On Saturday, March 31, 2018 at 6:49:25 AM UTC-7, Henri Sivonen wrote: > On Tue, Mar 27, 2018 at 1:36 PM, Henri Sivonen <[email protected]> wrote: > > Do we have numbers of how our installed base is split between earlier > > than Nehalem/Silvermont/Bulldozer vs. Nehalem/Silvermont/Bulldozer or > > later? > > Filed https://bugzilla.mozilla.org/show_bug.cgi?id=1449496
The answer is that our x86/x86_64 release population is 80% fast unaligned SSE2 and 20% slower unaligned SSE2. My conclusion is that I should keep code complexity that caters to the 20% on the most important code paths but that for the less important code paths it's probably OK to do unaligned-only. (encoding_rs does unaligned-only for UTF-16LE/UTF-16BE decode, which isn't used much on the Web, but will retain the alignment check complexity for other stuff.) _______________________________________________ dev-platform mailing list [email protected] https://lists.mozilla.org/listinfo/dev-platform
