[BUG] Bad output indices from `pointcloud_search()` with batched shading

[johnhaddon]

Describe the bug

When using pointcloud_search() with batched shader execution, the output indices array is not being filled correctly. It looks a little like OSL thinks the output is a uniform rather than a varying value, but I don't know enough about what is going on under the hood to be sure.

OSL version and dependencies

• OSL branch/version: 1.14.5.1
• OS: Rocky 8.8
• C++ compiler: GCC 11.2.1
• LLVM version: 15.0.7
• OIIO version: 3.0.6.1

To Reproduce

Steps to reproduce the behavior:

1. oslc search.osl (source code in section below).
2. testshade -g 16 16 -batched -param file cloud.geo search.oso -o Cout batched.png
3. testshade -g 16 16 -param file cloud.geo search.oso -o Cout unbatched.png
4. Compare batched.png to unbatched.png. The latter has the expected results, and the former does not.

In the above commands, cloud.geo is testsuite/pointcloud-fold/cloud.geo from the OSL repo.

Evidence

search.osl

shader search( string file = "", output color Cout = 0 )
{
    int indices[1];
    
    // Uncommenting this workaround gives the expected results in batched
    // mode. I suspect it works by tricking OSL into realising that `indices`
    // is a varying quantity.
    // indices[0] = P[0];

    if( pointcloud_search( file, P, 1, 1, "index", indices ) )
    {
        Cout = color( indices[0] / 15.0 );
    }
    else
    {
        Cout = color( 1, 0, 0 );
    }
}

batched.png :

unbatched.png :

[lgritz]

@AlexMWells is this something you can take a look at?