IBM Announces Mac OS X Compilers

[rjstanford]

Re: Recompiled Software

Originally posted by leo
Don't expect big performance improvements with performance-critical commercial software by recompiling it with these new compilers. Most of this stuff (mp3-/divx-codecs, Photoshop, 3D-software, Quartz-drawing and other critical parts of Mac OS X) is optimally hand-tuned already.

You'd be surprised as to just how little code is hand-tuned these days. The loss in maintainability is just not worth it except for the most critical path stuff, and even then it can be a tossup. If you can make a critical section 10% faster and introduce one obscure bug, that's generally a bad deal. Also, a lot of these optimizations tend to make it next to impossible to do a whole-process optimization down the road.

Not that it never happens, just rarely.

-Richard

 

[Spades]

Re: Re: Re: Small Comment on IBM Compilers and GCC

The question is, will anything in the license prevent somebody from compiling open source projects with XL and distributing the binaries? I highly doubt it. If IBM is going to restrict the code licenses that can be used with XL, I think open source licenses are the last ones they would restrict. Besides the license, the only constraint would be the time of actually compiling it. Compiling each release of more than a handful of projects is unrealistic. Anybody that buys XL should be able to compile open source projects and distribute them. Any volunteers?

XL compiled Gimp 2.0? Yes, please!

Originally posted by kherdin
The amount is trivial to pay for any decent commercial projects, as they will generate enough revenue to subsidize the cost, but this hurts freeware developers. Sure you might say if I make freeware/opensource software I might not need a different compiler from gcc, but that is not always true. There a few larger freeware projects that would benefit from the speed the XL compilers would provide, and this in the end benefits Mac OS X users. I think Apple should seriously consider bundling XL with Xcode, as this will make the platform much better when all apps run a good deal faster, and therefore generate sales and revenue.

 

[Rocketman]

Re: Re: Re: Small Comment on IBM Compilers and GCC

Originally posted by kherdin
The amount is trivial to pay for any decent commercial projects, as they will generate enough revenue to subsidize the cost, but this hurts freeware developers. Sure you might say if I make freeware/opensource software I might not need a different compiler from gcc, but that is not always true. There a few larger freeware projects that would benefit from the speed the XL compilers would provide, and this in the end benefits Mac OS X users. I think Apple should seriously consider bundling XL with Xcode, as this will make the platform much better when all apps run a good deal faster, and therefore generate sales and revenue.

In case there is an "experienced hobbiest programmer" that "needs" the environment, I will BUY it for them if only they will port over some old BASIC code to objective C in the process.

Rocketman

 

[secondshadow]

I keep seeing posts about "hand-tuned optimizations." Has anyone actually attempted to hand-tune machine code? As a general rule, a sufficiently good optimizing compiler will generate much better machine code than a person can especially for large code sections. Not only that, but have you ever tried to write any decently size projects in any type of assembly????? We're talking potentially hours for something that in a low-level high-level language (C) would take sufficiently less time and generate similar code as well as probably faster code as long as the optimizing compiler was well produced.

 

[bousozoku]

Originally posted by secondshadow
I keep seeing posts about "hand-tuned optimizations." Has anyone actually attempted to hand-tune machine code? As a general rule, a sufficiently good optimizing compiler will generate much better machine code than a person can especially for large code sections. Not only that, but have you ever tried to write any decently size projects in any type of assembly????? We're talking potentially hours for something that in a low-level high-level language (C) would take sufficiently less time and generate similar code as well as probably faster code as long as the optimizing compiler was well produced.

I had in my 6502 and 68000 days. It's less necessary with good, commercial compilers because they do the work. However, gcc 2.95 was not really that good at it. gcc 3.3x is much better at producing good code. Before Datalight and Watcom showed the world how to do more than peephole optimisations (1980s), it was necessary to unroll loops by hand, etc.

 

[iomar]

Wow!

This is great! I have been waiting for this day!

 

[wms121]

hmm..this means 64bit xlC at the developers conference right Arn?

..and a 64 bit Java (beta?)..Darwin(beta?)..anything else (beta?)

Is Apple going to give away OpenMp 64 if they give away xlC/C++?

Does Steve dream in color..must be rainbow apples and IBM paneling blues?

DoD needs 970's now.

 

[dashiel]

Re: Recompiled Software

Originally posted by leo
Don't expect big performance improvements with performance-critical commercial software by recompiling it with these new compilers. Most of this stuff (mp3-/divx-codecs, Photoshop, 3D-software, Quartz-drawing and other critical parts of Mac OS X) is optimally hand-tuned already.

if macromedia or adobe (to a slightly lesser extent) are hand tuning OS X apps i will eat my hat and then go buy a pc. it's unbelievably how slow they are compared to well written OS X apps.