Cool VL Viewer forum

It will be v1.26.1.0, and an alpha release, but for those who doubted it could be done, here are some proofs of the contrary:

Snapshot of the mesh viewer with shadows enabled
Snapshot of the mesh viewer without shadows

Currently, I still have to solve some issues with crippled APR pools (leading to memory leaks) and fast timers (so far used a quick hack to avoid systematic crashes), plus some crash bugs with some rendering debug modes, but it's coming... FAST !

To make the story short, after loosing a couple of months trying to convince the v1 renderer to play nice the new llmath library (got loads of issues with objects culling, spatial partitions, etc...), I finally adopted the pragmatic approach and "bulldozered" the v2.6-mesh viewer renderer into the Cool VL Viewer v1.26.0.15 in just five days of (hard) work (three for porting the code, and two figuring out the bugs and fixing them)... Note that without the hard work that was done over the 16 previous releases of v1.26.0, which brought progressively the Snowglobe v1.5 code on par with v2 viewers' for the parts that the new Mesh code required, the port won't have been possible (so, I'm alas unable to just provide a patch for other v1 TPVs...).

Of course, the viewer will not be as fast as a v1 renderer + mesh could have been, but the screen shots show that it's not slow either. On my main system (Quad Core Q6600 @ 3.4GHz and GTX460-1Gb) I experience a 10% loss in FPS when compared to v1.26.0 (with same settings, and of course shadows off in both viewers, and without Mesh objects around), but people with lower end graphic cards might well see a larger gap, because of the many shaders the v2 renderer is using...

Anyway... Stay tuned !

Awesome!

This sounds great but I have a general question.
Is the 10% drop in fps because of the code being ported instead of being directly in a V2 system ?
What I mean by this is I run a dual core laptop,my old laptop had a dedicated nvidia gpu which performed great and I was worried about it handling mesh,so I loggeed into beta grid and discovered it ran just fine. This was on Kirstens viewer at the time and was told mesh renders better than comparable items built with sculpties and prims becue it's more efficient (I'm a layperson,I don't know the in's and outs)
My old laptop died and I'm on one with dedicated 512mb Ati HD5400, which on paper should perform better than my old nvidia but in fact doesn't.

It doesn't perform as well in SL anyway and a 10% drop could make a big difference in club locations.

(as an afterthought, I'm on the phoenix beta team; if you need any beta testing done,bug reports etc, for win or linux, I'm more than happy to help out)
Mondy

Not at all because of the porting (nothing sacrificed for the backport). In my experience the Cool VL Viewer v1.26.1 is still faster than v2.

The new llmath library indeed uses SSE2 to speed things up, but on the other hand, the rendering pipeline is much more complex, and in deferred rendering (dynamic shadows) mode, it uses a lot more shaders. So, yes, the new renderer is faster in some particular cases, but overall, it's a little slower.
Not sure how well it will compare with Kirsten's viewer (Kirsten is well known for doing a great job in optimizing the deferred renderer), but it should compare favourably with v2/3 official viewers.

As a rule of thumb, ATI cards will perform noticeably poorer than Nvidia for OpenGL applications, mainly because of their poor OpenGL support in their driver...

Everyone will be able to test the Linux version today (got it ready to upload), and even perhaps the Windows version (currently trying to convince VS2005 to link it properly, which involved rebuilding llvertexdecomposition and colladadom for VS2005, and to fix issues with missing boost libraries in the pre-built ones provided by LL...).

And released just now, both Linux and Windows versions available. See the Announcements forum.

For TPV developers, there is one particular diff file that they may wish to have a look at, the diff between v1.26.0.16 v1.26.0.31 and v1.26.1.0 v1.26.2.10, but without the spacing changes (I tend to be anal with spacing and formatting in sources, and often run a few formatting macros in Nedit when I amend a source file...). Grab it here ! You may also have a look at the various diffs I produced for each version after v1.26.0 was forked from v1.25.0.34, since they contain important changes that are needed for Mesh support... See also the release notes for each v1.26.0 release in the announcement forum to see what changes are mandatory.

is it just me, or do mesh attachments only load on lowest detail?

I finally found a freebie mesh attachment and could test it. Yes, it looks like they don't load fully, unless you rez them first inworld (on the floor)... Strange, because I could see other avatars wearing mesh attachments and they looked fine... That's one issue to add to the list.

Also, another issue that was just reported to me: when basic shaders are switched off, avatars get completely distorted every few seconds... Just keep them on for now.

Yep,that basic shader issue had me confused for a moment. I looked like Sonic the Hedgehog on steroids )

thanks for your help inworld with that Henri.

*huggs*

I see the same issues with Basic Shaders and with Mesh as mentioned above.

In addition to that Deferred Rendering doesn't work for me. I see one of two things happening depending on my window size:
>2048 Pixel: Immediate viewer crash
<2048 Pixel: Everything looks VERY distorted. Like it is rotated around some angle. And not even the same angle for all objects.

Graphic card is Nvidia 580. Happens with the older driver version 275.33 and with the current 280.26 on 64-bit Windows 7. I had the same problem with window-size and deferred rendering on some early versions of Kirsten's viewer. The current one works fine though. I faintly remember that 2048 may have been the maximum texture-size in earlier OpenGL versions. Not sure if this is the cause though. Just guessing here.

In regular, non-deferred mode the framerate is actually better than before. Probably because I'm now missing anti-aliasing . From what I've seen that is normal for all v2 viewers?

These are all graphic driver issues, some of which might have been worked around in newer renderer (the one in use in v1.26.1.0 is the v2.6-mesh renderer, so v3 may work around more glitches).

I don't have a 2048 pixels wide screen (!) to test the crash issue, but here, everything works like a charm (no crash, no distortion, and the anti-aliasing works just fine). I've got a GTX460-1Gb and I'm running drivers v285.03 under Linux.

Please, also note that you *must* let the application decide of what OpenGL settings it shall use; some drivers allow you to override some settings (anti-aliasing is one of them, but also Anisotropic filtering and "Texture Sharpening" (that last one causes fonts to look blurry when enabled)). Go to your graphic driver control panel, and uncheck all these options, also selecting "Use application settings" (or "Let application decide") everywhere it is available.
The same is true with ATI drivers, by the way: their "AI" option is known to cause havoc with OpenGL apps.