diff --git a/coreblas/compute/CMakeLists.txt b/coreblas/compute/CMakeLists.txt
index 1285994bbbc7e54e149a863ceb7289b8feb82993..bec6c5aaf474aae765f84ac8f60e6d00c950fc62 100644
--- a/coreblas/compute/CMakeLists.txt
+++ b/coreblas/compute/CMakeLists.txt
@@ -68,8 +68,9 @@ set(ZSRC
core_zlanhe.c
core_zlansy.c
core_zlantr.c
- core_zlaset2.c
core_zlaset.c
+ core_zlaset2.c
+ core_zlaswp.c
core_zlatro.c
core_zlauum.c
core_zpamm.c
diff --git a/coreblas/compute/core_zlaswp.c b/coreblas/compute/core_zlaswp.c
new file mode 100644
index 0000000000000000000000000000000000000000..e28c82b1785c778b704eccab116b2be81bd93859
--- /dev/null
+++ b/coreblas/compute/core_zlaswp.c
@@ -0,0 +1,223 @@
+/**
+ *
+ * @file core_zlaswp.c
+ *
+ * @copyright 2012-2023 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ * Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon core_zgetrf with partial pivoting CPU kernel
+ *
+ * @version 1.3.0
+ * @author Mathieu Faverge
+ * @author Matthieu Kuhn
+ * @date 2023-08-31
+ * @precisions normal z -> c d s
+ *
+ */
+#include "coreblas/lapacke.h"
+#include "coreblas.h"
+
+/**
+ ******************************************************************************
+ *
+ * @ingroup CORE_CHAMELEON_Complex64_t
+ *
+ * CORE_zlaswp_get extracts the rows from the tile B that have been selected as
+ * pivot into the tile A.
+ *
+ *******************************************************************************
+ *
+ * @param[in] m0
+ * The index of the first row of the tile A into the larger matrix it
+ * belongs to.
+ *
+ * @param[in] m
+ * The number of rows of the matrix A.
+ *
+ * @param[in] n
+ * The number of columns of the matrices A and B.
+ *
+ * @param[in] k
+ * The number of rows of the matrix B. This is the number of potential
+ * pivot that can be extracted from A.
+ *
+ * @param[in] A
+ * On entry, the matrix A of dimension lda-by-n where to extract the
+ * pivot rows if some are selected in the range m0..m0+m.
+ *
+ * @param[in] lda
+ * The leading dimension of the array A. lda >= max(1,m).
+ *
+ * @param[inout] B
+ * On entry, a matrix of size ldb-by-n with 0s or already collected
+ * rows.
+ * On exit, B is filled with the selected rows from A, such that for
+ * each row i, B[i] = A[perm[i]-m0-1].
+ *
+ * @param[in] ldb
+ * The leading dimension of the array B. ldb >= max(1,k).
+ *
+ * @param[in] perm
+ * The permutation array of dimension k.
+ *
+ *******************************************************************************
+ *
+ * @retval CHAMELEON_SUCCESS successful exit
+ * @retval <0 if -i, the i-th argument had an illegal value
+ *
+ */
+int
+CORE_zlaswp_get( int m0, int m, int n, int k,
+ const CHAMELEON_Complex64_t *A, int lda,
+ CHAMELEON_Complex64_t *B, int ldb,
+ const int *perm )
+{
+ int i;
+
+ /* Check input arguments */
+ if (m0 < 0) {
+ coreblas_error(1, "Illegal value of m0");
+ return -1;
+ }
+ if (m < 0) {
+ coreblas_error(2, "Illegal value of m");
+ return -2;
+ }
+ if (n < 0) {
+ coreblas_error(3, "Illegal value of n");
+ return -3;
+ }
+ if (k < 0) {
+ coreblas_error(4, "Illegal value of k");
+ return -4;
+ }
+ if ((lda < chameleon_max(1,m)) && (m > 0)) {
+ coreblas_error(6, "Illegal value of lda");
+ return -6;
+ }
+ if ((ldb < chameleon_max(1,k)) && (k > 0)) {
+ coreblas_error(8, "Illegal value of ldb");
+ return -8;
+ }
+
+ /* Quick return */
+ if ((m == 0) || (n == 0) || (k == 0)) {
+ return CHAMELEON_SUCCESS;
+ }
+
+ for( i=0; i<k; i++ )
+ {
+ int idx = perm[i] - m0;
+
+ if ( ( idx >= 0 ) && (idx < m ) )
+ {
+ cblas_zcopy( n, A + idx, lda,
+ B + i, ldb );
+ }
+ }
+
+ return CHAMELEON_SUCCESS;
+}
+
+/**
+ ******************************************************************************
+ *
+ * @ingroup CORE_CHAMELEON_Complex64_t
+ *
+ * CORE_zlaswp_set copies the rows from the tile A into the tile B when they are
+ * the destination of the pivoted rows.
+ *
+ *******************************************************************************
+ *
+ * @param[in] m0
+ * The index of the first row of the tile B into the larger matrix it
+ * belongs to.
+ *
+ * @param[in] m
+ * The number of rows of the matrix B.
+ *
+ * @param[in] n
+ * The number of columns of the matrices A and B.
+ *
+ * @param[in] k
+ * The number of rows of the matrix A. This is the number of potential
+ * pivot that can be inserted into B.
+ *
+ * @param[in] A
+ * On entry, the matrix A of dimension lda-by-n where to read the
+ * pivoted rows.
+ *
+ * @param[in] lda
+ * The leading dimension of the array A. lda >= max(1,k).
+ *
+ * @param[inout] B
+ * On entry, a matrix of size ldb-by-n that may require some pivoted rows.
+ * On exit, B is updated with the pivoted rows it needs to receive, such that for
+ * each row i, A[i] = B[invp[i]-m0-1].
+ *
+ * @param[in] ldb
+ * The leading dimension of the array B. ldb >= max(1,m).
+ *
+ * @param[in] invp
+ * The inverse permutation array of dimension k.
+ *
+ *******************************************************************************
+ *
+ * @retval CHAMELEON_SUCCESS successful exit
+ * @retval <0 if -i, the i-th argument had an illegal value
+ *
+ */
+int
+CORE_zlaswp_set( int m0, int m, int n, int k,
+ const CHAMELEON_Complex64_t *A, int lda,
+ CHAMELEON_Complex64_t *B, int ldb,
+ const int *invp )
+{
+ int i;
+
+ /* Check input arguments */
+ if (m0 < 0) {
+ coreblas_error(1, "Illegal value of m0");
+ return -1;
+ }
+ if (m < 0) {
+ coreblas_error(2, "Illegal value of m");
+ return -2;
+ }
+ if (n < 0) {
+ coreblas_error(3, "Illegal value of n");
+ return -3;
+ }
+ if (k < 0) {
+ coreblas_error(4, "Illegal value of k");
+ return -4;
+ }
+ if ((lda < chameleon_max(1,k)) && (k > 0)) {
+ coreblas_error(6, "Illegal value of lda");
+ return -6;
+ }
+ if ((ldb < chameleon_max(1,m)) && (m > 0)) {
+ coreblas_error(8, "Illegal value of ldb");
+ return -8;
+ }
+
+ /* Quick return */
+ if ((m == 0) || (n == 0) || (k == 0)) {
+ return CHAMELEON_SUCCESS;
+ }
+
+ for( i=0; i<k; i++ )
+ {
+ int idx = invp[i] - m0;
+
+ if ( ( idx >= 0 ) && (idx < m ) )
+ {
+ cblas_zcopy( n, A + i, lda,
+ B + idx, ldb );
+ }
+ }
+
+ return CHAMELEON_SUCCESS;
+}
diff --git a/coreblas/compute/core_ztile.c b/coreblas/compute/core_ztile.c
index 6de9b2c63cd546ddb8216fa0f5c0bdfd522fdb1d..0383290fa13b9b02bb321c939d73fa72720ad2ad 100644
--- a/coreblas/compute/core_ztile.c
+++ b/coreblas/compute/core_ztile.c
@@ -9,11 +9,11 @@
*
* @brief Chameleon CPU kernel interface from CHAM_tile_t layout to the real one.
*
- * @version 1.2.0
+ * @version 1.3.0
* @author Mathieu Faverge
* @author Florent Pruvost
* @author Alycia Lisito
- * @date 2022-02-22
+ * @date 2023-08-31
* @precisions normal z -> c d s
*
*/
@@ -464,6 +464,24 @@ TCORE_zlaset2( cham_uplo_t uplo, int n1, int n2, CHAMELEON_Complex64_t alpha, CH
CORE_zlaset2( uplo, n1, n2, alpha, CHAM_tile_get_ptr( A ), A->ld );
}
+int
+TCORE_zlaswp_get( int m0, int m, int n, int k, CHAM_tile_t *A, CHAM_tile_t *B, const int *perm )
+{
+ coreblas_kernel_trace( A, B );
+ assert( A->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ assert( B->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ return CORE_zlaswp_get( m0, m, n, k, CHAM_tile_get_ptr( A ), A->ld, CHAM_tile_get_ptr( B ), B->ld, perm );
+}
+
+int
+TCORE_zlaswp_set( int m0, int m, int n, int k, CHAM_tile_t *A, CHAM_tile_t *B, const int *invp )
+{
+ coreblas_kernel_trace( A, B );
+ assert( A->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ assert( B->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ return CORE_zlaswp_set( m0, m, n, k, CHAM_tile_get_ptr( A ), A->ld, CHAM_tile_get_ptr( B ), B->ld, invp );
+}
+
int
TCORE_zlatro( cham_uplo_t uplo,
cham_trans_t trans,
diff --git a/coreblas/compute/core_ztile_empty.c b/coreblas/compute/core_ztile_empty.c
index ef1a09065093133f160a9b2891a202e8b82a3d6c..036c8326311cb1646690e9becc3c539158e33036 100644
--- a/coreblas/compute/core_ztile_empty.c
+++ b/coreblas/compute/core_ztile_empty.c
@@ -356,6 +356,24 @@ TCORE_zlaset2( cham_uplo_t uplo, int n1, int n2, CHAMELEON_Complex64_t alpha, CH
assert( A->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
}
+int
+TCORE_zlaswp_get( int m0, int m, int n, int k, CHAM_tile_t *A, CHAM_tile_t *B, const int *perm )
+{
+ coreblas_kernel_trace( A, B );
+ assert( A->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ assert( B->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ return 0;
+}
+
+int
+TCORE_zlaswp_set( int m0, int m, int n, int k, CHAM_tile_t *A, CHAM_tile_t *B, const int *invp )
+{
+ coreblas_kernel_trace( A, B );
+ assert( A->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ assert( B->format & (CHAMELEON_TILE_FULLRANK | CHAMELEON_TILE_DESC) );
+ return 0;
+}
+
int
TCORE_zlatro( cham_uplo_t uplo,
cham_trans_t trans,
diff --git a/coreblas/include/coreblas/coreblas_z.h b/coreblas/include/coreblas/coreblas_z.h
index 74382a8df5509d1ca8f2a9420fd7415a58e9dfc1..e5e68e2989f11f56517714e2692adb4ab95771a9 100644
--- a/coreblas/include/coreblas/coreblas_z.h
+++ b/coreblas/include/coreblas/coreblas_z.h
@@ -11,7 +11,7 @@
*
* @brief Chameleon CPU CHAMELEON_Complex64_t kernels header
*
- * @version 1.2.0
+ * @version 1.3.0
* @comment This file has been automatically generated
* from Plasma 2.5.0 for CHAMELEON 0.9.2
* @author Jakub Kurzak
@@ -22,7 +22,7 @@
* @author Cedric Castagnede
* @author Florent Pruvost
* @author Matthieu Kuhn
- * @date 2022-02-22
+ * @date 2023-08-31
* @precisions normal z -> c d s
*
*/
@@ -178,6 +178,14 @@ void CORE_zlaset2(cham_uplo_t uplo, int n1, int n2, CHAMELEON_Complex64_t alpha,
CHAMELEON_Complex64_t *tileA, int ldtilea);
void CORE_zlaswp(int N, CHAMELEON_Complex64_t *A, int LDA,
int I1, int I2, const int *IPIV, int INC);
+int CORE_zlaswp_get( int m0, int m, int n, int k,
+ const CHAMELEON_Complex64_t *A, int lda,
+ CHAMELEON_Complex64_t *B, int ldb,
+ const int *perm );
+int CORE_zlaswp_set( int m0, int m, int n, int k,
+ const CHAMELEON_Complex64_t *A, int lda,
+ CHAMELEON_Complex64_t *B, int ldb,
+ const int *invp );
int CORE_zlaswp_ontile( CHAM_desc_t descA, int i1, int i2, const int *ipiv, int inc);
int CORE_zlaswpc_ontile(CHAM_desc_t descA, int i1, int i2, const int *ipiv, int inc);
int CORE_zlatro(cham_uplo_t uplo, cham_trans_t trans,
diff --git a/coreblas/include/coreblas/coreblas_ztile.h b/coreblas/include/coreblas/coreblas_ztile.h
index 74cd413168d44653f27c408eee7dea77eca603ad..88d80d053b12e792beb61ea38d3a9171b3d8b2eb 100644
--- a/coreblas/include/coreblas/coreblas_ztile.h
+++ b/coreblas/include/coreblas/coreblas_ztile.h
@@ -7,11 +7,11 @@
*
* @brief Chameleon CPU kernel CHAM_tile_t interface
*
- * @version 1.2.0
+ * @version 1.3.0
* @author Mathieu Faverge
* @author Florent Pruvost
* @author Alycia Lisito
- * @date 2022-02-22
+ * @date 2023-08-31
* @precisions normal z -> c d s
*
*/
@@ -54,6 +54,8 @@ void TCORE_zlantr( cham_normtype_t norm, cham_uplo_t uplo, cham_diag_t diag, int
int TCORE_zlascal( cham_uplo_t uplo, int m, int n, CHAMELEON_Complex64_t alpha, CHAM_tile_t *A );
void TCORE_zlaset( cham_uplo_t uplo, int n1, int n2, CHAMELEON_Complex64_t alpha, CHAMELEON_Complex64_t beta, CHAM_tile_t *A );
void TCORE_zlaset2( cham_uplo_t uplo, int n1, int n2, CHAMELEON_Complex64_t alpha, CHAM_tile_t *A );
+int TCORE_zlaswp_get( int m0, int m, int n, int k, CHAM_tile_t *A, CHAM_tile_t *B, const int *perm );
+int TCORE_zlaswp_set( int m0, int m, int n, int k, CHAM_tile_t *A, CHAM_tile_t *B, const int *invp );
int TCORE_zlatro( cham_uplo_t uplo, cham_trans_t trans, int M, int N, const CHAM_tile_t *A, CHAM_tile_t *B );
void TCORE_zlauum( cham_uplo_t uplo, int N, CHAM_tile_t *A );
#if defined(PRECISION_z) || defined(PRECISION_c)