[cig-commits] [commit] devel: added a new comment about the s362ani + ATTENUATION bug (f5a3132)
cig_noreply at geodynamics.org
cig_noreply at geodynamics.org
Tue Apr 22 09:36:08 PDT 2014
Repository : ssh://geoshell/specfem3d_globe
On branch : devel
Link : https://github.com/geodynamics/specfem3d_globe/compare/bbdf11e34fb8e1b68a15e95b511f097103cf598e...1a5b4a182b85c2dbf7da31d567c33c1985eb474c
>---------------------------------------------------------------
commit f5a31324e86bab576f7ed2932cad6b8874cbadd1
Author: Dimitri Komatitsch <komatitsch at lma.cnrs-mrs.fr>
Date: Tue Apr 22 18:28:56 2014 +0200
added a new comment about the s362ani + ATTENUATION bug
>---------------------------------------------------------------
f5a31324e86bab576f7ed2932cad6b8874cbadd1
src/meshfem3D/meshfem3D_models.f90 | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/src/meshfem3D/meshfem3D_models.f90 b/src/meshfem3D/meshfem3D_models.f90
index bc380a5..e07e345 100644
--- a/src/meshfem3D/meshfem3D_models.f90
+++ b/src/meshfem3D/meshfem3D_models.f90
@@ -896,6 +896,22 @@
!! DK DK BUG
goto 777
+!! DK DK BUG
+!! DK DK BUG
+! the attenuation values for s362ani + ATTENUATION = .true. currently depend on the compiler used and on the compiler options used,
+! i.e. the lines below are unreliable. I am not sure why, it is probably either that elem_in_crust is not correctly set,
+! and / or that r_prem >(ONE-moho) is sensitive to roundoff noise and thus compiler-dependent,
+! in which case it should be changed to:
+!
+! if (abs(r_prem - (ONE-moho)) < some_epsilon_to_define) then...
+!
+! To be safe I have commented out the current code and checked that the bug (i.e. getting compiler-dependent results) is now gone.
+! If you fix that routine and check with Matthieu and Elliott that it then works fine on different machines
+! with different compilers please let us know and we will update the Git version
+! (including the master branch) and make an announcement to users.
+!! DK DK BUG
+!! DK DK BUG
+
select case (REFERENCE_1D_MODEL)
! case(REFERENCE_MODEL_PREM)
More information about the CIG-COMMITS
mailing list