Yesterday IBM released beta versions of their XL C/C++ and Fortran compilers for Mac OS X. See http://www-3.ibm.com/software/awdtools/ccompilers/ for details. These compilers are based on their highly acclaimed VisualAge for AIX compilers, and as such should potentially generate much higher performance code than any current PPC compiler for OS X (read: GCC and CodeWarrior).
As an example, take Craig Hunter's Jet3D fluid dynamics simulator. In his preliminary tests last month on a pre-release G5 using the Absoft Fortran compiler, Dr. Hunter attained scalar floating point performance on a single 2 Ghz G5 that was equivalent to a 2.66 Ghz P4 - decent but not great (note: this test does not make use of Altivec, and was specifically restricted to run on only one processor). See http://members.cox.net/craig.hunter/g5/
Now, Dr. Hunter has reported on an ArsTechnica forum that in the scalar benchmark, G5 performance increased by an incredible 210% when using the IBM XL Fortran compiler instead of the Absoft compiler. This would technically put it on par with an 8 Ghz P4 in this benchmark, if such a chip existed (again, this is NOT using Altivec or multiple processors). G4 scalar performance, and vector performance for the G4 and G5, also increased by significant margins (40 to 70 percent).
Needless to say, most applications would not see quite so large a benefit from using the IBM XL compiler instead of the current OS X compilers - a 210% increase is pretty insane. Nevertheless, many applications could still see substantial perfomance benefits from using the XL compiler. It is safe to say that the current compilers for Mac OS X (primarily GCC and CodeWarrior) are less than state-of-the-art. In contrast, while Intel might not always build very impressive microprocessors, they certainly do make excellent compilers, so this area (compiler technology) has always been a weak spot of the Mac OS X platform in comparison. But now it looks like IBM is willing to provide a world class compiler for OS X, and this can only be a good thing for the platform!
As an example, take Craig Hunter's Jet3D fluid dynamics simulator. In his preliminary tests last month on a pre-release G5 using the Absoft Fortran compiler, Dr. Hunter attained scalar floating point performance on a single 2 Ghz G5 that was equivalent to a 2.66 Ghz P4 - decent but not great (note: this test does not make use of Altivec, and was specifically restricted to run on only one processor). See http://members.cox.net/craig.hunter/g5/
Now, Dr. Hunter has reported on an ArsTechnica forum that in the scalar benchmark, G5 performance increased by an incredible 210% when using the IBM XL Fortran compiler instead of the Absoft compiler. This would technically put it on par with an 8 Ghz P4 in this benchmark, if such a chip existed (again, this is NOT using Altivec or multiple processors). G4 scalar performance, and vector performance for the G4 and G5, also increased by significant margins (40 to 70 percent).
Needless to say, most applications would not see quite so large a benefit from using the IBM XL compiler instead of the current OS X compilers - a 210% increase is pretty insane. Nevertheless, many applications could still see substantial perfomance benefits from using the XL compiler. It is safe to say that the current compilers for Mac OS X (primarily GCC and CodeWarrior) are less than state-of-the-art. In contrast, while Intel might not always build very impressive microprocessors, they certainly do make excellent compilers, so this area (compiler technology) has always been a weak spot of the Mac OS X platform in comparison. But now it looks like IBM is willing to provide a world class compiler for OS X, and this can only be a good thing for the platform!