matmul_gf2_spdn

cuquantum.stabilizer.jax.matmul_gf2_spdn(

A_rowoff: Array,

A_colidx: Array,

B_packed: Array,

*,

m: int,

n: int,

k: int,

) → Array

GF(2) sparse-dense matmul C = A @ B via cuStabilizer SpDn FFI.

Parameters:

• A_rowoff – (m+1,) uint64 CSR row offsets of A (device).

• A_colidx – (>= rowoff[m],) uint64 CSR column indices of A (device). nnz is read from A_rowoff[m]; trailing entries are ignored.

• B_packed – (k, n // 32) uint32 bit-packed dense matrix (device).

• m – rows of A/C.

• n – columns of B/C; must be a multiple of 32.

• k – columns of A / rows of B.

Returns:

(m, n // 32) uint32 bit-packed device tensor.

See also

For computing (A_rowoff, A_colidx) from a dense A:

• jax.experimental.sparse.CSR.fromdense() – jax-native.

• cupyx.scipy.sparse.csr_matrix().

• nvmath.bindings.cusparse cusparseDenseToSparse_*.