Sign in to follow this  
Alessio1989

Rendering commands container implementation dilemma

Recommended Posts

Alessio1989    4634

OK, I'm trying to clear my mind, make some order out of chaos in my spaghetti code and finally reorganize the code I learned (more or less) around before starting do something more challenging (like cool things with shaders, for that I just ordered a book).

So, one of the first things I want to implement to throw away my "dummy" forward rendering is a sort of rendering command managements.
And here comes the dilemma: in what kind of container store the commands?

Let me first (correctly I think) exclude some common containers type:
- "simple" array, well don't think I will never able to establish the maximum number of commands per frame.
- every associative container (for me is just unintuitive).

- a stack, why should just process commands in a LIFO policy?

- A heap, implemented into a vector, it could just add some sort of policy to chose the order of the commands, it doesn't seem a smart way to do the work.
 

Now let's try to think simple: could a FCFS policy feeds well the GPU?
If yes, these could be good candidates:
- A double linked list, easy to reuse some commands, but I never liked lists performance for "small" contents.

- A FIFO queue, maybe implemented with a deque? It should fit well the FCFS policy, but prevents reuse of commands.
- Directly a deque or a dynamic array, providing a little more freedom then a FIFO queue.
- My own custom container, maybe providing a good way to reuse commands? (how?!? ph34r.png)
 

Simple assumption: different container for different "living state" commands could be a better strategy (ie: a container for use-only-one commands and another for reusing commands).

Please, share your thoughts, opinions or a draft of partial solution to this wub.png  .

Edited by Alessio1989

Share this post


Link to post
Share on other sites
Ashaman73    13715

An other option:

Utilize the command queues of modern APIs. Add your rendering commands directly and concurrently to  command queue and send them later on to the GPU.

 

Else: FIFO

Edited by Ashaman73

Share this post


Link to post
Share on other sites
Alessio1989    4634

An other option:

Utilize the command queues of modern APIs. Add your rendering commands directly and concurrently to  command queue and send them later on to the GPU.

 

Else: FIFO

 

Unfortunately Direct3D 11 driver command lists + deferred contexts are broken and supported only by NVIDIA hardware and they do not work so well even on geforce afaik.

Using "software threads" in the correct way (multiple threads for creation and one for consumption) is usually better, or it is on what I usually read on slide and presentations; this is why I need a way to "group" the commands.

Edited by Alessio1989

Share this post


Link to post
Share on other sites
Giallanon    1893
I use a buffer, with a counter that "points" to the first free byte ( starts at 0 ends at buffer size -1). Every command is a byte (sort of command ID) followed by additional bytes (ie parameters) if needed. If I run out of space, I simple realloc the buffer giving it some more memory than before. During debug, I break on this reallocations so to have an idea of how much memory a buffer needs and with this data, I prealloc the buffer at startup. To my experience, prealloc a few kb (2 or 3) will be more than sufficient and rarely will run out of space.

Share this post


Link to post
Share on other sites
Alessio1989    4634

I use a buffer, with a counter that "points" to the first free byte ( starts at 0 ends at buffer size -1). Every command is a byte (sort of command ID) followed by additional bytes (ie parameters) if needed. If I run out of space, I simple realloc the buffer giving it some more memory than before. During debug, I break on this reallocations so to have an idea of how much memory a buffer needs and with this data, I prealloc the buffer at startup. To my experience, prealloc a few kb (2 or 3) will be more than sufficient and rarely will run out of space.

 

Thank you for the hint.

Share this post


Link to post
Share on other sites
Hodgman    51324
I do the same giallanon - allocate from a stack of bytes, each command is variable-sized with an ID byte on the front, keeping a pointer to the first item allocated lets you later iterate them in FIFO order, and for next frame you just reset the stack-alloc's cursor back to zero for fast reuse.

Share this post


Link to post
Share on other sites
Alessio1989    4634
Thank you too for the answer.
If I understand it correctly, you (both) use a buffer as a sort of commands FIFO container, where each commands is a "custom" aggregate of bytes (ID, parameters, a "magic" value for end of command), instead of directly storing pointers to callbacks. This strategy should also fit well interdependently of what use I do of Direct3D contexts, isn't it?
Edited by Alessio1989

Share this post


Link to post
Share on other sites
Giallanon    1893
You got it right. There's no need to store an end of command since, after you've read the command ID you should know exactly which params should follow,and then you should know how many bytes to read till the next command. You may instead want to use a special command and append it at the end of the stream so that, when reading the buffer, you can easily understand if you reached the end of the command list. This command list is API agnostic so you can use with dx/ogl, since it's up to you to define which command your list will support

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this