Line data    Source code

       1              : // Copyright 2024, UChicago Argonne, LLC
       2              : // All Rights Reserved
       3              : // Software Name: NEML2 -- the New Engineering material Model Library, version 2
       4              : // By: Argonne National Laboratory
       5              : // OPEN SOURCE LICENSE (MIT)
       6              : //
       7              : // Permission is hereby granted, free of charge, to any person obtaining a copy
       8              : // of this software and associated documentation files (the "Software"), to deal
       9              : // in the Software without restriction, including without limitation the rights
      10              : // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
      11              : // copies of the Software, and to permit persons to whom the Software is
      12              : // furnished to do so, subject to the following conditions:
      13              : //
      14              : // The above copyright notice and this permission notice shall be included in
      15              : // all copies or substantial portions of the Software.
      16              : //
      17              : // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
      18              : // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
      19              : // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
      20              : // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
      21              : // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
      22              : // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
      23              : // THE SOFTWARE.
      24              : 
      25              : #include <torch/autograd.h>
      26              : 
      27              : #include "neml2/tensors/functions/jacrev.h"
      28              : #include "neml2/tensors/assertions.h"
      29              : #include "neml2/tensors/Tensor.h"
      30              : #include "neml2/tensors/Scalar.h"
      31              : 
      32              : namespace neml2
      33              : {
      34              : std::vector<Tensor>
      35          150 : jacrev(const Tensor & y,
      36              :        const std::vector<Tensor> & xs,
      37              :        bool retain_graph,
      38              :        bool create_graph,
      39              :        bool allow_unused)
      40              : {
      41          150 :   std::vector<Tensor> dy_dxs(xs.size());
      42              : 
      43              :   // Return undefined Tensor if y does not contain any gradient graph
      44          150 :   if (!y.requires_grad())
      45           17 :     return dy_dxs;
      46              : 
      47              :   // Check batch shapes
      48          276 :   for (std::size_t i = 0; i < xs.size(); i++)
      49          143 :     neml_assert_dbg(y.batch_sizes() == xs[i].batch_sizes(),
      50              :                     "In jacrev, the output variable batch shape ",
      51              :                     y.batch_sizes(),
      52              :                     " is different from the batch shape of x[",
      53              :                     i,
      54              :                     "] ",
      55          143 :                     xs[i].batch_sizes(),
      56              :                     ".");
      57              : 
      58          133 :   const auto opt = y.options().requires_grad(false);
      59              : 
      60              :   // Flatten y to handle arbitrarily shaped output
      61          133 :   const auto yf = y.base_flatten();
      62          133 :   const auto G = Scalar::full(1.0, opt).batch_expand(yf.batch_sizes());
      63              : 
      64              :   // Initialize derivatives to zero
      65          266 :   std::vector<ATensor> xts(xs.begin(), xs.end());
      66          133 :   std::vector<Tensor> dyf_dxs(xs.size());
      67          276 :   for (std::size_t i = 0; i < xs.size(); i++)
      68          286 :     dyf_dxs[i] = Tensor::zeros(
      69          429 :         yf.batch_sizes(), utils::add_shapes(yf.base_size(0), xs[i].base_sizes()), opt);
      70              : 
      71              :   // Use autograd to calculate the derivatives
      72          402 :   for (Size i = 0; i < yf.base_size(0); i++)
      73              :   {
      74         1345 :     const auto dyfi_dxs = torch::autograd::grad({yf.base_index({i})},
      75              :                                                 {xts},
      76              :                                                 {G},
      77              :                                                 /*retain_graph=*/retain_graph,
      78              :                                                 /*create_graph=*/create_graph,
      79         2152 :                                                 /*allow_unused=*/allow_unused);
      80          269 :     neml_assert_dbg(dyfi_dxs.size() == xs.size(),
      81              :                     "In jacrev, the number of derivatives is ",
      82          269 :                     dyfi_dxs.size(),
      83              :                     " but the number of input tensors is ",
      84          269 :                     xs.size(),
      85              :                     ".");
      86          563 :     for (std::size_t j = 0; j < xs.size(); j++)
      87          294 :       if (dyfi_dxs[j].defined())
      88          873 :         dyf_dxs[j].base_index_put_({Size(i)}, dyfi_dxs[j]);
      89          269 :   }
      90              : 
      91              :   // Reshape the derivative back to the correct shape
      92          276 :   for (std::size_t i = 0; i < xs.size(); i++)
      93          143 :     dy_dxs[i] = dyf_dxs[i].base_reshape(utils::add_shapes(y.base_sizes(), xs[i].base_sizes()));
      94              : 
      95          133 :   return dy_dxs;
      96         1231 : }
      97              : 
      98              : Tensor
      99          146 : jacrev(const Tensor & y, const Tensor & x, bool retain_graph, bool create_graph, bool allow_unused)
     100              : {
     101          438 :   return jacrev(y, std::vector<Tensor>{x}, retain_graph, create_graph, allow_unused)[0];
     102          146 : }
     103              : } // namespace neml2