The main role of the project:

• CUDA'usage(each test code gives the implementation of C ++ and CUDA, respectively, and gives the calculation time for each method)
• TensorRT's usage
• Simd's usage Simd GitHub
• OpenMP's usage
• SIMD's usage
• Assembly Language's usage(MASM, NASM)
• Eigen's usage

CUDA test code(Note: depend on opencv):

• simple
  • vector add: C = A + B
  • matrix multiplication: C = A * B
  • dot product
  • Julia
  • ripple
  • green ball
  • ray tracking
  • heat conduction
  • calculate histogram
  • streams' usage
• layer(approximate)
  • channel normalize(mean/standard deviation)
  • reverse
  • prior_vbox
• image process
  • bgr to gray
  • bgr to bgr565
  • gray image histogram equalization(only C++ implementation)
  • gray image edge detection: Laplacian(only C++ implementation)

Eigen test code:

• transpose
• determinant
• inverse matrix
• norm
• eigenvalues/eigenvectors
• SVD(Singular Value Decomposition)
• pseudoinverse
• trace
• mean, variance, standard deviation
• covariance matrix

TensorRT 2.1.2 test code(only support linux):

• MNIST
• MNIST API(use api produce network)
• GoogleNet
• CharRNN
• Plugin(add a custom layer)
• MNIST Infer(serialize TensorRT model)

The project support platform:

• windows10 64 bits: It can be directly build with VS2022 in windows10 64bits(Except TensorRT).
• Linux:
  • CUDA supports cmake build(file position: prj/linux_cuda_cmake)
  • TensorRT support cmake build(file position: prj/linux_tensorrt_cmake)
  • Simd_Test support cmake build(file position: prj/linux_simd_cmake)
  • AssemblyLanguage_Test support cmake build(file position: prj/linux_assembly_language_cmake)
  • Eigen_Test support cmake build(file position: prj/linux_eigen_cmake)

Screenshot:

Blog: fengbingchun