Superbuild with ITK ZLIB module?

[cookpa]

Hi @stnava @ntustison @gdevenyi and other interested developers,

I am trying to build native Windows binaries, and had to find a solution for zlib. We've required system zlib for a long time, but ITK now has an updated zlib module based on zlib-ng.

Is there any reason why we would want to continue using system zlib?

If not, I'll try to use ITK's version consistently, and will test compression performance on image I/O.

[gdevenyi]

zlib-ng is superior in every way I know of

• hardware-accel
• improved algorithms
• code testing
• modern build support

I see no reason not to switch to it exclusively.

[dorianps]

I had asked about this topic some years ago, happy to see it implemented. #702

…

On Tue, Jan 10, 2023, 2:34 PM Gabriel A. Devenyi ***@***.***> wrote: zlib-ng is superior in every way I know of - hardware-accel - improved algorithms - code testing - modern build support I see no reason not to switch to it exclusively. — Reply to this email directly, view it on GitHub <#1481 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACFJU7MKUA34SCXF4VLZE23WRW2TLANCNFSM6AAAAAATXEVMLU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[cookpa]

Hi @dorianps , might you be able to try out the Windows binaries here?

https://github.com/cookpa/ANTs/releases/tag/testActions-0.1

I built the same source on Mac and it compiles OK, but I don't see a performance gain in I/O speed vs my system (zlib 1.2.11). Are there any use cases you or @gdevenyi know of where zlib-ng might perform better? Just curious, either way I still think it's a good thing to use ITK's version for multiple reasons outlined above.

[cookpa]

BTW the test I did was simply reading a largeish (104x104x72x478) image (.nii.gz) with ImageMath and then writing it back out in either .nii.gz or .mha format.

[gdevenyi]

I think the zlib-ng performance is "statistical", in the sense its in the low-percentage range right now. ZLIB has already had many optimizations done over the years. I also know that not all the hardware-acceleration features have been merged yet. We should also check the current state of the build config for ITK to ensure its both optimized and generic for distribution purposes.

[dorianps]

I never really used ANTs on Windows but can try if it helps. What exactly should I try? I can do I/o commands like imageRead/imageWrite but I would do that on a laptop compared to a Linux workstation, and there is a lot more than Ants changing between the two scenarios. Dorian P.s. I have not puty hands on Ants commands for a while now.

…

On Wed, Jan 11, 2023, 4:08 PM Gabriel A. Devenyi ***@***.***> wrote: I think the zlib-ng performance is "statistical", in the sense its in the low-percentage range right now. ZLIB has already had many optimizations done over the years. I also know that not all the hardware-acceleration features have been merged yet. We should also check the current state of the build config for ITK to ensure its both optimized and generic for distribution purposes. — Reply to this email directly, view it on GitHub <#1481 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACFJU7JRDMMKV2SBNRUUAVLWR4OLZANCNFSM6AAAAAATXEVMLU> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[cookpa]

Thanks, sorry for mixing streams @dorianps - just I was more wanting to test basic functionality of the Windows binaries, something like does antsRegistration.exe work for a somple test case.

ZLIB was kind of a means to an end, I'm not looking to do extensive benchmarking.

[dorianps]

Got it. The scripts will not work in windows but I can try few binaries. Would you mind pasting a couple of registration/bias correction commands? Setting up all iterations without the sh scripts is not straightforward for me.

…

On Wed, Jan 11, 2023, 6:09 PM Philip Cook ***@***.***> wrote: Thanks, sorry for mixing streams @dorianps <https://github.com/dorianps> - just I was more wanting to test basic functionality of the Windows binaries, something like does antsRegistration.exe work for a somple test case. ZLIB was kind of a means to an end, I'm not looking to do extensive benchmarking. I don't have a Windows VM handy — Reply to this email directly, view it on GitHub <#1481 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACFJU7M7IFBLZDHWUC5EWBLWR44SLANCNFSM6AAAAAATXEVMLU> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[cookpa]

I'm not sure I can even write a command in powershell (recent experience suggests not), but it occurs to me that I could try to run the built-in short tests - I'll try that first. Thanks!

[dorianps]

Ok. Fyi, the command in Windows will probably be identical to Linux. No need to even add .exe when calling the executable.

…

On Wed, Jan 11, 2023, 8:17 PM Philip Cook ***@***.***> wrote: I'm not sure I can even write a command in powershell (recent experience suggests not), but it occurs to me that I could try to run the built-in short tests - I'll try that first. Thanks! — Reply to this email directly, view it on GitHub <#1481 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACFJU7NCXM2EZJUR26FBCWTWR5LSJANCNFSM6AAAAAATXEVMLU> . You are receiving this because you were mentioned.Message ID: ***@***.***>