Author |
Message |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here is a new release of the Cool VL Viewer, available now from http://sldev.free.fr/: This is a maintenance release with significant bug fixes (especially with regards to MFA) and all users should update. Cool VL Viewer v1.28.2.61 (stable branch): New in this release: - Fixed (and could at last test) the MFA login procedure (one crash bug fixed, a MFA hash retention bug fixed and a double notification annoyance fixed). The Cool VL Viewer is now MFA-compatible; people already using MFA may force-enable it in the viewer during the grace period by setting the "UseMFAinSL" debug setting to TRUE; this setting will be auto-enabled whenever needed after the grace period will be over.
- Backported from LL's performance viewer a fix to physics shapes info display rendering and re-enabled the latter ("Advanced" -> "Rendering" -> "Info displays" -> "Physics shapes") in the code.
- Backported from LL's viewer "apply explicitly given physics shape only if the usual rules do not", "handle sub-mesh case of reusable physics model" and "only supply a default physics shape if it is named default_physics_shape" changes to the mesh upload floater.
- Made the mesh LOD validity checks less stringent (allowing NaNs) during the mesh uploader LOD optimization phase, to avoid abortions on error during that phase and the frustrating auto-closing of the mesh upload floater.
- Improved the no-mod attachments handling in the #RLV/ sub-folders: the auto-renaming of second level sub-folders containing such attachments is no more performed unless the said sub-folders are pristine (i.e. bearing the default "New Folder" name): instead, a new folder is created (named after the attachment point) and the attached object moved into it, avoiding confusion and folders tree unexpected changes for users not following Marine's folder placement recipe for such attachments.
- Added highlighting and tool tip support in the script editor for new "OBJECT_ACCOUNT_LEVEL" LSL constant.
- Backported from Alchemy viewer a fix to a potential issue in some OpenSim grids inventory default library loading (i.e. in grids where the library would not be bearing the owner UUID of Alexandria Linden).
- Linux only: added a "LL_NO_PUSEAUDIO" export to the cool_vl_viewer.conf file, to allow working around a bug in FMOD Studio which now (this was not the case in the past) lamentably crashes whenever a PulseAudio connection attempt is tried on a system without a running PulseAudio daemon.
- Updated the Lua library to v5.4.4.
- Code cleanup and minor optimizations.
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-03-12 10:43:01 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here is a new release of the Cool VL Viewer, available now from http://sldev.free.fr/: This is a maintenance/bugfix release and a recommended update. Cool VL Viewer v1.28.2.62 (stable branch): New in this release: - Got rid of the fragile (natively not thread-safe and therefore leaky, racy and crashy), slow, and totally superfluous LLAPRFile class (and associated LLVolatileAPRPool class which used mutexes in an attempt to make LLAPRFile thread safe, attempt largely ruined after the APR mutexes have been replaced with standard mutexes to solve circular APR cleanup/re-init issues and related crashes on shutdown), and rewrote LLFile to replace it. All file operations in the viewer are now performed using standard (and safe) system calls.
- Fixed a potential crash bug in LLGLManager::printGLInfoString().
- Fixed a potential crash bug in LLModel::sortVolumeFacesByMaterialName().
- Fixed a potential crash bug in LLVertexBuffer::drawArrays().
- Fixed potential crash bugs in LLViewerObjectList::updateApparentAngles().
- Lifted the experimental status for io_uring/IOCP (now always built too), and moved the enabling toggle from the "Advanced" menu into the "Preferences" floater, "Cool features" tab, "Misc." sub-tab (still off by default).
- Removed the now useless "RunMultipleThreads" debug setting.
- Windows only: fixed the issues with locked marker files not being readable by second (and further) viewer instances and preventing proper detection of sibling viewers to protect the debug settings and caches.
- Updated FMOD Studio to v2.02.06.
- Synced parallel_hashmap with latest upstream changes.
- Code cleanup.
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-03-19 10:19:09 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here is a new release of the Cool VL Viewer, available now from http://sldev.free.fr/: This is a maintenance/bugfix release and a recommended update. Cool VL Viewer v1.28.2.63 (stable branch): New in this release: - Fixed an uncommon crash on exit introduced in last week's release (race condition in a thread timer destruction).
- Fixed a rendering bug (introduced in v1.28.2.47: I am surprised no one saw it so far), that caused flashing water reflections in ALM mode, in certain conditions.
- Backported from LL's viewer a fix to wrong light attenuation for alphas and materials in deferred EE shaders.
- Backported from LL's viewer a fix to "llStopSound() does not stop sound if it is out of cutoff radius".
- Backported from LL's performance viewer the increase of VBO size from 512KB to 4MB, and removed my (less generous) work-around (mesh geometry VBO bytes discount) to avoid disappearing groups of mesh objects at close distance and high LOD.
- Windows only: backported from LL's performance viewer the attempt to create a GL context starting with OpenGL v4.6 (instead of v4.2 only).
- Windows only: fixed the failure to open "Help" -> "Lua scripting help...".
- Code cleanup and minor optimizations.
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-03-26 08:49:48 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here are two new releases of the Cool VL Viewer, available now from http://sldev.free.fr/: See the latest roadmap about the new experimental v1.29.0 viewer. Cool VL Viewer v1.28.2.64 (stable branch): This is a maintenance/bugfix release and a recommended update. New in this release: - Fixed the occurrence of rare crash on viewer exit during LLTextureCache shutdown.
- Use a safe hash-map container type to store skin infos, since we use pointers on these and faster but non-safe variants can cause pointers invalidation (crash !).
- Backported from LL's performance viewer a fix to rigged mesh bounding boxes, and made it so it does not break our workaround for bad LODs selection by the renderer during rigged mesh editing.
- Backported from LL's viewer a revert to "remove ability to set zero triangle limit" (i.e. allow zero again).
- Backported from LL's viewer a fix to the mesh upload floater when using the error threshold criteria with the meshoptimizer-based simplification methods (a 100 error factor, since the UI shows 0 to 100% while meshoptimizer takes 0.0 to 1.0 !).
- Backported from LL's viewer the displaying of rigged mesh attachments in the thumbnail previews of the "Appearance" edit floater. I made this configurable (allowing to revert to the mesh-less preview), via a new "PreviewAvatarWithRigged" debug setting, in case you need it the old way.
- Re-implemented my "mesh vertex discount" workaround (that I removed in last release) for disappearing group of objects with high mesh LOD, because no, LL's trick (larger VBO) was not enough to prevent such glitches.
- Backported from LL's viewer "show particle age info" in debug messages.
- Code cleanup, with removal of dead code, and minor optimizations.
Cool VL Viewer v1.29.0.0 (new experimental branch): This release is a first step, with "just" the renderer brought on par with LL's performance viewer. A second step will involve the backport (for Windows and macOS), and the implementation (for Linux) of the threaded OpenGL image creation (which, from what I saw under Windows with LL's performance viewer, will bring a very nice frame rates improvement, larger than the one brought by the new renderer alone, at least for NVIDIA's GPUs). Same as v1.28.2.64, plus: - Entirely gutted out the Windlight renderer and shaders (*). NOTE: while I implemented a quick fix for Windlight estate/region time and Lightshare support in OpenSim, I did not test them well, and they might be broken: please report any such breakage (when compared with v1.28.2 in Windlight rendering mode).
- Backported the new "performance viewer" shaders and renderer code. Removed various tweaks used to fix the old EE renderer bugs/glitches (e.g. water culling got at last fixed by LL and seems to be working fine without them now). This practically re-syncs the Cool VL Viewer renderer with LL's (but the pertinent fixes to the shaders and render pipeline have been kept).
- A few optimizations to LL's code.
(*) It would have been too much work to adapt the new "performance viewer" changes to them, and the Extended Environment renderer with these changes implemented is now faster than Windlight's anyway, at least when there is actually something to render, since its overhead is still larger than Windlight's (seen in my skybox, for example, where I get 700fps instead of 880... But so large a fps rate is useless anyway). Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-04-02 10:10:41 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here are two new releases of the Cool VL Viewer, available now from http://sldev.free.fr/: These are maintenance/bugfix releases and recommended updates. Cool VL Viewer v1.28.2.65 (stable branch): New in this release: - Backported from LL's viewer "ignore getIsAlphaMask() on rigged faces", i.e. disable rigged faces alpha-masking, which was bogus.
- Fixed the visually muted avatars jelly-dolls sometimes not rendering at all.
- Worked around the problem with GLOD crashing when the core GL profile is used, by automatically falling back to meshoptimizer methods in the model upload floater. Note: this is glitchy (you will have to reload the mesh file in the high LOD line of the floater after it opens); the real fix will be to use Nicky Dasmijn's patched GLOD (I'll have to work on this later).
- Made raw image stats accounting thread-safe (fixes potential bogus stats and prepares for threaded GL image creation/deletion in v1.29.0).
- Added a setting (defaulting to disabled) to support large avatar texture bakes in OpenSim grids (1024x1024 pixels per texture instead of 512x512), which was accidentally introduced and always enabled in last week's releases. This is a per-grid (and actually per-avatar account) setting that can be enabled via the new "Advanced" -> "Character" -> "Use large bakes (after restart)" menu entry. Be aware that this feature multiplies your avatar's bake time by a factor of 4 for everyone (4 times more bytes to transfer between the viewer and the sim server)...
- Removed the code and related settings ("RenderSync*") for synchronous render occlusions (the corresponding code was actually never executed).
- Added a "RenderBatchGlow" setting (defaulting to TRUE), to allow to disable glow textures batching (not really useful, but I was attempting to catch a rare render glitch with bogus glow faces, and wanted to see if it was related with textures batches; see below).
- Rationalized and made coherent over the render pipeline the flagging of glowing faces (they were sometimes considered as glowing above a 0 glow value or sometimes only above 1/255).
- Removed support for building the viewer with deprecated/antiquated CEF versions: the minimum version is now CEF 91 (i.e. LL's current build of CEF, which is itself too old already).
- Code cleanup and minor optimizations.
Cool VL Viewer v1.29.0.1 ( experimental branch): Same changes as in v1.28.2.65, plus: - Fixed the (badly) broken shadows rendering seen in v1.29.0.0.
- Backported from LL's "performance viewer" the threaded GL image creation/destruction feature, and implemented Linux support for it (with fixes). Added a build option to enable that code (only for Linux and macOS for now: the Windows backport is not yet ready); be aware however that it still is very glitchy...
- Adapted to the new renderer and re-enabled the @camdrawmin/@camdrawmax (vision limiting spheres) RestrainedLove commands that went missing in v1.29.0.0.
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-04-09 09:59:47 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here are two new releases of the Cool VL Viewer, available now from http://sldev.free.fr/: They fix crash bugs and all users should update. Cool VL Viewer v1.28.2.66 (stable branch): New in this release: - Fixed a rare crash bug, introduced in v1.28.2.64. See this thread for details.
- Fixed potential crash bugs in LLPipeline::stateSort().
- Backported from LL's viewer "fix a gesture upload crash".
- Backported from LL's viewer a fix to "viewer crashes after changing the environment settings".
- Gutted out the old, deprecated "Memory safety checks" features and associated code (which was totally bogus for modern 64 bits systems with over 4GB of RAM). Only kept the VRAM checks (now always performed on systems implementing the corresponding GL calls), but also made it possible to disable them (should you encounter issues with them) via a new "Disable VRAM checks" option in the "Preferences" floater, "Graphics" tab, "GPU/GL features" sub-tab (this option is not shown since not needed, when the driver does not support these checks).
- Improved threads core affinity setting by pushing the HTTP threads away from the main (renderer) thread cores.
- Windows only: implemented thread affinity support for Windows (code contribution from Kathrine Jansma).
- Fixed a bogus assert in LLImageRaw::clear(), which broke (always crashed) debug builds.
- Code cleanup and minor optimizations.
Cool VL Viewer v1.29.0.2 ( experimental branch): Same changes as in v1.28.2.66, plus: - Fixed the totally broken bump map textures rendering in ALM mode seen in v1.29.0.1 and v1.29.0.0.
- Made LLImageGLThread multi-threaded (instead of mono-threaded in LL's code), so to cope with the much higher throughput of the Cool VL Viewer multi-threaded image decoder.
- Added a limit to the number of queued GL image creations per frame with automatic spillover to the main thread should this limit be reached, to prevent queue trashing. Added a new "GLWorkerQueueSize" setting to allow adjusting this limit (this should not be needed at all: if you see trashing happening often, just increase the number of GL threads; see below).
- Added a line in the texture console to report the image creations number (immediate creations, threaded creations and for the latter, between parenthesis, the number of creations still in queue).
- Fixed three bad bugs in the threaded GL image creation/destruction feature and enabled the threaded worker code compilation. A new setting (in the "Preferences" floater, "Graphics" tab, "GPU/GL features" sub-tab) allows to enable the threaded GL worker (it is off by default). Note that there are still known (either potential or harmless) glitches with this code: please read this message about them.
- Windows only: backported support for the threaded GL image creation/destruction feature. (*)
(*) Note: I for now excluded the new threaded callbacks code for keyboard/mouse/window events that is also part (albeit not related with GL threading) of LL's "performance viewer", since it breaks window resizing/positioning and, much worst, some essential mouse pointer operations; I will need to work on this code later on (it does not really bring anything significant, performances-wise, anyway). Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-04-16 09:40:47 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here are two new releases of the Cool VL Viewer, available now from http://sldev.free.fr/: They are maintenance and bugfix releases. Cool VL Viewer v1.28.2.67 (stable branch): New in this release: - Backported from LL's viewer "fix triangle mesh overlays being generated twice when avatar animation is frozen".
- Backported from LL's viewer "use proper capability for setting pathfinding attributes".
- Backported from LL's viewer various performance improvements in octrees code.
- Backported from Alchemy a fix to the mesh vertex cache optimization method version based on the meshoptimizer library (Rye Mutt's code). Since this fix does make this meshoptimizer-based version functional, made it the default method and provided entries in the "Advanced" -> "Rendering" menu to switch between methods (or off). Please do read this message and provide feedback if necessary.
- Added a new "TextureFetchTricklingLevel" debug setting representing the number of texture fetches below which the amount of updated textures per frame is reduced to improve the frame rate once the full scene rezzing occurred.
- Replaced the GLOD library with a modified version (modifications by Nicky Dasmijn) that does not need fixed OpenGL functions to work (making it compatible with a usage under the core GL profile context), and modified the vertex buffer management code and mesh upload floater code accordingly.
- Updated the nghttp2 library to v1.47.0, the OpenSSL v1.1 library to v1.1.1n and recompiled libcurl v7.64.1 against them (note: these libraries are not used in official builds since they cause HTTP pipelining failures with bogus textures and meshes downloads).
- Code cleanup and some optimizations (much faster llformat(), especially).
Cool VL Viewer v1.29.0.3 ( experimental branch): Same changes as in v1.28.2.67, plus: - Fixed a crash in the worker thread on resolution changes (i.e. now properly rebuild the GL contexts on window recreation). As a side effect/benefit, you can now change the number of GL worker threads without any need to restart the viewer to get that change taken into account.
- Fixed crashes on exit of the worker thread pool, and properly destroy image GL thread objects, close the worker thread and cleanup everything.
- Worker thread code cleanup and minor optimizations.
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-04-23 09:00:07 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here are two new releases of the Cool VL Viewer, available now from http://sldev.free.fr/: They are maintenance and bugfix releases, and recommended updates. Cool VL Viewer v1.28.2.68 (stable branch): New in this release: - Backported from LL's viewer "do not do software skinning/octree building of a face during picking unless eligible for narrow-phase intersect check" and "only calculate the octree for a skinned mesh if intersection falls inside an on-the-fly calculated bounding box".
- Backported from LL's viewer "fix for bad lighting in upload preview render".
- Backported from LL's viewer proper accounting for any moved spatial bridge in avatar bounding box updates (I already implemented the crash fix part of this code in v1.28.2.64).
- Backported from LL's viewer a fix to "unable to move an avatar after fall on the ground" (problem with a wrongly forever-looping "fall on ground" female anim).
- Backported from LL's viewer "do not try initializing shaders if requirements are not met or if window does not exist" and, for Windows only "warn user before closing window" when failing to create the viewer window or OpenGL context.
- Backported from LL's viewer a putative fix for a potential crash on exit (*) during octrees destruction. (*) Yes, I really doubt that this fix makes any difference, and I never saw this crash happening myself...
- Fixed some alpha faces subtle render glitches by rationalizing and making coherent over the render pipeline the flagging of transparent or opaque faces (they were sometimes considered as transparent at 0.0 alpha channel value, or below 0.001, and opaque over 0.99 or 0.999). Do not push any more transparent alpha rigged faces to the invisible render batch (which causes underlying avatar body mesh to become invisible, something that should only be affected by Alpha wearables), but simply skip their rendering.
- When a face is transparent but has glow, do not optimize it out and always push it to the render pipeline (backport and improvement of Beq Janus' fix for render glitches with glowing transparent faces in Firestorm).
- Do not lower the "texture fetch boost factor" any more below 1.0 whenever textures are being deleted from the texture list (to avoid a slower freeing of the unused textures memory).
- More "elevenification" of the code by replacing all boost library calls dealing with shared and weak pointers with the C++11 standard library equivalents (generalization over the whole code of a very partial/timid change by LL in their own code).
- Got rid of LL's bogus algorithm/code for the mesh vertex cache optimization, and only kept the meshoptimizer library based method.
- Added syntax highlighting and tool tips support in the script editing floater for the new llGetVisualParams() LSL function and OBJECT_* and PARCEL_DETAILS_* LSL constants.
- Synced parallel_hashmap with latest upstream changes.
- Code cleanup and minor optimizations.
Cool VL Viewer v1.29.0.4 ( experimental branch): Same changes as in v1.28.2.68, plus: - Removed an irrelevant check on media textures creation that caused a crash due to the new way of threading these creations. See this message for details.
- Backported from LL's viewer "for real this time, maintain legacy draw order of alpha rigged meshes, batching be damned" (sic).
- Backported from LL's viewer the overriding of the rigged meshes bounding box with the avatar's when calculating the surface area (fixes blurry textures seen on some large rigged mesh avatars).
- Backported from LL's viewer "make Multitexture a requirement, if it is not present, we won't be able to run".
- GL image worker thread code cleanup.
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-04-30 09:53:10 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here are two new releases of the Cool VL Viewer, available now from http://sldev.free.fr/: They are maintenance and bugfix releases, with very significant fps rate gains: all users should update. Cool VL Viewer v1.28.2.69 (stable branch): New in this release: - Got rid of a (legacy v1 code) glFinish() call in the GL buffer swapping method of LLWindowSDL/LLWindowWin32 (macOS was not affected). This one-liner (!) change yields very large fps rate gains; it is a wonder that the Cool VL Viewer remained one of the fastest viewers all these years with this huge handicap at OpenGL level (it was a bit like trekking uphill with a 50kg backpack). A very nice secondary effect is that the OpenGL core profile is now much faster for NVIDIA GPUs than the compatibility profile, yielding massive fps rate gains when compared to former releases (up to +100%, depending on the scene, at least when there is actually something to render, since in a sky box it shows a slightly higher overhead in buffer swapping). As a result, the core GL profile setting is now exposed in the "Preferences" floater, "Graphics" tab, "GPU/GL features" sub-tab (and defaults to enabled for NVIDIA when applying the recommended settings); sadly, AMD and Intel GPUs/drivers do not seem to be benefitting from it... Please do read the tool tip for the "Enable core GL profile" check box.
- Backported from LL's viewer "fix 'empty leaf' crashes" in octrees.
- Backported from LL's viewer "speed up loading of CEF media plugins by removing follow-redirects for MIME type lookup".
- Backported from Alchemy (Rye Mutt's code) improvements in fonts kerning (characters spacing).
- Adopted std::mutex instead of boost::mutex for LLMutex ("elevenification" or the code).
- Properly release VAO buffers on GL shutdown.
- Do not attempt to restart GL while in core GL profile whenever a graphics settings is modified (e.g. anisotropic filtering) and requires a screen resolution change (this leads to nasty render glitches, such as missing terrain); instead, wait for a viewer restart to apply the changes.
- Updated Tracy to v0.8.1.
- Code cleanup and minor optimizations.
Cool VL Viewer v1.29.0.5 ( experimental branch): NOTE: this branch is nearing the promotion to the next stable branch; if you did not yet test it, do it now and report any issue you would encounter when compared with v1.28.2. Same changes as in v1.28.2.69, plus: - Fixed a crash bug seen with core GL profile enabled, due to the improper sharing of VAO buffers between GL threads. Kudos and thanks to Kathrine Jansma for finding that bug.
- Properly set a VAO for each shared context GL thread when actually needed.
- Fixed the slow down in peak frame rates encountered when enabling GL threads (removal of silly double-conversions of weak pointers into shared pointers, and implemented an atomic empty() method in LLThreadSafeQueue and LLWorkQueue to test for pending work before attempting to take a costly lock on the queue).
- Do not any more force GL threads off when their setting is set to "auto" (-1) for Intel iGPUs; contrarily to what LL says, shared GL context threads seem to work just fine (well, with my fixes, since LL's code is bogus anyway), at least under Linux with Mesa drivers on a Pentium Silver J5005 (UHD Graphics 605): feel free to report your success or failure in the experimental viewer feedback thread.
- Backported from LL's viewer a "fix for some particle systems with custom blend functions not rendering".
- Adopted (fibers-aware) boost::fibers::timed_mutex instead of std::timed_mutex for LLThreadSafeQueue (like in LL's performance viewer).
- Now manually create and destroy the "mainloop" work queue at the right time in the session, instead of letting the C++ compiler decide.
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-05-07 09:25:23 |
|
![](./styles/getaway/theme/images/spacer.gif) |
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5523
|
![Reply with quote Reply with quote](./styles/getaway/imageset/en/icon_post_quote.gif)
Here are two new releases of the Cool VL Viewer, available now from http://sldev.free.fr/: They are maintenance and bugfix releases and recommended update. Cool VL Viewer v1.28.2.70 (stable branch): New in this release: - Fixed a potential crash bug in libOpenJPEG, detected by gcc v12.1.
- Fixed an inverted test in LLFloaterModelPreview that could cause an improper loading of mesh models. Bug detected by gcc v12.1.
- Fixed various other potential bugs stemming from issues spotted by gcc v12.1 (dangling pointer in LLDirIterator, incorrect zeroing of a buffer in LLMD5, possible out of bound indices in LLStdStringTable and LLViewerTexture, double-free in LLImageGL).
- Fixed memory leaks in LLVorbisDecodeState encountered when an audio decode failed. Cleaned-up LLAudioDecodeMgr's and LLAudioEngine's code and re-synced it with LL's viewer.
- Made the objects visibility refresh feature faster (incurs a shorter "hiccup" after sim border crossings). Removed some unused code in it and the associated "OcclusionPauseDuration" debug setting.
- When the GL core profile is activated, do no allow the live changing of the renderer mode (WL or EE) but wait until restart; this avoids bogus rendering after the change.
- Updated Tracy to v0.8.2.
- Code cleanup.
Cool VL Viewer v1.29.0.6 ( experimental branch): NOTE: this branch is nearing the promotion to the next stable branch; if you did not yet test it, do it now and report any issue you would encounter when compared with v1.28.2. Same changes as in v1.28.2.70, plus: - Made LLImageGL and LLGLTexture LLThreadSafeRefCount'ed to plug potential thread-unsafety issues with GL worker threads.
- Made LLVertexBuffer thread-safe since it is indirectly used by the GL worker threads.
- Removed the useless workarounds for potential (but never seen) bogus GL textures and their associated debug settings ("GLWorkerUseFence" and "GLWorkerUseForBumpMaps").
- Backported from LL's viewer SL-16182 branch the threading via work queues of Vorbis audio decoding. Improved it by removing the use of LLLFSThread and using the simpler/faster LLFile instead for decoded file writes (that are now natively threaded and therefore do not need the use of yet another thread).
Enjoy ! ![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
|
2022-05-14 08:42:10 |
|
|
Who is online |
Users browsing this forum: No registered users and 21 guests |
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot post attachments in this forum
|
|