Steinar H. Gunderson explains __builtin_memcpy decompilation patterns

In the twelfth part of his decompilation patterns series, Steinar H. Gunderson discusses how GCC's __builtin_memcpy appears in decompiled output. He explains that what looks like manual loop-based data copying is often an optimized, unrolled memory copy. The article provides guidance on how to simplify these patterns back into readable C code.

Identifying unrolled memcpy

• Decompiled code often shows repetitive assignments (e.g., copying 16 bytes at a time) instead of a single function call.
• This pattern stems from GCC's __builtin_memcpy, which can occur implicitly during struct or array initializations.
• Depending on variable alignment, the resulting assembly may use standard load words (lw) or more complex sequences like lwr/lwl.
• To clean up the code, developers should replace these loops with a standard __builtin_memcpy call or direct array initialization.