microsoft
/
msccl
зеркало из https://github.com/microsoft/msccl.git
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность
399 коммитов 16 Ветки 44 Теги 3,4 MiB
Граф коммитов

399 Коммитов

Автор SHA1 Сообщение Дата
Olli Saarikivi 1306d086b5
Merge pull request #6 from parasailteam/olli/interpreter-template 
Extract shared parts of interpreter code into SCKLFunction template class.
 2021-04-09 13:01:29 -07:00
Olli Saarikivi 9327d76850 Extract shared parts of interpreter code 
Simple, LL and LL128 specify the differing parts with wrappers for the
*Primitives structs.
A SCKLFunction class template contains most of the interpreter logic.
 2021-04-09 02:52:41 -07:00
Saeed Maleki 252a9773a1 has_dependence 2021-04-09 00:56:21 +00:00
Saeed Maleki d83e777ac3 all protocols work 2021-04-08 23:37:17 +00:00
Saeed Maleki 007fc390c7 added sckl_interpreter 2021-04-08 06:20:41 +00:00
Saeed Maleki 1da0956a4a works 2021-04-08 06:01:36 +00:00
Saeed Maleki 8d8bc3c201 more progress 2021-04-07 21:04:43 +00:00
Saeed Maleki 0d1af8e26c added allgather support 2021-04-07 18:30:47 +00:00
Saeed Maleki 308b5abb88 not complete 2021-04-06 21:35:02 +00:00
Saeed Maleki e785648283
Merge pull request #5 from parasailteam/channel_management 
Channel management
 2021-04-06 10:52:48 -07:00
Saeed Maleki 5026d295e4 works across node as well 2021-04-06 06:04:14 +00:00
Saeed Maleki 2a7caa2bd0 works within a node 2021-04-05 23:50:48 +00:00
Saeed Maleki dea2ed3c1f bug fixes 2021-04-05 21:04:31 +00:00
Saeed Maleki 1968e78ce4 partially working 2021-04-05 20:24:09 +00:00
Saeed Maleki 4788b47b4d debugging 2021-04-05 18:31:39 +00:00
Saeed Maleki 90f3c3cf0d fixed a bug regarding rid 2021-04-03 03:27:55 +00:00
Saeed Maleki d01feb2275 compilable 2021-04-03 03:03:37 +00:00
Saeed Maleki d3dcbe7495 spacing fixes 2021-04-02 17:33:26 +00:00
Saeed Maleki 7e872304a1
Merge pull request #4 from parasailteam/cross_threadblocks 
Cross threadblocks
 2021-04-01 22:52:12 -07:00
Saeed Maleki 295c627e7f added a fence before the flag write 2021-04-02 05:48:58 +00:00
Saeed Maleki f2cf4f2858 fixed madan's comment regarding 3-tuple. also added a guard to run sckl async only when all are sckl algorithms 2021-04-02 02:51:34 +00:00
Saeed Maleki c29eca1568 added padding for sckl flags 2021-04-01 19:19:36 +00:00
Saeed Maleki 2cb6511754 added a check for NET_SHARED_BUFFERS 2021-04-01 18:59:09 +00:00
Saeed Maleki 3d60bd12d8 compilable version 2021-04-01 04:07:01 +00:00
Saeed Maleki b425f1cd91
Merge pull request #3 from parasailteam/across_node_setup 
Across node setup
 2021-03-31 18:00:08 -07:00
Saeed Maleki 80ba01581c fixed spacing issues 2021-04-01 00:59:25 +00:00
Saeed Maleki 7af77c04b8 done on 64 gpus 2021-03-31 01:17:19 +00:00
Saeed Maleki f7837e5614 works/msrhyper-weka/saemal/mpi/bin/mpirun -np 16 -N 8 -hostfile /job/hostfile --tag-output -x NCCL_NET_SHARED_BUFFERS=0 -x NCCL_MIN_NCHANNELS=8 -x NCCL_MAX_NCHANNELS=8 -x NCCL_DEBUG=INFO -x SCKL_XML_FILE=/msrhyper-weka/saemal/test2.xml -x NCCL_ALGO=SCKL -x NCCL_PROTO=Simple -x LD_LIBRARY_PATH=/msrhyper-weka/saemal/nccl-master/build/lib/ /msrhyper-weka/saemal/nccl-tests/build/alltoall_perf -w 10 -n 10 -b 1024 -e 1048576000 -f 2 -c 1 -g 1 -z 0 2021-03-30 23:19:51 +00:00
Saeed Maleki 944b1f2880 current state 2021-03-30 02:01:48 +00:00
Saeed Maleki 21d2e89564 fixed a bug regarding across node connection setup 2021-03-29 21:11:47 +00:00
Saeed Maleki 9a7af96ee3 a printf for NCCLCHECK 2021-03-27 00:59:54 +00:00
Saeed Maleki b3dc643313 fixed a bug 2021-03-26 05:27:35 +00:00
Saeed Maleki dabb6570e8 testing 2021-03-26 03:42:19 +00:00
Saeed Maleki 3694d84747 for all to all we need to directly copy from sendbuffer to recvbuffer for my rank portion 2021-03-25 20:53:54 +00:00
Saeed Maleki 7f55e284f3 done! it works 2021-03-25 08:01:51 +00:00
Saeed Maleki 36194ff2e4 working version within a node 2021-03-25 06:55:30 +00:00
Saeed Maleki b3dcc0d081 works within a node 2021-03-25 06:13:14 +00:00
Saeed Maleki b63029c21b deadlocks 2021-03-24 23:12:32 +00:00
Saeed Maleki 0d51c43749 more progress 2021-03-24 18:20:30 +00:00
Saeed Maleki 781bb7b302 ok version 2021-03-24 00:34:32 +00:00
Saeed Maleki 0959ebafc1 ok version 2021-03-24 00:33:15 +00:00
Saeed Maleki 99ad39b01b more progress 2021-03-23 23:23:36 +00:00
Saeed Maleki 08d3ea9909 not compilable 2021-03-23 06:48:26 +00:00
Saeed Maleki 62169be637 done with intra communication 2021-03-19 00:05:49 +00:00
Saeed Maleki 9581e3d1c3 finished developing 2021-03-18 22:50:56 +00:00
Saeed Maleki 102a204327
Merge pull request #2 from parasailteam/connection_setup 
Connection setup
 2021-03-17 15:31:21 -07:00
Saeed Maleki d12fca78a9 added one more comment 2021-03-17 22:30:54 +00:00
Saeed Maleki e276503b26 adding a new line 2021-03-17 22:24:06 +00:00
Saeed Maleki b69b341fca one more naming change 2021-03-17 22:21:34 +00:00
Saeed Maleki ee8e4c9d12 added some guards for corner cases 2021-03-17 19:37:57 +00:00