Intel Updates Compilers For 'Leopard'
Page 1 of 1
Intel today announced upgraded versions of its compilers and related technologies designed to support Apple's Mac OS X 10.5, a.k.a. "Leopard." Apple released its new operating system last month after some delay.
Intel has provided Mac compilers since January 2006, right after Apple switched its product line from Motorola's PowerPC processor to the Intel x86 line. The 10.1 versions of the Intel products include the C++ Compiler, Fortran Compiler, Intel Threading Building Blocks, Intel Math Kernel Libraries and Intel Performance Primitives.
All have been optimized for Leopard and the Xcode 3.0 development environment. Xcode allows developers to create binaries from a single code base for multiple platforms, including the old PowerPC-based Mac.
"Apple developers have been very aggressive in using multi-core because Macs have always been multi-core," he told InternetNews.com.
Intel also added functionality for out-of-core solvers, a fancy word for a library able to handle problems larger than can fit in system memory. The company found many scientific users were doing some heavy-duty computing on their computers, even on a MacBook Pro laptop, but the systems didn't have enough memory.
"You might say, 'Just swap it out to virtual memory,' but when you solve very large problems, the performance leaves a lot to be desired," Reinders said. As a result, Intel's Math Kernel Libraries break up the problem into sections and work on it piecemeal, thereby avoiding having to use virtual memory swapping.
Reinders added that the libraries and compilers are cross-platform compatible, so porting between Windows and Mac OS X should be easy -- theoretically.
Of course, he qualified that claim with the warning that it depends on how strongly the app is tied to the OS in terms of custom user interfaces (UI) and using platform specific technologies.
He said applications that aren't very UI-oriented -- more focused on performing tasks, such as scientific apps -- should make for a relatively smooth port. Applications with a heavy UI dependency would take a little longer to port.
In addition to the new libraries, Reinders said developers are seeing around a 10 percent improvement in performance just from moving their code from the old compiler to the new one, without making any further tweaks or optimization.
There are also big performance benefits coming from the auto parallelism, which examines the code and looks for places to parallelize the application, he said.
Intel's C++ Compiler Professional Edition for Mac OS X, featuring the Threading Building Blocks, Math Kernel Library, and Integrated Performance Primitives, is available for $599. The Standard Edition, which ships without the extras, is $449.
Intel's Fortran Compiler Professional Edition, which includes the Math Kernel Library, carries a $699 price tag, while the Standard Edition is $549.