diff --git a/include/chameleon/tasks_z.h b/include/chameleon/tasks_z.h
index 7027be42d7e821f468929f9ad9af4e9612a9faad..c5704884e1e11331008519bff1a2b955fd6e4321 100644
--- a/include/chameleon/tasks_z.h
+++ b/include/chameleon/tasks_z.h
@@ -497,4 +497,14 @@ void INSERT_TASK_zgetrf_nopiv_percol_trsm( const RUNTIME_option_t *options,
const CHAM_desc_t *A, int Am, int An,
const CHAM_desc_t *U, int Um, int Un );
+void INSERT_TASK_zgetrf_percol_diag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ws );
+
+void INSERT_TASK_zgetrf_percol_offdiag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ws );
+
#endif /* _chameleon_tasks_z_h_ */
diff --git a/runtime/CMakeLists.txt b/runtime/CMakeLists.txt
index dcab800afa3f5cf36c53308af492bbe6b78cf61f..e63a4dd5e7203333b4890a2aa09c27f71fda66c4 100644
--- a/runtime/CMakeLists.txt
+++ b/runtime/CMakeLists.txt
@@ -67,7 +67,7 @@ set(CODELETS_ZSRC
codelets/codelet_zgetrf_incpiv.c
codelets/codelet_zgetrf_nopiv.c
codelets/codelet_zgetrf_nopiv_percol.c
- codelets/codelet_zpanel.c
+ codelets/codelet_zgetrf_percol.c
codelets/codelet_zhe2ge.c
codelets/codelet_zherfb.c
codelets/codelet_zhessq.c
diff --git a/runtime/openmp/codelets/codelet_zgetrf_percol.c b/runtime/openmp/codelets/codelet_zgetrf_percol.c
new file mode 100644
index 0000000000000000000000000000000000000000..4a503f25f849e46b851e9279e1ca623ac68153ae
--- /dev/null
+++ b/runtime/openmp/codelets/codelet_zgetrf_percol.c
@@ -0,0 +1,52 @@
+/**
+ *
+ * @file openmp/codelet_zgetrf_percol.c
+ *
+ * @copyright 2012-2023 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ * Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon zgetrf_percol OpenMP codelets
+ *
+ * @version 1.3.0
+ * @comment Codelets to perform panel factorization with partial pivoting
+ *
+ * @author Mathieu Faverge
+ * @author Matthieu Kuhn
+ * @date 2023-08-22
+ * @precisions normal z -> c d s
+ *
+ */
+#include "chameleon_openmp.h"
+#include "chameleon/tasks_z.h"
+
+void INSERT_TASK_zgetrf_percol_diag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ assert( 0 );
+ (void)options;
+ (void)h;
+ (void)m0;
+ (void)A;
+ (void)Am;
+ (void)An;
+ (void)ipiv;
+}
+
+void INSERT_TASK_zgetrf_percol_offdiag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ assert( 0 );
+ (void)options;
+ (void)h;
+ (void)m0;
+ (void)A;
+ (void)Am;
+ (void)An;
+ (void)ipiv;
+}
diff --git a/runtime/parsec/codelets/codelet_zgetrf_percol.c b/runtime/parsec/codelets/codelet_zgetrf_percol.c
new file mode 100644
index 0000000000000000000000000000000000000000..f94717696c1d93378ac515ee77d3b6d3b6170b92
--- /dev/null
+++ b/runtime/parsec/codelets/codelet_zgetrf_percol.c
@@ -0,0 +1,52 @@
+/**
+ *
+ * @file parsec/codelet_zgetrf_percol.c
+ *
+ * @copyright 2012-2023 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ * Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon zgetrf_percol Parsec codelets
+ *
+ * @version 1.3.0
+ * @comment Codelets to perform panel factorization with partial pivoting
+ *
+ * @author Mathieu Faverge
+ * @author Matthieu Kuhn
+ * @date 2023-08-22
+ * @precisions normal z -> c d s
+ *
+ */
+#include "chameleon_parsec.h"
+#include "chameleon/tasks_z.h"
+
+void INSERT_TASK_zgetrf_percol_diag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ assert( 0 );
+ (void)options;
+ (void)h;
+ (void)m0;
+ (void)A;
+ (void)Am;
+ (void)An;
+ (void)ipiv;
+}
+
+void INSERT_TASK_zgetrf_percol_offdiag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ assert( 0 );
+ (void)options;
+ (void)h;
+ (void)m0;
+ (void)A;
+ (void)Am;
+ (void)An;
+ (void)ipiv;
+}
diff --git a/runtime/quark/codelets/codelet_zgetrf_percol.c b/runtime/quark/codelets/codelet_zgetrf_percol.c
new file mode 100644
index 0000000000000000000000000000000000000000..baea4553d52827ea63db58f487b00d8d119bee9a
--- /dev/null
+++ b/runtime/quark/codelets/codelet_zgetrf_percol.c
@@ -0,0 +1,52 @@
+/**
+ *
+ * @file quark/codelet_zgetrf_percol.c
+ *
+ * @copyright 2012-2023 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ * Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon zgetrf_percol Quark codelets
+ *
+ * @version 1.3.0
+ * @comment Codelets to perform panel factorization with partial pivoting
+ *
+ * @author Mathieu Faverge
+ * @author Matthieu Kuhn
+ * @date 2023-08-22
+ * @precisions normal z -> c d s
+ *
+ */
+#include "chameleon_quark.h"
+#include "chameleon/tasks_z.h"
+
+void INSERT_TASK_zgetrf_percol_diag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ assert( 0 );
+ (void)options;
+ (void)h;
+ (void)m0;
+ (void)A;
+ (void)Am;
+ (void)An;
+ (void)ipiv;
+}
+
+void INSERT_TASK_zgetrf_percol_offdiag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ assert( 0 );
+ (void)options;
+ (void)h;
+ (void)m0;
+ (void)A;
+ (void)Am;
+ (void)An;
+ (void)ipiv;
+}
diff --git a/runtime/starpu/codelets/codelet_zgetrf_percol.c b/runtime/starpu/codelets/codelet_zgetrf_percol.c
new file mode 100644
index 0000000000000000000000000000000000000000..e5887b02b7faee4ccd67b3e989b040f87b142f82
--- /dev/null
+++ b/runtime/starpu/codelets/codelet_zgetrf_percol.c
@@ -0,0 +1,164 @@
+/**
+ *
+ * @file starpu/codelet_zgetrf_percol.c
+ *
+ * @copyright 2012-2023 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ * Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon zpanel StarPU codelets
+ *
+ * @version 1.3.0
+ * @comment Codelets to perform panel factorization with partial pivoting
+ *
+ * @author Mathieu Faverge
+ * @author Matthieu Kuhn
+ * @date 2023-08-22
+ * @precisions normal z -> c d s
+ *
+ */
+#include "chameleon_starpu.h"
+#include "runtime_codelet_z.h"
+#include <coreblas/cblas_wrapper.h>
+
+CHAMELEON_CL_CB( zgetrf_percol_diag, cti_handle_get_m(task->handles[0]), 0, 0, M );
+CHAMELEON_CL_CB( zgetrf_percol_offdiag, cti_handle_get_m(task->handles[0]), 0, 0, M );
+
+#if !defined(CHAMELEON_SIMULATION)
+static void cl_zgetrf_percol_diag_cpu_func(void *descr[], void *cl_arg)
+{
+ int h, m0;
+ RUNTIME_sequence_t *sequence;
+ RUNTIME_request_t *request;
+ CHAM_tile_t *tileA;
+ int *ipiv;
+ cppi_interface_t *nextpiv;
+ cppi_interface_t *prevpiv;
+
+ starpu_codelet_unpack_args( cl_arg, &h, &m0,
+ &sequence, &request );
+
+ tileA = cti_interface_get(descr[0]);
+ ipiv = (int *)STARPU_VECTOR_GET_PTR(descr[1]);
+ nextpiv = (cppi_interface_t*) descr[2];
+ prevpiv = (cppi_interface_t*) descr[3];
+
+ if ( h > 0 ) {
+ cppi_display_dbg( prevpiv, stderr, "Prevpiv before call: " );
+ }
+ if ( h < tileA->n ) {
+ cppi_display_dbg( nextpiv, stderr, "Nextpiv before call: " );
+ }
+
+ /*
+ * Make sure the nextpiv interface store the right information about the
+ * column and diagonal row for the reduction
+ */
+ nextpiv->h = h;
+ nextpiv->has_diag = 1;
+
+ CORE_zgetrf_panel_diag( tileA->m, tileA->n, h, m0,
+ CHAM_tile_get_ptr( tileA ), tileA->ld,
+ ipiv, &(nextpiv->pivot), &(prevpiv->pivot) );
+
+ if ( h > 0 ) {
+ cppi_display_dbg( prevpiv, stderr, "Prevpiv after call: " );
+ }
+ if ( h < tileA->n ) {
+ cppi_display_dbg( nextpiv, stderr, "Nextpiv after call: " );
+ }
+}
+#endif /* !defined(CHAMELEON_SIMULATION) */
+
+/*
+ * Codelet definition
+ */
+CODELETS_CPU( zgetrf_percol_diag, cl_zgetrf_percol_diag_cpu_func );
+
+void INSERT_TASK_zgetrf_percol_diag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ struct starpu_codelet *codelet = &cl_zgetrf_percol_diag;
+ void (*callback)(void*) = options->profiling ? cl_zgetrf_percol_diag_callback : NULL;
+
+ int access_ipiv = ( h == 0 ) ? STARPU_W : STARPU_RW;
+ int access_npiv = ( h == ipiv->n ) ? STARPU_R : STARPU_REDUX;
+ int access_ppiv = ( h == 0 ) ? STARPU_NONE : STARPU_R;
+
+ rt_starpu_insert_task(
+ codelet,
+ STARPU_VALUE, &h, sizeof(int),
+ STARPU_VALUE, &m0, sizeof(int),
+ STARPU_VALUE, &(options->sequence), sizeof(RUNTIME_sequence_t*),
+ STARPU_VALUE, &(options->request), sizeof(RUNTIME_request_t*),
+ STARPU_RW, RTBLKADDR(A, CHAMELEON_Complex64_t, Am, An),
+ access_ipiv, RUNTIME_ipiv_getaddr( ipiv, An ),
+ access_npiv, RUNTIME_pivot_getaddr( ipiv, An, h ),
+ access_ppiv, RUNTIME_pivot_getaddr( ipiv, An, h-1 ),
+ STARPU_PRIORITY, options->priority,
+ STARPU_CALLBACK, callback,
+ STARPU_EXECUTE_ON_WORKER, options->workerid,
+#if defined(CHAMELEON_CODELETS_HAVE_NAME)
+ STARPU_NAME, "zgetrf_percol_diag",
+#endif
+ 0);
+}
+
+#if !defined(CHAMELEON_SIMULATION)
+static void cl_zgetrf_percol_offdiag_cpu_func(void *descr[], void *cl_arg)
+{
+ int h, m0;
+ RUNTIME_sequence_t *sequence;
+ RUNTIME_request_t *request;
+ CHAM_tile_t *tileA;
+ cppi_interface_t *nextpiv;
+ cppi_interface_t *prevpiv;
+
+ starpu_codelet_unpack_args( cl_arg, &h, &m0, &sequence, &request );
+
+ tileA = cti_interface_get(descr[0]);
+ nextpiv = (cppi_interface_t*) descr[1];
+ prevpiv = (cppi_interface_t*) descr[2];
+
+ nextpiv->h = h; /* Initialize in case it uses a copy */
+
+ CORE_zgetrf_panel_offdiag( tileA->m, tileA->n, h, m0,
+ CHAM_tile_get_ptr(tileA), tileA->ld,
+ &(nextpiv->pivot), &(prevpiv->pivot) );
+}
+#endif /* !defined(CHAMELEON_SIMULATION) */
+
+/*
+ * Codelet definition
+ */
+CODELETS_CPU(zgetrf_percol_offdiag, cl_zgetrf_percol_offdiag_cpu_func)
+
+void INSERT_TASK_zgetrf_percol_offdiag( const RUNTIME_option_t *options,
+ int h, int m0,
+ CHAM_desc_t *A, int Am, int An,
+ CHAM_ipiv_t *ipiv )
+{
+ struct starpu_codelet *codelet = &cl_zgetrf_percol_offdiag;
+
+ void (*callback)(void*) = options->profiling ? cl_zgetrf_percol_offdiag_callback : NULL;
+
+ rt_starpu_insert_task(
+ codelet,
+ STARPU_VALUE, &h, sizeof(int),
+ STARPU_VALUE, &m0, sizeof(int),
+ STARPU_VALUE, &(options->sequence), sizeof(RUNTIME_sequence_t *),
+ STARPU_VALUE, &(options->request), sizeof(RUNTIME_request_t *),
+ STARPU_RW, RTBLKADDR(A, CHAMELEON_Complex64_t, Am, An),
+ STARPU_REDUX, RUNTIME_pivot_getaddr( ipiv, An, h ),
+ STARPU_R, RUNTIME_pivot_getaddr( ipiv, An, h-1 ),
+ STARPU_PRIORITY, options->priority,
+ STARPU_CALLBACK, callback,
+ STARPU_EXECUTE_ON_WORKER, options->workerid,
+#if defined(CHAMELEON_CODELETS_HAVE_NAME)
+ STARPU_NAME, "zgetrf_percol_offdiag",
+#endif
+ 0);
+}