Set the IR's valid worlds to only the interpreter world by Technici4n · Pull Request #832 · chalk-lab/Mooncake.jl

Technici4n · 2025-11-01T20:22:37Z

Hopefully means that we don't need the optimize_OC workaround.

github-actions · 2025-11-01T20:27:10Z

Mooncake.jl documentation for PR #832 is available at:
https://chalk-lab.github.io/Mooncake.jl/previews/PR832/

codecov · 2025-11-01T20:30:16Z

Codecov Report

❌ Patch coverage is 75.00000% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/interpreter/forward_mode.jl	66.66%	1 Missing ⚠️
src/interpreter/ir_utils.jl	80.00%	1 Missing ⚠️
src/interpreter/reverse_mode.jl	75.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

github-actions · 2025-11-01T20:43:21Z

Performance Ratio:
Ratio of time to compute gradient and time to compute function.
Warning: results are very approximate! See here for more context.

┌────────────────────────────┬──────────┬──────────┬─────────────┬─────────┬─────────────┬────────┐
│                      Label │   Primal │ Mooncake │ MooncakeFwd │  Zygote │ ReverseDiff │ Enzyme │
│                     String │   String │   String │      String │  String │      String │ String │
├────────────────────────────┼──────────┼──────────┼─────────────┼─────────┼─────────────┼────────┤
│                   sum_1000 │ 100.0 ns │      1.8 │         1.9 │     1.2 │        5.51 │   8.21 │
│                  _sum_1000 │ 941.0 ns │     6.56 │       0.936 │  1520.0 │        34.7 │   1.08 │
│               sum_sin_1000 │  6.07 μs │     2.78 │        1.49 │    1.82 │        10.5 │   2.39 │
│              _sum_sin_1000 │  5.48 μs │     2.92 │         2.1 │   263.0 │        12.6 │   2.41 │
│                   kron_sum │ 298.0 μs │     43.9 │        2.92 │    5.44 │       226.0 │   9.41 │
│              kron_view_sum │ 323.0 μs │     43.0 │        3.55 │    11.4 │       227.0 │    6.8 │
│      naive_map_sin_cos_exp │  2.11 μs │     2.45 │         1.4 │ missing │        7.59 │   2.75 │
│            map_sin_cos_exp │  2.14 μs │     2.56 │        1.45 │    1.49 │        6.17 │   2.36 │
│      broadcast_sin_cos_exp │  2.28 μs │     2.34 │        1.38 │    2.31 │        1.46 │   2.28 │
│                 simple_mlp │ 408.0 μs │     4.65 │        3.02 │    1.63 │         6.6 │   3.29 │
│                     gp_lml │ 559.0 μs │     4.21 │        1.43 │    2.19 │     missing │   3.09 │
│ turing_broadcast_benchmark │  1.75 ms │     4.19 │         3.5 │ missing │        26.6 │   2.71 │
│         large_single_block │ 360.0 ns │     4.79 │        1.98 │  4420.0 │        34.3 │   2.36 │
└────────────────────────────┴──────────┴──────────┴─────────────┴─────────┴─────────────┴────────┘

Technici4n · 2025-11-03T10:38:38Z

Here are some 1.12 benchmark results: (with only the Mooncake backend)

┌────────────────────────────┬──────────┬──────────┬─────────────┬─────────┬─────────────┐
│                      Label │   Primal │ Mooncake │ MooncakeFwd │  Zygote │ ReverseDiff │
│                     String │   String │   String │      String │  String │      String │
├────────────────────────────┼──────────┼──────────┼─────────────┼─────────┼─────────────┤
│                   sum_1000 │   0.0 ns │      Inf │         Inf │ missing │     missing │
│                  _sum_1000 │ 600.0 ns │      5.0 │         1.0 │ missing │     missing │
│               sum_sin_1000 │   3.2 μs │     3.19 │        1.44 │ missing │     missing │
│              _sum_sin_1000 │   3.2 μs │      3.0 │        2.22 │ missing │     missing │
│                   kron_sum │ 712.0 μs │     18.7 │        2.15 │ missing │     missing │
│              kron_view_sum │ 647.0 μs │     20.9 │        2.58 │ missing │     missing │
│      naive_map_sin_cos_exp │   1.1 μs │     2.73 │        1.55 │ missing │     missing │
│            map_sin_cos_exp │   1.1 μs │     2.91 │        1.64 │ missing │     missing │
│      broadcast_sin_cos_exp │   1.1 μs │      3.0 │        1.64 │ missing │     missing │
│                 simple_mlp │ 232.0 μs │     5.64 │         2.5 │ missing │     missing │
│ turing_broadcast_benchmark │ 900.0 μs │     5.46 │        4.42 │ missing │     missing │
│         large_single_block │ 200.0 ns │      5.5 │         2.5 │ missing │     missing │
└────────────────────────────┴──────────┴──────────┴─────────────┴─────────┴─────────────┘

So this looks reasonable

mhauru · 2025-11-06T11:12:34Z

Just ran the same benchmarks as above, but on both this branch and the base branch of copilot/fix-521d0edf-b6f3-4f8f-810a-9b67505eccca on the same machine (my laptop). This branch:

┌────────────────────────────┬──────────┬──────────┬─────────────┐
│                      Label │   Primal │ Mooncake │ MooncakeFwd │
│                     String │   String │   String │      String │
├────────────────────────────┼──────────┼──────────┼─────────────┤
│                   sum_1000 │  83.0 ns │      2.0 │        2.01 │
│                  _sum_1000 │ 750.0 ns │     5.72 │        1.05 │
│               sum_sin_1000 │  3.29 μs │     3.43 │        2.01 │
│              _sum_sin_1000 │  3.38 μs │     3.25 │         2.0 │
│                   kron_sum │ 159.0 μs │     51.5 │        5.65 │
│              kron_view_sum │ 424.0 μs │     22.3 │        2.24 │
│      naive_map_sin_cos_exp │  2.71 μs │     1.34 │       0.862 │
│            map_sin_cos_exp │  2.71 μs │     1.46 │       0.862 │
│      broadcast_sin_cos_exp │  2.67 μs │     1.41 │       0.891 │
│                 simple_mlp │ 144.0 μs │     5.85 │        2.96 │
│ turing_broadcast_benchmark │ 973.0 μs │     7.14 │        2.39 │
│         large_single_block │ 209.0 ns │     8.97 │        3.19 │
└────────────────────────────┴──────────┴──────────┴─────────────┘

base:

┌────────────────────────────┬──────────┬──────────┬─────────────┐
│                      Label │   Primal │ Mooncake │ MooncakeFwd │
│                     String │   String │   String │      String │
├────────────────────────────┼──────────┼──────────┼─────────────┤
│                   sum_1000 │  42.0 ns │     4.95 │        3.95 │
│                  _sum_1000 │ 750.0 ns │     5.89 │        1.05 │
│               sum_sin_1000 │  3.33 μs │     3.41 │        1.99 │
│              _sum_sin_1000 │  3.38 μs │     3.19 │        1.98 │
│                   kron_sum │ 160.0 μs │     51.4 │         5.6 │
│              kron_view_sum │ 424.0 μs │     22.2 │        2.15 │
│      naive_map_sin_cos_exp │  2.71 μs │     1.34 │       0.846 │
│            map_sin_cos_exp │  2.67 μs │     1.47 │       0.875 │
│      broadcast_sin_cos_exp │  2.63 μs │     1.43 │        0.92 │
│                 simple_mlp │ 145.0 μs │      5.8 │        2.97 │
│ turing_broadcast_benchmark │ 973.0 μs │     7.11 │        2.39 │
│         large_single_block │   0.0 ns │      Inf │         Inf │
└────────────────────────────┴──────────┴──────────┴─────────────┘

Seems good to me.

mhauru

Looks all good as far as I can see.

* Initial plan * Add Julia v1.12 compatibility for BBCode and IR infrastructure Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Add 'pre' version to CI workflow Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Fix inlining_policy function name for Julia v1.12+ compatibility and format code Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * CompatHelper: bump compat for JET in [weakdeps] to 0.10, (keep existing compat) (#691) * CompatHelper: bump compat for JET in [weakdeps] to 0.10, (keep existing compat) * Set tangent_type for AbstractLogger to NoTangent (#710) * Initial plan * Set tangent_type for Logging.ConsoleLogger to NoTangent Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Apply suggestions from code review Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Add ConsoleLogger test to verify AbstractLogger tangent type coverage Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Add Base.CoreLogging.SimpleLogger to type union Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update test/tangents.jl Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Add better error messages for tangent accessors when types are passed (#707) * Initial plan * Add better error messages for tangent accessors when types are passed Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: Hong Ge <hg344@cam.ac.uk> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Tidy up (#715) * Tidy up error-generating methods * Remove redundant tests * Remove more redundant test cases * Bump patch version * add rule for complex logdet (#664) * working complex logdet * bump version * proper getrf! tests * improve tests * move tests * generalize tests * formatting and version --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> Signed-off-by: Mateus Araújo <maltusan@gmail.com> Co-authored-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Delete .github/copilot-instructions.md (#722) Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Uncomment b_binv_test_case for PlanarLayer (#727) Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Add DiffEq integration test for Mooncake (#726) * Initial plan * Add DiffEq integration test with original example from issue Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Finalize DiffEq integration test with comprehensive structure Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Remove @info and @test_broken statements to reduce clutter Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * fix formatting (#728) * Fix DiffEq integration test by restricting to reverse mode only (#731) * Initial plan * Fix DiffEq integration test by testing reverse mode only Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * formatting --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> Co-authored-by: Hong Ge <hg344@cam.ac.uk> * Fix IR docs duplication and improve documentation structure (#718) * Initial plan * Update IR docs title and remove duplication, add comparison section Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Address PR feedback: update title and simplify Optics reference Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Restore IRCode field details that were incorrectly removed Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Add robust Julia version-dependent JET version handling to CI workflow (#733) * Fix changelog version (#693) * remove `const M` (#696) * remove const M * fix formatting * Update Project.toml Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update README.md (#697) Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * minimal print io output (#699) * minimal print * vers bump * remove prev tests * include show in derived * Update algorithmic_differentiation.md (#703) I noticed a typo. I likely won't look at this again for a week or so, so please feel free to merge for me. Signed-off-by: Will Tebbutt <3628294+willtebbutt@users.noreply.github.com> * Fix 660 (#705) * Fix + tests * Bump patch version * Use _stable_typeof directly * Comment on the use of _stable_typeof * Extend arrayify for complex numbers (#706) * Fix + tests * Bump patch version * Use _stable_typeof directly * Add arrayify tests * Widen permitted types * Bump patch version again * Formatting * Uncomment blas tests * Create contribution guidelines in copilot-instructions.md (#713) Added contribution guidelines for the automatic differentiation package. Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Set tangent_type for AbstractLogger to NoTangent (#710) * Initial plan * Set tangent_type for Logging.ConsoleLogger to NoTangent Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Apply suggestions from code review Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Add ConsoleLogger test to verify AbstractLogger tangent type coverage Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Add Base.CoreLogging.SimpleLogger to type union Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update test/tangents.jl Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Add better error messages for tangent accessors when types are passed (#707) * Initial plan * Add better error messages for tangent accessors when types are passed Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: Hong Ge <hg344@cam.ac.uk> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Tidy up (#715) * Tidy up error-generating methods * Remove redundant tests * Remove more redundant test cases * Bump patch version * add rule for complex logdet (#664) * working complex logdet * bump version * proper getrf! tests * improve tests * move tests * generalize tests * formatting and version --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> Signed-off-by: Mateus Araújo <maltusan@gmail.com> Co-authored-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Delete .github/copilot-instructions.md (#722) Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Uncomment b_binv_test_case for PlanarLayer (#727) Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Add DiffEq integration test for Mooncake (#726) * Initial plan * Add DiffEq integration test with original example from issue Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Finalize DiffEq integration test with comprehensive structure Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Remove @info and @test_broken statements to reduce clutter Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * fix formatting (#728) * Fix DiffEq integration test by restricting to reverse mode only (#731) * Initial plan * Fix DiffEq integration test by testing reverse mode only Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * formatting --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> Co-authored-by: Hong Ge <hg344@cam.ac.uk> * Fix IR docs duplication and improve documentation structure (#718) * Initial plan * Update IR docs title and remove duplication, add comparison section Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Address PR feedback: update title and simplify Optics reference Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Restore IRCode field details that were incorrectly removed Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Initial plan * Add Julia version-dependent JET version selection to CI workflow Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Improve JET version handling to support complex constraints and remove need for explicit version on Julia 1.12+ Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Apply suggestions from code review Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update CI.yml Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> Signed-off-by: Will Tebbutt <3628294+willtebbutt@users.noreply.github.com> Signed-off-by: Mateus Araújo <maltusan@gmail.com> Co-authored-by: Penelope Yong <penelopeysm@gmail.com> Co-authored-by: Hong Ge <3279477+yebai@users.noreply.github.com> Co-authored-by: Astitva Aggarwal <84859349+AstitvaAggarwal@users.noreply.github.com> Co-authored-by: Will Tebbutt <3628294+willtebbutt@users.noreply.github.com> Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com> Co-authored-by: Hong Ge <hg344@cam.ac.uk> Co-authored-by: Mateus Araújo <maltusan@gmail.com> --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> Signed-off-by: Mateus Araújo <maltusan@gmail.com> Signed-off-by: Will Tebbutt <3628294+willtebbutt@users.noreply.github.com> Co-authored-by: CompatHelper Julia <compathelper_noreply@julialang.org> Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> Co-authored-by: Hong Ge <hg344@cam.ac.uk> Co-authored-by: Will Tebbutt <3628294+willtebbutt@users.noreply.github.com> Co-authored-by: Mateus Araújo <maltusan@gmail.com> Co-authored-by: Penelope Yong <penelopeysm@gmail.com> Co-authored-by: Astitva Aggarwal <84859349+AstitvaAggarwal@users.noreply.github.com> * Prevent exporting names already declared as public This is disallowed by the language and results in an error. * Extend required interface CC.add_edges_impl! * Adapt to stackless inference change (using CC.Future etc) * Always return a boolean from `src_inlining_policy` * Fix relocation of a few Experimental functions to CC * Fix `_ir_abstract_constant_propagation` rename * Adjust to `ir_abstract_constant_propagation` changes * Support `invoke` CodeInstance arguments * Adjust to opaque macro behavior change * Define better slotnames for debugging * Fix opaque closure construction * Adjust to 1.12 opaque closure creation change * More fixes * Fix IRCode construction bug * Adjust to binding changes for primal type extraction * Rules for new intrinsics * Set appropriate world bounds on the `CodeInfo` * Optimize opaque closures * Avoid failures on <1.12 * Don't attempt to set bounds if no inferred code is available * More <1.12 fixes * Don't optimize if IR interp can't be performed * More fixes/tests * Revert use of `Base.allocations` It is not defined on <1.12, and the macro form seems fine. * Fix bad line info information * Make line info logic more robust, fix a few more things * [WIP] add rules for `svec` and `_svec_len` * Don't add _svec_len rule for < 1.12 * Try to avoid error in IRInterp * Allow type unstability for `compute_oc_signature` * Attempt to fix allocations * Still use the same Tuple type for < 1.12 * Unconditionally set ir.argtypes[1] to an accurate Tuple * Fix `primal_type` bug * Fix construction error for IRCode * Fix more issues * Discard unwanted change * Make version check static * Add some @zero_derivatives for logging (#815) * Mark Base.fixup_stdlib_path as having @zero_derivative This comes up because in Julia 1.12 logmsg_code gets the folder of Core.Compiler if needed to resolve paths, which results in a ccall with no known rule. See JuliaLang/julia#57274. * Make Base.CoreLogging.handle_message_nothrow as having @zero_derivative This comes up in logmsg_code, because in Julia v1.12 handle_message_nothrow got a try-catch structure, which creates an UpsilonNode that Mooncake can't handle. * Update avoiding_non_differentiable_code.jl Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * fix format * Fix handle_message_nothrow @zero_derivative rule * Improve a comment * Fix duplicate svec rule after merge * Remove JET version adjustment from CI workflow Removed JET version adjustment step based on Julia version from CI workflow. Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Fix typo in 1.12- * Also test buildkit and ext on 1.12? * Use an eval to avoid duplicating BBCode * Remove useless line (thanks Copilot) * Update get_primal_type for GlobalRef * Update isva comment * Require Julia 1.12.1 or newer * Fix stdin reference not working in 1.11 * Update patch_for_319 following Julia 1.12 changes * Binding partitions are lazily populated! * Set the IR's valid worlds to only the interpreter world (#832) * Try to set the IR valid worlds to only the interpreter world * Add explanation to set_valid_worlds! * Tweak comment * Update dispatch_doctor.jl Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Import _Utils from DispatchDoctor in tests Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * reformat * Use a single world for inference * Disable JET tests for Julia 1.12 temporarily. Julia 1.12 has a few Base functions that are not type-stable, causing false-positive JET test failures in Mooncake. Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * re-enable JET test for Julia 1.12 and format. * Mark compute_oc_signature as unstable Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Add documentation for resolve_unbound_globalrefs function Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> * Add unstable block for version-specific functions Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update allocation tests for Julia version compatibility Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update frule allocation tests for Julia version compatibility * refactor count_allocs * Update JET version to include 0.11 (#841) Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * fix allocation counts for primals (plus `_new_` frule) (#843) * Fix allocation counters for `sum` and `rand` primals This fixes: - `sum(abs2, randn(11))` on 1.12 - `rand(rng, Float64)` on 1.10 and 1.11 * Use `eval` instead * Fix for `_new_` (I think) * Just get rid of the varargs in count_allocs * Add a warning * Fix all the horrible edge cases * Final fixes (please don't let me down) * Format * Skip certain stability and performance tests for 1.12 (#844) * Refactor stability rules for Float types in Julia Updated stability and allocation rules for Float types based on Julia version. Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Remove Float16 cases from test cases Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update avoiding_non_differentiable_code.jl Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * reformat --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Clarify GlobalRef resolution docs Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Fix for TwicePrecision allocs (#845) * Add rules for LAPACK.lacpy! and BLAS.trsv! (#839) * Add rules for LAPACK.lacpy! * LAPACK.lacpy! is exclusive to Julia 1.11+ * Add rule for BLAS.trsv! * Isolate sensitive trsm! rule --------- Co-authored-by: Hong Ge <3279477+yebai@users.noreply.github.com> * re-enable F16, only F32 and F64 are type unstable on 1.12 Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> * Update Project.toml Co-authored-by: Markus Hauru <mhauru@turing.ac.uk> Signed-off-by: Bruno Ploumhans <13494793+Technici4n@users.noreply.github.com> --------- Signed-off-by: Hong Ge <3279477+yebai@users.noreply.github.com> Signed-off-by: Mateus Araújo <maltusan@gmail.com> Signed-off-by: Will Tebbutt <3628294+willtebbutt@users.noreply.github.com> Signed-off-by: Bruno Ploumhans <13494793+Technici4n@users.noreply.github.com> Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: yebai <3279477+yebai@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: CompatHelper Julia <compathelper_noreply@julialang.org> Co-authored-by: Hong Ge <hg344@cam.ac.uk> Co-authored-by: Will Tebbutt <3628294+willtebbutt@users.noreply.github.com> Co-authored-by: Mateus Araújo <maltusan@gmail.com> Co-authored-by: Penelope Yong <penelopeysm@gmail.com> Co-authored-by: Astitva Aggarwal <84859349+AstitvaAggarwal@users.noreply.github.com> Co-authored-by: serenity4 <cedric.bel@hotmail.fr> Co-authored-by: Markus Hauru <mhauru@turing.ac.uk> Co-authored-by: Bruno Ploumhans <13494793+Technici4n@users.noreply.github.com>

Try to set the IR valid worlds to only the interpreter world

ab6e0be

github-actions bot assigned Technici4n Nov 1, 2025

Add explanation to set_valid_worlds!

1d428c0

Technici4n changed the title ~~Try to set the IR valid worlds to only the interpreter world~~ Set the IR's valid worlds to only the interpreter world Nov 3, 2025

Tweak comment

c51fd5b

mhauru mentioned this pull request Nov 5, 2025

Use set_valid_worlds! instead of manual optimisation TuringLang/Libtask.jl#205

Merged

yebai requested a review from mhauru November 5, 2025 14:13

mhauru approved these changes Nov 6, 2025

View reviewed changes

yebai merged commit 47e383a into copilot/fix-521d0edf-b6f3-4f8f-810a-9b67505eccca Nov 6, 2025
99 of 109 checks passed

yebai deleted the tech/single-ir-world branch November 6, 2025 15:48

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Set the IR's valid worlds to only the interpreter world#832

Set the IR's valid worlds to only the interpreter world#832
yebai merged 3 commits intocopilot/fix-521d0edf-b6f3-4f8f-810a-9b67505ecccafrom
tech/single-ir-world

Technici4n commented Nov 1, 2025

Uh oh!

github-actions bot commented Nov 1, 2025

Uh oh!

codecov bot commented Nov 1, 2025

Uh oh!

github-actions bot commented Nov 1, 2025 •

edited

Loading

Uh oh!

Technici4n commented Nov 3, 2025 •

edited

Loading

Uh oh!

mhauru commented Nov 6, 2025

Uh oh!

mhauru left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

Technici4n commented Nov 1, 2025

Uh oh!

github-actions bot commented Nov 1, 2025

Uh oh!

codecov bot commented Nov 1, 2025

Codecov Report

Uh oh!

github-actions bot commented Nov 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Technici4n commented Nov 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mhauru commented Nov 6, 2025

Uh oh!

mhauru left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

github-actions bot commented Nov 1, 2025 •

edited

Loading

Technici4n commented Nov 3, 2025 •

edited

Loading