When using batchFunc to perform element-wise multiplication, only the first column is actually multiplied when using a sparse matrix.
For example, if G is a square sparse matrix and nw is a dense column vector, the following code only applies the multiplication on the first column of G:
array batch_mult(const array &lhs, const array &rhs) {
return lhs * rhs;
}
array Cw = batchFunc(G, nw, batch_mult);
If G is dense, then everything works fine.
ArrayFire v3.5.1 (CUDA, 64-bit Linux, build 0a675e8)
Platform: CUDA Toolkit 8, Driver: 375.66
[0] Tesla P100-PCIE-16GB, 16277 MB, CUDA Compute 6.0
-1- Quadro K620, 1998 MB, CUDA Compute 5.0
Reactions are currently unavailable
When using batchFunc to perform element-wise multiplication, only the first column is actually multiplied when using a sparse matrix.
For example, if G is a square sparse matrix and nw is a dense column vector, the following code only applies the multiplication on the first column of G:
array batch_mult(const array &lhs, const array &rhs) { return lhs * rhs; } array Cw = batchFunc(G, nw, batch_mult);If G is dense, then everything works fine.
ArrayFire v3.5.1 (CUDA, 64-bit Linux, build 0a675e8)
Platform: CUDA Toolkit 8, Driver: 375.66
[0] Tesla P100-PCIE-16GB, 16277 MB, CUDA Compute 6.0
-1- Quadro K620, 1998 MB, CUDA Compute 5.0