dagmap.yaml

meta:
  version: 1
style:
  tags:
    performance:
      border: red
tasks:
  - label: Track allocations in functions
    id: track_alloc
    description: |
      Maintain a list of allocations inside each function which can be used
      for freeing things on return, and also for debugging memory usage.

  - label: Catch exceptions
    id: catch_exceptions
    description: |
      Allow exceptions raised in nopython mode to be caught in nopython mode.
    depends_on:
      - track_alloc


  - label: New IR
    id: new_ir
    description: |
      New intermediate representation for Numba that is backed by a dictionary

  - label: New Type Matching DSL
    id: type_matching
    description: |
      Replace the current DSL for Numba types with something more expressive
      that can match type patterns

  - label: Declarative type signatures\nfor @overload/@overload_method
    id: declarative_overload
    description: |
      Replace the current DSL for Numba types with something more expressive\n
      that can match type patterns
    depends_on:
      - type_matching

  - label: Rewrite "old-style" implementations
    id: rewrite_old_impls
    description: |
      Rewrite implementations of functions that use the old extension API that
      separates typing from implementation, and often uses the LLVM builder
      unnecessarily.
    depends_on:
      - declarative_overload
      - improve_test_suite_tooling
      - faster_pr_testing

  - label: Unify and add more test suite tooling
    id: improve_test_suite_tooling
    description: |
      Add tools to help with common patterns in testing and unify the ones we
      have, there's no need for 12 spellings of "is this Python 3" Also decide
      on "what to test", do all types need testing if inputs are being
      "as_array"'d?

  - label: Pipeline pass formalisation
    id: pass_formalisation
    description: |
      Decide on a formal description of a compiler pass and create supporting
      code for it

  - label: Array expression fusion pass
    id: new_array_expr_fusion_pass
    description:
        From parfors extract out the array expression fusion pass
    depends_on:
      - parfors_clean_up
      - pass_formalisation

  - label: LICM Pass
    id: new_licm_pass
    description: |
      Create a LICM Pass
    depends_on:
      - parfors_clean_up
      - pass_formalisation

  - label: Clean up Parfors
    id: parfors_clean_up
    description: |
      General clean up and refactoring of parfors ahead of any additional work

  - label: Mode based pipeline
    id: mode_based_pipeline
    description: |
      Switch the jit decorator to use a mode based pipeline with
      `nopython=True` equivalent as default.

  - label: Remove object mode fallback
    id: remove_objmode_fallback
    description: |
      Remove the deprecated object mode fallback
    depends_on:
        - mode_based_pipeline

  - label: Switch to ORC JIT
    id: orc_jit
    description: |
      MCJIT has been deprecated for some time.  Need to switch to the newer
      ORC JIT class.

  - label: Performance analysis suite
    id: perform_analysis_suite
    description: |
      Meta task for all performance analysis related functionality
    depends_on:
      - line_profiling
      - assembly_analysis_tooling
      - vectorisation_analysis

  - label: Vectorisation analysis
    id: vectorisation_analysis
    description: |
      Obtain LLVMs vectorisation reports and present these in a user friendly
      manner

  - label: Line profiling
    id: line_profiling
    description: |
      Support collection of profiling statistics from compiled machine code
      and map back to lines of Python.
    depends_on:
      - orc_jit
      - assembly_analysis_tooling

  - label: Assembly analysis tooling
    id: assembly_analysis_tooling
    description: |
      Tie generated assembly back to python lines and annotate instruction
      quality
    depends_on:
      - capstone

  - label: Build capstone against llvmdev
    id: capstone
    description: |
      Build capstone against llvmdev and create conda packages/wheels

  - label: Increase JIT class method performance
    id: jit_class_method_performance
    description: |
      Increase the performance of jitclass methods
    depends_on:
      - llvm_ref_count_pruning
      - new_licm_pass

  - label: LLVM level ref count pruning
    id: llvm_ref_count_pruning
    description: |
      Add a LLVM compiler pass to prune refcounts across entire functions

  - label: JITted coverage information
    id: jitted_coverage_info
    description: |
      Work out how to leverage gcov support in LLVM to enable coverage
      information
    depends_on:
      - compiler_rt

  - label: LLVM compiler_rt support
    id: compiler_rt
    description: |
      Work out how to build compiler_rt into LLVM and how to use it in Numba

  - label: Switch to pytest
    id: pytest
    description: |
      Make it possible to use pytest as test runner for Numba

  - label: Option to run modified tests only
    id: run_new_tests
    description: |
      Use / make pytest plugin to detect all test files which are new /
      changed relative to a given branch, and run only those tests
    depends_on:
      - pytest

  - label: Option to run 1/N slice of tests
    id: run_test_slice
    description: |
      Use / make pytest plugin to run 1/N of enumerated tests.
    depends_on:
      - pytest

  - label: Faster PR testing
    id: faster_pr_testing
    description: |
      Make automated PR testing with public CI services give faster feedback.
    depends_on:
      - run_new_tests
      - run_test_slice