Intermittent CUDA error in Ascent clip filter (cudaErrorSymbolNotFound) while image still renders

[aroccaPrediqtt]

I am seeing an intermittent failure when using Ascent’s clip pipeline filter with plane clips on a 3D CFD dataset.

Error:

TryExecute encountered an error: parallel_for failed: cudaErrorSymbolNotFound: named symbol not found
Failing functor: viskores::cont::detail::ScanExclusiveFunctor<viskores::worklet::Clip::PointBatchData>
Failing device: Cuda
...
Failing functor: viskores::cont::detail::ScanExclusiveFunctor<viskores::worklet::Clip::CellBatchData>
Failing device: Cuda

Important detail: the image is still rendered even when these errors are printed, and the clipping seems to be executed.

I am passing host data to Ascent, but the filter appears to execute on the CUDA backend. Ascent can use CUDA depending on build/runtime settings.

Tests performed:

• single clip only: still fails intermittently
• two clips: same behavior
• clip -> contour and contour -> clip: same behavior

This looks like a CUDA/backend issue in vtkh_clip / viskores::worklet::Clip rather than a deterministic scene-definition error. I attach an extract of the python wrapper I'm using

    def build_scene(self, plots_config, image_name=None, camera=None, x_clip=None):
        self.actions = conduit.Node()


        add_pipes = self.actions.append()
        add_pipes["action"] = "add_pipelines"

        pipes = add_pipes["pipelines"]
        pl1 = pipes["pl1"]

        if x_clip is not None:
            x1, x2 = map(float, x_clip)

            iso = pl1["f1"]
            iso["type"] = "contour"
            iso["params/iso_values"] = [0.02]
            iso["params/field"] = "q-criterion"

            c1 = pl1["f2"]
            c1["type"] = "clip"
            c1["params/topology"] = "volume"
            c1["params/plane/point/x"] = x1
            c1["params/plane/point/y"] = 0.0
            c1["params/plane/point/z"] = 0.0
            c1["params/plane/normal/x"] = 1.0
            c1["params/plane/normal/y"] = 0.0
            c1["params/plane/normal/z"] = 0.0

            c2 = pl1["f3"]
            c2["type"] = "clip"
            c2["params/topology"] = "volume"
            c2["params/plane/point/x"] = x2
            c2["params/plane/point/y"] = 0.0
            c2["params/plane/point/z"] = 0.0
            c2["params/plane/normal/x"] = 1.0
            c2["params/plane/normal/y"] = 0.0
            c2["params/plane/normal/z"] = 0.0
            c2["params/invert"] = "true"

            # c1 = pl1["f1"]
            # c1["type"] = "clip"
            # c1["params/topology"] = "volume"
            # c1["params/box/min/x"] = x1
            # c1["params/box/min/y"] = -10.0
            # c1["params/box/min/z"] = -10.0
            # c1["params/box/max/x"] = x2
            # c1["params/box/max/y"] =  10.0
            # c1["params/box/max/z"] =  10.0
        else:
            iso = pl1["f1"]
            iso["type"] = "contour"
            iso["params/iso_values"] = [0.02]
            iso["params/field"] = "q-criterion"

Mesh is the same as in:
#1669 (comment)

Has this been seen before?
Thanks!