diff options
Diffstat (limited to 'indra/newview/lltexturefetch.cpp')
-rw-r--r-- | indra/newview/lltexturefetch.cpp | 74 |
1 files changed, 69 insertions, 5 deletions
diff --git a/indra/newview/lltexturefetch.cpp b/indra/newview/lltexturefetch.cpp index dddfed097d..65a40a5b3a 100644 --- a/indra/newview/lltexturefetch.cpp +++ b/indra/newview/lltexturefetch.cpp @@ -850,15 +850,14 @@ bool LLTextureFetchWorker::doWork(S32 param) { if(mCanUseHTTP) { - const S32 HTTP_QUEUE_MAX_SIZE = 8; // *TODO: Integrate this with llviewerthrottle // Note: LLViewerThrottle uses dynamic throttling which makes sense for UDP, // but probably not for Textures. // Set the throttle to the entire bandwidth, assuming UDP packets will get priority // when they are needed F32 max_bandwidth = mFetcher->mMaxBandwidth; - if ((mFetcher->getHTTPQueueSize() >= HTTP_QUEUE_MAX_SIZE) || - (mFetcher->getTextureBandwidth() > max_bandwidth)) + if (mFetcher->isHTTPThrottled(mDesiredSize) || + mFetcher->getTextureBandwidth() > max_bandwidth) { // Make normal priority and return (i.e. wait until there is room in the queue) setPriority(LLWorkerThread::PRIORITY_NORMAL | mWorkPriority); @@ -1250,7 +1249,7 @@ bool LLTextureFetchWorker::deleteOK() if ((haveWork() && // not ok to delete from these states - ((mState >= SEND_HTTP_REQ && mState <= WAIT_HTTP_REQ) || + ((mState == WAIT_HTTP_REQ) || (mState >= WRITE_TO_CACHE && mState <= WAIT_ON_WRITE)))) { delete_ok = false; @@ -1511,6 +1510,11 @@ LLTextureFetch::LLTextureFetch(LLTextureCache* cache, LLImageDecodeThread* image { mMaxBandwidth = gSavedSettings.getF32("ThrottleBandwidthKBPS"); mTextureInfo.setUpLogging(gSavedSettings.getBOOL("LogTextureDownloadsToViewerLog"), gSavedSettings.getBOOL("LogTextureDownloadsToSimulator"), gSavedSettings.getU32("TextureLoggingThreshold")); + + for(S32 i = 0 ; i < TOTAL_TEXTURE_TYPES; i++) + { + mHTTPThrottleFlag[i] = FALSE ; + } } LLTextureFetch::~LLTextureFetch() @@ -1658,6 +1662,65 @@ void LLTextureFetch::removeFromHTTPQueue(const LLUUID& id) mHTTPTextureQueue.erase(id); } +void LLTextureFetch::clearHTTPThrottleFlag() +{ + static const F32 WAIT_TIME = 0.3f ; //seconds. + static LLFrameTimer timer ; + + if(timer.getElapsedTimeF32() < WAIT_TIME) //wait for WAIT_TIME + { + return ; + } + timer.reset() ; + + LLMutexLock lock(&mNetworkQueueMutex); + for(S32 i = 0 ; i < TOTAL_TEXTURE_TYPES; i++)//reset the http throttle flags. + { + mHTTPThrottleFlag[i] = FALSE ; + } +} + +//check if need to throttle this fetching request. +//rule: if a request can not be inserted into the http queue due to a full queue, +// block all future insertions of requests with larger fetching size requirement. +//because: +// later insertions are usually at lower priorities; and +// small textures need chance to be fetched. +bool LLTextureFetch::isHTTPThrottled(S32 requested_size) +{ + static const S32 SMALL_TEXTURE_MAX_SIZE = 64 * 64 * 4 ; + static const S32 MEDIUM_TEXTURE_MAX_SIZE = 256 * 256 * 4 ; + static const U32 MAX_HTTP_QUEUE_SIZE = 8 ; + + //determine the class of the texture: SMALL, MEDIUM, or LARGE. + S32 type = LARGE_TEXTURE ; + if(requested_size <= SMALL_TEXTURE_MAX_SIZE) + { + type = SMALL_TEXTURE ; + } + else if(requested_size <= MEDIUM_TEXTURE_MAX_SIZE) + { + type = MEDIUM_TEXTURE ; + } + + LLMutexLock lock(&mNetworkQueueMutex); + + if(mHTTPTextureQueue.size() >= MAX_HTTP_QUEUE_SIZE)//if the http queue is full. + { + if(!mHTTPThrottleFlag[type + 1]) + { + for(S32 i = type + 1 ; i < TOTAL_TEXTURE_TYPES; i++) //block all requests with fetching size larger than this request. + { + mHTTPThrottleFlag[i] = TRUE ; + } + } + + return true ; + } + + return mHTTPThrottleFlag[type] ; //true if this request can not be inserted to the http queue. +} + void LLTextureFetch::deleteRequest(const LLUUID& id, bool cancel) { lockQueue() ; @@ -1824,7 +1887,8 @@ S32 LLTextureFetch::update(U32 max_time_ms) lldebugs << "processed: " << processed << " messages." << llendl; } } - + clearHTTPThrottleFlag(); + return res; } |