Commit Graph

126 Commits

Author SHA1 Message Date
Olli Saarikivi
0cfe2dcffb Add allpairs allreduce test
To support this include separate source and destination offsets in the trigger.
Add functions for getting the rank and world size from a communicator.
2023-03-21 19:00:13 +00:00
Saeed Maleki
e2ee8d80b9 perf fix for multi-node allgather 2023-03-21 06:26:12 +00:00
Saeed Maleki
9df5077015 Merge pull request #15 from microsoft/chhwang/fix-trigger
Thread-safe trigger
2023-03-20 16:15:07 -07:00
Saeed Maleki
8b30121240 Merge branch 'main' into chhwang/fix-trigger 2023-03-20 23:12:58 +00:00
Saeed Maleki
fbc7b039e8 Merge pull request #17 from microsoft/saemal/fix-trigger
New trigger FIFO
2023-03-20 16:05:56 -07:00
Saeed Maleki
7cb2903799 some comment check ins 2023-03-20 21:07:58 +00:00
Saeed Maleki
93afed3e54 new allgather algorithm with both DMA and IB on a single node 2023-03-19 21:53:36 +00:00
Saeed Maleki
8a1ec28ff1 single node allgather works very well 2023-03-19 19:27:17 +00:00
Saeed Maleki
3e8f6758e5 both allgather algorithms 2023-03-19 06:35:40 +00:00
Saeed Maleki
17cbc84a14 both allgather algorithms 2023-03-19 06:35:32 +00:00
Saeed Maleki
5493e22633 fixed multinode bug 2023-03-19 06:09:07 +00:00
Saeed Maleki
4efc6e98db incorrect access fixed 2023-03-19 01:26:30 +00:00
Saeed Maleki
a485a7f238 single node works fine -- multinode is problematic 2023-03-19 01:08:05 +00:00
Saeed Maleki
9cc21f70e6 redesigning fifo 2023-03-17 22:51:11 +00:00
Saeed Maleki
73df12358f Merge branch 'main' of https://github.com/microsoft/mscclpp into main 2023-03-17 17:54:17 +00:00
Saeed Maleki
e86df92fa5 fixed a typo in debugging information 2023-03-17 17:52:53 +00:00
Changho Hwang
67dbbd1692 Thread-safe trigger 2023-03-17 09:46:23 +00:00
Saeed Maleki
2061ea91f7 Add allgather_test (#14) 2023-03-17 12:55:20 +08:00
Saeed Maleki
267d1f4f63 documentation for mscclpp.h 2023-03-14 21:08:39 +00:00
Saeed Maleki
88bbec997b Merge pull request #13 from microsoft/chhwang/shared-fifo
Shared FIFO
2023-03-14 12:02:14 -07:00
Changho Hwang
dc41c58769 Alloc proxy states on demand 2023-03-14 10:05:56 +00:00
Changho Hwang
c2859d258c Use aligned ld/st 2023-03-14 09:22:28 +00:00
Changho Hwang
135520a14a cleanups 2023-03-14 09:21:52 +00:00
Changho Hwang
aacee9727b trigger wrappers 2023-03-14 09:14:51 +00:00
Changho Hwang
75ec82d257 Store fifo tail in proxy state 2023-03-14 09:00:38 +00:00
Changho Hwang
e89d154503 Check run state periodically 2023-03-14 08:38:55 +00:00
Changho Hwang
ae01fa4958 Remove mscclpp_net.h and net.h 2023-03-14 08:32:19 +00:00
Saeed Maleki
2279a690d1 mscclpp_net.h is not required anywhere 2023-03-14 05:38:15 +00:00
Saeed Maleki
e000eb9177 some compilation clean up 2023-03-14 05:26:54 +00:00
Saeed Maleki
2bcf418b13 more comments 2023-03-14 03:36:38 +00:00
Saeed Maleki
ab9298d6e0 fixed the bits for trigger 2023-03-13 23:21:27 +00:00
Changho Hwang
9b124cabdb cleanup 2023-03-13 14:27:29 +00:00
Changho Hwang
86dd8e2e17 Remove unused macro 2023-03-13 14:21:37 +00:00
Changho Hwang
e357beef00 One fifo per proxy 2023-03-13 14:19:36 +00:00
Saeed Maleki
ea7134549e vector instructions for trigger 2023-03-13 07:02:26 +00:00
Changho Hwang
1be76d128d 128-bit trigger 2023-03-10 10:49:36 +00:00
Changho Hwang
85d92961a3 Remove MPI dependency 2023-03-10 08:26:38 +00:00
Changho Hwang
798759a225 Update README.md 2023-03-10 16:16:43 +08:00
Saeed Maleki
9b9706155e Merge pull request #11 from microsoft/chhwang/single-thread-proxy
Single thread proxy
2023-03-09 20:39:01 -08:00
Ubuntu
8f2831330c a few todos + some clean up in the test 2023-03-10 04:34:14 +00:00
Changho Hwang
6ac3c4c90f Relaxed sync 2023-03-09 07:24:09 +00:00
Changho Hwang
38a2adfe97 Single IB proxy per IB dev 2023-03-09 05:11:08 +00:00
Changho Hwang
e3fde881bc Single P2P proxy per GPU 2023-03-09 04:43:56 +00:00
Changho Hwang
3c800915b7 Merge pull request #10 from microsoft/saemal/proxyfifo
FIFO for proxy triggers
2023-03-09 10:55:29 +08:00
Saeed Maleki
8c4ab6dad6 merged with main 2023-03-09 02:06:04 +00:00
Saeed Maleki
6c3197a492 fifo for ib works as well 2023-03-09 02:02:14 +00:00
Saeed Maleki
160060ec77 fifo works now 2023-03-08 20:10:09 +00:00
Changho Hwang
2beedb342a Merge pull request #8 from microsoft/chhwang/bug-fix
Type bug fix
2023-03-07 20:47:17 +08:00
Changho Hwang
5403508ac4 Type bug fix 2023-03-07 12:41:56 +00:00
Changho Hwang
00b1fd5f42 Merge pull request #6 from microsoft/chhwang/ib-proxy-merge
Merge two IB proxy modes
2023-03-07 13:51:54 +08:00