Slashdot Mirror
Slashback: Letters, Time, Revision
Pardon me, do you have the time? Several months ago, we featured a short piece about investigations into clockless computing. Reader xenophrak writes with an update: "Sun Microsystems announces new technology that lets processors run various components of their internals in an asynchronous fashion. The 'FLEETzero' (warning, PDF) chips do not abide by a global clock pulse, and see lower power requirements and heat due to this new feature.
From the web page: 'At the ASYNC 2001 conference, Sun Microsystems Laboratories described FLEETzero, a prototype chip with raw speed roughly twice that of today's chips. Where today's chips use 'synchronous' circuits with a global clock to manage activity, the new, faster FLEETzero chip uses radical new circuits with low-power, asynchronous logic elements that produce timing signals only where and when needed.'
This could have some good impacts on embedded devices, and total processor throughput."
As usual, not so simple. On Saturday you read about Brian K. West, an ISP employee who claimed to be facing unfair threats of prosecution from the FBI for doing nothing more than accidentally discovering a security hole in a local newspaper. A followup posting at Politech indicates that the story isn't quite that simple. Specifically, the FBI's interest in West seems to stem more from alleged attempts at cracking into the violated site than from a simple "found a problem" report. If what the FBI says is true, it changes the story quite a bit.
Time to get a yardstick near the refrigerator ... f97hs writes "Yepps. Delayed almost a week due to regression bugs, the awaited bug-fix release is finally here. Unfortunately, it seems it still can't compile the KDE ARTS-lib (due to, I think, problems with virtual baseclasses). Worth noting is that in order to speed the compiler up, the default to -finline-limit has been lowered. This sometimes leads to considerably slower resulting code, so use -finline-limit=5000 if you compile something you want to be FAST. The mirrors are here and the official release letter from Mark Mitchell might also be worth a read."
- In terms of C++ standards compliance GCC is believed to be the first compiler to achieve full ISO compliance
Who believes that ???They don't support the export keyword for one.
C++ Standard Core Language Defect Reports
C++ Standard Library Defect Report List
Anonymous posts are filtered.
There are three issues
1. Speed of compiler.
2. Size of generated code.
3. Speed of generated code.
When comparing gcc 2.95 and gcc 3.0 with regard to inlining alone, the gcc 3.0 inliner is worse on all three counts. They changed the inliner to apply earlier (at the tree level instead of at the rtl level), which gives it far more oppertunities for inlining. This results (for C++ that uses STL) in order of magnitude slower compiles, several times larger binaries, and, because of cache misses and pipelining issues, significantly slower executables.
The problem is that the old inlining heuristics doesn't work with the new (and potentially much better) inliner. As a band-aid, they decreased one of the old parameters in 3.0.1, the inline limit. This avoids the huge compile times and binaries, but also sometimes misses important inlines. Exactly when you get the important inlines, but without the ridiculous inlines, depend on the application. Sometimes you can't.
For 3.1 the GCC developers will install all new inlining heuristics, which will hopefully be consistently better than 2.95. The potential is there with the new tree-based inliner.
In hindsight, it was probably a mistake to release gcc 3.0 before without the new inline heuristics, however 3.0 was already delayed, and is much better for most code.