VertexArray merged with culling

[Akasata]

Hi guys, I have a question related to drawing objects in sf::RenderWindow. As far as I know, calling single draw of sf::VertexArray is faster than calling multiple draws. That is why I decided to implement code able to draw tilemap stored in this container. Furthermore, I have read about culling, which is simple way to omit drawing objects situated outside of sf::View. There is no way to draw only part of sf::VertexArray, so implementation of culling is more complicated. I realized that I can split my tilemap into chunks and put sf::VertexArray into all of them. This solution seems to reduce number of draw calls and number of drawn objects, but I am not convinced enough. Is there any better way to merge culling with sf::VertexArray? I am looking forward your answers.

[Laurent]

There is no way to draw only part of sf::VertexArray

This is wrong, you can draw a sub-range of a vertex array with the overload of draw() that takes a sf::Vertex pointer and a vertex count; however you can only select a range of contiguous vertices in memory, so this may not be optimal for culling tiles (this means that you can cull rows or columns, but not both).

I realized that I can split my tilemap into chunks and put sf::VertexArray into all of them

This is often the best compromise.

[Akasata]

Thank you for you response. 

[Hapax]

You could approach this differently: instead of having a vertex array that is larger than the window and not drawing some of the vertices, have level data that is larger than the window but only a vertex array large enough to fill the window (probably with some extra padding - an extra tile column and row should be enough) and that vertex array is updated from the level data.

If your tilemap is grid-like and regular, this should be pretty easy. If not, consider trying Selba Ward's Tile Map class.

[Laurent]

You could approach this differently: instead of having a vertex array that is larger than the window and not drawing some of the vertices, have level data that is larger than the window but only a vertex array large enough to fill the window (probably with some extra padding - an extra tile column and row should be enough) and that vertex array is updated from the level data.

This also works but is more intensive for the CPU and the GPU memory bus. If you use static geometry and the new sf::VertexBuffer class, you have almost zero CPU overhead and no data transfer to the GPU after init.