Merge pull request #16556 from ChipKerchner:vectorizeIntegralSumPixels
authorChip Kerchner <49959681+ChipKerchner@users.noreply.github.com>
Fri, 28 Feb 2020 16:34:06 +0000 (11:34 -0500)
committerGitHub <noreply@github.com>
Fri, 28 Feb 2020 16:34:06 +0000 (19:34 +0300)
commit8c24af66bd18b6cadce602ac392b06a72fe229c1
tree449eb2ee550dba95d5cd0446813a22ebd879b8fc
parent7ffab23a531698a3d5c6d6b5c9105f7f7046321d
Merge pull request #16556 from ChipKerchner:vectorizeIntegralSumPixels

* Vectorize calculating integral for line for single and multiple channels

* Single vector processing for 4-channels - 25-30% faster

* Single vector processing for 4-channels - 25-30% faster

* Fixed AVX512 code for 4 channels

* Disable 3 channel 8UC1 to 32S for SSE2 and SSE3 (slower).  Use new version of 8UC1 to 64F for AVX512.
modules/imgproc/src/sumpixels.simd.hpp