Shadow framebuffer error, can't set shadows at "very high" level even when there is enough memory.

[nani]

0ad shadow buffer crashes if shadow setting is very high even if there is enough memory. My hunch is that is trying to allocate at the integrated graphics. Resolution 4k, integrated VRAM memory 500 MB, dedicated GPU 4GB.

(Already set all system/nvidia control panel to only use the dedicated gpu for 0ad)

Windows 10. Attachment with the system info.

[Vladislav Belov]

Could you try enable debug checks via adding the lines below to the config file?

renderer.backend.debugcontext = "true"
renderer.backend.debugmessages = "true"
renderer.backend.debuglabels = "true"
renderer.backend.debugscopedlabels = "true"

[nani]

Attached the logs with those debug lines added to user config.

Opened game, changed shadow  from "high" to "very high" multiple times. Error Framebuffer showed up on  the left top each time, shadows disappeared.

[Vladislav Belov]

So, the game doesn't crash itself and only shadows disappear?

Could you compile the game in Debug mode?

[nani]

Replying to Vladislav Belov:

So, the game doesn't crash itself and only shadows disappear? Could you compile the game in Debug mode?

No, no game crash. The shadows disappear, when trying to change them to "very high" level in the settings.

[Vladislav Belov]

Replying to nani:

No, no game crash. The shadows disappear, when trying to change them to "very high" level in the settings.

Have you had a chance to test the debug build?

[nani]

Added the new logs. This time I did the debug and release build against the last commit revision (18/4/2022).

[nani]

BTW: on the debug build it also happens the exact same thing, nothing different from what I can see.

[Vladislav Belov]

From logs:

WARNING: OpenGL | high: error source: the API id 1285: GL_OUT_OF_MEMORY error generated. Failed to allocate CPU address space mapping for texture (consider building 64-bit app).

So we have no other way but to switch to 64-bit for shadows with high resolution. Refs #2611

[nani]

Replying to Vladislav Belov:

Happy to say that this issue is not happening for the new release Alpha 26. I don't know if the fix was accidental from another change but now works without out of memory errors.

[Vladislav Belov]

Fixed in r26983.