log2(1) is 0 and log2(0) is -1. If we have the int64_t 0xffffffff then
log2(0xffffffff >> 31) = log2(0x1) = 0, so the current reduction code
would not shift. That's a bad idea, though, since 0xffffffff when
interpreted as an int32_t would become a negative number.
We need to always shift one more than the current code does to get a
safe reduction. This also means we can get rid of another compare/branch
since -1 is the smallest result log2() can return, so the shift can no
longer go negative now.
Signed-off-by: Julius Werner <jwerner@chromium.org>
Change-Id: Ib1eb6364c35c26924804261c02171139cdbd1034
Reviewed-on: https://review.coreboot.org/c/coreboot/+/42845
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Yu-Ping Wu <yupingso@google.com>
Reviewed-by: Joel Kitching <kitching@google.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
63358eaff6