Changeset ec43cf9

Timestamp:

Apr 5, 2021, 7:06:37 PM (2 years ago)

Author:

Thierry Delisle <tdelisle@…>

Branches:

ADT, arm-eh, ast-experimental, enum, forall-pointer-decay, jacob/cs343-translation, master, new-ast-unique-expr, pthread-emulation, qualifiedEnum

Children:

7039ab9, cd59d28

Parents:

a5a01faa

Message:

Kernel now keeps track of the how many threads each processor has contributed.
This isn't useful if it is not sampled regularly but it's a first step.

Location:

libcfa/src/concurrency

Files:

• kernel.cfa (modified) (4 diffs)
• stats.cfa (modified) (4 diffs)
• stats.hfa (modified) (1 diff)

libcfa/src/concurrency/kernel.cfa

@@ -357 +357 @@
                                 break RUNNING;
                         case TICKET_UNBLOCK:
+                                #if !defined(__CFA_NO_STATISTICS__)
+                                        __tls_stats()->ready.threads.threads++;
+                                #endif
                                 // This is case 2, the racy case, someone tried to run this thread before it finished blocking
                                 // In this case, just run it again.
@@ -370 +373 @@
 
         __cfadbg_print_safe(runtime_core, "Kernel : core %p finished running thread %p\n", this, thrd_dst);
+
+        #if !defined(__CFA_NO_STATISTICS__)
+                __tls_stats()->ready.threads.threads--;
+        #endif
 
         /* paranoid */ verify( ! __preemption_enabled() );
@@ -432 +439 @@
         if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
 
+        // Dereference the thread now because once we push it, there is not guaranteed it's still valid.
+        struct cluster * cl = thrd->curr_cluster;
+
         ready_schedule_lock();
-                // Dereference the thread now because once we push it, there is not guaranteed it's still valid.
-                struct cluster * cl = thrd->curr_cluster;
-
                 // push the thread to the cluster ready-queue
                 push( cl, thrd );
@@ -444 +451 @@
                 __wake_one( cl );
         ready_schedule_unlock();
+
+        #if !defined(__CFA_NO_STATISTICS__)
+                if( kernelTLS().this_stats ) {
+                        __tls_stats()->ready.threads.threads++;
+                }
+                else {
+                        __atomic_fetch_add(&cl->stats->ready.threads.threads, 1, __ATOMIC_RELAXED);
+                }
+        #endif
 
         /* paranoid */ verify( ! __preemption_enabled() );

libcfa/src/concurrency/stats.cfa

@@ -23 +23 @@
                 stats->ready.pick.pop .lsuccess = 0;
                 stats->ready.threads.migration = 0;
+                stats->ready.threads.threads   = 0;
                 stats->ready.sleep.halts   = 0;
                 stats->ready.sleep.cancels = 0;
@@ -61 +62 @@
                 __atomic_fetch_add( &cltr->ready.pick.pop .lsuccess, proc->ready.pick.pop .lsuccess, __ATOMIC_SEQ_CST ); proc->ready.pick.pop .lsuccess = 0;
                 __atomic_fetch_add( &cltr->ready.threads.migration , proc->ready.threads.migration , __ATOMIC_SEQ_CST ); proc->ready.threads.migration  = 0;
+                __atomic_fetch_add( &cltr->ready.threads.threads   , proc->ready.threads.threads   , __ATOMIC_SEQ_CST ); proc->ready.threads.threads    = 0;
                 __atomic_fetch_add( &cltr->ready.sleep.halts       , proc->ready.sleep.halts       , __ATOMIC_SEQ_CST ); proc->ready.sleep.halts        = 0;
                 __atomic_fetch_add( &cltr->ready.sleep.cancels     , proc->ready.sleep.cancels     , __ATOMIC_SEQ_CST ); proc->ready.sleep.cancels      = 0;
@@ -97 +99 @@
                         __cfaabi_bits_print_safe( STDOUT_FILENO,
                                 "----- %s \"%s\" (%p) - Ready Q Stats -----\n"
-                                "- total threads  : %'15" PRIu64 "run, %'15" PRIu64 "schd (%'" PRIu64 "ext, %'" PRIu64 "mig )\n"
+                                "- total threads  : %'15" PRIu64 "run, %'15" PRIu64 "schd (%'" PRIu64 "ext, %'" PRIu64 "mig, %'" PRId64 " )\n"
                                 "- push avg probe : %'3.2lf, %'3.2lfl (%'15" PRIu64 " attempts, %'15" PRIu64 " locals)\n"
                                 "- ext  avg probe : %'3.2lf, %'3.2lfl (%'15" PRIu64 " attempts, %'15" PRIu64 " locals)\n"
@@ -106 +108 @@
                                 , ready.pick.pop.success
                                 , ready.pick.push.success + ready.pick.ext.success
-                                , ready.pick.ext.success, ready.threads.migration
+                                , ready.pick.ext.success, ready.threads.migration, ready.threads.threads
                                 , push_len, lpush_len, ready.pick.push.attempt, ready.pick.push.local
                                 , ext_len , lext_len , ready.pick.ext .attempt, ready.pick.ext .local

libcfa/src/concurrency/stats.hfa

@@ -68 +68 @@
                 struct {
                         volatile uint64_t migration;
+                        volatile  int64_t threads; // number of threads in the system, includes only local change
                 } threads;
                 struct {