Update the qHash function for strings to use the CRC32 instruction
authorThiago Macieira <thiago.macieira@intel.com>
Fri, 20 Dec 2013 07:32:04 +0000 (23:32 -0800)
committerThe Qt Project <gerrit-noreply@qt-project.org>
Thu, 23 Jan 2014 00:06:34 +0000 (01:06 +0100)
commitea8e48a6799cf742ea23f4a30dcfc38a4988fe56
tree3c8c09dd3ded200ac2d8f3f76752e0dc5c509dba
parent506c34fdd2b0a7adfd79b0a42b91f0bcc7b045ec
Update the qHash function for strings to use the CRC32 instruction

According to my profiling of Qt Creator, qHash and the SHA-1 calculation
are the hottest spots remaining in QtCore. The current qHash function is
not really vectorizable. We could come up with a different algorithm
that is more SIMD-friendly, but since we have the CRC32 instruction that
can read 32- and 64-bit entities, we're set.

This commit also updates the benchmark for QHash and benchmarks both
the hashing function itself and the QHash class. The updated
benchmarks for the CRC32 on my machine shows that the hashing function
is *always* improved, but the hashing isn't always. In particular, the
current algorithm is better for the "numbers" case, for which the data
sample differs in very few bits. The new code is 33% slower for that
particular case.

On average, the improvement (including the "numbers" case) is:

 compared to          qHash only          QHash
Qt 5.0 function          2.54x            1.06x
Qt 4.x function          4.34x            1.34x
Java function            2.71x            1.11x

Test machine: Sandybridge Core i7-2620M @ 2.66 GHz with turbo disabled
for the benchmarks

Change-Id: Ia80b98c0e20d785816f7a7f6ddf40b4b302c7297
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
src/corelib/tools/qhash.cpp
tests/benchmarks/corelib/tools/qhash/main.cpp
tests/benchmarks/corelib/tools/qhash/main.h
tests/benchmarks/corelib/tools/qhash/outofline.cpp