Skip to content

v0.20.0

Compare
Choose a tag to compare
@github-actions github-actions released this 06 Jun 12:30
· 583 commits to refs/heads/main since this release
a83b98b

xDSL framework

Dialects

Transformations

  • transformations: Handle imperfect loop nest when lowering memref_stream.generic to loops @superlopuh (#2678)
  • transformations: add memref-stream-unnest-out-parameters @superlopuh (#2677)
  • transformations: linalg.generic to loops factor out some helpers @superlopuh (#2662)
  • transformations: linalg.generic to loops uses InsertPoint @superlopuh (#2661)
  • transformations: scf-for-loop-folding handle iter args @superlopuh (#2657)
  • transforms: Add function-constant-pinning pass to specialize functions @AntonLydike (#2389)
  • transformations: fold scf loops instead of riscv_scf @superlopuh (#2656)
  • transformations: Stencil shape inference fixes @PapyChacal (#2595)
  • transformations: CSE: Refactor for reuse-friendliness @PapyChacal (#2642)
  • transformations: Implement generic CSE. @PapyChacal (#2631)
  • dialects: arith.constant is Pure (no side effects) @PapyChacal (#2632)
  • transformations: fix memref_streamify non idempotency and wrong maps for outs @superlopuh (#2589)
  • transformations: (riscv) handle f64 lowering in int to float conversion @superlopuh (#2561)
  • transformations: (convert-snrt-to-riscv) Add lowerings for even more info ops @AntonLydike (#2428)
  • transformations: implement stencil-tensorize-z-dimension @n-io (#2516)
  • transforms: (convert-snrt-to-riscv) Add smaller number of ops @AntonLydike (#2559)
  • transformations: Fix mlir opt printing to stderr @AntonLydike (#2553)
  • transformations: stencil conversion fixes and improvements @PapyChacal (#2526)
  • transformation: stencil shape inference fixes @PapyChacal (#2522)
  • transformations: handle scalar arguments to generic ops when lowering @superlopuh (#2543)
  • transformations: Implement and test memref-to-gpu. @PapyChacal (#2542)
  • transformations: (riscv_scf) add a pass to fuse perfectly nested loops @superlopuh (#2540)
  • transformations: add malloc to riscv function call lowering @superlopuh (#2532)
  • dialects: Implement helpers on stencil attributes and simplify shape inference. @PapyChacal (#2517)
  • transformations: add onnx.Sub to linalg lowering @superlopuh (#2534)
  • transformations: Switch MPI lowering to use llvm.mlir.zero @PapyChacal (#2528)

Backend

  • backend: (csl) Added printing for param, comptime_struct and some builtins @dk949 (#2684)
  • backend: (csl) Maths and memory operation printing @dk949 (#2676)
  • backend: (csl) printing for functions, tasks and layout related things @dk949 (#2647)
  • backend: (riscv) reduce register pressure for stream configuration @superlopuh (#2621)
  • backend: (riscv) Don't make unreserved registers available @superlopuh (#2620)
  • backend: (riscv) Allocate ub and step before loop, and lb after @superlopuh (#2619)
  • backend: (csl) Add printing some scf and memref features @dk949 (#2610)
  • backend: (csl) Add printing for CSL dialect ops @AntonLydike (#2593)
  • backend: (riscv) flatten loops that don't use induction variables @superlopuh (#2585)
  • backend: (riscv) use fcvt.d.w for integer-valued float constants @superlopuh (#2583)
  • backend: (snitch) insert fmv ops when necessary when lowering streams @superlopuh (#2574)
  • backend: (riscv) move results of lowered riscv_scf.for loops @superlopuh (#2568)

Interpreter

🗎 Documentation

🐛 Bug Fixes

  • backend: (riscv) Don't make unreserved registers available @superlopuh (#2620)
  • backend: (riscv) Allocate ub and step before loop, and lb after @superlopuh (#2619)
  • core: Strip multi-digit suffixes in name_hint, not only single-digit. @PapyChacal (#2551)

Testing

Installation

Miscellaneous

Interactive

  • interactive: some minor typing fixes in xdsl-gui @superlopuh (#2649)
  • interactive: check for missing required fields before updating pipeline @superlopuh (#2557)