Hi all,

I'm currently implementing a light probe system. I'm passing the captured data of my 8 closest (specular) light probes to the shader as a TextureCubeArray. I'm using a workflow similar to clustered lighting to determine the list of which probes are needed.

Because I have a list, with a variable number of applicable probes per pixel, I need to use SampleGrad in my evaluation loop as the shader otherwise tries to enforce an unroll.

This is the call to my local specular probes (shader model 5 SampleGrad) :

float3 EnvColor = Loc_Spec.SampleGrad(LinearSampler, float4(dir, lightProbeIndex), SampleGrad_Probe_DDX, SampleGrad_Probe_DDY, LOD).rgb;

My problem: It always samples the first cube in the array. I can pass whatever for 'lightProbeIndex', it does not make a difference.

Capturing the executable however, shows that Loc_Spec is a correctly build array of 8 cubes, each with 6 correctly ordered faces, as expected.

Has anyone encountered a similar problem with SampleGrad before?

Greetings.

PS: Captured Texture/Sampler characteristics below for added info.

Texture

Type: Cube Array

Width/Height: 128

Depth: 1

Array Size: 48

Format: R11G11B10_FLOAT

Sampler

Addressing: UVW: CLAMP

Filter: ANISOTROPIC 16x

LOD Clamp: 0-FLT_MAX

Mip Bias: 0

That's exactly how I try to address them :) (Although, in desperation, I tried every other possibility as well)

It is an issue with SampleGrad. I previously had a working test with SampleLevel.

The WARP driver does not fix the issue either.

Yes, same result. Always first cube, but in that case also always highest mip of course.

There was a lack of a decent SRV description for the TextureCubeArray type. That slipped past since I had no hand in that part and the capture gave correct results. Thx for the hint :)