diff options
author | Thiago Macieira <thiago.macieira@intel.com> | 2014-01-26 23:45:27 -0800 |
---|---|---|
committer | The Qt Project <gerrit-noreply@qt-project.org> | 2014-05-21 17:10:56 +0200 |
commit | 309d3557ca832c42b8fbd372b957af51510b159e (patch) | |
tree | b2e5f2ede70647dcc1747e5537c46ba0d1ad4a3a /src/corelib/corelib.pro | |
parent | 4dba08eebf7db7e801ccbf83dec6f17369123f10 (diff) |
Improve a few string operations with AVX2
AVX2 brings the new PMOVZXBW instruction that extends from one 128-bit
SSE register to an 256-bit AVX register. With that, the main decoding
code is just two instructions (the loop requires a couple more to
maintain the offset counter and do the end-of-loop check).
This buys us another 4% performance improvement in the fromLatin1 code,
calculated on top of the VEX-encoded SSE2 code (which is already a little
better than plain SSE2).
Change-Id: I675fa24de4fa97683b662f19d146047251f77359
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@digia.com>
Diffstat (limited to 'src/corelib/corelib.pro')
0 files changed, 0 insertions, 0 deletions