[tkmd]

I was wondering if anyone could tell me what what optimizations are going to occur in Leopard? I know that the PPC is dead, so I fully expect Leopard to be as fast but probably slower (increase overhead or bloat) on the PPC but what is Apple doing for the intel machines? Is SSE3 used in Tiger if not is it going to be used more extensively in Leopard? Is the OS going to be more "tuned" to multi-core architecture?

I really don't care much about timemachine or the new color bouncy thing in ical or mail or whatever. I want performance - how is apple going to deliver this?

[Catfish_Man]

Like all previous OSX updates, Leopard will contain optimizations in many places in many ways, and should be faster overall on all machines.

[Blasphemy]

They're finally taking the debugging code out.

[mduell]

I doubt we'll see much performance increase... the new OpenGL threading will help a bit, but only in apps that depend heavily on it.
Perhaps a heavier reliance on SSE3/SSSE3 since C2D can execute vector instructions in 1 cycle.

Does anyone know what compilier Apple is using? I'm hoping icc, but I'm not optimistic.

[Angus_D]

Originally Posted by mduell 

Does anyone know what compilier Apple is using? I'm hoping icc, but I'm not optimistic.

At what point did any of the reasons that they used gcc in Tiger (and every other release before that) become less pertinent?

[P]

I know that Intel has a version of icc for OS X - it was part of the original deal that they would develop one - so I would guess that Apple is using it for at least some of the OS.

Architecturally specific speed gains come from two sources: compiler improvements and fine-tuned assembler code. There is little to no assembler code in OS X, so we can discount that. While compiler optimization is always a work in progress, and there is certainly some to be done to get Core2 to run even better, both icc and the Intel version of gcc were in very good shape already. There is little to be gained there. I don't know what sort of progress has been made on the PPC version of gcc lately, but there is certainly more to gain there, if anyone has been working on it.

More general speed gains can come from higher quality code and moving more tasks to other parts of the hardware. The former is always a possibility. The latter is happening slowly, but we haven't heard about anything specific about it happening for Leopard. Finally, older subsystems can be ripped out and replaced with newer, faster ones. One obvious candidate for this is HFS+, which is the biggest speed bump around for certain tasks, but that is also unconfirmed.

[OreoCookie]

Originally Posted by mduell 

Does anyone know what compilier Apple is using? I'm hoping icc, but I'm not optimistic.

It's gcc and won't change in the forseeable future. The reason is to maintain cross-platform compatibility (which has always been the purpose of project Marklar, i. e. OS X on Intel). I hope Apple will stick with gcc. Parts of the code doesn't compile with gcc -O2, let alone gcc -O3, so comparing highly optimized gcc code and icc code doesn't make sense in this situation.

[bowwowman]

Originally Posted by mduell 

the new OpenGL threading will help a bit, but only in apps that depend heavily on it.

as in OS X itself

Most if not all of the GUI functions and it's "eye candy" has already been optimized for OpenGL, and it's rendering has been moved to the GPU, where OpenGL can really show it's best performance........

[Catfish_Man]

Originally Posted by mduell 

I doubt we'll see much performance increase... the new OpenGL threading will help a bit, but only in apps that depend heavily on it.
Perhaps a heavier reliance on SSE3/SSSE3 since C2D can execute vector instructions in 1 cycle.

Does anyone know what compilier Apple is using? I'm hoping icc, but I'm not optimistic.

I'm under NDA, so I can't talk specifics, but I know of one important subsystem that is 4 times as fast in Leopard when used in a certain mode. That particular improvement will only apply to a very small class of applications though.

In general, I doubt platform-specific microoptimizations will be a big deal outside of Accelerate.framework, and perhaps the graphics system. Most improvement will come from better/fixed algorithms (example: the new cache implementation in webcore), or better use of existing APIs (example: clipping translucent CGLayers to the minimal size), or from allowing applications to disable expensive parts of the API that they don't need (example: the -isOpaque method on NSView, which allows you to avoid alpha blending for things that aren't transparent).