On Thu, 22 Oct 2020 20:40:00 GMT, CoreyAshford <github.com+51754783+coreyashf...@openjdk.org> wrote:
>> Thanks for this question. I also stumbled over it when reviewing. I guess a >> branch which gets mispredicted in ~22% of the cases leads to a big >> performance loss. (In addition, the branch target is not aligned.) > > Yes, it assumes uniformly random data, but also recall that the unencoded > data bytes get shifted by 2, 4, 6 bits into the encoded bytes, which I'm > guessing would tend to make the data somewhat more uniform, even if the > source data has low entropy. > > That said, I didn't actually benchmark it. I will do that to make sure there > is a gain, and if there isn't I will remove the conditional branch. > I took a look at the VSX algo. I haven't looked much beyond it. I had a few > questions I've inlined. It does look like a faithful VSX implementation of > the linked algo. I neglected to thank you for reviewing this code! I realize there's quite a time commitment required to review this properly, and because of that I was having difficulty finding a second reviewer for the PPC64 portion. Just to set expectations, I will be on vacation next week, so further commits won't be posted until the following week, but I will address all of your great feedback. Thanks again! ------------- PR: https://git.openjdk.java.net/jdk/pull/293
