Add finalizer to release the sparse matrix handle

Author: amontoison

deps/src/onemkl.cpp

@@ -4055,7 +4055,6 @@ extern "C" int onemklXsparse_init_matrix_handle(matrix_handle_t *p_spMat) {
 
 extern "C" int onemklXsparse_release_matrix_handle(syclQueue_t device_queue, matrix_handle_t *p_spMat) {
    auto status = oneapi::mkl::sparse::release_matrix_handle(device_queue->val, (oneapi::mkl::sparse::matrix_handle_t*) p_spMat, {});
-   __FORCE_MKL_FLUSH__(status);
    return 0;
 }
 
@@ -4157,6 +4156,12 @@ extern "C" int onemklXsparse_optimize_trmv(syclQueue_t device_queue, onemklUplo
    return 0;
 }
 
+extern "C" int onemklXsparse_optimize_trsv(syclQueue_t device_queue, onemklUplo uplo_val, onemklTranspose opA, onemklDiag diag_val, matrix_handle_t A) {
+   auto status = oneapi::mkl::sparse::optimize_trsv(device_queue->val, convert(uplo_val), convert(opA), convert(diag_val), (oneapi::mkl::sparse::matrix_handle_t) A, {});
+   __FORCE_MKL_FLUSH__(status);
+   return 0;
+}
+
 extern "C" int onemklSsparse_gemv(syclQueue_t device_queue, onemklTranspose opA, float alpha, matrix_handle_t A, float *x, float beta, float *y) {
    auto status = oneapi::mkl::sparse::gemv(device_queue->val, convert(opA), alpha, (oneapi::mkl::sparse::matrix_handle_t) A, x, beta, y, {});
    __FORCE_MKL_FLUSH__(status);

deps/src/onemkl.h

@@ -2473,6 +2473,9 @@ int onemklXsparse_optimize_gemv(syclQueue_t device_queue, onemklTranspose opA, m
 int onemklXsparse_optimize_trmv(syclQueue_t device_queue, onemklUplo uplo_val, onemklTranspose
                                 opA, onemklDiag diag_val, matrix_handle_t A);
 
+int onemklXsparse_optimize_trsv(syclQueue_t device_queue, onemklUplo uplo_val, onemklTranspose
+                                opA, onemklDiag diag_val, matrix_handle_t A);
+
 int onemklSsparse_gemv(syclQueue_t device_queue, onemklTranspose opA, float alpha,
                        matrix_handle_t A, float *x, float beta, float *y);
 

lib/mkl/wrappers_sparse.jl

@@ -1,3 +1,9 @@
+function sparse_release_matrix_handle(A::oneSparseMatrixCSR)
+    queue = global_queue(context(A.nzVal), device(A.nzVal))
+    handle_ptr = Ref{matrix_handle_t}(A.handle)
+    onemklXsparse_release_matrix_handle(sycl_queue(queue), handle_ptr)
+end
+
 for (fname, elty, intty) in ((:onemklSsparse_set_csr_data   , :Float32   , :Int32),
                              (:onemklSsparse_set_csr_data_64, :Float32   , :Int64),
                              (:onemklDsparse_set_csr_data   , :Float64   , :Int32),
@@ -18,7 +24,9 @@ for (fname, elty, intty) in ((:onemklSsparse_set_csr_data   , :Float32   , :Int3
             nnzA = length(At.nzval)
             queue = global_queue(context(nzVal), device(nzVal))
             $fname(sycl_queue(queue), handle_ptr[], m, n, 'O', rowPtr, colVal, nzVal)
-            return oneSparseMatrixCSR{$elty, $intty}(handle_ptr[], rowPtr, colVal, nzVal, (m,n), nnzA)
+            dA = oneSparseMatrixCSR{$elty, $intty}(handle_ptr[], rowPtr, colVal, nzVal, (m,n), nnzA)
+            finalizer(sparse_release_matrix_handle, dA)
+            return dA
         end
 
         function SparseMatrixCSC(A::oneSparseMatrixCSR{$elty, $intty})