This avoids triggering overflows in the filters, and avoids stray test failures in the approximate functions on x86; due to rounding differences, one implementation might overflow while another one doesn't. Signed-off-by: Martin Storsjö <martin@martin.st>
14 KiB
14 KiB