Stricter DEBUG checks when using GCC #857
Assignees
Comments
|
@jedwards4b points out that check=all is pretty noisy. We'll try to set equivalent debug flags to what we do for (e.g.) intel. |
|
I propose adding the following to the clause for gnu in config_compilers.xml: @jedwards4b and others: does this sound reasonable? |
rljacob
pushed a commit
that referenced
this issue
Feb 8, 2017
…(PR #857) jenkins_generic_job: Add -j flag Will allow parallelism to be dialed-down from Jenkins. [BFB] * jgfouca/scripts/jenkins_generic_job_j_opt: jenkins_generic_job: Add -j flag
This was referenced May 26, 2017
fischer-ncar
added a commit
that referenced
this issue
May 28, 2017
This adds more debug checks with gnu. I thought this was done in #1339 (which claimed to fix #857), but I realized that that only added flags for the C compiler, not the Fortran compiler. This PR adds debug flags for the Fortran compiler, too. Ideally, we would also have 'invalid' in the ffpe-trap list. But at least with some versions of gfortran (confirmed with 5.3.0), gfortran's isnan (which is called in cime via the CPRGNU-specific shr_infnan_isnan) causes a floating point exception when called on a signaling NaN. (This problem showed up in the unit tests on my Mac: Adding 'invalid' caused the shr_strconvert unit tests to fail (specifically, the tests toString_prints_snan_r4 and toString_prints_snan_r8).) Test suite: Ran nearly all gnu debug tests that I could find in all CESM test suites. This included: aux_clm45 yellowstone-gnu: all pass and are bit-for-bit The following yellowstone tests: ERS_D_Ly3.f09_g16_gl5.TG1.yellowstone_gnu.cism-noevolve SMS_D.T31_g37_gl20.IGCLM45.yellowstone_gnu.cism-test_coupling SMS_D.f09_g16_gl20.TG1.yellowstone_gnu SMS_D_Ly1.f09_g16_gl20.THISTG.yellowstone_gnu SMS_D_Ld5.f10_f10.IMCRUCLM50BGC.yellowstone_gnu.mosart-default SMS_D_Ld5.f10_f10.ICRUCLM50BGC.yellowstone_gnu.rtm-default SMS_D.f09_g16_gl20.TG1.yellowstone_gnu ERI_D.T31_g37_rx1.A.yellowstone_gnu SMS_D_Ld1.5x5_amazon.ITEST.yellowstone_gnu.clm-40default All pass except a build failure for SMS_D_Ld1.5x5_amazon.ITEST.yellowstone_gnu.clm-40default which doesn't look related to these new debug flags. This is a CLM40 test that isn't important. The following cheyenne tests: SMS_D.f09_g16_gl20.TG1.cheyenne_gnu ERI_D.T31_g37_rx1.A.cheyenne_gnu SMS_D_Ld1.5x5_amazon.ITEST.cheyenne_gnu.clm-40default The A test passed. The ITEST failed as for yellowstone. The TG1 test failed, unable to link some lapack routines (seems unrelated to the new build flags). I ran these tests from https://svn-ccsm-models.cgd.ucar.edu/clm2/branches/externals_update_cesm2_0_alpha06k at r85267, with cime changes rebased onto cime5.3.0-alpha.13 Test baseline: For aux_clm45 tests, I compared against the above version of CLM unmodified. For other tests, I didn't do baseline comparisons Test namelist changes: none Test status: bit for bit Fixes #1615 User interface changes?: none Code review:
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
From @quantheory on August 4, 2015 17:5
This is an extension of the issue here:
http://bugs.cgd.ucar.edu/show_bug.cgi?id=1773
Right now we really don't have any of the major run-time checks turned on for GNU compilers. We should probably use an option such as
-fcheck=all. (This prints unnecessary creation of array temporary warnings, but unlike Intel's equivalent option, it should only print once per code location, so it shouldn't produce as much spam.)Copied from original issue: CESM-Development#91
The text was updated successfully, but these errors were encountered: