CPlusPlus TOP Upgrades For 2022.20000 Builds

CPlusPlus TOPs compiled against headers from builds before and including 2021.30000 won't work in builds 2022.20000 and onwards, due to the changeover to Vulkan. This page contains information for the essential changes to be made to compile your code against the newer headers.

Changes

• getOutputFormat() is gone. You need to specify the output format from within execute() via TOP_CUDAOutputInfo for CUDA, and TOP_UploadInfo for CPU workflow. More flexibility offered this way.
• Pixel format is set in OP_TOPInputDownloadOptions::pixelFormat, replacing OP_TOPInputDownloadOptions::cpuMemPixelType
• The TOP_Context pointer isn't passed through execute() anymore. To get access to it, it is still passed in CreateTOPInstance() where you can store it in your TOP class during instantiation. This context can be used to create and work with a TOP_Buffer, to do CUDA operations, or work with python callbacks.
• For CPU workflow specifics:
  • OP_Inputs::getTOPDataInCPUMemory() is gone. To download texture from the input TOPs, Use OP_TOPInput::downloadTexture(), and specify the download options via OP_TOPInputDownloadOptions.
  • OP_TOPInputDownloadType is gone, and the option of either delaying downloading the frame, or instantly downloading the frame has to be done manually. Store the downloaded result OP_TOPDownloadResult as a member in your TOP class, then when you want to stall the thread to wait for the result, call getData() on it.
  • TOP_OutputFormatSpecs is gone. Use TOP_Output and TOP_UploadInfo. TOP_UploadInfo stores the format of the texture to be outputted. TOP_Output contains the methods for creating buffer via TOP_Context::createOutputBuffer() and uploading the texture to the GPU via TOP_Output::uploadBuffer(), offering lot more functionality.

• To use the CUDA workflow:
  • Store a CUDA stream object cudaStream_t in the operator class. Instantiate it via cudaStreamCreate(), and destroy via cudaStreamDestroy()
  • To access the input data, populate an object of OP_CUDAAcquireInfo, then get the input into OP_CUDAArrayInfo* inputArray via OP_TOPInput::getCUDAArray(). Note that the acquired inputArray->cudaArray will be a nullptr until beginCUDAOperations() has been called, more on this later.
  • To give the CUDA output back, populate an object of TOP_CUDAOutputInfo, then call TOP_Output::createCUDAArray() to create OP_CUDAArrayInfo* outputInfo that will store the output cudaArray. Again, outputInfo->cudaArray will remain null until beginCUDAOperations() is called.
  • CUDA operations can only begin once TOP_Context::beginCUDAOperations() has been called. The relevant input and output OP_CUDAArrayInfo::cudaArray objects will get valid addresses after this method has been called. Once you're done with the operations, call TOP_Context::endCUDAOperations().
  • Further, between beginCUDAOperations() and endCUDAOperations(), your OP_Inputs inputs object cannot be accessed, so call the OP_Inputs methods beforehand to get the values.

For a better understanding of all of these changes, refer to the examples in Samples\CPlusPlus\

See Also

Write a CPlusPlus TOP