I just optimized PyFloat_PackN() C functions in Python using __builtin_bswapN() to swap bytes. Problem: GCC does crash and clang miscompiles the 16-bit function which uses _Float16 on IBM s390x! GCC bug report: bugzilla.redhat.com/show_bug.cgi... clang bug report: github.com/llvm/llvm-pr...
bugzilla.redhat.com
2544649 – GCC crash on s390x when building a C function which uses _Float16/uint16_t, memcpy() and __builtin_bswap16()