5.2 multicore markdelay (#3029)

* Markdelay support for multi-domain programs

  - Domains now begin marking simultaneously at some minor GC during the major cycle

  - Before marking, the write barrier is disabled and fresh allocations are UNMARKED

  - Change to ephemeron logic: per-cycle ephemeron data structure reset is now done
    at marking start rather than at the start of the cycle. (In particular, this makes
    orphaned work simpler to deal with)

* Re-enable parallel tests

Includes several systhreads tests that rely on the interaction between
systhreads, domains and backup threads.

* Add a new test (tests/parallel/churn.ml) stressing cross-domain promotion

Author: stedolan

ocaml/otherlibs/str/str.ml

@@ -607,16 +607,6 @@ external re_search_forward: regexp -> string -> int -> int array
 external re_search_backward: regexp -> string -> int -> int array
      = "re_search_backward"
 
-module Domain = struct
-  module DLS = struct
-
-    (* CR ocaml 5 domains: Remove this proxy and use the real Domain.DLS *)
-    let[@inline always] new_key f = ref (f ())
-    let[@inline always] set k s = k := s
-    let[@inline always] get k = !k
-  end
-end
-
 let last_search_result_key = Domain.DLS.new_key (fun () -> [||])
 
 let string_match re s pos =

ocaml/runtime/caml/domain.h

@@ -109,14 +109,14 @@ int caml_try_run_on_all_domains_with_spin_work(
   int sync,
   void (*handler)(caml_domain_state*, void*, int, caml_domain_state**),
   void* data,
-  void (*leader_setup)(caml_domain_state*),
+  void (*leader_setup)(caml_domain_state*, void*),
   /* return nonzero if there may still be useful work to do while spinning */
   int (*enter_spin_callback)(caml_domain_state*, void*),
   void* enter_spin_data);
 int caml_try_run_on_all_domains(
   void (*handler)(caml_domain_state*, void*, int, caml_domain_state**),
   void*,
-  void (*leader_setup)(caml_domain_state*));
+  void (*leader_setup)(caml_domain_state*, void*));
 
 /* Function naming conventions for STW callbacks and STW critical sections.
 

ocaml/runtime/caml/major_gc.h

@@ -30,6 +30,7 @@ extern gc_phase_t caml_gc_phase;
 Caml_inline int caml_marking_started(void) {
   return caml_gc_phase != Phase_sweep_main;
 }
+extern atomic_uintnat caml_gc_mark_phase_requested;
 
 intnat caml_opportunistic_major_work_available (caml_domain_state*);
 void caml_opportunistic_major_collection_slice (intnat);
@@ -45,7 +46,7 @@ void caml_teardown_major_gc(void);
 void caml_darken(void*, value, volatile value* ignored);
 void caml_darken_cont(value);
 void caml_mark_root(value, value*);
-void caml_empty_mark_stack(void);
+void caml_mark_roots_stw(int, caml_domain_state**);
 void caml_finish_major_cycle(int force_compaction);
 #ifdef DEBUG
 int caml_mark_stack_is_empty(void);

ocaml/runtime/caml/minor_gc.h

@@ -87,7 +87,7 @@ extern void caml_realloc_ephe_ref_table (struct caml_ephe_ref_table *);
 extern void caml_realloc_custom_table (struct caml_custom_table *);
 struct caml_minor_tables* caml_alloc_minor_tables(void);
 void caml_free_minor_tables(struct caml_minor_tables*);
-void caml_empty_minor_heap_setup(caml_domain_state* domain);
+void caml_empty_minor_heap_setup(caml_domain_state* domain, void*);
 
 #ifdef DEBUG
 extern int caml_debug_is_minor(value val);

ocaml/runtime/domain.c

@@ -1031,8 +1031,6 @@ struct domain_startup_params {
 
 static void* backup_thread_func(void* v)
 {
-  // single-domain hack
-  caml_fatal_error("backup thread not allowed to run");
   dom_internal* di = (dom_internal*)v;
   uintnat msg;
   struct interruptor* s = &di->interruptor;
@@ -1591,7 +1589,7 @@ int caml_try_run_on_all_domains_with_spin_work(
   int sync,
   void (*handler)(caml_domain_state*, void*, int, caml_domain_state**),
   void* data,
-  void (*leader_setup)(caml_domain_state*),
+  void (*leader_setup)(caml_domain_state*, void*),
   int (*enter_spin_callback)(caml_domain_state*, void*),
   void* enter_spin_data)
 {
@@ -1659,7 +1657,7 @@ int caml_try_run_on_all_domains_with_spin_work(
   }
 
   if( leader_setup ) {
-    leader_setup(domain_state);
+    leader_setup(domain_state, data);
   }
 
 #ifdef DEBUG
@@ -1726,7 +1724,7 @@ int caml_try_run_on_all_domains_with_spin_work(
 int caml_try_run_on_all_domains(
   void (*handler)(caml_domain_state*, void*, int, caml_domain_state**),
   void* data,
-  void (*leader_setup)(caml_domain_state*))
+  void (*leader_setup)(caml_domain_state*, void*))
 {
   return
       caml_try_run_on_all_domains_with_spin_work(1,
@@ -1738,7 +1736,7 @@ int caml_try_run_on_all_domains(
 int caml_try_run_on_all_domains_async(
   void (*handler)(caml_domain_state*, void*, int, caml_domain_state**),
   void* data,
-  void (*leader_setup)(caml_domain_state*))
+  void (*leader_setup)(caml_domain_state*, void*))
 {
   return
       caml_try_run_on_all_domains_with_spin_work(0,