Being spun around in a “dance of the Endians,” Apple has recently decided it would garner more attention from Intel, as the chip supplier of choice, than it had from IBM. As a matter of economics, many industry watchers suspect IBM has given Sony its full attention. Mainly because its next-generation game console would ship in the tens of millions, requiring just as many processors. That view left Apple, with less than 5 million MAC shipments in 2005, out of focus for IBM. In terms of intrigue, one can also wonder how Microsoft will deal with Intel in the future. But that’s another story….

Are two of the new Intel devices in the new Intel-based Mac capable of keeping up with the PowerPC? That is the big question for application designers. Or, will the performance of two highfalutin PowerPCs dance circles around two Intel processors? So far it appears that the answer is Yes they will. The current MAC, with optimized applications running on PowerMacs with G4 and G5 PowerPCs, will need re-writing and recompiling to make effective use of the dual Intel processor’s features. Years spent by compiler writers, such as those from Metrowerks (CodeWarrior), with a lot of help from IBM, helped produce very high optimization capabilities that took advantage of one, two, or more PowerPC cores.

This transition follows on the heels of the recent transition of Mac OS 9 to Mac OS X, which features reliable UNIX underpinnings. Multimedia applications play an important role in Macs, as Macs are used for film editing, with Final Cut Pro, for example. Several high-end desktop publishing applications from Adobe, long renowned on the Mac, are preferred due to the PowerMac performance advantage. Newer, G5 PowerPCs have graphics optimizing instructions that bring chunks of compute-intensive data quickly to and from the processor. On a PowerPC, multi-byte variable data is organized into what’s known as a Big Endian (BE) data footprint. This arrangement of data bytes in memory is opposite to the orientation of the byte arrangement (termed “Little-Endian”) used by Intel processors. Little Endian means that the low-order byte of the number is stored in memory at the lowest address, and the high-order byte at the highest address. That’s mostly all people care about in the theme of Endianess. However, for Intel developers this means that the contents of multi-byte data arrays of existing PowerPC applications must be reversed if their contents are to be properly accessed on the Intel processor. The result is that the PowerPC translation program, “Rosetta”, only provides “less performance than PowerMacs running OS9 for many applications”, due to the required swapping of these bytes for execution on an Intel based machine. This will change, ok, - someday. Until then, software written and compiled for the PowerPC based MAC will run faster than a program simply “translated” to a version for the new Intel based MAC.

In addition to the byte-swapping, a later release of Mac OS X will have to be revised to make use of Intel's Streaming SIMD Extensions, or “SSE” (akin to its old MMX). These are all based on the “Single Instruction, Multiple Data” (SIMD) instructions for the Intel processor. SSE is quite similar in function to IBM/Motorola AltiVec SIMD instructions for the PowerPC. Revising high-end graphics programs that make heavy use of AltiVec instead of using SSE instructions is going to be an arduous task. Developers starting this porting process can spare themselves some grief by checking to can see if their mission-critical graphics algorithms are already within Apple’s Accelerate framework (ref: Apple Developer, Introduction to the vDSP Library).

So in conclusion, there are four areas of improvement possible before we can all say that the shift to Intel processors will be successful:
1 - Compilers and new profilers for the Intel based MAC developers need to be as optimized as the PowerPC versions.
2 - Translation via Rosetta or others has to take better care of Endian related processing.
3 - PowerPC based AltiVec enhancements in programs have to be seamlessly translated to the corresponding SSE executions.
If these can be implemented for the first Intel MACs, creative artists will have systems they can openly admit to using in their Apple oriented MAC tribe.
4 – Universal Binary Programming Guidelines, from Apple, will assist developers to modify Mac OS X applications to run as universal binaries. Universal binaries run natively on Macintosh computers using PowerPC or Intel microprocessors to deliver optimal performance for both architectures.

“With those caveats, let the dance commence.”

Richard Hooker is Chief Engineer at Westrex International. Previously he spent ten years, as advisory engineer/scientist with IBM’s advanced personal technology systems’ architecture and enablement groups in Essex Junction, VT and Waltham, MA. Also, while at IBM he was principle overseer of third-party developers and multimedia benchmarking for the PowerPC programs. Hooker also served three terms as vice chair of SPEC’s MediaMark, an industry-standard multimedia benchmark. Additionally, he has published articles and holds several patents in the field of software and benchmarking.