GNU Scientific Library - Bugs: bug #39152, make check errors with Intel icc...
You are not allowed to post comments on this tracker with your current authentication level.
bug #39152: make check errors with Intel icc 13.0.1
Submitter: | Patrick Alken <psa> | ||
Submitted: | Mon 03 Jun 2013 03:26:28 PM UTC | ||
Category: | Accuracy problem | Severity: | 3 - Normal |
Operating System: | Status: | None | |
Assigned to: | None | Open/Closed: | Open |
Release: |
Wed 05 Jun 2013 09:21:51 PM UTC, comment #4: |
Patrick Alken <psa>![]() |
Wed 05 Jun 2013 10:36:21 AM UTC, comment #3: Do you think a good approach would be to (a) determine exactly which Intel optimizations are problematic and to disable them or to (b) isolate the files for which GCC's -ffast-math causes problems, resolve those first, and revisit this? |
Rhys Ulerich <rhysu>![]() |
Mon 03 Jun 2013 04:43:38 PM UTC, comment #2: Intel icc turns on unsafe floating point optimizations by default, where gcc does not (gcc must specify -ffast-math for these). Compiling GSL with gcc -ffast-math produces many similar errors as seen by the icc logs below. To turn off unsafe optimizations in icc use -fp-model precise:
|
Patrick Alken <psa>![]() |
Mon 03 Jun 2013 03:37:05 PM UTC, comment #1: make CFLAGS="-O0" check passes all tests
|
Patrick Alken <psa>![]() |
Mon 03 Jun 2013 03:26:28 PM UTC, original submission:
see attached log |
Patrick Alken <psa>![]() |
Depends on the following items: None found
Items that depend on this one: None found
Follow 5 latest changes.
Date | Changed by | Updated Field | Previous Value | => | Replaced by |
---|---|---|---|---|---|
2013-06-03 | psa | Attached File | - | ![]() |
Added test-suite_sf.log, #28239 |
Attached File | - | ![]() |
Added test-suite_ode2.log, #28240 | ||
Attached File | - | ![]() |
Added test-suite_poly.log, #28241 | ||
Attached File | - | ![]() |
Added test-suite_multiroots.log, #28242 | ||
2013-06-03 | psa | Attached File | - | ![]() |
Added test-suite.log, #28238 |
Powered by Savane 3.12.
Corresponding source code
I made a new bug report (39171) to track the -ffast-math issue. For this report I think the main thing is to get make check working out of the box on intel compilers.
I like the idea of adding some autoconf code to add in -fp-model precise when icc is detected, though older icc's don't have this flag.