v2.0.0-beta1

What's Changed

• [Fix for Big Fortran SDFGs] Fix/framecode allocation placement by @ThrudPrimrose in #2640
• Bump the python-dependencies group across 1 directory with 12 updates by @dependabot[bot] in #2644
• No-View Nested SDFGs by @tbennun in #1696
• Fix fp16 mixed operations by @eliasachermann in #2648
• InlineSDFG: name an inlined transient apart from code-node connectors by @ThrudPrimrose in #2650
• SymbolPropagation: keep nested SDFG mappings in step with connected shapes by @ThrudPrimrose in #2654
• Pass only connector names to the nested SDFG of a library node expansion by @ThrudPrimrose in #2656
• Fix non-trivial copies by @eliasachermann in #2660
• Fix nest_sdfg_subgraph for a second region defining the same symbol by @tbennun in #2657
• Pure GEMV and batched matmul accept operands one device reaches by @ThrudPrimrose in #2661
• Broadcast and Merge library nodes, reachable from NumPy by @ThrudPrimrose in #2515
• Fix aligned new and delete by @eliasachermann in #2659
• Add categories to all properties by @tbennun in #2623
• Format with ruff instead of yapf by @ThrudPrimrose in #2667
• Connect the outer containers in the sequential reduce expansion by @ThrudPrimrose in #2653
• Install CSCS GPU dependencies with uv sync by @ThrudPrimrose in #2668
• codegen: keep a symbolically-sized register array on the stack by @ThrudPrimrose in #2504
• Lint codebase with ruff by @tbennun in #2671
• Canonicalize negative-step ranges from backward loop regions in memlet propagation by @edopao in #2672
• Drop the deprecated Indices subset in favor of single-element Range by @ThrudPrimrose in #2651
• Fix rebase_descendants not adding symbols to deeper nested SDFGs by @edopao in #2674
• Fix alloc: make aligned new/delete helpers host-only by @edopao in #2675
• Fix ScalarFission, LoopToMap and AugAssignToWCR on conditionals, constant arrays and Python updates by @ThrudPrimrose in #2681
• Fix four frontend and codegen bugs by @ThrudPrimrose in #2676
• Map only the symbols a nested SDFG reads when integrating it by @ThrudPrimrose in #2662
• Serialize each branch of a ConditionalBlock once by @ThrudPrimrose in #2682
• Map fusion: check intermediates symbolic size by @edopao in #2684
• Make states() the recursive state list of every control flow region and drop all_states() by @ThrudPrimrose in #2652
• Library nodes for Fortran's ALL, ANY, COUNT, MINLOC, MAXLOC and NORM2 by @ThrudPrimrose in #2678
• Fix fp16 mixed pow on CPU by @eliasachermann in #2685
• CShift library node and numpy.roll by @ThrudPrimrose in #2683
• Three small core fixes: scalar-access validation, nest_state_subgraph connectors, HPTT complex calls by @ThrudPrimrose in #2687
• MapReduceFusion: initialize a reduction inside an enclosing map in that scope by @ThrudPrimrose in #2688
• Stage length-1 arrays read by control flow and relax powers without a SymPy round trip by @ThrudPrimrose in #2689
• MatMul: transpose either operand without materializing it by @ThrudPrimrose in #2694
• Dot: conjugate the first operand on request (BLAS ?dotc) by @ThrudPrimrose in #2692
• MapFusionVertical: an empty Memlet into a scope accesses no data by @ThrudPrimrose in #2697
• Map fusion: relocate a scope connector that carries several edges once by @ThrudPrimrose in #2695
• Require memlet tree roots to be access nodes by @ThrudPrimrose in #2655
• Keep the other subset of copies propagated by MapFission by @ThrudPrimrose in #2700
• Work-depth: key every control-flow block by its region and index by @ThrudPrimrose in #2701
• MapFusionVertical: check NestedSDFG producers of the intermediate by @ThrudPrimrose in #2698
• Map fusion: handle Maps that share a dynamic map range by @ThrudPrimrose in #2696
• Fix three code generation bugs: method calls on expressions, containers named only in code, GPU outputs read by interstate edges by @ThrudPrimrose in #2691
• AugAssignToWCR: an update reads the written element of the same array by @ThrudPrimrose in #2707
• StateFusion: check every unordered access, and treat SubsetUnion overlap as unknown by @ThrudPrimrose in #2704
• STree->SDFG: Full support by @tbennun in #2590
• MapReduceFusion: keep the output offset, match a Reduce writing a map exit, refuse unsafe matches by @ThrudPrimrose in #2710
• StateFusionExtended: order a second-state write after the first state's readers by @ThrudPrimrose in #2705
• UniqueLoopIterators: give every loop iterator a unique name by @ThrudPrimrose in #2690
• LoopLocalMemoryReduction, FullMapFusion and MoveLoopIntoMap fixes by @ThrudPrimrose in #2703
• CPU codegen, MapExpansion and nest_state_subgraph fixes by @ThrudPrimrose in #2709
• GPU schedule groups: kernel and thread schedules, plus Chiplet, ThreadBlockCluster and Warp by @ThrudPrimrose in #2702
• One meaning per modulo spelling, and numeric casts in symbolic expressions by @ThrudPrimrose in #2706
• Keep a symbol free when a state reads it before any assignment by @ThrudPrimrose in #2712
• Array elimination of scalars by @romanc in #2566
• Map specialized symbols into nested SDFGs by @ThrudPrimrose in #2713
• nest_state_subgraph without self-copies; remove_memlet_path without None connectors by @ThrudPrimrose in #2714
• Fewer redundant walks in symbol resolution, expansion, serialization and code generation by @ThrudPrimrose in #2715
• InterstateEdge.replace_dict keeps an assignment target replaced by an expression by @ThrudPrimrose in #2721
• ScalarToSymbolPromotion keeps a transient that nothing writes by @ThrudPrimrose in #2717
• AI-generated implementations for DaCe library nodes by @tbennun in #2557
• Generalize the MPI library into a comm library with NCCL/RCCL expansions by @tbennun in #2686
• Pin pyFV3 CI checkout before NDSL 2026.10.00 by @ThrudPrimrose in #2723
• Python floor division and round-tripping float printing by @ThrudPrimrose in #2722
• Version Tag Update for v2.0.0 Beta 1 by @ThrudPrimrose in #2740

Full Changelog: v2.0.0-alpha10...v2.0.0-beta1