Quantinuum/hugr - CodSpeed

hugr

Blog Docs Changelog

Performance History

Latest Results

Remove default impl of with_scope_internal

acl/pass-scope-breaking

15 hours ago

Remove default impl of with_scope_internal

acl/pass-scope-breaking

15 hours ago

release-plz-2026-03-09T10-18-35Z

16 hours ago

feat(llvm)!: Upgrade to LLVM 21 (#2901) BREAKING CHANGE: This PR upgrades the Inkwell LLVM version flag to `llvm21-1` from `llvm14-0` (tested with LLVM version 21.1.8). The primary breakage between these versions is the removal of typed pointers from LLVM IR. It also updates the hugr MSRV to 1.91 to avoid [mysterious](https://github.com/Quantinuum/hugr/actions/runs/22420867639/job/64917996447) [errors](https://github.com/Quantinuum/hugr/actions/runs/22461963744/job/65058345857) when linking the new llvm_sys with older toolchains. ### Details - Builder functions to create loads and GEPs now need a separate type argument (they used to infer the type from the pointer). - The `CallableValue` trait was removed from Inkwell. Invocations of function pointers are now constructed using a separate `build_indirect_call` method which takes the function signature as a separate argument. - Pointer-to-pointer casts are now superfluous and have been removed. - The current code often uses typed pointers as the representative types for collections and aggregate types (e.g., `StaticArrayCodegen::static_array_type` returns the type "pointer to static_array\<T\>"). These pointers no longer carry type information, but are still used to determine the size in memory of a reference to these types. When type information is needed we acquire it elsewhere. - Similarly, the `TypingSession` currently maps HUGR function types to function pointers. After the upgrade, it now _maps all function types to the opaque pointer type_. This is unintuitive, but seems to work currently, because the output of the type mapping is only used to determine size in memory of a reference to function types (so the pointer type is correct). If this changes in future, we may need to introduce separate APIs to query the "storage type" and "underlying type" of a function. - After the upgrade, **one** JITed test case returns an incorrect value when called using `ExecutionEngine::run_function`. The same IR returns the correct value if called using a raw function pointer returned by `ExecutionEngine::get_function`. For that reason, I have changed `TestContext::exec_hugr` to use the raw pointer method for all JITed tests. This also requires being more specific (and accurate) about the return types of some JITed test functions. ### Tip for reviewers <img width="684" height="371" alt="Screenshot 2026-02-27 at 12 20 32 PM" src="https://github.com/user-attachments/assets/de98d090-5234-4b39-ab48-31579de53aac" /> Closes #2886 --------- Co-authored-by: Jake Arkinstall <65358059+jake-arkinstall@users.noreply.github.com>

main

16 hours ago

Remove nightly lints

george/llvmup

18 hours ago

Merge branch 'main' into george/llvmup... and apply/ignore some new lints

george/llvmup

19 hours ago

Update link to hugr webpage

ab/move-schema-and-exts

20 hours ago

release-plz-2026-03-09T10-18-35Z

20 hours ago

Active Branches

feat!: Update remainder of passes to use PassScope, drop default with_scope

last run

15 hours ago

0%

last run

16 hours ago

0%

docs: Move `spec/schema` and `spec/std_extensions` to `resources/`

last run

20 hours ago

0%

© 2026 CodSpeed Technology

Home Terms Privacy Docs