Changeset 953827a

Timestamp:

Jun 21, 2021, 3:42:47 PM (3 years ago)

Author:

Thierry Delisle <tdelisle@…>

Branches:

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

Children:

6ba6846, 929d925

Parents:

5614552a

Message:

Fixed cpu work stealing to properly fall back on other nodes.

File:

• libcfa/src/concurrency/ready_queue.cfa (modified) (3 diffs)

libcfa/src/concurrency/ready_queue.cfa

@@ -20 +20 @@
 
 
-#define USE_RELAXED_FIFO
+// #define USE_RELAXED_FIFO
 // #define USE_WORK_STEALING
+#define USE_CPU_WORK_STEALING
 
 #include "bits/defs.hfa"
@@ -341 +342 @@
                         }
                         proc->rdq.cutoff = min;
-                        proc->rdq.target = (map.start * READYQ_SHARD_FACTOR) + (__tls_rand() % (map.count* READYQ_SHARD_FACTOR));
+
+                        /* paranoid */ verify(lanes.count < 65536); // The following code assumes max 65536 cores.
+                        /* paranoid */ verify(map.count < 65536); // The following code assumes max 65536 cores.
+                        uint64_t chaos = __tls_rand();
+                        uint64_t high_chaos = (chaos >> 32);
+                        uint64_t  mid_chaos = (chaos >> 16) & 0xffff;
+                        uint64_t  low_chaos = chaos & 0xffff;
+
+                        unsigned me = map.self;
+                        unsigned cpu_chaos = map.start + (mid_chaos % map.count);
+                        bool global = cpu_chaos == me;
+
+                        if(global) {
+                                proc->rdq.target = high_chaos % lanes.count;
+                        } else {
+                                proc->rdq.target = (cpu_chaos * READYQ_SHARD_FACTOR) + (low_chaos % READYQ_SHARD_FACTOR);
+                                /* paranoid */ verify(proc->rdq.target >= (map.start * READYQ_SHARD_FACTOR));
+                                /* paranoid */ verify(proc->rdq.target <  ((map.start + map.count) * READYQ_SHARD_FACTOR));
+                        }
+
+                        /* paranoid */ verify(proc->rdq.target != -1u);
                 }
                 else {
@@ -378 +399 @@
                 processor * const proc = kernelTLS().this_processor;
                 unsigned last = proc->rdq.last;
+                if(last != -1u) {
+                        struct $thread * t = try_pop(cltr, last __STATS(, __tls_stats()->ready.pop.steal));
+                        if(t) return t;
+                        proc->rdq.last = -1u;
+                }
 
                 unsigned i = __tls_rand() % lanes.count;