[TomerStarkware]

What

Two Sierra types reported a TypeBuilder::layout that disagreed with the MLIR type TypeBuilder::build emits for them. Codegen bakes layout() into array strides, enum payload offsets and argument marshalling, while the generated code reads and writes memory according to the emitted LLVM type. The mismatch let valid programs access aggregate elements at the wrong offsets.

• Felt252DictEntry<T> is emitted as two pointers but its layout reported a felt252 followed by two pointers (48 bytes / align 16 instead of 16 / 8). An array or struct of entries then used the wrong stride and offsets.
• U96LimbsLessThanGuarantee<N> is emitted as 2 * N u96 limbs but was declared zero-sized with an empty layout. Boxing it allocated zero bytes in the arena and the store clobbered the previous allocation, which right after eval_circuit is the outputs array.

Regression guard

Both bugs are the same class: build and layout are maintained by hand and drifted apart. This adds find_type_layout_mismatches, which builds every type a program declares, lowers it to LLVM IR and compares LLVM's ABI size and alignment against layout(). A new test runs it over every Sierra snippet in the cairo repo's e2e libfunc test data, so any type a libfunc touches is checked, and new types are covered as soon as any snippet declares them.

Also factors the repeated LLVM target initialization into a shared ffi::initialize_llvm_targets helper.

Testing

• e2e_libfunc_type_layouts_match_llvm fails on the old layouts and passes with the fixes.
• Full dict suite, e2e compile test, and the circuit guarantee test pass.

🤖 Generated with Claude Code

---

This change is 

[orizi]

@orizi reviewed 6 files and all commit messages, and made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on TomerStarkware).

[github-actions]

Benchmarking results

Benchmark for program dict_insert

Open benchmarks

Command	Mean [s]	Min [s]	Max [s]	Relative
Cairo-vm (Rust, Cairo 1)	11.048 ± 0.052	10.964	11.138	5.99 ± 0.05
cairo-native (embedded AOT)	1.844 ± 0.014	1.819	1.862	1.00
cairo-native (embedded JIT using LLVM's ORC Engine)	1.872 ± 0.017	1.846	1.900	1.02 ± 0.01

Benchmark for program dict_snapshot

Open benchmarks

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
Cairo-vm (Rust, Cairo 1)	534.4 ± 8.3	523.8	546.8	1.00
cairo-native (embedded AOT)	1659.9 ± 11.1	1645.8	1681.7	3.11 ± 0.05
cairo-native (embedded JIT using LLVM's ORC Engine)	1691.4 ± 16.1	1671.5	1712.5	3.17 ± 0.06

Benchmark for program factorial_2M

Open benchmarks

Command	Mean [s]	Min [s]	Max [s]	Relative
Cairo-vm (Rust, Cairo 1)	4.930 ± 0.017	4.892	4.952	2.84 ± 0.04
cairo-native (embedded AOT)	1.734 ± 0.023	1.701	1.778	1.00
cairo-native (embedded JIT using LLVM's ORC Engine)	1.755 ± 0.011	1.741	1.780	1.01 ± 0.02

Benchmark for program fib_2M

Open benchmarks

Command	Mean [s]	Min [s]	Max [s]	Relative
Cairo-vm (Rust, Cairo 1)	4.842 ± 0.023	4.814	4.887	2.92 ± 0.02
cairo-native (embedded AOT)	1.658 ± 0.012	1.634	1.672	1.00
cairo-native (embedded JIT using LLVM's ORC Engine)	1.689 ± 0.019	1.665	1.724	1.02 ± 0.01

Benchmark for program linear_search

Open benchmarks

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
Cairo-vm (Rust, Cairo 1)	582.5 ± 9.0	570.2	596.7	1.00
cairo-native (embedded AOT)	1703.0 ± 11.6	1687.4	1721.7	2.92 ± 0.05
cairo-native (embedded JIT using LLVM's ORC Engine)	1737.5 ± 16.6	1718.8	1775.0	2.98 ± 0.05

Benchmark for program logistic_map

Open benchmarks

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
Cairo-vm (Rust, Cairo 1)	482.8 ± 7.6	469.6	493.9	1.00
cairo-native (embedded AOT)	1675.2 ± 12.0	1645.7	1688.3	3.47 ± 0.06
cairo-native (embedded JIT using LLVM's ORC Engine)	1697.6 ± 11.0	1676.7	1715.5	3.52 ± 0.06

[github-actions]

Benchmark results Main vs HEAD.

Base

Command	Mean [s]	Min [s]	Max [s]	Relative
base dict_insert.cairo (JIT)	1.784 ± 0.012	1.769	1.808	1.01 ± 0.01
base dict_insert.cairo (AOT)	1.765 ± 0.010	1.754	1.786	1.00

Head

Command	Mean [s]	Min [s]	Max [s]	Relative
head dict_insert.cairo (JIT)	1.866 ± 0.021	1.838	1.895	1.01 ± 0.02
head dict_insert.cairo (AOT)	1.856 ± 0.028	1.813	1.897	1.00

Base

Command	Mean [s]	Min [s]	Max [s]	Relative
base dict_snapshot.cairo (JIT)	1.627 ± 0.012	1.610	1.645	1.02 ± 0.01
base dict_snapshot.cairo (AOT)	1.595 ± 0.010	1.580	1.610	1.00

Head

Command	Mean [s]	Min [s]	Max [s]	Relative
head dict_snapshot.cairo (JIT)	1.720 ± 0.023	1.688	1.764	1.00
head dict_snapshot.cairo (AOT)	1.723 ± 0.026	1.692	1.773	1.00 ± 0.02

Base

Command	Mean [s]	Min [s]	Max [s]	Relative
base factorial_2M.cairo (JIT)	1.687 ± 0.010	1.666	1.695	1.01 ± 0.01
base factorial_2M.cairo (AOT)	1.670 ± 0.005	1.661	1.679	1.00

Head

Command	Mean [s]	Min [s]	Max [s]	Relative
head factorial_2M.cairo (JIT)	1.805 ± 0.022	1.781	1.839	1.03 ± 0.02
head factorial_2M.cairo (AOT)	1.758 ± 0.023	1.722	1.782	1.00

Base

Command	Mean [s]	Min [s]	Max [s]	Relative
base fib_2M.cairo (JIT)	1.629 ± 0.011	1.615	1.646	1.01 ± 0.01
base fib_2M.cairo (AOT)	1.609 ± 0.009	1.594	1.626	1.00

Head

Command	Mean [s]	Min [s]	Max [s]	Relative
head fib_2M.cairo (JIT)	1.730 ± 0.015	1.704	1.756	1.03 ± 0.02
head fib_2M.cairo (AOT)	1.680 ± 0.030	1.650	1.732	1.00

Base

Command	Mean [s]	Min [s]	Max [s]	Relative
base linear_search.cairo (JIT)	1.660 ± 0.007	1.643	1.668	1.02 ± 0.01
base linear_search.cairo (AOT)	1.633 ± 0.017	1.607	1.675	1.00

Head

Command	Mean [s]	Min [s]	Max [s]	Relative
head linear_search.cairo (JIT)	1.699 ± 0.031	1.656	1.760	1.01 ± 0.02
head linear_search.cairo (AOT)	1.688 ± 0.021	1.664	1.742	1.00

Base

Command	Mean [s]	Min [s]	Max [s]	Relative
base logistic_map.cairo (JIT)	1.626 ± 0.012	1.599	1.646	1.02 ± 0.01
base logistic_map.cairo (AOT)	1.601 ± 0.013	1.581	1.623	1.00

Head

Command	Mean [s]	Min [s]	Max [s]	Relative
head logistic_map.cairo (JIT)	1.680 ± 0.023	1.647	1.708	1.00 ± 0.02
head logistic_map.cairo (AOT)	1.680 ± 0.018	1.635	1.700	1.00