Excerpts from Nicholas Bamber's message of 2012-06-19 11:24:10 -0700: > On 19/06/12 19:08, Clint Byrum wrote: > > Excerpts from Nicholas Bamber's message of 2012-06-19 08:35:29 -0700: > >> I have tested with gcc-4.4 and it seems to work okay. So that is an option. > >> > > > > We should do some general performance benchmarks with 4.4 vs. 4.7 before > > we consider this option. It would be a shame to compromise the whole of > > mysqld's performance just to improve SSL performance, as only a small > > fraction of users actually make use of SSL and/or the encrypt functions. > > > > I think at this point I'm leaning toward TAOCRYPT_DISABLE_X86ASM on > > i386 as the short term fix. If Oracle figures out the ASM issue and > > can give us a patch for it soon, then we can apply that, but for now, > > this seems the solution that penalizes the fewest users. > > > Assuming we don't just stick one of thumbs in the air, how would you > plan to go about this? We could just compare the build times. We could > add some timestamps to debian/rules so we can focus on the test part of > the build. I don't think I have got much more stomach for any more > testing than that as I have dealt with a fair few RC bugs over the past > month.
I discussed this briefly with Adam Conrad whom knows a fair bit more about GCC than I do. He assured me that 4.4's i386 code would be mostly identical to 4.7's, so there shouldn't be risk of performance regression. So, if 4.4 is indeed staying, then it sounds like a good option to just build with gcc 4.4 on i386, and 4.7 on all other platforms. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
