On Mon, Jun 15, 2009 at 08:15:32PM -0400, Bill Nottingham wrote:
Richard W.M. Jones (rjones@redhat.com) said:
On Mon, Jun 15, 2009 at 08:01:09PM +0100, Jeremy Sanders wrote:
Bill Nottingham wrote:
- Faster and more consistent FP math by using SSE2 registers
- Allows for autovectorization by GCC where necessary
- More clearly delineates our support set of targets, sticking true to forwards innovation, not necessarily legacy support
Why not leave it be and suggest people move to the less brain dead x86-64 instead? Innovation and legacy support.
The slower x86 is, the more motivation there is to move to x86-64.
+1 ...
Well, then... let's build 32-bit x86 with -O0. Or add a few sleep()s strategically in glibc. That'll teach them.
Seriously, if there is a huge non-SSE2 (or, heck, non-SSE - that brings back in Athlon-XP/MP and P3) userbase, that I can understand. But saying 'let's not try and make it better when we can do so with trivial effort - let's leave it slow/make it slower'... that's just silly.
It's fairly sensible to suppose that, at this stage, people who want ix86 support aren't sitting around waiting for a 1% increase in performance.
If they were then they'd be using the latest, hotest hardware, which is all x86-64.
So compiling for baseline i386 isn't such a silly idea.
Rich.