Skip to content

Feature request: Shared fences on Vulkan for Profiling CudaInterop #484

Description

@julcst

I saw in the source code that shared fences are not functional for Vulkan:

NativeHandle Fence::getNativeHandle() const
{
    gfx::InteropHandle gfxNativeHandle = {};
    FALCOR_GFX_CALL(mGfxFence->getNativeHandle(&gfxNativeHandle));
#if FALCOR_HAS_D3D12
    if (mpDevice->getType() == Device::Type::D3D12)
        return NativeHandle(reinterpret_cast<ID3D12Fence*>(gfxNativeHandle.handleValue));
#endif
#if FALCOR_HAS_VULKAN
        // currently not supported
#endif
    return {};
}

Is this possible or planned? It is need to correctly profile Cuda kernels on Linux (tiny-cuda-nn). Without waiting for Cuda in the Render Pass that utilizes Cuda, the GPU time spent on Cuda spills to subsequent Render Passes, messing up the Profiling.

Or is there another possibility to profile CudaInterop in Falcor?

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions