perf: reduce per-instruction overhead in VM execution hot paths

[orizi]

Reduces per-instruction overhead in the VM execution hot paths:

• Memory::get: skip the relocation round-trip and redundant 40-byte copy for values that cannot be affected by relocation rules (everything except pointers into temporary segments). Hits 3–4 reads per instruction.
• Memory::insert: compute segment indexes once instead of twice, and compare MemoryCells directly (ignoring the ACCESS flag) instead of materializing a MaybeRelocatable for the consistency check.
• Range-check validation: validate inline instead of via a boxed closure, removing a heap allocation per range-checked value.
• step_instruction: drop the per-step mem::take + resize of the instruction cache; copy the (Copy) Instruction out instead. Also keeps the cache intact if run_instruction errors.
• deduce_op1: take op0 by reference, avoiding a felt clone per deduced operand.
• update_registers: take operands by reference instead of by value (~168-byte move per step).
• BuiltinHintProcessor: skip hashing the full hint source on every hint execution when no extra hints are registered.
• mark_as_accessed: batch the four per-step address marks, resolving each segment once for consecutive same-segment addresses.

Measured on cairo_programs/benchmarks (best-of-5, --layout all_cairo --proof_mode): −10% to −32% wall clock on most benchmarks (big_fibonacci −32%, pedersen −28%, linear_search −20%, keccak_integration −18%).

🤖 Generated with Claude Code

Performance

Expected: removes per-instruction overhead on every VM step — 3–4 relocation round-trips + 40-byte copies per step in Memory::get, a heap alloc per range-checked value, a mem::take+resize per step in the instruction cache, a felt clone per deduced operand, a ~168-byte operands move per step, a full-source hash per hint execution, and redundant index computations in insert/mark_as_accessed. Applies to essentially every workload; biggest where steps are cheap (tight loops) so per-step overhead dominates.

Measured (best-of-7, wall clock vs main, local; noise ±2.4%):

benchmark	delta		benchmark	delta
poseidon_integration	−20.2%		keccak_integration	−14.5%
pedersen	−20.4%		big_factorial	−13.2%
linear_search	−20.7%		memory_integration	−13.2%
big_fibonacci	−16.3%		math_integration	−9.3%
integration_builtins	−14.0%		secp_integration	−7.7%
uint256_integration	−14.0%		dict_integration	−7.3%

Every benchmark improves, −7% to −21%. An independent earlier measurement on a different machine (4-core container, best-of-5, before the insert/mark_as_accessed commits were added) showed the same shape: big_fibonacci −32%, pedersen −28%, linear_search −20%, keccak −18%, dict −16%, big_factorial −13%, math −10%.

Methodology: cairo_programs/benchmarks/*.cairo compiled with cairo-lang 0.14 --proof_mode, run via cairo-vm-cli <prog>.json --layout all_cairo --proof_mode, hyperfine best-of-7 per binary, local x86-64 Linux. Noise floor (identical-code control pair): ±2.4%.

---

This change is 

[github-actions]

**Hyper Thereading Benchmark results**




hyperfine -r 2 -n "hyper_threading_main threads: 1" 'RAYON_NUM_THREADS=1 ./hyper_threading_main' -n "hyper_threading_pr threads: 1" 'RAYON_NUM_THREADS=1 ./hyper_threading_pr'
Benchmark 1: hyper_threading_main threads: 1
  Time (mean ± σ):     24.669 s ±  0.013 s    [User: 24.210 s, System: 0.456 s]
  Range (min … max):   24.660 s … 24.679 s    2 runs
 
Benchmark 2: hyper_threading_pr threads: 1
  Time (mean ± σ):     20.630 s ±  0.021 s    [User: 20.191 s, System: 0.436 s]
  Range (min … max):   20.616 s … 20.645 s    2 runs
 
Summary
  hyper_threading_pr threads: 1 ran
    1.20 ± 0.00 times faster than hyper_threading_main threads: 1




hyperfine -r 2 -n "hyper_threading_main threads: 2" 'RAYON_NUM_THREADS=2 ./hyper_threading_main' -n "hyper_threading_pr threads: 2" 'RAYON_NUM_THREADS=2 ./hyper_threading_pr'
Benchmark 1: hyper_threading_main threads: 2
  Time (mean ± σ):     13.204 s ±  0.037 s    [User: 24.266 s, System: 0.456 s]
  Range (min … max):   13.178 s … 13.230 s    2 runs
 
Benchmark 2: hyper_threading_pr threads: 2
  Time (mean ± σ):     10.890 s ±  0.011 s    [User: 20.471 s, System: 0.442 s]
  Range (min … max):   10.883 s … 10.898 s    2 runs
 
Summary
  hyper_threading_pr threads: 2 ran
    1.21 ± 0.00 times faster than hyper_threading_main threads: 2




hyperfine -r 2 -n "hyper_threading_main threads: 4" 'RAYON_NUM_THREADS=4 ./hyper_threading_main' -n "hyper_threading_pr threads: 4" 'RAYON_NUM_THREADS=4 ./hyper_threading_pr'
Benchmark 1: hyper_threading_main threads: 4
  Time (mean ± σ):      9.778 s ±  0.303 s    [User: 35.794 s, System: 0.586 s]
  Range (min … max):    9.563 s …  9.992 s    2 runs
 
Benchmark 2: hyper_threading_pr threads: 4
  Time (mean ± σ):      8.703 s ±  0.153 s    [User: 31.436 s, System: 0.581 s]
  Range (min … max):    8.594 s …  8.811 s    2 runs
 
Summary
  hyper_threading_pr threads: 4 ran
    1.12 ± 0.04 times faster than hyper_threading_main threads: 4




hyperfine -r 2 -n "hyper_threading_main threads: 6" 'RAYON_NUM_THREADS=6 ./hyper_threading_main' -n "hyper_threading_pr threads: 6" 'RAYON_NUM_THREADS=6 ./hyper_threading_pr'
Benchmark 1: hyper_threading_main threads: 6
  Time (mean ± σ):      9.804 s ±  0.030 s    [User: 35.865 s, System: 0.610 s]
  Range (min … max):    9.783 s …  9.825 s    2 runs
 
Benchmark 2: hyper_threading_pr threads: 6
  Time (mean ± σ):      8.436 s ±  0.050 s    [User: 31.544 s, System: 0.607 s]
  Range (min … max):    8.401 s …  8.471 s    2 runs
 
Summary
  hyper_threading_pr threads: 6 ran
    1.16 ± 0.01 times faster than hyper_threading_main threads: 6




hyperfine -r 2 -n "hyper_threading_main threads: 8" 'RAYON_NUM_THREADS=8 ./hyper_threading_main' -n "hyper_threading_pr threads: 8" 'RAYON_NUM_THREADS=8 ./hyper_threading_pr'
Benchmark 1: hyper_threading_main threads: 8
  Time (mean ± σ):      9.519 s ±  0.067 s    [User: 36.542 s, System: 0.619 s]
  Range (min … max):    9.472 s …  9.567 s    2 runs
 
Benchmark 2: hyper_threading_pr threads: 8
  Time (mean ± σ):      8.399 s ±  0.046 s    [User: 31.990 s, System: 0.648 s]
  Range (min … max):    8.367 s …  8.432 s    2 runs
 
Summary
  hyper_threading_pr threads: 8 ran
    1.13 ± 0.01 times faster than hyper_threading_main threads: 8




hyperfine -r 2 -n "hyper_threading_main threads: 16" 'RAYON_NUM_THREADS=16 ./hyper_threading_main' -n "hyper_threading_pr threads: 16" 'RAYON_NUM_THREADS=16 ./hyper_threading_pr'
Benchmark 1: hyper_threading_main threads: 16
  Time (mean ± σ):      9.731 s ±  0.224 s    [User: 36.684 s, System: 0.700 s]
  Range (min … max):    9.572 s …  9.889 s    2 runs
 
Benchmark 2: hyper_threading_pr threads: 16
  Time (mean ± σ):      8.602 s ±  0.216 s    [User: 32.120 s, System: 0.684 s]
  Range (min … max):    8.449 s …  8.754 s    2 runs
 
Summary
  hyper_threading_pr threads: 16 ran
    1.13 ± 0.04 times faster than hyper_threading_main threads: 16

[github-actions]

Benchmark Results for unmodified programs 🚀

Command	Mean [s]	Min [s]	Max [s]	Relative
base big_factorial	1.977 ± 0.039	1.929	2.066	1.07 ± 0.02
head big_factorial	1.843 ± 0.010	1.826	1.858	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base big_fibonacci	1.905 ± 0.022	1.878	1.946	1.08 ± 0.02
head big_fibonacci	1.765 ± 0.021	1.741	1.803	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base blake2s_integration_benchmark	6.243 ± 0.051	6.150	6.299	1.16 ± 0.04
head blake2s_integration_benchmark	5.373 ± 0.187	5.193	5.827	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base compare_arrays_200000	2.013 ± 0.027	1.980	2.060	1.08 ± 0.03
head compare_arrays_200000	1.862 ± 0.053	1.828	1.995	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base dict_integration_benchmark	1.303 ± 0.034	1.266	1.363	1.05 ± 0.04
head dict_integration_benchmark	1.241 ± 0.035	1.215	1.324	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base field_arithmetic_get_square_benchmark	980.2 ± 21.6	959.4	1036.5	1.12 ± 0.04
head field_arithmetic_get_square_benchmark	876.3 ± 20.4	859.8	920.8	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base integration_builtins	6.262 ± 0.035	6.203	6.327	1.16 ± 0.04
head integration_builtins	5.409 ± 0.203	5.283	5.952	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base keccak_integration_benchmark	6.485 ± 0.162	6.319	6.880	1.19 ± 0.03
head keccak_integration_benchmark	5.463 ± 0.073	5.394	5.595	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base linear_search	2.045 ± 0.067	1.985	2.216	1.09 ± 0.04
head linear_search	1.868 ± 0.041	1.833	1.970	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base math_cmp_and_pow_integration_benchmark	1.443 ± 0.024	1.422	1.504	1.04 ± 0.03
head math_cmp_and_pow_integration_benchmark	1.391 ± 0.028	1.364	1.455	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base math_integration_benchmark	1.358 ± 0.031	1.324	1.418	1.05 ± 0.03
head math_integration_benchmark	1.291 ± 0.022	1.269	1.347	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base memory_integration_benchmark	1.166 ± 0.021	1.142	1.199	1.05 ± 0.03
head memory_integration_benchmark	1.115 ± 0.019	1.090	1.140	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base operations_with_data_structures_benchmarks	1.364 ± 0.013	1.342	1.389	1.04 ± 0.02
head operations_with_data_structures_benchmarks	1.314 ± 0.028	1.292	1.362	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base pedersen	426.5 ± 17.1	412.3	457.2	1.08 ± 0.05
head pedersen	395.6 ± 7.3	389.7	412.9	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base poseidon_integration_benchmark	520.7 ± 5.2	515.9	533.9	1.00 ± 0.04
head poseidon_integration_benchmark	519.3 ± 20.0	506.9	562.6	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base secp_integration_benchmark	1.467 ± 0.055	1.401	1.576	1.09 ± 0.05
head secp_integration_benchmark	1.345 ± 0.029	1.322	1.413	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base set_integration_benchmark	522.1 ± 14.7	511.0	558.6	1.07 ± 0.04
head set_integration_benchmark	487.1 ± 8.6	482.4	511.2	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base uint256_integration_benchmark	3.715 ± 0.056	3.634	3.848	1.13 ± 0.04
head uint256_integration_benchmark	3.288 ± 0.099	3.221	3.544	1.00

[codecov]

Codecov Report

❌ Patch coverage is 99.35897% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 96.23%. Comparing base (eeebcc9) to head (c610d41).

Files with missing lines	Patch %	Lines
vm/src/vm/vm_memory/memory.rs	99.06%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2391      +/-   ##
==========================================
+ Coverage   96.21%   96.23%   +0.01%     
==========================================
  Files         107      107              
  Lines       37925    37985      +60     
==========================================
+ Hits        36490    36555      +65     
+ Misses       1435     1430       -5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.