From 248427b8f956f3e879aeda9179b0479a6d778ab7 Mon Sep 17 00:00:00 2001 From: Rick Pasetto Date: Thu, 10 Dec 2009 14:21:48 -0800 Subject: PARTIAL DEV-43869: LLMediaDataClient now uses two queues This is a fairly major change that addresses the issue of an object with constantly-updating media. Before, that object would be put into our single queue and sorted to a particular spot, and since it continuously updates, it would "always be there". That means that nothing "behind" it would ever get serviced. This change introduces two queues for each MDC: one is the same "sorted" queue as before, and the other is unsorted, and "round-robins". New objects go into the sorted queue, objects whose media we already know about get put into the unsorted queue. The two queues are interleaved when serviced (one then the other is serviced -- if one is empty we try the other -- until they are both drained). The round-robin queue works a little differently: after an item is fetched from that queue (remember this would be an item we already know about), that request is marked and put back at the end of the queue. If that object gets a UDP update while in the queue, that mark is "cleared". When it gets to the front of the queue again, if it still marked, it is thrown away. If it is not marked, it is fetched, and again marked and put at the end. This makes the queue self-limiting in how big it can get. I have also made some other changes: - The sorting comparator now just delegates to the object for its "interest" calculation. A higher value = more interesting. LLVOVolume now uses its PixelArea for its "interest" calculation, which seems apparently better (the prior distance calculation was wrong anyway). - The score is cached before the sort operation is performed, so that it won't be expensive to sort - Now, the media version that is fetched is saved in the LLVOVolume, and we do not update if it is not newer (this is not very useful...yet.) - I've introduced hard limits (settable by debug settings) on the size of the queues. The sorted queue will be culled (after sort) to that count. NOTE: this will probably get removed in a later checkin, as I've already gotten feedback that this is not desirable - I've reorganized LLMediaDataClient so it makes more sense. - I've made the request object a little smaller, so the queue won't take up so much memory (more work could be done here) - Added a unit test for the two-queue case (though more tests are needed!) --- indra/newview/llvovolume.cpp | 74 ++++++++++++++++++++++++++++---------------- 1 file changed, 48 insertions(+), 26 deletions(-) (limited to 'indra/newview/llvovolume.cpp') diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index cf61994fea..ffadece615 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -92,7 +92,7 @@ static LLFastTimer::DeclareTimer FTM_GEN_VOLUME("Generate Volumes"); class LLMediaDataClientObjectImpl : public LLMediaDataClientObject { public: - LLMediaDataClientObjectImpl(LLVOVolume *obj) : mObject(obj) {} + LLMediaDataClientObjectImpl(LLVOVolume *obj, bool isNew) : mObject(obj), mNew(isNew) {} LLMediaDataClientObjectImpl() { mObject = NULL; } virtual U8 getMediaDataCount() const @@ -128,14 +128,14 @@ public: virtual bool hasMedia() const { return mObject->hasMedia(); } - virtual void updateObjectMediaData(LLSD const &data) - { mObject->updateObjectMediaData(data); } - - virtual F64 getDistanceFromAvatar() const - { return mObject->getRenderPosition().length(); } + virtual void updateObjectMediaData(LLSD const &data, const std::string &version_string) + { mObject->updateObjectMediaData(data, version_string); } - virtual F64 getTotalMediaInterest() const - { return mObject->getTotalMediaInterest(); } + virtual F64 getMediaInterest() const + { + F64 tmp = mObject->getTotalMediaInterest(); + return (tmp < 0.0) ? mObject->getPixelArea() : tmp; + } virtual std::string getCapabilityUrl(const std::string &name) const { return mObject->getRegion()->getCapability(name); } @@ -143,8 +143,15 @@ public: virtual bool isDead() const { return mObject->isDead(); } + virtual U32 getMediaVersion() const + { return LLTextureEntry::getVersionFromMediaVersionString(mObject->getMediaURL()); } + + virtual bool isNew() const + { return mNew; } + private: LLPointer mObject; + bool mNew; }; @@ -165,6 +172,7 @@ LLVOVolume::LLVOVolume(const LLUUID &id, const LLPCode pcode, LLViewerRegion *re mSpotLightPriority = 0.f; mMediaImplList.resize(getNumTEs()); + mLastFetchedMediaVersion = -1; } LLVOVolume::~LLVOVolume() @@ -190,7 +198,9 @@ void LLVOVolume::markDead() { if (!mDead) { - // TODO: tell LLMediaDataClient to remove this object from its queue + LLMediaDataClientObject::ptr_t obj = new LLMediaDataClientObjectImpl(const_cast(this), false); + sObjectMediaClient->removeFromQueue(obj); + sObjectMediaNavigateClient->removeFromQueue(obj); // Detach all media impls from this object for(U32 i = 0 ; i < mMediaImplList.size() ; i++) @@ -210,8 +220,12 @@ void LLVOVolume::initClass() const F32 queue_timer_delay = gSavedSettings.getF32("PrimMediaRequestQueueDelay"); const F32 retry_timer_delay = gSavedSettings.getF32("PrimMediaRetryTimerDelay"); const U32 max_retries = gSavedSettings.getU32("PrimMediaMaxRetries"); - sObjectMediaClient = new LLObjectMediaDataClient(queue_timer_delay, retry_timer_delay, max_retries); - sObjectMediaNavigateClient = new LLObjectMediaNavigateClient(queue_timer_delay, retry_timer_delay, max_retries); + const U32 max_sorted_queue_size = gSavedSettings.getU32("PrimMediaMaxSortedQueueSize"); + const U32 max_round_robin_queue_size = gSavedSettings.getU32("PrimMediaMaxRoundRobinQueueSize"); + sObjectMediaClient = new LLObjectMediaDataClient(queue_timer_delay, retry_timer_delay, max_retries, + max_sorted_queue_size, max_round_robin_queue_size); + sObjectMediaNavigateClient = new LLObjectMediaNavigateClient(queue_timer_delay, retry_timer_delay, + max_retries, max_sorted_queue_size, max_round_robin_queue_size); } // static @@ -406,7 +420,7 @@ U32 LLVOVolume::processUpdateMessage(LLMessageSystem *mesgsys, // If the media changed at all, request new media data LL_DEBUGS("MediaOnAPrim") << "Media update: " << getID() << ": retval=" << retval << " Media URL: " << ((mMedia) ? mMedia->mMediaURL : std::string("")) << LL_ENDL; - requestMediaDataUpdate(); + requestMediaDataUpdate(retval & MEDIA_FLAGS_CHANGED); } else { LL_INFOS("MediaOnAPrim") << "Ignoring media update for: " << getID() << " Media URL: " << @@ -1698,16 +1712,16 @@ LLVector3 LLVOVolume::getApproximateFaceNormal(U8 face_id) return result; } -void LLVOVolume::requestMediaDataUpdate() +void LLVOVolume::requestMediaDataUpdate(bool isNew) { - sObjectMediaClient->fetchMedia(new LLMediaDataClientObjectImpl(this)); + sObjectMediaClient->fetchMedia(new LLMediaDataClientObjectImpl(this, isNew)); } bool LLVOVolume::isMediaDataBeingFetched() const { // I know what I'm doing by const_casting this away: this is just // a wrapper class that is only going to do a lookup. - return sObjectMediaClient->isInQueue(new LLMediaDataClientObjectImpl(const_cast(this))); + return sObjectMediaClient->isInQueue(new LLMediaDataClientObjectImpl(const_cast(this), false)); } void LLVOVolume::cleanUpMediaImpls() @@ -1725,18 +1739,25 @@ void LLVOVolume::cleanUpMediaImpls() } } -void LLVOVolume::updateObjectMediaData(const LLSD &media_data_array) +void LLVOVolume::updateObjectMediaData(const LLSD &media_data_array, const std::string &media_version) { // media_data_array is an array of media entry maps + // media_version is the version string in the response. + U32 fetched_version = LLTextureEntry::getVersionFromMediaVersionString(media_version); - //llinfos << "updating:" << this->getID() << " " << ll_pretty_print_sd(media_data_array) << llendl; - - LLSD::array_const_iterator iter = media_data_array.beginArray(); - LLSD::array_const_iterator end = media_data_array.endArray(); - U8 texture_index = 0; - for (; iter != end; ++iter, ++texture_index) + // Only update it if it is newer! + if ( (S32)fetched_version > mLastFetchedMediaVersion) { - syncMediaData(texture_index, *iter, false/*merge*/, false/*ignore_agent*/); + mLastFetchedMediaVersion = fetched_version; + //llinfos << "updating:" << this->getID() << " " << ll_pretty_print_sd(media_data_array) << llendl; + + LLSD::array_const_iterator iter = media_data_array.beginArray(); + LLSD::array_const_iterator end = media_data_array.endArray(); + U8 texture_index = 0; + for (; iter != end; ++iter, ++texture_index) + { + syncMediaData(texture_index, *iter, false/*merge*/, false/*ignore_agent*/); + } } } @@ -1904,7 +1925,7 @@ void LLVOVolume::mediaNavigated(LLViewerMediaImpl *impl, LLPluginClassMedia* plu llinfos << "broadcasting navigate with URI " << new_location << llendl; - sObjectMediaNavigateClient->navigate(new LLMediaDataClientObjectImpl(this), face_index, new_location); + sObjectMediaNavigateClient->navigate(new LLMediaDataClientObjectImpl(this, false), face_index, new_location); } } @@ -1968,7 +1989,7 @@ void LLVOVolume::mediaEvent(LLViewerMediaImpl *impl, LLPluginClassMedia* plugin, void LLVOVolume::sendMediaDataUpdate() { - sObjectMediaClient->updateMedia(new LLMediaDataClientObjectImpl(this)); + sObjectMediaClient->updateMedia(new LLMediaDataClientObjectImpl(this, false)); } void LLVOVolume::removeMediaImpl(S32 texture_index) @@ -2063,7 +2084,7 @@ viewer_media_t LLVOVolume::getMediaImpl(U8 face_id) const F64 LLVOVolume::getTotalMediaInterest() const { - F64 interest = (F64)0.0; + F64 interest = (F64)-1.0; // means not interested; int i = 0; const int end = getNumTEs(); for ( ; i < end; ++i) @@ -2071,6 +2092,7 @@ F64 LLVOVolume::getTotalMediaInterest() const const viewer_media_t &impl = getMediaImpl(i); if (!impl.isNull()) { + if (interest == (F64)-1.0) interest = (F64)0.0; interest += impl->getInterest(); } } -- cgit v1.2.3 From 2d01d763d558a373b356ea445ddb3801e36ae457 Mon Sep 17 00:00:00 2001 From: Rick Pasetto Date: Thu, 10 Dec 2009 17:03:40 -0800 Subject: PARTIAL DEV-43869 - add "isInterestingEnough()" to the queue calculation This change bumps the queue sizes way up, because we think that the "isInterestingEnough()" call will prevent loading more media data than we think is necessary. Still need to implement it in LLVOVolume, though --- indra/newview/llvovolume.cpp | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'indra/newview/llvovolume.cpp') diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index ffadece615..b92d024ac9 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -136,6 +136,11 @@ public: F64 tmp = mObject->getTotalMediaInterest(); return (tmp < 0.0) ? mObject->getPixelArea() : tmp; } + virtual bool isInterestingEnough() const + { + // TODO: use performance manager to control this + return true; + } virtual std::string getCapabilityUrl(const std::string &name) const { return mObject->getRegion()->getCapability(name); } -- cgit v1.2.3 From 3089e1a17a1f5bac1487fa56796f2b52244886e9 Mon Sep 17 00:00:00 2001 From: "Nyx (Neal Orman)" Date: Fri, 11 Dec 2009 12:07:35 -0500 Subject: EXT-2718 avatar render cost higher in 2.0 than in 1.23 previous fix fixed the double-counting of texture costs. resulting ARC was ~99 points higher for most avatars. This patch makes the cost shoot up again, as ARC was improperly computed in 1.23 and before. This makes the cost for an avatar increase 10 points per prim instead of per-attachment, which is how we have documented it. Also used constants to eliminate magic numbers and increased ARC limit from 1024 to 2048. Will request feedback on change from BSI:STU Code reviewed by Bigpapi --HG-- branch : avatar-pipeline --- indra/newview/llvovolume.cpp | 29 +++++++++++++++++++++++++++-- 1 file changed, 27 insertions(+), 2 deletions(-) (limited to 'indra/newview/llvovolume.cpp') diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index cf61994fea..36d6d2c04b 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -2587,7 +2587,22 @@ const LLMatrix4 LLVOVolume::getRenderMatrix() const // children, and cost should only be increased for unique textures -Nyx U32 LLVOVolume::getRenderCost(std::set &textures) const { - U32 shame = 0; + // base cost of each prim should be 10 points + static const U32 ARC_PRIM_COST = 10; + // per-prim costs + static const U32 ARC_INVISI_COST = 1; + static const U32 ARC_SHINY_COST = 1; + static const U32 ARC_GLOW_COST = 1; + static const U32 ARC_FLEXI_COST = 8; + static const U32 ARC_PARTICLE_COST = 16; + static const U32 ARC_BUMP_COST = 4; + + // per-face costs + static const U32 ARC_PLANAR_COST = 1; + static const U32 ARC_ANIM_TEX_COST = 4; + static const U32 ARC_ALPHA_COST = 4; + + U32 shame = ARC_PRIM_COST; U32 invisi = 0; U32 shiny = 0; @@ -2663,7 +2678,17 @@ U32 LLVOVolume::getRenderCost(std::set &textures) const } } - shame += invisi + shiny + glow + alpha*4 + flexi*8 + animtex*4 + particles*16+bump*4+scale+planar; + + shame += invisi * ARC_INVISI_COST; + shame += shiny * ARC_SHINY_COST; + shame += glow * ARC_GLOW_COST; + shame += alpha * ARC_ALPHA_COST; + shame += flexi * ARC_FLEXI_COST; + shame += animtex * ARC_ANIM_TEX_COST; + shame += particles * ARC_PARTICLE_COST; + shame += bump * ARC_BUMP_COST; + shame += planar * ARC_PLANAR_COST; + shame += scale; LLViewerObject::const_child_list_t& child_list = getChildren(); for (LLViewerObject::child_list_t::const_iterator iter = child_list.begin(); -- cgit v1.2.3