Yeah, but VS Studio only Supports SSE2 and SSE3
But with OpenMP, you're right.
But with OpenMP, you're right.
DefenderX' ICC+OpenMP Dolphin Builds Windows+OS X [DISCONTINUED][UNOFFICIAL]
|
Yeah, but VS Studio only Supports SSE2 and SSE3
But with OpenMP, you're right. 04-02-2012, 10:39 PM
So for my Intel i5 2500K would I see better results with a ICC build or with a favor/O3 build? Thanks.
04-02-2012, 10:56 PM
(04-02-2012, 07:45 PM)DefenderX Wrote: Yeah, but VS Studio only Supports SSE2 and SSE3Automatic compiler optimizations for anything above SSE3 result in exactly zero speedup for Dolphin. OTOH our builtin SSE 4.2 optimizations (if we even have any of those) can be used with MSVC.
Yeah, it's explained in the Windows_Build guide on the googlecode page, how to activate those optimizations. but i'm not sure, if they're working. I didn't test it.
But on notebooks i'm pretty sure, that ICC builds are running faster than the normal builds. Just try it out. (04-02-2012, 10:39 PM)apcarr Wrote: So for my Intel i5 2500K would I see better results with a ICC build or with a favor/O3 build? Thanks. Just try it out, which build works best for you. This depends on many factors. such as CPU, GPU etc. In my opinion, you won't see much differences, because you've got an i5-2500k 04-02-2012, 11:48 PM
ssse3 incluido ?
AMD Phenom X4 9650, ECS A780GM-A v1.1, 4GB Kreton 800MHZ,ATI Radeon HD 5800 Series twin frozr ii 1GB, pFonte Atx 3r System Iceage Ia500hp80 500w + Heat Pipe + 80% HD 500GB
Yes, my ICC builds have SSSE3 instructions included as well
btw: Pls try to write in English. 04-03-2012, 01:23 AM
This last build ( 3.0-583 ) is ICC + OpenMP + favor/O3 for X64 intel?
If not have to make a test? Your /favor build is stable and fast in x64.
Ok, thx for testing
btw: It's not possible to add ICC instructions + /favor in one build. (thx to Lectrode for this hint). (04-02-2012, 01:40 AM)DefenderX Wrote: Edit: Dolphin 3.0-583 /favor:INTEL64 + /O3 build up! -> http://www.mediafire.com/download.php?pnzwliquqd2flil Pls test it. So this build have no ICC optimizations. 04-03-2012, 07:06 AM
(04-02-2012, 10:56 PM)neobrain Wrote: Automatic compiler optimizations for anything above SSE3 result in exactly zero speedup for Dolphin. OTOH our builtin SSE 4.2 optimizations (if we even have any of those) can be used with MSVC.ICC does auto-vectorization so it actually uses the SIMD part of SSE outside of intrinsics functions. An ICC SSE2 build should still be faster than MSVC. Without a reliable way to benchmark your probably right compiler created code for SSE3+ and any gain is probably negated by use of the dispatcher since everyone seems to using multiple /Qax switches now. 04-03-2012, 07:14 AM
ya. I'm not questioning at all that ICC might produce faster builds than MSVC in certain scenarios, I just wanted to point out that it's kinda stupid to advertise "SSE 4.2 optimizations omgwtfbbq" just because you toggled some compiler flag which doesn't change anything in the end, anyway. In fact, it just confuses users around here, it's spreading misinformation (as if we didn't have enough of that around here) and, frankly said, it's kinda annoying thus.
|
« Next Oldest | Next Newest »
|