Fix the incorrect MKLDNN/MKL logic in cmake #14877

yinghu5 · 2019-05-05T07:02:55Z

This is a re-creation of #14829 . I've accidentally deleted my fork of incubator-mxnet.

Description

We actually want to unlock the MKLDNN (MKLML) and MKL dependency, which means use MKLDNN and USE_BLAS=mkl completely independent.

when MKLDNN is on, it will download MKLML automatically, don't use full MKL package.
when USE_BLAS=mkl, only use the full MKL (MKL_ROOT), don't detect MKLML (from MKLDNN)

update MKL_ROOT

anirudhacharya · 2019-05-06T17:18:06Z

@mxnet-label-bot add [pr-awaiting-review]

zixuanweeei · 2019-05-07T05:11:48Z

ci/jenkins/Jenkins_steps.groovy

+        ws('workspace/build-cpu-nomkldnn-mkl') {
+          timeout(time: max_time, unit: 'MINUTES') {
+            utils.init_git_win()
+            powershell 'py -3 ci/build_windows.py -f WIN_CPU_NOMKLDNN_MKL'


The build flavor(-f) should be WIN_CPU_MKL instead of WIN_CPU_NOMKLDNN_MKL since it has been modified from the latter in ci/build_windows.py (line 93).

thank you for fix. change it now.

zixuanweeei · 2019-05-07T05:42:24Z

ci/build_windows.py

@@ -218,6 +246,8 @@ def main():
            os.environ["OpenCV_DIR"] = "C:\\Program Files\\OpenCV-v3.4.1\\build"
        if 'CUDA_PATH' not in os.environ:
            os.environ["CUDA_PATH"] = "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v9.2"
+        if 'MKL_ROOT' not in os.environ:
+            os.environ["MKL_ROOT"] = "C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\mkl"


Not sure about the effects of the escape characters here. But for the path string, I think using a raw string prefix flag (r'the/path/string'), in this case, is better.

we test both format should be ok. anyway change it now. thanks

fix nomkldnn issue.

change MKL_ROOT

yinghu5 · 2019-05-08T13:32:10Z

@Chancebair @marcoabreu Could you help to install a full MKL library for window CI? we supposed the full MKL will be installed to C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\mkl on windows. You can get the executable here: http://registrationcenter-download.intel.com/akdlm/irc_nas/tec/15247/w_mkl_2019.3.203.exe thanks

Chancebair · 2019-05-08T13:35:51Z

@yinghu5 I created a ticket with my teams oncall to look into this

TaoLv · 2019-05-09T05:32:31Z

Hi @Chancebair, do you have any update for the ticket? I don't want to rush you but I hope this PR can catch the code freeze of 1.5.0 release as I see there are quite a few issues about Windows build. It will also help improving the coverage of existing CI system. Thank you.

Chancebair · 2019-05-09T09:57:43Z

MKL will be installed on windows sometime today thanks to @lebeg

TaoLv · 2019-05-10T00:00:43Z

Great! Thank you @Chancebair and @lebeg . @yinghu5 please try if the windows CI works now~

yinghu5 · 2019-05-10T08:37:37Z

@Chancebair and @lebeg , thank you a lot! is the MKL installed to C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\mkl ?
In CI we define the environment MKL_ROOT
=C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\mkl

the current CI test shows,
Could NOT find MKL (missing: MKL_INCLUDE_DIR MKL_RT_LIBRARY). seems MKL still not finded.
Please help check, thanks.

windows md5

yinghu5 · 2019-05-13T00:51:57Z

I test again in local machine today. it works when >set MKL_ROOT=C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\mkl

thanks

export MKLROOT

powershell 'cd $Env:MKL_ROOT'

lebeg · 2019-05-13T12:07:55Z

I've updated the scripts for AMI creation, but currently we have a blocker on @Chancebair side. Once he deploys the new AMI's MKL should be available at the exact path you've mentioned.

perdasilva · 2019-05-14T20:00:22Z

I've updated the windows instances to the new AMI. It may take some time to cycle them with the autoscaler, but please give it a go.

trigger MKL_ROOT test

test MKL_ROOT line 552

update 538 $Env:MKL_ROOT

…to yinghu5-cmake-1

yinghu5 · 2019-05-15T08:26:04Z

@perdasilva @marcoabreu May I have your comments on the CI status? Seems it doesn't work now.

perdasilva · 2019-05-15T10:17:18Z

@yinghu5 I've checked the windows instances running on prod, and they are running with the new AMIs.
There's an unrelated issue with TensorRT, however. I've taken a look at your latest run, and it seems there is a problem in the PowerShell. Could you please address that and try again?

perdasilva · 2019-05-15T10:20:42Z

I also see that there's been a successful run since the update. Could it be that the TensorRT issue was throwing you off? The current error is on the powershell side.

lebeg · 2019-05-15T10:22:07Z

This line probably doesn't like backslashes.

marcoabreu · 2019-05-15T11:51:39Z

ci/jenkins/Jenkins_steps.groovy

+        ws('workspace/build-cpu-mkl') {
+          timeout(time: max_time, unit: 'MINUTES') {
+            utils.init_git_win()
+            powershell '$Env:MKL_ROOT="C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\mkl"'


Could you move this code into build_windows.py? We prefer to have these scripts self-contained and the Jenkinsfile just being an orchestration tool.

yinghu5 · 2019-05-16T02:43:55Z

@lebeg @marcoabreu @TaoLv @pengzhao-intel thank you very much for the support! after remove the redundancy test MKL_ROOT code as you suggested, the CI was triggered today. and all windows CPU parts, windows-cpu, which the PR want to fix passed successfully. Thank you!

pengzhao-intel · 2019-05-16T02:53:45Z

It's great @yinghu5, please rebase the code again. The GPU failure is fixed now.

yinghu5 · 2019-05-16T03:25:03Z

All CI passed, except the unix-gpu tensorRT test error. thanks

…to yinghu5-cmake-1

pengzhao-intel · 2019-05-16T06:12:09Z

@lebeg @marcoabreu @perdasilva the Windows CI is built, thanks all of your help.
Please help to take a final review :) I will merge the PR if no further comment in order to keep the timeline of r1.5.

perdasilva

LGTM - maybe @lebeg could you have a look at the cmake changes?

pengzhao-intel · 2019-05-16T13:26:51Z

It's a nice first PR for MXNet, thanks @yinghu5.
Merging now.

* fix cmake for win and mkl * Update build_windows.py update MKL_ROOT * Update Jenkins_steps.groovy fix nomkldnn issue. * Update build_windows.py change MKL_ROOT * msshadow_blas * update mshasow * Update DownloadMKLML.cmake windows md5 * Update DownloadMKLML.cmake windows md5 * Update Jenkins_steps.groovy export MKLROOT * Update Jenkins_steps.groovy powershell 'cd $Env:MKL_ROOT' * Update Jenkins_steps.groovy trigger MKL_ROOT test * Update Jenkins_steps.groovy test MKL_ROOT line 552 * Update Jenkins_steps.groovy update 538 $Env:MKL_ROOT * update mshadow * revert gitmodules * remove comments * remove test code * remove test powershell

fix cmake for win and mkl

634bfea

yinghu5 requested a review from szha as a code owner May 5, 2019 07:02

Update build_windows.py

2f8adc0

update MKL_ROOT

This was referenced May 6, 2019

Fix the incorrect MKLDNN/MKL logic in cmake #14829

Closed

Cmake blas #14871

Closed

marcoabreu added the pr-awaiting-review PR is waiting for code review label May 6, 2019

zixuanweeei reviewed May 7, 2019

View reviewed changes

yinghu5 added 2 commits May 8, 2019 16:12

Update Jenkins_steps.groovy

90eb723

fix nomkldnn issue.

Update build_windows.py

44ef104

change MKL_ROOT

This was referenced May 9, 2019

Cmake : could not find mkl (missing MKLDNN_LIBRARY) #13881

Closed

[Discussion] 1.5.0 Roadmap #14619

Closed

yinghu5 added 2 commits May 10, 2019 15:51

msshadow_blas

b4c8ddd

update mshasow

b22ce7d

yinghu5 added 2 commits May 12, 2019 21:48

Update DownloadMKLML.cmake

5b2a7cb

windows md5

Update DownloadMKLML.cmake

d9bd487

windows md5

yinghu5 added 2 commits May 13, 2019 11:09

Update Jenkins_steps.groovy

8e6b4ad

export MKLROOT

Update Jenkins_steps.groovy

4e54ca8

powershell 'cd $Env:MKL_ROOT'

yinghu5 mentioned this pull request May 14, 2019

Update mkldnn window build instructions in MKLDNN_README.md yinghu5/incubator-mxnet#2

Open

7 tasks

yinghu5 mentioned this pull request May 15, 2019

Update mkldnn window build instructions in MKLDNN_README.md #14952

Merged

7 tasks

yinghu5 added 8 commits May 15, 2019 10:32

Update Jenkins_steps.groovy

4247d58

trigger MKL_ROOT test

Update Jenkins_steps.groovy

c343e69

test MKL_ROOT line 552

Update Jenkins_steps.groovy

6cc35d6

update 538 $Env:MKL_ROOT

Merge branch 'master' of https://github.com/apache/incubator-mxnet in…

ab0cd04

…to yinghu5-cmake-1

update mshadow

af5edc3

revert gitmodules

dc414d2

remove comments

33561d9

remove test code

29d0bb9

marcoabreu reviewed May 15, 2019

View reviewed changes

remove test powershell

6bd886d

Merge branch 'master' of https://github.com/apache/incubator-mxnet in…

712a1f9

…to yinghu5-cmake-1

perdasilva approved these changes May 16, 2019

View reviewed changes

marcoabreu approved these changes May 16, 2019

View reviewed changes

TaoLv approved these changes May 16, 2019

View reviewed changes

lebeg approved these changes May 16, 2019

View reviewed changes

pengzhao-intel merged commit d87bd2a into apache:master May 16, 2019

This was referenced May 17, 2019

Default build with CMake not using MKL #14670

Closed

[MKLDNN] Question on installation and use of MKLDNN #14335

Closed

[Bug]Cannot compile mxnet on windows #14203

Open

Default cmake build uses openblas instead of MKL #14085

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix the incorrect MKLDNN/MKL logic in cmake #14877

Fix the incorrect MKLDNN/MKL logic in cmake #14877

yinghu5 commented May 5, 2019 •

edited

Loading

anirudhacharya commented May 6, 2019

zixuanweeei May 7, 2019

yinghu5 May 8, 2019

zixuanweeei May 7, 2019

yinghu5 May 8, 2019

yinghu5 commented May 8, 2019

Chancebair commented May 8, 2019

TaoLv commented May 9, 2019

Chancebair commented May 9, 2019

TaoLv commented May 10, 2019

yinghu5 commented May 10, 2019

yinghu5 commented May 13, 2019

lebeg commented May 13, 2019

perdasilva commented May 14, 2019

yinghu5 commented May 15, 2019

perdasilva commented May 15, 2019

perdasilva commented May 15, 2019

lebeg commented May 15, 2019

marcoabreu May 15, 2019

yinghu5 commented May 16, 2019

pengzhao-intel commented May 16, 2019

yinghu5 commented May 16, 2019

pengzhao-intel commented May 16, 2019

perdasilva left a comment •

edited

Loading

pengzhao-intel commented May 16, 2019

Fix the incorrect MKLDNN/MKL logic in cmake #14877

Fix the incorrect MKLDNN/MKL logic in cmake #14877

Conversation

yinghu5 commented May 5, 2019 • edited Loading

Description

anirudhacharya commented May 6, 2019

zixuanweeei May 7, 2019

Choose a reason for hiding this comment

yinghu5 May 8, 2019

Choose a reason for hiding this comment

zixuanweeei May 7, 2019

Choose a reason for hiding this comment

yinghu5 May 8, 2019

Choose a reason for hiding this comment

yinghu5 commented May 8, 2019

Chancebair commented May 8, 2019

TaoLv commented May 9, 2019

Chancebair commented May 9, 2019

TaoLv commented May 10, 2019

yinghu5 commented May 10, 2019

yinghu5 commented May 13, 2019

lebeg commented May 13, 2019

perdasilva commented May 14, 2019

yinghu5 commented May 15, 2019

perdasilva commented May 15, 2019

perdasilva commented May 15, 2019

lebeg commented May 15, 2019

marcoabreu May 15, 2019

Choose a reason for hiding this comment

yinghu5 commented May 16, 2019

pengzhao-intel commented May 16, 2019

yinghu5 commented May 16, 2019

pengzhao-intel commented May 16, 2019

perdasilva left a comment • edited Loading

Choose a reason for hiding this comment

pengzhao-intel commented May 16, 2019

yinghu5 commented May 5, 2019 •

edited

Loading

perdasilva left a comment •

edited

Loading