Code

Test bench#127

Code

Open

novaCoder-zrk wants to merge 116 commits intoNVIDIA:masterNVIDIA/nccl-tests:masterfrom Panlichen:test_benchPanlichen/nccl-tests:test_benchCopy head branch name to clipboard

+10,430-49

Commits

Commits on Jul 8, 2022

• add all_reduce_group test

  Panlichen

  committed
  93a5cda

  Copy full SHA for 93a5cda

Commits on Jul 13, 2022

• simple group allreduce

  Panlichen

  committed
  725b0d7

  Copy full SHA for 725b0d7

Commits on Jul 14, 2022

• nccl group bigger

  Panlichen

  committed
  06c97da

  Copy full SHA for 06c97da

Commits on Jul 16, 2022

• log for ncclGroupStart/End

  Panlichen

  committed
  560f9eb

  Copy full SHA for 560f9eb

• add non group all_reduce in simple

  Panlichen

  committed
  abe8f5c

  Copy full SHA for abe8f5c

• half ofccl_all_reduce

  Panlichen

  committed
  6990dc3

  Copy full SHA for 6990dc3

• Merge branch 'master' of github.com:Panlichen/nccl-tests

  Panlichen

  committed
  44c40d8

  Copy full SHA for 44c40d8

• Simple not really necessary, yet no harm to keep it

  Panlichen

  committed
  b8a749a

  Copy full SHA for b8a749a

Commits on Jul 18, 2022

• ofccl test file

  Panlichen

  committed
  02a914e

  Copy full SHA for 02a914e

• run startColl exactly as we want

  Panlichen

  committed
  8eba16f

  Copy full SHA for 8eba16f

• ofccl_all_reduce.cu

  Panlichen

  committed
  d6a4d47

  Copy full SHA for d6a4d47

Commits on Jul 19, 2022

• add log

  Panlichen

  committed
  5e3c98d

  Copy full SHA for 5e3c98d

• add -M option: use seprate ncclComm for different coll op, even with the same dev set

  Panlichen

  committed
  d3de021

  Copy full SHA for d3de021

• remove log

  Panlichen

  committed
  76b1cd7

  Copy full SHA for 76b1cd7

Commits on Jul 28, 2022

• use prepare and done in nccl-tests

  Panlichen

  committed
  a466914

  Copy full SHA for a466914

Commits on Aug 9, 2022

• check no reused ncclComm in ofcclCommList

  Panlichen

  committed
  9b362b5

  Copy full SHA for 9b362b5

• invoke ofcclDestroy

  Panlichen

  committed
  818c8e3

  Copy full SHA for 818c8e3

Commits on Aug 23, 2022

• use ofcclRunAllReduce

  Panlichen

  committed
  67e70b9

  Copy full SHA for 67e70b9

Commits on Aug 26, 2022

• use callback

  Panlichen

  committed
  97f58bc

  Copy full SHA for 97f58bc

• use func-ptr for callback, instead of std::function and lambda

  Panlichen

  committed
  8a3d5f8

  Copy full SHA for 8a3d5f8

Commits on Aug 28, 2022

• stuck

  Panlichen

  committed
  a3a1aea

  Copy full SHA for a3a1aea

• completeColl in warmup result in stuck

  Panlichen

  committed
  7ff3ea5

  Copy full SHA for 7ff3ea5

• Merge pull request #1 from Panlichen/aggregate_waiting_cqes

  Show description for 8c5df9e
  Panlichen

  authored
  8c5df9e

  Copy full SHA for 8c5df9e

• +lock

  Panlichen

  committed
  ee76beb

  Copy full SHA for ee76beb

Commits on Aug 29, 2022

• tidy log

  Panlichen

  committed
  3f0a8fe

  Copy full SHA for 3f0a8fe

Commits on Sep 5, 2022

• nccl-tests run exactly once

  Panlichen

  committed
  0bd6d6a

  Copy full SHA for 0bd6d6a

Commits on Sep 8, 2022

• ad-hoc check

  Panlichen

  committed
  9a35e7f

  Copy full SHA for 9a35e7f

Commits on Sep 9, 2022

• wierd check

  Panlichen

  committed
  22b869f

  Copy full SHA for 22b869f

• activate -n, can run multi-iters

  Panlichen

  committed
  732f8bd

  Copy full SHA for 732f8bd

Commits on Sep 26, 2022

• + warmup

  Panlichen

  committed
  85d5cbd

  Copy full SHA for 85d5cbd

• fix completeColl in warmup

  Panlichen

  committed
  090185c

  Copy full SHA for 090185c

Commits on Sep 30, 2022

• try context

  Panlichen

  committed
  17197fa

  Copy full SHA for 17197fa

Commits on Oct 2, 2022

• bugfix: seenCqe[miter] = 0; in warmup

  Panlichen

  committed
  5f399fd

  Copy full SHA for 5f399fd

Commits on Oct 6, 2022

• polish callback

  Panlichen

  committed
  5cd2cb8

  Copy full SHA for 5cd2cb8

Commits on Oct 9, 2022

• check OK

  Panlichen

  committed
  0148628

  Copy full SHA for 0148628

• check ok

  Panlichen

  committed
  b6027be

  Copy full SHA for b6027be

• restore semi-original NCCL's BenchTime

  Panlichen

  committed
  24290c6

  Copy full SHA for 24290c6

Commits on Oct 12, 2022

• run check smoothly

  Panlichen

  committed
  d609039

  Copy full SHA for d609039

Commits on Oct 13, 2022

• finalize check

  Panlichen

  committed
  4cd2091

  Copy full SHA for 4cd2091

• adapt to volunteer quit

  Panlichen

  committed
  b8dc018

  Copy full SHA for b8dc018

• Merge branch 'master' of github.com:Panlichen/nccl-tests

  Panlichen

  committed
  4ebe9e5

  Copy full SHA for 4ebe9e5

• adapt to volunteer quit

  Panlichen

  committed
  bd10523

  Copy full SHA for bd10523

Commits on Oct 14, 2022

• keep the report log

  Panlichen

  committed
  74d4f0d

  Copy full SHA for 74d4f0d

Commits on Oct 17, 2022

• try pure inplace

  Panlichen

  committed
  ed7f645

  Copy full SHA for ed7f645

• log format

  Panlichen

  committed
  eed57ca

  Copy full SHA for eed57ca

Commits on Oct 18, 2022

• manual buffer size done

  Panlichen

  committed
  0ef76cc

  Copy full SHA for 0ef76cc

Commits on Oct 19, 2022

• adjust log

  Panlichen

  committed
  d6cad8e

  Copy full SHA for d6cad8e

Commits on Oct 20, 2022

• + nccl_manual_size

  Panlichen

  committed
  cec88ef

  Copy full SHA for cec88ef

• nccl manual size seems ok

  Panlichen

  committed
  34f1b12

  Copy full SHA for 34f1b12

• fix manual size bug

  Panlichen

  committed
  c84dd89

  Copy full SHA for c84dd89

• non-homogeneous nccl manual size

  Panlichen

  committed
  93668dc

  Copy full SHA for 93668dc

Commits on Oct 22, 2022

• + cudadev in cbArgs for ofccl manual size

  Panlichen

  committed
  71b40c7

  Copy full SHA for 71b40c7

Commits on Oct 24, 2022

• 161 maunal size from resnet

  Panlichen

  committed
  f2b285d

  Copy full SHA for f2b285d

• accurate damie

  Panlichen

  committed
  a32587b

  Copy full SHA for a32587b

• .

  Panlichen

  committed
  5d07bca

  Copy full SHA for 5d07bca

Commits on Oct 31, 2022

• aggressive no sync

  Panlichen

  committed
  4f06775

  Copy full SHA for 4f06775

Commits on Nov 7, 2022

• a new permutation from oneflow

  Panlichen

  committed
  40fbb70

  Copy full SHA for 40fbb70

Commits on Nov 12, 2022

• log

  Panlichen

  committed
  e98b271

  Copy full SHA for e98b271

Commits on Nov 14, 2022

• suit 8 cards

  Panlichen

  committed
  3598de4

  Copy full SHA for 3598de4

Commits on Nov 18, 2022

• use prepareDone

  Panlichen

  committed
  bcf3b87

  Copy full SHA for bcf3b87

Commits on Nov 20, 2022

• nccl ms different order

  Panlichen

  committed
  0ccfcc9

  Copy full SHA for 0ccfcc9

Commits on Nov 26, 2022

• usleep

  Panlichen

  committed
  2b19a59

  Copy full SHA for 2b19a59

• + ofccl_test.sh

  Panlichen

  committed
  b3b6323

  Copy full SHA for b3b6323

Commits on Nov 29, 2022

• scripts

  Panlichen

  committed
  e2bfe2e

  Copy full SHA for e2bfe2e

Commits on Nov 30, 2022

• script

  Panlichen

  committed
  34cd275

  Copy full SHA for 34cd275

Commits on Dec 1, 2022

• scripts

  Panlichen

  committed
  4e81620

  Copy full SHA for 4e81620

Commits on Dec 5, 2022

• scripts

  Panlichen

  committed
  0c3718e

  Copy full SHA for 0c3718e

Commits on Dec 7, 2022

• scripts

  Panlichen

  committed
  dba5947

  Copy full SHA for dba5947

Commits on Dec 8, 2022

• scripts

  Panlichen

  committed
  10fefc6

  Copy full SHA for 10fefc6

Commits on Dec 9, 2022

• little ms

  Panlichen

  committed
  7a79d98

  Copy full SHA for 7a79d98

Commits on Dec 19, 2022

• + nccl_test.sh

  Panlichen

  committed
  7b37cea

  Copy full SHA for 7b37cea

• fix bug in nccl_tests.sh

  Panlichen

  committed
  5bb88a1

  Copy full SHA for 5bb88a1

Commits on Dec 21, 2022

• + run multi test scripts

  Panlichen

  committed
  d9f1a55

  Copy full SHA for d9f1a55

• +order

  Panlichen

  committed
  f505977

  Copy full SHA for f505977

Commits on Dec 22, 2022

• 28 is occupied

  Panlichen

  committed
  1c9a007

  Copy full SHA for 1c9a007

• fix bug in nccl-tests/src_manual_size/ofccl_all_reduce_ms.cu

  Panlichen

  committed
  54ff526

  Copy full SHA for 54ff526

Commits on Dec 23, 2022

• 第一次完成 auto_test 开发

  Panlichen

  committed
  57875d6

  Copy full SHA for 57875d6

• 文件命名bug修复

  Panlichen

  committed
  23a25ea

  Copy full SHA for 23a25ea

• deltaSec看起来统计得偏大了

  Panlichen

  committed
  9daa76c

  Copy full SHA for 9daa76c

• meaningless NEW_TIMER

  Panlichen

  committed
  1f58d48

  Copy full SHA for 1f58d48

• check frequency

  Panlichen

  committed
  fc09438

  Copy full SHA for fc09438

• update xls name and ndev

  Panlichen

  committed
  266d3c8

  Copy full SHA for 266d3c8

• Merge branch 'master' into auto_test

  Panlichen

  committed
  b034d10

  Copy full SHA for b034d10

• Merge pull request #2 from Panlichen/auto_test

  Show description for f0cf272
  Panlichen

  authored
  f0cf272

  Copy full SHA for f0cf272

• add log

  Panlichen

  committed
  4702178

  Copy full SHA for 4702178

• update env

  Panlichen

  committed
  5f7b4bf

  Copy full SHA for 5f7b4bf

Commits on Dec 24, 2022

• +nccl ofccl run.sh

  Panlichen

  committed
  5ced1a0

  Copy full SHA for 5ced1a0

Commits on Dec 25, 2022

• report rank 0 avg time

  Panlichen

  committed
  875d1d5

  Copy full SHA for 875d1d5

• scripts

  Panlichen

  committed
  386ee92

  Copy full SHA for 386ee92

• nccl show each kernel time

  Panlichen

  committed
  13567f2

  Copy full SHA for 13567f2

Commits on Dec 27, 2022

• 能处理均值

  

  novaCoder-zrk

  committed
  12969b5

  Copy full SHA for 12969b5

• time页表 R列-O列

  

  novaCoder-zrk

  committed
  29d300b

  Copy full SHA for 29d300b

• nccl kern 求平均

  

  novaCoder-zrk

  committed
  2b4b937

  Copy full SHA for 2b4b937

Commits on Dec 28, 2022

• 输出 ori

  

  novaCoder-zrk

  committed
  93b9ddc

  Copy full SHA for 93b9ddc

• 编译 QE_oricpp 增加一列实际的byte数，增加 Ex-Ox，修改average

  

  novaCoder-zrk

  committed
  4321b54

  Copy full SHA for 4321b54

Commits on Dec 29, 2022

• + in order ms

  Panlichen

  committed
  d3f652d

  Copy full SHA for d3f652d

• 输出 totalCnt

  

  novaCoder-zrk

  committed
  5135aa3

  Copy full SHA for 5135aa3

• script

  Panlichen

  committed
  ffcbfba

  Copy full SHA for ffcbfba

• Merge branch 'master' into auto_test2

  Panlichen

  authored
  8dab62a

  Copy full SHA for 8dab62a

• Merge pull request #3 from Panlichen/auto_test2

  Show description for c0454ff
  Panlichen

  authored
  c0454ff

  Copy full SHA for c0454ff

Commits on Dec 30, 2022

• scripts

  Panlichen

  committed
  57ee21f

  Copy full SHA for 57ee21f

Commits on Jan 1, 2023

• scripts

  Panlichen

  committed
  9ee7202

  Copy full SHA for 9ee7202

Commits on Jan 6, 2023

• scripts

  Panlichen

  committed
  665de43

  Copy full SHA for 665de43

• scripts

  Panlichen

  committed
  b5a42cc

  Copy full SHA for b5a42cc

Commits on Jan 8, 2023

• scripts

  Panlichen

  committed
  1970187

  Copy full SHA for 1970187

Commits on Jan 10, 2023

• scripts

  Panlichen

  committed
  ceb3a5a

  Copy full SHA for ceb3a5a

Commits on Jan 11, 2023

• datatype in cmd; MY_NUM_DEV as cmd line param

  Panlichen

  committed
  ac30fd4

  Copy full SHA for ac30fd4

Commits on Jan 13, 2023

• scripts

  Panlichen

  committed
  82c2a8e

  Copy full SHA for 82c2a8e

Commits on Jan 14, 2023

• + occl AllGather

  Panlichen

  committed
  c760c82

  Copy full SHA for c760c82

• 5555 remove NCCL_MIN_NCHANNELS limit T^T TAT T_T T-T

  Panlichen

  committed
  abd0c14

  Copy full SHA for abd0c14

Commits on Jan 15, 2023

• + ofccl ReduceScatter

  Panlichen

  committed
  bbb04c6

  Copy full SHA for bbb04c6

• + occl reduce

  Panlichen

  committed
  1f099a1

  Copy full SHA for 1f099a1

• +ofccl_broadcast; fix DEBUG_NT

  Panlichen

  committed
  6c075fd

  Copy full SHA for 6c075fd

Commits on Jan 19, 2023

• 精简前

  

  novaCoder-zrk

  committed
  37b2463

  Copy full SHA for 37b2463

Commits on Jan 20, 2023

• 测试 五种操作

  

  novaCoder-zrk

  committed
  d81c257

  Copy full SHA for d81c257

Commits on Jan 27, 2023

• 去除 xlrd

  

  zrk

  committed
  3bfb750

  Copy full SHA for 3bfb750