On Sat, 12 Aug 2017 11:30:46 -0600 Michael Sartain <mikes...@fastmail.com> wrote:
> Signed int values were being used where the original code used uint32_t types: > > http://www.azillionmonkeys.com/qed/hash.html > > Right shifting negative int values has implementation-defined and left > shifting > has undefined behavior. > > On my platform (x86_64) right shifting was doing sign extension and filling > high bits with 1s, which is different than the original algorithm. > Heh, nice catch. Although the hash was never used for anything too important. Mostly just colors of the graph. > Signed-off-by: Michael Sartain <mikes...@fastmail.com> > --- > trace-hash-local.h | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/trace-hash-local.h b/trace-hash-local.h > index b2a1002..b3f9b06 100644 > --- a/trace-hash-local.h > +++ b/trace-hash-local.h > @@ -22,7 +22,7 @@ > > static inline unsigned int trace_hash(int val) > { > - int hash, tmp; > + unsigned int hash, tmp; > > hash = 12546869; /* random prime */ > > @@ -34,7 +34,7 @@ static inline unsigned int trace_hash(int val) > */ > > hash += (val & 0xffff); > - tmp = (val >> 16) ^ hash; > + tmp = ((unsigned int)val >> 16) ^ hash; Why not just have val be unsigned int, and not do the typecast? -- Steve > hash = (hash << 16) ^ tmp; > hash += hash >> 11; >
