Forum > Windows

optimized sources

<< < (107/179) > >>

Jason G:
Go to the development thread  ;)

_heinz:
for all others who have no access to the developer area:

What we can expect ?

testrun against our latest public published astropulse  ap_5.05r168_SSE3.exe
Quick timetable
 
WU : ap_18se08aa_B6_P1_00046_1LC25.wu
ap_5.05r168_SSE3.exe : 410.609 secs CPU
ap_5.05r303_SSE3_ICC_ATOM.exe : 340.719 secs CPU
Speedup     : 17.02%
Ratio       : 1.21 x
 
WU : Raistmer's_tiny.wu
ap_5.05r168_SSE3.exe : 150.547 secs CPU
ap_5.05r303_SSE3_ICC_ATOM.exe : 144.781 secs CPU
Speedup     : 3.83%
Ratio       : 1.04 x
 
WU : sigind_v5.wu
ap_5.05r168_SSE3.exe : 912.922 secs CPU
ap_5.05r303_SSE3_ICC_ATOM.exe : 727.484 secs CPU
Speedup     : 20.31%
Ratio       : 1.25 x

All results strongly similar.
This is basis for our further optimization process.
Some more in the developer area.
 ;)

KarVi:
I have been running the various R303 SSE3 builds on my Phenom.

Strangely enough none of the Atom builds work proberly, allthough they are SSE3, and should be compatible.

Some quick results:

Sigind.wu

ap_5.05r168_SSE3.exe : 845.703 secs CPU
ap_5.05r293_SSE.exe : 775.766 secs CPU
ap_5.05r303_SSE3_ICC_Qopt.exe : 694.078 secs CPU
ap_5.05r303_ATOM_ICC_Qopt.exe : 17.031 secs CPU
ap_5.05r303_SSE3_ICC_ATOM.exe : 0.031 secs CPU

The first 3 give strongly similar, the last 2 clearly don't.

But still 303_SSE3 is much faster than r168_SSE3 and 293_SSE.

[attachment deleted by admin]

Jason G:

--- Quote from: KarVi on 03 Jan 2010, 06:34:36 pm ---Strangely enough none of the Atom builds work proberly, allthough they are SSE3, and should be compatible.
...

--- End quote ---
  'Should be', though I believe ATOM has an extra instruction (MOVBE) which is available in our 45nM Core2's (at least)... So ATOM builds are really ATOM specific, though they should run on later Intels OK. (The SSE3 Qopt one uses the generic SSE3 options ... nice to know it works on Phenom II)

KarVi:
Thats OK, but then they should be marked (S)SSE3 or something to that effect, since they are not really SSE3 compatible.

But I do like the improvements of the real SSE3 build, its all very promising :)

Navigation

[0] Message Index

[#] Next page

[*] Previous page