

GT4Py v1.1.4

Summary of changes since v1.1.3

Cartesian

• Fixed issue in program bindings by locking pybind11 to 2.x

All changes

• refactor[next]: Refactor metrics by @egparedes in #2450
• refactor[next-dace]: Use import alias for dace.nodes by @edopao in #2461
• feat[next]: Compiled variant for field operators by @tehrengruber in #2368
• feat[next-dace]: Updated MoveDataflowIntoIfBody by @philip-paul-mueller in #2460
• refactor[next-dace]: New Optimization Scheme in Intra-Map Optimization by @philip-paul-mueller in #2457
• build[cartesian]: Keep pybind11 to 2.x by @FlorianDeconinck in #2468
• build[next-dace]: Updated DaCe Dependency by @philip-paul-mueller in #2471
• fix[next-dace]: Fix Memory Layout for CPU by @philip-paul-mueller in #2459
• fix[next-dace]: Avoid adding instrumention to programs without GPU schedule by @iomaganaris in #2473
• docs: Update Slack join link by @havogt in #2470
• feat[next-dace]: Enable setting gpu_maxnreg attribute in maps by @iomaganaris in #2464
• fix[next]: Fix different static args after with_backend by @tehrengruber in #2475
• fix[next]: remove indeterminism in closure_var extraction by @havogt in #2476
• refactor[next]: Rename and refactor toolchain type definitions by @egparedes in #2474
• ci: Update Ubuntu version to 24.04 for beverin, disable MPS on santis by @havogt in #2455
• fix[next-dace]: Added a Check for Symbol Conflicts Upon relocation by @philip-paul-mueller in #2472
• fix[next]: Fix segfault for nanobind >=2.10 by @tehrengruber in #2431
• feat[next]: Add instrumentation package and user-defineable hooks by @egparedes in #2437
• perf[next-dace]: RemoveScalarCopies and FuseHorizontalConditionBlocks transformations by @iomaganaris in #2469
• build[next]: Update dace to version 2026.02.12 by @edopao in #2479
• Releasing v1.1.4 by @edopao in #2480

Full Changelog: v1.1.3...v1.1.4

GT4Py v1.1.3

Summary of changes since v1.1.2

Cartesian

• New features:
  • Support for K iterator access in numpy backend
  • New dace_KJI backend that operates on fields with Fortran memory layout
  • Automatic match for dace:X backends between layout and schedule, cache-optimal by default

All changes

• ci: fix daily CI task for python 3.14 by @egparedes in #2415
• fix[next][dace]: Support lowering of let-lambdas inside an iterator expression by @edopao in #2420
• fix[next][dace]: Remove isolated access nodes to unused lambda args by @edopao in #2418
• build[next]: Update dace version by @edopao in #2419
• fix[next][dace]: Remove isolated access nodes for unused args in let-lambda by @edopao in #2422
• fix[next]: Reuse parameters in direct fo calls by @SF-N in #2375
• fix[cartesian] Remove the vloop_sections for a more unqiue id(node) by @FlorianDeconinck in #2427
• ci: try node sharing by @havogt in #2135
• fix[next]: don't cse literal expressions (e.g. scan's init) by @havogt in #2421
• feat[cartesian]: Layout & Schedule pairing for dace:X by @FlorianDeconinck in #2426
• fix[next]: type checking with named collections in scans by @havogt in #2416
• fix[next]: Support named collections with multiple output domains by @havogt in #2428
• fix[next][dace]: Better usage of SubgraphContext during SDFG lowering by @edopao in #2413
• build[next]: Update dace version to 2026_01_12 by @edopao in #2432
• build[dace][next]: Added Custom Python Package for DaCe in GT4Py.Next by @philip-paul-mueller in #2423
• Add NVTX marker instrumentation by @iomaganaris in #2345
• feat[next][dace]: Enable async memory alloc on DaCe-HIP backend by @edopao in #2433
• feat[cartesian]: numpy backend support for K iterator access by @FlorianDeconinck in #2430
• fix[next][dace]: Make all node labels unique by @edopao in #2436
• feat[next]: Enable unrolling scan loops by @iomaganaris in #2434
• feat[dace][next] Updated gt_inline_nested_sdfg() by @philip-paul-mueller in #2385
• feat[dace][next]: Deterministic gt_split_access_nodes() by @philip-paul-mueller in #2383
• fix[next-dace]: Remove isolated node generated from splitting by @edopao in #2444
• ci[cartesian]: Disable maocOS test in daily CI by @edopao in #2438
• refactor[next]: Add config variable GT4PY_ADD_GPU_TRACE_MARKERS by @edopao in #2440
• fix[next-dace]: Addressed Memlet Caching Issue by @philip-paul-mueller in #2445
• fix[next]: Test cleanups by @havogt in #2441
• refactor[cartesian]: remove dead code from backend/base.py by @romanc in #2446
• fix[next-dace]: Update connectivities in fastcall by @edopao in #2449
• build[next-dace]: Update DaCe Version to 2026_01_21 by @edopao in #2451
• build[next-dace]: Use dace setting for compiler setting by @edopao in #2453
• Releasing v1.1.3 by @edopao in #2448

Full Changelog: v1.1.2...v1.1.3

GT4Py v1.1.2

Summary of changes since v1.1.1

General

• Added support for Python 3.14.

Cartesian

• New feature: Runtime Interval Bounds

All changes

• refactor[next][dace]: Simplify visitor of concat_where expressions by @edopao in #2394
• fix[next][dace]: Use explicit dataflow in let-lambdas by @edopao in #2396
• ci[next]: skip test_compile_variants_args_and_kwargs with dace backend and ROCM device by @edopao in #2400
• build[dace][next]: Updated DaCe Dependency by @philip-paul-mueller in #2401
• ci[next]: skip test_compile_variants_tuple with dace backend and ROCM device by @edopao in #2402
• feat[cartesian]: Runtime Interval Bounds by @twicki in #2395
• style[cartesian]: Cleanup in error messaging by @katrinafandrich in #2387
• build[dace][next]: Updated DaCe Dependency by @philip-paul-mueller in #2405
• bug[next]: Source location agnostic compilation hash by @tehrengruber in #2397
• fix[next][dace]: Use symbol mapping for stride propagation across nested SDFGs by @edopao in #2404
• ci/build/refactor[all]: python 3.14 support by @DropD in #2399
• ci[cartesian,next]: Upgrade scipy min version for py3.14 by @edopao in #2408
• refactor[eve,next]: UIDGenerator -> SequentialIDGenerator by @havogt in #2407
• fix[next]: apply cse in fuse_as_fieldop by @havogt in #2257
• refactor[next]: unroll_reduce deduce shift via typesystem by @havogt in #2267
• build: remove support for python releases breaking networkx by @egparedes in #2410
• fix[next][dace]: Unconditional execution of else-branch in if-statements by @edopao in #2412
• refactor[next][dace]: Move SDFG-lowering modules into subpackage by @edopao in #2409
• Releasing v1.1.2 by @havogt in #2414

New Contributors

• @katrinafandrich made their first contribution in #2387

Full Changelog: v1.1.1...v1.1.2

GT4Py v1.1.1

Summary of changes since v1.1.0

Cartesian

• Allow self-assignment with offset in K dimension in sequential vertical loops
• Bug fixes:
  • Skip implicit upcasting in (explicit) cast operations
  • Respect precision of Constants

All changes

• fix[cartesian]: upcast arguments of cast operations by @romanc in #2369
• docs[next][dace]: Updated CompiledDaceProgram by @philip-paul-mueller in #2374
• Fix warning in thread block size setting by @iomaganaris in #2371
• fix[next][dace]: Use correct dtype for connectivity tables by @edopao in #2373
• feat[dace][next]: Added DoubleWriteRemover Transformation by @philip-paul-mueller in #2370
• fix[cartesian]: respect percision of constants by @romanc in #2377
• refactor[next][dace]: Rename array symbols for shape and strides by @edopao in #2378
• fix[next][dace]: Generate temporary array ids in determistic way by @edopao in #2380
• fix[next][dace]: Avoid name conflict between tasklet connector and SDFG data by @edopao in #2382
• refactor[next][dace]: Introduce SDFGBuilder.add_nested_sdfg by @edopao in #2372
• perf[next]: Refactor of Program call arguments canonicalization by @egparedes in #2379
• fix[next][dace] Set properly strides of data in CPU by @iomaganaris in #2384
• feat[next]: adding custom named collections of scalars and fields by @egparedes in #2232
• fix[cartesian]: Allow self-assignment with offset in K dimension in sequential vertical loops by @romanc in #2388
• bug[next]: disallow implicit bool conversion of fields by @havogt in #2393
• fix[gt4py]: Transform enums in tuple static args by @SF-N in #2389
• Releasing v1.1.1 by @havogt in #2391

Full Changelog: v1.1.0...v1.1.1

GT4Py v1.1.0

Summary of changes since v1.0.10

Cartesian

• New experimental feature: 2D temporaries.
• Removed deprecated cuda backend.
• Bug fixes:
  • Absolute field-access in while-condition
  • Upcasting of expressions on the lhs of assignments

Versioning

• Added a fallback version of the form 1.0.10+unknown.version.details, when no version is available from git.

Next

See commit history.

All changes

• fix[dace][next]: Fixed DeadMapElimination by @philip-paul-mueller in #2340
• feat[dace][next]: Added MapToCopy by @philip-paul-mueller in #2311
• ci[cartesian]: increase time limit for cartesian / dace tests on gh200 by @romanc in #2335
• feat[cartesian]: 2D temporaries [Experimental] by @FlorianDeconinck in #2314
• refactor[cartesian]: Remove cuda backend by @romanc in #2337
• cartesian[fix]: Remove hard check on ADR existing for experimental feature by @FlorianDeconinck in #2343
• cartesian[fix]: Temporary annotation parsing by @FlorianDeconinck in #2347
• Improve setting of thread block size for 1D maps by @iomaganaris in #2344
• refactor[next][dace]: Refactoring tests of GTIR to SDFG lowering by @edopao in #2346
• feat[dace][next]: Map Splitter by @philip-paul-mueller in #2304
• build[dace][next]: Updated DaCe Dependency by @philip-paul-mueller in #2348
• feat[dace][next]: Modified How Degenerated 1D Maps Are Handled by @philip-paul-mueller in #2349
• perf[dace][next]: Updated Top Level Dataflow Optimization Stage by @philip-paul-mueller in #2284
• build: add script to set the default fallback version to latest release by @egparedes in #2341
• perf[next]: cache low level buffer information in fields by @egparedes in #2342
• refactor[next][dace]: Update lowering of domain by @edopao in #2350
• feat[dace][next]: Included TrivialTaskletelimination in gt_simplify() by @philip-paul-mueller in #2316
• feat[dace][next]: Block Inlining of NestedSDFG Containing Scans by @philip-paul-mueller in #2315
• build[dace][next]: Updated DaCe Dependency by @philip-paul-mueller in #2352
• refactor[next][dace]: Cleanup SDFG lowering of tuples by @edopao in #2354
• feat[dace][next]: Strides of Views by @philip-paul-mueller in #2334
• fix[cartesian]: Fix absolute field-access in while-conditionals by @twicki in #2357
• fix[next][dace]: More strict check on node sequencing in gt_create_local_double_buffering by @edopao in #2359
• fix[dace][next]: Fixed Setting Strides of Views by @philip-paul-mueller in #2358
• feat[next]: Multiple output domains by @SF-N in #2225
• feat[dace][next]: Using the canonicalize_memlet_trees() function by @philip-paul-mueller in #2351
• fix[cartesian]: do upcasting on both sides of assignments by @romanc in #2361
• test[cartesian]: absolute k access with index field and computation by @romanc in #2362
• perf[next][dace]: Remove temporary inside scan map scope by @edopao in #2355
• feat[dace][next] More Roboust Calling by @philip-paul-mueller in #2353
• bug[next]: fix timer precision loss by @havogt in #2364
• refactor[next]: Add check for GPU errors in debug mode by @edopao in #2365
• build[dace][next]: Updated DaCe Dependency by @philip-paul-mueller in #2367
• feat[next][dace]: Targeted transformation to remove copies in ICON4Pys vertically implicit solvers by @iomaganaris in #2329
• Releasing v1.1.0 by @havogt in #2366

Full Changelog: v1.0.10...v1.1.0