Thursday, August 8, 2013

Thank you for holding, please continue to hold

I'm trying to get 22.0.1 out but one of the security patches made the browser somewhat unstable and I'm trying to figure out which one. I've got it narrowed down to one of three. I apologize for this taking so long, so I'll throw in a couple of performance boosters stolen from 24-25 as a way of making amends for being tardy (these are very safe enhancements, and make gradients and CSS menus a bit snappier). ETA this weekend because I want to test it thoroughly before I make a release. Fortunately, I know it's something I did personally, because 22.0 itself is just fine, so it's just a matter of backing things out until it sticks.

BaselineCompiler is now to the point where it will generate the final inline cache code before terminating successfully on our agonizingly simple test case ("var i=0"). I haven't finished debugging that yet, but the light at the end of the tunnel is finally visible, if very, very distant. There seems to be some glitch with the JSContext * BaselineCompiler is passing and logging that I haven't figured out yet, but I've worked around that temporarily to get the meat of the work moving along. If I can get this passing tests by the first or second week of September -- which is doable, but a very big if -- then we might be able to get ESR24 up before ESR17 runs out. I would like this very trivial test case to be fully functional by this weekend, and then I will make a code dump for those who wish to collabourate. However, I am on standby call this weekend for my actual paying job, and 22.0.1 has priority above 24, so I have rather less time right now to work on this than I would like.

I've been so busy trying to get BaselineCompiler to lift off that I haven't really had time to work on the transition from Google Code to SourceForge. For the time being, we will continue to clear out old alpha/beta/test releases periodically to keep us below quota. Once we are out of space and out of things to clear out, or January 2014 comes and Google refuses to let us upload anything, then we start uploading to SourceForge. Old versions will remain here until Google kills that too someday. I really haven't had time to go through issues and manually move them over, let alone the wiki, so we're going to leave them where they are for the foreseeable future.

In the news category, IBM is looking to open up the POWER architecture a la ARM through a new group called the OpenPower Consortium, starting with the next-generation POWER8 series. Recall that PowerPC came out of the AIM alliance (that is, Apple, IBM and Motorola), back in the deep dark ages when dinosaurs roamed the earth and used Power Mac 7100s to hunt mammals. However, Freescale (Motorola "today") has since gone their own way with embedded PowerPC, not a market IBM has historically desired, and Apple is of course completely out of the POWER racket, so IBM is looking for more partners to keep Power Architecture viable. One of the names mentioned is, intriguingly, Google, suggesting that the Googleplex may be interested in making their own CPUs as well as their own servers. Another one is Nvidia, which will enhance its CUDA GPU parallel computing environment for Power Architecture to make it more attractive as a big-iron number crunching platform (and that's a market IBM does like being in). Server builder Tyan is presumably looking at customizing their servers and motherboard lines for POWER CPUs, though their press release is a little opaque. While existing Power licensees will still be doing what they're doing, it's an interesting move to get the big iron POWER chips a little more circulation: while IBM will maintain control of the ISA, like ARM does, OpenPower members will get to customize the hardware and add more system-on-a-chip functionality, and presumably do so at favourable licensing rates. And, hey, look how well ARM did. Time will tell.

No comments:

Post a Comment

Due to an increased frequency of spam, comments are now subject to moderation.