GPU-Accelerated Synthesis of Mixed-Boolean Arithmetic: Beyond Caching

This collection of PL papers contains several items with clear security relevance. Key points: (1) SIMBA introduces a GPU-accelerated, cache-free synthesizer for Mixed-Boolean Arithmetic (MBA) expressions, which materially speeds up program deobfuscation and can aid cryptanalysis and reverse engineering. (2) IRIS-14B demonstrates an LLM fine-tuned to translate GCC GIMPLE to LLVM IR with strong accuracy; this can enable cross-toolchain workflows but also risks silent semantic mismatches or miscompilations if used as a drop-in translator. (3) Elevator presents a deterministic, fully-static x86-_