mirror of
https://github.com/overte-org/overte.git
synced 2025-04-20 01:24:03 +02:00
cr cleanup
This commit is contained in:
parent
2a746ec111
commit
08c7c486d9
1 changed files with 5 additions and 65 deletions
|
@ -168,45 +168,9 @@ void eraseVoxelTreeAndCleanupAgentVisitData() {
|
|||
}
|
||||
|
||||
|
||||
void voxelDistributeHelper(AgentList* agentList, AgentList::iterator& agent, VoxelAgentData* agentData, ViewFrustum& viewFrustum) {
|
||||
// A quick explanation of the strategy here. First, each time through, we ask ourselves, do we have voxels
|
||||
// that need to be sent? If not, we search for them, if we do, then we send them. We store our to be sent voxel sub trees
|
||||
// in a VoxelNodeBag on a per agent basis. The bag stores just pointers to the root node of the sub tree to be sent, so
|
||||
// it doesn't store much on a per agent basis.
|
||||
//
|
||||
// There could be multiple strategies at play for how we determine which voxels need to be sent. For example, at the
|
||||
// simplest level, we can just add the root node to this bag, and let the system send it. The other thing that we use
|
||||
// this bag for is, keeping track of voxels sub trees we wanted to send in the packet, but wouldn't fit into the current
|
||||
// packet because they were too big once encoded. So, as we run though this function multiple times, we start out with
|
||||
// voxel sub trees that we determined needed to be sent because they were in view, new, correct distance, etc. But as
|
||||
// we send those sub trees, if their child trees don't fit in a packet, we'll add those sub trees to this bag as well, and
|
||||
// next chance we get, we'll also send those needed sub trees.
|
||||
// If we don't have nodes already in our agent's node bag, then fill the node bag
|
||||
void voxelDistributor(AgentList* agentList, AgentList::iterator& agent, VoxelAgentData* agentData, ViewFrustum& viewFrustum) {
|
||||
// If the bag is empty, fill it...
|
||||
if (agentData->nodeBag.isEmpty()) {
|
||||
|
||||
// To get things started, we look for colored nodes. We could have also just started with the root node. In fact, if
|
||||
// you substitute this call with something as simple as agentData->nodeBag.insert(rootNod), you'll see almost the same
|
||||
// behavior on the client. The scene will appear.
|
||||
//
|
||||
// So why do we do this extra effort to look for colored nodes? It turns out that there are subtle differences between
|
||||
// how many bytes it takes to encode a tree based on how deep it is relative to the root node (which effects the octal
|
||||
// code size) vs how dense the tree is (which effects how many bits in the bitMask are being wasted, and maybe more
|
||||
// importantly, how many subtrees are also included). There is a break point where the more dense a tree is, it's more
|
||||
// efficient to encode the peers together with their empty parents. This would argue that we shouldn't search for these
|
||||
// sub trees, and we should instead encode the parent for dense scenes.
|
||||
//
|
||||
// But, there's another important side effect of dense trees related to out maximum packet size. Namely... if a tree
|
||||
// is very dense, then you can't fit as many branches in a single network packet. Because when we encode the parent and
|
||||
// children in a single packet, we must include the entire child branch (all the way down to our target LOD) before we
|
||||
// can include the siblings. Since dense trees take more space per ranch, we often end up only being able to encode a
|
||||
// single branch. This means on a per packet basis, the trees actually _are not_ dense. And sparse trees are shorter to
|
||||
// encode when we only include the child tree.
|
||||
//
|
||||
// Now, a quick explanation of maxSearchLevel: We will actually send the entire scene, multiple times for each search
|
||||
// level. We start at level 1, and we scan the scene for this level, then we increment to the next level until we've
|
||||
// sent the entire scene at it's deepest possible level. This means that clients will get an initial view of the scene
|
||||
// with chunky granularity and then finer and finer granularity until they've gotten the whole scene. Then we start
|
||||
// over to handle packet loss and changes in the scene.
|
||||
int maxLevelReached = randomTree.searchForColoredNodes(agentData->getMaxSearchLevel(), randomTree.rootNode,
|
||||
viewFrustum, agentData->nodeBag);
|
||||
agentData->setMaxLevelReached(maxLevelReached);
|
||||
|
@ -225,58 +189,34 @@ void voxelDistributeHelper(AgentList* agentList, AgentList::iterator& agent, Vox
|
|||
if (!agentData->nodeBag.isEmpty()) {
|
||||
static unsigned char tempOutputBuffer[MAX_VOXEL_PACKET_SIZE - 1]; // save on allocs by making this static
|
||||
int bytesWritten = 0;
|
||||
|
||||
// NOTE: we can assume the voxelPacket has already been set up with a "V"
|
||||
int packetsSentThisInterval = 0;
|
||||
|
||||
while (packetsSentThisInterval < PACKETS_PER_CLIENT_PER_INTERVAL) {
|
||||
if (!agentData->nodeBag.isEmpty()) {
|
||||
VoxelNode* subTree = agentData->nodeBag.extract();
|
||||
|
||||
// Only let this guy create at largest packets equal to the amount of space we have left in our final???
|
||||
// Or let it create the largest possible size (minus 1 for the "V")
|
||||
bytesWritten = randomTree.encodeTreeBitstream(agentData->getMaxSearchLevel(), subTree, viewFrustum,
|
||||
&tempOutputBuffer[0], MAX_VOXEL_PACKET_SIZE - 1,
|
||||
agentData->nodeBag);
|
||||
|
||||
// if we have room in our final packet, add this buffer to the final packet
|
||||
if (agentData->getAvailable() >= bytesWritten) {
|
||||
agentData->writeToPacket(&tempOutputBuffer[0], bytesWritten);
|
||||
} else {
|
||||
// otherwise "send" the packet because it's as full as we can make it for now
|
||||
agentList->getAgentSocket().send(agent->getActiveSocket(),
|
||||
agentData->getPacket(), agentData->getPacketLength());
|
||||
|
||||
// keep track that we sent it
|
||||
packetsSentThisInterval++;
|
||||
|
||||
// reset our finalOutputBuffer (keep the 'V')
|
||||
agentData->resetVoxelPacket();
|
||||
|
||||
// we also need to stick the last created partial packet in here!!
|
||||
agentData->writeToPacket(&tempOutputBuffer[0], bytesWritten);
|
||||
}
|
||||
} else {
|
||||
// we're here, if there are no more nodes in our bag waiting to be sent.
|
||||
// If we have a partial packet ready, then send it...
|
||||
if (agentData->isPacketWaiting()) {
|
||||
agentList->getAgentSocket().send(agent->getActiveSocket(),
|
||||
agentData->getPacket(), agentData->getPacketLength());
|
||||
|
||||
// reset our finalOutputBuffer (keep the 'V')
|
||||
agentData->resetVoxelPacket();
|
||||
|
||||
}
|
||||
|
||||
// and we're done now for this interval, because we know we have not nodes in our
|
||||
// bag, and we just sent the last waiting packet if we had one, so tell ourselves
|
||||
// we done for the interval
|
||||
packetsSentThisInterval = PACKETS_PER_CLIENT_PER_INTERVAL;
|
||||
packetsSentThisInterval = PACKETS_PER_CLIENT_PER_INTERVAL; // done for now, no nodes left
|
||||
}
|
||||
}
|
||||
|
||||
// Ok, so we're in the "send from our bag mode"... if during this last pass, we emptied our bag, then
|
||||
// we want to move to the next level.
|
||||
// if during this last pass, we emptied our bag, then we want to move to the next level.
|
||||
if (agentData->nodeBag.isEmpty()) {
|
||||
if (agentData->getMaxLevelReached() < agentData->getMaxSearchLevel()) {
|
||||
agentData->resetMaxSearchLevel();
|
||||
|
@ -314,7 +254,7 @@ void *distributeVoxelsToListeners(void *args) {
|
|||
|
||||
viewFrustum.calculate();
|
||||
|
||||
voxelDistributeHelper(agentList, agent, agentData, viewFrustum);
|
||||
voxelDistributor(agentList, agent, agentData, viewFrustum);
|
||||
}
|
||||
}
|
||||
|
||||
|
|
Loading…
Reference in a new issue