Jump to content

  • Log In with Google      Sign In   
  • Create Account

Asynchronous Loading - Dependencies?


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
6 replies to this topic

#1 Telios   Members   -  Reputation: 398

Like
0Likes
Like

Posted 22 August 2012 - 10:13 AM

Hiya,

I posted a question about asynchronous resource loading previously, and got some really useful replies. I've run into a problem with resource dependencies, and I was hoping to see what others think.

My original system had several resource types, and some were dependent on others - for example, a Shader resource is dependent on a VertexShader and FragmentShader, and a Material is dependent on a Shader.

I solved this by providing the loaders access to caches of the dependent resources, which works fine when everything is synchronous:

ResourceLoader<Shader> shaderLoader;
ResourceCache<Shader> shaderCache(&shaderLoader);

// Allow the material loader to resolve shader references:
ResourceLoader<Material> materialLoader(&shaderCache);

// Now a material request will automatically load the shader if necessary.
ResourceCache<Material> materialCache(&materialLoader);
MaterialPtr material = materialCache.get("test.mat");

With asynchronous loading this isn't so straightforward. A material will be loaded and asynchronously load its dependent shader, but the shader won't be loaded right away and might even encounter an error - making the material invalid.

I could design some complicated dependency system, but I'm hoping there is a simpler way. Does anyone have any suggestions to point me in the right direction?

Many thanks!

Sponsor:

#2 web383   Members   -  Reputation: 793

Like
1Likes
Like

Posted 22 August 2012 - 11:43 AM

As an example, during the load of a Shader, if the VertexShader it is dependent on hasn't been loaded yet, you could do the following:

1) Shader load() returns false - also sending out the dependency it requires.
2) The VertexShader the Shader requires is added to the resource loader if it hasn't been added already.
3) The Shader is re-added to the resource loader

With this in mind, the Shader load() could execute again before the VertexShader and the same thing happens. This could be ok, but will also keep you in an infinite loop if there is a problem locating the VertexShader. In order to handle these cases, I would probably add a timeout or a max fail count to each loader, and log the information.

You could also adjust the thread priority of each loader when a dependency is encountered, but I would try to keep things simple at first and work from there.

#3 bronxbomber92   Members   -  Reputation: 275

Like
1Likes
Like

Posted 22 August 2012 - 09:42 PM

Perhaps the easiest solution is to have a serial queue of resource loading requests that's get consumed by a second thread that is concurrent to your main thread. Just enqueue each resource in the same order as your original synchronous code and your resource loading is now synchronous relative to other resources being loaded. A callback notification system can then easily be created on top of this so other parts of your application will know when the resource's they need have been loaded.

#4 Hodgman   Moderators   -  Reputation: 31843

Like
1Likes
Like

Posted 22 August 2012 - 11:00 PM

Yeah it might be a good idea to queue/batch up your "on loaded" callbacks until after all of the files that need to be loaded have been loaded, then make sure they're in the right order (e.g. shaders before materials), then perform all the "on loaded" calls.
e.g. quick, untested, psuedo C++ with no thread-safety considerations:[source lang=cpp]struct Asset{ std::vector<const char*> dependencies; virtual void Parse();};struct AssetLoadBatch{private: friend class Loader; int itemsRemaining; std::vector<Asset*> loaded;};struct LoadRequest{ const char* name; AssetLoadBatch* batch; int index;};class Loader{ void AsyncLoadDone( LoadRequest* req, Asset* asset ) { AssetLoadBatch* batch = req->batch; ASSERT( !batch->loaded[req->index] ); batch->loaded[req->index] = asset; for( int i=0, end=asset->dependencies.size(); i!=end; ++i ) { StartLoad( asset->dependencies[i], batch ); } if( --batch->itemsRemaining == 0 ) { sortByDependencies( batch->loaded );//make sure dependencies are parsed before the assets that depend on them for( int i=0, end==batch->loaded.size(); i!=end; ++i ) { ASSERT( batch->loaded[i] ) batch->loaded[i]->Parse(); } } delete req; }public: void StartLoad( const char* name, AssetLoadBatch* batch ) {//todo - if name already cached or in progress as part of this batch, then do nothing. LoadRequest* req = new LoadRequest; req->name = name; req->batch = batch; req->index = batch->loaded.size(); batch->loaded.push_back( 0 ); batch->itemsRemaining++; StartAsyncLoad( req, this, &AsyncLoadDone ); }};[/source]

To augment your design, I'd give the MaterialLoader access to the ShaderLoader, not the ShaderCache -- when loading a material, you can also instruct the shaderLoader to load the dependent shaders (or quickly fetch them from the cache).

Edited by Hodgman, 22 August 2012 - 11:11 PM.


#5 Telios   Members   -  Reputation: 398

Like
0Likes
Like

Posted 23 August 2012 - 10:52 AM

Thanks for the replies :)

...


Yep, that seems sensible. I'll try and base the design around this.

(example)


Thanks Hodgman. The only part I'm missing is that without parsing a material, for example, I can't tell what shaders it references. So I guess the parse() method will need to be able to back out and say "I'm missing these dependencies". It could then be pushed to the back of the batch, after the dependencies it needs.

Cheers.

#6 bronxbomber92   Members   -  Reputation: 275

Like
1Likes
Like

Posted 23 August 2012 - 11:48 AM

Thanks Hodgman. The only part I'm missing is that without parsing a material, for example, I can't tell what shaders it references. So I guess the parse() method will need to be able to back out and say "I'm missing these dependencies". It could then be pushed to the back of the batch, after the dependencies it needs.

Why not just load all shaders at once and ensure it happens before material resource loading? You possibly might get speed improvements this way too, since you'll have better instruction cache coherency and the shader compiler may under the hood be able to do multiple shader compilations in parallel.

#7 Telios   Members   -  Reputation: 398

Like
0Likes
Like

Posted 23 August 2012 - 11:59 AM

Ah thanks, that's a good idea. I'll go for that.

To add to this - I might try out a slightly different idea along with that.

My assets currently have a 'state' enum specifying whether they are loaded, pending, or encountered an error. This implies, although I hadn't previously realised, that it is valid for an asset to exist in the cache without being loaded - there are no null pointers, just empty handles.

So I can begin by loading up all the assets in bulk, in the correct order.

When a material is loaded it can kick off jobs to load any missing shader and texture dependencies if needed, get back the 'pending' handles, and then forget about them. There's only a small handful of code which actually cares whether an asset is loaded or not - so I can manually check there, which will only be the cost of testing the state enum. If the asset isn't loaded I can ignore the operation and/or return null, false or whatever - and log the problem.

I can imagine this being very useful - for example, if you start the game with a vertex shader file missing, the objects using that shader just won't draw because the asset will be marked with an error state. This would also be useful with asset swapping at runtime. If an asset change is picked up but fails to load, the objects just stop being drawn or the sound stops playing, and a warning is written to the log.

Thanks for all your help. I'll try down this path for a while and see what happens :)

Cheers!




Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS