bgfx module/target

[Ferdi]

I have no preferation about this. But i can see devs using several atlas where names collide.
Anyway someone could always use a selfmade solution. I would have implented it differently but each dev has their own way of doing things.

I would love to hear how you would have done this. I am really bad at creating API

 

[MikeHart]

I would return an id when you load the atlas. This way you have them identifiable and there will be no duplicates when you add this as a parameter to your load image command.

 

[dawlane]

@Ferdi: What you need to do is read up on game asset management to find a suitable solution to files with the same name.

 

[Ferdi]

I would return an id when you load the atlas. This way you have them identifiable and there will be no duplicates when you add this as a parameter to your load image command.

*slap forehead* Thanks Mike. I think you better check my code. Make sure I don't do silly things like this.

@dawlane: I like Mike's solution. I had a look at how LibGDX does it. I did not like it. It returns Stack<AtlasImage>, but it also can return individual AtlasImage.

So the code now becomes:

Local atlasId:Int = LoadAtlas("td_spritesheet.txt")
' Local atlasId:Int = ATLAS_NULL ' = -1, this is if you want to load individual files

destroyerImg = LoadAtlasImage(atlasId, "sprites/Destroyer.png")
ekstraLiveImg = LoadAtlasImage(atlasId, "sprites/EkstraLive.png")
goldImg = LoadAtlasImage(atlasId, "sprites/Gold.png")

 

[Ferdi]

Example 6 bump texture. I had to load the texture using bimg, and it has conflict with Cerberus because both were using stb_image. It is working, but need to do ALOT of clean up underneath. Anyway source code is attached and screenshot:

 

[MikeHart]

That looks interesting. Cant wait to play with that target.

 

[Ferdi]

That looks interesting. Cant wait to play with that target.

Hi Mike, I'll think of something so others can play with the target.

 

[Ferdi]

I was going to give a github link, but I decided otherwise. So hopefully this works and it is easy to use.

I have put a highly experimental build together. It is Cerberus + bgfx (compiled) + MinGW (64-bit). It is 207 MB download. It is Windows only. And it is 64-bit only. It is mainly to try bgfx. Here is the link:

https://www.dropbox.com/s/vhvd88j5ulfs00n/ceberus+bgfx-v2018-12-18.zip?dl=0 (207 MB)

Several notes:

1. It is ~700MB when unzipped. Each build directory that you make is ~150MB. So remember to delete your build directory when you have tried the example!

2. I have put the MinGW in it because I compiled bgfx, bimg and bx with that compiler. To be precise, this is the MinGW I am using/downloaded:

https://sourceforge.net/projects/mingw-w64/files/Toolchains targetting Win64/Personal Builds/mingw-builds/8.1.0/threads-posix/seh/

I have set the config.winnt.txt to point to the MinGW:

MINGW_PATH="${CERBERUSDIR}/3rdparty/x86_64-8.1.0-release-posix-seh-rt_v6-rev0/mingw64"

3. If there are any bugs, I am going to ask for source code, so I don't have to waste time reproducing the problem. But I am still going to change the core of bgfx (because of all examples I still need to do) and I will definitely going to introduce more problems.

4. This build includes bgfx in modules, mojo1bgfx in modules and Glfw 3 + bgfx in target. (desktop & Windows only)

5. Bgfx examples are in modules/bgfx/examples/... To compile these examples, Cerberus must use the "Glfw 3 + bgfx" target. (desktop & Windows only)

6. To compile Mojo examples with bgfx, you have to change "Import mojo" to "Import mojo1bgfx" and set the target to "Glfw 3 + bgfx". Remember only Mojo 1.

7. The Github link is at: https://github.com/Ferddi/cxbgfx

 

[Ferdi]

Example 7 hooks 2 classes during init. I have done this before to get ReadPixels working. I think what I am going to do is just give functions (like SaveScreenshot or SavePng etc) instead of giving the option to hook 2 classes.

Example 8 updating texture is working. Here is a screenshot:


The source code is at: https://github.com/Ferddi/cxbgfx/tree/bgfx/modules/bgfx/examples/08-update

I did several changes, you can read it in the commits section. And I also merged the latest develop branch from Cerberus main repository to the bgfx branch. So it should be all the latest stuff there

 

[Ferdi]

I have managed to get example 9-hdr working. Here is a screenshot:


Source code: https://github.com/Ferddi/cxbgfx/blob/bgfx/modules/bgfx/examples/09-hdr/hdr.cxs

I had problem with calling bgfxSetViewOrder. I was passing in an array of ViewId as and Int (which in Cerberus is 32-bit). I was wondering why it wasn't drawing after 5 frame buffers. It is because ViewId is 16-bit. Took me a while to figure this one out. Once I created a DataBuffer and use PokeShort to set the ViewId in an array, the example started working.

I looked very carefully at this bit of code in the header file while I was debugging:

#define BGFX_HANDLE_T(_name) \
    typedef struct _name##_s { uint16_t idx; } _name##_t

BGFX_HANDLE_T(bgfx_dynamic_index_buffer_handle);
BGFX_HANDLE_T(bgfx_dynamic_vertex_buffer_handle);
BGFX_HANDLE_T(bgfx_frame_buffer_handle);
BGFX_HANDLE_T(bgfx_index_buffer_handle);
BGFX_HANDLE_T(bgfx_indirect_buffer_handle);
BGFX_HANDLE_T(bgfx_occlusion_query_handle);
BGFX_HANDLE_T(bgfx_program_handle);
BGFX_HANDLE_T(bgfx_shader_handle);
BGFX_HANDLE_T(bgfx_texture_handle);
BGFX_HANDLE_T(bgfx_uniform_handle);
BGFX_HANDLE_T(bgfx_vertex_buffer_handle);
BGFX_HANDLE_T(bgfx_vertex_decl_handle);

#undef BGFX_HANDLE_T

Basically it creates every handle using a macro. And that handle is basically a uint16. Currently handle I wrapped in a class. So I had a think about this for several days now. I think I am going to remove the classes and try to change it as an Int (32-bit). Unfortunately this is will be a big change, but I think it will make the code ALOT easier to read. The next update will probably be really boring.

FYI, I have merge the develop branch into the bgfx branch.

 

[MikeHart]

Sorry for not responding earlier. I definitely have to try what you have posted before.

 

[Ferdi]

Sorry for not responding earlier. I definitely have to try what you have posted before.

May be I do another build once I change it to Int. May be wait for that.

 

[MikeHart]

May be I do another build once I change it to Int. May be wait for that.

Ok will do so.

 

[Ferdi]

Sorry for the delay. I actually change all the handle from class to int 3 weeks ago. But I wanted to use it, to make sure it is the right decision. It is a big change. Also I had a really bad bug when I was doing example 13 stencil, took like a week to solve it.

I have committed my changes and I have merged the develop branch from https://github.com/KrautApps/cerberus.

Since we are now passing an int, the C++ code does casting magically! Here is a C++ code for destroying vertex buffer:

void _bgfx_destroy_vertex_buffer( uint16_t _handle )
{
    bgfx_destroy_vertex_buffer( *(bgfx_vertex_buffer_handle_t *)&_handle ); // <-- from uint16_t to a bgfx_vertex_buffer_handle_t struct
}

I tried using int as handle on 2 examples, example 12 lod and example 13 stencil.

Here is a screenshot of example 12 lod:

And here are two screenshots of example 13 stencil:


Notes and problems:

1. 32-bit int array of size 2 for 64-bit int is causing headache. For example, example 9 hdr I need to create an if statement like this:

        If (BGFX_CAPS_TEXTURE_BLIT[1]|BGFX_CAPS_TEXTURE_READ_BACK[1]) = (supported[1] & (BGFX_CAPS_TEXTURE_BLIT[1]|BGFX_CAPS_TEXTURE_READ_BACK[1]) ) Then

It works but not readable at all! And will probably cause problem down later on.

Also in example 13 stencil there is a table of render states:

static RenderState s_renderStates[RenderState::Count] =
{
    { // StencilReflection_CraftStencil
        BGFX_STATE_WRITE_RGB
        | BGFX_STATE_WRITE_Z
        | BGFX_STATE_DEPTH_TEST_LESS
        | BGFX_STATE_MSAA
        , UINT32_MAX
        , BGFX_STENCIL_TEST_ALWAYS         // pass always
        | BGFX_STENCIL_FUNC_REF(1)         // value = 1
        | BGFX_STENCIL_FUNC_RMASK(0xff)
        | BGFX_STENCIL_OP_FAIL_S_REPLACE
        | BGFX_STENCIL_OP_FAIL_Z_REPLACE
        | BGFX_STENCIL_OP_PASS_Z_REPLACE   // store the value
        , BGFX_STENCIL_NONE
    },
    ...

Currently I just hard coded these render states:

    rs = s_renderStates[ RENDER_STATE_STENCIL_REFLECTION_CRAFT_STENCIL ]
    rs.m_state[0]        = $01000040
    rs.m_state[1]        = $00000017
    rs.m_blendFactorRgba = $ffffffff
    rs.m_fstencil        = $2228ff01
    rs.m_bstencil        = $00000000
    ...

I am not sure how to solve this problem.

2. In example 13 stencil, bgfxFrame is not called at the end:

            // Setup view rect and transform for all used views.
            setViewRectMask(s_viewMask, 0, 0, uint16_t(m_viewState.m_width), uint16_t(m_viewState.m_height) );
            setViewTransformMask(s_viewMask, m_viewState.m_view, m_viewState.m_proj);
            s_viewMask = 0;

            // Advance to next frame. Rendering thread will be kicked to
            // process submitted rendering primitives.
            bgfx::frame();

            //reset clear values on used views
            clearViewMask(s_clearMask, BGFX_CLEAR_NONE, m_clearValues);
            s_clearMask = 0;

            return true;
        }

In Cerberus, I called bgfxFrame() in glfw bgfx target. This is so that Mojo 1 program does not need to call bgfxFrame.

3. Currently Mojo 1 uses bgfx dynamic create buffer. But after doing some of the example, I think I may need to use bgfx transient buffer. I am not sure what is the difference. I need to look into it.

 

[MikeHart]

Wow. Lots of work you put into this. Looks great for sure. Regarding readability and usability. Having the user deal with it directly is indeed difficukt and it screams for a much more simple wrapper luke mojo does.

 

[MikeHart]

@Ferdi Still alive?

 

[Ferdi]

@Mike sorry I have been sick. I am still sick, but getting better. I updated it in Github (https://github.com/Ferddi/cxbgfx/commits/bgfx) before easter and wanted to do the particle example going, but I got sick. I have been resting. I really want to do the particle example first before I update.

 

[MikeHart]

No problem Ferdi. Get well. Health is more important than anything.

 

[MikeHart]

I hope you are ok @Ferdi. Just tried to run your examples from your latest and OMG, you did an awesome job. Thank you for sharing!

 

[Ferdi]

Hi @MikeHart, I am ok, just been taking it easy. I am slowly getting back into it. I got an example working 2 weeks ago, but nothing to report about. It is working but not fully. Now that I know bgfx better, I am going to try to refactor the mojo 1 code that I was working on. My big problem is the current interface between CX and C++ is slow. I model it like DataBuffer. But calling the method PokeFloat for every x y z co-ordinate + normal + texture ... etc is just slow. I tried other weird method like casting a float into an int and putting everything into an int array, it works but I think it is complicated. I want to try writing it in C++ to see what speed increase I get.