Intel Core 2, Core i7 Optimizations For GCC 4.6

[kraftman]

Are they going to boycott AMD processors as Intel Compiler always did?

http://www.agner.org/optimize/blog/read.php?i=49

You probably don't have to worry:

http://developer.amd.com/cpu/gnu/Pages/default.aspx

I wonder if llvm also gets such optimizations? Thankfully there's only one way - BSD with clauses or not → GPL and no way back.

[deanjo]

It just happens that nobody is in for making AMD optimizations,

What sandbank has your head been hiding in? AMD has done of optimizations for GCC on their processor line. They aren't the only ones. PGI, Sun, Rapidmind, Pathscale, Absoft, Microsoft, etc. AMD has a long history of commiting patches to GCC as well as other projects such as open64 for processor optimizations with each family of processor release. Intel traditionally has done very little for optimizations with GCC.

[deanjo]

Are they going to boycott AMD processors as Intel Compiler always did?

http://www.agner.org/optimize/blog/read.php?i=49

Well they never "Boycotted" them, they did however cripple support for them. They can't do that however after the AMD/intel antitrust. It was part of the settlement.

[deanjo]

Isn't the case with GCC that AMD themselves work on it, but Intel contracts it out?

If only intel farmed out their graphics development as well. You might see some decent performance then in linux from their graphics instead of the crippled performance they put out now.

[fabiank22]

If only intel farmed out their graphics development as well. You might see some decent performance then in linux from their graphics instead of the crippled performance they put out now.

Nope. They outsourced Poulsbo, both hardware and driver and it's a mess.

[V!NCENT]

@deanjo:
I was hiding in Gnome cave for a while. Needless to say that teleporting back to the KDE Enterprice was the right thing to do

[blackshard]

Well, if these optimization can be applied to AMD cpu's they will be. It's not like with Intel's own compiler where they can cripple the codepath for anything not genuine Intel. For Intel to use CodeSourcery seems smart since they have worked on GCC for a very long time and also has a member on the GCC steering commitee. My Core i7 looks forward to these optimizations

Oh it would be nice if those optimizations are generic for all processors. Actually it would be an optimization for the whole GCC.

The fact I'm a bit worried about is that those optimizations are for Core 2 and Core i7 processors: "Unaligned vector instructions" exist even in the first SSE instruction set, so why they say they are for Core 2 and Core i7? Are they going to put SSE4 in GCC (actually unsupported on AMD, but soon they will be supported), are they going to put a cpu dispatcher like in the Intel Compiler that enables the usage of those instructions on just Intel processors or are they just making some marketing?

[Svartalf]

Nope. They outsourced Poulsbo, both hardware and driver and it's a mess.

Well, it's not WHOLLY their fault. If they'd not outsourced Poulsbo's GPU to Imagination Technologies, and kept the 3D part in-house it'd probably been better. I fault them for THAT particular decision they made on that platform (What were you THINKING guys? Oh, wait...you WEREN'T...). The driver outsourcing was done with the main go-to guys for 3D driver support for Linux- but there's only so much one can do with the PowerVR stuff. It's a nice chip and all, but even ImgTec took forever to get OpenGL ES and X11 support sorted out on the OMAP3 in a manner that was usable for all. Couple that with it being closed and you have a disaster like we've seen with things there.

The main problem they have with driver performance is that doing shader related work is complicated- which is why Gallium3D moves somewhat slower than we'd like for it to.