diff --git a/compute/pzgemm.c b/compute/pzgemm.c
index f0c77ad3f4773654701a41b2aae4c757e5a7d41a..5c6563d2d6d6280f0f08a35b9a25f266eb5e9764 100644
--- a/compute/pzgemm.c
+++ b/compute/pzgemm.c
@@ -233,7 +233,7 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
options,
ChamUpperLower, tempkk, tempmm,
A( k, m ),
- WA( m, (k % C->q) + lq ) );
+ WA( m, (m % C->q) + lq ) );
RUNTIME_data_flush( sequence, A( k, m ) );
@@ -241,8 +241,8 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempkk, tempmm,
- WA( m, ((k+q-1) % C->q) + lq ),
- WA( m, ((k+q) % C->q) + lq ) );
+ WA( m, ((m+q-1) % C->q) + lq ),
+ WA( m, ((m+q) % C->q) + lq ) );
}
}
}
@@ -273,7 +273,7 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
options,
ChamUpperLower, tempnn, tempkk,
B( n, k ),
- WB( (k % C->p) + lp, n ) );
+ WB( (n % C->p) + lp, n ) );
RUNTIME_data_flush( sequence, B( n, k ) );
@@ -281,8 +281,8 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempnn, tempkk,
- WB( ((k+p-1) % C->p) + lp, n ),
- WB( ((k+p) % C->p) + lp, n ) );
+ WB( ((n+p-1) % C->p) + lp, n ),
+ WB( ((n+p) % C->p) + lp, n ) );
}
}
}
diff --git a/compute/pzgepdf_qdwh.c b/compute/pzgepdf_qdwh.c
index de7a82cb1a1f8e99c79bbea4a3551e8c5549d835..420ad6f502ceccea3a253363e5b00333c5f49aea 100644
--- a/compute/pzgepdf_qdwh.c
+++ b/compute/pzgepdf_qdwh.c
@@ -36,7 +36,7 @@ static int _zgepdf_qdwh_opt_genD = 0;
#endif
static int _zgepdf_qdwh_opt_qr = 1;
-static int _zgepdf_qdwh_opt_id = 1;
+static int _zgepdf_qdwh_opt_id = 1; // There is a numerical issue when combining this optimization and the StarPU lacpy
static int _zgepdf_qdwh_verbose = 0;
/**
@@ -719,13 +719,13 @@ chameleon_pzgepdf_qdwh( cham_mtxtype_t mtxtype, CHAM_desc_t *descU, CHAM_desc_t
it++;
last = ( it >= itconv );
+ chameleon_sequence_wait( chamctxt, sequence_it );
if ( params[2] > 100 ) {
int do_qr = (!_zgepdf_qdwh_opt_qr) || (it > 1);
if ( (chamctxt->scheduler == RUNTIME_SCHED_PARSEC) &&
( sequence_it != sequence_qr ) )
{
- chameleon_sequence_wait( chamctxt, sequence_it );
sequence_it = sequence_qr;
request_it = &request_qr;
}
@@ -753,7 +753,6 @@ chameleon_pzgepdf_qdwh( cham_mtxtype_t mtxtype, CHAM_desc_t *descU, CHAM_desc_t
if ( (chamctxt->scheduler == RUNTIME_SCHED_PARSEC) &&
( sequence_it != sequence_po ) )
{
- chameleon_sequence_wait( chamctxt, sequence_it );
sequence_it = sequence_po;
request_it = &request_po;
}
@@ -796,10 +795,10 @@ chameleon_pzgepdf_qdwh( cham_mtxtype_t mtxtype, CHAM_desc_t *descU, CHAM_desc_t
}
}
+ chameleon_sequence_wait( chamctxt, sequence_it );
if ( (chamctxt->scheduler == RUNTIME_SCHED_PARSEC) &&
( sequence_it != sequence ) )
{
- chameleon_sequence_wait( chamctxt, sequence_it );
chameleon_sequence_destroy( chamctxt, sequence_qr );
chameleon_sequence_destroy( chamctxt, sequence_po );
}
diff --git a/compute/pzhemm.c b/compute/pzhemm.c
index b47dda5baa2ac08d4904d564a73a98b4cecdf515..12269d34a3abf4de45163eb75657441cceac5e74 100644
--- a/compute/pzhemm.c
+++ b/compute/pzhemm.c
@@ -339,7 +339,7 @@ chameleon_pzhemm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempam, tempak,
A( Am, Ak ),
- WA( m, (k % C->q) + lq ) );
+ WA( m, (Ak % C->q) + lq ) );
RUNTIME_data_flush( sequence, A( Am, Ak ) );
@@ -347,8 +347,8 @@ chameleon_pzhemm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempam, tempak,
- WA( m, ((k+q-1) % C->q) + lq ),
- WA( m, ((k+q) % C->q) + lq ) );
+ WA( m, ((Ak+q-1) % C->q) + lq ),
+ WA( m, ((Ak+q) % C->q) + lq ) );
}
}
@@ -496,7 +496,7 @@ chameleon_pzhemm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempak, tempan,
A( Ak, An ),
- WB( (k % C->p) + lp, n ) );
+ WB( (Ak % C->p) + lp, n ) );
RUNTIME_data_flush( sequence, A( Ak, An ) );
@@ -504,8 +504,8 @@ chameleon_pzhemm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempak, tempan,
- WB( ((k+p-1) % C->p) + lp, n ),
- WB( ((k+p) % C->p) + lp, n ) );
+ WB( ((Ak+p-1) % C->p) + lp, n ),
+ WB( ((Ak+p) % C->p) + lp, n ) );
}
}
diff --git a/compute/pzsymm.c b/compute/pzsymm.c
index 47632f6ce07bd4a0254cff94ad7b1a258f3d14e4..944ac75ea2daef41f206207d420e9d0543fd0ae1 100644
--- a/compute/pzsymm.c
+++ b/compute/pzsymm.c
@@ -340,7 +340,7 @@ chameleon_pzsymm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempam, tempak,
A( Am, Ak ),
- WA( m, (k % C->q) + lq ) );
+ WA( m, (Ak % C->q) + lq ) );
RUNTIME_data_flush( sequence, A( Am, Ak ) );
@@ -348,8 +348,8 @@ chameleon_pzsymm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempam, tempak,
- WA( m, ((k+q-1) % C->q) + lq ),
- WA( m, ((k+q) % C->q) + lq ) );
+ WA( m, ((Ak+q-1) % C->q) + lq ),
+ WA( m, ((Ak+q) % C->q) + lq ) );
}
}
@@ -497,7 +497,7 @@ chameleon_pzsymm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempak, tempan,
A( Ak, An ),
- WB( (k % C->p) + lp, n ) );
+ WB( (Ak % C->p) + lp, n ) );
RUNTIME_data_flush( sequence, A( Ak, An ) );
@@ -505,8 +505,8 @@ chameleon_pzsymm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempak, tempan,
- WB( ((k+p-1) % C->p) + lp, n ),
- WB( ((k+p) % C->p) + lp, n ) );
+ WB( ((Ak+p-1) % C->p) + lp, n ),
+ WB( ((Ak+p) % C->p) + lp, n ) );
}
}
diff --git a/runtime/starpu/codelets/codelet_zlacpy.c b/runtime/starpu/codelets/codelet_zlacpy.c
index 2d227b37b7167fa28f687652cdcc4b5e734b3299..aa8d73ed3774d60b34f1b9d6641c984db59fe820 100644
--- a/runtime/starpu/codelets/codelet_zlacpy.c
+++ b/runtime/starpu/codelets/codelet_zlacpy.c
@@ -151,7 +151,7 @@ void INSERT_TASK_zlacpyx( const RUNTIME_option_t *options,
if ( (uplo == ChamUpperLower) &&
(tileA->m == m) && (tileA->n == n) &&
(tileB->m == m) && (tileB->n == n) &&
- (displA == 0) && (displB == 0) && 0 )
+ (displA == 0) && (displB == 0) )
{
#if defined(CHAMELEON_USE_MPI)
insert_task_zlacpy_on_remote_node( options,
@@ -227,7 +227,7 @@ void INSERT_TASK_zlacpy( const RUNTIME_option_t *options,
/* Insert the task */
if ( (uplo == ChamUpperLower) &&
(tileA->m == m) && (tileA->n == n) &&
- (tileB->m == m) && (tileB->n == n) && 0 )
+ (tileB->m == m) && (tileB->n == n) )
{
#if defined(CHAMELEON_USE_MPI)
insert_task_zlacpy_on_remote_node( options,
diff --git a/testing/CTestLists.cmake b/testing/CTestLists.cmake
index c185e50b525c719a7b62422f89d5d5b9a259c435..c8d012141de283ea45c4ebe3a4ae3f9270d7e435 100644
--- a/testing/CTestLists.cmake
+++ b/testing/CTestLists.cmake
@@ -21,6 +21,10 @@ if (CHAMELEON_SIMULATION)
endif()
endif()
+set( SINGLE_PRECISIONS s c )
+# list all tests that have a specific input file for single precision computations
+set( SINGLE_TESTS gepdf_qdwh genm2 )
+
if (NOT CHAMELEON_SIMULATION)
foreach(prec ${CHAMELEON_PRECISION})
@@ -82,7 +86,11 @@ if (NOT CHAMELEON_SIMULATION)
endif()
foreach( test ${TESTSTMP} )
- add_test( test_${cat}_${prec}${test} ${PREFIX} ${CMD} -c -t ${THREADS} -g ${gpus} -P 1 -f input/${test}.in )
+ if ( ${test} IN_LIST SINGLE_TESTS AND ${prec} IN_LIST SINGLE_PRECISIONS )
+ add_test( test_${cat}_${prec}${test} ${PREFIX} ${CMD} -c -t ${THREADS} -g ${gpus} -P 1 -f input/${test}_32.in )
+ else()
+ add_test( test_${cat}_${prec}${test} ${PREFIX} ${CMD} -c -t ${THREADS} -g ${gpus} -P 1 -f input/${test}.in )
+ endif()
endforeach()
if ( CHAMELEON_SCHED_STARPU )
@@ -111,11 +119,15 @@ if (NOT CHAMELEON_SIMULATION)
list( REMOVE_ITEM TESTSTMP print gepdf_qr )
- foreach( test ${TESTSTMP} )
- if ( NOT (${cat} STREQUAL "mpi"))
- add_test( test_${cat}_${prec}${test}_std ${CMD} -c -t ${THREADS} -g ${gpus} -P 1 -f input/${test}.in --api=1 )
- endif()
- endforeach()
+ if ( NOT (${cat} STREQUAL "mpi"))
+ foreach( test ${TESTSTMP} )
+ if ( ${test} IN_LIST SINGLE_TESTS AND ${prec} IN_LIST SINGLE_PRECISIONS )
+ add_test( test_${cat}_${prec}${test}_std ${CMD} -c -t ${THREADS} -g ${gpus} -P 1 -f input/${test}_32.in --api=1 )
+ else()
+ add_test( test_${cat}_${prec}${test}_std ${CMD} -c -t ${THREADS} -g ${gpus} -P 1 -f input/${test}.in --api=1 )
+ endif()
+ endforeach()
+ endif()
endforeach()
endforeach()
endforeach()
diff --git a/testing/input/genm2_32.in b/testing/input/genm2_32.in
new file mode 100644
index 0000000000000000000000000000000000000000..3d2a04f54b973ff0c0a78b2228ccfc4649b64839
--- /dev/null
+++ b/testing/input/genm2_32.in
@@ -0,0 +1,19 @@
+# You can enumerate each parameter's values as an explicit list separated by commas or by a range start:end[:step]
+# Not given parameters will receive default values
+
+# GENM2
+# mtxfmt
+# nb: Tile size
+# M: Number of rows of matrix A
+# N: Number of columns of matrix A
+# LDA: Leading dimension of matrix A
+# cond: The condition number
+# mode: the mode values for latms
+
+op = genm2
+nb = 16, 17
+m = 15, 25, 37
+n = 13, 23, 35
+lda = 41
+cond = 1., 1.e6
+mode = 1:6
diff --git a/testing/input/gepdf_qdwh_32.in b/testing/input/gepdf_qdwh_32.in
new file mode 100644
index 0000000000000000000000000000000000000000..80de2094db4a8f7ea63f49cfa4b0a57ed65fd835
--- /dev/null
+++ b/testing/input/gepdf_qdwh_32.in
@@ -0,0 +1,23 @@
+# You can enumerate each parameter's values as an explicit list separated by commas or by a range start:end[:step]
+# Not given parameters will receive default values
+
+# GEPDF_QDWH
+
+# nb: Tile size
+# ib: Inner tile size
+# m: Number of rows of the A matrix
+# n: Number of columns of the A matrix
+# lda: Leading dimension of the A matrix
+# ldb: Leading dimension of the H matrix
+# cond: The condition number
+# mode: the mode values for latms
+
+op = gepdf_qdwh
+nb = 8
+ib = 3
+m = 8, 32, 64
+n = 8, 16, 32
+lda = 79
+ldb = 78
+cond = 1., 1.e6
+mode = 1:6