perf: Add extensive_hints feature to prevent performance regression for the common use-case

[fmoletta]

Gates behaviour added by #1491 under extensive_hints feature to prevent performance regressions on most use cases

[github-actions]

Benchmark Results for unmodified programs 🚀

Command	Mean [s]	Min [s]	Max [s]	Relative
base big_factorial	2.809 ± 0.039	2.781	2.905	1.00 ± 0.01
head big_factorial	2.802 ± 0.013	2.787	2.820	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base big_fibonacci	2.402 ± 0.008	2.390	2.413	1.00
head big_fibonacci	2.413 ± 0.021	2.386	2.463	1.00 ± 0.01

Command	Mean [s]	Min [s]	Max [s]	Relative
base blake2s_integration_benchmark	8.571 ± 0.233	8.447	9.219	1.23 ± 0.03
head blake2s_integration_benchmark	6.942 ± 0.052	6.882	7.077	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base compare_arrays_200000	2.876 ± 0.031	2.853	2.959	1.15 ± 0.01
head compare_arrays_200000	2.507 ± 0.012	2.484	2.527	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base dict_integration_benchmark	1.875 ± 0.006	1.866	1.885	1.10 ± 0.01
head dict_integration_benchmark	1.703 ± 0.007	1.695	1.714	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base field_arithmetic_get_square_benchmark	1.379 ± 0.003	1.375	1.386	1.17 ± 0.01
head field_arithmetic_get_square_benchmark	1.179 ± 0.010	1.166	1.201	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base integration_builtins	6.924 ± 0.134	6.853	7.303	1.00 ± 0.02
head integration_builtins	6.908 ± 0.036	6.875	6.977	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base keccak_integration_benchmark	8.644 ± 0.134	8.567	9.015	1.23 ± 0.02
head keccak_integration_benchmark	7.026 ± 0.034	6.988	7.113	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base linear_search	2.897 ± 0.010	2.883	2.910	1.14 ± 0.01
head linear_search	2.545 ± 0.007	2.538	2.560	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base math_cmp_and_pow_integration_benchmark	1.881 ± 0.026	1.857	1.934	1.12 ± 0.02
head math_cmp_and_pow_integration_benchmark	1.673 ± 0.009	1.663	1.692	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base math_integration_benchmark	1.797 ± 0.015	1.779	1.832	1.12 ± 0.01
head math_integration_benchmark	1.604 ± 0.008	1.586	1.613	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base memory_integration_benchmark	1.596 ± 0.020	1.577	1.649	1.12 ± 0.02
head memory_integration_benchmark	1.426 ± 0.017	1.414	1.472	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base operations_with_data_structures_benchmarks	1.746 ± 0.007	1.738	1.760	1.12 ± 0.01
head operations_with_data_structures_benchmarks	1.557 ± 0.010	1.538	1.568	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base pedersen	618.7 ± 8.5	612.8	641.3	1.00 ± 0.02
head pedersen	616.5 ± 4.5	613.2	629.1	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base poseidon_integration_benchmark	1.231 ± 0.005	1.222	1.238	1.03 ± 0.01
head poseidon_integration_benchmark	1.196 ± 0.012	1.179	1.224	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base secp_integration_benchmark	2.304 ± 0.011	2.287	2.323	1.09 ± 0.01
head secp_integration_benchmark	2.112 ± 0.007	2.103	2.126	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base set_integration_benchmark	1.212 ± 0.007	1.202	1.220	1.03 ± 0.01
head set_integration_benchmark	1.181 ± 0.003	1.177	1.185	1.00

Command	Mean [s]	Min [s]	Max [s]	Relative
base uint256_integration_benchmark	5.062 ± 0.010	5.048	5.083	1.17 ± 0.01
head uint256_integration_benchmark	4.331 ± 0.047	4.267	4.405	1.00

[codecov]

Codecov Report

Attention: 2 lines in your changes are missing coverage. Please review.

Comparison is base (db708e3) 96.82% compared to head (c023ce7) 96.82%.

Files	Patch %	Lines
vm/src/types/program.rs	98.27%	1 Missing ⚠️
vm/src/vm/runners/cairo_runner.rs	96.15%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1503   +/-   ##
=======================================
  Coverage   96.82%   96.82%           
=======================================
  Files          96       96           
  Lines       39684    39744   +60     
=======================================
+ Hits        38423    38482   +59     
- Misses       1261     1262    +1     

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

[Oppen]

I suggest we first try to see if we can alleviate the hit with a faster hash. It may not be as fast as the array approach, but it will be significantly easier to maintain than having two separate cases and could come quite close.
Popular alternatives:
https://crates.io/crates/ahash
https://crates.io/crates/ritehash
https://crates.io/crates/fnv_rs

We can also make the input to the hash much smaller by simply converting the Relocatable to its compact form before hashing, for example:

impl Hash for Relocatable {
    fn hash<H: Hasher>(&self, state: &mut H) {
        (self.segment_index.unsigned_abs() as u64 << 48 | self.offset as u64).hash(state);
    }
}

This halves the input bytes to the hash for nearly zero effort, which is typically about linear in CPU time. I used unsigned_abs for convenience, but of course this would cause collisions for temporary segments sharing the same offset. I'm not sure we store temporary segments on any HashMap, but if we do we could use the MSB to store the sign if we need.

[Oppen]

Just tried the suggestions in my previous comment on the draft PR #1505. The speedup is quite similar with ahash. I can clean that one up so we don't add the extra feature flag, it should Just Work (TM) for both cases.