[OpenMP] Add num_threads clause list format and strict modifier suppo…

…rt (#85466)

Add support to the runtime for 6.0 spec features that allow num_threads
clause to take a list, and also make use of the strict modifier.
Provides new compiler interface functions for these features.

openmp/runtime/src/dllexports

@@ -1268,6 +1268,11 @@ kmp_set_disp_num_buffers                    890
     __kmpc_atomic_val_8_cas_cpt            2158
     %endif
 
+    # No longer need to put ordinal numbers
+    __kmpc_push_num_threads_list
+    __kmpc_push_num_threads_strict
+    __kmpc_push_num_threads_list_strict
+
 %endif
 
 __kmpc_set_thread_limit

openmp/runtime/src/kmp.h

@@ -532,6 +532,15 @@ enum clock_function_type {
 enum mic_type { non_mic, mic1, mic2, mic3, dummy };
 #endif
 
+// OpenMP 3.1 - Nested num threads array
+typedef struct kmp_nested_nthreads_t {
+  int *nth;
+  int size;
+  int used;
+} kmp_nested_nthreads_t;
+
+extern kmp_nested_nthreads_t __kmp_nested_nth;
+
 /* -- fast reduction stuff ------------------------------------------------ */
 
 #undef KMP_FAST_REDUCTION_BARRIER
@@ -2965,6 +2974,12 @@ typedef struct KMP_ALIGN_CACHE kmp_base_info {
   /* The data set by the primary thread at reinit, then R/W by the worker */
   KMP_ALIGN_CACHE int
       th_set_nproc; /* if > 0, then only use this request for the next fork */
+  int *th_set_nested_nth;
+  bool th_nt_strict; // num_threads clause has strict modifier
+  ident_t *th_nt_loc; // loc for strict modifier
+  int th_nt_sev; // error severity for strict modifier
+  const char *th_nt_msg; // error message for strict modifier
+  int th_set_nested_nth_sz;
 #if KMP_NESTED_HOT_TEAMS
   kmp_hot_team_ptr_t *th_hot_teams; /* array of hot teams */
 #endif
@@ -3206,6 +3221,7 @@ typedef struct KMP_ALIGN_CACHE kmp_base_team {
   void *t_stack_id; // team specific stack stitching id (for ittnotify)
 #endif /* USE_ITT_BUILD */
   distributedBarrier *b; // Distributed barrier data associated with team
+  kmp_nested_nthreads_t *t_nested_nth;
 } kmp_base_team_t;
 
 // Assert that the list structure fits and aligns within
@@ -3542,15 +3558,6 @@ extern enum mic_type __kmp_mic_type;
 extern double __kmp_load_balance_interval; // load balance algorithm interval
 #endif /* USE_LOAD_BALANCE */
 
-// OpenMP 3.1 - Nested num threads array
-typedef struct kmp_nested_nthreads_t {
-  int *nth;
-  int size;
-  int used;
-} kmp_nested_nthreads_t;
-
-extern kmp_nested_nthreads_t __kmp_nested_nth;
-
 #if KMP_USE_ADAPTIVE_LOCKS
 
 // Parameters for the speculative lock backoff system.
@@ -3785,6 +3792,11 @@ extern void ___kmp_thread_free(kmp_info_t *th, void *ptr KMP_SRC_LOC_DECL);
   ___kmp_thread_free((th), (ptr)KMP_SRC_LOC_CURR)
 
 extern void __kmp_push_num_threads(ident_t *loc, int gtid, int num_threads);
+extern void __kmp_push_num_threads_list(ident_t *loc, int gtid,
+                                        kmp_uint32 list_length,
+                                        int *num_threads_list);
+extern void __kmp_set_strict_num_threads(ident_t *loc, int gtid, int sev,
+                                         const char *msg);
 
 extern void __kmp_push_proc_bind(ident_t *loc, int gtid,
                                  kmp_proc_bind_t proc_bind);
@@ -4423,6 +4435,18 @@ KMP_EXPORT kmp_int32 __kmpc_in_parallel(ident_t *loc);
 KMP_EXPORT void __kmpc_pop_num_threads(ident_t *loc, kmp_int32 global_tid);
 KMP_EXPORT void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid,
                                         kmp_int32 num_threads);
+KMP_EXPORT void __kmpc_push_num_threads_strict(ident_t *loc,
+                                               kmp_int32 global_tid,
+                                               kmp_int32 num_threads,
+                                               int severity,
+                                               const char *message);
+
+KMP_EXPORT void __kmpc_push_num_threads_list(ident_t *loc, kmp_int32 global_tid,
+                                             kmp_uint32 list_length,
+                                             kmp_int32 *num_threads_list);
+KMP_EXPORT void __kmpc_push_num_threads_list_strict(
+    ident_t *loc, kmp_int32 global_tid, kmp_uint32 list_length,
+    kmp_int32 *num_threads_list, int severity, const char *message);
 
 KMP_EXPORT void __kmpc_push_proc_bind(ident_t *loc, kmp_int32 global_tid,
                                       int proc_bind);

openmp/runtime/src/kmp_csupport.cpp

@@ -237,6 +237,50 @@ void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid,
   __kmp_push_num_threads(loc, global_tid, num_threads);
 }
 
+void __kmpc_push_num_threads_strict(ident_t *loc, kmp_int32 global_tid,
+                                    kmp_int32 num_threads, int severity,
+                                    const char *message) {
+  __kmp_push_num_threads(loc, global_tid, num_threads);
+  __kmp_set_strict_num_threads(loc, global_tid, severity, message);
+}
+
+/*!
+@ingroup PARALLEL
+@param loc source location information
+@param global_tid global thread number
+@param list_length number of entries in the num_threads_list array
+@param num_threads_list array of numbers of threads requested for this parallel
+construct and subsequent nested parallel constructs
+
+Set the number of threads to be used by the next fork spawned by this thread,
+and some nested forks as well.
+This call is only required if the parallel construct has a `num_threads` clause
+that has a list of integers as the argument.
+*/
+void __kmpc_push_num_threads_list(ident_t *loc, kmp_int32 global_tid,
+                                  kmp_uint32 list_length,
+                                  kmp_int32 *num_threads_list) {
+  KA_TRACE(20, ("__kmpc_push_num_threads_list: enter T#%d num_threads_list=",
+                global_tid));
+  KA_TRACE(20, ("%d", num_threads_list[0]));
+#ifdef KMP_DEBUG
+  for (kmp_uint32 i = 1; i < list_length; ++i)
+    KA_TRACE(20, (", %d", num_threads_list[i]));
+#endif
+  KA_TRACE(20, ("/n"));
+
+  __kmp_assert_valid_gtid(global_tid);
+  __kmp_push_num_threads_list(loc, global_tid, list_length, num_threads_list);
+}
+
+void __kmpc_push_num_threads_list_strict(ident_t *loc, kmp_int32 global_tid,
+                                         kmp_uint32 list_length,
+                                         kmp_int32 *num_threads_list,
+                                         int severity, const char *message) {
+  __kmp_push_num_threads_list(loc, global_tid, list_length, num_threads_list);
+  __kmp_set_strict_num_threads(loc, global_tid, severity, message);
+}
+
 void __kmpc_pop_num_threads(ident_t *loc, kmp_int32 global_tid) {
   KA_TRACE(20, ("__kmpc_pop_num_threads: enter\n"));
   /* the num_threads are automatically popped */