Blend between splits and fade in last one

[TaaTT4]

Ciao,

I'm far from being a shaders guru, so I ask for your help.
Is it possible to fade in the last PSSM split based on shadow distance from the camera?


What i want to avoid is the shadows pop out effect when the last split becomes visible.

I've seen some other games (Assetto Corsa for example) doing something similar.


[edit] Title changed to reflect content of my own reply post

[TaaTT4]

I've solved things by myself.
A little step toward being a shaders guru

Replace this block in PixelShader_ps

Code: Select all

@property( hlms_pssm_splits )
    float fShadow = 1.0;
    if( inPs.depth <= passBuf.pssmSplitPoints@value(CurrentShadowMap) )
        fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL0, passBuf.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
@foreach( hlms_pssm_splits, n, 1 )	else if( inPs.depth <= passBuf.pssmSplitPoints@value(CurrentShadowMap) )
        fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@n, passBuf.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
@end @end @property( !hlms_pssm_splits && hlms_num_shadow_maps && hlms_lights_directional )
    float fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL0, passBuf.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
@end

with this one

Code: Select all

@property( hlms_pssm_splits )
@psub( pssm_split_n, hlms_pssm_splits, 1 )
@pset( pssm_split_blend_m, 0 )
@padd( pssm_split_blend_m_plus_1, pssm_split_blend_m, 1 )
	float pssm_split_blend_point0 = lerp( 0.0, passBuf.pssmSplitPoints@value(pssm_split_blend_m), 0.9 );
@foreach( pssm_split_n, n, 1 )	float pssm_split_blend_point@n = lerp( passBuf.pssmSplitPoints@value(pssm_split_blend_m), passBuf.pssmSplitPoints@value(pssm_split_blend_m_plus_1), 0.9 );
	@add( pssm_split_blend_m, 1 )
	@add( pssm_split_blend_m_plus_1, 1 )
@end
    float fShadow = 1.0;
	float fShadowNextSplit = 1.0;
    if( inPs.depth <= passBuf.pssmSplitPoints@value(CurrentShadowMap) )
	{
        fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL0, passBuf.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
		if( inPs.depth > pssm_split_blend_point0 )
		{
			fShadowNextSplit = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@value(CurrentShadowMap), passBuf.shadowRcv[@value(CurrentShadowMap)].invShadowMapSize );
			fShadow = lerp( fShadow, fShadowNextSplit, (inPs.depth - pssm_split_blend_point0) / (passBuf.pssmSplitPoints0 - pssm_split_blend_point0) );
		}
	}
@foreach( pssm_split_n, n, 1 )	else if( inPs.depth <= passBuf.pssmSplitPoints@value(CurrentShadowMap) )
	{
        fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@n, passBuf.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
		if( inPs.depth > pssm_split_blend_point@n )
		{
			fShadowNextSplit = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@value(CurrentShadowMap), passBuf.shadowRcv[@value(CurrentShadowMap)].invShadowMapSize );
			fShadow = lerp( fShadow, fShadowNextSplit, (inPs.depth - pssm_split_blend_point@n) / (passBuf.pssmSplitPoints@n - pssm_split_blend_point@n) );
		}
	}
@end
@psub( pssm_split_m, pssm_split_n, 1 )
	else if( inPs.depth <= passBuf.pssmSplitPoints@value(pssm_split_n) )
	{
		fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@value(CurrentShadowMap), passBuf.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
		float pssmSplitPointFade = lerp( passBuf.pssmSplitPoints@value(pssm_split_m), passBuf.pssmSplitPoints@value(pssm_split_n), 0.5 );
		if( inPs.depth > pssmSplitPointFade )
		{
			fShadow = lerp( fShadow, 1.0, (inPs.depth - pssmSplitPointFade) / (passBuf.pssmSplitPoints@value(pssm_split_n) - pssmSplitPointFade) );
		}
	}
@end @property( !hlms_pssm_splits && hlms_num_shadow_maps && hlms_lights_directional )
    float fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL0, passBuf.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
@end

and enjoy your last splits fading in.

As a bonus track, you'll receive also blending between the various splits.
Check the "Blend between Cascades" section at this link to more info about what I'm talking about.


[edit]
Fixed an issue when spotlights are used (CurrentShadowMap value not updated)
Fade in starts at the half of the last split (instead of beginning)

[GlowingPotato]

That's so cool!

You should submit a pull request. Kudos!

[al2950]

That's so cool!

You should submit a pull request. Kudos!

+1

[dark_sylinc]

Fading between all splits? how's the performance impact? Noticeable? acceptable? ok? unnoticeable?

[TaaTT4]

Do you mean blending?
The fade-in effect happens just in the last split and is pretty much "free" as it just consist in a lerp.

Anyway, I didn't see any appreciable performance impact.
Since the blending addition mostly depends on shadows far clip, splits count and how much of the n-1 first splits must be blended with the next one, with different parameters the performance impact could become significant.

To test those additions we have used:
- 4 splits of 8K, 4K, 4K and 4K (from the nearest to the farthest)
- 200m as shadows far clip
- 10% of evey split blended with the next one

[dark_sylinc]

Do you mean blending?

Yes, sorry. I meant blending.

Anyway, I didn't see any appreciable performance impact.

Very interesting

[xrgo]

This is great! I just put it in my shaders and its working great!
I only made the last split to fadeout just the last 10%, no everything, and changed lerp to mix for glsl:

Code: Select all

					@property( hlms_pssm_splits )
						@psub( pssm_split_n, hlms_pssm_splits, 1 )
						@pset( pssm_split_blend_m, 0 )
						@padd( pssm_split_blend_m_plus_1, pssm_split_blend_m, 1 )
						float pssm_split_blend_point0 = mix( 0.0, pass.pssmSplitPoints@value(pssm_split_blend_m), 0.9 );
						@foreach( hlms_pssm_splits, n, 1 )   
							float pssm_split_blend_point@n = mix( pass.pssmSplitPoints@value(pssm_split_blend_m), pass.pssmSplitPoints@value(pssm_split_blend_m_plus_1), 0.9 );
							@add( pssm_split_blend_m, 1 )
							@add( pssm_split_blend_m_plus_1, 1 )
						@end
							float fShadow = 1.0;
							float fShadowNextSplit = 1.0;
							if( inPs.depth <= pass.pssmSplitPoints@value(CurrentShadowMap) )
							{
								fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL0, pass.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
								if( inPs.depth > pssm_split_blend_point0 )
								{
									fShadowNextSplit = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@value(CurrentShadowMap), pass.shadowRcv[@value(CurrentShadowMap)].invShadowMapSize );
									fShadow = mix( fShadow, fShadowNextSplit, (inPs.depth - pssm_split_blend_point0) / (pass.pssmSplitPoints0 - pssm_split_blend_point0) );
								}
							}
						@foreach( pssm_split_n, n, 1 )
							else if( inPs.depth <= pass.pssmSplitPoints@value(CurrentShadowMap) )
							{
								fShadow = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@n, pass.shadowRcv[@counter(CurrentShadowMap)].invShadowMapSize );
								if( inPs.depth > pssm_split_blend_point@n )
								{
									fShadowNextSplit = getShadow( texShadowMap[@value(CurrentShadowMap)], inPs.posL@value(CurrentShadowMap), pass.shadowRcv[@value(CurrentShadowMap)].invShadowMapSize );
									fShadow = mix( fShadow, fShadowNextSplit, (inPs.depth - pssm_split_blend_point@n) / (pass.pssmSplitPoints@n - pssm_split_blend_point@n) );
								}
							}
						@end
						@psub( pssm_split_m, pssm_split_n, 1 )
					    else if( inPs.depth <= pass.pssmSplitPoints@value(pssm_split_n) )
					    {
							fShadow = getShadow( texShadowMap[@value(pssm_split_n)], inPs.posL@value(pssm_split_n), pass.shadowRcv[@value(pssm_split_n)].invShadowMapSize );
					    	if( inPs.depth > pssm_split_blend_point@value(pssm_split_n) )
							{
								fShadow = mix( fShadow, 1.0, (inPs.depth - pssm_split_blend_point@value(pssm_split_n)) / (pass.pssmSplitPoints@value(pssm_split_n) - pssm_split_blend_point@value(pssm_split_n)) );
					    	}
					    }
					@end

Thank you soooo much!!

have to test performance yet

[TaaTT4]

I've edited the original source code fixing a bug when spotlights are used too and starting the fade in at the half of the last splits (thanks @xrgo).

[TaaTT4]

Better late than never, I've finally find some time to make a PR of this: https://bitbucket.org/sinbad/ogre/pull- ... prove-pssm

Enjoy!

[dark_sylinc]

It's been approved. Thank you for this wonderful contribution!

Now shadows look prettier!

PS: Worked out of the box on Metal.

[zxz]

This sounds really nice. I'm looking forward to trying it out. Do you have any before/after shots?

[dark_sylinc]

Screenshots can work, but it's much better to appreciate in motion. Particularly the last split as they now smoothly fade in/out as you move in/away, instead of being stopped by an invisible wall that cuts shadows.