Add AVX2 and AVX512 optimization#1552
Conversation
How was that measured? Total performance gain, or just on the DWT? |
Total performance gain |
|
This breaks at least Windows x64 regression tests. |
…n MSVC 2015 toolset
Thanks for reply, I was able to fix it. Root cause is that CI use MSVC 2015 toolset that does not have _mm256_extract_epi32 and _mm256_insert_epi32 intrinsic defined, |
|
@tszumski Can you fix the formatting of the code according to the instructions at end of https://github.com/uclouvain/openjpeg/actions/runs/10718601930/job/29724100308?pr=1552, that is |
Fixed |
|
@tszumski I don't have hardware to test AVX512F and github CI doesn't seem to have machines with it. Could you paste somewhere (here or a github gist if too large):
|
|
@rouault
Failed Tests 1147/1148 are also failing on master branch without my changes |
yes, those are "expected" (we have some a file in CI to ignore them ok, so everything looks good. Merging |
Encoder: performance gain ~0.1%
Decoder: performance gain ~2.5%