What's the secret configuration for ICC? Lectrode?

xenofears · 12-16-2010, 02:38 AM #1

Wow.. using ICC 12 XE results in ABYSMAL performance on this i7.. what gives, what's the secret?

Building using VC2008 with default project settings, tried turning SSE[4.2] on/off, off only made it worse.

**Diddy Kong** · 12-16-2010, 03:29 AM #2

Maybe you should PM the guy if no one can give you a straight forward answer, should you even get one. lol I have no idea what you're askin'

hungry man · 12-16-2010, 05:35 AM #3

I'm curious as well. I'm about to start building on my own so this might be relevant =p

xenofears · (This post was last modified: 12-16-2010, 06:12 AM by xenofears.)

I got one piece of advice (I just sent a PM): SSE optimizations to OpenGL cause horrible performance.. I tried turning off the Intel-specific SSE optimizations and it didn't work, but let's see what happens when I try DirectX.

I am also using a newer more elaborate version of ICC than Lectrode, from inside VC2008 (it's integrated in Parallel Studio XE 12).

There's also heaps and heaps of other optimizations besides the standard-fare /02 and the desired /Qx<sse..> (you wouldn't want /Qax, would you?)

Standard stuff is Intrinsics (/Oi), Fiber-Safe Optimizations (/GT)..

The Intel has plenty more, for example:
Interprocedural Optimization (single-file /Qip, multi-file /Qipo)
Optimize for Windows Application [EXE] /GA (most of the stuff is DLL's..)
Floating-Point Speculation /Qfp-spectulation:...
Matrix Multiple Library Call /Qopt-matmul
Parallelization /Qparallel
**Intel Optimized Headers** /Quse-intel-optimized-headers

There's plenty more. For some reason VC2008 doesn't seem to have easy options to enable IPP, Intel's Math Library (There are several options here).. again don't know if it is a good idea to use *ANY* of this Smile

Unnecessary optimizations tend to actually slow it down..

ICC also doesn't seem to know how to not come dangerously close to overrunning my entire 4gb of RAM..

Lectrode · 12-16-2010, 06:33 AM #5

(12-16-2010, 05:47 AM)xenofears Wrote: There's also heaps and heaps of other optimizations besides the standard-fare /02 and the desired /Qx<sse..> (you wouldn't want /Qax, would you?)

Just a quick note: programs compiled with /Qx will ONLY run on Intel processors, while /Qax will run on both AMD and Intel.

I am now banning myself from internet usage before I get more distracted lol. stupid finals...
I should be back Friday night if not sooner.

xenofears · 12-17-2010, 02:40 AM #6

It's funny how they don't tell you that unless you dig the documentation, you'd think SSEx is SSEx, just Intel being proprietary. If you build with /arch:SSE2 /Qax:SSEx you'll get a build that can run both regular (SSE2) and SSEx optimized Smile

.. too bad you can only add one extra set of optimizations... but it would probably take FOREVER to build otherwise anyway.

x64 SSE4.2 Dx9 was crashing on my computer, sent you a PM. x86 SSE3 didn't. Trying x64 SSE3.

Man ICC uses a boatload of RAM, it's all in the linking where it does most optimizing. It's only using 1 core per link and I can't run more than one at a time because each one uses over 3GB of RAM, Dolphin itself easily pushing the 4GB limit!