Question on shared memory concurrency

[Andy Valencia]

On Monday, 4 March 2024 at 18:08:52 UTC, Andy Valencia wrote:
> For any other newbie dlang voyagers, here's a version which 
> works as expected using the system memory allocator.  On my 
> little i7 I get 1.48 secs wallclock with 5.26 CPU seconds.
>...

Using a technique I found in a unit test in std/concurrency.d, I 
managed to share process memory without GC.  It counted up to 
1,000,000,000 on my low end i7 in:

real    0m15.666s
user    0m59.913s
sys     0m0.004s



import core.atomic : atomicFetchAdd;
import std.concurrency : spawn, send, receiveOnly, ownerTid;
import core.thread : Thread;

const uint NSWEPT = 1_000_000_000;
const uint NCPU = 4;

void
doadd()
{
     auto val = receiveOnly!(shared(int)[]);
     for (uint count = 0; count < NSWEPT/NCPU; ++count) {
         atomicFetchAdd(val[0], 1);
     }
     ownerTid.send(true);
}

void
main()
{
     static shared int[] val = new shared(int)[1];

     // Parallel workers
     for (int x = 0; x < NCPU; ++x) {
         auto tid = spawn(&doadd);
         tid.send(val);
     }

     // Pick up all completed workers
     for (int x = 0; x < NCPU; ++x) {
         receiveOnly!(bool);
     }
     assert(val[0] == NSWEPT);
}