Changeset b9376fe for libcfa/src

Timestamp:

May 5, 2021, 2:12:36 PM (3 years ago)

Author:

Peter A. Buhr <pabuhr@…>

Branches:

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

Children:

7f54356, a67c5b6

Parents:

4026d1be (diff), f302d80 (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.

Message:

Merge branch 'master' of plg.uwaterloo.ca:software/cfa/cfa-cc

Location:

libcfa/src

Files:

• Makefile.am (modified) (1 diff)
• concurrency/ready_queue.cfa (modified) (5 diffs)
• concurrency/ready_subqueue.hfa (modified) (3 diffs)
• containers/list2.hfa (added)

libcfa/src/Makefile.am

@@ -59 +59 @@
         concurrency/iofwd.hfa \
         containers/list.hfa \
+        containers/list2.hfa \
         containers/queueLockFree.hfa \
         containers/stackLockFree.hfa \

libcfa/src/concurrency/ready_queue.cfa

@@ -398 +398 @@
 
                 if(proc->rdq.target == -1u) {
-                        proc->rdq.target = __tls_rand() % lanes.count;
-                        unsigned it1  = proc->rdq.itr;
-                        unsigned it2  = proc->rdq.itr + 1;
-                        unsigned idx1 = proc->rdq.id + (it1 % READYQ_SHARD_FACTOR);
-                        unsigned idx2 = proc->rdq.id + (it2 % READYQ_SHARD_FACTOR);
+                        _Static_assert(READYQ_SHARD_FACTOR == 2);
+                        unsigned idx1 = proc->rdq.id + 0;
+                        unsigned idx2 = proc->rdq.id + 1;
                         unsigned long long tsc1 = ts(lanes.data[idx1]);
                         unsigned long long tsc2 = ts(lanes.data[idx2]);
-                        proc->rdq.cutoff = min(tsc1, tsc2);
-                        if(proc->rdq.cutoff == 0) proc->rdq.cutoff = -1ull;
+                        proc->rdq.target = __tls_rand() % lanes.count;
+
+                        // WARNING: std::min is polymorphic and therefore causes 500% slowdown instead of the expected 2%
+                        proc->rdq.cutoff = tsc1 < tsc2 ? tsc1 : tsc2;
                 }
                 else {
@@ -418 +418 @@
 
                 for(READYQ_SHARD_FACTOR) {
-                        unsigned i = proc->rdq.id + (--proc->rdq.itr % READYQ_SHARD_FACTOR);
+                        unsigned i = proc->rdq.id + (proc->rdq.itr++ % READYQ_SHARD_FACTOR);
                         if($thread * t = try_pop(cltr, i __STATS(, __tls_stats()->ready.pop.local))) return t;
                 }
@@ -469 +469 @@
         // Actually pop the list
         struct $thread * thrd;
-        thrd = pop(lane);
+        unsigned long long tsv;
+        [thrd, tsv] = pop(lane);
 
         /* paranoid */ verify(thrd);
@@ -481 +482 @@
 
         #if defined(USE_WORK_STEALING)
-                lanes.tscs[w].tv = thrd->link.ts;
+                lanes.tscs[w].tv = tsv;
         #endif
 
@@ -663 +664 @@
                         while(!is_empty(lanes.data[idx])) {
                                 struct $thread * thrd;
-                                thrd = pop(lanes.data[idx]);
+                                unsigned long long _;
+                                [thrd, _] = pop(lanes.data[idx]);
 
                                 push(cltr, thrd);

libcfa/src/concurrency/ready_subqueue.hfa

@@ -53 +53 @@
 // Push a thread onto this lane
 // returns true of lane was empty before push, false otherwise
-void push( __intrusive_lane_t & this, $thread * node ) {
+static inline void push( __intrusive_lane_t & this, $thread * node ) {
         /* paranoid */ verify( node->link.next == 0p );
         /* paranoid */ verify( node->link.ts   == 0  );
@@ -77 +77 @@
 // returns popped
 // returns true of lane was empty before push, false otherwise
-$thread * pop( __intrusive_lane_t & this ) {
+static inline [* $thread, unsigned long long] pop( __intrusive_lane_t & this ) {
         /* paranoid */ verify( this.anchor.next != 0p );
         /* paranoid */ verify( this.anchor.ts   != 0  );
 
         // Get the relevant nodes locally
+        unsigned long long ts = this.anchor.ts;
         $thread * node = this.anchor.next;
         this.anchor.next = node->link.next;
@@ -94 +95 @@
         /* paranoid */ verify( node->link.next == 0p );
         /* paranoid */ verify( node->link.ts   == 0  );
-        return node;
+        return [node, ts];
 }