test: add comprehensive unit tests for vermicelli and noodle accelerators

[byeonguk-jeong]

Summary

Add extensive unit tests for vermicelli and noodle accelerator functions to improve coverage and catch edge-case regressions.

Tests Added

vermicelli (~750 lines, new file)

• Single-byte small buffer (1–31 bytes) match/no-match
• Exact VECTORSIZE boundary and VECTORSIZE+1 tail path
• Per-position match sweep (forward and reverse)
• Non-alphabetic character matching
• Negated vermicelli (NVermicelli) small buffers and nocase
• Reverse and reverse-negated vermicelli
• Double vermicelli: small buffers, partial match at end, nocase
• Reverse double vermicelli: no-match, small buffers, multiple matches
• Double vermicelli masked: small buffers and partial-bit masking
• Alignment stress test (varying offset × varying length)
• Cross-SIMD-boundary double match sweep
• Forward + reverse consistency verification

noodle (~413 lines)

• Early termination via callback (single and double char)
• No-match scenarios for single and double patterns
• Empty and minimal-length buffer handling
• Large buffer scanning (multi-vector iteration)
• Case-insensitive matching for single and double patterns
• Unaligned buffer scanning
• Various alignment boundary conditions
• All-match dense buffers for single and double patterns

gtest: uninitialized variable warning

• Initialize dummy variable in StackGrowsDown() to suppress -Wuninitialized.

sheng: DFA state transition table size mismatch

• Add missing sentinel element to state transition vectors for both 16-state and 32-state DFA test configurations. The alpha_size includes a sentinel entry, so each state's next vector must have alpha_size elements.

@AhnLab-OSS @AhnLab-OSSG

[markos]

There are conflicts could you please rebase against develop?

[byeonguk-jeong]

There are conflicts could you please rebase against develop?

Done :)

[markos]

Several SVE2 related tests have failed, perhaps because the fixes were in a separate PR (which is now merged). Could you please look into them or just rebase against current develop to see if they persist? Thanks again.

[byeonguk-jeong]

Several SVE2 related tests have failed, perhaps because the fixes were in a separate PR (which is now merged). Could you please look into them or just rebase against current develop to see if they persist? Thanks again.

Okay, I just rebased it.

[byeonguk-jeong]

Several SVE2 related tests have failed, perhaps because the fixes were in a separate PR (which is now merged). Could you please look into them or just rebase against current develop to see if they persist? Thanks again.

It seems that vermicelliDoubleMakedExec() is implemented only with NEON, not SVE. It requires at least 16 bytes.
This function is always called with buffer larger than 16 bytes from run_accel(), so it is safe now. However, eventually it has to be refactored with SVE, or at least a fallback has to be implemented for small buffer less than 16 bytes.
For now, I will comment out the unit test with small buffer case.

[markos]

Nice set of unit tests! Again thank you for your contribution!