Old problem with compilation of FoX fortran library with Ifort

Old problem with compilation of FoX fortran library with Ifort

Hi all!
I would like to draw you attention to one issue, which is rather serious, I guess.
There exists old problem with compilation of FoX Fortral library with ifort. I saw some discussions on a forum, but no real (receipt like) conclusion was elaborated.
The most iteresting point is that things are getting worse with time. What do I mean
ifort 10.1.015 - can compile FoX lib, all tests passed
ifort 11.0.074 - can compile Fox lib, tests are PARTIALLY passed
ifort 11.1.059 - CAN NOT compile FoX lib.

If I will try to compile any FoX using application I will get
ifort 10.1.015 - operational program
ifort 11.0.074 - program crashes
ifort 11.1.059 - COMPILER crashes
Compiler error messages can be different, depending on the presence of PRIVATE statement in FoX modules AND the order of USE statements in main program (not the case for 10.1), i.e. internal error or stack segmentation.

Apparently there is a huge problem for ifort 11.xx with recursive declaration used in FoX. When compiled with 11.0 compiler the entry points for FoX procedures are visible in the exec program body, but they are referencing nothing (easy to see with, for example IDB. By the way IDB hangs).

So, my questions are. Is this problem a real concern for Intel software developers? For how long will it stay?

There is no problem at all with GNU.

Best wishes,
Sergei

P.S. All you need to investigate a case is http://www.uszla.me.uk/FoX/source/

14 帖子 / 0 全新
最新文章
如需更全面地了解编译器优化,请参阅优化注意事项

Hi Sergei - Thanks for providing this summary. I've handled at lest three unique internal compiler errors w/11.1 involving the FoX library of late. I do not recall any reports of run-time errors. Yes, these are of concern to us. Some of the reported errors will be addressed in upcoming 11.1 updates while others are still being investigated.

Thank you for the pointer to the source. I will investigate what run-time failures exist with any included self-check tests. There's a version 4.0.4 available, but it appears the current released version is 4.0.2 which is the version I intend to work with.

Just curious, what version are you using?

Quoting - Kevin Davis (Intel)

Hi Sergei - Thanks for providing this summary. I've handled at lest three unique internal compiler errors w/11.1 involving the FoX library of late. I do not recall any reports of run-time errors. Yes, these are of concern to us. Some of the reported errors will be addressed in upcoming 11.1 updates while others are still being investigated.

Thank you for the pointer to the source. I will investigate what run-time failures exist with any included self-check tests. There's a version 4.0.4 available, but it appears the current released version is 4.0.2 which is the version I intend to work with.

Just curious, what version are you using?

Hello Kevin!

You will see runtime errors during FoX tests.

I am using FoX as a part of fortran quantum mechanical calculations package Siesta. I can not say 100% for sure that there is no small bugs in 4.0.4 lib. From the first glance everything is OK. At least I have not seen anything alarming. So I am using 4.0.4 version.

Very glad to hear that you are working on the issue.

And, please, spend half an hour to IDB non-functional FoX test. I never saw such a strange thing. You will need a only a couple of steps.

Wishes,
Sergei

For those watching/waiting for updates regarding the FoX library, the imminent Intel Fortran Pro 11.1 update 4 contains a fix for the internal error involving file FoX_sax.f90 (Internal tracking id: DPD200140042), however, a subsequent internal error occurs for file m_dom_dom.F90. The subsequent internal error will be fixed in the next update, 11.1 update 5, tentatively scheduled for January.

(Internal tracking id: DPD200141787 - internal error m_dom_dom.F90)

After clearing the deck of internal compiler errors, I find there are still a number of failures with the self-check tests now under investigation:

RESULT utils/ Test Results:
RESULT utils/ Passed: 0
RESULT utils/ Failed: 5
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 43
RESULT wxml/ Failed: 65
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 8
RESULT wcml/ Failed: 113
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

I will keep this post updated as I learn more.

Quoting - Kevin Davis (Intel)

For those watching/waiting for updates regarding the FoX library, the imminent Intel Fortran Pro 11.1 update 4 contains a fix for the internal error involving file FoX_sax.f90 (Internal tracking id: DPD200140042), however, a subsequent internal error occurs for file m_dom_dom.F90. The subsequent internal error will be fixed in the next update, 11.1 update 5, tentatively scheduled for January.

(Internal tracking id: DPD200141787 - internal error m_dom_dom.F90)

After clearing the deck of internal compiler errors, I find there are still a number of failures with the self-check tests now under investigation:

RESULT utils/ Test Results:
RESULT utils/ Passed: 0
RESULT utils/ Failed: 5
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 43
RESULT wxml/ Failed: 65
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 8
RESULT wcml/ Failed: 113
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

I will keep this post updated as I learn more.

Hello Kevin!

Really nice to hear things are moving.

Now all we have is to wait for your investigaton to be finished.

Could you please to keep us informed (migh be a bit more details - just for pleasure to know/understand).

Thank you.

Sergei

Quoting savinovsv

Quoting - Kevin Davis (Intel)

For those watching/waiting for updates regarding the FoX library, the imminent Intel Fortran Pro 11.1 update 4 contains a fix for the internal error involving file FoX_sax.f90 (Internal tracking id: DPD200140042), however, a subsequent internal error occurs for file m_dom_dom.F90. The subsequent internal error will be fixed in the next update, 11.1 update 5, tentatively scheduled for January.

(Internal tracking id: DPD200141787 - internal error m_dom_dom.F90)

After clearing the deck of internal compiler errors, I find there are still a number of failures with the self-check tests now under investigation:

RESULT utils/ Test Results:
RESULT utils/ Passed: 0
RESULT utils/ Failed: 5
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 43
RESULT wxml/ Failed: 65
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 8
RESULT wcml/ Failed: 113
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

I will keep this post updated as I learn more.

Hello Kevin!

Really nice to hear things are moving.

Now all we have is to wait for your investigaton to be finished.

Could you please to keep us informed (migh be a bit more details - just for pleasure to know/understand).

Thank you.

Sergei

Hello Kevin!

Sad news again.

The latest releale 11.1.069 is still not capable of compiling FoX.
Waiting...

Thank you.
Sergei

Quoting Kevin Davis (Intel)

For those watching/waiting for updates regarding the FoX library, the imminent Intel Fortran Pro 11.1 update 4 contains a fix for the internal error involving file FoX_sax.f90 (Internal tracking id: DPD200140042), however, a subsequent internal error occurs for file m_dom_dom.F90. The subsequent internal error will be fixed in the next update, 11.1 update 5, tentatively scheduled for January.

(Internal tracking id: DPD200141787 - internal error m_dom_dom.F90)

After clearing the deck of internal compiler errors, I find there are still a number of failures with the self-check tests now under investigation:

RESULT utils/ Test Results:
RESULT utils/ Passed: 0
RESULT utils/ Failed: 5
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 43
RESULT wxml/ Failed: 65
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 8
RESULT wcml/ Failed: 113
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

I will keep this post updated as I learn more.

Hi Sergei,

My apologies for not updating earlier. I'm not sure what you're seeing with 11.1 Update 5 (11.1.069 - Linux), but that update contains the fix for the final internal compiler error affecting m_dom_dom.F90 (Internal tracking id: DPD200141787).I see many additional tests passing (vs. 11.1 Update 4) but I haven't had time to investigate the remaining failures further. I will try and post again once I know more.

Current results with 11.1 update 5:

RESULT utils/ Test Results:
RESULT utils/ Passed: 4
RESULT utils/ Failed: 1
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 108
RESULT wxml/ Failed: 0
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 74
RESULT wcml/ Failed: 47
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

Quoting Kevin Davis (Intel)

Quoting Kevin Davis (Intel)

For those watching/waiting for updates regarding the FoX library, the imminent Intel Fortran Pro 11.1 update 4 contains a fix for the internal error involving file FoX_sax.f90 (Internal tracking id: DPD200140042), however, a subsequent internal error occurs for file m_dom_dom.F90. The subsequent internal error will be fixed in the next update, 11.1 update 5, tentatively scheduled for January.

(Internal tracking id: DPD200141787 - internal error m_dom_dom.F90)

After clearing the deck of internal compiler errors, I find there are still a number of failures with the self-check tests now under investigation:

RESULT utils/ Test Results:
RESULT utils/ Passed: 0
RESULT utils/ Failed: 5
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 43
RESULT wxml/ Failed: 65
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 8
RESULT wcml/ Failed: 113
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

I will keep this post updated as I learn more.

Hi Sergei,

My apologies for not updating earlier. I'm not sure what you're seeing with 11.1 Update 5 (11.1.069 - Linux), but that update contains the fix for the final internal compiler error affecting m_dom_dom.F90 (Internal tracking id: DPD200141787).I see many additional tests passing (vs. 11.1 Update 4) but I haven't had time to investigate the remaining failures further. I will try and post again once I know more.

Current results with 11.1 update 5:

RESULT utils/ Test Results:
RESULT utils/ Passed: 4
RESULT utils/ Failed: 1
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 108
RESULT wxml/ Failed: 0
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 74
RESULT wcml/ Failed: 47
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

Dear Kevin!

Please find below the results of ifort (Linux) 11.1.073 FoX lib compilation tests. Note the first (utils results).

RESULT utils/ Test Results:
RESULT utils/ Passed: 0
RESULT utils/ Failed: 5

RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score

RESULT wcml/ Test Results:
RESULT wcml/ Passed: 74
RESULT wcml/ Failed: 47

RESULT wxml/ Test Results:
RESULT wxml/ Passed: 108
RESULT wxml/ Failed: 0

RESULT wcml/ Test Results:
RESULT wcml/ Passed: 74
RESULT wcml/ Failed: 47

Thank you for your kind attention.
I will wait for further improvements. Hopefully it will be done one day.

Sincerely yours,
Sergei

One step forward, two steps back.

The test results for this library really fluctuate with Intel compilers after 10.1. I will investigate and update when I know more.

Quoting Kevin Davis (Intel)
One step forward, two steps back.

The test results for this library really fluctuate with Intel compilers after 10.1. I will investigate and update when I know more.

Dear Kevin!

Here is the new round. Composer Studio XE against FoX-4.0.4.

RESULT utils/ Test Results:
RESULT utils/ Passed: 0
RESULT utils/ Failed: 5
RESULT utils/ See utils/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 452 passed.score
RESULT common/ Failed: 68 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT wxml/ Test Results:
RESULT wxml/ Passed: 108
RESULT wxml/ Failed: 0
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 74
RESULT wcml/ Failed: 47
RESULT wcml/ See wcml/test/failed.out for details of failed tests.

This is really strange. I had a feeling that the problem is somewhat important, but ...
As the authors of the FoX commented -"We do NOT trust to Intel compilers. They can NOT compile perfectly valid Fortran code".

Sincerely Yours,
Sergei

I wonder if they have reported those issues?

If not, we would appreciate having those reported so we can investigate and fix those if that's the case. That could also certainly help expedite resolving the constantly changing self-test results for this library with Intel compilers and make the Intel compilers reliable now and going forward for this library.

A few words on the nature of these tests: they seem (I have spent only a few minutes looking at them, I confess) to consist of comparing formatted output from the test programs with reference output produced by code compiled by a specific compiler.

Since list-directed I/O is not nailed down by the Fortran standard, output can vary slightly among different compilers. Even with format-directed I/O, there can be variations, e.g., n.nnnnEsmm versus 0.nnnnEmm and 0.nnnnesmm.

As your report in #8 shows, the the 12.0 compiler compiled all the tests and they were run, so the statement "They cannot compile perfectly valid Fortran code" is an exaggeration. Fortran standards do not guarantee the output of a valid Fortran program 100 percent, especially if floating point calculations are involved.

Strangely enough, three non-Intel compilers that I tried also led to reports of "all tests failed" for the utils/Tests checks. That is why I raised the objections in the first two paragraphs of this response.

What they need in making the checks more useful is to use text diff rules and tools that are more sensible than "match every character (including CR/LF?)".

You can help, as Kevin Davis stated, by spending the time to make a case for what the Intel compiler ought to do with one specific test. If you can say something along the lines of "according to the Fortran standard, lines mm-nn of the output should be exactly identical to <***>", then there will be something for the compiler developers to work with.

By the way, the original URL cited in #0 is dead. Here is the current source location URL:

http://www.ctcp.caltech.edu/galacticus/tools/FoX-4.0.4-full.tar.gz

Dear Kevin!
Here is another round. Composer Studio XE 2013 against FoX-4.1.2.

RESULT wxml/ Test Results:
RESULT wxml/ Passed: 141
RESULT wxml/ Failed: 29
RESULT wxml/ See wxml/test/failed.out for details of failed tests.
RESULT wcml/ Test Results:
RESULT wcml/ Passed: 124
RESULT wcml/ Failed: 6
RESULT wcml/ See wcml/test/failed.out for details of failed tests.
RESULT wkml/ Test Results:
RESULT wkml/ Passed: 129
RESULT wkml/ Failed: 6
RESULT wkml/ See wkml/test/failed.out for details of failed tests.
RESULT common/ Test Results:
RESULT common/ Passed: 518 passed.score
RESULT common/ Failed: 2 failed.score
RESULT common/ See common/test/failed.out for details of failed tests.
RESULT utils/ Test Results:
RESULT utils/ Passed: 5
RESULT utils/ Failed: 1
RESULT utils/ See utils/test/failed.out for details of failed tests.

I would comletely agree with mecej4 if not one thing. I am not doing coding of FoX. I am using it as a part of big project: SIESTA DFT package.
And the whole program crashes because of FoX library used fot XML output. So, I am critisizing ifort as a user, not as a developer.

By the way, I am using ifort. And I ever prefer it in comparison with gfortran. I am ringing a bell in attempt to make ifort better, not to accuse developing team.

Sergei

Thank you for the update, Sergei. Your continued interest and use of ifort is appreciated. I'll have a try with the newer versions.

登陆并发表评论。