Minor issue: daal produces lots of warnings in pedantic modes

[emmenlau]

Describe the bug
This is not really a bug, rather a suggestion for an improvement. The issue is that when we include DAAL 2019.5 headers, we get a huge number of compiler warnings on all platforms.

This is due to our strict compiler settings, like -Wall -Wextra -pedantic on Linux and /W4 on Windows. It seems DAAL can currently not be used nicely with such strict compiler settings, because we get thousands of lines of warnings. This makes our work a bit more complicated because our internal code review requests warning-free code.

Most warnings seem quite simple to fix. For example there are many cases of extra ; that could be removed.

To Reproduce
Steps to reproduce the behavior:

1. Set Linux compiler warnings to -pedantic -Wall -Wextra -Werror=return-type -Werror=uninitialized -Werror=unknown-pragmas or Windows compiler warnings to /W4
2. Include standard daal headers
3. Build the code
4. There are typically 100-10000 lines of warnings per source file

Expected behavior
Ideally, DAAL would only show a few or now warnings at all.

Output/Screenshots
Here an example from Ubuntu 18.04 x86_64 with gcc-7.4.0:

 In file included from /data/Tools/include/algorithms/neural_networks/layers/dropout/dropout_layer_backward.h:30:0,
                  from /data/Tools/include/algorithms/neural_networks/layers/dropout/dropout_layer.h:31,
                  from /data/Tools/include/daal.h:256,
                  from /home/builds/include/Regression.hh:23,
                  from /home/builds/src/Regression.cc:20:
/data/Tools/include/algorithms/neural_networks/layers/dropout/dropout_layer_backward_types.h:116:38: warning: extra ';' [-Wpedantic]
      DECLARE_SERIALIZABLE_CAST(Result);

Environment:

• OS: Ubuntu 18.04 x86_64, macOS 10.13 XCode 9.4, macOS 10.14 XCode 11.3, MS Visual Studio 2017, MS Visual Studio 2019
• Version: 2019 Update 5

[SmirnovEgorRu]

@emmenlau, thank you for reporting the issue.
Usually we build our examples without "-Wpedantic" flag and don't observe what you reported. We are looking at the the problem and how it can be resolved.

@dr-pain is assigned for this work.

[emmenlau]

Thanks a lot @SmirnovEgorRu for the positive feedback! I hope the issue is not causing hazzle, and its not very urgent. But we're looking forward if something can be improved!

[dr-pain]

@emmenlau
I have fixed all gcc warnings (or most of them). See #408.
Check it please.
Andrey

[emmenlau]

Awesome @dr-pain, thanks! I'm currently using the binaries from official Intel, so I'll have to wait a bit for the 2020 Update 1. But I will happily close the issue as fixed! Thanks!

[dr-pain]

Hello @emmenlau !
Sorry, but we don`t plan to commit fixes to 2020 product line - use 2021 (master, releases) instead.
Andrey

[emmenlau]

Ok I'll check them out! Will they be released as binary download for all platforms too (like the current Intel downloads from the official website), or are these source releases?

[dr-pain]

yes, we have binary releases for 2021.
You may download them from "releases" folder (but for linux and macos only).
2021 oneDAL is also available at official web site in oneAPI KIT package https://software.intel.com/en-us/oneapi .
Andrey

[emmenlau]

Thanks Andrey! I'm really sorry and I feel very stupid to ask this, but: could you point me to the binary download for 2021? I've considered

• The branch "releases" from https://github.com/intel/daal/tree/releases
• The "releases" page from https://github.com/intel/daal/releases
• The Installation section from https://github.com/intel/daal#installation

But in all cases, I do not see anything to download for 2021, neither classic DAAL nor oneDAL. I did not consider oneAPI KIT package because we're not ready for this yet, but anything else. Am I doing something wrong?

[dr-pain]

Im really sorry, but I missed fact that we are not publishing oneDAL binary releases at github. The only way to get changes in binaries is - download oneAPI KIT and install oneDAL. It should be possible to install oneDAL only and skip all other tools. Dont afraid to use oneDAL instead of classic DAAL. Now they are similar from several points of view. The only one big change happened - neural networks were removed from oneDAL.
Changes should be available in beta06 release. it is next release after upcoming beta05 release. current available release is beta04.
Andrey