On Feb 17, 2006, at 5:04 PM, Michael Rice wrote: On Feb 17, 2006, at 10:55 AM, Chris Espinosa wrote: On Feb 16, 2006, at 5:09 PM, Christopher Drum wrote: Now that the Intel Macs are shipping, has anyone seen compile time tests on PowerPC vs. Intel in Xcode?
Our internal tests show that big C++ apps build nearly as fast on a Core Duo as they do on a quad G5. and Xcode on a MacBook Pro builds faster than CodeWarrior 10.
Can you attribute how much of that is due to the processor and how much is due to a different compiler? I know that a lot of work goes into tuning GCC performance for Intel platforms and not nearly as much work for other platforms, such as PowerPC and especially MIPS.
I can't really give a breakdown; we haven't, for example, built the entire gcc/Xcode toolchain in -O0 v. -Os and measured the percent improvement gained by optimization, which would show to some degree how much the compiler affects the code execution speed. I have heard (unsubstantiated) assertions that the PowerPC benefits more from high levels of optimization, where Intel executes poorly-optimized code better. We have also seen real cases where on certain algorithms the PowerPC beats the Intel chip hands down.
And of course this only is valid for highly CPU-bound operations. Once you start hitting the cache or going out to disk, the performance of the CPU bus, memory, and disk start affecting the numbers much more than raw instructions per second. Cache misses and swapping caused by code bloat will also affect relative performance; a tight but unoptimized PowerPC loop that fits in cache might outrun a highly optimized but fatter Intel loop that causes cache misses; a large optimized app might cause more swapping than a smaller unoptimized one.
So we really are only going to cite specifics of controlled benchmarks and generalities of overall scenarios, and not try to give detailed breakdowns, as the details are highly idiomatic.
Chris |