Commit Graph

23 Commits

Author SHA1 Message Date
Eric Holk
58fdbdc4ef Avoid some more copies. 2012-05-30 17:39:53 -07:00
Eric Holk
f0c345841c Added a factory version of par::mapi, which avoids the need for share_arc, and copies arcs between tasks better. 2012-05-30 17:39:53 -07:00
Eric Holk
f05040f17a Use std::deque now that the bugs are fixed. Also, avoid a few copies. 2012-05-30 17:39:53 -07:00
Eric Holk
ccb54f0ce0 Move par into libstd. Closes #2441. 2012-05-30 12:03:51 -07:00
Eric Holk
f49b891ce5 Include capture clauses when generating metadata.
This may fix the bug that's blocking Issue #2441.
2012-05-30 12:03:51 -07:00
Eric Holk
aa2efc05ea Adding shared_arc to hide some of the tedium of sharing an arc between tasks.
Also added some asserts and logging to trans.

Modified graph500 to use the shared_arc, but this unfortunately doesn't work due to #2444.
2012-05-29 10:43:19 -07:00
Brian Anderson
45d0fbf6a1 bench: Increase the difficulty of some benchmarks 2012-05-26 16:00:35 -07:00
Brian Anderson
4fec4cd8f5 bench: Add hard mode to benchmarks. Activate with RUST_BENCH.
RUST_BENCH is on automatically when running `make perf`
2012-05-23 22:55:41 -07:00
Brian Anderson
8caf1403be bench: Reduce some constants to make graph500-bfs faster 2012-05-22 18:20:23 -07:00
Eric Holk
037a166ea7 Updating pbfs for new kind rules. 2012-05-22 15:51:11 -07:00
Eric Holk
b62d92c993 Removed accidental machine size dependencies. 2012-05-22 15:31:39 -07:00
Eric Holk
44bea03614 Updating bfs to work with the new random methods. 2012-05-22 15:31:39 -07:00
Eric Holk
5a4e2ae706 More unsafe pointers to share immutable structures between tasks. This version has a 2.8 to 3x speedup! 2012-05-22 15:31:39 -07:00
Eric Holk
7f05f088c7 Be less careful about checking that lengths make sense, for slightly more performance. 2012-05-22 15:31:39 -07:00
Eric Holk
77f6099606 Using unsafe pointers to views to try to reduce some vector copy and allocation time. Doesn't seem to have made much difference. 2012-05-22 15:31:39 -07:00
Eric Holk
ea889742d7 Using unsafe pointers to share closures.
This prevents the tons of copying problems we were having before, which means we only have a 5x slowdown now.
2012-05-22 15:31:39 -07:00
Eric Holk
65abe2c6dc Using libstd future instead.
Turning off sequential versions for tracing.
2012-05-22 15:31:38 -07:00
Eric Holk
cba2761cc4 A new version of the sequential BFS. This uses the same approach as the parallel one, but without using the parallel vector oeprators. 2012-05-22 15:31:38 -07:00
Eric Holk
4312fa44b9 BFS now tries a number of search keys. 2012-05-22 15:31:38 -07:00
Eric Holk
a79f5174ea A parallel BFS. 2012-05-22 15:31:38 -07:00
Eric Holk
a785f3fc95 Adding a module with parallel vector operations.
This should go in libstd, but currently resolve bugs make this not
work.
2012-05-22 15:31:38 -07:00
Eric Holk
d485f23a1a Added most of the Graph500 validation routine. It's very slow. 2012-05-22 15:31:38 -07:00
Eric Holk
f5b9ebffa0 Starting on a BFS benchmark, but ran into problems with the deque module, so I used ports and channels as a queue in the simple sequential algorithm. 2012-05-22 15:31:38 -07:00