05-30-2009, 03:38 PM
|
 |
Panther
|
|
Join Date: Jun 2008
Posts: 194
|
|
|
28 May 2009
28 May 2009
Quote:
Big news update today. Voodoo 2 port is going smoothly so far, and Intel-specific parts have been ported already.
The port seems quite stable except for an elusive kernel panic which occurs sporadically, usually when starting Terminal.app shortly after booting up. I'm trying to find and fix this problem. Apart from this, real-world stress test was done with about 1 hour of gaming (Nexuiz) and another hour of multimedia (Logic Pro), and there has been no other regression. Keep in mind this is a ground-up reworking of almost everything, so compared to Voodoo 1.0, new issues might crop up and old issues might be fixed (e.g. "seatbelt.kext" problem is now fixed, and the breakpoints-not-working issue has been identified and fixed in the current Intel build - though it will need to be revisited after AMD specific changes are made).
Focus in this release will be: cutting down on even more patches to make the modifications as clean and minimal as possible while still maintaining the same level of hardware support.
The most important change so far is in the Real-time clock and SpeedStep support. We've totally rewritten the Speedstep and rtclock support in the kernel, this time the changes are quite minimal. Instead of defaulting all processors to use the "slow" method (long multiplication and division) to support clock speeds lower than 1 GHz, what we've done is to switch over to the "fast" version (32-bit scale and implicit shift), but with an upscale factor (currently this is 4 bits, which means 1/16th of 1GHz = processor speeds down to 62.5 MHz are now supported using the same 'fast' method that Apple's stock kernels use). This means higher performance in multimedia/timer intensive apps (specially games!). No benchmark has been made yet, though the performance should be almost on par with vanilla kernels (only a couple of extra bit shifts have been added compared to the stock rtclock code).
Next up we plan to remove the copious amount of TSC related code (The 6-7 bus ratio paths!) and consolidate all of it into one "fail-proof" method: timing the TSC over a finite duration. This is already present in Voodoo 1.0 but we are planning to totally remove all other methods as it is just duplicating what your EFI bootloader has already done.
The challenge after this is going to be porting over the AMD-specific changes. kaitek, the developer who wrote most of the on-the-fly opcode patcher, 64bit support on AMD and other bits, is currently not reachable. Because of this porting this part of the code (which is 60% of the entire set of changes made in Voodoo) will be quite difficult, specially because Apple have changed a lot of the same parts in the kernel that we did.
There is currently no ETA for the final release, however alpha testing of at least the Intel builds should start in first week of June. Note that we will be doing alpha tests iteratively (ie. Intel users test their parts first before we proceed with AMD testing) to make it easy to isolate and track down problems.
Further updates will be made available in due time, as and when we have more to report on. Please direct queries directly via email to voodoo [AT] mercurysquad.com (but no questions about the timeline as we are ourselves unsure of it).
Cheers, mercurysquad
|
💡 Deploy cloud instances seamlessly on DigitalOcean. Free credits ($100) for InfMac readers.
|
Linear Mode
