sumMatrixGPU function
GPUTensor<Scalar>
sumMatrixGPU(
- GPUTensor<
Matrix> m, - CommandBuffer tape, {
- GPUTensor<
Scalar> ? outTensor,
Appends commands for summing all elements of matrix m to tape.
Implementation
GPUTensor<Scalar> sumMatrixGPU(GPUTensor<Matrix> m, CommandBuffer tape,
{GPUTensor<Scalar>? outTensor}) {
int numRows = m.shape[0];
int numCols = m.shape[1];
GPUTensor<Scalar> out = outTensor ?? GPUTensor<Scalar>(0.0);
tape.putInt(OP_SUM_REDUCE);
tape.putString(m.id);
tape.putString(out.id);
out.creator = GPUNode(
[m],
(CommandBuffer bTape) {
bTape.putInt(OP_SUM_REDUCE_BACKWARD);
bTape.putString(out.gradId);
bTape.putString(m.gradId);
},
opName: 'sum_matrixGPU',
cost: numRows * numCols,
);
return out;
}