
On Tuesday 2017-08-15 17:56, Richard Biener wrote:
It keeps to work by accident and then people are suprised that suddenly fips check start to break stuff...
Also this has lovely benefit of speeding all the 32bit stuff like steam for 64bit users, so by this you are saying that you preffer 12y+ old hw over the current one.
Please back your claim up by numbers. The important parts, such as glibc, should be using run-time detection already.
Recently 'gd' came up as an example which has a configure test whether it can enable -mfpmath=sse (which it can with -march=x86-64 but not with -march=i586). I suspect there are quite some packages that might benefit.
Well ok, so that's *that* kind of speedup, i.e. replacing x87 floating point math like multiply and divide. My number - from the year 2006 - was that vorbis-tools improved by ~17% when enabling (just) SSE1 this way. Then again, are the low-power machines the place where one would (still) do encoding today? :-)
Note that Leap 15 will (likely) have this change so it's going to be somewhat odd that Tumbleweed doesn't.
Still waiting for somebody to confirm he's actually running such old machine on Tumbleweed ;)
I have a working vintage SSE1-only Athlon XP system on an L7S7A2 board with 1536 MB memory installed, which is 2x more than what it was sold with decades ago. I also have a vintage TM5800 system which I recall likely did not have SSE - or either CMOV. I should check which when getting home. Has the system max of 448MB installed. -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org