Playing around with OpenCL from C# using OpenCL.NET.

• FpConfig
  • Dump out device floating point configuration info
    • CL_DEVICE_SINGLE_FP_CONFIG
• RunKernel
  • Run a simple kernel
• SaveBinaries
  • Build a program and save the CL binaries to files
    • CL_PROGRAM_BINARY_SIZES
    • CL_PROGRAM_BINARIES
• WorkGroupInfo
  • Dump out kernel work group info
    • CL_KERNEL_WORK_GROUP_SIZE
    • CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE
    • CL_KERNEL_COMPILE_WORK_GROUP_SIZE
    • CL_KERNEL_LOCAL_MEM_SIZE
    • CL_KERNEL_PRIVATE_MEM_SIZE
• ReductionScalar
  • Implement reduction using scalar float (see section 10.2 Numerical Reduction in OpenCL in Action)
• ReductionVector
  • Implement reduction using vector float4 (see section 10.2 Numerical Reduction in OpenCL in Action)
• ReductionVectorComplete
  • Like ReductionVector but do the final summing on the OpenCL device using a second kernel (see section 10.3 Synchronizing work-groups in OpenCL in Action)
  • Note: when I run this against my "Intel(R) Core(TM) i7-4720HQ CPU @ 2.60GHz" device it gives crazy (i.e. wrong) results
    • UPDATE: See this discussion on the Intel forums for an explanation of why this happens (race condition)
    • I get the same results using the C code from the book's download (VS2015 project can be found here)
    • I get the same results in C++ too (fixed code can be found here)

• OpenCL - The open standard for parallel programming of heterogeneous systems
• OpenCL (Wikipedia)
• OpenCL 1.2 Reference Pages
  • (I am using version 1.2 - current version is 2.1)
• OpenCL.NET
• OpenCL.NET (NuGet)
• OpenCL in Action (Manning Publications Co.)
• Simon McIntosh-Smith
  • Head of the Microelectronics Group and Bristol University Business Fellow
  • Senior Lecturer in High Performance Computing and Architectures
  • OpenCL: A Hands-on Introduction
  • COMPILING OPENCL KERNELS
• OpenCL™ Optimization Case Study: Simple Reductions