Add ELPA_AUTOTUNE_EXTENSIVE; use for stripe_width

elpa/elpa_constants.h.in

@@ -107,7 +107,8 @@ enum ELPA_CONSTANTS {
 #define ELPA_FOR_ALL_AUTOTUNE_LEVELS(X, ...) \
         X(ELPA_AUTOTUNE_NOT_TUNABLE, 0) \
         X(ELPA_AUTOTUNE_FAST, 1) \
-        X(ELPA_AUTOTUNE_MEDIUM, 2)
+        X(ELPA_AUTOTUNE_MEDIUM, 2) \
+	X(ELPA_AUTOTUNE_EXTENSIVE, 3)
 
 enum ELPA_AUTOTUNE_LEVELS {
         ELPA_FOR_ALL_AUTOTUNE_LEVELS(ELPA_ENUM_ENTRY)

src/elpa2/elpa2_trans_ev_tridi_to_band_template.F90

@@ -278,24 +278,24 @@
           thread_width = (l_nev-1)/max_threads + 1 ! number of eigenvectors per OMP thread
 
 #if REALCASE == 1
-          ! call obj%get("stripe_width_real",stripe_width, error)
+          call obj%get("stripewidth_real",stripe_width, error)
 
 #ifdef DOUBLE_PRECISION_REAL
-          stripe_width = 48 ! Must be a multiple of 4
+          !stripe_width = 48 ! Must be a multiple of 4
 #else
-          !stripe_width = stripe_width * 2
-          stripe_width = 96 ! Must be a multiple of 8
+          stripe_width = stripe_width * 2
+          !stripe_width = 96 ! Must be a multiple of 8
 #endif
 #endif /* REALCASE */
 
 #if COMPLEXCASE == 1
-          !call obj%get("stripe_width_complex",stripe_width, error)
+          call obj%get("stripewidth_complex",stripe_width, error)
 
 #ifdef DOUBLE_PRECISION_COMPLEX
-          stripe_width = 48 ! Must be a multiple of 2
+          !stripe_width = 48 ! Must be a multiple of 2
 #else
-          !stripe_width = stripe_width * 2
-          stripe_width = 48 ! Must be a multiple of 4
+          stripe_width = stripe_width * 2
+          !stripe_width = 48 ! Must be a multiple of 4
 #endif
 #endif /* COMPLEXCASE */
 
@@ -384,18 +384,23 @@
 
         else ! useGPU
 #if REALCASE == 1
+          call obj%get("stripewidth_real",stripe_width, error)
+
 #ifdef DOUBLE_PRECISION_REAL
-          stripe_width = 48 ! Must be a multiple of 4
+          !stripe_width = 48 ! Must be a multiple of 4
 #else
-          stripe_width = 96 ! Must be a multiple of 8
+          !stripe_width = 96 ! Must be a multiple of 8
+          stripe_width = 2 * stripe_width
 #endif
 #endif /* REALCASE */
 
 #if COMPLEXCASE == 1
+          call obj%get("stripewidth_complex",stripe_width, error)
+
 #ifdef DOUBLE_PRECISION_COMPLEX
-          stripe_width = 48 ! Must be a multiple of 2
+          !stripe_width = 48 ! Must be a multiple of 2
 #else
-          stripe_width = 48 ! Must be a multiple of 4
+          !stripe_width = 48 ! Must be a multiple of 4
 #endif
 #endif /* COMPLEXCASE */
 

src/elpa_index.c

@@ -222,9 +222,9 @@ static const elpa_index_int_entry_t int_entries[] = {
 
         INT_ENTRY("blocking_in_band_to_full", "Loop blocking, default 3", 3, ELPA_AUTOTUNE_MEDIUM, ELPA_AUTOTUNE_DOMAIN_ANY,
                         band_to_full_cardinality, band_to_full_enumerate, band_to_full_is_valid, NULL, PRINT_YES),
-        INT_ENTRY("stripewidth_real", "Stripewidth_real, default 48. Must be a multiple of 4", 48, ELPA_AUTOTUNE_MEDIUM, ELPA_AUTOTUNE_DOMAIN_REAL,
+        INT_ENTRY("stripewidth_real", "Stripewidth_real, default 48. Must be a multiple of 4", 48, ELPA_AUTOTUNE_EXTENSIVE, ELPA_AUTOTUNE_DOMAIN_REAL,
                         stripewidth_real_cardinality, stripewidth_real_enumerate, stripewidth_real_is_valid, NULL, PRINT_YES),
-        INT_ENTRY("stripewidth_complex", "Stripewidth_complex, default 96. Must be a multiple of 8", 96, ELPA_AUTOTUNE_MEDIUM, ELPA_AUTOTUNE_DOMAIN_COMPLEX,
+        INT_ENTRY("stripewidth_complex", "Stripewidth_complex, default 96. Must be a multiple of 8", 96, ELPA_AUTOTUNE_EXTENSIVE, ELPA_AUTOTUNE_DOMAIN_COMPLEX,
                         stripewidth_complex_cardinality, stripewidth_complex_enumerate, stripewidth_complex_is_valid, NULL, PRINT_YES),
 
         INT_ENTRY("max_stored_rows", "Maximum number of stored rows used in ELPA 1 backtransformation, default 63", 63, ELPA_AUTOTUNE_MEDIUM, ELPA_AUTOTUNE_DOMAIN_ANY,