Changeset b808625 for libcfa/src

Timestamp:

May 19, 2021, 1:37:49 PM (4 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:

77f1265

Parents:

9cac0da

Message:

Added option to ready-queue to push ignoring locality.

Location:

libcfa/src/concurrency

Files:

• kernel.cfa (modified) (2 diffs)
• kernel_private.hfa (modified) (1 diff)
• ready_queue.cfa (modified) (6 diffs)

libcfa/src/concurrency/kernel.cfa

@@ -544 +544 @@
         /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
 
-
+        const bool local = thrd->state != Start;
         if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
 
@@ -552 +552 @@
 
         // push the thread to the cluster ready-queue
-        push( cl, thrd );
+        push( cl, thrd, local );
 
         // variable thrd is no longer safe to use

libcfa/src/concurrency/kernel_private.hfa

@@ -261 +261 @@
 // push thread onto a ready queue for a cluster
 // returns true if the list was previously empty, false otherwise
-__attribute__((hot)) void push(struct cluster * cltr, struct $thread * thrd);
+__attribute__((hot)) void push(struct cluster * cltr, struct $thread * thrd, bool local);
 
 //-----------------------------------------------------------------------

libcfa/src/concurrency/ready_queue.cfa

@@ -249 +249 @@
         }
 
-        __attribute__((hot)) void push(struct cluster * cltr, struct $thread * thrd) with (cltr->ready_queue) {
+        __attribute__((hot)) void push(struct cluster * cltr, struct $thread * thrd, bool push_local) with (cltr->ready_queue) {
                 __cfadbg_print_safe(ready_queue, "Kernel : Pushing %p on cluster %p\n", thrd, cltr);
 
-                const bool external = (!kernelTLS().this_processor) || (cltr != kernelTLS().this_processor->cltr);
+                const bool external = !push_local || (!kernelTLS().this_processor) || (cltr != kernelTLS().this_processor->cltr);
                 /* paranoid */ verify(external || kernelTLS().this_processor->rdq.id < lanes.count );
 
@@ -279 +279 @@
                 push(lanes.data[i], thrd);
 
-                        // Unlock and return
-                        __atomic_unlock( &lanes.data[i].lock );
+                // Unlock and return
+                __atomic_unlock( &lanes.data[i].lock );
 
                 // Mark the current index in the tls rng instance as having an item
@@ -336 +336 @@
 #endif
 #if defined(USE_WORK_STEALING)
-        __attribute__((hot)) void push(struct cluster * cltr, struct $thread * thrd) with (cltr->ready_queue) {
+        __attribute__((hot)) void push(struct cluster * cltr, struct $thread * thrd, bool push_local) with (cltr->ready_queue) {
                 __cfadbg_print_safe(ready_queue, "Kernel : Pushing %p on cluster %p\n", thrd, cltr);
 
                 // #define USE_PREFERRED
                 #if !defined(USE_PREFERRED)
-                const bool external = (!kernelTLS().this_processor) || (cltr != kernelTLS().this_processor->cltr);
+                const bool external = !push_local || (!kernelTLS().this_processor) || (cltr != kernelTLS().this_processor->cltr);
                 /* paranoid */ verify(external || kernelTLS().this_processor->rdq.id < lanes.count );
                 #else
                         unsigned preferred = thrd->preferred;
-                        const bool external = (!kernelTLS().this_processor) || preferred == -1u || thrd->curr_cluster != cltr;
+                        const bool external = push_local || (!kernelTLS().this_processor) || preferred == -1u || thrd->curr_cluster != cltr;
                         /* paranoid */ verifyf(external || preferred < lanes.count, "Invalid preferred queue %u for %u lanes", preferred, lanes.count );
 
@@ -365 +365 @@
                         else {
                                 #if !defined(USE_PREFERRED)
-                                processor * proc = kernelTLS().this_processor;
-                                unsigned r = proc->rdq.its++;
-                                i =  proc->rdq.id + (r % READYQ_SHARD_FACTOR);
-                #else
+                                        processor * proc = kernelTLS().this_processor;
+                                        unsigned r = proc->rdq.its++;
+                                        i =  proc->rdq.id + (r % READYQ_SHARD_FACTOR);
+                                #else
                                         i = start + (r++ % READYQ_SHARD_FACTOR);
                                 #endif
@@ -378 +378 @@
                 push(lanes.data[i], thrd);
 
-                        // Unlock and return
-                        __atomic_unlock( &lanes.data[i].lock );
+                // Unlock and return
+                __atomic_unlock( &lanes.data[i].lock );
 
                 #if !defined(__CFA_NO_STATISTICS__)
@@ -666 +666 @@
                                 [thrd, _] = pop(lanes.data[idx]);
 
-                                push(cltr, thrd);
+                                push(cltr, thrd, true);
 
                                 // for printing count the number of displaced threads