(create_func1, create_func2, shape, nnz1, nnz2, nz_dim)
| 112 | @pytest.mark.parametrize("nnz2", [1, 14]) |
| 113 | @pytest.mark.parametrize("nz_dim", [None, 3]) |
| 114 | def test_spspmul(create_func1, create_func2, shape, nnz1, nnz2, nz_dim): |
| 115 | dev = F.ctx() |
| 116 | A = create_func1(shape, nnz1, dev, nz_dim) |
| 117 | B = create_func2(shape, nnz2, dev, nz_dim) |
| 118 | C = mul(A, B) |
| 119 | assert not C.has_duplicate() |
| 120 | |
| 121 | DA = sparse_matrix_to_dense(A) |
| 122 | DB = sparse_matrix_to_dense(B) |
| 123 | DC = DA * DB |
| 124 | |
| 125 | grad = torch.rand_like(C.val) |
| 126 | C.val.backward(grad) |
| 127 | DC_grad = sparse_matrix_to_dense(val_like(C, grad)) |
| 128 | DC.backward(DC_grad) |
| 129 | |
| 130 | assert torch.allclose(sparse_matrix_to_dense(C), DC, atol=1e-05) |
| 131 | assert torch.allclose( |
| 132 | val_like(A, A.val.grad).to_dense(), DA.grad, atol=1e-05 |
| 133 | ) |
| 134 | assert torch.allclose( |
| 135 | val_like(B, B.val.grad).to_dense(), DB.grad, atol=1e-05 |
| 136 | ) |
| 137 | |
| 138 | |
| 139 | @pytest.mark.parametrize( |
nothing calls this directly
no test coverage detected