sldev-traffic no 8 - THE "OCHO"
sldev discussions through April 20, 2007
One of the more involved and difficult-to-summarize developments in the past week, Rob Lanphier (Linden) reports further favorable results from the in-world bug triages. Recent triages are available at 2007-04-16 and 2007-04-13, with the next triage scheduled for Monday. Rob has listed steps for advance participation at Bug_triage.
A significant number of bugs have been tested, filtered, and imported, with the promise of many more if subsequent sessions go as well as the past two. Listmembers and Lindens alike have stepped in to help pound these out. Get involved if you can!
James Cook (Linden) did some profiling with Shark and found hot spots in a vector matrix transform, vector matrix rotation, and LLViewerJointMesh::updateGeometry(). He posted some patches that cut up to 1ms off of each frame on an unspecified Mac, asking for advice from people with SSE experience and postulating that these, as well as the matrix lerp(), could be further enhanced by vectorization. Callum Lerwick suggested enabling automatic vectorization, which lead to a discussion about how many platforms were still supported without streaming SIMD extensions -- the 1GHz Athlon Thunderbird falling within stated minimal requirements seems to be the proverbial fly in the ointment for SSE-only builds. James indicated that he'd research whether many of these were in use in order to make a decision about enabling SSE globally. There wasn't any answer on this yet, but Dzonatas Sol and James continued conversation along with some experiments showing potential further speedup.
Argent Stonecutter asked why automatic vectorization would be turned off on the Mac and didn't get an answer. Presumably all supported Macs have AltiVec instructions.
John Hurliman posted a list of 117 message packets that the viewer doesn't need to be aware of, suggesting that these might be used by back-end services exclusively. These may be of use to anyone writing his or her own client.
Source Drops Going Smoothly
Recent source drops have built much more smoothly on multiple platforms. If anyone has been hesitant to jump in for fear of time spent working past build issues, now is a fantastic time!
OpenSL for Solaris on Intel... SPARC - Almost
Dana Fagerstrom (Sun Microsystems) announced work on a Solaris build for x86/x64 and SPARC. Things seemed quite near working, save problems with Dana's graphics card not supporting a 32-bit framebuffer for testing the SPARC build. David Fries offered up a patch that should help with other bit depths, but there wasn't any reply yet. Once working, Dana hopes to spend some time profiling Solaris performance in DTrace.
John Hurliman noted some changes in the message protocol, of interest to anyone writing his or her own client. John noted that most seemed to increase precision on existing variables, and Joshua Bell (Linden) explained the remainder.
Jesse Malthus passed away this past Thursday. Jesse, also known by Jesse Nesbitt (on-list) and Jesse Higginbotham (meatspace) will be missed. Jesse was memorialized in-world, on-list, in most SL-related IRC channel topic lines, and even on multiple Second Life blogs and Twitter-blogs. Jason Giglio broke the sad news on-list, and Rob Lanphier linked to an article on Second Life Insider. Jesse was 17.
Several people have passed around a snapshot of Jesse in-world, wearing a rocket pack at 102km above a sim, far above us all and still climbing fast.