[macintologist]

http://en.wikipedia.org/wiki/Streaming_SIMD_Extensions

There's SSE, SSE2 and SSE3. These are sort of like Altivec equivalents in the x86 world.

When is Apple going to start optimizing the x86 binaries of their apps with these SIMD instructions?

Or do they already?

[mduell]

OSX requires a minimum of SSE3, although it has been hacked to support machines that only have SSE2. The system vector libraries support the use of SSE (1/2/3) in addition to Altivec.

[Big Mac]

I think Apple already is. As mduell points out, an official requirement is SSE3 because Rosetta uses it. And I'm sure it's used elsewhere.

[Binarymix]

All the intel binaries are optimized for SSE2 when they are compiled. Some also include SSE3 instructions, but since there are equivelents in SSE2 for all SSE3 instructions, the kernel can be recompiled/modified to emulate SSE3 using SSE2 instructions...

[tkmd]

Did Intel release its own compiler for OS X or is Apple still using gcc from 10.4.4 to now?

[OreoCookie]

The compilers automatically use SSE, unless you explicitly tell them not to. AFAIK the fpu is actually integrated into the SSE unit of the P4 at least.

[tkmd]

Originally Posted by OreoCookie

The compilers automatically use SSE, unless you explicitly tell them not to. AFAIK the fpu is actually integrated into the SSE unit of the P4 at least.

OK but do you know if apple is using Intels' compliers or gcc when compiling for the Intel versions of Mac's in 10.4.5?

[OreoCookie]

Originally Posted by tkmd

OK but do you know if apple is using Intels' compliers or gcc when compiling for the Intel versions of Mac's in 10.4.5?

I assume they use gcc, although I don't know.
In any case, it doesn't make a difference for the topic of this thread since both use SSE instructions, unless you explicitly tell them not to.

Hence to answer your initial questions: SSE is already used heavily in Apple's software for Intel.

Probably you confuse this a bit with the way SSE (or any sort of SIMD instruction set, be it AltiVec or SSE3) can be used to dramatically accelerate media encoding for instance. Some of the code is manually adjusted to increase the speed of the encoding process. Judging from the various QuickTime benchmarks, Apple has still some way to go. On the other hand, AltiVec seems to be more powerful than SSE (in that is easier to use and hence easier to optimize for), so we'll have to see in a year or two what Apple does to close the gap.

[Ken_F2]

Apple is using GCC for the OS and its applications at the moment.

Intel has a new compiler for the Mac in beta (final release this fall) that offers a substantial increase in performance, particularly on the Core Duo and Conroe platforms. It's unknown whether Apple will adopt this compiler, but it would net a 5% to 40% boost (depending on app) in performance if they did so.

[TheTraveller]

That'd be cool, a nice 5-40% speed bump, although my Mini Core Duo already is pretty smokin' compared to every other Mac I've ever used.

[Eug]

Originally Posted by Binarymix

All the intel binaries are optimized for SSE2 when they are compiled. Some also include SSE3 instructions, but since there are equivelents in SSE2 for all SSE3 instructions, the kernel can be recompiled/modified to emulate SSE3 using SSE2 instructions...

I'm confused....

I remember reading in some Apple doc that they recommended developers target SSE2 (not SSE3). From what you're saying it seems that Apple itself is just assuming everyone with an Intel Mac has SSE3 (which is true), and isn't really targetting SSE2. Am I correct?

In any case, I don't see the harm in targetting SSE3, because AMD chips (which Apple could switch to in the future if necessary) support SSE3 now too.

Originally Posted by Ken_F2

Apple is using GCC for the OS and its applications at the moment.

Intel has a new compiler for the Mac in beta (final release this fall) that offers a substantial increase in performance, particularly on the Core Duo and Conroe platforms. It's unknown whether Apple will adopt this compiler, but it would net a 5% to 40% boost (depending on app) in performance if they did so.

It's several months later now, so you wouldn't have known this at the time, but Apple is at least using Intel's compilers for its SPEC benchmark submissions.

[RevEvs]

I'm not 100% sure, but I think I read that developers use the Acceleration (?) Framework. This Framework then uses the correct optimizations depending on the hardware (i.e. Altivec on PPC and SSE on Intel).

If thats correct, then if apple update the Acceleration Framework, all apps that use it will benefit.

So if theres a SSe version, then all Altivec enhanced apps will get the SSE optimizations

[lookmark]

Originally Posted by Ken_F2

Apple is using GCC for the OS and its applications at the moment.

Intel has a new compiler for the Mac in beta (final release this fall) that offers a substantial increase in performance, particularly on the Core Duo and Conroe platforms. It's unknown whether Apple will adopt this compiler, but it would net a 5% to 40% boost (depending on app) in performance if they did so.

Curious: is it nontrivial for Apple to go and back forth between say GGC and Intel's compiler, or would adopting Intel's compiler in effect tie Apple much closer to using Intel chips?

Also: does anyone know what compiler MS uses for Windows?

[Millennium]

Originally Posted by lookmark

Curious: is it nontrivial for Apple to go and back forth between say GGC and Intel's compiler, or would adopting Intel's compiler in effect tie Apple much closer to using Intel chips?

It's nontrivial: this codebase has been using GCC since the NeXTStep days.

Also: does anyone know what compiler MS uses for Windows?

MS uses its own compiler.

[lookmark]

Oops, meant GCC, not GGC. Thanks for the response.

Seems the best path would be to offer both compiling options in XCode, and let developers choose, per app.

[Eug]

OK, here is some documentation from the AltiVec/SSE Migration Guide:

Detecting SSE3
Instruction Overview

This may be old, but they haven't updated it in the docs in any case. They are still recommending that if you use SSE3, you should add code to detect it first.

Why? I dunno, cuz all Intel Macs ever released already have SSE3, including the first Intel developer boxes.

[RevEvs]

Originally Posted by Eug

OK, here is some documentation from the AltiVec/SSE Migration Guide:

Detecting SSE3
Instruction Overview

This may be old, but they haven't updated it in the docs in any case. They are still recommending that if you use SSE3, you should add code to detect it first.

Why? I dunno, cuz all Intel Macs ever released already have SSE3, including the first Intel developer boxes.

because making assumptions in your code always comes back to bite you on the ass

the number of times ive had to fix a bug caused by a dev assuming a situation could not possibly exist...

[Eug]

I understand your point, but why not require detecting SSE2 too then? Or even SSE and MMX for that matter?

I mean, even Cyrix chips support SSE3 now.