/cuad

Automatic Differentiation try with cuda

Primary LanguageCuda

cuad

Automatic Differentiation try with cuda

  1. Run in apollo docker image.

Reference Performance Comparison Test Data

Jacobian Matrix Computation Based On ADOLC

  1. 坐标转换，使用数据量为1024的数据点，输入变量数3，输出变量数3；
  2. 耗时仅包含求导的计算过程部分，不包含结果的获取与打印输出，保持统一标准。

Coordinate Transform Example Codes:

  • CPU version: coordinate_transform/tapeless_scalar.cpp
  • CUDA version : coordinate_transform/my_codes.cu

CPU (first row) and CUDA(second row) test result table:
1.82727 1.9159 2.38486 2.51743 2.193 6.74031 2.35056 2.28936 1.72139 6.13136
0.355435 0.31505 0.303447 0.30466 0.397284 0.399476 0.317374 0.341977 0.304162 0.306204

The average execution time of CPU is 2.787844ms, some value of CUDA is 0.3345069ms, Speed up times 8.33.