[no sq] Texture/rendering fixes and improvements

[sfan5]

followup to #15396

context for commit 2 is here:

Another great trick at the cost of some performance is to use a higher precision depth buffer. Most depth buffers have a precision of 24 bits, but most GPUs nowadays support 32 bit depth buffers, increasing the precision by a significant amount. So at the cost of some performance you'll get much more precision with depth testing, reducing z-fighting.

To do

This PR is Ready for Review.

How to test

first commit:

• apply

diff --git a/src/irrlicht_changes/CGUITTFont.h b/src/irrlicht_changes/CGUITTFont.h
--- a/src/irrlicht_changes/CGUITTFont.h
+++ b/src/irrlicht_changes/CGUITTFont.h
@@ -143,7 +143,7 @@ namespace gui
 				bool flgmip = driver->getTextureCreationFlag(video::ETCF_CREATE_MIP_MAPS);
 				driver->setTextureCreationFlag(video::ETCF_CREATE_MIP_MAPS, false);
 				bool flgcpy = driver->getTextureCreationFlag(video::ETCF_ALLOW_MEMORY_COPY);
-				driver->setTextureCreationFlag(video::ETCF_ALLOW_MEMORY_COPY, true);
+				driver->setTextureCreationFlag(video::ETCF_ALLOW_MEMORY_COPY, false);
 
 				// Set the texture color format.
 				switch (pixel_mode)
@@ -176,6 +176,8 @@ namespace gui
 			{
 				if (!dirty) return;
 
+				fprintf(stderr, "updating glyph texture\n");
+
 				void* ptr = texture->lock();
 				video::ECOLOR_FORMAT format = texture->getColorFormat();
 				core::dimension2du size = texture->getOriginalSize();

• ensure that "updating glyph texture" shows up and that the font rendering works fine

other commits:
check that everything still works, in particular:

• sun/moon tonemapping
• minimap
• dynamic shadows

[grorp]

For the texture downloading code, tested with Sharpnet texture pack (because it has a sun_tonemap.png)

video_driver = ogles2

• works on "master" (ETCF_ALLOW_MEMORY_COPY true)

• broken on "master" with ETCF_ALLOW_MEMORY_COPY false

  nicebroken.mp4

  (perhaps epilepsy warning or something)

• works on this PR (also ETCF_ALLOW_MEMORY_COPY false)

works 👍

[grorp]

For preferring a 24-bit depth buffer over a 32-bit buffer:

• Do you have data how much it improves performance / have you observed a performance improvement?

• Looking through the issue tracker, it seems there have been some issues with z-fighting in the past. Have you tested this to make sure there are no new problems?

[sfan5]

I have done no testing beyond checking that world rendering generally works. No performance data either.
This is all just going off "24-bit seems to be the default in GL so that should be enough for us".

If x2048 was still here we could ask him why he preferred a 32-bit depth buffer but 🤷

Looking through the issue tracker, it seems there have been some issues with z-fighting in the past

links please

[grorp]

OK, looking into this further, the non-post-processing version of the pipeline has always used a 24-bit depth buffer, so this should be fine.

I don't see a noteworthy performance improvement on my phone, but I do see a noteworthy performance improvement when removing the ECF_A16B16G16R16F branch so it uses ECF_A8R8G8B8 instead. The data I collected is not of very high quality though.

[sfan5]

thx for testing

I do see a noteworthy performance improvement when removing the ECF_A16B16G16R16F branch so it uses ECF_A8R8G8B8 instead

OpenGL has a 10-bit color format and I have wondered before if we should maybe try using it, in case it's faster than 16-bit.
Not exactly sure what the higher precision is for anyway. It might be so bloom, tonemapping and light effects can work correctly? (see: HDR in video)