WIP: Dist testing

I was testing GEMM with the Tile interface on a distributed setting (p = 1, q = 2 or even p = 2, q = 1) and found that, given a LAPACK matrix A, Tile_to_Lapack(Lapack_to_Tile(A)) != A.

Setting the environment variable DEBUG prints the matrix created in the middle of the process at ztile.c. Even though the descriptor B (used to temporarily store the matrix in a single process) is correct, the resulting descriptor after chameleon_pzlacpy is not.

I'm running the test with mpirun --np 2 $CHAM_FOLDER/build/testing/stesting 2 0 2 2 GEMM 4 4, being the two last arguments the size of the square matrix created and its leading dimension.

compute/ztile.c

@@ -55,6 +55,8 @@
  * @sa CHAMELEON_sLapack_to_Tile
  *
  */
+
+
 int CHAMELEON_zLapack_to_Tile( CHAMELEON_Complex64_t *Af77, int LDA, CHAM_desc_t *A )
 {
     CHAM_context_t *chamctxt;
@@ -79,6 +81,11 @@ int CHAMELEON_zLapack_to_Tile( CHAMELEON_Complex64_t *Af77, int LDA, CHAM_desc_t
                                 LDA, A->n, 0, 0, A->m, A->n, 1, 1,
                                 chameleon_getaddr_cm, chameleon_getblkldd_cm, NULL );
 
+    if(getenv("DEBUG")) {
+        printf("[%d]Matrix B (created inside zLapack_to_Tile; always p=1 q=1)\n", CHAMELEON_My_Mpi_Rank());
+        chameleon_zprint_node_matrix(B, 1);
+    }
+
     /* Start the computation */
     chameleon_sequence_create( chamctxt, &sequence );
 
@@ -89,6 +96,11 @@ int CHAMELEON_zLapack_to_Tile( CHAMELEON_Complex64_t *Af77, int LDA, CHAM_desc_t
 
     chameleon_sequence_wait( chamctxt, sequence );
 
+    if(getenv("DEBUG")) {
+        printf("\n[%d]Matrix A (created by pzlacpy using B; p and q come from the descriptor created at testing_zgemm)\n", CHAMELEON_My_Mpi_Rank());
+        chameleon_zprint_node_matrix(A, 2);
+    }
+
     /* Destroy temporary B descriptor */
     CHAMELEON_Desc_Destroy( &B );
 
@@ -171,3 +183,74 @@ int CHAMELEON_zTile_to_Lapack( CHAM_desc_t *A, CHAMELEON_Complex64_t *Af77, int
     chameleon_sequence_destroy( chamctxt, sequence );
     return status;
 }
+
+
+
+
+
+
+#include <stdio.h>
+
+/* Prints a tile if it's stored in this process */
+void chameleon_zprint_tile(CHAM_desc_t *desc, int i, int j) {
+
+    int rank = CHAMELEON_My_Mpi_Rank();
+    int x,y;
+
+    /* leading dimension of the tile */
+    int ld = desc->get_blkldd(desc, i);
+    int tempmm = i == desc->mt-1 ? desc->m - i * desc->mb : desc->mb;
+    int tempnn = j == desc->nt-1 ? desc->n - j * desc->nb : desc->nb;
+
+    printf("[P%d]Tile %d,%d,%d\n", rank, i, j, ld);
+
+
+    if (chameleon_desc_islocal( desc, i, j)) {
+        CHAMELEON_Complex64_t *tile;
+        tile= desc->get_blkaddr(desc, i, j);
+
+        for (x = 0; x < tempmm; x++) {
+            for (y = 0; y < tempnn; y++) {
+                printf("[P%d]%e ", rank, tile[y*ld+x]);
+            }
+            printf("\n");
+        }
+    }
+
+    else {
+        for (x = 0; x < tempmm; x++) {
+            for (y = 0; y < tempnn; y++) {
+                printf("[P%d]x ", rank);
+            }
+            printf("\n");
+        }
+    }
+}
+
+
+/* Prints every tile stored in this process */
+void chameleon_zprint_node_matrix(CHAM_desc_t *desc, int tiles) {
+    int i, j;
+
+    for (i = 0; i < desc->mt; i++) {
+        for (j = 0; j < desc->nt; j++) {
+            chameleon_zprint_tile(desc, i, j);
+        }
+    }
+}
+
+
+void chameleon_zprint_matrix(char *id, int m, int n, CHAMELEON_Complex64_t *A, int lda) {
+
+    printf("====================================\n");
+    printf("%s:\n", id);
+    int i, j;
+    for (i = 0; i < m; i++) {
+        for (j = 0; j < n; j++) {
+            /* The matrix is stored in col-major but we print it in row-major */
+            printf("%e ", A[j*lda+i]);
+        }
+        printf("\n");
+    }
+    printf("====================================\n");
+}