Simplify softlight glow blending math.

[allenwp]

Would someone mind checking my math on this one? It seems that the softlight blend mode has a number of math operations that will never happen.

I've taken a screenshot with this blend mode before and after the change and it resulted in the exact same render.

(I also fixed a comment in another file that incorrectly references the "softlight" term.)

Edit: I've updated the PR to use a better simplification from @Lielay9

[Lielay9]

Is there really a problem with 0.5; won't both branches simplify to color as (2.0f * glow.r - 1.0f) becomes 0?

Otherwise the function looks correct but the form is a bit odd. Personally I'd move to use the more widely known version (unless there's a precision benefit or smth that I'm unaware of) for the left side:

(color.r + (2.0f * glow.r - 1.0f) * (((16.0f * color.r - 12.0f) * color.r + 4.0f) * color.r - color.r))

If you want to be space efficient combine the common parts and place the condition in the middle:

(color.r + (2.0f * glow.r - 1.0f) * ((color.r <= 0.25f ? ((16.0f * color.r - 12.0f) * color.r + 4.0f) * color.r : sqrt(color.r)) - color.r))

EDIT: Concentrated on the functions at first but if you want to simplify further:

glow = glow * vec3(0.5f) + vec3(0.5f);
(2.0f * glow.r - 1.0f);

these lines are inverse of one another, so the combined function (the short version for example) could become:

(color.r + glow.r * ((color.r <= 0.25f ? ((16.0f * color.r - 12.0f) * color.r + 4.0f) * color.r : sqrt(color.r)) - color.r))

[allenwp]

Is there really a problem with 0.5; won't both branches simplify to color as (2.0f * glow.r - 1.0f) becomes 0?

Ah, I think you're right; I had just seen (0,undefined) in the graphing calculator and blindly trusted it instead of checking the math myself.

[clayjohn]

Let's look at the case where glow.r is < 0.5.

Assume glow.r = 0.4 and color.r = 0.8

Before:
(color.r - (1.0f - 2.0f * glow.r) * color.r * (1.0f - color.r))
result = (0.8 - (1.0f - 2.0f * 0.4) * 0.8 * (1.0f - 0.8))
result = (0.8 - (0.2) * 0.8 * (0.2))
result = 0.768

After
result = 0.8

[Lielay9]

Let's look at the case where glow.r is < 0.5.

Assume glow.r = 0.4 and color.r = 0.8

Before: (color.r - (1.0f - 2.0f * glow.r) * color.r * (1.0f - color.r)) result = (0.8 - (1.0f - 2.0f * 0.4) * 0.8 * (1.0f - 0.8)) result = (0.8 - (0.2) * 0.8 * (0.2)) result = 0.768

After result = 0.8

Glow can never be below 0.5 as it is first clamped to 0 to 1 range and then 0.5 is added. The path for below 0.5 is completely unnecessary as we only want to use the part of soft-light blending that lightens (i.e > 0.5 range).

Edit: For further clarity: even in your example it doesn't make sense that glow 0.4 + color 0.8 would result in darker color (0.768) i.e. negative glow.

[clayjohn]

Aaaaaaah right. I should have read the surrounding code

[allenwp]

these lines are inverse of one another, so the combined function (the short version for example) could become:

(color.r + glow.r * ((color.r <= 0.25f ? ((16.0f * color.r - 12.0f) * color.r + 4.0f) * color.r : sqrt(color.r)) - color.r))

Wow, I didn't intend to nerd-snipe like this, but this new function looks great to me! I've pushed the changes to this PR with you as co-author.

[allenwp]

I've included the changes from this PR in #110671 and added co-authors to it as well.

[clayjohn]

Superseded by #110671