[B2-devel] Improvements to AltiVec emulation

[Gwenole B. <gbe...@us...>]

Hi,

I have just committed a few changes to the AltiVec emulation. First, I=20=

totally nuked the use of GCC intrinsics as some people had a bugous=20
compiler, that would have caused some incorrect results. Next, I=20
noticed I wasn't using SSE2 optimizations, that's now fixed. As a=20
bonus, my system now flies at 1025 Megaflops with AltiVec Fractal=20
Carbon (vs. 807 MegaFlops previously):
<http://gwenole.beauchesne.free.fr/benchmarks/altivec.html>

Does someone know about another altivec benchmark that may use integer=20=

vectors? The G4 HINT benchmark can be interesting but I don't have a=20
CodeWarrior new enough to compile it.

Some slight improvements are possible but I will see if the Windows=20
port can be brought up to date.

Bye,
Gwenol=E9.=