The code fails for test cases with non-zero bits in other positions

[quickfur]

Example: movmask(0x7f7f7f7f) (binary 01111111011111110111111101111111) should return 0x00 (upper bit of each byte in input is 0) but it returns 0xff instead.

Solution: mask out the irrelevant bits before performing the multiplication:

const uint64_t result = (uint64_t)(input & 0x80808080) * mult;

This now produces the correct result in all cases.

[WojciechMula]

You are right, thank you for reporting this issue. There are assertions in demo.c & proof.py that enforce bit layout of input word, but it must be also included in the main procedure.

[WojciechMula]

Fixed in b68de8fad45939ef1770724abad2e479ab5f4ed7