sumReduceRowsGPU function

GPUTensor<Vector> sumReduceRowsGPU(
  1. GPUTensor<Matrix> m,
  2. CommandBuffer tape, {
  3. GPUTensor<Vector>? outTensor,
})

Sums all matrix elements of m via their row order returning the sum of each row as a vector. The operation is appended to tape for execution.

Implementation

GPUTensor<Vector> sumReduceRowsGPU(GPUTensor<Matrix> m, CommandBuffer tape,
    {GPUTensor<Vector>? outTensor}) {
  int rows = m.shape[0];

  List<int> outShape = <int>[rows];
  GPUTensor<Vector> out = outTensor ?? GPUTensor<Vector>.empty(outShape);

  tape.putInt(OP_SUM_REDUCE_ROWS);
  tape.putString(m.id);
  tape.putString(out.id);

  out.creator = GPUNode(
    <GPUTensor>[m],
    (CommandBuffer bTape) {
      bTape.putInt(OP_BROADCAST_ADD);
      bTape.putString(m.gradId);
      bTape.putString(out.gradId);
      bTape.putString(m.gradId);
    },
    opName: 'sumReduceRowsGPU',
  );

  return out;
}