On Sat, 18 Dec 2021 16:00:24 GMT, Nir Lisker <nlis...@openjdk.org> wrote:
>> modules/javafx.graphics/src/main/native-prism-d3d/hlsl/psMath.h line 106:
>>
>>> 104:
>>> 105: void computeLight(float i, float3 n, float3 refl, float specPower,
>>> float3 L, float3 lightDir, in out float3 d, in out float3 s) {
>>> 106: if (gLightAttenuation[i].w < 0.5) {
>>
>> Is there any performance difference between `< 0.5` and `!= 0` ? I suspect
>> not, but in any case, you might consider the latter (as I also mentioned in
>> the java files). The latter is more clear, so if you choose to stick with
>> what you have, I'd like to see a comment added.
>
> Initially I went with the equality test, but it did not work well, probably
> since we are comparing floating points. This just seemed safer. If you want
> to try the equality test on you machine(s) it would be interesting, but I
> don't know if I would trust it. Same with the ES shaders. On the Java side I
> would trust it, but at this point it's a matter of consistency.
>
> I suspect that this also messed up with the computation approaches of spot
> light that check `falloff != 0` where we should be checking with a small
> delta: `abs(falloff) < EPSILON` instead.
> Performance should be revisited anyway I think at a later stage after the
> emissive color is added (where we need to figure out splitting shaders, which
> also has to do with the removal of the 3 lights restriction).
It seems safer then to either leave it the way you have it (in which case,
please a comment), or change it to `abs(gLightAttenuation[i].w) < EPSILON` both
here and in the Java code (for consistency). I'll leave it up to you.
For spot light, you might want to file a new bug to address the `falloff != 0`
problem.
I agree about deferring the performance question.
-------------
PR: https://git.openjdk.java.net/jfx/pull/548
