Fixed distributed sparse mat

Cargo.toml

@@ -26,6 +26,7 @@ blas = "0.22"
 lapack = "0.19"
 thiserror = "2"
 serde = "1"
+bempp-ghost = { git = "https://github.com/bempp/ghost.git" }
 
 coe-rs = "0.1.2"
 pulp = { version = "0.20" }

examples/distributed_csr_matmul.rs

@@ -0,0 +1,73 @@
+//! Demonstrate the creation of a distributed CSR matrix and its multiplication with a vector.
+
+use mpi::traits::Communicator;
+use rlst::{assert_array_relative_eq, prelude::*};
+
+use rlst::io::matrix_market::{read_array_mm, read_coordinate_mm};
+
+fn main() {
+    let universe = mpi::initialize().unwrap();
+    let world = universe.world();
+
+    let rank = world.rank() as usize;
+
+    let dist_mat;
+
+    let domain_layout = DefaultDistributedIndexLayout::new(313, 1, &world);
+
+    let mut dist_x = DistributedVector::<f64, _>::new(&domain_layout);
+
+    let range_layout = DefaultDistributedIndexLayout::new(507, 1, &world);
+
+    let mut dist_y = DistributedVector::<f64, _>::new(&range_layout);
+
+    if rank == 0 {
+        // Read the sparse matrix in matrix market format.
+        let sparse_mat = read_coordinate_mm::<f64>("mat_507_313.mm").unwrap();
+
+        // Read the vector x. Note that the matrix market format mandates two dimensions for arrays.
+        // So the vector is returned as two-dimensional array with the column dimension being 1.
+        let x = read_array_mm::<f64>("x_313.mm").unwrap();
+
+        // Read the expected result vector in matrix market format.
+        let y_expected = read_array_mm::<f64>("y_507.mm").unwrap();
+
+        // Create a new vector to store the actual matrix-vector product.
+        let mut y_actual = rlst_dynamic_array1!(f64, [507]);
+
+        // Execute the matrix-vector product on just the first rank.
+        sparse_mat.matmul(1.0, x.data(), 0.0, y_actual.data_mut());
+
+        // Make sure that the actual result matches the expected result.
+        // We need to slice the expected result to remove the column dimension.
+        assert_array_relative_eq!(y_actual, y_expected.view().slice(1, 0), 1E-12);
+
+        dist_mat = DistributedCsrMatrix::from_serial_root(
+            sparse_mat,
+            &domain_layout,
+            &range_layout,
+            &world,
+        );
+
+        dist_x.scatter_from_root(x.view().slice(1, 0));
+        dist_mat.matmul(1.0, &dist_x, 0.0, &mut dist_y);
+
+        // It is not needed to set y_actual to zero. But we want to make sure that
+        // the correct data is not already contained in the vector for testing purpose.
+        y_actual.set_zero();
+        dist_y.gather_to_rank_root(y_actual.view_mut());
+
+        // Make sure that the actual result matches the expected result.
+        // We need to slice the expected result to remove the column dimension.
+        assert_array_relative_eq!(y_actual, y_expected.view().slice(1, 0), 1E-12);
+
+        println!("Distributed matrix-vector product successfully executed.");
+    } else {
+        dist_mat = DistributedCsrMatrix::from_serial(0, &domain_layout, &range_layout, &world);
+        dist_x.scatter_from(0);
+
+        dist_mat.matmul(1.0, &dist_x, 0.0, &mut dist_y);
+
+        dist_y.gather_to_rank(0);
+    }
+}