How does it come that SSE/SSE2 instructions don't work under DOS with my AMD Athlon XP+? Do they generally don't work under DOS, or are there some MSR to change to enable these instructions?

If yes, are they disabled for energy saving reasons, or just to keep compatibility?
well , they don't destroy any compatibility if proper programming, just like MMX works fine under DOS!

Are there perhaps any AMD/Intel engineers out there who know about this?

I discovered another thing:
Right after starting, SSE/SSe2 instr. are disabled, but after returning from Windows into DOS, these instr. are still enabled, so probably Windows switch something from what I have no idea.

I have already compared lots of MSR/CPUID entries from before and after the windows start, but there are no differences!

HELP!!!

You must enable them in a control register - iirc it has to do with context switches and generated faults, but I'm probably wrong and the intel manuals holds your answers.

Well ,I actually already tried many control registers, but none of CR0,CR2,CR3 and CR4 was different before and after a Windows start.
Perhaps some undocumented control registers?

HEEEEEEEEEEELP, han't anyone a specific idea?

look at the intel.com

i have taken a look yesterday and the only example i have found on the fist visit was: how to optimize 3d on SSE



MATRIX

I FINALLY FOUND OUT!

It's as simple as effective: simply

SET BIT 9 of CR4, an everything is alright!!!
disabling SSE by resetting this bit works also!

(I simply compared these registers before and after Windows started. I think I must have missed this one before)

Till now, it is only tested on my Athlon XP+.

Has anyone an idea if this works on other CPUs? Anyone an idea?

Intel Systems Programming Guide, section 11.4.1: Enabling SIMD Extensions Support... Does *all* SSE instructions fails when bit9 isn't set, or only FXSAVE/FXRSTOR?

On my Athlon XP, ALL SSE-instructions fail, not just FXSAVE/FXRSTOR.