I have several (potentially thousands) of scenes I would like to render in order to perform queries on what is drawn. The problem I am running into is that calls to glGetQueryObjectiv() are expensive, so I'd like to figure out a way to render several scenes in advance while I wait for the results of the queries to become available.
I have read a bit about Framebuffer Objects and Pixel Buffer Objects, but mostly in the context of saving to a file using glReadPixels() and I haven't been able to track down an example of using either of these objects in asynchronous queries. Is there any reason why the setup for performing glGetQueryObjectiv() would be different from glReadPixels() as in this example (e.g., should I use an FBO or a PBO)?
Note:
This is NOT for a graphics application. All I'm interested in is the result of the GL_SAMPLES_PASSED query (i.e. how many pixels were drawn?).
The specific application is estimating how much sunlight is striking a surface when other surfaces are casting shadows. If you are interested, you can read about it here.
Neither Framebuffer Objects nor (Pixel) Buffer Objects are the proper class of objects to use in an (asynchronous) query; GL has an entirely separate class of objects called Query Objects. You can feed the results of certain queries into Buffer Objects, but that is about as far as their relationship goes.
Basically the idea with asynchronous queries is that you put the query into the graphics pipeline and let some time pass before you try to read it back. This gives the GPU time to actually finish all of the commands in between the beginning and end of your query first. The alternative is a synchronous query, which means that your calling thread stops doing anything useful and waits for the GPU to finish everything necessary to complete your query.
The simplest way of implementing a query that does not force synchronization would be to start and finish your query as you would typically, but rather than reading the result immediately, in your application's main loop periodically check the status of GL_QUERY_RESULT_AVAILABLE for your query object. When this gives you GL_TRUE, it means you can read the results from the query object without disrupting the render pipeline. If you do try to get the results from a query object before it is in this state, you are going to force a synchronous query.
Related
I am making a simple STG engine with OpenGL (To be exact, with LWJGL3).In this game, there can be several different types of items(called bullet) in one frame, and each type can have 10-20 instances.I hope to find an efficient way to render it.
I have read some books about modern OpenGL and find a method called "Instanced Rendering", but it seems only to work with same instances.Should I use for-loop to draw all items directly for my case?
Another question is about memory.Should I create an VBO for each frame, since the number of items is always changing?
Not the easiest question to answer but I'll try my best anyways.
An important property of OpenGL is that the OpenGL context is always bound to a single thread. So every OpenGL-method has to be called within that thread. A common way of dealing with this is using Queuing.
Example:
We are using Model-View-Controller architecture.
We have 3 threads; One to read input, one to handle received messages and one to render the scene.
Here OpenGL context is bound to rendering thread.
The first thread receives a message "Add model to position x". First thread has no time to handle the message, because there might be another message coming right after and we don't want to delay it. So we just give this message for the second thread to handle by adding it to second thread's queue.
Second thread reads the message and performs the required tasks as far as it can before OpenGL context is required. Like reads the Wavefront (.obj)-file from the memory and creates arrays from the received data.
Our second thread then queues this data to our OpenGL thread to handle. OpenGL thread generates VBOs and VAO and stores the data in there.
Back to your question
OpenGL generated Objects stay in the context memory until they are manually deleted or the context is destroyed. So it works kind of like C, where you have to manually allocate memory and free it after it's no more used. So you should not create new Objects for each frame, but reuse the data that stays unchanged. Also when you have multiple objects that use the same model or texture, you should just load that model once and apply all object specific differences on shaders.
Example:
You have an environment with 10 rocks that all share the same rock model.
You load the data, store it in VBOs and attach those VBOs into a VAO. So now you have a VAO defining a rock.
You generate 10 rock entities that all have position, rotation and scale. When rendering, you first bind the shader, then bind the model and texture, then loop through the stone entities and for each stone entity you bind that entity's position, rotation and scale (usually stored in a transformationMatrix) and render.
bind shader
load values to shader's uniform variables that don't change between entities.
bind model and texture (as those stay the same for each rock)
for(each rock in rocks){
load values to shader's uniform variables that do change between each rock, like the transformation.
render
}
unbind shader
Note: You don't need to unbind/bind shader each frame if you only use one shader. Same goes for VAO's and every other OpenGL object as well. So the binding will also stay over each rendering cycle.
Hope this will help you when getting started. Altho I would recommend some tutorial that might have a bit more context to it.
I have read some books about modern OpenGL and find a method called
"Instanced Rendering", but it seems only to work with same
instances.Should I use for-loop to draw all items directly for my
case?
Another question is about memory.Should I create an VBO for each
frame, since the number of items is always changing?
These both depend on the amount of bullets you plan on having. If you think you will have less than a thousand bullets, you can almost certainly push all of them to a VBO each frame and upload and your end users will not notice. If you plan on some obscene amount, then don't do this.
I would say that you should write everything each frame because it's the simplest to do right now, and if you start noticing performance issues then you need to look into instancing or some other method. When you get to "later" you should be more comfortable with OpenGL and find out ways to optimize it that won't be over your head (not saying it is over your head right now, but more experience can only help make it less complex later on).
Culling bullets not on the screen either should be on your radar.
If you plan on having a ridiculous amount of bullets on screen, then you should say so and we can talk about more advanced methods, however my guess is that if you ever reach that limit on today's hardware then you have a large ambitious game with a zoomed out camera and a significant amount of entities on screen, or you are zoomed up and likely have a mess on your screen anyways.
20 objects is nothing. Your program will be plenty fast no matter how you draw them.
When you have 10000 objects, then you'll want to ask for an efficient way.
Until then, draw them whichever way is most convenient. This probably means a separate draw call per object.
I don't properly understand how to parallelize work on separate threads in Vulkan.
In order to begin issuing vkCmd*s, you need to begin a render pass. The call to begin render pass needs a reference to a framebuffer. However, vkAcquireNextImageKHR() is not guaranteed to return image indexes in a round robin way. So, in a triple-buffering setup, if the current image index is 0, I can't just bind framebuffer 1 and start issuing draw calls for the next frame, because the next call to vkAcquireNextImageKHR() might return image index 2.
What is a proper way to record commands without having to specify the framebuffer to use ahead of time?
You have one or more render passes that you want to execute per-frame. And each one has one or more subpasses, into which you want to pour work. So your main rendering thread will generate one or more secondary command buffers for those subpasses, and it will pass that sequence of secondary CBs off to the submission thread.
The submissions thread will create the primary CB that gets rendered. It begins/ends render passes, and into each subpass, it executes the secondary CB(s) created on the rendering thread for that particular subpass.
So each thread is creating its own command buffers. The submission thread is the one that deals with the VkFramebuffer object, since it begins the render passes. It also is the one that acquires the swapchain images and so forth. The render thread is the one making the secondary CBs that do all of the real work.
Yes, you'll still be doing some CB building on the submission thread, but it ought to be pretty minimalistic overall. This also serves to abstract away the details of the render targets from your rendering thread, so that code dealing with the swapchain can be localized to the submission thread. This gives you more flexibility.
For example, if you want to triple buffer, and the swapchain doesn't actually allow that, then your submission thread can create its own extra images, then copy from its internal images into the real swapchain. The rendering thread's code does not have to be disturbed at all to allow this.
You can use multiple threads to generate draw commands for the same renderpass using secondary command buffers. And you can generate work for different renderpasses in the same frame in parallel -- only the very last pass (usually a postprocess pass) depends on the specific swapchain image, all your shadow passes, gbuffer/shading/lighting passes, and all but the last postprocess pass don't. It's not required, but it's often a good idea to not even call vkAcquireNextImageKHR until you're ready to start generating the final renderpass, after you've already generated many of the prior passes.
First, to be clear:
In order to begin issuing vkCmd*s, you need to begin a render pass.
That is not necessarily true. In command buffers You can record multiple different commands, all of which begin with vkCmd. Only some of these commands need to recorded inside a render pass - the ones that are connected with drawing. There are some commands, which cannot be called inside a render pass (like for example dispatching compute shaders). But this is just a side note to sort things out.
Next thing - mentioned triple buffering. In Vulkan the way images are displayed depends on the supported present mode. Different hardware vendors, or even different driver versions, may offer different present modes, so on one hardware You may get present mode that is most similar to triple buffering (MAILBOX), but on other You may not get it. And present mode impacts the way presentation engine allows You to acquire images from a swapchain, and then displays them on screen. But as You noted, You cannot depend on the order of returned images, so You shouldn't design Your application to behave as if You always have the same behavior on all platforms.
But to answer Your question - the easiest, naive, way is to call vkAcquireNextImageKHR() at the beginning of a frame, record command buffers that use an image returned by it, submit those command buffers and present the image. You can create framebuffers on demand, just before You need to use it inside a command buffer: You create a framebuffer that uses appropriate image (the one associated with index returned by the vkAcquireNextImageKHR() function) and after command buffers are submitted and when they stop using it, You destroy it. Such behavior is presented in the Vulkan Cookbook: here and here.
More appropriate way would be to prepare framebuffers for all available swapchain images and take appropriate framebuffer during a frame. But You need to remember to recreate them when You recreate swapchain.
More advanced scenarios would postpone swapchain acquiring until it is really needed. vkAcquireNextImageKHR() function call may block Your application (wait until image is available) so it should be called as late as possible when You prepare a frame. That's why You should record command buffers that don't need to reference swapchain images first (for example those that render geometry into a G-buffer in deferred shading algorithms). After that when You want to display image on screen (like for example some postprocessing technique) You just take the approach describe above: acquire an image, prepare appropriate command buffer(s) and present the image.
You can also pre-record command buffers that reference particular swapchain images. If You know that the source of Your images will always be the same (like the mentioned G-buffer), You can have a set of command buffers that always perform some postprocess/copy-like operations from this data to all swapchain images - one command buffer per swapchain image. Then, during the frame, if all of Your data is set, You acquire an image, check which pre-recorded command buffer is appropriate and submit the one associated with acquired image.
There are multiple ways to achieve what You want, all of them depend on many factors - performance, platform, specific goal You want to achieve, type of operations You perform in Your application, synchronization mechanisms You implemented and many other things. You need to figure out what best suits You. But in the end - You need to reference a swapchain image in command buffers if You want to display image on screen. I'd suggest starting with the easiest option first and then, when You get used to it, You can improve Your implementation for higher performance, flexibility, easier code maintenance etc.
You can call vkAcquireNextImageKHR in any thread. As long as you make sure the access to the swapchain, semaphore and fence you pass to it is synchronized.
There is nothing else restricting you from calling it in any thread, including the recording thread.
You are also allowed to have multiple images acquired at a time. Assuming you have created enough. In other words acquiring the next image before you present the current one is allowed.
I'm wondering how e.g. graphic (/game) engines do their job with lot's of heterogeneous data while a customized simple rendering loop turns into a nightmare when you have some small changes.
Example:
First, let's say we have some blocks in our scene.
Graphic-Engine: create cubes and move them
Customized: create cube template for vertices, normals, etc. copy and translate them to the position and copy e.g. in a vbo. One glDraw* call does the job.
Second, some weird logic. We want block 1, 4, 7, ... to rotate on x-axis, 2, 5, 8, ... on y-axis and 3, 6, 9 on z-axis with a rotation speed linear to the camera distance.
Graphic-Engine: manipulating object's matrix and it works
Customized: (I think) per object glDraw* call with changing model-matrix uniform is not a good idea, so a translation matrix should be something like an attribute? I have to update them every frame.
Third, a block should disappear if the distance to the camera is lower than any const value Q.
Graphic-Engine: if (object.distance(camera) < Q) scene.drop(object);
Customized: (I think) our vbo is invalid and we have to recreate it?
Again to the very first sentence: it feels like engines do those manipulations for free, while we have to rethink how to provide and update data. And while we do so, the engine (might, but I actually don't know) say: 'update whatever you want, at least I'm going to send all matrizes'.
Another Example: What about a voxel-based world (e.g. Minecraft) where we only draw the visible surface, and we are able to throw a bomb and destroy many voxels. If the world's view data is in one huge buffer we only have one glDraw*-call but have to recreate the buffer every time then. If there are smaller chunks, we have many glDraw*-calls and also have to manipulate buffers, which are smaller.
So is it a good deal to send let's say 10MB of buffer update data instead of 2 gl*-calls with 1MB? How many updates are okay? Should a rendering loop deal with lazy updates?
I'm searching for a guide what a 60fps application should be able to update/draw per frame to get a feeling of what is possible. For my tests, every optimization try is another bottleneck.
And I don't want those tutorials which says: hey there is a new cool gl*Instance call which is super-fast, buuuuut you have to check if your gpu supports it. Well, I also rather consider this an optimization than a meaningful implementation at first.
Do you have any ideas, sources, best practices or rule of thumb how a rendering/updating routine best play together?
My questions are all nearly the same:
How many updates per frame are okay on today's hardware?
Can I lazy-load data to have it after a few frames, but without freezing my application
Do I have to do small updates and profile my loop if there are some microseconds left till next rendering?
Maybe I should implement a real-time profiler which gets a feeling over time, how expensive updates are and can determine the amount of updates per frame?
Thank you.
It's unclear how any of your questions relate to your "graphics engines" vs "customized" examples. All the updates you do with a "graphics engines" are translated to those OpenGL calls in the end.
In brief:
How many updates per frame are okay on today's hardware?
Today's PCIe bandwidth is huge (can go as high as 30 GB/s). However, to utilize it in its entirety you have to reduce the number transactions via consolidating OpenGL calls. The exact number of updates entirely depends on the hardware, drivers, and the way you use them, and graphics hardware is diverse.
This is the kind of answer you didn't want to hear, but unfortunately you have to face the truth: to reduce the number of OpenGL calls you have to use the newer version APIs. E.g. instead of setting each uniform individually you are better to submit a bunch of them through uniform shader buffer objects. Instead of submitting each MVP of each model individually, it's better to use instanced rendering. And so on.
An even more radical approach would be to move to a lower-level (and newer) API, i.e. Vulkan, which aims to solve exactly this problem: the cost of submitting work to the GPU.
Can I lazy-load data to have it after a few frames, but without freezing my application
Yes, you can upload buffer objects asynchronously. See Buffer Object Streaming for details.
Do I have to do small updates and profile my loop if there are some microseconds left till next rendering?
Maybe I should implement a real-time profiler which gets a feeling over time, how expensive updates are and can determine the amount of updates per frame?
You don't need any of these if you do it asynchronously.
I'm making a game and I'm actually on the generation of the map.
The map is generated procedurally with some algorithms. There's no problems with this.
The problem is that my map can be huge. So I've thought about cutting the map in chunks.
My chunks are ok, they're 512*512 pixels each, but the only problem is : I have to generate a texture (actually a RenderTexture from the SFML). It takes around 0.5ms to generate so it makes the game to freeze each time I generate a chunk.
I've thought about a way to fix this : I've made a kind of a threadpool with a factory. I just have to send a task to it and it creates the chunk.
Now that it's all implemented, it raises opengl warnings like :
"An internal OpenGL call failed in RenderTarget.cpp (219) : GL_INVALID_OPERATION, the specified operation is not allowed in the current state".
I don't know if this is the good way of dealing with chunks. I've also thought about saving the chunks into images / files, but I fear that it take too much time to save / load them.
Do you know a better way to deal with this kind of "infinite" maps ?
It is an invalid operation because you must have a context bound to each thread. More importantly, all of the GL window system APIs enforce a strict 1:1 mapping between threads and contexts... no thread may have more than one context bound and no context may be bound to more than one thread. What you would need to do is use shared contexts (one context for drawing and one for each worker thread), things like buffer objects and textures will be shared between all shared contexts but the state machine and container objects like FBOs and VAOs will not.
Are you using tiled rendering for this map, or is this just one giant texture?
If you do not need to update individual sub-regions of your "chunk" images you can simply create new textures in your worker threads. The worker threads can create new textures and give them data while the drawing thread goes about its business. Only after a worker thread finishes would you actually try to draw using one of the chunks. This may increase the overall latency between the time a chunk starts loading and eventually appears in the finished scene but you should get a more consistent framerate.
If you need to use a single texture for this, I would suggest you double buffer your texture. Have one that you use in the drawing thread and another one that your worker threads issue glTexSubImage2D (...) on. When the worker thread(s) finish updating their regions of the texture you can swap the texture you use for drawing and updating. This will reduce the amount of synchronization required, but again increases the latency before an update eventually appears on screen.
things to try:
make your chunks smaller
generate the chunks in a separate thread, but pass to the gpu from the main thread
pass to the gpu a small piece at a time, taking a second or two
I'm interested in getting threading into the small engine I'm working on in my spare time, but I'm curious over what the best approuch is. I'm curious about the recommended way to sync the physics thread with the rest of the engine, similar to ThisGuy. I'm working with the Bullet Physics SDK, which already use the data copy method he was describing, but I was wondering, once bullet goes through one simulation then syncs the data back to the other threads, won't it result in something like vertical sync, where the rendering thread, half way through processing data suddenly starts using a newer and different set of information?
Is this something which the viewer will be able to notice? What if an explosion of some sort appears with the object that is meant to be destroyed?
If this is an issue, what is then is the best way to solve it?
Lock the physics thread so it can't do anything until the rendering thread (And basically every other thread) has gone through its frame? That seems like it would waste some CPU time. Or is the preferable method to triple buffer, copy the physics data to a second location, continue the physics simulation then copy that data to the rendering thread once its ready?
What approaches do you guys recommend?
The easiest and probably most used variant is to run physic, render, ai, ... threads in parallel and syncronise them after each of them has finished with a frame/timestep.
This is not the fastest solution, but the one with the fewest problems.
Writing back the data to the rendering thread while this is running, leads to massive syncronisation problems (e.g. you have to lock each vector/matrix while updating it).
To make the paralellisation efficent, you have to minimize the amount of data to syncronize, e.g. only write data to the render thread, that can possible be rendered.
When not synronizing after each frame, you can probably get the effect, that the physic/ai uses all the cpu power producing 60fps, while the renderer only renders 10fps, which in most cases is not, what you want.
A double buffering would also increase performance, but you still need to syncronize your threads. A problem is ai and physic or similar threads, because they possible want modify the same data