fix(mem2reg): Do not attempt to analyze an instruction simplified to a global

[vezenovm]

Description

Problem*

Resolves #9880
Resolves #9861

Summary*

We were panicking here noir/compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs

An index OOB when accessing an instruction on the DFG directly is usually caused by attempting to a mismatch between the global and function graphs and using the ID from one in the other.

Instead of using get_local_or_global_instruction I chose to instead just check whether we have simplified to a global value. We know globals cannot have referencing so we do not need to go through this analysis. We also use the InstructionId to fetch results when analyzing instructions so I felt it was easier to simply skip global handling rather than try and make analyze_possibly_simplified_instruction check for whether we have a global.

Additional Context

For some reason I am not getting an AST from the fuzzer. I ran the following:

NOIR_AST_FUZZER_SEED=0xe6851ab700100000 NOIR_AST_FUZZER_SHOW_AST=1 RUST_BACKTRACE=1 cargo test -p noir_ast_fuzzer_fuzz pass_vs_prev 

But I am not getting an AST, so I was unable to produce a simplified regression. Running locally, the seed above passes on this PR but not on master.

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[aakoshh]

For some reason I am not getting an AST from the fuzzer.

Sorry about that, the printing of the AST could use some overhaul. The NOIR_AST_FUZZER_SHOW_AST is used by create_ssa_or_die to print the AST just before running the SSA pipeline on it, which happens during the generation of arbitrary test inputs. However that's not used for pass_vs_prev, where we generate the AST, then the initial SSA, then run some passes, but not the whole pipeline. For no apparent reason, the logging of the AST and the SSA happens during exec rather than the arbitrary generation, probably because this way it stayed together and in the generic part of the tester. Because SSA generation panicked, it never got to exec. Unlike the other cases, it also uses log::debug!, so to it needs to be enabled by RUST_LOG=debug. It would probably be better if all of them used that, rather than SHOW_AST variants.

[aakoshh]

LGTM!

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Execution Time'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: b5bec32	Previous: 13fbc6f	Ratio
rollup-root	0.005 s	0.004 s	1.25

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

[vezenovm]

Now that we have the AST output I wanted to add a reduced regression test. I added a small unit test that would fail without the checks included in this PR.

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Test Suite Duration'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: 2ddaf34	Previous: 13fbc6f	Ratio
test_report_zkpassport_noir_rsa_	2 s	1 s	2

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench