adding amd and symrcm to the mix

owensgroup · Sep 16, 2024 · 6b7b075 · 6b7b075
1 parent 3a64de4
commit 6b7b075
Show file tree

Hide file tree

Showing 3 changed files with 77 additions and 19 deletions.
diff --git a/apps/MCF/benchmark.cmd b/apps/MCF/benchmark.cmd
@@ -18,20 +18,20 @@ set input_dir4=C:\Github\RXMesh\input\SI
 
 
 rem Flag to indicate whether to start processing files
-set start_processing=1
+set start_processing=0
 
 rem Define the file name to start processing from
-rem set start_file=C:\Github\RXMesh\input\SI\usnm_1149322-20m.obj
+set start_file=C:\Github\RXMesh\input\SI\bell_x1-complete_with_vane-smooth.obj
 
 rem Loop over each directory
-for %%d in (%input_dir4%) do (
+for %%d in (%input_dir4% %input_dir1% %input_dir2% %input_dir3%) do (
     for %%f in (%%d\*.obj) do (
 		echo %%f
         if not "!start_processing!"=="1" (
-            rem if "%%f"=="%start_file%" (
-            rem     set start_processing=1
-			rem 	echo !start_processing!
-            rem )
+            if "%%f"=="%start_file%" (
+                set start_processing=1
+				echo !start_processing!
+            )
         ) else (
             if exist "%%f" (
                 echo %exe% -input "%%f" -device_id %device_id% -perm_method symamd

diff --git a/apps/NDReorder/test_all_permutations.cu b/apps/NDReorder/test_all_permutations.cu
@@ -31,7 +31,7 @@ void no_permute(rxmesh::RXMeshStatic& rx, const EigeMatT& eigen_mat)
 
     fill_with_sequential_numbers(h_permute.data(), h_permute.size());
 
-    //render_permutation(rx, h_permute, "No_PERM");
+    // render_permutation(rx, h_permute, "No_PERM");
 
     int nnz = count_nnz_fillin(eigen_mat, h_permute, "natural");
 
@@ -130,15 +130,15 @@ void with_metis(rxmesh::RXMeshStatic&          rx,
     EXPECT_TRUE(
         rxmesh::is_unique_permutation(h_permute.size(), h_permute.data()));
 
-    //render_permutation(rx, h_permute, "METIS");
+    // render_permutation(rx, h_permute, "METIS");
 
     int nnz = count_nnz_fillin(eigen_mat, h_iperm, "metis");
 
     RXMESH_INFO(" With METIS Nested Dissection NNZ = {}", nnz);
 }
 
 template <typename EigeMatT>
-void with_mgnd(rxmesh::RXMeshStatic& rx, const EigeMatT& eigen_mat)
+void with_gpumgnd(rxmesh::RXMeshStatic& rx, const EigeMatT& eigen_mat)
 {
     std::vector<int> h_permute(eigen_mat.rows());
 
@@ -147,15 +147,15 @@ void with_mgnd(rxmesh::RXMeshStatic& rx, const EigeMatT& eigen_mat)
     EXPECT_TRUE(
         rxmesh::is_unique_permutation(h_permute.size(), h_permute.data()));
 
-    //render_permutation(rx, h_permute, "MGND");
+    // render_permutation(rx, h_permute, "GPUMGND");
 
-    int nnz = count_nnz_fillin(eigen_mat, h_permute, "mgnd");
+    int nnz = count_nnz_fillin(eigen_mat, h_permute, "gpumgnd");
 
-    RXMESH_INFO(" With MGND NNZ = {}", nnz);
+    RXMESH_INFO(" With GPUMGND NNZ = {}", nnz);
 }
 
 template <typename EigeMatT>
-void with_cuda_nd(rxmesh::RXMeshStatic& rx, const EigeMatT& eigen_mat)
+void with_gpu_nd(rxmesh::RXMeshStatic& rx, const EigeMatT& eigen_mat)
 {
     std::vector<int> h_permute(eigen_mat.rows());
 
@@ -164,11 +164,57 @@ void with_cuda_nd(rxmesh::RXMeshStatic& rx, const EigeMatT& eigen_mat)
     EXPECT_TRUE(
         rxmesh::is_unique_permutation(h_permute.size(), h_permute.data()));
 
-    //render_permutation(rx, h_permute, "CUDA_ND");
+    // render_permutation(rx, h_permute, "GPUND");
 
-    int nnz = count_nnz_fillin(eigen_mat, h_permute, "cuda_nd");
+    int nnz = count_nnz_fillin(eigen_mat, h_permute, "gpund");
 
-    RXMESH_INFO(" With CUDA ND NNZ = {}", nnz);
+    RXMESH_INFO(" With GPUND NNZ = {}", nnz);
+}
+
+template <typename T, typename EigeMatT>
+void with_amd(rxmesh::RXMeshStatic&    rx,
+              rxmesh::SparseMatrix<T>& rx_mat,
+              const EigeMatT&          eigen_mat)
+{
+    std::vector<int> h_permute(eigen_mat.rows());
+
+    rx_mat.permute_alloc(rxmesh::PermuteMethod::SYMAMD);
+    rx_mat.permute(rx, rxmesh::PermuteMethod::SYMAMD);
+
+    const int* h_perm = rx_mat.get_h_permute();
+
+    std::memcpy(h_permute.data(),
+                rx_mat.get_h_permute(),
+                h_permute.size() * sizeof(int));
+
+    // render_permutation(rx, h_permute, "AMD");
+
+    int nnz = count_nnz_fillin(eigen_mat, h_permute, "amd");
+
+    RXMESH_INFO(" With AMD NNZ = {}", nnz);
+}
+
+template <typename T, typename EigeMatT>
+void with_symrcm(rxmesh::RXMeshStatic&    rx,
+                 rxmesh::SparseMatrix<T>& rx_mat,
+                 const EigeMatT&          eigen_mat)
+{
+    std::vector<int> h_permute(eigen_mat.rows());
+
+    rx_mat.permute_alloc(rxmesh::PermuteMethod::SYMRCM);
+    rx_mat.permute(rx, rxmesh::PermuteMethod::SYMRCM);
+
+    const int* h_perm = rx_mat.get_h_permute();
+
+    std::memcpy(h_permute.data(),
+                rx_mat.get_h_permute(),
+                h_permute.size() * sizeof(int));
+
+    // render_permutation(rx, h_permute, "symrcm");
+
+    int nnz = count_nnz_fillin(eigen_mat, h_permute, "symrcm");
+
+    RXMESH_INFO(" With SYMRCM NNZ = {}", nnz);
 }
 
 TEST(Apps, NDReorder)
@@ -207,11 +253,15 @@ TEST(Apps, NDReorder)
 
     no_permute(rx, eigen_mat);
 
+    with_amd(rx, rx_mat, eigen_mat);
+
+    with_symrcm(rx, rx_mat, eigen_mat);
+
     with_metis(rx, rx_mat, eigen_mat);
 
-    with_mgnd(rx, eigen_mat);
+    with_gpumgnd(rx, eigen_mat);
 
-    with_cuda_nd(rx, eigen_mat);
+    with_gpu_nd(rx, eigen_mat);
 
     // polyscope::show();
 }

diff --git a/include/rxmesh/matrix/sparse_matrix.cuh b/include/rxmesh/matrix/sparse_matrix.cuh
@@ -797,6 +797,14 @@ struct SparseMatrix
 
     /* --- LOW LEVEL API --- */
 
+    /**
+     * @brief return a pointer to the host memory that holds the permutation
+     */
+    __host__ IndexT* get_h_permute()
+    {
+        return m_h_permute;
+    }
+
     /**
      * @brief allocate all temp buffers needed for the solver low-level API
      */