Index buffer offset not being set with dynamic index buffer and indirect call about bgfx HOT 25 OPEN

Sorry if the screenshot quality is bad

from bgfx.

It doesn't look like there is going to be an easy fix for this in opengl, as the functions to offset buffers don't work on instance buffers, and glMultiDrawElementsIndirect doesn't have an offset either. However Im pretty sure a pure bgfx solution would work in Vulkan and D3D

from bgfx.

And this may not be specific to dynamic buffers either, just any index buffers that are subbuffers & need to be offsetted.

from bgfx.

Can you fix your example to actually work as any other example with shaders binaries being in runtime instead in source directory. I just ran example you provided and it just crashes.

from bgfx.

Sorry. Thought that was a build thing not working for me. Ill fix it

from bgfx.

I cant build the hlsl or metal shaders so ill just write a script in the cubes dir to do that if you're using windows. Also, so far this issue has only occurred on OpenGL and Vulkan.

from bgfx.

GL shaders + SPIRV are enough, I'm looking at GL bug, SPIRV just for VK reference.

from bgfx.

Should be good now.

from bgfx.

It doesn't work...

If you just name your shaders vs_, fs_, cs_ it would build with provided makefile, and it would just output where it needs to be.

from bgfx.

ok. cant see why but, ill fix that now.

from bgfx.

I'm also on Linux, so .bat doesn't work there.

from bgfx.

yeah, that was just for windows. cause i didn't see your message about D3D till id already finished it.

from bgfx.

hopefully it works now

from bgfx.

Failed to load shaders/glsl/vs_bug.bin

from bgfx.

Did you build the shaders?

from bgfx.

Cause I just put the names so they should build with the system.

from bgfx.

diff --git a/examples/01-cubes/fs_bug.sc b/examples/01-cubes/fs_bug.sc
index 4b544fe..226f516 100644
--- a/examples/01-cubes/fs_bug.sc
+++ b/examples/01-cubes/fs_bug.sc
@@ -1,7 +1,5 @@
-vec4 vec4_splat(float input)
-{
-       return vec4(input, input, input, input);
-}
+
+#include <bgfx_shader.sh>
 
 void main()
 {
diff --git a/examples/01-cubes/vs_bug.sc b/examples/01-cubes/vs_bug.sc
index 36c3aa7..2748563 100644
--- a/examples/01-cubes/vs_bug.sc
+++ b/examples/01-cubes/vs_bug.sc
@@ -1,6 +1,8 @@
 $input a_position
 
+#include <bgfx_shader.sh>
+
 void main()
 {
-       gl_Position = vec4(a_position, 1);
+       gl_Position = vec4(a_position, 1.0);
 }

from bgfx.

Ok, so now when I run GL I should see red triangle. And when I run VK I should see something else?

from bgfx.

No, its a really hard bug to show

from bgfx.

But it only shows up in a rendering debugger. If you read my initial comment youll see what I mean. Also, after reading that if you have a better idea on how to confirm a bug like that then Ill implement it.

from bgfx.

Just visually i mean

from bgfx.

Ok, I'll try to understand what's going on... It's usually better to have repro to be obvious when something is wrong.

from bgfx.

yeah no, sorry. I 100% agree. I will say, I think my thought is that when you use dynamic index buffers bgfx packs all of the data into one index buffer, and an offset is used to make sure the right buffer is selected. But when you do a indirect call there is no way for this offset to be set for index buffers (because all index buffer offsets are part of the indirect call, which is created on the gpu). I can't say why it seems to get the vertex buffer right as I would assume this issue happens on the vertex buffers as well (but it doesn't).

from bgfx.

I implemented a potential temporary fix here, on the master branch https://github.com/oxheron/bgfx, if you want to take a look at it. It just provides an interface for users to get the index buffer offset, so it puts it on the users to add that into their indirect buffer calls. This should probably be tested more, but it worked for me on Vulkan so far (and seems that it could be a just bgfx side call anyways).

from bgfx.

I don't know if this would be a good long term fix or not however.

from bgfx.