Cpp extensions do not work about fluidnet_cxx HOT 2 CLOSED

At execution, in the function Advect Scalar, called from advection.py input arguments behave as Variable type (grad is computed) whereas intermediate objects in the function are just Tensor type.

at::Tensor advectScalar
(
  float dt, T src, T U, T flags,
  const std::string method_str,
  int bnd,
  const bool sample_outside_fluid,
  const float maccormack_strength
) {
  // Size checking done in python side
  int bsz = flags.size(0);
  int d = flags.size(2);
  int h = flags.size(3);
  int w = flags.size(4);

  bool is3D = (U.size(1) == 3);

  src.set_requires_grad(false);
  U.set_requires_grad(false);
  flags.set_requires_grad(false);

  T s_dst = zeros_like(src);

  T fwd = zeros_like(src);
  T bwd = zeros_like(src);
  T fwd_pos = zeros_like(U);
  T bwd_pos = zeros_like(U);

  std::cout << "Is variable?" << std::endl;
  std::cout << src.is_variable() << std::endl;
  // returns 1, should be 0
  std::cout << U.is_variable() << std::endl;
  // returns 1, should be 0
  std::cout << flags.is_variable() << std::endl;
  // returns 1, should be 0

All should be of Tensor type as no backpropagation is required in the advection part.

@beniz Hello! Could you have a look at this, as I am a bit stucked trying to install fluidnet with pytorch 1.0.

from fluidnet_cxx.

Closed a2a49ac

from fluidnet_cxx.