]> git.sesse.net Git - plocate/commit
projects / plocate / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 4f4d842) | patch
Unroll and specialize decode_bitmap_sse2().
authorSteinar H. Gunderson <steinar+git@gunderson.no>
Tue, 6 Oct 2020 19:30:08 +0000 (21:30 +0200)
committerSteinar H. Gunderson <steinar+git@gunderson.no>
Wed, 7 Oct 2020 22:44:35 +0000 (00:44 +0200)
commit58320e42bd3152aee9ebbc4743229d1eb390ec87
tree1e095464d98086f35bd34c9076186603172a93d1tree | snapshot
parent4f4d842fa2baf7ab2218445a3dafcc3acc901c24commit | diff
Unroll and specialize decode_bitmap_sse2().

By asking GCC to unroll the loop, and specializing for the bit width
using templatizing, we can get rid of a lot of the control overhead.
This takes us up from 60% to 80% of reference performance, still
without requiring anything more than SSE2.
turbopfor.h diff | blob | history
Unnamed repository; edit this file 'description' to name the repository.