Skip to content

v0.22.0
Compare
Choose a tag to compare
@github-actions github-actions released this 02 Sep 15:12
· 273 commits to refs/heads/main since this release
v0.22.0
8a70781
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Learn about vigilant mode.

xDSL framework

Dialects

Transformations

  • transformations: (stencil-tensorize-z-dim) Support loops @n-io (#3118)
  • transforms: Fix constant pinning breaking for external functions @AntonLydike (#3131)
  • bug: (csl-stencil-to-csl-wrapper) Reverse generated operands @n-io (#3112)
  • transformations: (csl-stencil-to-csl-wrapper) Generate symbol export and no memref.globals @n-io (#3101)
  • transformations: Translate memref to dsd @n-io (#3092)
  • transformations: Support safe inplace stencil bufferization. @PapyChacal (#3077)
  • transformations: Implement control-flow-hoist @PapyChacal (#3103)
  • transformations: (csl-stencil-to-csl-wrapper) Support reading properties from bufferized csl-stencil @n-io (#3083)
  • transformations: (csl-stencil-bufferize) Fold csl_stencil.access that have no effect @n-io (#3084)
  • transformations: stencil-bufferize side-effect analysis fix @PapyChacal (#3076)
  • dialects: stencil: Allow loading and storing from a same field. @PapyChacal (#3071)
  • transformations: Enable bufferized dmp.swap lowering. @PapyChacal (#3073)
  • transformations: Forgotten xdsl/transforms/shape_inference_patterns/init.py @PapyChacal (#3072)
  • transformations: (stencil-to-csl-stencil) Support also bufferized dmp.swap @n-io (#3068)
  • transformations: Enable dmp.swap stencil bufferization. @PapyChacal (#3066)
  • transformations: Tensorize new result of dmp.swap @n-io (#3063)
  • dialects: Update dmp.swap @PapyChacal (#3056)
  • transformations: Decouple distribute-stencil from shape-inference @PapyChacal (#3054)
  • dialects: dmp: make strategies attributes and carry them in IR @PapyChacal (#3050)
  • transformations: (csl-stencil-to-csl-wrapper) Translate func args to exportable memref.global @n-io (#3051)
  • transformations: update stencil lowering to properly accomodate bufferized stencils. @PapyChacal (#3016)
  • transformations: Implement shape-inference pass @PapyChacal (#3047)
  • transformations: stencil shape inference polish @PapyChacal (#3046)
  • transformations: stencil bufferization polish @PapyChacal (#3001)
  • transformations: Add linalg-to-csl pass @n-io (#3028)
  • transformations: (csl-stencil-bufferize) Inject iter_arg into linalg compute @n-io (#3033)
  • transformations: (csl_stencil) Add bufferization pass @n-io (#3004)
  • transformations: First implementation of stencil-bufferize @PapyChacal (#2983)
  • transformations: Enable bufferized stencil lowering. @PapyChacal (#2984)
  • transformations: (stencil-tensorize-z-dim) Scaling z-shifts to avoid negative offsets @n-io (#2975)
  • transform: (stencil-tensorize-z-dimension) Tensorize arith.constant directly @n-io (#2970)
  • transformations: lift arith to linalg named ops @n-io (#2963)
  • (transforms): dont have tensor inside memref @n-io (#2962)
  • (transform): csl_wrapper program module init @n-io (#2891)
  • (transform): csl_stencil canonicalization pass @n-io (#2814)
  • transformations: (memref_stream) add memref_stream.generic legalization pass @nazavode (#2929)

Backend

  • fix: Param names were not correctly referenced in the printer @dk949 (#3108)
  • backend: (riscv) print list of registers to regalloc stats @superlopuh (#3111)
  • backend: (JAX) add some syntax sugar for typed compiled functions @superlopuh (#3058)
  • backend: (JAX) add export to JAX executable @superlopuh (#3057)
  • backend: (riscv) fix register allocation stats @superlopuh (#2961)
  • backend: (riscv) add option to insert regalloc stats @superlopuh (#2958)
  • backend: (snitch) add an optimization-level option to linalg-snitch pass @superlopuh (#2956)
  • backend: (snitch) unify all the test passes as a single end-to-end pass @superlopuh (#2955)
  • dialects: (riscv_snitch) lower arith.addf on vector<4xf16> to vfadd.h @nazavode (#2934)

Interpreter

API

🗎 Documentation

🐛 Bug Fixes

  • core: Assembly format shouldn't require $*SegmentSizes @alexarice (#3123)
  • tests: Increase timeout on lit tests @alexarice (#3125)
  • bug: remove duplicate isattr definition @superlopuh (#3088)
  • transformations: stencil-bufferize side-effect analysis fix @PapyChacal (#3076)
  • dependencies: fix numpy version for subprojects that depend on it @superlopuh (#3062)
  • interpreter: fix int64 format in ptr dtypes @superlopuh (#3035)
  • bug: (snitch_stream) fix stride pattern repeat param parsing @superlopuh (#2986)
  • transformations: (stencil-tensorize-z-dim) Scaling z-shifts to avoid negative offsets @n-io (#2975)
  • dialects: (arith, core) Constant op to accept tensor and memref @n-io (#2969)
  • backend: (riscv) fix register allocation stats @superlopuh (#2961)
  • (bug): linalg named ops to include hidden regions in generic print @n-io (#2957)
  • (bug): fix csl_stencil canonicalise @n-io (#2949)

Testing

Continuous Integration

Installation

Miscellaneous

  • misc: remove type annotations from operand definitions [NFC] @superlopuh (#3120)
  • core: Assembly format shouldn't require $*SegmentSizes @alexarice (#3123)
  • core: (irdl) Add support for regions in optional groups to declarative assembly format @kimxworrall (#3106)
  • bug: (csl-stencil-to-csl-wrapper) Reverse generated operands @n-io (#3112)
  • transformations: (csl-stencil-to-csl-wrapper) Support reading properties from bufferized csl-stencil @n-io (#3083)
  • core: Implement speculability trait and helper. @PapyChacal (#3102)
  • transformations: (csl-stencil-bufferize) Fold csl_stencil.access that have no effect @n-io (#3084)
  • transformations: stencil-bufferize side-effect analysis fix @PapyChacal (#3076)
  • dialects: Update dialects with SameSize IRDL options. @PapyChacal (#3070)
  • transformations: Forgotten xdsl/transforms/shape_inference_patterns/init.py @PapyChacal (#3072)
  • dialects: Update tensor side-effect traits @n-io (#3069)
  • transformations: (stencil-to-csl-stencil) Support also bufferized dmp.swap @n-io (#3068)
  • transformations: Tensorize new result of dmp.swap @n-io (#3063)
  • installation: Update nix flake @alexarice (#3059)
  • rewriting: Make modify_block_argument_type modify_value_type @PapyChacal (#3045)
  • misc: Fix typo achorable -> anchorable @alexarice (#3048)
  • core: CanonicaliZation naming consistency @PapyChacal (#3040)
  • misc: remove mentions of black formatter @superlopuh (#3032)
  • misc: make some pyright ignores more specific [NFC] @superlopuh (#3029)
  • misc: Add Dialect.split_name method @alexarice (#3003)
  • misc: don't set pass spec arguments from instance when they are the default value @superlopuh (#3005)
  • transformations: Enable bufferized stencil lowering. @PapyChacal (#2984)
  • transformations: (stencil-tensorize-z-dim) Scaling z-shifts to avoid negative offsets @n-io (#2975)
  • transform: (stencil-tensorize-z-dimension) Tensorize arith.constant directly @n-io (#2970)
  • dialects: (tensor) Incorrect default 'dynamic index' on insert_slice op @n-io (#2974)
  • misc: don't import riscv and x86 in xdsl-opt @superlopuh (#2967)
  • misc: use specific print method in printer.py @superlopuh (#2964)
  • (transforms): dont have tensor inside memref @n-io (#2962)
  • misc: move IRDLDialectLoader to own file and remove circular dependency @superlopuh (#2946)
  • misc: use more isattr checks @superlopuh (#2952)
  • misc: make version lazy @superlopuh (#2950)
  • misc: remove some pylance errors throughout the project [NFC] @superlopuh (#2953)
  • (transform): csl_stencil canonicalization pass @n-io (#2814)
  • misc: move print_csl import away from top level @superlopuh (#2948)
  • misc: use constraints in more operation definitions @superlopuh (#2939)

Contributors

superlopuh, nazavode, and 15 other contributors
Assets 2
Loading