mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Added infrastructure to spin for a bit on recv. A spin count > 0 makes bench/pingpong.rs about 10x faster, but makes msgsend-ring-pipes unbearably slow.

Browse Source
This commit is contained in:
Eric Holk 2012-07-24 09:35:44 -07:00
parent 35576168dc
commit f5be06fa1f
2 changed files with 18 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
src/libcore/pipes.rs
View File

@ -20,6 +20,8 @@ export select, select2, selecti, select2i, selectable;
export spawn_service, spawn_service_recv;
export stream, port, chan, shared_chan, port_set, channel;
const SPIN_COUNT: uint = 0;
macro_rules! move {
{ $x:expr } => { unsafe { let y <- *ptr::addr_of($x); y } }
}
@ -272,7 +274,7 @@ unsafe fn get_buffer<T: send>(p: *packet_header) -> ~buffer<T> {
class buffer_resource<T: send> {
let buffer: ~buffer<T>;
new(+b: ~buffer<T>) {
let p = ptr::addr_of(*b);
//let p = ptr::addr_of(*b);
//#error("take %?", p);
atomic_add_acq(b.header.ref_count, 1);
self.buffer = b;
@ -280,7 +282,7 @@ class buffer_resource<T: send> {
drop unsafe {
let b = move!{self.buffer};
let p = ptr::addr_of(*b);
//let p = ptr::addr_of(*b);
//#error("drop %?", p);
let old_count = atomic_sub_rel(b.header.ref_count, 1);
//let old_count = atomic_xchng_rel(b.header.ref_count, 0);
@ -345,6 +347,7 @@ fn try_recv<T: send, Tbuffer: send>(-p: recv_packet_buffered<T, Tbuffer>)
rustrt::task_clear_event_reject(this);
p.header.blocked_task = some(this);
let mut first = true;
let mut count = SPIN_COUNT;
loop {
rustrt::task_clear_event_reject(this);
let old_state = swap_state_acq(p.header.state,
@ -352,7 +355,18 @@ fn try_recv<T: send, Tbuffer: send>(-p: recv_packet_buffered<T, Tbuffer>)
alt old_state {
empty {
#debug("no data available on %?, going to sleep.", p_);
wait_event(this);
if count == 0 {
wait_event(this);
}
else {
count -= 1;
// FIXME (#524): Putting the yield here destroys a lot
// of the benefit of spinning, since we still go into
// the scheduler at every iteration. However, without
// this everything spins too much because we end up
// sometimes blocking the thing we are waiting on.
task::yield();
}
#debug("woke up, p.state = %?", copy p.header.state);
}
blocked {

2
src/test/bench/pingpong.rs
View File

@ -137,7 +137,7 @@ fn main() {
let unbounded = do timeit { unbounded(count) };
io::println(#fmt("count: %?\n", count));
io::println(#fmt("bounded: %? s\t(%? μs/message)",
io::println(#fmt("bounded: %? s\t(%? μs/message)",
bounded, bounded * 1000000. / (count as float)));
io::println(#fmt("unbounded: %? s\t(%? μs/message)",
unbounded, unbounded * 1000000. / (count as float)));