Various fixes for transmittance effect

[clayjohn]

Fixes: #88083

Fixes: #73946

Use correct shadow sampling for omni and spot lights. We changed how shadow maps were sampled in #51335, but we forgot to update the code for transmittance, so it was sampling from the totally wrong position. This resulted in the bug reported in #88083

Disable transmittance if shadows are disabled. Transmittance requires reading from the shadow map. It used to be calculated at the same time as shadows but was moved in #45023. It has been broken ever since. I noticed this while working on something unrelated.

Correct DirectionalLight transmittance bias to match shadow bias (its still pretty sensitive though). The calculation of bias was updated in #51025 but it wasn't updated for Transmittance. The result is that the default bias is way too large. Users need to manually adjust bias to a tiny amount for it to work at all.

before

after

[BastiaanOlij]

LGTM, I'm not too familiar with the transmittance code so purely judging on code style.

Only question I have seeing you mention in your description that transmittance requires shadows to be enabled, why you're not doing a #if defined(LIGHT_TRANSMITTANCE_USED) && !defined(SHADOWS_DISABLED) and skip over the code entirely?

[clayjohn]

LGTM, I'm not too familiar with the transmittance code so purely judging on code style.

Only question I have seeing you mention in your description that transmittance requires shadows to be enabled, why you're not doing a #if defined(LIGHT_TRANSMITTANCE_USED) && !defined(SHADOWS_DISABLED) and skip over the code entirely?

Just because the first couple variables are still needed for the light compute function:

#ifdef LIGHT_TRANSMITTANCE_USED
	float transmittance_z = transmittance_depth; //no transmittance by default
	transmittance_color.a *= light_attenuation;
#ifndef SHADOWS_DISABLED

So the defines are slightly separated, but in practice it should be fine

[akien-mga]

Thanks!

[akien-mga]

Cherry-picked for 4.3.1.