Thomas Daede
Thomas Daede
A long time coming, but I've rebased this and fixed the nits.
Would you prefer that the error message for width/height be updated to indicate that it should be less than 65535 or max_width, or that a new error message for explicitly...
OK, it now displays: ``` !! Invalid configuration: invalid width 720 (expected >= 16,
(we should probably remove that redundant CLI error printing)
Ah indeed I forgot to update a parameter of the format string. Now displays: ``` !! Invalid configuration: invalid width 1280 (expected >= 16,
Bumping this for later as the compression gain is small.
It's totally okay to have it in the repo (that way it's version tracked with the code), just someone needs to update it regularly. I don't have any particular preferences...
Resulting assembly: ``` example::first_max_element_avx2: vmovdqu xmm0, xmmword ptr [rdi] vpmaxsd xmm0, xmm0, xmmword ptr [rdi + 16] vpshufd xmm1, xmm0, 238 vpmaxsd xmm0, xmm0, xmm1 vpshufd xmm1, xmm0, 85 vpmaxsd...
It looks like the autovectorizer fails to use ymm registers for the max. Maybe an intrinsic is needed there too?
Ah yeah sorry, yeah the first max comparison can only be 4 vs 4.