next IFX release

[David_DiLaura1]

When is the next release of IFX scheduled to be available?

The latest version of IFX still throws internal compiler errors. The code compiles without incident using Ifort or the old Parallel Studio XE release 2019.5 

[JohnNichols]

The hints given by Intel are April, usually it shows up a week earlier than the announcement in the updates if you watch control panel. 

[TobiasK]

@David_DiLaura1 

can you please point to some error message and ideally a reproducer?

[Andrew_Smith]

I have found the IFX compiler is free of internal compiler errors once the code does not contain illegal instructions. One I found was where I declared a module procedure interface and then declared different arguments in the submodules implementation of the procedure. But I too have not been able to switch over to IFX entirely and eagerly await the next update.

[Ron_Green]

We stopped posting release dates some years ago.  The situation is that the Intel Fortran compiler is part of the larger Intel oneAPI product releases.  As an example, we had our compiler code freeze in mid-December.  That code has passed all compiler QA tests, regression tests and so on. That is when we delivered our compilers to the larger org. 

From that point, the "downstream" oneAPI products use this drop to build.  MKL, IPP, and all the other libraries, Intel MPI, etc etc etc.  So they have time to validate their software with our candidate compilers.  Mostly the other teams use the ICX compiler.  But MKL  and IPP teams, and our SciPy in our distro of Python, use the Fortran compiler as part of their builds.  Sometimes they find something that needs a fix.  Finally the entire Toolkits are bundled, tested as a group, QA'ed several times.  Then uploaded to repositories, IRC, etc.  In flight all the Documentation gets updated in drafts and quality checked.  A few days before the official release, we post all the docs online, repos go live and another full test of these public packages are tested to make sure the checksums are correct, links work, documents are accessible, etc.  THEN, if everything checsk out, the announcement goes out.  BUT occasionally these early public but unannounced packages will have an issue.  Rare, but it has happened.  In that case we'd pull everything offline, do a fix, repackage and QA, and try again.  We've not had this happen because of anything the Fortran team delivered (just to be clear). This is why I'd hold off downloading anything until the public announcement, even if you happen to see the docs and packages 'go live' before the official announcement.

Hopefully you can see that there are a lot of dependences and things that can need addressing that will delay a release.  

We have about a 3 month or more gap from when we drop OUR Fortran code and the time it gets to you.  Intel is working to improve the turnaround.  It's been a problem and we know it.   For example, 2024.0.2 IFX compiler:  if you print the -what -V logo information you will see:

ifx -what -V -c hello.f90

Intel(R) Fortran Compiler for applications running on Intel(R) 64, Version 2024.0.2 Build 20231213

Copyright (C) 1985-2023 Intel Corporation. All rights reserved.

Intel(R) Fortran 24.0-1238.2

The build date can be deceiving - it is when the compilers were bundled together and run through QA for the patch release 0.2.  It is not when we delivered our IFX and IFORT code for that update.

The -what information is more telling: 24.0-1238.2
The "24.0" is the major number for this compiler.  not so interesting.

The "1238.2" is our "Edit Number".  Each fix ( pull request) bumps that number.  It's a rough indication of the actual source version we used for the front-end of ifx.  1238 was from around August/early September 2023.  We put in 2 special fixes past that for a couple of issues found in testing, hence the ".2" on top of 1238. 

The next update, 2024.1.0 package, aka 2024 Update 1, will have edit number 1472.x.   So that is 224 edits in 2024.1 over the initial code base used for 2024.0.   Some are new features, some are esoteric fixes to warning and error messages, some are bugs reported here, some are bugs from key customers, some are bugs we find while fixing other things, some are bugs found in testing.   

[David_DiLaura1]

Thanks for the explanation of the the compiler release process. There's no denying it's complicated!

FYI: After more than a day of searching, we have found that in some of our routines, the size parameter used in the align compiler directives were 128.  Changing this to 32 resulted in a clean compile. IFX throws an internal compiler error only when the code is compiled for release (full optimization, threading, vectorization, etc). In debug mode (no optimization or threading) no internal compiler error was generated when the align size was set to 128.

[Barbara_P_Intel]

There will be an announcement on this Forum when the new compiler versions are released.